South Asia Development Matters Hidden Debt Solutions to Avert the Next Financial Crisis in South Asia Martin Melecky Hidden Debt    iii Hidden Debt Solutions to Avert the Next Financial Crisis in South Asia Martin Melecky © 2021 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW, Washington, DC 20433 Telephone: 202-473-1000; Internet: www.worldbank.org Some rights reserved 1 2 3 4 24 23 22 21 This work is a product of the staff of The World Bank with external contributions. The findings, interpreta- tions, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy, completeness, or currency of the data included in this work and does not assume responsibility for any errors, omissions, or discrepancies in the information, or liability with respect to the use of or failure to use the information, methods, processes, or conclusions set forth. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Nothing herein shall constitute or be construed or considered to be a limitation upon or waiver of the privileges and immu nities of The World Bank, all of which are specifically reserved. Rights and Permissions This work is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) http://creativecommons.org/licenses/by/3.0/igo. Under the Creative Commons Attribution license, you are free to copy, distribute, transmit, and adapt this work, including for commercial purposes, under the following conditions: Attribution—Please cite the work as follows: Melecky, Martin. 2021. Hidden Debt: Solutions to Avert the Next Financial Crisis in South Asia. South Asia Development Matters. Washington, DC: World Bank. doi:10.1596/978-1-4648-1667-3. License: Creative Commons Attribution CC BY 3.0 IGO Translations—If you create a translation of this work, please add the following disclaimer along with the attribution: This translation was not created by The World Bank and should not be considered an official World Bank translation. The World Bank shall not be liable for any content or error in this translation. Adaptations—If you create an adaptation of this work, please add the following disclaimer along with the attribution: This is an adaptation of an original work by The World Bank. Views and opinions expressed in the adaptation are the sole responsibility of the author or authors of the adaptation and are not endorsed by The World Bank. Third-party content—The World Bank does not necessarily own each component of the content contained within the work. The World Bank therefore does not warrant that the use of any third-party-owned individual component or part contained in the work will not infringe on the rights of those third parties. The risk of claims resulting from such infringement rests solely with you. If you wish to re-use a component of the work, it is your responsibility to determine whether permission is needed for that re-use and to obtain permission from the copyright owner. Examples of components can include, but are not limited to, tables, figures, or images. All queries on rights and licenses should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; e-mail: pubrights@worldbank.org. ISBN (paper): 978-1-4648-1667-3 ISBN (electronic): 978-1-4648-1668-0 DOI: 10.1596/978-1-4648-1667-3 Cover design: Sergio Andrés Moreno Tellez, World Bank Group Library of Congress Control Number: 2021909369 Contents Acknowledgments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi Executive Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii   Spotlight ES.1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xviii Abbreviations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxiii Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Analytical Framework. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Empirical Findings. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Policy Recommendations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Notes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 1 Public-Private Partnerships in South Asia: Managing the Fiscal Risks from Hidden Liabilities While Delivering Efficiency Gains . . . 23 The Need to Carefully Manage the Fiscal and Economic Risks of PPPs. . . . . . . . . . . . . . . . . . . 23 Balancing the Efficiency Gains from PPPs against Their Risks and Liabilities . . . . . . . . . . . . . . 25 Booming Infrastructure PPPs, Their Country and Sector Distribution, and Signs of Distress in South Asia. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 Fiscal Risks from Contingent Liabilities Due to Early Termination of PPPs . . . . . . . . . . . . . . . . 30 Features of Contract Design That Matter: Exploring the Link between PPP Contract Design and Early Terminations of Highway PPPs in India. . . . . . . . . . . . . . . . . . . . . . 42 Improving Government Capacity, Due Diligence, and Contract Design to Better Manage the Fiscal Risks of the Growing PPP Programs in South Asia. . . . . . . . . . . . . . . 44 Annex 1A. Methodology to Determine the Value at Risk of a Public-Private Partnership. . . . . . 45 Annex 1B. Definitions of Variables. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 Annex 1C. Distribution of South Asian Public-Private Partnership Projects by Sector . . . . . . . . 48 Annex 1D. Imputing the Missing Values for Predictions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 Annex 1E. Model Selection. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 Annex 1F. Estimation Tables. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 Notes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 v vi   C o n t e n t s 2 State-Owned Banks versus Private Banks in South Asia: Agency Tensions, Susceptibility to Distress, and the Fiscal and Economic Costs of Distress. . . . . 57 The Upsides and Downsides of State-Owned Commercial Banks. . . . . . . . . . . . . . . . . . . . . . . . 57 The Omnipresence of State-Owned Commercial Banks in South Asia. . . . . . . . . . . . . . . . . . . . 59 Bank Business Models by Ownership Type: The Example of India. . . . . . . . . . . . . . . . . . . . . . . 60 Understanding Bank Distress and Its Main Factors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 Analyzing the Effect of Firms’ Banking with SOCBs Compared with Private Banks . . . . . . . . . 75 Policy Recommendations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 Annex 2A. Methodology for Determining Bank Distress. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 Annex 2B. Regression Tables: Probability of Distress for South Asian Banks and Adjustments to Distress, 2009–18. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 Annex 2C. Regression Tables for South Asian Scheduled Commercial Banks: Country Results, 2009–18. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 Notes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 3 South Asia’s State-Owned Enterprises: Surprise Liabilities versus Positive Externalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 The Importance of Paying More Attention to the Hidden Liabilities of SOEs in South Asia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 Describing the Opaque and Complex SOE Sector in South Asia Using Data. . . . . . . . . . . . . . . 103 Analyzing the Roots and Extent of Hidden Liabilities in South Asian SOEs. . . . . . . . . . . . . . . 107 What Drives the Contingent Liabilities from SOEs? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 The SOE Sector Has a Role to Play in South Asia, Such as through Its Long-Term Investment in R&D and Positive Spillovers on Private Firms. . . . . . . . . . . . . . . . . . . . . . . . . . . 116 Only a Combination of Internal and External Policy Reforms Can Help Better Manage Contingent Liabilities from SOEs in South Asia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 Annex 3A. Sources of Data about South Asian SOEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 Annex 3B. Summary Statistics and Estimations for Indian Enterprises. . . . . . . . . . . . . . . . . . . 121 Annex 3C. Productivity Estimation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 Notes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 4 Subnational Governments in South Asia: Balancing the Fiscal Risks of Government Decentralization with the Returns. . . . . . . . . . . . . . . . . . . . . . . . 133 The Promise and Risks of Fiscal Decentralization in South Asia. . . . . . . . . . . . . . . . . . . . . . . . 133 The Unclear Extent of Subnational Fiscal Liabilities and Rising Fiscal Risks in South Asia . . . 136 Fiscal Responsibility Legislation and Subnational Fiscal Risks. . . . . . . . . . . . . . . . . . . . . . . . . 140 Subnational Debt, Data, and Transparency: Lessons from Pakistan. . . . . . . . . . . . . . . . . . . . . 140 Estimating Contingent Liability Shocks, Adjustment Costs, and Mitigating Factors Using Data for India . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145 Results: Examining the Occurrence of Contingent Liability Shocks. . . . . . . . . . . . . . . . . . . . . 150 Improved Transparency and Fiscal Rules, the Disciplining Role of Markets, and Better Intergovernmental Frameworks Are Needed to Achieve Better Subnational Fiscal Outcomes in South Asia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158 Annex 4A. Methodology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162 Annex 4B. The Kalman Filter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 Annex 4C. Regression Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164 Notes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166 References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 C o n t e n t s   vii Boxes ES.1 Applying the Purpose, Incentives, Transparency, and Accountability (PITA) Recommendations in Fragile and Conflict-Affected Contexts. . . . . . . . . . . . . . . . . . . xvi 1.1 The Hidden Debt of National Highways in India . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 1.2 Low-, Medium-, and High-Risk Scenarios for Computing Losses to the Government from Contingent Liabilities of Public-Private Partnerships . . . . . . . . . . . . 38 2.1 Main Findings of the Overall Analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 4.1 Recommendations for Improving Fiscal Reporting and Transparency in Pakistan . . . 160 Figures O.1 Some South Asian Governments (India, Pakistan) Use State-Owned Commercial Banks, State-Owned Enterprises, and Public-Private Partnerships More Commonly Than the Global Benchmark While Others (Bangladesh, Sri Lanka) Are Catching Up . . . . 3 O.2 Analytical Framework: Links from Distress to Adjustments to Impacts . . . . . . . . . . . . . 6 O.3 Highlights of the Report’s Findings on Distress, Adjustments, and Impacts . . . . . . . . . . 9 O.4 State-Owned Commercial Banks Adjust Differently from Private Banks in Times of Distress, 2009–18. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 O.5 Annual Government Support for South Asian State-Owned Enterprises Could Account for More Than 2 Percent of GDP, on Average, Depending on the Country, 2015–17. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 O.6 A Profound Macrofinancial Crisis Could Trigger Failures among Public-Private Partnerships That Would Cost South Asian Governments up to 4 Percent of Revenues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 O.7 The Liabilities of Loss-Making State-Owned Enterprises in India, Pakistan, and Sri Lanka Have Been Huge, but More Than 80 Percent of Losses in Each Country Have Occurred in Only the Top 10 Loss-Makers . . . . . . . . . . . . . . . . . . . . . . 14 O.8 Local Investments in Indian States Fall Significantly with a Contingent Liability Shock, Keep Dropping the Year After, and Stay Well Below the Trend for Three Years . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 O.9 Checks and Balances on Government Executives Help Prevent Distress of Public-Private Partnerships . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 1.1 Active Portfolio of Public-Private Partnerships in Infrastructure in South Asia, 1990–2018. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 1.2 Sectoral Distribution of Public-Private Partnership Projects with Financial Closure in South Asia, by Country and Number of Cancellations, 1990–2018 . . . . . . . . . . . . . 29 1.3 Number of National Highway Public-Private Partnership Projects in India, by Year of Financial Closure, 2001–18. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 1.4 Traditional versus Public-Private Partnership Procurement of Infrastructure in India, 2001–17. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 1.5 Distribution of the Percentage of Contract Period Elapsed, 1990–2018. . . . . . . . . . . . 32 1.6 Distribution of Failures of Public-Private Partnerships over the Contract Period, 1990–2018. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 1.7 Estimates of Survival and Cumulative Hazard for Public-Private Partnership Projects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 1.8 Factors That Predict the Likelihood of Public-Private Partnership Distress. . . . . . . . . . 34 1.9 Distribution of Predicted Probabilities of Distress for Public-Private Partnerships in South Asia, from 2020 to the End of Contractual Period. . . . . . . . . . . . . . . . . . . . . . . 36 1.10 Composition of Public-Private Partnership Financing for Active Projects in South Asia, by Country, 1990–2018 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 viii   C o n t e n t s 1.11 Estimated Total Fiscal Costs from Early Termination of Public-Private Partnership Portfolio in South Asia, as a Percentage of GDP, 2020–24 . . . . . . . . . . . . . . . . . . . . . . 39 1.12 Estimated Total Fiscal Costs from Early Termination of the Public-Private Partnership Portfolio in South Asia, as a Percentage of Government Revenues for a Single Year. . . . . 40 1.13 Estimated Fiscal Costs from Early Termination of the Public-Private Partnership Portfolio in South Asia over Different Periods as a Percentage of Expected Government Revenues, 2020–24. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 1.14 Estimated Fiscal Costs from Early Termination of the Public-Private Partnership Portfolio Assuming Profound Macrofinancial Shocks, as a Percentage of Government Revenues, 2020–24. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 1.15 Number of Indian Highway Projects Canceled versus Not Canceled, by Contract Type and Financial Closure Year, 2010–14. . . . . . . . . . . . . . . . . . . . . . . . . . . 43 1E.1 Baseline Hazard Profile Estimates Using Semi-parametric Methods. . . . . . . . . . . . . . . 50 1E.2 Baseline Hazard Profile Estimates Using Parametric Methods. . . . . . . . . . . . . . . . . . . . 51 1E.3 Baseline Hazard Profile Estimates Using Flexible Parametric Methods. . . . . . . . . . . . . 51 2.1 South Asia: Share of State-Owned Commercial Bank Assets in Total Banking Assets, 2016. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 2.2 Bangladesh, India, and Pakistan: State-Owned Commercial Banks’ Underperformance Relative to Domestic and Foreign Private Banks, 2009–18 Average . . . . . . . . . . . . . . . . . . 60 2.3 India: Branch Networks and Total Credit, 2018 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 2.4 India: Selected Funding and Credit Indicators, 2018. . . . . . . . . . . . . . . . . . . . . . . . . . . 63 2.5 India: Selected Business Model, Performance, and Soundness Indicators, 2018 . . . . . . 64 2.6 South Asia’s Four Main Economies: Business Models and Strategies of State-Owned Commercial Banks versus Privately Owned Commercial Banks, 2009–18 . . . . . . . . . . 65 2.7 India: Characteristics of the Average Client Firms of Scheduled Commercial Banks, 2009–18. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 2.8 India, Bangladesh, Pakistan, and Sri Lanka: Interest Coverage Ratio by Bank Type, 2009–18. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 2.9 Differences in How State-Owned Commercial Banks and Domestically Owned Private Banks Adjust in Times of Distress. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 2.10 Capital Injections by the Indian Government to Distressed State-Owned Commercial Banks, FY2009–FY2020. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 3.1 Total Number and Average Revenue of South Asian State-Owned Enterprises, 2017 . . . 105 3.2 State-Owned Enterprise Revenue by Sector in India, Pakistan, and Sri Lanka, 2016–17. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 3.3 Net Profit/Loss of South Asian State-Owned Enterprises, 2014–17. . . . . . . . . . . . . . . 106 3.4 Share of State-Owned Enterprises That Reported a Loss in India, Pakistan, Sri Lanka, and Bangladesh, 2017. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 3.5 Breakdown of Loss-Making State-Owned Enterprises in India and Pakistan by the Total Number of Years in Which They Made a Loss, 2012–17. . . . . . . . . . . . . . . 107 3.6 Average Annual Government Support for South Asian State-Owned Enterprises, 2015–17. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 3.7 Share of Distressed Firms in India, 1989–2017. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 3.8 Share of Persistently Distressed Firms in India, 1991–2017. . . . . . . . . . . . . . . . . . . . . 109 3.9 Total Liabilities and Debts for South Asian State-Owned Enterprises, 2017. . . . . . . . 110 3.10 Outstanding Government Guarantees to State-Owned Enterprises, 2015–17. . . . . . . 110 3.11 Total Liabilities of Financially Distressed Central Public Sector Enterprises in India, 2000–17 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 3.12 Total Liabilities of Loss-Making State-Owned Enterprises in India, Pakistan, and Sri Lanka, 2005–17. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 3.13 Central Public Sector Enterprise Share of Industry Gross Fixed Assets and Industry Research and Development Expenses in India, 2016. . . . . . . . . . . . . . . . . . . 116 C o n t e n t s   ix 4.1 The Relationship between Private Investment and Fiscal Decentralization. . . . . . . . . 134 4.2 Share of Subnational Expenditure in General Government Expenditure. . . . . . . . . . . 136 4.3 India’s Subnational Debt in Comparison with Other Federations and Its Growing Access to Market Loans. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 4.4 Composition, Currency Denomination, and Interest Rate Structure of Provincial Debt, Pakistan, 2019. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142 4.5 Discrepancies and Understatements in the Accounting for Provincial Debt, Pakistan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 4.6 Public Debt Levels for Balochistan and Khyber Pakhtunkhwa, 2007/08–2017/18 . . . 144 4.7 Indian States’ Sources of Borrowing, 2018. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 4.8 Aggregate Subnational Debt and UDAY Debt, India. . . . . . . . . . . . . . . . . . . . . . . . . . 147 4.9 Outstanding Guarantees and Investment in India’s Guarantee Redemption Fund. . . . 148 4.10 Distribution of Above-the-Line and Below-the-Line Shocks to Subnational Debt, India. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149 4.11 Notable Subnational Shocks in India, 1990–2020. . . . . . . . . . . . . . . . . . . . . . . . . . . . 150 4.12 Estimated Fiscal Adjustments by Indian Subnational Governments to Contingent Liability Shock. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 4.13 Estimated Assistance from the Indian Central Government to Subnational Governments Hit by Contingent Liability Shocks. . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 4.14 Decreases in Indian Subnational Governments’ Gross Fixed Capital Formation following Contingent Liability Shocks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 4.15 Occurrence of Contingent Liability Shocks around Indian State Legislative Assembly Elections. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 4.16 Adoption of Transparency Measures and Fiscal Rules by Indian States, 2001–19. . . . 155 4.17 Occurrence of Contingent Liability Shocks around the Publication of Debt Reports in India. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 4.18 Occurrence of Contingent Liability Shocks around the Enactment of Fiscal Rule. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156 4.19 Breaching the Fiscal Rule: The Effect on Interest Rates Paid by Indian States. . . . . . . 157 4.20 Reliance of Indian States on Central Government Revenues and Share of State Debt, 2017. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162 Tables O.1 Implementing the High-Level Policy Recommendations for Public-Private Partnerships, State-Owned Commercial Banks, State-Owned Enterprises, and Subnational Governments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 1B.1 Project-Level Variables. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 1B.2 Institutional Variables. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 1B.3 Macroeconomic Variables. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 1C.1 Sectoral Distribution of Public-Private Partnership Projects with Financial Closure in South Asia, by Country. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 1E.1 Akaike Information Criterion and Bayesian Information Criterion under Different Orders of Flexible Parametric Methods. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 1F.1 Estimated Fiscal Costs (at the 99th Percentile) from Early Termination of the Public-Private Partnership Portfolio in South Asia, 2020–24. . . . . . . . . . . . . . . . . . . . . 52 1F.2 Logit Regression Estimates of Likelihood of Cancellation of Indian National Highway Public-Private Partnerships. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 2B.1 Probability of Distress for Indian Banks: Baseline Regression Results, 2009–18. . . . . . 83 2B.2 Probability of Distress for Indian Banks: Robustness Test Using z-Score, 2009–18. . . . 84 2B.3 Probability of Distress for South Asian Banks, 2009–18. . . . . . . . . . . . . . . . . . . . . . . . 85 2B.4 Baseline Regressions: Adjustments Given Distress, 2009–18. . . . . . . . . . . . . . . . . . . . . 85 x   C o n t e n t s 2B.5 Adjustments in Distress Compared with Private Banks Using Alternative Indicators of Distress, 2009–18. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 2B.6 Bank Adjustments in Distress across Bangladesh, India, Pakistan, and Sri Lanka, 2009–18. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 2B.7 Effect of Borrowing from State-Owned Commercial Banks on Investment by Client Firms, 2009–18. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 2C.1 Pooled Data Set for South Asia: Summary Statistics for Scheduled Commercial Banks, 2009–18. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 2C.2 Bangladesh: Summary Statistics for Scheduled Commercial Banks, 2009–18. . . . . . . . 90 2C.3 India: Summary Statistics for Scheduled Commercial Banks, 2009–18. . . . . . . . . . . . . 91 2C.4 Pakistan: Summary Statistics for Scheduled Commercial Banks, 2009–18. . . . . . . . . . 92 2C.5 Sri Lanka: Summary Statistics for Scheduled Commercial Banks, 2009–18. . . . . . . . . 93 2C.6 India: Average Characteristics of the Client Firms of Commercial Banks, 2009–18 . . . . 94 3A.1 Definitions/Categorization of State-Owned Enterprises Used in This Report. . . . . . . 121 3B.1 Summary Statistics of Prowess Data for Indian Central Public Sector Enterprises, 2016. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 3B.2 Probability of Indian Central Public Sector Enterprises Being Financially Distressed. . . . 122 3B.3 Volatility of Sales and Profit for Indian Central Public Sector Enterprises. . . . . . . . . . 122 3B.4 State Ownership and Financial Performance of Indian Central Public Sector Enterprises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 3B.5 Productivity for Indian Central Public Sector Enterprises and Non-Central Public Sector Enterprises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 3B.6 Corporate Governance Ratings and Financial Performance of Indian Central Public Sector Enterprises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 3B.7 Cumulative Losses and Debt-to-Asset Ratio of Indian Central Public Sector Enterprises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 3B.8 Distress and the Growth Rate of Paid-in Capital, Debt, and Fixed Assets of Indian Central Public Sector Enterprises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 3B.9 Probability of Negative Shocks to Sales and Profit for Indian Central Public Sector Enterprises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 3B.10 Sales Shock and the Growth Rate of Paid-up Capital, Debt, and Fixed Assets for Indian Central Public Sector Enterprises. . . . . . . . . . . . . . . . . . . . . . . . . . . 126 3B.11 Research and Development Expenditure: Comparing Indian Central Public Sector Enterprises to Other Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 3B.12 Relationship between Public Sector Research and Development and Private Sector Performance in India. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 4.1 Categorization of Fiscal Risks at the Subnational Level. . . . . . . . . . . . . . . . . . . . . . . . 137 4.2 Legal Authority for Subnational Governments in South Asia to Borrow and Issue Guarantees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 4.3 Vertical Imbalances between Subnational and Central Governments in South Asia. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 4.4 Long-Established Fiscal Rules for the Central Government of Several South Asian Countries. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141 4C.1 Effect of Contingent Liability Realizations on Fiscal Variables. . . . . . . . . . . . . . . . . . 164 4C.2 Effect of Contingent Liability Realizations on Assistance from the Central Government . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164 4C.3 Effects of Contingent Liability Realizations on Gross Fixed Capital Formation. . . . . 165 4C.4 Effect of Adoption and Transparency of Fiscal Rules on the Likelihood of Contingent Liability Shocks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165 Acknowledgments T his report was prepared by a team led prepa ration: M a r tin R a ma, Z oubida by Martin Melecky (Lead Economist) Allaoua, and the Global Practice leadership under the guidance of Hans Timmer team of Vivien Foster, Mary C. Hallward- (Chief Economist) and Hartwig Schafer Driemeier, Alfonso Garcia Mora, Marcello (Vice President). Core members of the team Estevão, Demetrios Papathanasiou, and included Matias Herrera Dappe and Burak Guangzhe Chen; for chapter 1: Alvaro Turkgulu (chapter 1), Katie Kibuuka (chap- Pedraza Morales, Fernanda Ruiz Nuñez, ter 2), Siddharth Sharma (chapter 3), and Diane Dorothy Menville, Deblina Saha, Florian Blum and Pui Shen Yoong (chapter Reenu Aneja, Tema Alawari Kio-Michael, 4). Able research support and inputs were Cigdem Aslam, Samuel Maimbo, Shomik provided by Martin Brun, Mario di Filippo, Raj Mehndiratta; for chapter 2: Davide Yunfan Gu, Shaheen Malik, Tom O’Keefe, Mare, Esperanza Lasagabaster, Marius Viviana Maria Eugenia Perego, Di Yang, V ism a nt as , I l ias Sk a m nelos , S a n ke t and Kyoung Yang Kim. Tobias Akhtar Mohapat ra , a nd pa r t icipa nt s i n t he Haque and Alex Sundakov contributed to CA FR A L –World Bank Conference on the boxes on fragility, conflict, and violence “State Intervention in the Financial Sector”; country context and successful public-pri- for chapter 3: Ana Cusolito, Francesca de vate partnerships in South Asia, respectively. Nicola, Anoma Kulathunga, Nazmus Sadat Neelam Chowdhry provided able admin- Khan, and Dhananath Fernando; for chapter istrative support to the report team. The 4: Frederico Gil Sander, Manuela Francisco, report was edited by Nancy Morrison. Zehra Aslam, Adnan Ashraf Ghumman, We are grateful to the peer reviewers who Nyda Mukhtar, Rangeet Ghosh, and Fritz supported the preparation of this report from Bachmair; for the decision review draft: the concept stage, through the authors’ work- Najy Benhassine, Faris H. Hadad-Zervos, shop, to the decision stage: Ila Patnaik, Ugo Henry Kerali, Hideki Mori, and Mercy Panizza (both external), David Duarte, Eva Tembon; and for the concept review draft: Gutierrez, and Abha Prasad (all World Bank). Enrique Blanco Armas, Fei Deng, Jorge We thank the following colleagues for Jose Escurra, Alexander Anthony Ferguson, their guidance, suggestions, comments, Sereen Juma, Luc Lecuit, Tatiana Nenova, and inputs at various stages of the report’s Janardan Prasad Singh, and Volker Treichel. xi xii   A c k n o w l e d g m e n t s Our thanks also go to the communica- We gratefully acknowledge financial sup- tions and production teams, including Elena port from the United Kingdom’s Foreign, Karaban, Yann Doignon, Diana Ya-Wai Commonwealth and Development Office Chung, Mark McClure, Jewel McFadden, through the Program for Asia Connectivity and Sergio Andrés Moreno Tellez. and Trade. Executive Summary T   he COVID-19 (coronavirus) crisis than problems compared with SOCBs and has expanded public direct interven- SOEs. Some PPPs in South Asia have been tions through state banks and enter- highly successful in terms of both public ben- prises to aid economies, but with possible efits and private returns (see spotlight ES.1 at risks to debt sustainability, long-term pro- the end of this summary). ductivity, and equality. The ongoing COVID-19 crisis, which has Compared with other regions, South Asia is sent economies in South Asia and other more exposed to the risk of hidden debt and parts of the world into a deep recession, has mounting contingent liabilities from SOEs, created the need for public relief efforts. It SOCBs, and PPPs because of its greater reli- has also raised South Asia’s debt levels and ance on state off-balance sheet operations. future contingent liabilities. This once-in- South Asia has the largest share of SOCBs in a-century shock and the subsequent vast terms of banks’ total assets across developing deployment of public resources have come country regions, has the highest use of SOEs on the back of the debt wave that has formed together with East Asia and Pacific,1 and is over the past decade. Large amounts of debt among the top three regions in the use of PPPs obligations are likely to resurface for central in infrastructure (along with Eastern and governments as many troubled state-owned Central Europe and Sub-Saharan Africa). For enterprises (SOEs) and state-owned com- instance, the SOE sector in both India and mercial banks (SOCBs) will call for support Pakistan is more than twice as large as the through bailouts. Prematurely terminated international benchmark, controlling for size public-private partnerships (PPPs) may of the economy. Overall, India and Pakistan require large public payments as settlements. are among the biggest users of public agents Increased leveraging of SOEs, SOCBs, such as SOEs, SOCBs, and PPPs. Other South and PPP interventions during the COVID- Asian countries are prominent users of one or 19 crisis must be safeguarded against pos- two of the public agents, such as Bhutan’s and sible exploitation by elites for their own Sri Lanka’s heavy use of SOCBs. Not only do benefit and must minimize misallocations of South Asian central governments rely heavily resources in the economy that could reduce on off–balance sheet operations to aid eco- productivity in the medium to long term. On nomic development, but so do s ­ ubnational balance, PPPs have yielded more successes governments (SNGs). xiii xiv  E x e c u t i v e Summary The high degree of reliance on SOEs, whether private or public—could have been SOCBs, and PPPs reflects the intention to in state of distress between 2009 and 2018.2 help accelerate inclusive economic develop- Importantly, the greater the ownership share ment through direct state interventions, but of government in an SOCB, the higher the the negative repercussions have been largely probability of bank distress. ignored. Because subnational governments are con- The strong preference of South Asian govern- strained from borrowing autonomously, they ments for direct interventions in the economy do not experience overall distress. They do, and markets comes with a price. The reality however, experience significant shocks from is that, on average, the efficiency of South triggered contingent liabilities: that is, when Asian SOEs, SOCBs, and PPPs is well below their own off–balance sheet operations— the international benchmark. South Asian including subnational SOEs, SOCBs, and governments thus face a trade-off when they PPPs—go bad. Over the past two decades, rebuild after the current crisis. They must a contingent liability shock has hit a South balance the tension between using SOEs, Asian subnational government about 10 SOCBs, and PPPs to maximize socially ben- percent of the time, on average. eficial investments and minimizing the risk of large surprise liabilities due to inefficiencies When public agents enter distress, govern- and mismanagement of risks. South Asia gov- ments face the daunting dilemma between ernments must ensure that their off–­ balance bailout and reduced economic activity. sheet operations and their mutual intercon- Typically, they resort to bailouts. nectedness do not become the source of Overall, SOEs and SOCBs enjoy soft bud- the next financial crisis in the region. This get constraints and central government report studies the trade-off between tackling bailouts after they get into distress. The development challenges through direct state same holds for subnational governments presence in the market and avoiding unsus- that e­ xperience a financial shock after tainable debt due to economic inefficiencies their own off–balance sheet operations of off–balance sheet operations. have failed. These soft budget constraints and bailouts could be partly motivated by Financial distress of public agents is not a “guilt” stemming from the government rare event in South Asia. Off–balance sheet having set unclear mandates for SOCBs operations of both national and subnational and SOEs through ad hoc requests to governments become distressed frequently help with economic stimulus and other SOCBs fare the worst. political agendas. These unclear mandates In South Asia and globally, 8 percent of PPPs hinder financial accountability, hard bud- are canceled early (terminated) before their get constraints, and fair monitoring of contracts expire. Railroad, treatment plant, performance. and toll road projects under PPP arrange- The public agents and subnational gov- ments appear to be the most vulnerable to ernments on one side of the social contract distress. SOEs enter distress even more often. and the central government on the other have For instance, India’s central public sector settled in a bad equilibrium—one that the enterprises (CPSEs) are 15 percentage points economist would characterize as the tragedy to 21 percentage points more likely to enter of the commons: that is, a situation when a distress than similar private firms. Other common pool of fiscal resources is overused South Asian countries may confront even for self-interest. More transparent setting of greater distress problems in their SOE sec- purpose and better design, incentives, and tors—even if those problems are seemingly monitoring of SOEs, SOCBs, and PPPs— well hidden because data are not available. including a clearer definition of social versus Across Bangladesh, Pakistan, and Sri Lanka, commercial mandates—can help create a bet- almost half of the banks—regardless of ter equilibrium. E x e c u t i v e S u m m a r y    xv The fiscal costs of failing off–balance sheet function of SOCBs for the credit cycle, how- operations are sizable and can markedly ever, comes at the cost of significant credit reduce the fiscal space available to South misallocation—away from successful firms Asian governments. and especially SMEs. It helps unproductive This report estimates that a systemic macro- zombie firms linger in the economy in the financial crisis could trigger PPP failures that medium term and stalls needed reallocation would cost South Asian governments more of capital (and labor) to enable productive than 4 percent of revenues—and the fallout investments. from the current COVID-19 crisis could be At the subnational level, local invest- even more severe. The potential fiscal costs ments in the Indian states fall significantly from distressed SOEs have been even more in the year of a contingent liability shock, overwhelming. In Pakistan, the total liabilities continue to decline in the year after, and of chronic loss-makers—defined as SOEs that remain below the trend for three years after made a loss in three out of the five past years— the event. Low fiscal capacity and perhaps have been about 8 percent to 12 percent of the greater reliance of local private invest- GDP in recent years, several times more than ments on complementary public investments the country’s public spending on education in can drive the adverse impact of contingent FY2019/20.3 In Sri Lanka, the liabilities of liability shocks on total local investments. loss-making SOEs have hovered at 4 percent For these reasons, recklessly leveraged to 5 percent of GDP. Interestingly, in every public capital through badly designed off– country studied, just the top 10 loss-making balance sheet operations of subnational SOEs account for more than 80 percent of governments is very costly for local econo- the total losses in the SOE sector—suggest- mies and communities in South Asia. ing that the problem could be managed. In India, the cumulative recapitalization of How can the downside risks of leveraging SOCBs from FY2016 through FY2020 was public capital be mitigated and the upside equivalent to almost one and a half times the benefits enjoyed? How can overleveraging and country’s planned public spending on health uneconomical use of public capital be pre- care in FY2021/22.4 The recapitalization vented and thus the threat of a financial cri- needs are estimated to increase markedly in sis minimized? Through purpose, incentives, FY2021/22, including due to repercussions of transparency, and accountability (PITA). the COVID-19 crisis. Four interconnected principles form the basis of the reform agenda to ensure that public Distressed public agents also inflict sub- capital is leveraged responsibly in South Asia stantial costs on the real economy and local and to minimize the threat of a financial cri- business. When a subnational government sis originating from government off–balance is hit by a contingent liability shock, local sheet operations (and their interconnections): investments suffer for several years. Bailouts of SOCBs can help supply credit to the econ- P = Purpose. The purpose of off–balance omy in crises, but they also pave the way sheet operations and leveraging of public cap- for an uneven and frail recovery—one that ital through SOEs, SOCBs, and PPPs must be is especially unfair to small and medium clearly defined by the central government or enterprises (SMEs). subnational government as the establisher, In episodes of systemic shocks—such as the owner, or sponsor. This includes formulating global financial crisis and the COVID-19 cri- a clear vision or mission, setting time-bound sis—many banks experience distress. While objectives, and defining corresponding key private banks deleverage and curtail lend- performance indicators (KPIs). ing, SOCBs receive capital and debt support from the state to continue (or even increase) I = Incentives. Institutions, rules, and con- lending. This short-term, positive stabilizing tracts must be structured in a way that creates xvi  E x e c u t i v e Summary proper incentives to perform in line with the Economic transparency is also required. It defined purpose. The operational costs of should start with publicly disclosing the pol- SOEs, SOCBs, and PPPs often exceed mar- icy and purpose of SOEs, SOCBs, and PPPs, ket costs to fulfill their purpose and may thus and enforcing the requirement that each pub- require fiscal subsidies. The nature and extent lic agent publish its theory of change for ful- of these operational costs and subsidies must filling its objective and purpose. be determined and linked to the government’s budgetary and debt management frameworks. A = Accountability. The electorate, civil society organizations (CSOs), industry T = Transparency. Two types of transpar- associations, media, and financial markets ency are needed. Debt transparency and rel- must support reforms to enact the prin- evant data collection are critical to enabling ciples of purpose, incentives, and trans- both central and subnational governments to parency (PIT) so that off–balance sheet assess the big picture of how SOEs, SOCBs, operations of governments cannot be used and PPPs shape the fiscal space and con- for political self-interest or side deals—or tribute to the overall public debt—including at least make it harder to do so. Once the direct obligations and explicit and implicit reforms are implemented, the electorate, guarantees. CSOs, industry associations, and financial BOX ES.1 Applying the Purpose, Incentives, Transparency, and Accountability (PITA) Recommendations in Fragile and Conflict-Affected Contexts Fragility, conflict, and violence (FCV)-afflicted conflict, the informality of the private sector rises. countries and jurisdictions are characterized by This response and the state’s various inabilities can weak institutions and thus present challenging lead to the development of a “gray” economy, in contexts for effectively operating, reforming, or which private sector activity is irregular and largely privatizing state-owned enterprises (SOEs) and opportunistic and operates without regulation. In state-owned commercial banks (SOCBs), as well financial markets, beyond local short-term tradi- as structuring successful public-private partner- tional lending markets based on family or kinship, ships (PPPs). credit is provided at either steep rates or is con- In FCV countries such as Afghanistan, persis- fined to the individuals and businesses best- tent security and governance constraints on the connected to officials controlling SOCBs. effective delivery of central government functions Regularly using public funds to bail out or and services can prompt the local private sector recapitalize inefficient SOEs and SOCBs creates or communities to provide these services instead. tension because it competes against other and Further, the state may be unable to provide the higher priorities by the use of public money. This institutional underpinnings of markets—such as tension may propel efforts by the state to divest property rights and contract protections—as well or fully privatize state enterprises, banks, and as basic infrastructure and services. State institu- other assets, as occurred in Bosnia and tions are often captured by political elites to Herzegovina and Mozambique during past peri- extract rents rather than to serve the public inter- ods of fragility and conflict. This generally posi- est because proper accountability mechanisms tive move is, however, not without risk. For are missing, among other problems. instance, the privatization of SOCBs and the At the same time, in response to disrupted social licensing a private banks to recapitalize a failing networks and formal structures during periods of banking system have sometimes been highly (Box continues on next page) E x e c u t i v e S u m m a r y    xvii BOX ES.1 Applying the Purpose, Incentives, Transparency, and Accountability (PITA) Recommendations in Fragile and Conflict-Affected Contexts (continued) nontransparent—especially when initiated ­ during rule of law and weakly enforced contracts; (3) wartime. In the former Yugoslavia, this process binding capacity constraints, including for the resulted in large asset transfers to war criminals. governance and management of SOEs, SOCBs, By contrast, with a lag of several years following and PPPs; and (4) fragmented and dysfunctional the cessation of civil war, Mozambique was able financial markets lacking the capacity to price to privatize its commercial banks by bringing in risk and exert market discipline. foreign partners. In FCV areas, it is challenging to identify and Development agencies are increasingly con- carefully align the appropriate reform path with scious of the ways that development interven- the country context. In contexts in which mar- tions—including public sector reforms (such as kets are distorted and dysfunctional, reforms privatization of SOEs and civil service reforms), should not rely on unrealistic assumptions that major infrastructure projects, and community- exposure to market discipline is possible—even level decentralization programs—can “do harm” via full or partial privatization. In contexts in and contribute to tensions in already fraught situ- which policy makers are not necessarily held ations unless they are well at tuned to knowledge accountable by citizens, reforms should reflect a of local context and political economy. realistic assessment of how and through which The four-pronged reform agenda advanced channels increased transparency can change by this report and centering around purpose, incentives. Planned reforms should be informed incentives, transparency, and accountability by analyses of the broader context of the politi- (PITA) also pertains to FCV contexts in South cal economy as well as the specific incentives Asia when sensitized to the local context and affecting decision makers and how these can be political economy. Sensitizing proposed concrete influenced using the available policy levers. actions is perhaps even more important in International development organizations can Afghanistan than in other countries of the play a key role in creating incentives for reform, region. Some of the challenges that set enabling sustained change leadership, and pro- Afghanistan apart involve (1) political systems viding required technical assistance over the dominated by patronage that can reduce the medium to long term. effectiveness of transparency measures; (2) weak Source: World Bank. markets must remain vigilant and active. contexts, including areas affected by fragility These actors need to keep testing the jus- and conflict (see box ES.1). tifications for continuing the off–balance In closing, while the government must sheet operations—such as the existence lead in reform, it takes a concerted effort by of SOEs and SOCBs, as well as the use of society to ensure that the off–balance sheet PPPs for the right (socially beneficial) pur- operations of government serve the right pose and with desirable results. socioeconomic purpose and responsibly lever- These PITA principles must be based on age public capital for the sake of more rapid realistic assessments of the national and sub- and more equitable development. Falling national contexts and available policy levers. short of this task, South Asian countries face With care, they can be applied to a variety of the threat of possible financial crises soon. xviii  E x e c u t i v e Summary Spotlight ES.1: Examples of Highly Successful PPPs in South Asia Public-private partnerships (PPPs) that are capacity, from 250,000 twenty-foot equivalent highly successful allocate risk between the units (TEUs) to 1.1 million TEUs per year; and public sector and the private sector in a (2) improve efficiency in terms of the number sustainable way. These transactions should of containers handled per hour and the aver- instill confidence, so that the private capital is age waiting time for container vessels. mobilized for a particular project and further South Asia Gateway Terminals (Private) private participation is encouraged. From the Limited (SAGT partnership) won the 30-year government’s viewpoint, a highly successful BOT contract. The SAGT partnership has PPP project must not only be fiscally sustain- financed infrastructure improvements and able, but must also ensure delivery of the leases the terminal from SLPA, which also demanded infrastructure services—avoiding earns income as a minority equity holder. situations of expensive, unnecessary “white While the partnership is majority-owned by elephant” projects. A PPP project can be domestic entities, an international terminal partially successful if it meets some but not operator (APM Terminals) provides key oper- all the criteria for a high degree of success. ational expertise. The increased traffic to the For example, the government may succeed Port of Colombo also provides additional in protecting its fiscal position by transfer- benefits to SLPA through port fees charged ring risk to the private sector, but the proj- directly to vessels that use the terminal. ect could be a financial failure for the private SAGT handled 2 million TEU throughput investors, causing loss of confidence in a PPP in 2018, and productivity increased greatly. program. The four examples that follow, Time spent loading, offshoring, and reposi- ranging from the local to the national level, tioning cargo rose from 12 gross gantry moves illustrate features of highly successful PPPs in per hour in 1998 to 30 by 2003. Waiting time four South Asian countries. for vessels in berth decreased from 6.9 hours in 1997 to 0.9 hours by 2003. SAGT has invested in advanced terminal handling equip- Sri Lanka ment and tracking technology. The South Asia Gateway Terminal (SAGT) During recent, and somewhat controver- is a 30-year build-operate-transfer (BOT) sial, attempts to secure PPP contracts for container terminal project within the Port of other terminals at the Port of Colombo, pub- Colombo signed in September 1999. The proj- lic debate identified SAGT as an undoubted ect was conceived when the government rec- success and centered on whether new PPPs ognized that the Port of Colombo—then fully were sufficiently closely modeled on the operated by the Sri Lanka Ports Authority SAGT approach. (SLPA)—was performing poorly. In addi- tion, SLPA did not have access to financing to Nepal enable additional dredging and other invest- ment required to handle the largest container The Khimti I Hydropower Project (Khimti) is vessels. By the mid-1990s, projections indi- the first power sector PPP and the first foreign- cated that trans-shipment business would start owned and foreign-operated power project in diverting to ports outside Sri Lanka. Nepal, developed using a build-own-operate- To remain competitive, the government of transfer (BOOT) contract. Khimti is a run-of- Sri Lanka sought private expertise and the the-river hydroelectric power generation plant private operation of one of the port terminals. with an installed capacity of 60 megawatts. The PPP included extending the quay, repair- The project is located about 100 kilometers ing and upgrading existing facilities, and east of Kathmandu along a tributary of one of purchasing new equipment to (1) quadruple Nepal’s major rivers. E x e c u t i v e S u m m a r y    xix Khimti is owned and operated by Himal the project is well positioned to take adaptive Power Limited (HPL), a special purpose com- measures to address climate change risks. pany. The government of Nepal and HPL entered into both a project agreement and a power purchase agreement (PPA) in January India 1996. The government issued HPL a 50-year project license that gives it the right of unin- The Hyderabad Rajiv Gandhi International terrupted water flow of the Khimti River. At Airport (Hyderabad Airport) was the first the end of the 50-year period, HPL will trans- greenfield PPP airport in India. It is primar- fer the project to the government. The Nepal ily owned and operated by the GMR Group, Electricity Authority (NEA) and HPL also one of India’s largest listed infrastructure entered into a 20-year PPA from the commer- companies, through a 30-year concession cial operation date, which was in July 2000. agreement with the government of India At the end of the PPA period (July 2020), (through the Ministry of Civil Aviation). HPL was to transfer 50 percent ownership The concession was signed in December to NEA. The remaining 50 percent will be 2004, and the Hyderabad Airport was inau- transferred at the end of the 50-year project gurated in March 2008. license period. However, the initial trans- H I AL , the project company for the fer process remains delayed due COVID-19 Hyderabad Airport PPP was a joint venture issues, and the project is operating under an between the GMR Group (63 percent); the interim agreement. government of India (13 percent); the gov- HPL was established on February 21, ernment of Telangana (13 percent); and 1993, by one of Nepal’s leading private com- Malaysia Airports Holding Bhd (11 percent). panies—Butwal Power Company Limited— Under the PPP contract, HIAL was responsi- and three Norwegian companies: Statkraft ble for designing, financing, building, com- SF; ABB Energi AS (now ABB Kraft); and missioning, operating, maintaining, and Kvaerner AS (now GE Hydro). The World managing the Hyderabad Airport. HIAL Bank Group’s Multilateral Investment won the concession by promising to pay the Guarantee Agency (MIGA) issued a US$32.8 highest concession fee as well as by assum- million guarantee in 1996 to the Norwegian ing all commercial and financing risks. investors to insured their equity against cur- The Hyderabad Airport PPP was c ­ onceived rency transfer limitations, expropriation, and to replace an existing airport in Hyderabad. war risks. The construction was financed In return for the government's equity share, half by debt and half by equity, with the the concession agreement required the gov- World Bank Group’s International Finance ernment to (1) close the existing airport to Corporation (IFC) among the parties provid- commercial operations and pay any resulting ing project finance. costs or claims; (2) not allow a new or exist- The project was successful in transferring ing airport to be developed within 150 aerial construction cost risks associated with such kilometers of Hyderabad Airport within a hydropower projects to the private sector. The certain period; (3) not provide other major Khimti project had an unusually large tun- airports with an unfair competitive advan- neling component, with an 890-meter access tage; and (4) allow HIAL to propose amend- tunnel and 7,620-meter headrace tunnel. ments to the concession agreement, should Despite this, it was completed on time and a change in law lead to an increase in costs on budget. Participation by highly capable or impair the financial position of HIAL international specialist firms was a key ele- exceeding a certain amount. ment of success. The project remained fully The concession agreement capped the operational following an earthquake in April aggregate liability of the government of India 2015. A recent review by IFC concluded that in respect to any breach, default, or change in xx  E x e c u t i v e Summary law to approximately US$14 million (Rs 100 Various sections of the road are main- crore). tained by private companies—such as the Hyderabad Airport was commissioned mobile operator Grameen, Banglalink, and in a record time of 31 months. The air- the stainless steel manufacturer KSRM—that port’s initial capacity was 12 million pas- would otherwise be renting advertising space sengers per year, but before the onset of on the sides of private buildings. the COVID-19 pandemic, the airport was Public opinion surveys indicate more handling more than 21 million passengers. than 80 percent satisfaction with the out- In March 2021, the government of India comes. This simple PPP model enables KCC announced its intention to sell its stake in to ensure maintenance that would otherwise the Hyderabad Airport (among others) as not have happened. part of its ambitious program to monitize assets. Lessons Learned Successful PPPs can range in size from Bangladesh micro solutions to local problems to large- K hu l n a i s t h e t h i rd - l a r g e s t c it y i n scale, transformative infrastructure projects. Bangladesh. It is connected to Bangladesh’s Common features of successful PPPs involve capital, Dhaka, and other regional cities realistic and clear delineations of responsibil- through rail, road, water, and air transport. ities between the public and private sectors, Improved transport links have resulted in a careful selection of private sector counter- large influxes of people from nearby cities, parts (including appropriate combinations of affecting the city’s liveability and ­ putting international skill and capability with local pressure on the city’s infrastructure and capital), and transparency regarding the risks services. Within the city, major roads and and rewards allocated to the private sector. urban landscaping are constructed by the Roads and Highways Department (RHD) and the Khulna Development Authority Bibliography (KDA). Sri Lanka As is common, the city budget for roads A DB (A si a n D evelopment B a n k). 2 02 0. and related infrastructure tends to be allo- “Democratic Socialist Republic of Sri Lanka: cated for new construction projects, with National Port Master Plan.” Technical limited funds available for maintenance. As Assistance Consultant’s Report prepared by a result, roads and road medians tend to Maritime & Transport Business Solutions B.V. deteriorate, reducing traffic safety and urban (MTBS), February. https://www.adb.org/sites/ amenities. default ​/files/project-documents/50184/50184​ In 2016 and 2017, the Khulna City -001-tacr-en_7.pdf. Corporation (KCC) entered into a series of Bisbey, J. “Public-Private Partnerships for innovative agreements with multiple private Sustainable Port Development.” UNESCAP parties to take over responsibility for main- (Un ited Nations E conom ic and S ocial Commission for Asia and the Pacific). https:// taining 4.6 kilometers of city road medians www.unescap.org/sites/default/files/1.4%20 in return for the right to install advertising PPP%20for%20sustainable%20port%20dev​ billboards on the section allocated to each _Jyoti%20Bijbey_ESCAP.pdf. private party. The objectives of these small- SAG T web site: ht t p s: // w w w. s ag t .com .l k​ scale PPPs are to ensure that landscaping and /­about-us/journey-of-sagt.html. maintenance make road crossing safer as U N DP (United Nations Development well as to improve the attractiveness of the Programme). 2012. “Colombo, Sri Lanka Case streetscape. The positioning and size of the Study (Port Expansion).” November. https:// permitted billboards are specified, to be con- www.esc-pau.fr/ppp/documents/featured​ sistent with those objectives. _projects/sri_lanka.pdf. E x e c u t i v e S u m m a r y    xxi Nepal India Bhatta, S. “Public Private Partnership in Nepal.” Hyderabad Airport PPP Concession Agreement. National Planning Commission. https://www​ Dow n load able f rom World B a n k PPP .unescap.org/sites/default/files/Nepal%20PPT​ Legal Resource Center website: https://ppp​ .pdf. .worldbank.org/public-private-partnership​ Himal Power Limited website: library/concession-agreement-development​ ­ https://hpl.com.np/projects/khimti-power​ -construction-operation-and-maintenance​ -plant/ -hyderabad-internatio. ­ https://hpl.com.np/projects/project-license​ -agreement-2/ ­ https://hpl.com.np/projects/project-license​ -agreement-2/power-purchase-agreement-ppa/ Bangladesh ­ https://hpl.com.np/projects/project-license​ -agreement-2/project-generation-and-cost/ Haque, M. N., M. Saroar, M. A. Fattah, MIGA website: and S. R. Morshed. 2020. “Public-Private ­ https://www.miga.org/project/himal​- power​ Pa r t nersh ip for Ach iev i ng Sust a i nable -limited Development Goals: A Case Study of Khulna, ­ https://www.miga.org/project/himal-power​ Bangladesh.” Public Administration and -limited-0 Policy 23 (3): 283–98. xxii  E x e c u t i v e Summary Notes and taxes (EBIT) by its interest expense during a given period. Regions are as defined by the World Bank. 1.  Pakistan’s public expenditure on education 3.  Regional comparisons of SOEs are difficult. as a percentage of GDP is estimated at 2.3 This study uses the database of SOEs main- percent for FY2019/20, making it the lowest tained by the Organisation for Economic in the region. Co-operation and Development (OECD) for Recapitalization of public banks entailed a 4.  several regions outside South Asia. Yet in that cumulative capital infusion of Rs 3.16 lakh database, for instance, East Asia and Pacific crore from FY2016 through FY2020. By con- is represented only by China and Eastern and trast, the Union Budget 2021–22 p ­ roposed a Central Europe by Poland. significantly smaller outlay of Rs 2.23 lakh crore That is, their interest coverage ratio (ICR)—a 2.  toward health and well-being. For FY2020/21, measure used to determine how easily a com- the Indian credit rating agency ICRA estimates pany can pay interest on its outstanding debt— that the budgeted capital of Rs 20,000 crore, lingered below 1, on average. A ratio of 1.5 is along with the external equity raised of around considered healthy. The ratio is calculated by Rs 7,500 crore by a few public sector banks, dividing a company’s earnings before interest will be sufficient for public banks. Abbreviations AG Auditor General AIC Akaike information criterion BIC Bayesian information criterion CPSE central public sector enterprise CRAR capital to risk-weighted assets ratio CSO civil society organization DFI development financial institution EU European Union FB foreign commercial bank FCV fragility, conflict, and violence FY fiscal year GDP gross domestic product GFCF gross fixed capital formation ICR interest coverage ratio ICT information and communications technology IMF International Monetary Fund KPI key performance indicator LGD loss given distress M&E monitoring and evaluation MRPK marginal revenue product of capital MRPL mar­ ginal revenue product of labor MRPM marginal revenue product of material inputs MSMEs micro, small, and medium enterprises NHAI National Highways Authority of India NPA nonperforming asset NPL nonperforming loan NPV net present value OECD Organisation for Economic Co-operation and Development OLS ordinary least squares PCB privately owned commercial bank PH proportional hazards xxiii xxiv   A b b r e v i a t i o n s PITA purpose, incentives, transparency, and accountability PO proportional odds PPI Private Participation in Infrastructure PPP public-private partnership PSB public sector bank PVTB domestically owned private bank R&D research and development ROA return on assets RRB regional rural bank SBI State Bank of India SCB scheduled commercial bank SFA stock-flow adjustment SFB small finance bank SMEs small and medium enterprises SNG subnational government SOCB state-owned commercial bank SOE state-owned enterprise TFPR revenue total factor productivity UNDP United Nations Development Programme WDI World Development Indicators Overview T he recent COVID-19 (coronavirus) sheet operations by national and subna- pandemic, a once-in-a-century global tional governments. They have helped gov- shock, has sent economies in South ernments address important development Asia and the rest of the world into a deep challenges and rapidly deliver relief mea- recession. It has also cloaked future develop- sures. However, because of their inefficien- ments in a deep cloud of uncertainty. cies, they have been an important way in Governments have deployed numerous relief which public debt has accumulated. Over measures to buttress the economy and liveli- time, part of the debt generated by off–­ hoods. The indirect measures have come balance sheet operations is revealed as it hits through regulatory forbearance, while the the central government budget and debt direct ones have involved hefty social trans- stock, but at a given time, a large part fers and financial support programs. remains hidden under the radar of the exist- This big shock and deployment of public ing financial disclosure standards. resources have come on the back of the latest Because of the economic importance that global debt wave. Since 2010, emerging hidden debt carries for South Asia and ­ m arket and developing economies have beyond, this report studies the trade-offs experienced the largest, fastest, and most between addressing development challenges broad-based increase in debt in the past 50 directly through a state presence in the mar- years (Kose et al. 2020). Many of the debt kets and the risk of accumulating unsustain- increases have been pushed by the activation able debt through the economic inefficiencies of contingent liabilities—obligations of off–balance sheet operations. incurred by governments off their balance This report offers some insights regarding sheets that have triggers for payment. Such the ongoing COVID-19 crisis. Specifically, indirect (hidden) debt has been historically the crisis is likely to exacerbate problems large in South Asia. many SOCBs, SOEs, and PPPs confronted At the heart of the rising debt wave and even before the COVID-19 shock because of pandemic response have been state-owned their opaque contracts and distorted incen- commercial banks (SOCBs), state-owned tives, operational inefficiencies, and substan- enterprises (SOEs), and public-private part- dard management of risks. As a result, large nerships (PPPs) as well as other off–balance amounts of debt obligations are likely to 1 2   H IDDEN DEBT resurface for central governments. PPPs in typically de-risk through explicit or implicit South Asia and around the world are likely to guarantees. For economic reasons, govern- terminate early as the COVID-19 crisis ments aim to leverage public capital off their strains partnerships and project viability; balance sheets to maximize the development many private partners can exploit force impact of those resources. majeure clauses as expected project revenues Three prominent agents through which plunge. SOCBs and SOEs will require injec- governments leverage public capital in South tions of liquidity through debt and equity Asia are SOCBs, SOEs, and, more recently, bailouts to sustain their operations. This fis- PPPs. This leveraging of public capital cal cost may be well justified to help SOEs through off–balance sheet operations can continue to invest and SOCBs continue to happen at the level of both central and subna- lend. However, adverse side effects are likely. tional governments. As decentralization For instance, SOCBs’ positive countercyclical increases in India and Pakistan, and more lending in the short term is likely to trigger recently in Bhutan, Maldives, and Nepal, sub- capital (and labor) misallocation in the national off–­ balance sheet operations could medium term, and in turn, create conditions grow considerably. For instance, as of 2020, for inequitable and unproductive recovery. India had three times as many subnational Similar strains are likely to resurface at the SOEs than it had federal SOEs—and the level of subnational governments (SNGs) due financial performance of SOEs is much worse to triggered contingent liabilities. Central at the subnational level. governments will be obliged to come to the South Asian countries use direct inter- rescue with bailout loans and tax transfers. ventions in the markets through off–­ balance However, this may not be enough to sustain sheet operations more heavily than the SNG expenditures, including public invest- international benchmark. India and ments. As SNG expenditures shrink, local Pakistan are among the biggest users of all investments (both public and private) are three public agents considered in this report likely to contract significantly for several (SOCBs, SOEs, and PPPs). Other countries years. On the positive side, by exposing vul- stand out in using one or two of the agents, nerabilities and the urgency for reform, the such as Bhutan and Sri Lanka for SOCBs. COVID-19 crisis can also help policy makers Figure O.1 shows how the use of SOCBs, push for change in the key areas that this SOEs, and PPPs depends on the size of the report highlights. economy as measured by the real GDP (in In general, governments run operations logs). It sheds some light on whether gov- off their balance sheets for two main rea- ernments that develop public policies for sons. The first is to address market failures bigger markets tend to utilize direct tools and help create markets by encouraging more heavily—in addition to utilizing indi- (crowding in) the private sector to make rect interventions, such as regulations. investments with positive spillovers that ben- The use of SOCBs increases with the size efit the public. The second is to expand the of the economy (figure O.1, panel a). India, pool of public finance by turning direct debt Bhutan, Sri Lanka, and, to a lesser extent, obligations into a larger pool of indirect debt Bangladesh are outliers compared with the obligations that may or may not have to be international average marked by the trend met, depending on future events (contingent line. Interestingly, there is no significant code- liabilities).1 Jointly, the two approaches aim pendence between how many SOEs govern- to leverage public capital financially and ments deploy and the size of the economy economically to advance development. For (figure O.1, panel b). However, even here financial reasons, governments leverage pub- India and Pakistan stand out, exceeding the lic capital off their balance sheets to mobilize international benchmark (the trend line). greater resources from the private sector and The use of PPPs increases only marginally with foreign savings—extra resources they the size of the economy ( ­figure O.1, panel c). OVERVIEW  3 PPPs may have advantages for both smaller FIGURE O.1  Some South Asian Governments (India, Pakistan) and larger economies. De-risking may be Use State-Owned Commercial Banks, State-Owned Enterprises, needed in small m ­ arkets. Bigger markets and Public-Private Partnerships More Commonly Than the Global present more opportunities for risk pooling, Benchmark While Others (Bangladesh, Sri Lanka) Are Catching Up but also can experience bigger coordination a. Share of SOCB assets to total banking system assets, 2017–19 failures that governments may need to 80 resolve—including through de-risking. 70 Pakistan has a large share of PPPs compared India 60 Bhutan with the benchmark, while India has only 50 marginally more. However, given India’s— Percent Sri Lanka and Bangladesh’s—large pipelines of infra- 40 structure PPPs, the two countries can be 30 Bangladesh expected to increase their shares significantly 20 Pakistan in the near future. y = 4.7324x – 38.368 10 R² = 0.065 Overall, it seems that South Asian govern- ments have a strong preference for direct 0 8 9 10 11 12 13 14 intervention in the economy and markets, but Size of the economy, log of GDP in 2016 it comes with a price. The reality is that the (constant 2017 international dollars, million) efficiency of South Asian SOCBs, SOEs, and b. Number of SOEs, 2015 PPPs is well below the international bench- 400 mark, on average. Some notable examples of inefficiencies and surprise liabilities include the following: 300 Number of SOEs India •  In 2013, the Pakistan government cleared the circular debt of energy companies 200 Pakistan that stemmed from arrears between enter- prises in an attempt to clear hidden debt R2 = 0.0011 100 y = 3.32x + 27.07 once and for all. The cost was estimated Sri Lanka Bangladesh at 1.5 percent of GDP (Bova et al. Bhutan 0 2016)—but the problem continues today. In 2014, the Sri Lankan government had 10 11 12 13 14 to inject about 1.2 percent of GDP (SL Rs Size of the economy, log of GDP in 2015 (constant 2017 international dollars, million) 123 ­billion) from the budget into its stra- tegic SOEs (Government of Sri Lanka c. Share of PPPs in infrastructure projects, 2011 2014). The SOE sector then generated net 60 losses in two out of the next three years. •  Recapitalization of SOCBs has been an ongoing issue over the last two decades in most South Asian countries, including 40 Percent Afghanistan, Bangladesh, India, Nepal, Pakistan and Sri Lanka. In Bangladesh, for instance, R2 = 0.002 a single branch of Sonali Bank (an SOCB) 20 India y = 4.97 + 0.77x extended loans valued at about $454 Sri Lanka Bangladesh million based on fraudulent documents. ­ Nepal Bhutan Maldives The massive fraud led to a nonperforming 0 loan ratio of 37 percent at the SOCB in 8 9 10 11 12 13 2014. These loans have invariably Size of the economy, log of GDP in 2011 defaulted and have created a big hole in (constant 2017 international dollars, million) the bank’s capitalization (World Bank Sources: Original figures for this report. Data for panel b are from the Organisation for Economic 2020b). ­Co-operation and Development (OECD) and World Bank World Development Indicators. Note: The diagonal lines (trend lines) in each panel indicate the international average at ­varying l­evels of real GDP (in logs). In panel b, the number of SOEs for India does not include ­subnational SOEs. PPPs = public-private partnerships; SOCBs = state-owned commercial banks; SOEs = state-owned enterprises. 4   H IDDEN DEBT •  Overoptimistic bidding on PPP contracts underinvestment in externalities and natural has led to many cancellations in the PPP monopolies, but potentially at the cost of portfolio of the National Highways exposure to large financial risks and poten- Authority of India. In turn, these cancella- tial surprise liabilities. At the level of SNGs, tions have increased the level of nonper- the trade-off concerns the tension between forming assets in India’s banking sector. offering rewards for good performance and The State Bank of India, which holds the providing bailout support in bad times: that greatest nominal amount of debt related is, increasing the fiscal autonomy of SNGs to Indian highways, reported that about with good fiscal performance to boost the 20 percent of loans to ports and highways efficiency of local public spending (including were in nonperforming status by the end through subnational SOEs and PPPs), while of 2016, with the trend increasing limiting bailout support from the central throughout 2016 (ADB et al. 2018). government to exceptional cases and retract- •  At the subnational level, to resolve the ing some fiscal autonomy if a subnational problems of long-running underperfor- government systematically underperforms in mance and overindebtedness at power dis- normal times. tribution companies (subnational SOEs), These and similar tensions have led econo- Indian states unexpectedly increased their mists to formulate three complementary debt stock by about 5 percent, on average, views about the character of SOCBs, SOEs, between FY2015 and FY2018 through and PPPs and the challenges for managing the Ujwal DISCOM Assurance Yojana their performance (World Bank 2020b): (UDAY) scheme, according to Reserve Bank of India data. In 2017 alone, India’s •  Social view. Public agents (SOCBs, SOEs, state public sector enterprises—the SOEs and PPPs) are created by government to owned by subnational governments—lost address market failures and improve social an amount equal to 0.5 percent of welfare, mixing profitability goals with GDP. In FY2018, eight Indian states social objectives. These mixed objectives provided farm loan waivers amounting to ­ create challenges for monitoring outcomes 0.32 percent of GDP. and performance. •  Agency view. Because of the inability to As these experiences highlight, South monitor public agencies, an agency prob- Asian governments face a trade-off between lem emerges involving a discrepancy using PPPs, SOCBs, and SOEs to maximize between the objectives of managers (the socially beneficial investments and minimiz- agents) and owners (the principals). ing the risk of large surprise liabilities due to While governments (principals) may seek inefficiencies and mismanagement of risks. to maximize social welfare, their agents Specifically, through PPPs, the governments (and private partners) may lack the incen- try to minimize the inefficiency of project tive to maximize the use of resources execution and leverage public capital, but toward this end. potentially at the cost of assuming too much •  Political economy view. Social objectives risk—and at times allowing for moral haz- might be corrupted by politicians who ard on the part of the private sector. Through pursue their personal interests. That is, in SOCBs, the government tries to reach the some cases, the public agents can become financially underserved and finance the mechanisms for politicians to pursue their economy even when big shocks hit, but individual goals, often at the cost of eco- potentially at the cost of mismanagement nomic distortion or inequitable distribu- of risks by SOCBs and misallocation of tion of resources. capital in the economy. Through SOEs, the ­ government addresses market failures These views explain the high operational related to risky, long-term investments, or inefficiencies, reckless risk management, OVERVIEW  5 and problematic governance and political wider economic costs. The latter, through economy issues that can be the leading rea- lower economic activity and tax revenues, sons behind the financial underperformance among others, come back to weaken the of public agents (SOCBs, SOEs, and PPPs). government’s fiscal stance and increase pub- The problems can occur at the level of pub- lic indebtedness. To investigate financial lic agents, but also at the level of the politi- distress among public agents, trace the fiscal cal government (the principal of these costs and costs to the real economy of such agents). distress, and better understand the implica- For instance, SOEs are often promised tions for public policy reform, this report subsidies to run costly government has devised an analytical framework. p rograms—such as advancing access to ­ electricity to underserved populations and small enterprises—that are not received on Analytical Framework time. SOCBs are asked to run government If governments run off–balance sheet opera- p rograms—such as to advance financial ­ tions through SOCBs, SOEs, and PPPs that inclusion or lend to underserved and riskier financially underperform or are otherwise micro, small, and medium enterprises financially vulnerable, these operations are (MSMEs)—but without receiving the sub- likely to experience periodic distress. The sidy for the expected and unexpected losses SOCBs, SOEs, and PPPs will be forced to that private markets avoid. They are also adjust—including with the help of financial asked to help stimulate economies during bailouts or by curtailing their activity. In turn, downturns or financially support large PPPs these adjustments will generate adverse with concentrated risks: that is, take risk impacts on the fiscal stance by triggering con- that they cannot diversify away. They are tingent liabilities and/or adverse impacts on often asked to perform these functions in an the real economy by depriving the firms and ad hoc fashion and without prior consider- individuals of some services delivered by ation of costs and risks—for which fiscal SOCBs, SOEs, or PPPs. Figure O.2 traces transfers (subsidies, extra capitalization) these pathways through an analytical frame- must be arranged and delivered. They are work that the report adopts and follows in its often tasked with the impossible: to cross- analysis. subsidize the related losses from the profits on their commercial portfolio and Distress activities. In so doing, the government (the principal) A public agent enters distress when its finan- becomes part of the problem. Governments cial condition has worsened to the point that at the national and subnational levels origi- it cannot perform some of its common func- nate frictions that complicate effective finan- tions. For instance, an SOCB in distress can- cial management of the public agents (among not lend to its clients at the same amount as others) by being unclear about the purpose(s) before, or an SOE in distress cannot continue that the agents should serve and by being investing in new infrastructure to reach inconsistent over time when confronting underserved population. In empirical analy- political and systemic shocks. ses, the state of distress can be determined As a result of financial underperformance using a threshold for financial ratios com- due to various tensions and shocks, South puted from accounting data. In situations of Asia’s SOCBs, SOEs, and PPPs face periodic adequate financial transparency, this financial distress. At such times, they need approach is the preferred way of measuring to adjust—with more or less help from the distress—mostly because of its simplicity and ­ g overnment (the owner or sponsor). equal treatment across similar types of agent, These adjustments can inflict fiscal and such as SOCBs and SOEs. 6   H IDDEN DEBT FIGURE O.2  Analytical Framework: Links from Distress to Adjustments to Impacts Distress Adjustments Impacts • What is the probability? • What are the main • What is the effect on • What are the drivers? channels? fiscal stance? • How do public agents • What is the channels’ • How are the private differ from their relative intensity? sector and local counterparts in their • How soft are the economies affected? response? budget constraints? Source: World Bank. As an indicator of financial distress for dynamics are determined by past levels of SOCBs and SOEs, this report uses the interest SNG debt and planned fiscal balances. While coverage ratio (ICR) along with alternatives indirect, this identification by econometric that serve as robustness checks. The ICR association has been successful and validated reveals whether the revenue that the agent by some publicly recognized distress events, generates suffices to cover the interest such as the unexpected increase in the debt of ­ payments on its debt. When the ICR falls Indian states through the UDAY scheme. below 1, an SOCB or SOE is considered to be With respect to distress, the analyses in the distressed. report tackle questions such as the following: Some PPP projects could be hard to com- What is the probability of distress for a given pare with SOCBs or SOEs because they can public agent? Which factors can help predict enter distress early in their investment cycle nearing distress? How do public agents differ before they start operating, that is, perform- from their private counterparts? ing their functions. This distress could occur because the public and private partners dis- Adjustment in Times of Distress agree and engage in a dispute that cannot be resolved; in such instances, the PPP is can- A public agent becomes distressed because celed or terminated before the PPP contract its financial situation becomes unsustain- expires. As the measure of distress for PPPs, able. Something needs to change—and usu- this report uses early terminations of PPPs. ally several things. These changes concern The empirical data to capture PPP distress is the public agent’s financials and business retrieved from the Private Participation in operations. They must adjust to resolve the Infrastructure (PPI) database of the World unsustainable financial situation and exit Bank.2 distress. The adjustments could be fast— When financial transparency is impaired such as instant recapitalization of the dis- and accounting data are unavailable or unre- tressed SOCB, SOE, or PPP. Or they can be liable, econometric methods can be used to protracted—such as if the resolution determine the distress events empirically. This depends on the result of an investigation of report uses such an econometric approach to a committee tasked with deciding on behalf empirically define distress—such as triggering of the public how to proceed and what of major contingent liabilities—for SNGs. adjustments to make. Unresolved and pro- Because the financial transparency of SNGs— tracted distress may require bailouts and/or especially concerning their off–balance sheet adjustments that are ultimately more costly operations—is inadequate, the report associ- and/or severe. ates SNG distress with an unexpected increase The adjustments in times of distress can in SNG debt. Here the expected debt occur through various channels. Typically, the OVERVIEW  7 public agent uses a combination of channels public agent could seek help and time to rather than a single channel. For instance, recover under the bankruptcy protections even if recapitalization is promised, it may and/or be liquidated—including if it fulfilled not be instant. The distressed SOCB may need its purpose and there is no rationale for its to curtail its lending for some time. The report further existence. considers the following five adjustment With respect to adjustments, the analyses channels: in the report tackle questions such as the following: What are the main adjustment ­ 1. Request a bailout by injection of public channels that public agents in South Asia use equity or debt financing from the central when resolving situations of distress? How government. intensively are these different channels used? 2. Approach financial markets to raise new How soft (binding) are the budget constraints private debt or equity, or mobilize addi- that public agents in South Asia face? tional deposits. 3. Reallocate or forgo planned expenditure​ — such as postponing investment to cover Impacts unexpected expenditures or curtailing The forced adjustment in times of distress lending to meet unexpected needs for inflicts losses on the central government and/ liquidity. or the economy. Hence, the report considers 4. Sell assets to cover unexpected expenditures. and studies two main impacts of SOCB, SOE, 5. Enter bankruptcy and/or liquidation or and PPP distress. The first is the direct impact cancel the partnership and/or the project. on the fiscal stance (that is, the fiscal deficit or This list is by no means exhaustive; it public debt). The second is the impact on the focuses on the main adjustments observed in economy (that is, the macroeconomy, indus- practice. For instance, the first channel could trial activity, and/or local economic activity), include less transparent forms of bailout, such which in turn affects the fiscal stance as bailout purchases from the central govern- ­ indirectly, for example, through lower tax ment against delivery of services in the future, revenues because of lower economic activity. overinvoicing and underinvoicing of transac- We consider these two impacts at the levels tions among SOEs, and overpricing of assets of both the central and subnational purchased by central government entities. governments. Under the second channel, the extreme case If the central government or a subnational could be outright privatization—including government decides that the SOCBs, SOEs, or through conversion of private debt into pri- PPPs need to recover from distress with the vate equity and ownership. Under the third help of a bailout, its fiscal expenditures or channel, the reallocated or foregone expendi- debt will increase and its future fiscal space tures often involve maintenance expenditure will shrink. By issuing new public debt, gov- that can severely impair the quality of core ernments may discourage (crowd out) private assets and of service provision by the public sector investment by increasing borrowing agent—generating a second round of distress costs for private borrowers and/or shrinking pressure—or reduced wage expenditures also the limited pool of funding in the economy. resulting in firing of employees. Under the For instance, Huang, Pagano, and Panizza fourth channel, the sale of noncore assets (2020) find that in China, increasing local could be revitalizing because it can cleanse the public debt crowded out the investment of public agent from unfit and distracting busi- private firms by tightening their funding con- ness lines. By contrast, the sale of core assets straints. Importantly, such increases in gov- could impair the bottom line of the business ernment expenditure and debt cover only the and service provision by the public agent, accumulated losses of the public agent and including by decreasing its economies of scale are not new investments or purchases of and productivity. Under the fifth channel, the any kind. They merely help restore the 8   H IDDEN DEBT functioning of the public agent, restructure it, in distress because of a systemic risk event or liquidate it. (such as the global financial crisis or the One type of restructuring is full privatiza- COVID-19 pandemic) when most private tion, which turns the public agent into a firms in the same tier of creditworthiness are ­ private entity, essentially ceases its noncom- trying to raise funds as well. Second, this mercial social functions (that is, the public crowding-out effect can occur in a segment production of socially beneficial externali- of financial markets other than government ties), and, in principle, releases the central or securities markets—a segment in a lower subnational government from any official creditworthiness tier to which the distressed exposure to this entity. Here, two observa- SOEs and other public agents rightfully tions are warranted. First, even after full migrate. This segment—involving banks as privatization, the government may not be well as private debt and equity—also serves completely released from exposure because of small and medium enterprises (SMEs), perceived reputation risks if the privatized whose access to finance can worsen. Hence, entity gets into distress or otherwise fails soon such a crowding-out effect may be even after privatization. The government may still more detrimental for equal access to extend support to privatized firms. Second, opportunities. while forced adjustment involving liquidation The second way—by limiting activities or or privatization may impose short-term costs assets—could also be costly for the economy on the economy, it could have positive long- if public agents help create markets (through term effects if the public agents had been dis- positive supply side effects, such as spillovers torting private markets—such as by crowding in investment in research and development) out private activity, pursuing undue competi- or perform socially beneficial functions that tive advantage, or mispricing production help sustain or stimulate the demand side of inputs or final goods and services. markets—such as connecting buyers to infor- If the central government decides not to mation and communication technology (ICT) bail out (or only partially bail out) the PPPs, infrastructure and e-commerce or providing SOCBs, or SOEs in distress—including for credit after disasters. Consider the example reasons of limited fiscal space—the agents of PPPs. If PPPs are terminated early (can- must financially adjust by themselves. They celed), not only is the government likely to can do so in two broad ways: increasing their lose directly because it will have to compen- liabilities by raising equity or debt financing sate the private partner, but the project will from the private sector; or limiting their activ- not be realized or will be realized under full ities or asset growth. public ownership and operations. When the The first way can crowd out private needed project is not realized, the cost is financing available for the private sector apparent. However, even if the government because investors might prefer to allocate decides to finish the project on its own—as funds to public agents with their (perceived) the sole financier and overseer—the efficien- implicit government guarantee—even if cies that the private sector could generate by those agents are distressed. The uncompeti- managing the project implementation tive advantage on the better risk-adjusted through PPP will also be lost.3 For instance, return can cause the crowding out of private highways will be of lower quality, thereby financing even if the central (or subnational) increasing the maintenance expense, and will government itself does not issue new debt. be inefficiently operated. Another example of Of course, SOEs and other public agents costly adjustment for the economy is cur- also borrow when they are not distressed— tailed lending by SOCBs. When SOCBs in possibly inducing some crowding out distress adjust by decreasing their lending, as well. However, their borrowing in times the economy will be deprived of credit and of distress can have more adverse effects for businesses will decrease their activity (invest- two reasons. First, the public agents may be ments, production, and purchase of inputs). OVERVIEW  9 Again, the distributional effect can more adversely hit the segments that are riskier to In South Asia, off–balance sheet lend to, such as MSMEs. operations by governments are With respect to impacts, the report exam- common, but produce mixed results ines such questions as the following: What impact does the distress experienced by public and may cost societies too much. agents have on the fiscal stance? How are the private sector and local economies affected by the forced adjustment of public agents in are summarized around the topics of distress times of distress? of public agents (SOCBs, SOEs, PPPs); their adjustment in times of distress; and fiscal and economic impacts. Highlights of the main Empirical Findings findings are presented in figure O.3, in line Figure O.2 and the previous discussion frame with the report’s framework. The respective some key questions addressed by this report. chapters on SOCBs, SOEs, and PPPs as well This section highlights some answers to these as the chapter on the off–balance sheet opera- questions based on the in-depth analyses pre- tions of SNGs provide the full analysis and sented in the report’s chapters. Key findings detail. FIGURE O.3  Highlights of the Report’s Findings on Distress, Adjustments, and Impacts Distress Adjustments Impacts • About 4% to 10% of the • SOCBs and SOEs tend to get • The liabilities of chronically time, PPPs, SOCBs, and SOEs bailed out with access to distressed federal SOEs could enter distress, including at new debt and equity in account for up to 5% of GDP the SNG level. times of distress. Hence, they in India. The cost of • About 92% of PPP projects mostly keep their business recapitalizing SOCBs is survive until the end of their running as usual, while increasing in India, trending contract period. For their private firms and banks in beyond $50 billion over success, macrofinancial distress must deleverage, 2019–20.a A macrofinancial stability is fundamental, decrease activity, or curtail crisis could trigger fiscal as is contract design and lending. losses from PPPs of around structuring. • PPPs in distress can 4% of government revenue terminate early. Because PPP in Pakistan, Bangladesh, and • The main drivers are contracts are individually India. operational inefficiency; weak governance, structured, adjustments in • Successful SMEs with high institutions, and contracts; distress can be based on sales growth suffer the most and poor risk management. distorted incentives that are from impaired access to unexpected and finance when linked with • SOCBs and SOEs do not nontransparent. SOCBs. take more risk than private firms. SOCBs manage credit • SNGs hit by triggered off– • In Indian states that risk worse, and SOEs balance sheet commitments experienced triggering of overemploy. Their reduce their fiscal spending contingent liabilities, the conditions are likely worse but also receive bailouts local investment activity is at the subnational level. through tax allocations from depressed for the next four the central government. years, on average. Source: World Bank. Note: The highlights follow the analytical framework for the report presented in figure O.2. PPPs = public-private partnerships; SNGs = subnational governments; SMEs = small and medium enterprises; SOCBs = state-owned commercial banks; SOEs = state-owned enterprises. a. The Economic Times, https://economictimes.indiatimes.com/news/economy/policy/indian-banks-may-need-20-50-bn-capital-over-next-1-2-years-as-bad-loans-set-to-rise​ /articleshow/76043255.cms?from=mdr. 10   H IDDEN DEBT Distress of Public Agents more likely to experience distress than private banks, on average (controlling for bank char- Distress is not a rare event in South Asia. The acteristics). Moreover, compared with an off–balance sheet operations of government average SOCB, an SOCB with majority gov- examined by this report enter distress fre- ernment ownership share (of more than quently. There is an 8 percent likelihood that 70 percent) is about 24 percentage points PPPs in the region and beyond will terminate more likely to experience distress than a simi- early—meaning that the partnership will be lar private bank. canceled before its contract expires. In other Because SNGs are constrained from auton- words, only about 92 percent of PPP projects omous borrowing, they do not experience survive until the end of their contract period. overall distress. But they do experience sig- Railroad, treatment plant, and toll road proj- nificant shocks from triggered contingent lia- ects under PPP arrangements appear the most bilities—that is, when their own off–­ balance vulnerable to distress. sheet operations, including subnational PPPs, SOEs enter distress more often, especially SOEs, and SOCBs—go bad. A contingent lia- India’s central public sector enterprises bility shock hits a South Asian SNG about 10 (CPSEs), which are majority owned by the percent of the time (based on results for federal government. Regression analysis for Indian states). Fiscally weaker states, such as this report estimates that a CPSE is 15 per- India’s special category states,4 are shocked centage points to 21 percentage points more even more often—about 13 percent of the likely to enter distress than similar private time. firms. CPSEs in the manufacturing sector are significantly more vulnerable to distress. Pakistan has an even larger SOE sector than Adjustments in Times of Distress: India judging by the sector’s total liabilities as Bailouts versus Reduced Activity a share of GDP. Similarly, based on SOE debt as a share of GDP, Bhutan has a larger SOE How do SOCBs and SOEs adjust in times of sector than India, with Bangladesh and Sri distress compared with private banks and Lanka following not far behind India. firms? Do they raise new equity and debt to However, the lack of data prevents deeper cover unexpected losses? How big is the analyses for South Asian countries other than bailout that the central government pro- India. Nevertheless, the dearth of data and vides? How big is the adjustment on the lower transparency suggest that other South business and investment side of SOCBs and Asian countries may confront even greater SOEs? distress problems in their SOE sectors than The adjustment of SOCBs in times of dis- India—even if those problems seem well tress differs significantly from that of private hidden. banks (figure O.4). If private banks get into As for banks, our analysis indicates that in distress, they reduce lending much more than Bangladesh, Pakistan, and Sri Lanka, half of state banks in distress, which continue lending the banks—regardless of whether they are at the same or marginally higher rate—­ private or public—could have been in distress compared with healthy SOCBs or private between 2009 and 2018, that is, have an banks (compare the long blue negative interest coverage ratio (our baseline indicator “­lending” bar for distressed private banks of distress) lingering below one, on average. with the shorter positive orange bar for dis- The situation appears significantly better in tressed SOCBs). When in distress, SOCBs India. While India’s SOCBs are likely to be in enjoy softer budget constraints and readily distress 25 percent of the time, old private obtain state equity and debt support (compare banks fare progressively better and new pri- the positive orange “capital” and “debt” bars vate banks even more so. Overall, SOCBs in for distressed SOCBs with the negative blue South Asia are about 11 percentage points bars for distressed private banks). OVERVIEW  11 The softer budget constraint, as well as FIGURE O.4  State-Owned Commercial Banks Adjust Differently conditions of government recapitalization, from Private Banks in Times of Distress, 2009–18 enable SOCBs to sustain investments in times of distress. However, the soft budget con- Investment straints impose substantial fiscal costs and erode market discipline. The policy question is whether this costly insurance and risk- Lending absorption function of SOCBs pays off in terms of wider economic benefits, such as sus- tained investment by firms that bank with Debt and borrow from SOCBs. For SOEs, the report finds that distress does not restrain these public agents from Capital investing and acquiring new fixed assets to the same extent as it does private firms. As a –4 –2 0 2 4 condition of their recapitalization or other Private banks SOCBs bailouts, SOEs could also be required to Source: Original calculations for this report. expand their investment and stimulate the Note: The bars depict the t-score of the estimated adjustment coefficients. Like the z-score for the economy. Focusing on some of the shocks population average, the t-score for a regression coefficient helps illustrate the “economic” signifi- cance of the estimate by combining its magnitude and associated degree of uncertainty. That is, triggering distress, this report finds that pri- longer bars in the figure denote that the estimates are generally large and certain, while small bars vate firms confronting a negative revenue denote that the estimate is small or uncertain. For instance, the two top bars show that, on average, one can be confident to expect that SOCBs will receive a sizable capital injection (bailout) in times shock reduce investments, debt, and paid-in of distress, while private banks in distress will not be able to raise capital and will have to write off capital. Naturally, the availability of funds to some capital. SOCBs = state-owned commercial banks. finance asset growth is sensitive to revenue shocks because, for banks and investors, rev- renegotiate the PPP, the government often enue shocks are indicative of repayment needs to fully compensate the private partners capacity. However, for SOEs, the access to for the debt, equity, and foregone earnings financing—whether through equity or debt— from the failed PPP project. The government is much less sensitive to revenue shocks than can decide to form another PPP to implement for private firms. Therefore, SOEs do not the project, implement the project on its own, need to adjust to such shocks by reducing or not implement the project at all. These their activity to the same extent that private adjustments are potentially very costly both firms do. Private firms infer a negative shock fiscally and economically in terms of essential as a market signal to slow down their bor- infrastructure that the PPP projects aim to rowing and investment, but SOEs can largely build and operate and that is supposed to ignore such market signals. For this reason, reach the most vulnerable firms and during 2015–17, government support to communities. SOEs in India and Pakistan averaged between At the subnational level, contingent liabil- 1.2 percent and 1.7 percent of GDP per year ity shocks due to distressed off–balance sheet (figure O.5). operations have direct budgetary impacts. The adjustments of distressed PPPs are less They require unplanned expenditures if not clear because the contracts that uphold such adequately provisioned for; increase SNG partnerships are not standardized or transpar- indebtedness; raise their borrowing costs; and ent. Often their details and the contractual shrink their overall fiscal space. The shocks clauses that create various obligations for the necessitate a fiscal policy adjustment. This government are not disclosed or adequately adjustment can involve reducing expenditure shared—even with the public debt managers. or increasing revenue or both. SNGs may also And when PPPs are renegotiated, the fiscal receive assistance from the central govern- implications are rarely reported. When the ment, through either increased transfers or parties do not reach an agreement to loans. Our estimations reveal that after a 12   H IDDEN DEBT FIGURE O.5  Annual Government Support for South Asian State-Owned Enterprises Could Account for More Than 2 Percent of GDP, on Average, Depending on the Country, 2015–17 1.0 0.92 0.80 0.8 0.65 Percent of GDP 0.6 0.51 0.4 0.29 0.24 0.24 0.22 0.2 0.12 0.06 0.01 0 Capital Loans Grants Capital Loans Grants Capital Loans Grants Capital Loans Grants Capital Loans Grants injection and injection and injection and injection and injection and subsidies subsidies subsidies subsidies subsidies India SOEs India SPSEs Pakistan Sri Lanka Bangladesh Source: Data from government reports, averaged over 2015–17. Note: Indian state-owned enterprises (SOEs) include both central public sector enterprises (CPSEs) and state public sector enterprises (SPSEs). contingent liability shock, state governments are the distresses of public agents and their in India reduce expenditures—split approxi- adjustments? What are the costs for the real mately equally between capital and revenue economy and local economic activity? These expenditures—and increase revenue through impacts are explored next. taxes in the subsequent year. Moreover, the states receive assistance from the central gov- The Fiscal Impacts of Distress in ernment in the form of loans and increased Public Agents tax devolution. Overall, SOCBs, SOEs, and SNGs enjoy The fiscal implications of failing off–balance soft budget constraints and central govern- sheet government operations are sizable and ment bailouts after SOCBs or SOEs get into could notably reduce the fiscal space available distress or SNGs experience a financial shock to South Asian governments. For instance, the after their own off–balance sheet operations report estimates the likely fiscal costs from have failed. These soft budget constraints and PPP projects that are terminated early by sim- bailouts could be partly motivated by the gov- ulating the effect of a profound macroeco- ernment’s “guilt” stemming from unclear nomic crisis in 2020—combining a large mandates it sets for SOCBs and SOEs through depreciation of local currency with banking ad hoc requests to help with economic stimu- and debt crises. Such a profound macrofinan- lus and other political agendas. Given this cial crisis would dramatically increase the fis- pattern, accountability, hard budget con- cal costs from early termination of PPPs, straints, and fair monitoring of performance particularly in 2021. The estimated fiscal are often not possible together. The public costs over the 2020–21 period could reach agents and SNGs on the one side of the social 1.0 percent to 4.3 percent of government rev- contract, and the central government on the enues. Specifically, they could be as high as other, have settled in a bad equilibrium—one 4.3 percent of government revenues in that the economist would characterize as the Pakistan, 3.9 percent in Bangladesh, and tragedy of the commons: that is, a situation 3.7 percent in India (figure O.6). In Nepal, when a common pool of fiscal resources is early termination of PPPs could require up to overused for self-interest. How fiscally costly 3.3 percent of government revenues, while in OVERVIEW  13 FIGURE O.6  A Profound Macrofinancial Crisis Could Trigger Failures among Public-Private Partnerships That Would Cost South Asian Governments up to 4 Percent of Revenues a. Low scenario b. Medium scenario c. High scenario 5 5 5 4 4 4 3 3 3 Percent 2 2 2 1 1 1 0 0 0 2020 2021 2022 2023 2024 2020 2021 2022 2023 2024 2020 2021 2022 2023 2024 Afghanistan Bangladesh Bhutan India Nepal Pakistan Sri Lanka Source: Original figures for this report. Note: Estimated fiscal costs as a percentage of government revenues from early termination of public-private partnership portfolios assuming a profound macrofinancial crisis. If a more conservative measure is adopted The huge problem of SOE losses can and only chronic loss-makers—defined as be addressed by focusing on the SOEs that made a loss in three out of the five largest, chronic loss-makers. past years—are considered, this number remains between 8 percent and 12 percent of GDP. In Sri Lanka, the liabilities of loss- Bhutan and Sri Lanka, it could require around making SOEs have hovered between 1 percent of government revenues. These sim- 4 percent and 5 percent of GDP. If these per- ulations underestimate the effect of the crisis centages are deducted from the calculations because government revenues are kept of available fiscal space, the debt sustainabil- ­ constant—even though they would contract ity picture for South Asian countries notably during such a profound macrofinancial crisis. deteriorates. Interestingly, in every country The potential fiscal costs from distressed studied, the top 10 loss-­ m aking SOEs SOEs are even more overwhelming. account for more than 80 percent of the Unfortunately, due to the unavailability of total losses in the SOE sector. Focusing on data, the report cannot measure SOE-level these heavy loss-makers means that this huge distress based on the interest coverage ratio problem can be addressed! for South Asian countries other than India. However, it examined the total liabilities of The Economic Impacts of Distressed loss-making SOEs using the limited SOE- Public Agents level data available for Pakistan and Sri Lanka (­ figure O.7). In Pakistan, the total One key measure of economic development is liabilities of loss-making SOEs have ranged the rate at which firms and the local economy from 12 percent to 18 percent of GDP in invest. The report focuses on such a measure recent years—a remarkably high percentage. when estimating the economic impact of 14   H IDDEN DEBT FIGURE O.7  The Liabilities of Loss-Making State-Owned Enterprises in India, Pakistan, and Sri Lanka Have Been Huge, but More Than 80 Percent of Losses in Each Country Have Occurred in Only the Top 10 Loss-Makers 20 17.8 16 14.7 13.7 12.0 12.4 Percent of GDP 12 8 4.6 4.5 5.3 4.9 4.5 4.5 4.3 4.5 4 3.3 3.2 3.3 1.9 1.7 1.8 2.0 1.1 0 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 India CPSEs Pakistan Sri Lanka Source: Data from Prowess and government reports, various years. Note: Total liabilities of loss-making state-owned enterprises in India, Pakistan, and Sri Lanka (as percentage of GDP). CPSEs = central public sector enterprises. literature that shows that SOCB lending Recklessly leveraged public capital by induces credit misallocation and does not subnational governments has been very costly generally serve credit-constrained SMEs. 7 for local communities and must be remedied. Our findings therefore suggest that in epi- sodes of systemic shocks—such as the global financial crisis or the COVID-19 pandemic— frequent distress at public agents and failing when many banks experience distress simul- off–balance sheet operations of governments. taneously, private banks deleverage and For instance, the report finds that when curtail lending, while SOCBs receive capital firms start borrowing from SOCBs (as and debt support from the state to continue opposed to private banks), they systematically (or even increase) lending. This short-term, invest less. Even firms with high growth of positive stabilizing function of SOCBs for the sales and greater investment potential invest credit cycle, however, comes at the cost of sig- less when engaging with SOCBs. The implica- ­ isallocation—away from suc- nificant credit m tions of banking with SOCBs are particularly cessful firms and especially SMEs. This strong for SMEs. Using the Indian Chamber short-term positive effect makes for an of Commerce definition of SMEs, the report unequal and unfair recovery and helps unpro- finds that the SMEs that started banking with ductive, zombie firms linger in the economy SOCBs invest much less than other SMEs. in the medium term. It stalls the necessary This negative SME effect is more significant reallocation of capital (and labor) to enable than the effect of firm size or age. Even more productive investments.8 than average SMEs, SOCBs hinder successful At the subnational level, how does the dis- SMEs with high sales growth from realizing tress of off–balance sheet operations and the their investment potential. It seems that resulting contingent liability shocks affect SOCBs are particularly challenged by screen- local investments in South Asia? Such shocks ing the creditworthiness of opaque SMEs and reduce public capital expenditure and enabling their potential for investment.5,6 decrease public investment. Consequently, This finding is in line with the existing private investment that relies on the execution OVERVIEW  15 of public investment and is typically crowded FIGURE O.8  Local Investments in Indian States Fall Significantly in by public investment also declines. with a Contingent Liability Shock, Keep Dropping the Year After, Moreover, contingent liability shocks dampen and Stay Well Below the Trend for Three Years local investments indirectly: for instance, by 0.4 Percentage difference between treatment raising the tax burden and thus discouraging private investment or by reducing the viability 0.2 of investment projects, firm creditworthiness, and local lending by banks. The report con- firms that local investments in Indian states 0 and control fall significantly in the year of a contingent liability shock, continue to decline in the year −0.2 after, and remain significantly below the trend for three years after the event (figure O.8). −0.4 Interestingly, low fiscal capacity and possibly greater reliance on the c ­ rowding-in effect of −0.6 public investments can drive the adverse −5 −4 −3 −2 −1 0 1 2 3 4 5 impact of contingent liability shocks on local Years relative to distress event investments. For instance, in India’s g ­ eneral category states, local investments ­ contract Source: Original calculations for this report. Note: The figure plots decreases in subnational governments’ gross fixed capital formation only marginally after the shock. However, in ­following contingent liability shocks. The blue dots in the figure mark the value of the estimated special category states, investments contract effect (regression coefficients) of the occurrence of a contingent liability shock on gross fixed capital formation in the state (in logs). The orange line intersecting each dot marks the 80 percent by more than 60 percent in the year after a confidence interval associated with the estimated effect. The underlying regression controls for the contingent liability shock. Therefore, reck- confounding effect of business cycle shocks by using common time dummies. lessly leveraged public capital through badly designed off–balance sheet operations of SNGs is a very costly affair for local econo- that PPPs will not fail. Institutionalized mies and communities in South Asia. This checks and balances on the decision- practice must be urgently remedied by making powers of chief government execu- informed policy reforms. tives reduce PPPs’ vulnerability to expropriation by the government, such as through a change in policy or direct politi- Factors That Can Help Explain Distress, cal interference. Direct government support Inform Policy Reform, and Improve to PPPs—involving capital and revenue Outcomes subsidies as well as in-kind transfers—low- ers the probability of distress, perhaps What are some of the main factors that can thanks to more effective de-risking of the predict distress and help inform policies to underlying projects. mitigate distress of public agents in South The analysis in the report shows that Asia? PPPs executed by subnational governments •  PPPs. Larger PPP projects are more prone are less likely to face early termination than to distress (figure O.9). The exception is the PPPs with central governments. Perhaps largest projects—which could perhaps ben- local authorities understand local problems efit from more checks and balances of more better or oversee projects better because stakeholders. Distress is more likely in cer- they are nearby. By contrast, national gov- tain sectors, particularly railroads, treat- ernments may engage in riskier projects ment plants, and toll roads. Preserving because they can bear the termination risk macroeconomic ­ stability—​in particular, from an individual PPP project thanks to preventing the occurrence of local currency their more diversified PPP portfolio and devaluations and banking and debt c ­ rises— greater fiscal resources.9 In terms of con- can significantly increase the probability tract design, PPP contracts based on 16   H IDDEN DEBT FIGURE O.9  Checks and Balances on Government Executives Help Prevent Distress of Public-Private Partnerships Debt crisis Banking crisis Currency devaluation Checks & balances for executives Investment size Direct government support Water utility Treatment plant Toll roads Railroads Airports –4 –3 –2 –1 0 1 2 3 4 5 Statistical significance (t-score) Factors that reduce distress probability Factors that increase distress probability Source: Original calculations for this report. Note: The bars depict the t-scores of the estimated adjustment coefficients. Like the z-score for the population average, the t-scores for the estimated regres- sion coefficient help illustrate the “economic” significance of the estimate by combining its estimated magnitude and associated degree of uncertainty. That is, longer bars in the figure denote that the estimates are generally large and certain, while small bars denote that the estimate is small or uncertain. premium payments to government may and more vulnerable to distress. Credit create an unsound incentive structure and risk culture and management can help provoke overly optimistic bids from the explain the more frequent distress at private sponsor to win a PPP contract. SOCBs. Interestingly, SOCBs do not •  SOCBs. The extent of government owner- appear to take on more risk than private ship matters in the frequency of distress at banks. The organizational culture, possi- SOCBs. SOCBs with a government share bly from formative experiences in shel- of between 50 and 70 percent can be less tered markets, explains the patterns of prone to distress than SOCBs in which slower adoption of credit scoring technol- government has more than 70 percent ogy and inferior risk management among ownership. SOCBs can be more fragile by India’s SOCBs relative to new private design (Calomiris and Haber 2014). That banks (Mishra, Prabhala, and Rajan is, the overall governance around and at 2019). But this report’s findings also impli- SOCBs can expose them to more or cate broader governance issues and politi- greater shocks, such as directed lending, cal economy influences as important directed support of government programs, factors in shaping the structures and deci- political interference in management, sions underpinning credit risk manage- forced overemployment, and unqualified ment in SOCBs. employment (Cole 2009; Ashraf, Arshad, •  SOEs. South Asia’s SOEs do not engage in and Yan 2018; Richmond et al. 2019). inherently more risky activities than pri- The likelihood of distress increases as vate firms. For instance, India’s SOEs do bank size decreases. Therefore, smaller not have more volatile sales or profits than SOCBs with more concentrated business comparable private firms, nor are the models are the most prone to ­ distress. SOEs concentrated in sectors that have Banks—and SOCBs in particular—that are lower profit margins. So, what factors not able to intermediate the volume of explain SOE underperformance and recur- deposits they mobilize are less efficient ring distress? SOEs overemploy capital OVERVIEW  17 and labor. Controlling for size, age, and sector, the revenue-to-wage bill ratio of The negative risks of leveraging public SOEs is 85.8 log points lower, and their capital can be mitigated and the revenue-to-fixed-assets ratio is 21.5 log possible benefits enhanced through points lower than comparable private four principles: purpose, incentives, firms. Thus, SOEs earn less per unit of transparency, and accountability (PITA). labor cost and per unit of capital than their private sector comparators. This is despite the SOE’s higher debt-to-asset they permanently reduce the likelihood of ratio and financial leverage.10 It has long contingent liability shocks. In addition, been argued that SOEs underperform due financial markets do not help exert disci- to various internal management prob- pline on the states by effectively using the lems.11 Our findings support the argument disclosed information in their pricing. that SOEs are constrained from adjusting Although fiscal rules have immediate miti- labor use. Based on the idea that corpo- gating effects, these effects are short lived rate governance reforms could improve and more significant in fiscally weaker SOE performance, such reforms have states (such as the special category states emerged in South Asian countries. in India). Because contingent liability However, this report finds that a higher shocks have triggered support from the corporate governance rating for an SOE central government in the past, the states does not significantly correlate with better engage in some moral hazard by failing to SOE performance. Hence, improvements optimize their efforts to properly manage in corporate governance must be comple- the risk from contingent liabilities. mented by broader reforms in the govern-   Overall, the report finds ample evidence ing environment around SOEs. One aspect of issues related to unclear objectives of of this environment are soft loans and off–balance sheet operations, distorted implicit guarantees that distort the incen- incentives, weak transparency, and lack of tives of SOEs to monitor debt levels and monitoring or faulty monitoring. These act early to improve performance. and other issues can be addressed by the •  SNGs. At the subnational level, when off– recommendations that follow. balance sheet operations of governments go bad, they trigger contingent liabilities that shock the government fiscal stance. Policy Recommendations These shocks do not have a purely exter- The report’s findings suggest that the reform nal origin; they are induced endogenously agenda to leverage public capital responsibly as a response to political incentives. The in South Asia can be framed through four report finds that during the run-up to state principles: purpose, incentives, transparency, elections,12 SNGs assume debt from off– and accountability (PITA). balance sheet operations, such as debt of subnational SOEs, to secure jobs in the •  Purpose. The purpose of off–balance sheet short term. At the same time, SNGs delay operations and leveraging of public capital recognizing some other debt shocks until through SOCBs, SOEs, or PPPs must be after elections because the required adjust- clearly defined by the central government ments and the impact on the local econ- or subnational government as the estab- omy may cause political fallout. The lisher, owner, or sponsor. This includes contingent liability shocks can be miti- formulating a clear vision or mission, set- gated through increased transparency, ting time-bound objectives, and defining such as through the publication of debt- corresponding key performance indicators related information. Such measures take (KPIs). For example, when a government time to become effective, but once they do, council formulates the vision and mission 18   H IDDEN DEBT for SOCBs and SOEs, the government termination—and most important, deliv- entity/unit representing the state as the ers efficiency gains in the construction and owner of SOCBs and SOEs in turn can operation of the infrastructure. Fiscal rules formulate the objectives for each SOCB for SNGs must be binding and their adher- and SOE (or for each cluster by similar ence or breach reflected in the degree of purpose)—such as advancing financial fiscal autonomy the central government inclusion in rural areas or access to elec- awards the SNG. For example, on the tricity by SMEs. The government owner- back of limited transparency, SNGs have ship entity/unit can further define the used off–balance sheet operations and corresponding measurable or verifiable contingent liabilities to escape from fiscal KPIs. The KPIs can combine commercial rules. indicators (such as the return on equity) •  Transparency. Two types of transparency and development outcomes (such as accel- are needed. Debt transparency and rele- erated growth in newly opened and vant data collection are critical to enable actively used payment accounts by the both central governments and SNGs to adult population in rural areas). pull together the big picture of how •  Incentives. Institutions, rules, and con- SOCBs, SOEs, and PPPs shape the fiscal tracts must be structured in a way that space and contribute to the overall public creates proper incentives to perform in debt—including direct obligations line with the defined purpose. The opera- and explicit and implicit guarantees. Of tional costs of SOCBs, SOEs, and PPPs importance, a meaningful system for set- often exceed market costs in order to ful- ting the probabilities that guarantees are fill their purpose. The nature and extent of triggered (conversion probabilities) and these operational costs need to be deter- become direct obligations of the govern- mined and linked to the government’s ment must be developed, regulated, and budgetary and debt management frame- enforced. This will require that the works. For example, expanding connec- accounting and other back office systems tive infrastructure—energy, transport, and of SOCBs, SOEs, PPPs, and SNGs can ICT—to underserved areas may generate communicate with the central govern- very low commercial returns or losses over ment’s back office systems for debt man- the time horizon during which the typical agement. Further, South Asia should move private firm would maximize profits. The from the cash-based fiscal accounting length of the horizon over which the activ- standards toward accrual accounting to ity would become profitable, or the inabil- disclose debt and contingent liability risks ity to secure all the returns from the when they accrue, not when they material- activity, may require an ongoing budget- ize, to allow for adequate budgeting, deci- ary subsidy. This subsidy must be assigned sion making, and market response. in the budget and specified in the medium- Economic transparency is also required. term fiscal framework. The boards of It should start with publicly disclosing the directors for SOCBs and SOEs must be policy and purpose of SOCBs, SOEs, and properly staffed to deliver a skill mix to PPPs and enforcing the requirement that effectively guide the SOCBs and SOEs in each public agent publish its theory of fulfilling the twin objectives of generating change for fulfilling its objective and both profitability and developmental p urpose. Furthermore, monitoring and ­ impact—possibly over longer horizons evaluation (M&E) frameworks (central, than commercial private firms. Likewise, cluster-based, or individual) need to be PPP contracts must be structured in a way developed to inform the necessary data col- that encourages competitive and responsi- lection and to demonstrate economic ble bidding as well as fair restructuring of (development) impact. M&E processes and PPPs in distress—rather than their early outcomes should be periodically audited. OVERVIEW  19 For both the financial and economic (devel- continuing the off–balance sheet opera- opment) impact audits, the auditor general tions—such as the existence of SOCBs and of the government and the fiscal council SOEs as well as the use of PPPs for the have a crucial role to play in ensuring the right purpose and with desirable results. thoroughness and quality of these audits These actors must periodically ask and and proper functioning of the monitoring demand public answers to questions such system. as the following: South Asian countries—and many •  CSOs, for example, can question other nations—are in the early stages of whether SOCBs and SOEs expand the developing financial and debt transpar- reach of public and commercial services ency. The availability and quality of data to undeserved households and busi- on SOEs and PPPs—especially subnational nesses (MSMEs). CSOs can also ask ones—and the data quality for SNGs is whether it is time for the public agents very low in South Asia. In Pakistan, for to gradually exit some market segments example, neither the provinces nor the and give way to the private sector to Ministry of Finance publishes a time series ensure that the quality of service is of the provinces’ debt that is harmonized, improved on a commercial basis. unified, and centrally audited. The total •  Industry associations must ask whether liabilities of subnational SOEs are gener- SOCBs and SOEs can help stabilize the ally not known and could be in some cases market, set the strategic direction for even greater than those of the federal the industry to decrease investment SOEs. While the quality of financial data uncertainty, or generate positive spill- for SOCBs is slightly better than that of overs for the rest of the industry (such SOEs, the economic transparency of as through their R&D investments). SOCBs is often murkier. For the sake of Industry associations can point out that transparency, the government’s medium- market distortions, such as in funding term fiscal framework—at both the c ­ entral and pricing, as well as product and and subnational levels—should account ­ s ervice competition are becoming so for contingent liabilities from PPPs, harsh that streamlining or the exit of SOCBs, and SOEs by assessing the public state ownership from the industry is agent’s debt trajectories and their sensitiv- warranted. ity to shocks as well as keep track of likely •  Financial markets need to have enough government commitments in case of information to differentiate good per- distress. formers from bad ones among SOCBs, •  Accountability. The electorate, civil soci- SOEs, PPPs, and SNGs, for example, by ety organizations (CSOs), industry associ- pricing the debt of worse performers ations, media, and financial markets must higher than that of good performers. take action to support reform that imple- Along with necessary transparency and ments the PIT principles so that off–bal- disclosure, various other steps could ance sheet operations of governments help, including developing markets for cannot be used for political self-interest project bonds to ensure pooled, local- (such as increasing reelection prospects) or currency funding and market monitor- side deals (“I’ll scratch your back if you ing of PPPs; requiring listing on stock scratch mine”)—or at least make it harder exchanges and public trading of the to do so. Once the reforms are imple- debt and equity of the agents; issuing of mented, the electorate, CSOs, industry debt (and bail-in instruments) by associations, and financial markets must SOCBs and SOEs; and improving mar- remain vigilant and active. The actors kets for subnational bonds to price the need to keep testing the justifications for risk of SNG financial performance. 20   H IDDEN DEBT TABLE O.1 Implementing the High-Level Policy Recommendations for Public-Private Partnerships, State-Owned Commercial Banks, State-Owned Enterprises, and Subnational Governments The PITA State-owned principles Public-private partnerships commercial banks State-owned enterprises Subnational governments Purpose To create efficiency in public To help create markets To help create markets To expand the local efficiency of projects through well- for financial services and provide an alternative SNG operations and help create incentivized private sector by addressing market to ineffective regulation local infrastructure, markets, and participation. Should not be failures. Typically of natural monopolies in public services using off–balance used primarily to expand fiscal combine social and some sectors. Typically sheet operations at the SNG level. space (public funding) because commercial objectives. combine social and Experience suggests caution in infrastructure PPPs are ultimately Purely commercial commercial objectives. expanding these operations rapidly funded through tax revenues or SOCBs could be used Purely commercial SOEs at the SNG level. user fees, which could have been to expand government could be used to expand collected by the government if capacity to generate government capacity to the infrastructure were publicly revenue. generate revenue. provided. Incentives Improve de-risking of projects Establish fiscal provisions Include fiscal provisions in Make the ability to run sizable off– and risk sharing between to cover the above- medium-term expenditure balance sheet operations through the government and private market operating costs and debt management PPPs, SOCBs, and SOEs an earned sector. Risk must be addressed and risk taking needed frameworks to cover the privileged that responsible SNGs by the government and to pursue legitimate above-market operating obtain with greater autonomy. So efficiently assigned between objectives. Avoid using costs needed to pursue far, the incentive for off–balance the public and private partners, commercial operations legitimate objectives. sheet operations has largely been not simply passed on to the to cross-subsidize Ensure that the distribution the escape from subnational fiscal private partner. Establish social functions. Couple of transfers is timely to rules. Consider empirical evidence checks and balances on assurances that fiscal keep incentives aligned. that subnational PPP projects have the powers of executives to transfers will cover Couple assurances that been more successful—thanks to mitigate expropriation risks legitimately higher fiscal transfers will cover more efficient local supervision— and corruption and strengthen losses with binding legitimately higher losses than subnational SOCBs and SOEs governance around PPPs. rules and hard budget with binding rules and because of their more concentrated Ensure that the contract design constraints. Ensure hard budget constraints. geographic and industry risks, encourages competitive but proper supervision by an as well as weaker governance of responsible bidding. independent regulator. subnational SOCBs and SOEs. Transparency South Asian governments should move from cash-based fiscal accounting standards toward accrual accounting to disclose debt and contingent liability risks when those risks accrue, not when they materialize, to allow for adequate budgeting, decision making, and market response. Disclose and gradually Publicly disclose SOCB Better assess and monitor Collect and consolidate information standardize contracts for PPPs. lending to and funding the fiscal risks from SOEs. on debt and other contingent Link all contingent liabilities from SOEs, together Incorporate them into obligations through a single entity from PPP contracts to medium- with the policy/directed fiscal planning and debt at the subnational level, such as a term expenditure and debt lending share of the management frameworks. specialized debt management unit management frameworks. SOCB loan portfolio, Improve collection of within the Finance Department. Publicly disclose the entire within the audited financial data of subnational Further consolidate data at the compensation of private financial statements SOEs so that, for example, central government level to disclose partners in case of success or of SOCBs. Shift SOE the total liabilities of all the big picture. Audit, analyze, and failure. Publish private bids investment borrowing SOEs are disclosed. Ensure publicize the data on consolidated to enable monitoring by the from SOCBs to capital adequate provisions to debt and contingent obligations public and competitors to markets through meet each contingent through an independent national enforce bidding that is both the issuance of SOE liability, and all contingent agency, such as the fiscal council, to competitive and responsible. corporate bonds or liabilities, without disrupting ensure the consistency and accuracy government bonds. public spending plans. of the data. Accountability The electorate, civil society organizations, industry associations, media, and financial markets must take action to support reform that implements the PIT (purpose-incentives-transparency) principles so that off–balance sheet operations of governments cannot be used for political self-interest or side deals—or at least make it harder to do so. Once the reforms are implemented, all these actors must remain vigilant and active and keep testing the justifications for continuing off–balance sheet operations. Source: World Bank. Note: PITA = purpose, incentives, transparency, accountability; PPPs = public-private partnerships; SNGs = subnational governments; SOCBs = state-owned commercial banks; SOEs = state-owned enterprises. OVERVIEW  21 In closing, while public policy must lead, it overall or their willingness to lend to SMEs takes a concerted effort by society to ensure only for working capital needs. Future that the off–balance sheet operations of gov- research could examine this. ernment make sense and responsibly leverage 6. Anecdotal evidence suggests that SOCBs focus more on meeting lending quotas for the public capital for the sake of more rapid and volume of extended credit than they focus on more equitable development. Table O.1 sum- the quality of project screening. These quotas marizes the high-level policy recommenda- are more easily met by serving larger firms— tions discussed in this report and organizes including SOEs implicitly backed by a gov- them into a matrix with the PITA principles ernment guarantee—than opaque and risker in the rows and the types of public agents in SMEs. Therefore, the combination of more the columns. The report chapters discuss frequent distress with the inability to take these recommendations in detail. informed risks and manage them makes SOCB operations problematic for private sector development (Mishra, Prabhala, and Notes Rajan 2019). 1. Note that contingent liabilities can originate 7. For example, studies for India (Cole 2009), from many sources, such as potential bailouts Pakistan (Khwaja and Mian 2005), and of systemically important banks, unexpected Brazil (Carvalho 2014) show that SOCBs costs from litigation against the government, induce significant credit misallocation in the natural disasters, and schemes that the govern- economy. Besides being more politicized and ment may run (pensions and health insurance inefficient, the lending of SOCBs may not obligations or social transfers in recessions). reach more credit-constrained economic For instance, Bova and others (2016) estimate agents such as SMEs (see Berger et al. 2008; that the largest average fiscal cost of contingent Ongena and Sendeniz-Yüncü 2011). liability realizations for 80 countries sampled 8. However, this is not to deny the successes of came from financial sector support and bank SOCBs in mobilizing deposits, advancing bailouts in financial crises, followed by the financial inclusion in digital payments, or unexpected litigation costs. This report focuses facilitating relief after disasters (World Bank on a narrower set of potential contingent liabil- 2020a). ities related to PPPs, SOEs, and SOCBs, includ- 9. Highway projects in India provide an inter- ing at the subnational government level. esting example: All the highway projects that 2. See the PPI Data page at https://ppi​ were canceled between 2012 and 2015 were .worldbank.org/en/ppidata. PPPs with the central government. At the 3. When the government finishes the project on same time, however, state governments con- its own as the sole financier and overseer, it tinued to form successful PPPs for road con- may either use SOEs or contract private firms struction and operation. to build, operate, and/or maintain the infra- 10. Interestingly, CPSEs compensate for the over- structure. Because the government is unable use of labor by “underusing” other inputs. to manage the project implementation as effi- For example, SOEs could be using more ciently as the private sector, a large part of manual processes that consume less power. the implementation efficiencies will be lost 11. It is harder to align the incentives of manage- even if the government contracts private ment and owners in the public sector (Ehrlich firms to build, operate, and/or maintain the et al. 1994). The compensation of managers infrastructure. is weakly linked to the SOEs’ market perfor- 4. Special category status is a classification mance (Borisova, Salas, and Zagorchev given by India’s central government to assist 2019), and SOE managers are prevented in the development of states that confront from making optimal choices, for example, geographical and socioeconomic disadvan- because of a government mandate leading to tages, such as hilly terrains, strategic interna- excessive hiring (Shleifer and Vishny 1994). tional borders, economic and infrastructural 12. We focus on the state legislative assembly backwardness, and nonviable state finances. (Vidhan Sabha) elections, which largely 5. It is not clear whether this finding might be determine the state-level governments, which due to SOCBs not lending enough to SMEs hold fiscal authority. 22   H IDDEN DEBT References Huang, Y., M. Pagano, and U. Panizza. 2020. “Local Crowding-Out in China.” Journal of ADB, DFID, JICA, and World Bank (Asian Finance 75 (6): 2855–98. Development Bank, UK Department for Khwaja, A. I., and A. Mian. 2005. “Do Lenders International Development, Japan International Favor Politically Connected Firms? Rent Cooperation Agency, and the World Bank). Provision in an Emerging Financial Market.” 2018. The WEB of Transport Corridors in Quarterly Journal of Economics 120 (4): South Asia. Washington, DC: World Bank. 1371–411. Ashraf, B. N., S. Arshad, and L. Yan. 2018. “Do Kose, A. M., P. Nagle, F. Ohnsorge, and N. Better Political Institutions Help in Reducing Sugawara. 2020. Global Waves of Debt: Causes Political Pressure on State-Owned Banks? and C ons e que nc e s . Was hington, DC : Evidence from Developing Countries.” Journal World Bank. of Risk and Financial Management 11 (3): 1–18. Mishra, P., N. Prabhala, and R. G. Rajan. 2019. Berger, A, L. Klapper, M. Martinez Peria, and “The Relationship Dilemma: Organizational R. Zaidi. 2008. “Bank Ownership Type and Culture and the Adoption of Credit Scoring Banking Relationships.” Journal of Financial Technology in Indian Banking.” Johns Intermediaries 17 (1): 37–62. Hopkins Carey Business School Research Borisova, G., J. M. Salas, and A. Zagorchev. 2019. Paper no. 19-03, Johns Hopkins University, “CEO Compensation and Government Baltimore, MD. Ownership.” Corporate Governance: An Ongena, S., and I. Sendeniz-Yüncü. 2011. “Which International Review 27 (2): 120–43. Firms Engage Small, Foreign, or State Banks? Bova, E., M. Ruiz-Arranz, F. Toscani, and H. Elif And Who Goes Islamic? Evidence from Ture. 2016. “The Fiscal Costs of Contingent Turkey.” Journal of Banking & Finance Liabilities: A New Dataset.” IMF Working 35 (12): 3213–24. Paper WP/16/14, International Monetary Fund, Richmond, C. J., D. Benedek, E. Cabezon, Washington, DC. B. Cegar, P. A. Dohlman, M. Hassine, B. Jajko, Calomiris, C. W., and S. H. Haber. 2014. Fragile P. Kopyrski, M. Markevych, J. A. Miniane, by Design: The Political Origins of Banking F. J. Parodi, G. Pula, J. Roaf, M. Song, Crises and Scarce Credit . Princeton, NJ: M. Sviderskaya, R. Turk Ariss, and S. Weber. Princeton University Press. 2019. “Reassessing the Role of State-Owned Carvalho, D. 2014. “The Real Effects of Enterprises in Central, Eastern and Southeastern Government-Owned Banks: Evidence from an Europe.” IMF Departmental Paper 19/11, Emerging Market.” Journal of Finance 69 European Department, International Monetary (2, April): 577–609. Fund, Washington, DC. Cole, S. 2009. “Fixing Market Failures or Fixing Shleifer, A., and R. W. Vishny. 1994. “Politicians Elections? Agricultural Credit in India.” and Firms.” Quarterly Journal of Economics American Economic Journal: Applied 109 (4): 995–1025. Economics 1 (1): 219–50. World Bank. 2020a. South Asia Economic Focus, Ehrlich, I., G. Gallais-Hamonno, Z. Liu, and Spring 2020: The Cursed Blessing of Public R. Lutter. 1994. “Productivity Growth and Banks. Washington, DC: World Bank. Firm Ownership: An Analytical and Empirical World Bank. 2020b. State Your Business! An Investigation.” Journal of Political Economy Evaluation of World Bank Group Support to 102 (October): 1006–38. the Reform of State-Owned Enterprises, Government of Sri Lanka. 2014. Department of FY08–18 . Washington, DC: World Bank, Public Enterprises–Performance Report 2014. Independent Evaluation Group. Department of Public Enterprises. Public-Private Partnerships in South Asia: Managing the Fiscal 1 Risks from Hidden Liabilities While Delivering Efficiency Gains S The Need to Carefully Manage the ince the early 1990s, public-private partnerships (PPPs) to provide infra- Fiscal and Economic Risks of PPPs structure have been expanding around the world, including in South Asia. Well- Worldwide, nearly 1 billion people lack structured PPPs can unleash efficiency gains electricity, 1 billion live more than 2 kilometers in the provision of infrastructure, but PPPs from an all-season road, and many are unable can also create liabilities for governments, to access work and educational opportunities among them contingent liabilities: that because transport services are not available or is, liabilities triggered by a specific event. are too costly. In South Asia, estimates of the Providing infrastructure through PPPs is annual investment needs to close the infra- preferred to public provision if the effi- structure gap range from 7.5 percent of GDP ciency gains offset the higher cost of private (Rozenberg and Fay 2019) to 8.8 percent of financing and the public liabilities that PPPs GDP (ADB 2017). To meet these investment may create. This chapter assesses the fiscal needs, infrastructure spending will have to risks from contingent liabilities assumed by increase by 3.5 percent to 4.3 percent of South Asian governments through their cur- GDP from its current level.1 rent stock of PPPs in infrastructure. First, it Different approaches can be used to analyzes the drivers of PPP distress. Second, provide infrastructure services. In the tradi- it simulates scenarios of possible fiscal costs tional, public provision approach, line minis- for South Asian governments that could tries, government agencies, or state-owned stem from risky PPPs. Third, it studies spe- enterprises directly procure the infrastructure. cific PPP contract designs and their rela- In the private provision approach, regulated tionship to early termination in South Asia or unregulated private companies that own to draw lessons for structuring future PPP the infrastructure assets provide infrastruc- contracts. ture. Infrastructure provision through PPPs Note: This chapter draws on the background research paper: Herrera Dappe, M., M. Melecky, and B. Turkgulu. 2020. “PPP Distress and Fiscal Contingent Liabilities in South Asia.” Background paper for Hidden Debt. World Bank, Washington, DC. 23 24   H IDDEN DEBT falls in between the public and private approach. In a PPP, the private party controls It is important not to overestimate the rights to the infrastructure during the con- the efficiency gains or underestimate tract term and returns the infrastructure to the risks and liabilities of a PPP. the government when the contract term expires. PPPs can help emerging market economies opacity of financial records, confidentiality of and developing countries expand their infra- most PPP contracts, and prevalence of cash structure stock, build required infrastructure rather than accrual accounting systems in more efficiently, and maintain infrastructure emerging markets and developing economies better in the long term. The potential e ­ fficiency hide the fiscal risks for government finances gains can be seized through an appropriate until the contingent liability materializes. design of PPP contracts that bundle various This chapter assesses the fiscal risks aspects of the infrastructure project and allo- South Asian governments assume when an cate risks according to the partners’ ability to infrastructure PPP is terminated early. There manage them. Economies such as Brazil, are three major reasons for early termination of China, India, South Africa, and Turkey have PPPs: the government’s default or voluntary used PPP arrangements extensively to boost termination of the project; the private partner’s their infrastructure investments. default or breach of contract; or force majeure Infrastructure PPPs are no free lunch. They (unforeseen circumstances). For the assess- create liabilities for governments, including ment, the study adopts the value-at-risk meth- contingent (hidden) ones. To share risk appro- odology (see annex 1A). The expected loss priately between the public and private parties, from a PPP project is gauged using the proba- governments tend to provide explicit guaran- bility of distress, exposure of the government in tees to the private party, such as revenue or the event of distress, and the loss given distress. credit guarantees. The government, as the ulti- Using data from the World Bank Private mate guarantor of the public infrastructure Participation in Infrastructure (PPI) database, service, also provides an implicit guarantee to World Bank World Development Indicators backstop the fiscal and economic conse- (WDI), the Polity IV Project, and the banking quences of any failures by the partnership. crises data set of Laeven and Valencia (2018), At the center of the PPP approach rests a the study identifies systematic contractual, trade-off between the efficiency of execution institutional, and macroeconomic factors that and the efficiency of financing. The private can help predict the probability that a PPP partners bring the efficiency of execution project will be terminated early (see annex 1B). because they can better monitor the project. Factors that contribute to the early termi- However, governments can achieve a greater nation of PPPs. The analysis finds that PPPs efficiency of financing because they can finance in developing countries have a lower proba- a project at cheaper (sovereign) rates than the bility of early termination when they are con- private partners in PPPs, which need to pay a tracted by subnational entities. Direct support funding premium on top of the sovereign rate from the government—whether capital to cover extra risk.2 If the efficiencies in project grants, revenue subsidies, and/or in-kind execution are systematically overestimated or transfers—decreases the financing risk of the the contingent liabilities due to risk and uncer- project. The probability of early termination tainty are underestimated, the government is also lower for PPP contracts in countries may be better off executing investments with greater constraints on executive power through conventional contracting of the pri- in government. Large physical investments vate sector or even state-owned enterprises. and macrofinancial shocks increase the likeli- The rising popularity of PPPs, and thus the hood of early terminations. In particular, increase in the contingent liabilities associated unexpected currency depreciations, and with them, warrant careful management of incidences of debt and systematic banking cri- the fiscal and economic risks they pose. The ses, increase the rate of project cancellation. P UBLI C - P RIV A TE P A RTNERS H I P S IN SOUT H A SI A    25 The analysis also derives lessons on ways and responsibilities associated with different to structure contracts for better results by stages of the project to maximize the value for examining the PPP highway sector in India, money. Outsourcing of responsibilities to the for which rich data are available. PPPs for private sector and bundling of investment and national highways are more likely to termi- service provision can yield efficiency gains. nate early if, through their contractual obli- Outsourcing allows the public sector to lever- gations, they put the private sponsor under age private sector expertise and gain organi- larger financial commitments—namely, zational efficiency in service provision. The higher payments to the government and a potential of knowledge and technology spill- larger share of debt financing. The ­ pattern of overs from foreign sponsors may be better higher payments to the government might harnessed by the host country within a PPP reflect a perverse incentive structure that relationship (ITF 2018). Competitive pro- encourages private partners to make overop- curement to select the private partner can also timistic bids on payments to the government drive the cost down compared to in-house in order to win tenders on PPP contracts, public sector provision. which have been financed largely by public PPPs bundle investment and service banks. provision—that is, financing, design, con- struction, rehabilitation, operation, and maintenance—into a single long-term con- Balancing the Efficiency Gains tract. This contrasts with traditional procure- from PPPs against Their Risks ment practices, in which the government and Liabilities separates the contractual relationships for A PPP is an organizational arrangement that each phase of the infrastructure investment enables public and private institutions to and operation. The idea behind bundling is to cooperate in providing a public project— combine the two major stages of a typical which in the context of this chapter is an infrastructure project (investment and service infrastructure project. As Grimsey and Lewis provision) to achieve efficiency gains. When (2017) point out, a PPP is an enduring and these two stages are bundled, the private relational partnership, with each partner party has the incentive to adopt improve- bringing something of value (money, prop- ments during the design and construction erty, authority, reputation) to the partner- stages that reduce operation and maintenance ship.3 A key defining feature of a PPP is the costs or increase the quality of services and sharing of responsibilities and risks of out- revenues during operation, as long as the comes between the partners. Underpinning additional construction costs are offset by the partnership is a framework contract that higher returns in the latter stage.4 sets out the “rules of the game” delineating Efficiency gains from PPPs may also arise each partner’s rights and obligations. Because from mobilizing private finance. Private of uncertainties inherent in long-term proj- finance may provide outside expertise in valu- ects, PPP contracts are incomplete: that is, ing risks and monitoring effort that public they do not cover all possible scenarios, and finance lacks. Hence, when private creditors they leave room for renegotiation (Guasch that are specialized in project finance are 2004). involved in financing, private finance may PPPs can offer numerous efficiency gains. resolve uncertainty and agency problems faced Well-structured PPPs have the potential to by the government (Iossa and Martimort 2012, provide infrastructure services at a relatively 2015).5 By providing incentives for efficient low cost to society. This can increase a coun- termination, private finance may also resolve try’s capacity to invest in infrastructure, given soft budget constraints (privileged access to that some investments would potentially be additional financing due to the implicit guaran- feasible only under a PPP arrangement tee of unconditional government support) (Iossa and Martimort 2012). PPPs aim to effi- through which governments can keep bad ciently allocate among the partners the risks projects alive (de Bettignies and Ross 2009).6 26   H IDDEN DEBT The funding structure in a PPP affects the Private finance still poses a fiscal burden allocation of risks and costs and the quality on government. Sometimes policy makers of service provision. Demand risk—the risk and development practitioners claim that a that demand for the infrastructure services benefit of mobilizing private finance is that will fall short of a forecast and hence so will it allows governments to invest in infra- revenues—is a major risk of infrastructure ­ structure when the government has no fiscal PPPs. The allocation of demand risk affects space (budgetary room that allows a the financing cost of the project and the oper- government to provide resources for public ator’s incentives to ensure adequate service purposes without undermining fiscal sus- under conditions of imperfect monitoring and tainability). This argument is based on con- regulation. The funding structure determines fusion between funding and financing. the allocation of demand risk between the Private financing, by itself, does not reduce government and the private sponsor the fiscal burden on the government because (operator). A contract funded purely by user either through future availability payments fees collected by the operator allocates the or foregone user fees, the government ends entire demand risk on the private sponsor. In up directly or indirectly funding the provi- contrast, the government assumes the demand sion of the infrastructure service over the risk under an availability payment scheme lifetime of the project (Grout 1997; Hart (committing to provide a fixed payment to 2003; Engel, Fischer, and Galetovic 2013). the private provider/operator as long as the It could be argued that in a developing performance of the project meets agreed per- economy when the government is facing formance metrics). The operator in a user fee temporary credit constraints but must scheme would face higher costs of capital invest in critical infrastructure needed right than in an availability payment scheme away, mobilizing private financing could be because creditors would require compensa- beneficial—if international private sponsors tion for the extra risk. The operator in the with well-diversified portfolios and good former case would have an incentive to attract credit ratings can obtain financing at a more users through better services to increase low cost (de Bettignies and Ross 2010; its revenues, while in the latter case the public Yehoue 2013). partner would have to monitor the quality of A better framework is needed for service to ensure a certain level of service. valuating and reporting the liabilities cre- Intermediate arrangements—whereby the ated by PPPs. PPPs can create liabilities for government guarantees a minimum revenue governments based on how risks are shared from user fees or provides some availability with the private partner. A good way to cat- payments and allows the operator to charge egorize these liabilities is to use a fiscal risk reduced user fees—are common. matrix, which categorizes the government’s PPPs have the potential to weed out a bad liabilities as direct or contingent and project when it is funded by user fees and the explicit or implicit (Polackova 1998; government commits not to fund it through Budina, Polackova Brixi, and Irwin 2007). public sources (tax revenues). When the Direct explicit liabilities created by PPPs demand risk is effectively transferred to the are contractual or legal promises by the private party, the project will only attract pri- government in the event that all stages of vate sponsors and external creditors if the the project go according to the schedule project is financially profitable. This means foreseen in the contract. Availability and that bridges to nowhere would not be built capacity payments, shadow tolls, and under a PPP arrangement funded by user fees. energy payments in power purchasing However, this market test is less useful than it agreements where the public party has no appears. It fails to indicate whether projects, control over energy generated 7 are exam- even if they are unprofitable, yield benefits to ples of direct explicit liabilities for the society. government.8 P UBLI C - P RIV A TE P A RTNERS H I P S IN SOUT H A SI A    27 Contingent liabilities can be explicit or burdens that could not be foreseen at the implicit. Explicit contingent liabilities created beginning of the contract. If the renegotiation by PPPs are the contractual or legal guaran- process fails, the contract may be terminated tees by the government contingent on the as a last resort, and the government may be occurrence of an exogeneous event. For left to settle the claims of the creditors and the example, the government may commit to a private sponsors at a rate beyond what is minimum revenue guarantee for a toll road or specified in the termination clauses. for an independent power producer in the The current public sector accounting PPP contract. Implicit contingent liabilities principles do not provide an adequate frame- created by PPPs are the noncontractual work for valuating and reporting the liabilities liabilities that the partnership and incomplete created by PPPs. This uncertainty exacerbates contracts create in various states of the world. the fiscal risks from PPPs (see box 1.1). Cash- For example, even though the government based accounting practices, which are still might not contractually promise any guaran- popular in many developing countries, do not tees to the private party in the event of a provide a way to include the liabilities in gov- default, given that the government is the ulti- ernment finances. Even the financial account- mate guarantor of public services in most ing frameworks recommended by the societies, the government might have to bail International Monetary Fund (IMF) and the out the private party or assume the remaining European Union (EU) limit their inclusion debt and service obligations of the private based on assessment of the risks and control party to avoid service disruption. This means borne by the government (Heald and Georgiou that when a PPP contract is agreed upon, the 2010; de Vries 2013). Furthermore, govern- government assumes the ultimate insolvency ments facing fiscal constraints tend to increase risk (Irwin 2007). The most common implicit their levels of PPP investments as a percentage contingent liabilities of PPPs stem from of GDP without instituting proper institu- renegotiation and early termination. tional mechanisms to deal with the liabilities Renegotiations can lead to additional fiscal they create (Reyes-Tagle and Garbacik 2016). BOX 1.1  The Hidden Debt of National Highways in India The annual funding needs of the National support through budgetary transfers. In short, Highways Authority of India (NHAI) are the Ministry of Finance covers full debt service approved by the Ministry of Road Transport and and payment obligations throughout the govern- Highways and the Ministry of Finance through ment’s public-private partnerships (PPPs), but the Union Budget. This includes a contribution such debt and liabilities are not explicitly stated from the Central Road and Infrastructure Fund on the government’s balance sheet—and thus (CRIF) and approval for market borrowings become part of the “hidden debt.” This arrange- (with implicit or explicit sovereign guarantee). ment differs from other state-owned enterprises However, debt service for market borrowings (SOEs), which are corporatized and have inde- and the payment obligations under the hybrid pendent balance sheets. annuity model are also part of the annual funding Source: World Bank staff, based on inputs from World Bank experts. needs of NHAI. Thus, NHAI, as a national Note: Under the hybrid annuity model, the project company is entitled to authority, raises debt and undertakes long-term receive from the NHAI both semi-annual availability payments during the liabilities (PPP hybrid annuity model payment operation of the road and a capital grant during the construction phase. obligations) not on the strength of its financials, The project company is responsible for the construction and the mainte- but on the assurance of government of India nance of the road, but it is the authority’s responsibility to collect tolls. 28   H IDDEN DEBT Booming Infrastructure PPPs, Their increased the number of projects as well as Country and Sector Distribution, their average size. and Signs of Distress in South Asia India has the most infrastructure PPPs in South Asia, by far. India accounted for more PPPs have grown rapidly in South Asia. The than three-quarters of the 1,232 infrastruc- use of PPPs in infrastructure in the region ture PPPs in South Asia. Pakistan and grew exponentially from the early 1990s to Sri Lanka implemented 81 and 79 PPPs, the early 2010s, but has slowed down since respectively, followed by Bangladesh with 45 2012. PPP investments accelerated between and Nepal with 38 during 1990–2018. 2005 and 2012, increasing the value of the Afghanistan and Bhutan implemented only active portfolio by more than five-fold, from 2, while Maldives had 1 PPP in infrastruc- $45 billion to $267 billion, or from ture. In terms of value, PPP investments in 3.9 percent to 11.4 percent of the region’s India—amounting to $283 billion—account GDP (figure 1.1, panel a). After 2012, invest- for more than 85 percent of the $328 billion ment growth slowed, falling behind GDP in aggregate investment in PPPs in the region. growth, as indicated by the decline in invest- India is followed by Pakistan and Bangladesh ments as percentage of GDP observed in the with total investments of $31 billion and same panel. At the end of 2018, cumulative $6.9 billion, respectively. Sri Lanka’s PPP investment in the active portfolio of PPP proj- program had investments of $3.4 billion and ects was just over $320 billion,9 according to Nepal’s had $2.9 billion. The PPP programs the World Bank PPI Database, 10 which of Maldives, Bhutan, and Afghanistan have corresponds to 8.9 percent of the region’s been the smallest, with investments of $469 total GDP. million, $240 million, and $39 million, The number of active projects in South respectively (figure 1.2, panel a). Asia increased exponentially between the Most PPPs in South Asia are in the energy early 1990s and early 2010s, but slowed and transport sectors. Of the 1,232 PPP proj- down after 2012 (figure 1.1, panel b). The ects in infrastructure with financial closure in increase in the number of projects was the region since 1990,11 97 percent are in the slower between 2005 and 2012 than the energy or transport sectors (figure 1.2, increase in investment volumes, which indi- panel a). The remaining 38 projects are in cates that the countries in the region the water and sewerage sector and the FIGURE 1.1  Active Portfolio of Public-Private Partnerships in Infrastructure in South Asia, 1990–2018 a. Cumulative investment b. Number of projects 350 12 1,400 300 10 1,200 2019 US$, billion 250 1,000 8 Percent 200 800 6 150 600 4 100 400 50 2 200 0 0 0 1990 1994 1998 2002 2006 2010 2014 2018 1990 1994 1998 2002 2006 2010 2014 2018 In monetary terms (left axis) As percentage of GDP (right axis) Sources: Private Participation in Infrastructure database; World Development Indicators. Source: Private Participation in Infrastructure database. Note: Dashed lines indicate the start (2005) and the end (2012) of the boom period of Note: Dashed lines indicate the start (2005) and end (2012) of the boom PPP ­investments. period of PPP investments. P UBLI C - P RIV A TE P A RTNERS H I P S IN SOUT H A SI A    29 FIGURE 1.2  Sectoral Distribution of Public-Private Partnership Projects with Financial Closure in South Asia, by Country and Number of Cancellations, 1990–2018 a. Sectoral distribution of PPP projects b. Number of cancellations 2 projects Afghanistan Afghanistan $39 million 45 projects Bangladesh 3 Bangladesh $6.9 million 2 projects Bhutan Bhutan $240 million 984 projects India 31 India $282.9 billion 1 project Maldives 1 Maldives $469 million 38 projects Nepal Nepal $2.9 billion 81 projects Pakistan Pakistan $31.2 billion 79 projects Sri Lanka Sri Lanka $3.4 billion 1,232 projects Total Total $328 billion 35 0 10 20 30 40 50 60 70 80 90 100 0 20 40 Percent Energy Transport Water and sewerage ICT Source: Private Participation in Infrastructure database. Note: Investment totals are in 2019 US$, million. ICT = information and communications technology; PPP = public-private partnership. See also annex 1C. information and communications technology India, there are 543 PPPs in the transport (ICT) sector. The composition by country sector and 414 in the energy sector. shows that aside from India, Bhutan, and While the share of cancellations in Maldives, the country programs are domi- South Asia is low and similar to the global nated by investments in the energy sector. In share, the transport sector in India has a 30   H IDDEN DEBT disproportionate number of cancellations The distressed projects in India also coin- (figure 1.2, panel b). Only 2.8 percent of the cide with the slowdown in PPP investments in infrastructure PPPs initiated in South Asia South Asia, as can be seen in figures 1.2 and have been canceled. This share is similar to 1.3. Although it can be argued that it is coin- the global share, which is 3.7 percent. The cidental, there is evidence that the cancella- cancellations in South Asia have mostly tions led to more caution in initiating new been in the Indian transport sector, particu- PPPs. For example, a federal statute that larly national highways, which accounts for came into effect in 2014 set the requirement 27 of the 35 canceled PPPs in South Asia that a PPP seeking to acquire land must (figure 1.2, panel b, and table 1C.1). Even obtain the consent of at least 70 percent of though the highest number of cancellations the affected persons and made compensation occurred in the transport sector, the sector more generous—and thus more costly for the with the largest share of canceled PPPs is the project. Earlier projects were undertaken ICT sector, with 19 percent. with as little as 30 percent of the land having The PPP national highways program in been secured (Pratap and Chakrabarti 2017). India showed signs of distress in 2013 and Furthermore, after 2010 in India, traditional 2014. Twenty-four of the 27 cancellations in procurement of infrastructure became the Indian national highways sector occurred the preferred choice at an increasing rate in 2013 and 2014. At that time, there were (figure 1.4). about $7 billion of highway PPPs in opera- tion and roughly $34 billion of highway PPPs under construction. About one-third of Fiscal Risks from Contingent the PPPs under operation and two-thirds of Liabilities Due to Early those under construction showed signs of Termination of PPPs distress. Infrastructure developers and banks This section analyzes the fiscal risks from (mainly public) that financed highway PPPs contingent liabilities that are realized when were also stressed. As a result, the national an infrastructure PPP project is distressed. In highways development program experienced this analysis, distress refers to early termina- a sudden stop in 2013 and 2014 after a tion of the PPP project. The fiscal costs esti- period of rapid growth between 2010 and mated in this section are the costs that a 2012 (figure 1.3). government incurs in the event that a project FIGURE 1.3  Number of National Highway Public-Private Partnership Projects in India, by Year of Financial Closure, 2001–18 40 30 Number of projects 20 10 0 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 Source: Private Participation in Infrastructure database. Note: Year of financial closure refers to the year in which the sponsor secured financing for the project. PPP = public-private partnership. P UBLI C - P RIV A TE P A RTNERS H I P S IN SOUT H A SI A    31 FIGURE 1.4  Traditional versus Public-Private Partnership Procurement of Infrastructure in India, 2001–17 5,000 4,000 Rs, billion 3,000 2,000 1,000 0 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 PPP Traditional Sources: Department of Economic Affairs, Ministry of Finance, India; Private Participation in Infrastructure database. Note: For PPPs, year is the year of the concession agreement, financial closure, or the appointed date, whichever is available, in that order. For traditional procurement, the year is the project award year. PPP = public-private partnership. is terminated early. To value the fiscal risks, data for all characteristics. For the variables the study adopts a value-at-risk methodology essential for the analysis—namely, contract (see annex 1A for details). period and the level of government that granted the contract—missing data were added for all projects using the individual Predicted Probabilities of Distress for project descriptions provided in the database, Active PPP Projects in South Asia if available. The institutional characteristics Data come from four sources: the World Bank of a country are drawn from the Polity Private Participation in Infrastructure Project IV data, using variables on yearly executive (PPI) database;12 the Polity IV Project;13 the recruitment, the concept of constraints on the World Bank’s World Development Indicators executive, and the concept of political compe- (WDI);14 and the Systemic Banking Crises tition. From the WDI, annual series of per data set of Laeven and Valencia (2018) capita growth rate and nominal exchange (see annex 1B). rates are used to create series of detrended The PPI database includes data on project and demeaned series of per capita growth characteristics as they were agreed at the time rates and exchange rate shocks using the filter of the signing of the PPP contract or at the suggested by Hamilton (2018). The data on time of financial closure. These characteristics financial crises come from the Systemic include the type of project, sector, contract Banking Crises data set of Laeven and period, government level (national or subna- Valencia (2018). tional) granting the contract, identities of the The econometric estimation uses the data sponsors, types of government support, on all PPP projects in low- and middle-income amount of investment commitments, and countries. After estimating equation (1A.2), financing information. The PPI database also in annex 1A, predicted probabilities of dis- provides the current status of the project as tress are obtained for the PPP projects in active, concluded, distressed, or canceled. South Asia using the predictions implied by The PPI database is sourced from publicly the survival analysis.15 available information, such as press reports. The PPI database records 7,979 projects in As a result, some projects might not be cap- emerging markets and developing countries, tured in the database, and a considerable encompassing 127 economies, with financial number of projects in the database lack the closure dates from 1990 to 2019. The sample 32   H IDDEN DEBT includes projects from five sectors: ICT, The contract periods of some projects in the energy, transport, water and sewerage, and PPI database have been completed although municipal solid waste. Because the municipal they are labeled as still active. Possible reasons solid waste data have been a recent addition are that successfully concluded projects have to the database, and only cover the currently gone unnoticed because they are not covered in active projects with financial closures starting the news or that project companies have in 2009, they were dropped from the sample obtained contract extensions after fulfilling the to make all projects comparable. ICT proj- terms of their initial contract. The projects ects, merchant and rental greenfield proj- with completed contract periods but still ects,16 management and lease projects, and labeled as active are kept in the analysis and divestitures were also excluded from the relabeled as concluded. sample because they are not PPPs as this study Figure 1.5 shows the distribution of the per- defines them. centage of contract period elapsed within the In the analysis, distress is defined as estimation sample. It includes 3,977 projects, early termination of a project. The PPI of which 167 were canceled. database labels any project “canceled” if An overwhelming majority of the projects the private party has exited by selling or included in the sample have not passed the turning over its shares back to government halfway mark in their contract periods. One or has ceased operations. The database reason is that PPPs have been originated in labels any project “distressed” if it is in larger numbers only recently compared to the international arbitration or either the gov- median contract period of a PPP in the sam- ernment or the private party has requested ple, which is 25 years. Another reason is that that the contract be terminated. Using news some of the older projects in the PPI database articles and other public online sources, the are missing crucial information, such as the current status of each distressed project contract period and level of contracting was determined and all projects were government; hence they were dropped from relabeled as “canceled,” “concluded,” or the sample used for the econometric “active.” When no definitive information estimation. about the resolution of the distress could be found, the project was dropped from Estimation Results: When Are Projects the sample. Most Likely to Fail? PPP projects are most likely to fail during the FIGURE 1.5  Distribution of the Percentage of Contract early portion of their contract periods, and Period Elapsed, 1990–2018 risks accumulated during their contract 700 periods are not trivial, non-parametric esti- 600 mates show (figure 1.6). The risk of early termination for a project increases rapidly 500 until around 20 percent of the project’s con- tract period elapses. It plateaus at this level Frequency 400 and declines slightly until it reaches 300 50 percent. Beyond 50 percent of the contract 200 period, except for a small increase at around the 80 percent mark, the risk of early termi- 100 nation decreases until the project approaches 0 the end of the contract period. The cumulative hazard curve shows that 10 0 0 0 0 0 00 0 0 0 –2 –3 –4 –5 –6 –7 –8 –9 0– –1 10 20 30 40 50 60 70 80 the accumulated probability of distress 90 Percent of contract period elapsed increases steadily, but its pace decreases after Source: Private Participation in Infrastructure database. reaching around 50 percent of the contract P UBLI C - P RIV A TE P A RTNERS H I P S IN SOUT H A SI A    33 period (figure 1.7, panel a). The survival curve FIGURE 1.6  Distribution of Failures of Public-Private Partnerships estimate mirrors the profile and implies that over the Contract Period, 1990–2018 the probability of an average project has a 0.0015 92 percent likelihood of surviving until the end of its contract period (figure 1.7, panel b).17 0.0010 Hazard Estimation Results: What Factors Determine the Likelihood of Distress? 0.0005 The estimation examines the explanatory power of project-level variables, institutional 0 20 40 60 80 100 variables, and macroeconomic variables. Details on selection of the econometric model Percent of contract period are contained in annex 1E. Figure 1.8 Source: Herrera Dappe, Melecky, and Turkgulu 2020. presents the results from estimating the Note: The figure shows the smoothed hazard function estimate over the percentage of the ­contract period completed, using a Gaussian kernel with optimal bandwidth. PPPs = public-private econometric model in equation (1A.2). partnerships. Positive coefficients indicate factors that increase the cumulative hazard and ultimately the probability of distress. Negative coeffi- brought about the rapid demise of the con- cients indicate factors that decrease the tract (Marin 2009). cumulative hazard and ultimately the proba- Large versus small projects. Larger projects bility of distress. are associated with a higher probability of distress, except for the largest PPP projects. Increases in the committed investment in Project-Level Factors physical assets are associated with higher Brownfield versus greenfield projects. The probability of distress as long as the invest- probability of distress from a brownfield proj- ment is less than $3.4 billion. For investments ect is not statistically different from that of a in physical assets above $3.4 billion (about greenfield project, the estimation results sug- 1 percent of the sample), the higher the invest- gest.18 Private sponsors tend to express a pref- ment, the lower the probability of distress. erence for brownfield projects over greenfield Direct versus indirect government support. projects because the returns of the latter proj- Government support that reduces financing ects are uncertain. The results show that ex risk is the most effective in preventing ante uncertainty about the return of projects is distress. Direct government support, which not associated with higher risk of distress. includes capital and revenue subsidies and in- Sector. Natural gas, railroad, toll road, kind transfers, is associated with lower prob- treatment plant, and water utility projects ability of distress. Indirect government are associated with higher hazard rates rel- support, which includes various guarantees to ative to electricity projects. PPPs in water the sponsors and support from multilateral utility and treatment plant projects experi- organizations, also reduces the likelihood of enced high rates of early termination distress—but the coefficients are not statisti- because of difficulties in adapting the con- cally significant at the 10 percent level. tracts to changing conditions, contract Subnational versus national governments. designs that were not viable, and a bidding PPPs with subnational governments are less process that led to unrealistic financial con- likely to face early termination than PPPs ditions. For example, a concession in with central governments. This finding could Cochabamba (Bolivia) required substantial be related to better project selection at the tariff hikes to make the large investment local level because the local authorities may required from the private operator viable— understand the local problems better or which proved socially unsustainable and oversee the project better because it is nearby. 34   H IDDEN DEBT FIGURE 1.7  Estimates of Survival and Cumulative Hazard for Public-Private Partnership Projects a. Nelson-Aalen cumulative hazard curve b. Kaplan-Meier survival curve 0.10 1.00 Cumulative survival probability 0.08 0.98 Cumulative hazard 0.06 0.96 0.04 0.94 0.02 0.92 0 0.90 0 20 40 60 80 100 0 20 40 60 80 100 Percent of contract period Percent of contract period Source: Herrera Dappe, Melecky, and Turkgulu 2020. FIGURE 1.8  Factors That Predict the Likelihood of Public-Private Partnership Distress Brownfield Natural gas Airports Railroads Toll roads Seaports Treatment plant Water utility Direct government support Indirect government support Multilateral support Subnational government contract Physical investment (US$ billion) Physical investment squared (US$ billion) Executive recruitment concept Executive constraints concept Political competition concept Annual GDP per capita growth rate (detrended, previous year) Annual depreciation (detrended, previous year) Banking crisis occurred (previous year) Debt crisis occurred (previous year) Exchange rate crisis occurred (previous year) –4 –3 –2 –1 0 1 2 3 4 5 z-statistic Source: Herrera Dappe, Melecky, and Turkgulu 2020. Note: Vertical lines indicate the thresholds for significance at the 90 percent confidence level. Blue color indicates significance at the 10 percent level. The base category for sectoral indicators of projects (natural gas, airports, railroads, toll roads, seaports, treatment plant, water utility) is the electricity sector. P UBLI C - P RIV A TE P A RTNERS H I P S IN SOUT H A SI A    35 It could also be the case that national govern- Systematic banking and debt crises. The ments tend to engage in risky projects as they occurrence of systematic banking and debt can bear the termination risk from an crises are associated with higher hazard rates individual PPP project because they have a for PPP projects. A systematic banking crisis more diversified PPP portfolio and fiscal undermines the ability of financial institutions resources. The highway projects in India to provide the financing that is necessary to provide an interesting example: all the high- sustain long-term infrastructure projects. way projects that were canceled between A debt crisis can limit the government’s abil- 2012 and 2015 were PPPs with the central ity to fund PPP projects according to the government. At the same time, state govern- terms of the contracts. It may also hinder the ments continued to enter into successful PPPs ability of a local private party to secure debt for road construction and operation. financing through the market and increase the cost of its outstanding debt, leading to early Institutional Factors termination of PPP projects. Because of the Constraints on the executive branch of gov- long-term nature of PPPs and the high trans- ernment. Greater constraints on the execu- action costs of preparing, procuring, and tive are associated with lower probability of awarding them, both parties try to negotiate distress. When the government can exercise changes to the contract or some kind of com- authority without adequate checks and pensation in response to macrofinancial balances, it leaves PPPs vulnerable to expro- shocks. Early termination happens only if the priation by the government through a change parties cannot reach an agreement, which in policy or direct interference. Hence, the explains the lag in the impact of macrofinan- project becomes susceptible to policy and cial shocks. political risks (Irwin 2007; Grimsey and Lewis 2017). More generally, when the con- Predicting the Probabilities of Distress for straints on the executive are not stringent Currently Active PPPs in South Asia enough, the contract loses its value in mediat- Probabilities of distress for currently active ing the relationship between the government projects in the PPI database in South Asia and the private party, making the project have been predicted using equation (1A.9), in more susceptible to distress. annex 1A, assuming all active projects have survived until the end of 2019. Missing values Macroeconomic Factors for project-specific variables essential for pre- Unexpected currency depreciation. A surprise diction were collected. When such efforts depreciation in the local currency—a devia- proved fruitless, values were imputed using tion of the annual depreciation rate from its the characteristics of similar projects. The long-term average—is associated with a details of the imputation process for all vari- higher risk of distress. Exchange rate risk ables are presented in annex 1D. affects infrastructure investment in two ways, Active projects in South Asia are far from as Irwin (2007) notes. First, many infrastruc- being riskless. Figure 1.9 shows the distribu- ture PPPs, such as those in power generation, tion of the predicted probability of distress use inputs priced in foreign currency. Second, at the 99th percentile (that is, the maximum given the insufficient local savings and loss with 99 percent confidence) of active underdeveloped local currency markets in projects over the remainder of their contract most low- and middle-income economies, periods after 2020. The predictions assume financing of long-term infrastructure projects that no banking or debt crisis occurs and most often relies on debt denominated in for- that the local currency does not depreciate eign currency, but the revenues of the operators beyond its trend against the US dollar. In are in local currency. The currency mismatch addition, institutional variables are assumed between revenues and costs can push the proj- to remain at their 2018 levels. The median ect company to insolvency very quickly if the probability of distress is 0.03 and the mean local currency depreciates sharply. is 0.049. 36   H IDDEN DEBT FIGURE 1.9  Distribution of Predicted Probabilities of Distress for Public-Private Partnerships in South Asia, from 2020 to the End of Contractual Period 35 30 25 20 Percent 15 10 5 0 0 0.09 0.19 0.29 0.39 0.49 Probability of distress for a project at the 99th percentile Source: Herrera Dappe, Melecky, and Turkgulu 2020. Note: The 99th percentile refers to the confidence level with which the computed loss is not exceeded. What Are the Fiscal Risks to South Asian missing values were imputed as discussed in Governments If PPPs Terminate Early? annex 1D. It is important to note that the analysis can be underestimating a country’s Calculating Exposure at Distress exposure at distress if some PPPs are not In general, governments are exposed to obli- captured in the PPI database. gations from the debt and equity financing of The total debt used to finance the cur- an infrastructure project when the PPP is ter- rently active 1,056 greenfield and brownfield minated. As the ultimate guarantor of the PPP projects in energy, transport, and water public infrastructure service, the government and sewerage in South Asia is estimated to steps in to resolve the matter. As indicated in be $218 billion, and the total equity financ- World Bank’s Guidance on PPP Contractual ing is estimated to be $77 billion.19 The aver- Provisions, the market practice in the event a age leverage ratio, at the time of financial PPP contract is terminated is that both the closure, among the active PPP projects in lenders and the equity owners must be com- South Asia is 3.22. Debt financing makes up pensated if distress occurred without any fault more than 70 percent of total physical of either party (World Bank 2019). Without investments in India, Nepal, and Pakistan, such explicit or implicit guarantees, especially while it is between 60 percent and 70 percent in emerging market economies and develop- of total physical investment in Bangladesh, ing countries, private finance cannot be Bhutan, and Sri Lanka (figure 1.10). Capital effectively mobilized. grants from government play a larger role in The PPP portfolio captured by the PPI Afghanistan than in the other countries in database is used to value the fiscal risk from South Asia. early termination of PPPs for each country, and the exposure at distress due to each proj- Calculating Value at Risk ect is calculated using the debt and equity The losses from debt and equity obligations in data in the PPI database. The PPI database the event of distress depend on the causes of provides the shares of the physical investment distress. In the case of government default or that have been financed through debt, equity, voluntary termination of the project, the mar- or capital grant from the government. In the ket practice is for the government to compen- case of missing values, data were collected. sate the private party for the full amount When information could not be found, the invested in the project (debt and equity) P UBLI C - P RIV A TE P A RTNERS H I P S IN SOUT H A SI A    37 FIGURE 1.10  Composition of Public-Private Partnership Financing for Active Projects in South Asia, by Country, 1990–2018 100 80 60 Percent 40 20 0 Afghanistan Bangladesh Bhutan India Nepal Pakistan Sri Lanka Debt Equity Capital grant Sources: World Bank staff calculations; Private Participation in Infrastructure database. Note: PPP = public-private partnership. plus the equity return it had forecasted developed countries, collected by the Data (World Bank 2019). Alliance Project Finance Consortium show In the case of termination due to the pri- that the average ultimate recovery rate is vate partner’s default or breach of contract, 79.3 percent (Moody’s Investors Service 2019; the market practice is to provide some see box 1.2).20 This might be a low estimate in amount of compensation. The justification the context of South Asia because the model for compensation is that if there is no com- PPP concession agreements in the road sector pensation, the government might be seen as in India guarantee as much as 90 percent of enjoying windfall gains unfairly and would the debt financing, even in the cases of the have a hard time attracting lenders and inves- project company’s default or force majeure. tors for PPP projects in general (EPEC 2013; The data on compensation of private World Bank 2019). Even if a private partner equity are even scarcer than data on recovery defaults, the private partner may legally allege rates of bank loans. The government’s loss government responsibility, so the government involving private equity in the event of dis- becomes liable to compensate the private tress depends on the reason for termination, party or otherwise incur additional legal costs the explicit clauses in the contract, and— (World Bank 2019). potentially—the negotiation at the time of In the case of force majeure, because the termination. Road concession agreements in event triggering distress is outside both par- India offer a range of possibilities depending ties’ control, the risk should be shared on the source of termination of the contract. between both parties. As such, the govern- If the project company defaults, the conces- ment is liable for less than full compensation sion agreements do not foresee any compen- and has the right to take over the relevant sation on equity, but if the public authority asset, while the private partner loses any defaults, the contract entitles the private return on its invested equity and possibly sponsor to 150 percent of its equity. If a force some of the invested equity (EPEC 2013; majeure event that is indirectly caused by a World Bank 2019). political event occurs, the private party is Only limited data are available on losses entitled to 110 percent of the equity it invested incurred by governments in cases of early ter- in the project. Anecdotal evidence also sug- mination of PPPs. The data on the recovery gests that no matter the cause for termination, rates of bank loans to PPP projects, mainly in governments might pay a premium on the 38   H IDDEN DEBT BOX 1.2  Low-, Medium-, and High-Risk Scenarios for Computing Losses to the Government from Contingent Liabilities of Public-Private Partnerships In the event that a project is terminated, it is safe not compensate the private party for the foregone to assume that the government will lose the entire return. public equity. Based on the considerations dis- In the high-risk scenario, in addition to cussed so far in this chapter, equation (1A.1) in compensating for all the debt (LGDDebt = 1), the ­ annex 1A can be written as follows, assuming that ­ government compensates the private party three scenarios—low, medium, and high risk—are project 150 percent of the equity it invested in the ­ considered for the loss given distress: (LGDEquity = 1.5), in line with the aforemen- tioned contract terms for India’s road sector. ELi,99%= PDi,99% (Debti LGDDebt + The expected losses from contingent ­liabilities Public Equityi + Private due to PPPs in country c, reported with Equityi LGDEquity). (B1.2.1) 99 percent confidence—that is, with 99 percent confidence that the maximum annual loss will In the low-risk scenario, the loss on debt given not exceed the calculated amount—are the sum distress of the government from a project of the expected losses within the set of all active (LGDDebt) is assumed to be the recovery rate esti- projects in the country, Ic: mated by Moody’s Investors Service (2019): (LGDDebt = 0.793). Assuming this, the government  does not cover the loss of private equity and only ELc ,99% = ∑ i∈Ic ELi ,99%. (B1.2.2) loses its own equity in the PPP, (LGDEquity = 0). In the medium-risk scenario, the loss on both Note: In the calculation of ELi ,99% for each project, correlations across projects debt and equity given government distress from a in the same country are taken into account via both the distress probabilities, which control for country-specific factors, and their standard errors, c ­ alculated project is assumed to be 1 ( LGD Debt = 1; using the delta method. The delta method uses the variance-­ covariance LGDEquity = 1). That is, the government guaran- matrix of the regression analysis, in which the observations have been tees the total financing of the project, but does assumed to be correlated within the country (clustered at the country level). equity to compensate the private party for the investments, the fiscal cost estimates in Sri loss of expected return on its investment. Lanka are less than one-sixth of the esti- mates in Bangladesh and less than 80 percent Estimating Fiscal Costs of Active PPPs of the estimates in Nepal. The main reason is in South Asia that the PPPs in Sri Lanka have mostly India, Pakistan, and Bangladesh have the passed 40 percent of their contract period, highest estimated fiscal costs from early ter- while the portfolios in Bangladesh and Nepal mination of active PPPs in the region, while are younger.21 Afghanistan and Bhutan have the lowest. In most South Asian countries, about The estimated fiscal cost from early termina- 39 percent to 50 percent of the fiscal costs tion over the remainder of the contract from early termination of active projects are periods of the PPPs ranges from $9.7 billion due to the risks of early termination during to $18.5 billion in India; $1 billion to the 2020–24 period. Nepal is the exception $2 billion in Pakistan; and $379 million to to this trend because the costs increase at a $730 million in Bangladesh (see table 1F.1 in slower pace. In the low scenario, for India the annex 1F). Even though Sri Lanka’s current fiscal cost due to the risk from PPP cancella- PPP portfolio is about half the size of tions is estimated not to exceed $853 million Bangladesh’s portfolio in terms of the num- (8.8 percent of total costs) in 2020, with ber of projects, and almost 50 percent larger 99 percent confidence. In the 2020–24 than Nepal’s portfolio in terms of total period, the estimated fiscal cost (value at risk P UBLI C - P RIV A TE P A RTNERS H I P S IN SOUT H A SI A    39 FIGURE 1.11  Estimated Total Fiscal Costs from Early Termination of Public-Private Partnership Portfolio in South Asia, as a Percentage of GDP, 2020–24 0.7 0.6 0.5 0.4 Percent 0.3 0.2 0.1 0 Afghanistan Bangladesh Bhutan India Nepal Pakistan Sri Lanka Low Medium less low High less medium Source: Herrera Dappe, Melecky, and Turkgulu 2020. Note: The estimated fiscal costs are based on the value at risk at 99 percent over the entire contract period and are expressed as percentage of GDP of a single year. The 99th percentile refers to the confidence level with which the computed maximum loss is not exceeded during the relevant period. PPP = public-private partnership. at 99 percent) is about $3.8 billion (39 India’s revenues represent 20.5 percent of percent of total costs). GDP.22 Pakistan faces the most significant Nepal, India, and Pakistan have the high- fiscal challenge from early termination of est estimated fiscal costs from early termina- PPPs in South Asia—slightly less than 4 tion of active PPPs as a share of GDP percent of the government’s annual reve- (figure 1.11). The cumulative fiscal costs nues. Even though Bangladesh’s estimated estimated for the entire lifetime of the PPP fiscal costs are low relative to the size of its portfolio in Nepal (based on the value at risk economy, they are high compared to the at 99 percent) ranges from 0.38 percent to annual government revenues (figure 1.12), 0.67 percent of annual GDP; in India, from posing a significant fiscal challenge to the 0.35 percent to 0.67 percent; and in Pakistan, country, if only the revenues of a single year from 0.33 percent to 0.61 percent. are available to absorb the costs of early ter- Bangladesh, Bhutan, and Sri Lanka follow, mination of PPPs. with estimated fiscal costs from early The probability that an entire PPP port- termination of active PPPs ranging from folio is terminated in a single year is very 0.14 percent to 0.26 percent, 0.16 percent to low. Hence, a more realistic analysis is to 0.22 percent, and 0.06 to 0.12 percent of compare the estimated fiscal costs over a annual GDP, respectively. These estimates period of time with an estimate of the gov- give an idea of the resources that would be ernment revenues over the same period of needed in case of early termination of the PPP time. The estimated fiscal costs from early portfolio relative to the size of the economy. termination of PPPs as a percentage of gov- South Asian governments are quite differ- ernment revenues tend to decrease over the ent in terms of their revenue mobilization 2020–24 period. Figure 1.13 presents the capacity, which helps determine their ability estimated fiscal costs for different periods, to absorb the fiscal costs from early termina- all starting at the beginning of 2020, as the tion of PPPs in infrastructure. The govern- ratio of government revenues of that ment of Bangladesh’s revenues represent period.23 The estimated fiscal costs decline only 9.6 percent of GDP, while the govern- for most countries because active projects ment of Pakistan’s revenues represent 15.6 get older and some reach the end of their percent of GDP, and the government of contract periods. 40   H IDDEN DEBT FIGURE 1.12  Estimated Total Fiscal Costs from Early Termination of the Public-Private Partnership Portfolio in South Asia, as a Percentage of Government Revenues for a Single Year 4 3 Percent 2 1 0 Afghanistan Bangladesh Bhutan India Nepal Pakistan Sri Lanka Low Medium less low High less medium Source: Herrera Dappe, Melecky, and Turkgulu 2020. Note: Low, medium, and high correspond to the three scenarios. The estimated fiscal costs are based on the value at risk at the 99th percentile and are over the entire contract period and are expressed as percentage of government revenue of a single year. The 99th percentile refers to the confidence level with which the computed maximum loss is not exceeded during the relevant period. PPPs = public-private partnerships. FIGURE 1.13  Estimated Fiscal Costs from Early Termination of the Public-Private Partnership Portfolio in South Asia over Different Periods as a Percentage of Expected Government Revenues, 2020–24 a. Low scenario b. Medium scenario c. High scenario 0.4 0.4 0.4 0.3 0.3 0.3 Percent 0.2 0.2 0.2 0.1 0.1 0.1 0 0 0 2020 2021 2022 2023 2024 2020 2021 2022 2023 2024 2020 2021 2022 2023 2024 Afghanistan Bangladesh Bhutan India Nepal Pakistan Sri Lanka Source: Herrera Dappe, Melecky, and Turkgulu 2020. Note: Each data point represents the estimated fiscal costs from early termination of the PPP portfolio based on the 99 percent value at risk over the period starting at the beginning of 2020 and ending at the end of the corresponding year, as a percentage of the government revenue over the same period. The analysis assumes that annual government revenues during the 2020–24 period are the same as in 2019. PPP = public-private partnership. P UBLI C - P RIV A TE P A RTNERS H I P S IN SOUT H A SI A    41 Estimating the Effects of Adverse a banking and a debt crisis in 2020.25 Such a Macrofinancial Shocks profound macroeconomic crisis is similar to Multiple types of macroeconomic shocks some crises in emerging market economies and have significant effects on the probability of developing countries that led to early termina- early termination (figure 1.8). The estimated tion of many PPPs. fiscal costs presented so far assume there will A profound macroeconomic crisis in 2020 be no macrofinancial shock, such as deprecia- would significantly increase the estimated tion of local currency or a stress in the fiscal costs from early termination of PPPs, banking sector. particularly in 2021. The estimated fiscal This section simulates the results of a mac- costs over the 2020–21 period could be as rofinancial shock. The Systematic Banking high as 4.3 percent of government revenues Crises data set of Laeven and Valencia (2018) in Pakistan, 3.9 percent in Bangladesh, and identifies 104 banking crisis episodes among 3.7 percent in India (figure 1.14). In Nepal, the countries included in the PPI data set, of early termination of PPPs could require up to which 13 also involved both sovereign debt 3.3 percent of the government revenues and currency crises. In the episodes of banking, under a scenario of severe macrofinancial debt, and currency crises, the maximum yearly crisis, while in Bhutan and Sri Lanka, it deviation in the depreciation rate from its long- could require around 1 percent of govern- term average ranged between 15.1 percentage ment revenues. These estimates underesti- points and 116 percentage points, with an mate the effect of the crisis because average of 48.3 percentage points.24 The simu- government revenues are kept constant— lation assumes a 48.3 percentage-point cur- even though they would contract in such a rency depreciation shock and the occurrence of profound economic crisis. FIGURE 1.14  Estimated Fiscal Costs from Early Termination of the Public-Private Partnership Portfolio Assuming Profound Macrofinancial Shocks, as a Percentage of Government Revenues, 2020–24 a. Low scenario b. Medium scenario c. High scenario 5 5 5 4 4 4 3 3 3 Percent 2 2 2 1 1 1 0 0 0 2020 2021 2022 2023 2024 2020 2021 2022 2023 2024 2020 2021 2022 2023 2024 Afghanistan Bangladesh Bhutan India Nepal Pakistan Sri Lanka Source: Herrera Dappe, Melecky, and Turkgulu 2020. Note: Each data point represents the estimated fiscal costs from early termination of the PPP portfolio based on the 99 percent value at risk over the period starting at the beginning of 2020 and ending at the end of the corresponding year, as a percentage of government revenue over the same period. The analysis assumes that government revenues during the 2020–24 period are each year the same as in 2019. PPP = public-private partnership. 42   H IDDEN DEBT Features of Contract Design That toll-based and annuity-based projects. Toll- Matter: Exploring the Link between based projects entitled the project company to PPP Contract Design and Early charge tolls. Annuity-based projects entitled Terminations of Highway PPPs in the project company to semi-annual availabil- India ity payments from the NHAI after the road was built or rehabilitated. PPPs for national highways in India experi- All contracts were awarded through com- enced several early terminations between 2013 petitive auctions. For toll-based projects, the and 2015. Anecdotal evidence points to the outcome of the auction could be either an incentives created by some contract designs as annual premium payment from the private a potential reason for early termination (ADB sponsor to the government, which would esca- et al. 2018; Pratap and Chakrabarti 2017). late at 5 percent yearly, or an upfront capital This section takes advantage of the existence grant from the government for the PPP project of different contract designs for national high- through a scheme known as Viability Gap way PPPs in India during the early 2010s to Funding—based on the expected profitability identify contract features that can help explain of the roads. Sponsors would bid either the the large number of early terminations. highest premium they would pay or the lowest Narrowing the sample to national highway capital grant they would require. For annuity- PPPs in India allows the analysis to examine based projects, the sponsors would bid the contract features that are not available for the lowest annuity payments they require.26 global and cross-sectoral sample used in previ- The procurement process ends in three ous sections. The analysis in this section can types of contracts, each of which presents a help improve the design of road PPPs and their different set of risks. While both types of contract structure and inform the design of contracts for toll-based projects expose the PPPs in other sectors with similar incentive project company to demand risk, capital- structures. grant contracts alleviate the financing risk. The analysis uses those national highway Annuity-based contracts insulate the PPPs in India with financial closure or conces- project company from the demand risk, but sion agreement years from 2010 to 2014 the company is still exposed to the full included in the PPI database. Of the financing risk. 125 projects, 26 were canceled, and 24 of In 2012, an unusually high number of PPP those cancellations occurred between 2013 contracts were awarded based on premium and 2014. One cancellation occurred in 2015 payments, and about half of them were can- as a result of a lengthy court process over the celed (figure 1.15). This increase occurred terms of termination after the project com- partly because the NHAI decreased the devel- pany decided to withdraw from the project in opment of annuity-based projects after 2011, 2014. The time frame allows the analysis to and private sponsors started bidding more compare canceled projects with contempora- aggressively for toll-based contracts. The neous active projects. decrease in the capital-grant contracts indi- Data on the investment amount, the project cates that private sponsors were more opti- status, the length of the road to be constructed, mistic about the projects offered in 2012 and the data on financing mix were extracted compared to the earlier toll-based projects. from the PPI database. When the data for a A logistic regression model was estimated project were missing in the PPI database, the to identify the contract characteristics affect- data were collected from the concession agree- ing early termination of national highway ments published on the National Highways PPPs. The contract characteristics are intro- Authority of India (NHAI) website, and in the duced in the regression in three different case of financing data, from online news arti- ways. First, they are introduced as indicator cles or private sponsors’ annual reports. variables for each type of contract. Second, During this period, two types of PPP the net present value (NPV) of payments to contracts were mainly initiated by NHAI: the government is introduced as a continuous P UBLI C - P RIV A TE P A RTNERS H I P S IN SOUT H A SI A    43 FIGURE 1.15  Number of Indian Highway Projects Canceled versus Not Canceled, by Contract Type and Financial Closure Year, 2010–14 25 20 Number of projects 15 10 5 0 2010 2011 2012 2013 2014 2010 2011 2012 2013 2014 2010 2011 2012 2013 2014 Premium Capital grant Annuity Not canceled Canceled Sources: Private Participation in Infrastructure database; National Highways Authority of India. Note: The concession agreement year is used if financial closure had not been achieved by the time of cancellation. indicator of the net cost of the contract to the the total investment financed by debt sponsor.27 Third, the net present value of pay- (columns 4–9). The findings indicate that as ments is rescaled by the amount of invest- the sponsor and the project company’s finan- ment. Two additional variables are introduced cial burden increases, the likelihood of cancel- in all cases: road length and fraction of the lation of the project increases. Alternatively, total investment financed by debt. The debt the premium-payment contracts may create variable only applies to projects that secured an unsound incentive structure that encour- financing. Table 1F.2 in annex 1F presents the ages overoptimistic bids from the private regression results. sponsor to win the PPP contract in order to The size of premium payments and tender access funds that can be channeled to other payments matters. The larger the tendered construction companies in the same financial premium payments to the government, the group (ADB et al. 2018; Pratap and higher the probability that national highway Chakrabarti 2017). PPPs were later canceled. Similarly, the The likelihood of cancellation increases smaller the tendered payment from the gov- with the share of the investment financed ernment to the sponsor as capital grant or through debt. This finding is important from annuity payment, the higher the probability a fiscal risk perspective because roughly that national highway PPPs were later can- 80 percent of the debt to finance the NHAI celed. Premium-payment contracts are more portfolio has come from public sector banks likely to be canceled than annuity-based con- in which the government owns more than tracts when no other project characteristics 50 percent of capital shares. The large number are included (column 1 in table 1F.2). of cancellations in the NHAI portfolio have However, the relationship is not statistically contributed to the rising number of nonper- significant when controlling for the length of forming assets in India’s banking sector. The the road (column 2). When using the net pres- State Bank of India, which holds the greatest ent value of the premium payments to gov- nominal amount of Indian highway-related ernment, either by itself or as a share of debt, reported that about 20 percent of loans investment in the project, the coefficient again to ports and highways were nonperforming turns statistically significant—even when con- by the end of 2016, with the trend increasing trolling for road length and fraction of in 2016 (ADB et al. 2018). 44   H IDDEN DEBT The likelihood of cancellation increases Second, the sector in which a PPP project with the length of the highway. The finding operates, and the size of a project, in terms that longer highways are more likely to be of its physical investment, matter in evaluat- canceled is consistent with the observation ing riskiness of a project. PPPs in the power that many of the cancellations have been and seaport sectors are less likely to expe- related to the government having problems rience early termination than PPPs in the with securing the right of way. other infrastructure sectors analyzed. Large physical investments lead to increased rates of early terminations. Third, delegating PPP contracting and Improving Government Capacity, monitoring to state and local governments Due Diligence, and Contract Design should be considered when it is institution- to Better Manage the Fiscal Risks ally and economically possible. PPPs show of the Growing PPP Programs in lower probability of early termination when South Asia they are contracted and monitored by state The fiscal risks from the current infrastruc- and local entities. The subnational govern- ture PPP programs in South Asia are not neg- ments could have better information to ligible for some countries. Under a severely monitor and could be held more directly adverse scenario, the potential fiscal costs accountable for effective implementation of from early termination in 2020–21 amount to the project—but even here risk could arise 3.3 percent to 4.3 percent of the government (see chapter 4). revenues in Bangladesh, India, Nepal, and Fourth, PPPs tend to have reduced rates of Pakistan. The pipeline of PPP projects in early termination when the contract is exe- South Asia is quite large, particularly in cuted in a country with stronger constraints Bangladesh, which has the same number (30) on the power of the executive branch. These of infrastructure projects in the pipeline as it constraints could deliver the required disci- has active projects. Such an expansion of the pline and decrease uncertainty during proj- PPP program can lead to a significant increase ect implementation. in fiscal risks. Therefore, an important agenda Fifth, macrofinancial shocks are an in South Asia is improving the design and important cause of early termination of management of infrastructure PPPs to miti- PPPs, highlighting the importance of macro- gate the corresponding fiscal risks, while economic management in enabling sustain- ensuring timely implementation of financially able funding and financing of PPP projects. responsible infrastructure projects to address the infrastructure deficit. Recommendations for Building Government Capacity and Undertaking Five Overarching Lessons Rigorous Due Diligence, Assessments, and Feasibility Studies The analysis identifies five overarching lessons Government capacity to prepare, procure, from the global PPP data. and manage PPP projects must be built to First and foremost, several factors that ensure that the expected efficiency gains are increase the risk of early termination of PPP indeed achieved and that the fiscal risks from projects are related to the financing risk contingent liabilities are contained and prop- of the project. Governments may alleviate erly managed. Good practices for the prepa- some of the financial risk of PPPs by pro- ration, procurement, and management of viding support (helping de-risk the projects) PPPs can help governments improve their through capital grants, revenue subsidies, capabilities to take advantage of PPPs at more or in-kind transfers, which lead to reduced acceptable levels of risks (World Bank 2019). rates of early termination. An important good practice is to ascertain the P UBLI C - P RIV A TE P A RTNERS H I P S IN SOUT H A SI A    45 fiscal implications of PPPs, including specify- use these changes opportunistically by either ing their budgetary, accounting, and reporting the private partner or the procuring treatment. The Ministry of Finance or central authority. budget authority should approve the long- Specific circumstances—force majeure, term financial implications of the project materially adverse government action, change (World Bank 2019). in the law, refinancing—that may arise during Given the complexity, magnitude, and the life of the contract should be also expressly inherently long-term nature of PPP projects, regulated. Dispute resolution mechanisms the procuring authority should exercise a should be in place allowing the parties to good amount of due diligence and perform resolve differences in an efficient and satisfac- rigorous assessments to gauge the viability of tory manner without adversely affecting the infrastructure projects before deciding on a project (ADB et al. 2018). PPP procurement. Giving lenders step-in rights for cases when A sound PPP preparation starts with the the private partner is at risk of default or identification of potential infrastructure proj- if the PPP contract is under threat of termina- ects that could be procured as PPPs. The tion for failure to meet service obligations is results from the econometric analysis suggest another good practice to avoid early that it is important to undertake feasibility termination and reduce the fiscal costs. studies to inform the structure of the PPP Having well-defined grounds for termina- project, including assessing and deciding on tion of the PPP contract and its associated the allocation of risks, and sounding out the consequences can also reduce fiscal costs from market to gauge its appetite and capacity, early termination (World Bank 2019). which can reduce the probability of early termination. Annex 1A. Methodology to Addressing Issues in Contract Design Determine the Value at Risk of a Public-Private Partnership In India, PPPs for national highways were The study employs the value-at-risk method- more likely to terminate early if the contract ology. Accordingly, the fiscal risk from project put the private sponsor under larger financial i is valued as the maximum expected loss with commitments through higher premium 99 percent confidence: payments to the government or lower capital grants or annuity payments from the govern- ment. The reason could be exposure to ELi,99% = PDi,99% * EADi * LGDi , demand risk and a perverse incentive struc- (1A.1) ture that encourages overly optimistic bids on where PDi,99% is the probability of distress for payments to the government in order to win project i such that there is 1 percent chance tenders on PPP contracts that will be financed that the probability will exceed PDi,99%; EADi with loans from state-owned banks with is the government’s exposure to project i at the weak due diligence capacity. time of distress; and LGDi is the government’s loss given project i distress. Recommendations for Improving Contract The PDi,99%, EADi , and LGDi are obtained Design separately. The probability of distress of a The success of project implementation will public-private partnership (PPP) is predicted determine whether the project delivers the using a flexible parametric survival model of expected value for money and whether fiscal the realization of distress conditional on risks have been properly managed. Following project-specific and country-specific institu- ­ established good practices, modification and tional variables and macroeconomic shocks. renegotiation of the contract should be The EADi is estimated based on the debt and expressly regulated to lower the incentives to equity invested in the project. The LGDi is 46   H IDDEN DEBT determined based on different practices by event-time distribution, and am = (kmax − km)/ countries in case of termination of PPP con- (kmax − kmin). Then, the likelihood function is tracts and the recovery rates of defaulted loans to infrastructure projects. N S (t j | X j β , γ ) ( ) ( ) d L γ ,β = ∏  h t | X β ,γ  j , S (t j 0 | X j β , γ )  j j  Econometric Model for Calculating the j =1 Probability of Distress (1A.3) The probability of distress of a PPP at a specific where j denotes each observation used in the point during its contract period is predicted by analysis. The number of observations in the estimating a flexible parametric proportional estimation sample is the number of project hazards model (Royston and Parmar 2002; years because the time-variant variables are Royston and Lambert 2011).28 The model is all annual. The variable tjo denotes the begin- an extension of the parametric proportional ning of the period for a specific o ­ bservation hazards model with a Weibull baseline hazard ending at time tj, during which the covariates function. The generalization allows for a non- remain constant at Xj. The variable dj equals monotonic baseline hazard function using 1 if the project goes into distress at time tj, restricted cubic splines. Accordingly, coeffi- and dj = 0 if the project survives time tj. cients of the following equation are estimated To estimate the probability of distress at a to maximize the likelihood of the observed dis- specific interval (t0,t), the survival function tribution of failure times: S (·) and the hazard function h (·) can be restated in terms of ln H(t/Xit) in equation 2 (1A.2) using the relations ( ) ( ) ln H t|X it = γ 0 + ∑ γ m zm ln t + X i , proj β proj m =1 + X it ,inst β inst + X it , macro β macro ,  ( ) S t|X it = exp − exp  {  ln H t|X it   , ( (1A.4) )} d (1A.2) (1A.5) dt ( h t|X it = exp  )  ln H t|X it  , ( ) where ln H ( t | X it ) is the log cumulative hazard at time t for project i conditional on ­ Then, after obtaining γ and β, the probabil- Xit = (Xi,proj, Xit,inst, Xit,macro). Xi,proj is the vec- ity of distress for any period (t0,t) for any tor of project-specific time-invariant covari- project i can be recovered using the relation- ates. Xit,inst is the vector of country-specific, ship between the survival and the cumulative time-varying institutional covariates. Xit,macro hazard functions is the vector of country-specific, time-varying ( ) { ( )} macroeconomic shocks. The terms under the ˆ t |X  , ˆ t | X = exp − exp  ln H S summation operator represent the set of j it j   j it j   restricted cubic spline terms in log time scale, z m(ln t ). The time scale is chosen as the (1A.6) percentage of contract period elapsed. The restricted cubic spline functions are ˆ Si t|t0 = ∏ (1A.7) Sˆ t |X j ( ) it j . ( ) ˆ t |X j S j0 it ( j ) ( ) z1 ln t = ln t where j is any observation for project i ( ln t ) = ( ln t − k ) − a ( ln t − k ) 3 3 zm m + m min + such that ( t j0, t j) ⊆ ( t 0, t ). Then, the proba- − (1 − a )( ln t − k )  for m > 1, bility of distress between (t0,t) can be writ- 3 m max + ten as where km are the interior knots, which are  ˆ t|t . PD i = 1 − S (1A.8) i 0 ( ) picked at the centiles of the uncensored log P UBLI C - P RIV A TE P A RTNERS H I P S IN SOUT H A SI A    47 The probability of distress at the 99th per- Annex 1B. Definitions of Variables centile is calculated using the standard errors obtained via the delta method:29 The following variables were used in the study to identify systematic contractual, insti-    + z × s.e.  PD i ,99% = PD i 1% PD i . (1A.9) ( ) tutional, and macroeconomic factors that can help predict the probability that a PPP project will be terminated early. TABLE 1B.1  Project-Level Variables Variable Measure Source Distress 1, if project is canceled;a PPI database and authors’ collection for 0, otherwise. Herrera Dappe, Melecky, and Turkgulu 2020 Contract period Percentage of contract period elapsed until PPI database and authors’ collection for distress (as defined in the first row of the table); Herrera Dappe, Melecky, and Turkgulu 2020 otherwise, until the end of 2018. Type Greenfield if the special purpose vehicle (SPV) PPI database builds and operates a new facility; Brownfield if the SPV takes over an existing asset and either rehabilitates or expands it. Sector Electricity, natural gas, telecom, airports, railroads, PPI database toll roads, seaports, treatment plants, water utility. Direct government 1, if capital, revenue, or in-kind subsidies exist; PPI database support 0, otherwise. Indirect 1, if payment, debt, revenues, guarantee, etc. PPI database government exist; support 0, otherwise. Multilateral 1, if a multilateral bank provides financial support; PPI database and authors’ collection for support 0, otherwise. Herrera Dappe, Melecky, and Turkgulu 2020 Subnational 1, if local or provincial/state government PPI database and authors’ collection for government grants the contract; Herrera Dappe, Melecky, and Turkgulu 2020 contract 0, otherwise. Physical Total investment in physical assets, in 2019 PPI database and authors’ collection for investment US$, billion. Herrera Dappe, Melecky, and Turkgulu 2020 Source: Herrera Dappe, Melecky, and Turkgulu 2020. Note: The PPI database is the World Bank Private Participation in Infrastructure Project (PPI) database, Version 2019 H1, available at https://ppi.worldbank​ .org/en/ppidata. “Authors’ collection” refers to the recoding of cases labeled as distressed in the data set as active or canceled based on information avail- able through public sources, such as news articles. a. A project is canceled if the private party has exited by selling or turning over its shares to the government or by ceasing operations. TABLE 1B.2 Institutional Variables Variable Measurea Source Executive recruitment Index of openness of executive recruitment, ranging from Polity IV Project concept 1 (succession by birthright) to 8 (competitive election). Executive constraints Index of the degree of constraints on the executive, ranging from Polity IV Project concept 1 (unlimited authority) to 7 (executive parity or subordination). Political competition Index of the degree of competition in politics, ranging from Polity IV Project concept 1 (suppressed) to 10 (institutional electoral). Note: “Executive” refers to the executive branch of government. a. The values for the interruption, interregnum, and transition periods, during which the institutional variables take values outside of their regular ranges in Polity IV, were linearly interpolated. Specifically, when the political structure faces some sort of irregularity (interruption, interregnum, or transition), the database assigns the variable a value of –66, –77, or –88, depending on the type of irregularity. Interpolation implicitly assumes a continuous rather than an abrupt resolution to any sort of political irregularity. 48   H IDDEN DEBT TABLE 1B.3  Macroeconomic Variables Variable Measure Source Annual GDP per capita Annual GDP per capita growth rate of the Herrera Dappe, Melecky, and Turkgulu growth rate country, detrended and demeaned using 2020, based on Hamilton 2018; World the filter suggested by Hamilton (2018). Development Indicators (WDI) Previous year’s value, calculated using the “GDP per capita growth (annual %)” series from the WDI. Annual depreciation Depreciation rate of the local currency against Herrera Dappe, Melecky, and Turkgulu (detrended, previous the US dollar, detrended and demeaned 2020, based on Hamilton 2018; WDI year) using the filter suggested by Hamilton (2018). Previous year’s value, calculated using the “DEC alternative conversion factor (LCU per US$)” series, from the WDI. Banking crisis occurred 1, if a systematic banking crisis occurred in the Laeven and Valencia 2018 (previous year) country during the previous year; 0, otherwise. Debt crisis occurred 1, if a debt crisis occurred in the country Laeven and Valencia 2018 (previous year) during the previous year; 0, otherwise. Exchange rate crisis 1, if an exchange rate crisis occurred in the Laeven and Valencia 2018 occurred (previous year) country during the previous year; 0, otherwise. Annex 1C. Distribution of South Asian Public-Private Partnership Projects by Sector TABLE 1C.1  Sectoral Distribution of Public-Private Partnership Projects with Financial Closure in South Asia, by Country Water and Country ICT Energy Transport sewerage Total Afghanistan Projects 0 2 0 0 2 US$, million 0 39 0 0 39 Bangladesh Projects 5 (2) 31 8 (1) 1 45 (3) US$, million 76 5,372 1,097 333 6,877 Bhutan Projects 1 1 0 0 2 US$, million 0 240 0 0 240 India Projects 7 (1) 414 (3) 543 (27) 20 984 (31) US$, million 1,624 162,375 117,595 1,337 282,932 Maldives Projects 0 0 1 (1) 0 1 (1) US$, million 0 0 469 0 469 Nepal Projects 2 34 1 1 38 US$, million 12 2,491 378 0 2,880 Pakistan Projects 0 73 8 0 81 US$, million 0 28,159 3,005 0 31,164 Sri Lanka Projects 1 75 3 0 79 US$, million 81 2,244 1,062 0 3,387 Total Projects 16 (3) 630 (3) 564 (29) 22 1,232 (35) US$, million 1,764 200,919 123,605 1,670 327,988 Source: Private Participation in Infrastructure database. Note: The number of canceled projects appears in parentheses. Investment totals are in 2019 US$, million. ICT = information and communications technology; PPP = public-private partnerships. P UBLI C - P RIV A TE P A RTNERS H I P S IN SOUT H A SI A    49 Annex 1D. Imputing the Missing Contract Period Values for Predictions Four kind of projects have been identified as Physical Investment missing contract period information: energy projects, airport projects, seaport projects, The missing values for investments in physi- and toll road projects in India. The missing cal infrastructure for small hydro projects in contract period data were imputed using the Nepal and Sri Lanka and for the wind and following regressions and were rounded to solar energy projects in India were imputed the nearest integer. The regression for energy by estimating the amount of investment projects in South Asia is needed per megawatt for the same type of energy projects in the same country. Hence, the imputations are obtained from the fol- β0 + Typei + Countryi Periodi =  lowing regression: + Financial Closure Yeari + ui , where i represents each energy project in Physical Investmenti = β × Capacityi + ui, South Asia. A total of 391 observations were used to impute 194 missing values. where i stands for project; β is the regression The regression for airport projects in South coefficient; and ui is the residual. Asia is •  In the case of Sri Lankan small hydro proj- Periodi = β0 + ui , ects, 41 observations were used to impute 3 missing values. where i represents each airport project in •  In the case of Nepalese small hydro proj- South Asia. This regression essentially finds ects, 22 observations were used to impute the average contract length for airport proj- 4 missing values. ects without any predictors. The choice of the •  In the case of Indian wind projects, model is due to sample limitations. Eight 91 observation were used to impute 4 observations were used to impute three miss- missing values. ing values. •  In the case of Indian solar projects, 97 obser- The regression for seaport projects in vations were used to impute 1 missing value. South Asia is β0 + Typei + Financial Periodi =  Debt-to-Physical-Investment Ratio Closure Yeari + ui , To obtain the estimates for the missing financing variables—debt and equity—the where i represents each toll road project in debt-to-physical-investment ratio has been India. A total of 53 observations were used to predicted using type, sector, country, and impute one missing value. financial closure year dummies, using the The regression for toll road projects in following regression: India is    Debt to Physical Investment Ratioi β0 + Typei + Financial Closure Periodi =        = β0 + Typei + Sectori + Countryi + Yeari + Subnationali + ui ,       Financial Closure Yeari + ui , where i represents each toll road project in where i represents each project in South India. The inclusion of the subnational con- Asia. A total of 737 observations have been tract indicator is due to the differences in the used to impute 344 missing values. The handling of contracts between the National physical investment has been apportioned Highways Authority of India and the state according to the predicted ratio between highways authorities. A total of 393 observa- debt and equity. tions were used to impute 14 missing values. 50   H IDDEN DEBT Annex 1E. Model Selection baseline hazard function does not differ dras- tically compared to the non-parametric esti- The model selection in the context of survival mation of the overall hazard presented in the analysis is to choose the correct parametric main chapter. However, observe that the base- model for the purposes of the study. This line is scaled down to some extent. The para- study estimates a nontraditional model—a metric model chosen needs to be able to flexible parametric model—rather than one of reasonably replicate the major characteristics the usual out-of-the-box models using distri- of these curves. butions such as the Weibull or the loglogistic Two of the common specifications for the models. The main reason is that the regular baseline hazard are Weibull and loglogistic parametric models fail to fit the pattern of the functions. These models are known to have baseline hazard function in this study. The proportional hazards properties and propor- baseline hazard is defined as the hazard pro- tional odds properties, respectively. When the file of the project for which all the dummies model is fit using Weibull and loglogistic haz- are at their base cases, all shocks are set to ard functions, the resulting estimates of the zero, and all other continuous variables are baseline hazard function are monotonically set to their means. increasing, neither of which captures the uni- To get an idea about the baseline hazard modal concave aspect of the baseline hazard profile, a semi-parametric Cox model can be captured using the semi-parametric methods fit, and the baseline hazard can be extracted (see figure 1E.2). This result is potentially due as a residual. Alternatively, the total contract to the fact that the sample overrepresents period of the project can be partitioned based projects in the early stages of their contract on the centiles of distress and—assuming that periods, and the models, in their attempt to fit hazard is constant within each partition—a the data, underpenalize the lack of fit in the step function that would indicate the hazard later stages. profile over the contract period of a project. Royston and Parmar (2002) provide gener- Figure 1E.1 shows the resulting baseline haz- alizations of the parametric survival models ard profile using the two methods. using Weibull and loglogistic functions. These Figure 1E.1 shows that the baseline hazard models partition the analysis period must be increasing and then decreasing into multiple periods and, using restricted around 15 percent to 30 percent of the splines, make the relationship between the contract period. The overall shape of the strict functional forms of Weibull and FIGURE 1E.1  Baseline Hazard Profile Estimates Using Semi-parametric Methods a. Using Cox regression b. Assuming constant hazard within centiles of distress 0.0007 0.0007 0.0006 0.0006 0.0005 0.0005 Hazard Hazard 0.0004 0.0004 0.0003 0.0003 0.0002 0.0002 0.0001 0.0001 0 20 40 60 80 100 0 20 40 60 80 100 Percent of contract period Percent of contract period Source: Herrera Dappe, Melecky, and Turkgulu 2020. Source: Herrera Dappe, Melecky, and Turkgulu 2020. Note: Using Gaussian kernel and optimal bandwith. Note: Using 11 interior knots. P UBLI C - P RIV A TE P A RTNERS H I P S IN SOUT H A SI A    51 loglogistic functions and the analysis period parametric approach. Observe that the rela- more flexible. The underlying distributions are tionship can still be improved. However, no longer Weibull or loglogistic, but their pro- this quest may result in overfitting the sam- portionality properties are preserved. Hence, ple. To this end, flexible parametric propor- the flexible models are characterized by the tional hazards (PH) and proportional odds proportionality property and the degree of free- (PO) models with different degrees of free- dom that they provide. Figure 1E.3 presents the dom have been fit, and the Akaike informa- baseline hazard functions using the flexible tion criterion (AIC) and the Bayesian parametric methods with only a single interior information criterion (BIC) are used to knot, which means that two parameters are select the appropriate model. Table 1E.1 used in characterizing the baseline. Even intro- shows the resulting AIC and BIC values. ducing only one interior knot, the model cap- Observe that the PH(2) model yields the tures the unimodal relationship implied by the lowest AIC and BIC, so the analysis overall semi-parametric methods in figure 1E.1. uses PH(2) model. The choice of the model More interior knots can be introduced to does not have substantial qualitative or obtain a better fit using the flexible quantitative effects on the estimates. FIGURE 1E.2  Baseline Hazard Profile Estimates Using Parametric Methods a. Weibull b. Loglogistic 0.0007 0.0007 0.0006 0.0006 0.0005 0.0005 Hazard Hazard 0.0004 0.0004 0.0003 0.0003 0.0002 0.0002 0 20 40 60 80 100 0 20 40 60 80 100 Percent of contract period Percent of contract period Source: Herrera Dappe, Melecky, and Turkgulu 2020. FIGURE 1E.3  Baseline Hazard Profile Estimates Using Flexible Parametric Methods a. Proportional hazards: PH(2) b. Proportional odds: PO(2) 0.0007 0.0007 0.0006 0.0006 0.0005 0.0005 0.0004 0.0004 Hazard Hazard 0.0003 0.0003 0.0002 0.0002 0.0001 0.0001 0 0 0 20 40 60 80 100 0 20 40 60 80 100 Percent of contract period Percent of contract period Source: Herrera Dappe, Melecky, and Turkgulu 2020. Note: The analysis uses one interior knot. 52   H IDDEN DEBT TABLE 1E.1  Akaike Information Criterion and Bayesian Information Criterion under Different Orders of Flexible Parametric Methods PH(1) PO(1) PH(2) PO(2) PH(3) PO(3) PH(4) PO(4) PH(5) PO(5) AIC 1,504.0 1,510.7 1,483.4 1,492.6 1,485.3 1,494.5 1,489.8 1,497.1 1,490.6 1,499.8 BIC 1,578.8 1,585.6 1,561.4 1,570.6 1,566.3 1,575.5 1,580.2 1,584.4 1,584.1 1,593.3 Source: World Bank staff calculations. Note: The number in parentheses refers to the model number. The bold PH(2) column marks the order at which AIC and BIC are minimized. AIC = Akaike information criterion; BIC = Bayesian information criterion; PH = proportional hazards; PO = proportional odds. Annex 1F. Estimation Tables TABLE 1F.1  Estimated Fiscal Costs (at the 99th Percentile) from Early Termination of the Public-Private Partnership Portfolio in South Asia, 2020–24 (2019 US$, million) From the beginning of 2020 to the end of: Number Country/ Contract of Scenario period 2020 2021 2022 2023 2024 projects Afghanistan Low 0.51 0.06 0.11 0.16 0.21 0.25 2 Medium 1.16 0.14 0.26 0.38 0.47 0.56 High 1.42 0.17 0.32 0.46 0.58 0.69 Bangladesh Low 379 49 92 130 162 191 30 Medium 643 79 151 214 268 316 High 730 89 170 241 302 356 Bhutan Low 4.11 0.35 0.69 1.01 1.32 1.62 1 Medium 5.32 0.45 0.89 1.31 1.71 2.10 High 5.59 0.48 0.93 1.37 1.80 2.21 India Low 9,663 853 1,656 2,409 3,110 3,766 848 Medium 16,371 1,444 2,803 4,076 5,263 6,371 High 18,510 1,631 3,166 4,605 5,945 7,198 Nepal Low 112 9 17 24 30 36 35 Medium 178 13 26 37 47 56 High 197 15 29 41 52 62 Pakistan Low 1,046 96 189 277 359 434 72 Medium 1,746 159 314 460 595 720 High 1,959 178 352 515 667 806 Sri Lanka Low 58 6 12 17 22 26 66 Medium 95 10 19 28 36 44 High 107 11 22 32 41 49 Source: Herrera Dappe, Melecky, and Turkgulu 2020. Note: The 99th percentile refers to the confidence with which the calculated loss is not exceeded during a given period. PPPs = public-private partnerships. P UBLI C - P RIV A TE P A RTNERS H I P S IN SOUT H A SI A    53 TABLE 1F.2  Logit Regression Estimates of Likelihood of Cancellation of Indian National Highway Public-Private Partnerships (1) (2) (3) (4) (5) (6) (7) (8) (9) Base case: Annuity Capital grant 0.68 0.58 1.16 (0.82) (0.83) (1.37) Premium 1.40* 1.18 1.95+ (0.812) (0.84) (1.20) NPV of payments 0.003** 0.002* 0.002* to government (0.001) (0.001) (0.001) NPV of payments 0.74** 0.54+ 0.92* to government as (0.36) (0.37) (0.50) share of project investment Road length (log) 1.20** 2.219*** 0.978* 1.93** 1.14** 2.10** (0.49) (0.823) (0.53) (0.90) (0.50) (0.83) Debt financing 9.302* 8.98* 8.923* share of project (5.271) (4.62) (4.73) investment Constant −2.20*** −7.68*** −20.54*** −1.42*** −5.93** −17.62*** −1.31*** −6.65*** −18.39*** (0.75) (2.43) (6.06) (0.24) (2.48) (5.66) (0.23) (2.37) (5.48) Observations 123 123 109 123 123 109 123 123 109 Source: Herrera Dappe, Melecky, and Turkgulu 2020. Note: To assess which contract features could be associated with cancellations, the following equation is estimated using the maximum likelihood method: Canceledi = f( β0+Xi,projβ1 + ui ), (1F.1) where the dependent variable is the event of cancellation, f(·) is the logistic function, and X i, proj are project-specific variables. The 99th percentile refers to the confidence level with which the computed maximum loss is not exceeded during the reference period (such as one year or the duration of the project). NPV = net present value. Standard errors in parentheses. + p < 0.15, * p < 0.1, ** p < 0.05, *** p < 0.01. year is not an enduring and relational part- Notes nership and therefore not a PPP.   1. Fay and others (2019) estimate that infra-   4. The incentive to adopt improvements structure spending in South Asia was 4.5 requires that either the service quality can be percent of GDP in 2011. specified by contract or the demand for the   2. In some cases, companies in developing service depends on its quality. If quality is not countries are able to borrow at lower costs contractible, PPPs may lead to innovations than the government. Such companies tend during the design stage that reduce both costs to have substantial export earnings and/or a and quality. This might lead to the provision close relationship with either a foreign firm of some service with the least cost but not at or the home government (Durbin and Ng socially optimal levels if the social benefits 2005; Grandes, Panigo, and Pasquini 2017). from the innovations are larger than the net   3. For example, a government buying goods or savings incurred by the firm (Hart 2003; services from the same supplier year after Martimort and Pouyet 2008). 54   H IDDEN DEBT   5. The  agency problem  is a conflict of interest 15. Data limitations preclude estimation using inherent in any relationship in which one only the South Asian countries. party is expected to act in another’s best 16. Merchant greenfield projects in the PPI data interests. set are projects in which the private sponsor   6. Mobilizing private finance may also resolve a constructs and operates a new facility at its political problem hindering investment in own risk in a liberalized market. A rental infrastructure by the incumbent provider. For greenfield project in the PPI data set refers to example, when fiscal rules prevent public projects in which the government agrees to financing of infrastructure, PPPs might still rent the facility and the service from a private be politically and legally feasible by shifting party as a temporary measure. liabilities off the central government’s bal- 17. Equations (1A.4) and (1A.5) in annex 1A pres- ance sheet and/or into the future (Budina, ent the relationships among hazard, cumulative Polackova Brixi, and Irwin 2007). Another hazard, and cumulative survival probability possibility is that when a different political presented in figures 1.5, 1.6, and 1.7. faction can bar public financing of infrastruc- 18. In a greenfield project, a new facility or asset ture within an incumbent’s district, the is built and operated for the period specified incumbent can utilize private finance to skirt in the project contract. In a brownfield proj- the political constraint. ect, instead of building a new asset, the pri-   7. Availability payments are periodic pay- vate entity takes over an existing asset and ments to the sponsor conditional on the usually makes an improvement to it or availability of service at a prespecified level. expands it. Capacity payments are periodic payments 19. This figure of 1,056 active projects excludes from the government to the sponsor for the projects whose imputed or observed con- upholding a certain level of capacity. tract period has elapsed. Shadow tolls are payments per user to the 20. Only 209 of the 1,970 PPP projects included sponsor from the government. Power pur- in the data set, which spans 1983 to 2017, chasing agreements are contractual prom- are from emerging market economies and ises of prespecified levels of energy developing countries (Moody’s Investors purchases by the government. Service 2019).   8. PPPs typically do not create direct implicit 21. Figure 1.6 depicts a decreasing hazard profile liabilities unless payments to the private for projects past 20 percent of their contract party are expected to continue due to non- periods. contractual or noncontingent reasons, such 22. Percentages are as of 2017, from World Revenue as continued political relationship even after Longitudinal Data (WoRLD), ­ available at http:// the contract is over (Budina, Polackova Brixi, data.imf.org/?sk=77413F1D-1525-450A​ and Irwin 2007). -A23A-47AEED40FE78.   9. All dollar figures in the study are expressed in 23. The analysis assumes that the annual govern- 2019 US dollars inflated using the US ment revenues during the 2020–24 period Consumer Price Index (CPI) series in the are the same as in 2019. World Development Indicators. 24. The statistics exclude two episodes in 10. The analysis used Version 2019 H1, available Ecuador because the adjusted currency con- at https://ppi.worldbank.org/en/ppidata. version factor remained stable, even though 11. Financial closure refers to the securing of the official rate of depreciation was 56.8 financing for the project by the sponsor. percent and 77.2 percent, in the banking cri- 12. The study uses Version 2019 H1, available at ses of early 1980s and the late 1990s, https://ppi.worldbank.org/en/ppidata. respectively. 13. Polity IV Project, Political Regime 25. The scenario abstracts from the sequencing Characteristics and Transitions, 1800–2018, of different macro events because all events is available at http://www.systemicpeace.org​ are assumed to occur in the same year. /­inscrdata.html. 26. In the chosen time frame, only two projects 14. The World Development Indicators are avail� - did not adhere to these two schemes. The able at http://datatopics.worldbank.org/world​ first is a project that was awarded based on -development​-indicators/. payment of a percentage of toll revenues. P UBLI C - P RIV A TE P A RTNERS H I P S IN SOUT H A SI A    55 The  second is a hybrid annuity project that Spreads.” Journal of International Money and entitled the project company to both avail- Finance 24 (4): 631–49. ability payments and a capital grant. The Engel, E., R. Fischer, and A. Galetovic. 2013. “The hybrid annuity model has become the pre- Basic Public Finance of Public-Private ferred method more recently. Both projects Partnerships.” Journal of the European were excluded from the analysis. Economic Association 11 (1): 83–111. 27. An annual discount rate of 8 percent has EPEC (European PPP Expertise Centre). 2013. been assumed. “Termination and Force Majeure Provisions in 28. The model is estimated in the Stata Journal PPP Contracts: Review of Current European using the stpm2 routine, which has been Practice and Guidance.” EPEC, Luxembourg. authored by Lambert and Royston (2009). Fay, M., H. I. Lee, M. Mastruzzi, S. Han, and 29. The delta method provides a way to estimate M. Cho. 2019. “Hitting the Trillion Mark: standard errors for nonlinear predictions. The A Look at How Much Countries Are Spending method is based on a central limit theorem on Infrastructure.” Policy Research Working that shows that the prediction is distributed Paper 8730, World Bank, Washington, DC. with a standard normal distribution under Grandes, M., D. T. Panigo, and R. A. Pasquini. certain conditions. The method calculates the 2017. “Corporate Credit Spreads and the standard errors using the variance-covariance Sovereign Ceiling in Latin America.” Emerging matrix of the coefficient estimates. Markets Finance and Trade 53 (5): 1217–40. Grimsey, D., and M. K. Lewis. 2017. Global Developments in Public Infrastructure References Procurement: Evaluating Public-Private ADB (Asian Development Bank). 2017. Meeting Partnerships and Other Procurement Options. Asia’s Infrastructure Needs. Manila: ADB. Cheltenham, United Kingdom: Edward Elgar. ADB, DFID, JICA, and World Bank (Asian Grout, P. A. 1997. “The Economics of the Private Development Bank, Department for Finance Initiative.” Oxford Review of International Development, Japan International Economic Policy 13 (4): 53–66. Cooperation Agency, and the World Bank). Guasch, J. L. 2004. Granting and Renegotiating 2018. The WEB of Transport Corridors in Infrastructure Concessions: Doing It Right. South Asia. Washington, DC: World Bank. WBI Development Studies. Washington, DC: Budina, N., H. Polackova Brixi, and T. Irwin. World Bank. 2007. “Public-Private Partnerships in the New Hamilton, J. D. 2018. “Why You Should Never EU Member States: Managing Fiscal Risks.” Use the Hodrick-Prescott Filter.” Review of Working Paper, World Bank, Washington, DC. Economics and Statistics 100 (5): 831–43. de Bettignies, J-E, and T. W. Ross. 2009. “Public– Hart, O. 2003. “Incomplete Contracts and Public Private Partnerships and the Privatization of Ownership: Remarks, and an Application to Financing: An Incomplete Contracts Public-Private Partnerships.” Economic Journal Approach.” International Journal of Industrial 113 (486): C69–C76. Organization 27 (3): 358–68. Heald, D., and G. Georgiou. 2010. “Accounting de Bettignies, J-E, and T. W. Ross. 2010. “The for PPPs in a Converging World.” In Economics of Public-Private Partnership: Some International Handbook of Public-Private Theoretical Contributions.” In International Partnerships , edited by Graeme A. Hodge, Handbook on Public-Private Partnerships , Carsten Greve, and Anthony E. Boardman, edited by Greame A. Hodge, Carsten Greve, 237–61. Cheltenham, United Kingdom: and Anthony E. Boardman, 132–58. Edward Elgar. Cheltenham, United Kingdom: Edward Elgar. Herrera Dappe, M., M. Melecky, and B. Turkgulu. de Vries, P. 2013. “Public Budget Norms and PPP: 2020. “PPP Distress and Fiscal Contingent An Anomaly.” In The Routledge Companion to Liabilities in South Asia.” Background Public-Private Partnerships, edited by Piet de paper for Hidden Debt . World Bank, Vries and Etienne B. Yehoue, 301–21. Oxon, Washington, DC. United Kingdom: Routledge. Iossa, E., and D. Martimort. 2012. “Risk Allocation Durbin, E., and D. Ng. 2005. “The Sovereign and the Cost of Public-Private Partnerships.” Ceiling and Emerging Market Corporate Bond RAND Journal of Economics 43 (3): 442–74. 56   H IDDEN DEBT Iossa, E., and D. Martimort. 2015. “The Simple Research Working Paper 1989, World Bank, Microeconomics of Public-Private Washington, DC. Partnerships.” Journal of Public Economic Pratap, K. V., and R. Chakrabarti. 2017. Public- Theory 17 (1): 4–48. Private Partnerships in Infrastructure: Irwin, T. C. 2007. Government Guarantees: Managing the Challenges. Singapore: Allocating and Valuing Risk in Privately Springer. Financed Infrastructure Projects. Directions in Reyes-Tagle, G., and K. Garbacik. 2016. Development: Infrastructure. Washington, DC: “Policymakers’ Decisions on Public-Private World Bank. Partnership Use: The Role of Institutions and ITF (International Transport Forum). 2018. Private Fiscal Constraints.” IDB Technical Note no. Investment in Transport Infrastructure: Dealing 1169, Inter-American Development Bank with Uncertainty in Contracts. Paris: ITF. Washington, DC. Laeven, L., and F. Valencia. 2018. “Systematic Royston, P., and P. C. Lambert. 2011. Flexible Banking Crises Revisited.” IMF Working Paper Parametric Survival Analysis Using Stata: WP/18/206, International Monetary Fund, Beyond the Cox Model. College Station, TX: Washington, DC. Stata Press. Lambert, P. C., and P. Royston. 2009. “Further Royston, P., and M. K. B. Parmar. 2002. “Flexible Development of Flexible Parametric Models for Parametric Proportional-Hazards and Survival Analysis.” Stata Journal 9 (2): 265–90. Proportional-Odds Models for Censored Marin, P. 2009. Public-Private Partnerships for Survival Data, with Application to Prognostic Urban Water Utilities: A Review of Experiences Modelling and Estimation of Treatment Effects.” in Developing Countries. Washington, DC: Statistics in Medicine 21 (15): 2175–97. World Bank. Rozenberg, J., and M. Fay, eds. 2019. Beyond Martimort, D., and J. Pouyet. 2008. “To Build or the Gap: How Countries Can Afford the Not to Build: Normative and Positive Theories Infrastructure They Need While Protecting the of Public–Private Partnerships.” International Planet. Washington, DC: World Bank. Journal of Industrial Organization 26 (2): World Bank. 2019. Guidance on PPP Contractual 393–411. Provisions. Washington, DC: World Bank. Moody’s Investors Service. 2019. “Default and Yehoue, E. B. 2013. “Financial and Sovereign Debt Recovery Rates for Project Finance Bank Loans, Crises and PPP Market Structure.” In The 1983–2017.” Moody’s Investors Service, Routledge Companion to Public-Private New York. Partnerships, edited by Piet de Vries and Etienne Polackova, H. 1998. “Contingent Government B. Yehoue, 349–70. Oxon, United Kingdom: Liabilities: A Hidden Risk for Stability.” Policy Routledge. State-Owned Banks versus Private Banks in South Asia: 2 Agency Tensions, Susceptibility to Distress, and the Fiscal and Economic Costs of Distress T he pros and cons of state banking are which state-owned and private banks in dis- vigorously debated. State-owned tress use these channels. The study examines commercial banks (SOCBs) can be how firms’ links with SOCBs versus private established to help create markets and fulfill banks affect their investment—focusing on social goals and support fiscal policy (by small and medium enterprises (SMEs) and raising additional revenues for public invest- successful firms with high growth of sales. ment), but the operation of SOCBs has South Asian economies rely on their bank- downsides because of possible inefficiencies, ing system to help allocate resources for misuse, and financial distress. The upsides greater productivity and employment as well and downsides of SOCBs are increasingly as financial inclusion for greater access to being scrutinized by policy makers and the opportunities. These functions cannot be per- global community. formed effectively when banks are distressed. This chapter contributes to the debate by Therefore, understanding the drivers of dis- examining episodes of distress at SOCBs and tress is important for devising and implement- private banks, the drivers of distress, bank ing effective policy remedies for SOCBs. The adjustments in times of distress, and the costs chapter therefore concludes with recommen- of bank distress to the real economy in dations to strengthen SOCBs and the finan- Bangladesh, India, Pakistan, and Sri Lanka. cial sector for the benefit of economies and Distinguishing banks by ownership type is societies overall. important because state commercial banking is prevalent in these countries, and in South Asia overall. The analysis identifies episodes The Upsides and Downsides of of bank distress, explores several adjustment State-Owned Commercial Banks channels through which banks cope with dis- By 2017, three countries stood out in the tress, and examines the relative intensity with world because of the dominance of SOCBs in Note: This chapter draws on the background research paper: Kibuuka, K., and M. Melecky. 2020. “State-Owned versus Private Banks in South Asia: Agency Tensions, Distress Factors, and Real Costs of Distress.” Background paper for Hidden Debt. World Bank. Washington, DC. 57 58   H IDDEN DEBT their banking systems: Belarus, Iceland, and of i­nterest that bureaucrats/technocrats India. In each country, SOCBs held close to can have when tasked with managing govern- 70 percent of total banking ­ system assets ment-owned banks. The conflict is between (Cull, Martinez Peria, and Verrier 2017).1 the government’s interest in maximizing social Across South Asia, SOCBs hold a very high welfare and the bureaucrats’ or technocrats’ share of total banking system assets—about interest in extracting benefits. This conflict 40 percent, on average. However, the use of gives rise to red tape, operational inefficien- SOCBs is not confined to developing econo- cies, and misallocation of resources (Banerjee mies. In Germany, for instance, the share also 1997; Hart, Shleifer, and Vishny 1997). hovers above 40 percent. Politicians can misuse SOCBs to pursue their On the upside, using commercial and own interests, such as reelection and personal hybrid SOCBs can reflect state efforts to profit, by pushing to finance their supporters address market failures and create positive or those willing to pay the highest bribes. externalities (Atkinson and Stiglitz 1980; This misuse induces resource misallocation Stiglitz 1993; Cull, Martinez Peria, and and economic inefficiency (Shleifer and Verrier 2017).2 Specifically, the state could Vishny 1994; Shleifer 1998). Politicians are use the SOCBs to (1) promote competition, more likely to favor government bank owner- extend the reach of service delivery, and spur ship when public accountability and judicial new markets in the financial sector (Cull, independence are low because they can Martinez Peria, and Verrier 2017; Ferrari, extract more benefits with fewer personal Mare, and Skamnelos 2017; Mazzucato and consequences, Perotti and Vorage (2010) Penna 2016); (2) help resolve coordination suggest. failures (de la Torre, Gozzi, and Schmukler The upsides and downsides of using 2007); and (3) play countercyclical and safe- SOCBs create tensions in practice. For haven roles in crises after markets have failed instance, when SOCBs allocate credit ineffi- to internalize individual contributions to sys- ciently, their countercyclical role can be temic risk (Micco and Panizza 2006; Bertay, uncertain (Bertay, Demirgüç-Kunt, and Demirgüç-Kunt, and Huizinga 2015; Duprey Huizinga 2015; Coleman and Feler 2015). To 2015). The state may use SOCBs to create address such tensions, several studies have positive externalities by (1) financing projects reviewed and proposed some good practices with high nonmonetary social returns that to improve SOCB operations (Gutierrez et al. have negative net present value (that is, their 2011; de la Torre, Gozzi, and Schmukler internal rate of return does not cross the pri- 2007).3 vate sector hurdle rate for investable projects) (Levy-Yeyati, Micco, and Panizza 2007), and Understanding the drivers of distress is (2) promoting strategically important indus- tries, jumpstarting economic development, important for devising and implementing helping create new markets and national effective policy remedies for state- champions, and providing a source of revenue owned commercial banks. for social investments (Gerschenkron 1962; Ferrari, Mare, and Skamnelos 2017). For a Data suggest that, in practice, SOCBs global overview of development, hybrid, and rarely have explicitly defined roles in terms of commercial state-owned banks with an addressing market failures or creating positive explicit social mandate and at least 30 percent externalities—at least in Europe and Central ownership stake by the state, see de Luna- Asia (Ferrari, Mare, and Skamnelos 2017). If Martinez and Vicente (2012). SOCBs have some social mandate, it may On the downside, using SOCBs involves change over time: for instance, when the risks of inefficiency and misallocation costs underlying market failure has been overcome due to agency problems and political misuse or when policy makers reweigh competing (Cull, Martinez Peria, and Verrier 2017). social priorities. Unlike government-owned The agency problem relates to the conflict development financial institutions (DFIs), ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    59 SOCBs most often operate without an explicit The methodology is described in annex 2A. social m­ andate—including in several South Annex 2B presents regression results for the Asian countries such as India and Bangladesh. probability of distress for banks and their Thus, economists increasingly worry about adjustments to distress. the downside of SOCB operations. However, relatively little research has explored the downside behavior of SOCBs in distress and The Omnipresence of State-Owned the costs of distress for central government Commercial Banks in South Asia and the economy. This chapter helps fill this The South Asia region has the highest share gap. of SOCB assets in terms of total banking Using bank-level and firm-level data for assets in the world, followed by the Europe India and bank-level data for Bangladesh, and Central Asia region (see ­ figure 2.1).4 In Pakistan, and Sri Lanka, this chapter identi- South Asia, the share of SOCB assets is par- fies episodes of distress at banks using a rule- ticularly high in India, Bhutan, and Sri Lanka of-thumb threshold for the interest rate when compared with the regional average. coverage ratio (ICR), which indicates whether Despite a fair share of SOCB assets, a bank has enough revenues to cover its inter- Bangladesh and Pakistan fall below the est expense. As a robustness check, the analy- regional average—but still above the average sis also includes three additional indicators of for the lower-middle-income country group. financial performance and soundness. In general in South Asia, SOCBs appear to Highlighting the role of bank ownership, this be performing poorly when compared with examination takes into consideration the fac- privately owned commercial banks (PCBs) tors behind bank distress and bank adjust- (figure 2.2). PCBs are better capitalized than ments in distress as well as the wider economic SOCBs within the same country. The same impact of ownership on investment in client goes for asset quality, profitability, and effi- firms. The latter analysis relies on bank–firm ciency measures (figure 2.2). Across coun- matched data from the Prowess database tries, Pakistan’s SOCBs are, on average, for India for the period 2009–18. performing better than SOCBs in Bangladesh FIGURE 2.1  South Asia: Share of State-Owned Commercial Bank Assets in Total Banking Assets, 2016 a. South Asia has the highest share of b. Within South Asia, SOCBs among world regions India leads in state banking 80 70 60 60 50 40 Percent Percent 40 30 20 20 10 0 0 EAP ECA HIC LAC MENA SAR SSA India Bhutan Sri Lanka Bangladesh Pakistan SOCB asset share South Asia average LMIC average Source: World Bank Survey on Banking Supervision. Note: EAP = East Asia and Pacific; ECA = Europe and Central Asia; HIC = high-income countries; LAC = Latin America and the Caribbean; LMIC = lower-middle-income countries; MENA = Middle East and North Africa; SAR = South Asia; SSA = Sub-Saharan Africa; SOCBs = state-owned commercial banks. 60   H IDDEN DEBT FIGURE 2.2  Bangladesh, India, and Pakistan: State-Owned Commercial Banks’ Underperformance Relative to Domestic and Foreign Private Banks, 2009–18 Average a. Bangladesh b. India Cost to income Cost to income Net interest margin Net interest margin Return on equity Return on equity Gross NPL ratio Gross NPL ratio Capital to RWA Capital to RWA –15 –10 –5 0 5 10 15 20 25 30 –25 –20 –15 –10 –5 0 5 10 15 20 25 30 Percent Percent c. Pakistan Cost to income Net interest margin Return on equity Gross NPL ratio Capital to RWA 0 5 10 15 20 25 30 35 Percent SOCBs Foreign PCBs Domestic PCBs Source: World Bank staff calculations using Fitch Connect data. Note: Cost to income rescaled by dividing by 10. NPLs = nonperforming loans; PCBs = privately owned commercial banks; RWA = risk-weighted assets; SOCBs = state-owned ­commercial banks. and India. The average capital adequacy ratio ­ egative profitability. The performance of n is well above the 10 percent national pruden- India’s SOCBs is mixed: capital adequacy tial threshold. Even though elevated at ratios are well above the 9 percent national 12 percent, Pakistan’s average nonperforming prudential minimum and the Basel mini- loan (NPL) ratio is the lowest compared with mum, but NPL ratios hover at a worrying Bangladesh and India. SOCBs in Pakistan are, 17 percent. Despite the lowest cost- on average, the only profitable SOCBs in the t o-income ratio, profitability is strongly ​ comparison group. They have the lowest cost- negative—­ particularly the return to equity.5 to-income ratios, which are on par with more operationally efficient PCBs in South Asia. Meanwhile, SOCBs in Bangladesh appear Bank Business Models by to have the weakest performance indicators: Ownership Type: The Example capital ratios below the 10 percent national of India prudential threshold and the Basel mini- As shown in the previous section, bank per- mum; the highest NPL ratios (28 percent) formance and operations vary greatly by type among the South Asian peer group; the high- of ownership. To explore this further, this sec- est cost-to-income ratios and, as a result, tion focuses on India because more detailed ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    61 bank data are readily available for it, and Given their large branch network, PSBs bank characteristics can be linked to firm- can mobilize large amounts of retail deposits, level investment data—a real outcome vari- which comprise the largest component of PSB able of interest for this chapter. funding (figure 2.4). Loan-to-deposit ratios The banking system assets of India’s sched- are higher in other banks compared with uled commercial banks (SCBs) amounted to PSBs, further reflecting their ability to mobi- about 80 percent of GDP in 2018.6 SOCBs lize greater amounts of deposits. Other banks (called public sector banks, PSBs, in India) must rely more on costlier modes of raising dominate the banking sector in terms of funds. For instance, SFBs rely largely on assets, credit, and branches. PSBs hold lines of credit to fund their lending activities 66 percent of total SCB assets, while domesti- figure 2.4). Unlike other PSBs, SBI, as India’s (­ cally owned private banks (PVTBs) have largest bank and a government corporation about 28 percent; foreign commercial banks statutory body, can readily raise funds outside (FBs) hold about 6 percent; and small finance of India by borrowing from international banks (SFBs) control a minimal 0.3 percent.7 global markets. Thus, total SBI borrowings In terms of credit, PSBs control about (10 percent of total liabilities) are higher than 63 percent of total banking credit, PVTBs the borrowings of other PSBs (7 percent of control about 29 percent, and other SCBs total liabilities). Meanwhile, leverage, as mea- represent about 8 percent of total banking sured by the tier 1 capital-to-total-assets ratio, credit. By the end of 2018, PSBs operated is above the prudential minimum of 4 percent 92,362 branches across India, three times for systemically important banks and 3.5 more than the domestic and foreign private percent for other banks—stricter limits than banks combined. The largest commercial the Basel minimum of 3 percent. Leverage is bank by far is the State Bank of India (SBI), less than 6 percent for PSBs (at 5.6 percent for which controls 23 percent of total banking SBI and 5.1 for other PSBs) and above assets and 20 percent of total banking credit, 10 percent for other banks, indicating that and operates the largest branch network, with PSBs are more leveraged than other banks. more than 23,382 branches and a dominant As noted, SOCBs do not tend to have rural presence (figure 2.3). explicit mandates to address market failures FIGURE 2.3 India: Branch Networks and Total Credit, 2018 a. SOCBs have much more extensive branch networks b. SOCBs extend most of the credit volume Number of branches Total credit outstanding in scheduled commercial banks SFBs 1% RRBs 3% FBs 3% SBI 20% SBI 23,382 PSBs Other PVTBs 63% PSBs 43% 29% Other PSBs 68,980 Source: Reserve Bank of India. Note: FBs = foreign commercial banks; PSBs = public sector banks; PVTBs = domestically owned private banks; RRBs = regional rural banks; SBI = State Bank of India; SFBs = small finance banks; SOCBs = state-owned commercial banks. 62   H IDDEN DEBT or create positive externalities. Data on the trend has been declining profitability. This is sectoral allocation of credit and lending to partially explained by lower levels of effi- typically underserved segments (such as small ciency and rising costs and expenses (includ- borrowers) and priority sectors (as identified ing staff costs and expenses), as well as rising by the Reserve Bank of India, RBI) show that nonperforming loans (figure 2.4). PSBs do not focus on lending to these groups In 2015, following the RBI’s accelerated or sectors more than private banks.8 In fact, efforts to ensure that losses expected from most PSB credit goes to large borrowers and distressed debt were adequately recognized to the industry sector, a nonpriority sector. and provisioned, an asset quality review was However, given their size, PSBs provide the conducted. It revealed a higher level of NPLs largest absolute volume of lending to small than previously reported across the banking borrowers. PSBs tend to lend much more to sector—most notably at the SOCBs. Many of public sector entities compared with other these NPLs were attributed to infrastructure banks, even though this lending comprises projects that had turned sour and accrued less than 10 percent of total loans (figure 2.4). during a period when PSBs benefitted from Smaller banks—namely, SFBs and regional regulatory forbearance. During 2019, NPLs rural banks (RRBs)—do target priority sec- remain the highest in PSBs in the Indian tors and small borrowers. More than ­ banking s ­ ystem—at about 10 percent in SBI 40 percent of their total credit is devoted to and more than 17 percent in other PSBs— these segments. while the ratio is on average less than Overall, banks earn most of their income 4 percent for other banks. Since the discovery from their lending activities. Foreign commer- of high levels of NPLs, many banks have cial banks tend to earn more through invest- worked hard to write off and resolve out- ments, as well as from fee-based and foreign standing problem assets. However, legal exchange services. PSB business models tend delays, inadequate infrastructure, and a large to be more traditional, focusing on earning pipeline of insolvency cases have stretched income through government securities and out and will c ­ ontinue to lengthen resolution similar investments (30 percent of total timelines. In response to these legal bottle- income) and lending (more than 50 percent of necks, the g ­ overnment in July 2019 increased total income). Unlike most PSBs, SBI earns the resolution time frame to 330 days, from almost 10 percent of its income from fee- the p­ reviously stipulated 270 days. based services—compared with an average of In addition to declining asset quality, capi- 3 percent earned by other PSBs. Because of its tal positions have been weak within PSBs cheap source of borrowing, SBI has a higher and have affected their lending capacity net interest margin (2.4 percent) than other (­figure 2.5). Even before the NPL finding, PSBs (2.0 percent). By contrast, net interest capital buffers were low, and the government margin indicators for all other banks exceed developed a public-private recapitalization 2.5 percent (reaching 6.7 percent for SFBs). response—the Indradhanush plan— Other efficiency indicators show that PSBs announced in August 2015. Given the limited tend to be less efficient, devoting a higher private participation in this plan, government share of their wage bill to intermediation ownership in PSBs has increased because of costs (figure 2.4). This could imply overcom- the state capital injections to prop up these pensation of the management or overemploy- banks. However, following the NPL d ­ iscovery, ment. The literature suggests the latter PSBs’ capital positions deteriorated again as (Kumbhakar and Sarkar 2003). provisioning increased substantially with the In recent years, declining profitability has need to adequately cover higher NPLs. resulted in negative returns on capital and Making matters worse, the introduction of assets. Although some PSBs still have positive Basel III starting in 2019 and the subsequent profit indicators—return on assets (ROA) phase-in of implementation has led to higher and return on equity (ROE)—the general prudential capital requirements. For the 2020 ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    63 FIGURE 2.4 India: Selected Funding and Credit Indicators, 2018 a. PSBs are mostly funded by retail deposits, b. The cost of funding is generally comprising more than 60 percent of total liabilities lower for PSBs than other banks 100 12 35 10 7 16 10 30 80 15 38 19 25 8 60 Percent Percent Percent 62 65 34 20 18 6 40 15 51 4 20 23 10 20 6 9 2 10 10 5 10 18 14 0 6 6 0 0 SBI PSBs PVTBs FBs SFBs SBI PSBs PVTBs FBs SFBs Equity Wholesale deposits Borrowings Cost of funds Cost of deposits Government deposits Retail deposits Other Cost of borrowings CRAR (right axis) c. PSBs have lower leverage and funding ratios d. PSBs do not appear to have a sectoral mandate, than other banks but most lending is to the industry sector 180 2 1 16 SBI 7 31 12 39 9 150 2 PSBs 10 16 16 38 7 9 3 12 120 PVTBs Percent Percent 7 25 9 30 10 11 4 4 90 2 8 FBs 14 12 42 17 9 5 60 0.5 4 RRBs 2 13 65 33 6 7 30 1 SFBs 6 20 10 19 16 11 17 0 0 SBI PSBs PVTBs FBs SFBs 0 20 40 60 80 100 Percent Leverage (tier 1 capital to total assets) Finance Personal loans Agriculture Cash to deposit Industry Services Trade Loan to deposit (right axis) Transport Other e. PSBs mostly lend to the nonpriority sector, but f. Less than 10 percent of total PSB credit is most credit to the public sector is from PSBs allocated to small borrowers 60 100 4,000 60 outstanding to priority sectors Percent of total loans Percent of total credit 80 3,000 40 40 Rs, billion 60 Percent 2,000 40 20 20 1,000 20 0 0 0 0 SBI PSBs PVTBs FBs SFBs SBI PSBs PVTBs FBs RRBs SFBs Priority sector Public sectors Total credit to small borrowers (left axis) Agriculture (right axis) MSMEs (right axis) Share of small borrower credit to total credit and substitute (right axis) Source: Reserve Bank of India. Note: CRAR = capital to risk-weighted assets ratio; FBs = foreign commercial banks; MSMEs = micro, small, and medium enterprises; PSBs = public sector banks; PVTBs = domestically owned private banks; RRBs = regional rural banks; Rs = Indian rupees; SBI = State Bank of India; SFBs = small finance banks. 64   H IDDEN DEBT FIGURE 2.5 India: Selected Business Model, Performance, and Soundness Indicators, 2018 a. Lending is the main income generator for PSBs, b. PSBs tend to earn less fee-based income and are less but investments contribute signi cantly e cient than most other banks 100 12 60 17 14 18 13 21 80 10 50 16 28 20 30 8 40 60 36 Percent Percent Percent 6 30 40 72 4 20 58 62 53 20 44 2 10 0 0 0 SBI PSBs PVTBS FBs SFBs SBI PSBs PVTBs FBs SFBs Loans Net interest margin Other interest income Fee-based Income to total income Non-interest income Wage bill/Intermediation cost (%, right axis) c. PSBs are struggling to make a pro t d. PSBs also have lower asset quality and capital bu ers 15 Gross NPA ratio Basel CRAR Bank group 10 (median) (median) 5 SBI 10.9 12.6 Return on assets (%) 0 PSBs 17.2 11.0 –5 PVTBs 3.6 14.9 –10 FBs 2.6 28.9 –15 SFBs 2.4 19.7 –20 –25 –30 –2.0 –1.5 –1.0 –0.5 0 0.5 1.0 1.5 Return on equity (%) SBI PSBs PVTBs FBs SFBs Source: Reserve Bank of India. Note: CRAR = capital to risk-weighted assets ratio; FBs = foreign commercial banks; NPA = nonperforming assets; PSBs = public sector banks; PVTBs = domestically owned private banks; SBI = State Bank of India; SFBs = small finance banks. financial year, the government budgeted for Economy. The database consists of data US$10 billion in PSB capital injections. reported by firms registered with the Registrar General of Companies. It is an unbalanced panel that covers the 1989 to 2018 period Data for Analysis of SOCBs in India and and contains detailed annual financial state- Other South Asian Countries ment data as well as performance information For further in-depth bank-level analysis in on firms in India—both financial and nonfi- relation to firms, we used an Indian firm-level nancial. In addition, specifically for banks, database called Prowess, which is managed Prowess provides an expanded set of financial by the Center for Monitoring the Indian soundness indicators. Prowess covers about ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    65 80 percent to 90 percent of SCBs in India, observations on SOCBs and nearly 700 obser- which account for most banking sector assets. vations on private banks. In the regression Using Prowess data, we constructed a bal- analyses, about 204 observations are used for anced, bank-level panel including key finan- Bangladesh, 243 for India, 153 for Pakistan, cial soundness indicators and bank and 86 for Sri Lanka. In pooled regressions, characteristics. As many of the key variables, about 686 were used for the four countries particularly financial soundness indicators, together. Figure 2.6 shows the pooled statis- were missing prior to 2009, the data set cov- tics for banks of South Asia’s four main econ- ers the 2009–18 period. For each year, there omies. It summarizes and contrasts the are 74 banks. However, not all banks have different business models and strategies of data for each of the key variables. 9 SOCB versus PCBs, using the mean across Table 2C.3, in annex 2C, presents the sum- banks for the 2009–18 period divided by mary statistics of key variables in this data set standard deviation. Detailed summary statis- for PSBs (panel a) and PVTBs (panel b). tics and tests of differences between SOCB For other South Asian countries, a detailed and PCB characteristics are reported pooled database that enables linking banks and firms and by country in annex 2C. is not readily available, to our knowledge. The data suggest that private banks take Therefore, we used the Fitch Connect data- more risks in lending than SOCBs: their ratio base to analyze banks in the four biggest of risk-weighted assets to total assets is economies of South Asia: Bangladesh, India, greater than those of SOCBs. Perhaps also for Pakistan, and Sri Lanka. Together, they this reason, they keep greater liquidity buffers account for more than 90 percent of South as self-insurance against greater risk. Private Asia’s GDP. We constructed a panel data set banks lend more from the deposits that for each country covering the 2009–18 period they mobilize and employ nondeposit fund- and most banks in the system. The panel data ing—including foreign savings. The greater set includes nearly 300 panel data interest margins and profitability of private FIGURE 2.6  South Asia’s Four Main Economies: Business Models and Strategies of State-Owned Commercial Banks versus Privately Owned Commercial Banks, 2009–18 Risk-weighted assets/Total assets (%) Liquid assets/Total assets (%) Loans/Deposits (%) z-score Net interest margin (%) Return on average equity (ROAE) (%) Return on average assets (ROAA) (%) Impaired loans (NPLs)/Gross loans (%) Total regulatory capital ratio (%) Interest coverage ratio (ICR) 0 0.30 0.60 0.90 1.20 1.50 Median statistics SOCBs PCBs Sources: World Bank staff calculations using Fitch Connect data. Note: Pooled data set for Bangladesh, India, Pakistan, and Sri Lanka. NPLs = nonperforming loans; PCBs = privately owned commercial banks; SOCBs = state-owned commercial banks. 66   H IDDEN DEBT banks may reflect returns commensurate with selection of borrowers (“reverse cherry-pick- their greater risk taking, as well as their ing”). For example, firms that maintain greater capacity to manage credit risk (lower exclusive relationships with government- median NPL ratios). Finally, the solvency owned banks can enjoy privileged borrowing standing (regulatory capital ratio) and two status with those banks: that is, the firms’ broader measures of distress based on balance sensitivity of investment to cash flows (their sheet solvency (z-score) and cash flows (ICR) financing constraints) is lower. But such suggest greater resilience of private banks to firms can be in worse financial condition risk. relative to other firms—for instance, be more Nevertheless, caution is warranted in inter- leveraged, invest less, be less profitable, and preting aggregate median numbers for the have worse growth prospects (Srinivasan region. The situation can vary considerably and Thampy 2017). Such adverse selection across countries and within countries across of borrowers can increase the default risk of individual banks. For instance, risk taking is government-owned banks because they lend lower than the regional median for SOCBs in to weaker firms on average. Pakistan and even more so for SOCBs in For this reason, we also controlled for Sri Lanka, with median ratios of risk- characteristics of client firms by linking banks weighted assets to total assets of 51 percent to firms. For each bank, we constructed a cli- and 47 percent, respectively, compared with ent firm portfolio and calculated average the regional median of 59 percent for SOCBs characteristics of this portfolio, such as aver- and 70 percent for private banks. In contrast, age firm size (total assets); leverage (debt-to- SOCBs in Bangladesh are the least profit- equity ratio); investment orientation able of the SOCBs in the region, based on the (investments to assets); and profitability median statistics for the return on (return on assets). Because of data availability equity—3.6 percent for Bangladesh compared constraints, we controlled for client firm char- with 9.3 percent for the regional average acteristics only for Indian banks. The main (median). Indian SOCBs appear to sit com- characteristics of the client firms of PVTBs fortably around the median for the region on and SOCBs are summarized in figure 2.7. all considered SOCB characteristics except Detailed summary statistics, along with statis- the ICR. Indian SOCBs, with a median ICR tical tests of difference in average characteris- of 0.57, fall well below the regional median tics, are reported in table 2C.6, in annex 2C. of 0.66. In none of the four major South The statistics suggest that client firms of Asian countries does the median ICR for SOCBs could be much less profitable, more SOCBs stand above 1. At the 75th percentile, leveraged, and perhaps ­ bigger—reflecting the the better performing SOCBs have ICRs findings of the recent literature. However, well above 1 in Bangladesh, Pakistan, these median characteristics of banks’ client and Sri Lanka. However, India’s better firms mask significant variations. Therefore, ­ p erforming SOCBs still fail to cross the the statistical test of difference between client ­t hreshold—­p ossibly indicating systemic firm characteristics of SOCBs and private ­ distress in the sector from 2009 to 2018. banks shows that only leverage differs. That Recent literature argues that bank distress is, client firms of SOCBs are significantly can arise because the firms and households more leveraged. that banks serve are in distress. For instance, uncertainty about economic policy can boost the default risk of both firms and house- Understanding Bank Distress and holds, which is then transmitted to banks Its Main Factors (Ashraf and Shen 2019; Gopalakrishnan Conceptually, our econometric framework and Mohapatra 2019). The literature also builds on the value at risk (VaR) methodol- highlights the possibility that government- ogy. It examines two types of losses. The first owned banks may engage in adverse is the financial loss that could be passed on ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    67 FIGURE 2.7 India: Characteristics of the Average Client Firms of Scheduled Commercial Banks, 2009–18 Return on assets Investment to assets Debt to assets Debt to equity Total assets 0 0.50 1.00 1.50 2.00 2.50 3.00 3.50 PVTBs SOCBs Sources: World Bank staff calculations using Prowess data. Note: All numbers are in percent except for total assets, which are in Rs, 100 billion. PVTBs = domestically owned private banks; SOCBs = state-owned com- mercial banks. to the central government in part or entirely, A bank is considered to be in distress when depending on the budget constraints SOCBs it does not have enough revenues to cover its face—that is, softer versus harder budget interest due: that is, when its ICR drops constraints. Gauging this loss involves esti- below 1. As robustness checks, we used three mating the probability of SOCB distress other indicators: ROA dropping below zero (PD) and financial loss given SOCB distress percent; the bank capital to risk-weighted (loss given distress, LGD). The second type is assets ratio (CRAR) measured against a the economic loss from SOCB distress due to threshold related to the amount of capital forced adjustments by distressed SOCBs, above the minimum prudential requirement such as in the form of changes in capital, that banks want to keep;10 and the bank’s debt, lending, or investments, which in turn z-score, a popular solvency indicator in the can affect firms, consumers, and the govern- literature (Laeven and Levine 2009; Ashraf ment. Here, the focus is on the loss of private and Shen 2019). The average annual proba- firms’ investment due to the frequent distress bility of distress for a given group of banks— of SOCBs: that is, unrealized investments such as private banks or SOCBs—could compared with the counterfactual of private be estimated as the average probability of firms being able to make investments ­ distress using historical data on identified through financing from banks not in distress d istress events (see equation (2A.1), in ­ regardless of whether those banks are pri- annex 2A, for details). vate or state owned. For the big picture of the analyzed data, figure 2.8 plots the ICRs for banks in India (panel a) using Prowess data—disaggregating Identifying Distress Using Financial new and old private banks—and in Soundness Indicators Bangladesh, Pakistan, and Sri Lanka (panels We define a distress event as the breach of a b, c, and d) using Fitch Connect data.11 threshold. In principle, the threshold could be One can observe that in India, the ICR of determined by an economic relationship or a new private banks is comfortably above 1 practical rule of thumb. The threshold value most of the time for most of the individual together with an actual value of an indicator banks—even though some outliers fall variable then help identify a distress event. below 1. The situation is progressively worse 68   H IDDEN DEBT FIGURE 2.8 India, Bangladesh, Pakistan, and Sri Lanka: Interest Coverage Ratio by Bank Type, 2009–18 a. India b. Bangladesh 2.0 6 1.5 4 Percent Percent 1.0 2 0.5 0 0 New private Old private Public PSB PVTB c. Pakistan d. Sri Lanka 6 10 8 4 Percent Percent 6 4 2 2 0 0 PSB PVTB PSB PVTB Sources: For panel a, World Bank staff calculations using Prowess data; for panels b, c, and d, World Bank calculations using Fitch Connect data. Note: There are 400 observations for India, 237 for Bangladesh, 141 for Pakistan, and 98 for Sri Lanka. In panel a, New private bank, Old private bank, and Public are defined as in Mishra, Prabhala, and Rajan (2019). PSBs = public sector banks; PVTBs = domestically owned private banks. for old private banks and the SOCBs (or pub- the drivers of distress is important for devel- lic banks). Not only do outliers occur more oping and implementing effective remedies. frequently, skewing the distribution more toward zero, but the 25th percentile of the Examining the Distress Factors distribution (the lower T arm of the candle graph) reaches way below 1—showing that To assess the difference in the probability of these distress events repeat frequently. distress between private banks and SOCBs In Bangladesh, Pakistan, and Sri Lanka, and whether certain bank characteristics private banks do not differ significantly from could drive bank vulnerability to distress, we the SOCBs, and their situation could be ran a logit regression of the probability of equally concerning. Figure 2.8 shows that distress on bank characteristics including half the time, half the banks—regardless of size, age, and ownership type. We also whether private or public—would be included characteristics indicating the bank ­ considered in a state of distress: that is, have funding and business model (such as the an ICR below 1. Given how much these econ- loan-to-deposit ratio, net foreign exchange omies rely on their banking system to help exposure, and ratio of risk-weighted assets with resource allocation for greater produc- to total assets) and macrofinancial shocks tivity and jobs, as well as financial inclusion (commodity price shocks, portfolio capital for greater access to opportunities, it is flows). Equation (2A.2), in annex 2A, unlikely that these functions can be performed ­ p rovides mode details of this regression effectively by banks in distress. Understanding estimation. ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    69 In running the logit regression, we were Overall Results, with a Focus on India most interested in uncovering whether Our baseline logit regression focused on the SOCBs are on average more prone to distress likelihood of bank distress and in particular than privately owned banks—controlling for on distress of SOCBs compared with other factors such as size, funding models, and types of banks (table 2B.1, in annex 2B). average bank specific effects. The nondis- The results suggest that, on average, SOCBs tressed private banks serve as the control in India (PSBs) are significantly more likely group in this difference-in-difference estima- to experience distress compared with tion. In addition, we singled out foreign PVTBs. banks and old private banks; the latter are Interestingly, the result is not driven by SBI thought to be the closest control group for because this bank is not significantly more or SOCBs because they existed during the less prone to distress, as shown in column 2 nationalization period (from 1969 to 1980) of table 2B.1. New private banks and foreign when majority state ownership was estab- banks are not significantly different from old lished in the existing SOCBs (Mishra, private banks—although the lower probabil- Prabhala, and Rajan 2019). ity of distress for foreign banks could be bor- Broader public governance issues could derline significant. Given that SBI is the drive the inherent weaknesses in PSBs. largest bank in India, its failure or perception However, in a vicious cycle, weak banks with of its failing would significantly shake confi- poor governance structures that have suffered dence in the system. As such, this bank is very reoccurring episodes of distress have been the likely to receive rapid state attention in the main recipients of government capital injec- event of any signs of distress. Indeed, a high tions—which have in turn increased the share propensity for extraordinary state support of government ownership. In addition, once underpins this bank’s credit rating by Fitch, these banks receive additional capital, they among others.12 The coefficient on the SBI are expected to increase lending mainly to dummy is positive, suggesting that, on aver- support priority sectors or government pro- age, SBI could be more prone to distress com- grams that are not always viable, which fur- pared with old PVTBs; however, the result is ther increases their risk of distress. Therefore, not statistically significant. This finding is the results could simply reflect the recurring likely driven by the overall weakening of higher probability of weak banks being regu- financial soundness indicators, particularly in larly in distress, of receiving recurring govern- recent years, with the nonperforming assets ment support, of increasing their directed ratio breaching the 10 percent threshold and lending, and of lacking decisive intervention the ROA turning negative in 2018. to resolve their underlying problems. If addi- Importantly, the likelihood of distress tional capital injections are not coupled with increases with the share of government own- meaningful reforms, then these SOCBs may ership, as shown in table 2B.1, column 3.13 continue to exhibit recurring or even intensi- We estimated that SOCBs with a government fying distresses. share of at least 50 percent but less than This government bailout dynamic intro- 70 percent could be less prone to distress than duces a possible selection bias that can result SOCBs in which government has more than in overestimates of the coefficient on the 70 percent share of ownership. SOCB/PSB dummy. To address this issue, we These findings suggest that SOCBs could be simply retained the classification of PSBs, old more fragile by design (Calomiris and Haber and new PVTBs, and foreign banks used at 2014). That is, the overall governance sur- the beginning of the sample and kept it fixed. rounding SOCBs potentially exposes them to Future research could allow for the owner- more or greater shocks, such as from directed ship type to change over time and adjust for lending, directed support of government pro- the described issue using selection bias correc- grams, political interference in management, tion (or an instrumental variable approach). forced overemployment, or unqualified 70   H IDDEN DEBT employment (Cole 2009; Ashraf, Arshad, and foreign currency lending within the domestic Yan 2018; Richmond et al. 2019).14 banking system and firm access to foreign The estimations also indicate that smaller currency via other forms of financing, such ­ banks are relatively more prone to distress as international capital markets. Therefore, a than their larger counterparts. This is not sur- bank that has access to sizeable foreign cur- prising given the characteristics and more con- rency liabilities would need to be sound and centrated business models of smaller banks. capable of competing with international Notably, smaller banks rely more on borrow- financiers. ings to fund their activities; can be more Although SOCBs are significantly differ- exposed to riskier segments of the market ent from private banks in terms of several (such as priority sectors and small borrowers); bank characteristics, as shown in table 2B.3, and have limited diversification to help pool in annex 2B, inclusion of these characteris- shocks. SOCBs are larger, on average, and can tics does not diminish the significance of the therefore manage concentrated risk better. SOCB dummy in explaining greater proba- The results further suggest that banks with bility of distress at SOCBs. However, inclu- a greater loan-to-deposit ratio are more prone sion of one characteristic does remove that to distress. The ratio could also be a proxy significance: the nonperforming loan ratio. for the bank’s funding structure and risk: that This finding suggests that credit risk manage- is, how much of its loan book is funded by ment and culture are key to the more fre- own deposits relative to other sources of quent distress at SOCBs. It coincides with funding. Higher loan-to-deposit ratios can the finding of Mishra, Prabhala, and Rajan thus indicate less diverse sources of funding, (2019), who studied the pace of adoption of which can increase the likelihood of distress. credit scoring technology and found that In particular, banks with loan-to-deposit while new private banks adopt scoring ratios above 100 percent are more exposed to quickly for all borrowers, SOCBs (PSBs) liquidity shocks—for instance, because of adopt scoring quickly for new borrowers but their borrowing exposures to the money mar- not for existing borrowers. They conjecture ket and private market credit lines. This find- that organizational culture, possibly from ing dovetails with that of the International formative experiences in sheltered markets, Monetary Fund’s Global Financial Stability explains the patterns of adoption of credit Report (IMF 2013) that higher loan-to- scoring technology. In addition, our finding deposit ratios (greater reliance on wholesale does not rule out broader governance issues funding) are, across the board, linked to and political economy influence in forming higher levels of distress in banks in advanced the structures and decisions underpinning and emerging economies. The result with credit risk management in SOCBs. respect to loan-to-deposit ratios is relatively Our results are robust to using the bank more important for private sector banks z-score as the measure of distance to distress because SOCBs in India have a loan-to- and rerunning the regression with ordinary deposit ratio significantly below 100 percent, least squares (OLS; see table 2B.2). Here, the on average. In India, the correct interpreta- significant effect of the degree of government tion could involve lower capacity to interme- shareholding survives the inclusion of the NPL diate deposits. That is, banks that are not ratio in the regression. In addition, the SBI able to intermediate the volume of deposits effect becomes significant and survives the they mobilize are less efficient and more vul- inclusion of the NPL ratio in the regression. nerable to distress. This additional finding keeps open the ques- The age of a bank and its foreign currency tion of the effect of broader governance issues exposure do not appear to significantly affect beyond credit risk management. These issues the likelihood of distress. The estimated neg- could include problems with staffing and ative coefficient on foreign currency expo- career management of executives, soft budget sure may coincide with the low levels of constraints, cronyism, and political links. ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    71 Broadening the Analysis to Include diminish the probability of distress across the Bangladesh, Pakistan, and Sri Lanka region. Here, apart from greater business We now broaden the analysis to cover the diversification, we can also conjecture that main economies of South Asia, in addition to more intensive supervision of larger and sys- India—Bangladesh, Pakistan, and Sri Lanka— tematically important banks helps diminish the although we have less data available. To ana- probability of distress at larger banks. The ear- lyze the distress at SOCBs relative to private lier observation for India that banks that can banks in these countries, we ran similar logit intermediate less of the volume of mobilized regressions using the distress events identified deposits are more vulnerable to distress is also by ICR falling below 1 as the dependent vari- valid for Bangladesh and Pakistan. It is not ables. Here, we could only control for bank valid, however, for Sri Lanka, where the characteristics because the Fitch Connect median loan-to-deposit ratio is about 100 database does not allow the mapping of percent—significantly higher than in the other banks to their clients (as the Prowess data set three countries—and whose coefficient is esti- does for India). Also, some of the bank char- mated to be positive and borderline significant. acteristics that we control for could be slightly With at least 50 percent of banks in the Sri different. Table 2B.3, in annex 2B, reports the Lankan system running loan-to-deposit ratios estimation results for each country (columns above 100 percent, the indicator plays a more 1–4) and for all countries pooled (South Asia traditional role. It reflects the exposure to regional estimation) (column 5). liquidity risk. The higher the exposure, the On the pooled basis that could be repre- greater the probability of distress (that is, of sentative of the South Asia region, the proba- ICR falling below 1). The ratio of liquid assets bility of distress is higher for SOCBs than to total assets has a similar interpretation for private banks—even if controlling for several all countries: greater liquidity buffers lower the bank characteristics on which SOCBs differ probability of distress. Greater risk taking— from private banks (see panel c of table 2C.1, reflected in a higher ratio of risk-weighted in annex 2C). However, this pooled estima- assets to total assets—is a sign of resilience tion masks some country differences. In Sri among South Asian banks (“Pooled” column, Lanka, the probability of SOCB distress is table 2B.3) perhaps i ­ndicating relatively lower higher than that of private banks and lending to SOEs, especially those in Bangladesh ­ significant—as in India. By contrast, such a and to some extent India. Risk taking does not high probability cannot be confirmed in seem to be a trait of resilient banks in Sri Bangladesh and Pakistan, where state owner- Lanka, however. ship does not seem to play a role in explaining the probability of distress. This result could Understanding How Banks Adjust be due to several controls that pick up main in Distress differences between SOCBs and private banks in terms of bank characteristics, or it could Next, we examine the adjustments that banks simply be that SOCBs in Bangladesh and make in episodes of distress to shed light on Pakistan are less different from private banks. the possible fiscal and economic losses that (Compare the absolute value of the t statistic arise as distressed banks—particularly for the differences between SOCBs and pri- SOCBs—adjust their capital, lending, debt, vate banks in panel c of tables 2C.2, 2C.3, and investment. That is, assuming that some and 2C.4. They are much higher for India budgetary constraint exists for banks, and than for Bangladesh and Pakistan.) Still, at perhaps softer constraints for SOCBs, we least for India and Sri Lanka, the issues with assess how SOCBs adjust relative to private broader governance and political economy banks. We define the loss given distress (LGD) seem valid. as the monetary loss due to all forced adjust- From the estimates on the control variables, ments that the PSB in distress must perform we observe that bank size continues to to survive, restructure, or close. The LGD for 72   H IDDEN DEBT SOCBs is estimated relative to the control distressed PVTBs (see tables 2B.4 and 2B.5, group of similar private banks for our in annex 2B, for detailed results). This find- ­difference-in-differences estimation. ing could reflect the government’s efforts to The LGD can be estimated based on the promptly recapitalize at least systemically monetary value of all the adjustments that important public banks—most notably, happen when a distress occurs. For SOCBs, India’s SBI. It can also reflect the softer bud- we focus on the following five categories of get constraints that SOCBs as a group enjoy adjustments: compared with private banks. These softer budget constraints can then increase moral 1. Percent change in capital hazard among SOCBs and distort their 2. Percent change in provisioning incentives to properly manage credit and 3. Percent change in debt other risk taking, as well as act in a timely 4. Percent change in lending manner to restore their performance when 5. Percent change in fixed assets (including it declines. sale of fixed assets). To a lesser extent, during the initial year To estimate the adjustment size for each of of distress, SOCBs tend to increase fixed these five categories for distressed SOCBs rel- assets (invest)—or at least do not drop ative to private banks, we regress each vari- their plans to accumulate fixed assets. This able in categories 1–5 on the distress dummy, result could be linked to the government the interaction of the distress dummy with the capital injections that often come with the SOCB dummy, and control variables—includ- conditionality to continue supporting pri- ing bank and time fixed effects (see appendix ority lending sectors and government pro- 3A for a detailed description of the estimation grams and stimulate economic growth. If methodology). SOCBs are unable to stimulate growth through lending—for instance, because Overall Results, with a Focus on India breaching prudential requirements can Our estimation results, summarized in trigger regulations that prohibit increasing f igure 2.9, suggest that compared with ­ the lending portfolio—the SOCBs can use distressed PVTBs (the control group), dis- their investments to help stimulate growth tressed SOCBs tend to adjust to distress by and meet government conditions of increasing capital relatively more than recapitalization. FIGURE 2.9  Differences in How State-Owned Commercial Banks and Domestically Owned Private Banks Adjust in Times of Distress Capital Debt Lending Investment –4 –3 –2 –1 0 1 2 3 4 5 6 t statistic PVTBs SOCBs Source: Tables 2B.4 and 2B.5, in annex 2B; Kibuuka and Melecky 2020. Note: The bars depict the t statistics of the estimated adjustment coefficients. PVTBs = domestically owned private banks; SOCBs = state-owned ­commercial banks. ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    73 Moreover, our estimation results show that turns negative. In contrast, distressed SOCBs distressed PVTBs tend to reduce lending sig- do not seem to significantly increase these nificantly in the initial year of distress. buffers during similar distress events because Compared with distressed PVTBs, distressed they appear to rely on capital infusions to SOCBs do not decrease their lending signifi- address these issues. This difference in adjust- cantly in the initial year of distress. PVTBs do ment channels in the face of similar distress not appear to use other adjustment channels events could reflect the difference in gover- significantly. It may be more difficult and nance in these banks—private banks rely on expensive for them to raise additional equity their own resources, while SOCBs rely on or borrow in times of distress; the estimation government resources to address the distress results for debt somewhat support this con- issue. The robustness checks broadly concur jecture. Namely, PVTBs can somewhat reduce with the baseline results—although they are their debt financing during periods of distress not as statistically significant (CRAR) or their compared with SOCBs, whose borrowing can timing is slightly different (ROA). be relatively less affected or even increase in Regarding debt dynamics, distressed periods of distress relative to PVTBs. Note, PVTBs reduce their debt borrowing in times however, the possible offsetting results for dis- of distress, while distressed SOCBs enjoy tressed SOCBs’ debt across the contempora- softer borrowing conditions than distressed neous and lagged PSB dummy (table 2B.4, PVTBs. The difference between SOCBs’ and panel b, column 2). With the disciplining PVTBs’ debt borrowings in times of distress pressure of difficult access to additional debt could also relate to the prevailing type of debt in times of distress in mind, PVTBs could be instruments the banks use. SOCBs tend to keeping larger buffers to draw on and serve as borrow from public institutions and agencies a cushion during episodes of distress—which such as the RBI, while PVTBs tend to access could mitigate the likelihood of distress in the and are more exposed to foreign capital mar- first place. (For instance, the capital adequacy kets and their stricter covenants, and thus ratio and net interest margins of PVTBs are discipline. higher than those of SOCBs; see table 2C.1, Other significant results indicate that pri- in annex 2C.) Also for this reason, PVTBs’ vate banks reduce lending during the initial preferred adjustment channel can be to reduce year of distress when their CRAR falls below lending to clients—both by a prior strategic 11 percent. Because these banks tend to main- choice and necessity in times of distress. tain higher CRARs, they may prioritize build- Our estimation results are robust to alter- ing capital buffers over increasing lending native threshold measures of distress (table when these ratios fall below 11 percent. In 2B.5): namely, the ROA turning negative and addition, because their income sources are the CRAR falling below 11 percent.15 During more diversified relative to SOCBs, private the initial year of distress, distressed SOCBs banks can reduce lending without signifi- tend to increase capital when the ICR falls cantly affecting their income. below 1, or profitability turns negative, or We now turn to fixed assets and investment when the ROA turns negative. When CRAR dynamics during times of distress. Perhaps falls below 11 percent, distressed SOCBs may because PVTBs find it difficult to adjust invest- also increase total capital, but this estimation ment plans in the near term, ­ distressed PVTBs result is not statistically significant. This may with a CRAR below 11 percent or negative be the case because when CRAR levels reach profitability (ROA) tend to reduce investment around 11 percent, they still exceed pruden- in fixed assets only in the year following dis- tial requirements of 9 percent; thus, SOCBs tress—compared with distressed SOCBs, may opt to adjust in other ways than increas- which can sustain investment. The differences ing capital. in adjustment by PVTBs and SOCBs when Distressed private banks tend to increase faced with a CRAR below 11 percent or nega- provisions during the initial year of distress as tive profitability (ROA) further illustrate well as the subsequent period as profitability PVTBs’ focus on self-reliance compared with 74   H IDDEN DEBT SOCBs, which enjoy softer budget constraints and SOCBs get recapitalized. The results by and backing by government capital to support country reveal that the private bank recapital- their operations and survival. The self-reliance ization happens in Bangladesh, India, and of private banks versus potential moral haz- Sri Lanka, but less so in Pakistan. SOCBs are ard of SOCBs can severely undermine market also promptly recapitalized, most significantly discipline and the efficient functioning of in Bangladesh, less so in Pakistan, and least so financial markets. in Sri Lanka. Although the recapitalization In sum, sound SOCBs may help sustain appears statistically more significant for pri- lending to firms throughout the cycle and in vate banks, it is almost three times larger for the face of financial shocks. However, weaker publics banks, on average. The same hypoth- SOCBs could fall into distress more often esis and result for India of SOCBs having than private banks and reduce lending in softer budget constraints than private banks times of distress compared with sound are thus broadly confirmed for other main SOCBs. However, if private banks get into South Asian economies. Provisions are distress—including because of common released by private banks after the distress macro shocks—they reduce lending much event, while SOCBs do not release accumu- more than distressed SOCBs and even more lated provisions—which could further con- so than sound SOCBs. Significantly reducing firm their greater reliance on new capital lending is the adjustment private banks select injections. in times of their less frequent distress. SOCBs Lending by private banks declines signifi- have softer budget constraints regarding both cantly after the distress event, while SOCBs equity injections and additional debt borrow- continue to increase lending. Again, the con- ings. Compared with private banks, the softer ditions of SOCB recapitalization may often budget constraint and conditions of govern- require them to continue to stimulate the ment recapitalization (to help stimulate economy—and if not breaching prudential growth) could encourage SOCBs to sustain rules, increase their lending. This pattern is their investments (fixed asset accumulation) most significant in Bangladesh and less so in even when distressed. However, the soft bud- India and Pakistan. In Sri Lanka, SOCBs cur- get constraints inflict substantial fiscal costs tail lending after distress—even more so than and erode discipline and competition in the private banks. This could relate back to low financial market. recapitalization of SOCBs in Sri Lanka and to their harder budget constraints relative to Broadening the Analysis to Include other SOCBs in the region. This result high- Bangladesh, Pakistan, and Sri Lanka lights the trade-off between increasing fiscal Can the adjustment patterns estimated for discipline through harder budget constraints Indian banks be generalized to the three other and having SOCBs absorb risks and continue main South Asian economies? To address this lending even when big shocks hit. question, we ran the same estimations on a After episodes of distress, investments pool of data obtained from Fitch Connect for increase at SOCBs, while they decrease mildly Bangladesh, India, Pakistan, and Sri Lanka. at private banks. This finding could again The timing of the data and the estimation reflect the conditions of recapitalization that variables were adjusted to better correspond require SOCBs to stimulate the economy. If with the fiscal year timing of the Prowess such stimulation is not possible through database for India. The estimation results are increased lending because of prudential con- reported in table 2B.6, in annex 2B. straints, increasing fixed investments is Our estimation results show that entering another way to implement such stimulus. a period of distress, the capitalization of both These results are consistent with the Prowess SOCBs and private banks is declining. Unlike estimation for India. This pattern is also the Prowess data estimate, the Fitch Connect strongly evident in Bangladesh. By contrast, data estimates suggest that both private banks this investment stimulus of SOCBs in distress ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    75 and after recapitalization is absent in Pakistan characteristics, such as total firm assets, firm and Sri Lanka. age, and sector, as well as key indicators of In Bangladesh, Pakistan, and Sri Lanka, banks, notably bank ownership type. debt financing of private banks is unaffected While our analysis relates banks to all or decreases slightly right before and during firms, it focuses on successful firms with episodes of distress. After the period of dis- higher sales growth and SMEs to understand tress, SOCBs also appear to decrease debt whether SOCBs can help reallocate capital to financing. This finding contrasts with the esti- more productive firms and whether they can mates for India, which showed increasing effectively work with more opaque SMEs as debt financing of SOCBs in distress and con- well as with large corporations. Namely, do firmed the hypothesis of soft budget con- successful firms with high sales growth get straints. This disparity may in part arise enough credit from both private and state because the Prowess and Fitch Connect data banks to invest and realize their potential? are not entirely compatible in their definition Controlling for their size and age, do SMEs of categories and dating, as mentioned. get adequate access to finance from both pri- vate and state banks to invest and grow? And do successful SMEs with growing sales get Analyzing the Effect of Firms’ adequate lending support from SOCBs— Banking with SOCBs Compared compared with support from private banks— with Private Banks to grow and create productive jobs? SOCB distress can have vital economic Our regression analysis controlled for impacts on firm financing and private invest- firm-specific effects and common shocks ments. For instance, if SOCBs are more prone using firm-level fixed effects and year dum- to distress than private banks, and in distress, mies. It focused on one key outcome measure predominantly adjust by reducing longer-term relating to firms: their ability to sustain lending to SMEs, small private firms doing investments, measured as (log-log) change in business primarily with SOCBs will suffer a fixed assets. Because firms use multiple banks, greater loss of access to financing or unreal- we used two types of dummy variables in our ized investments over time. In contrast, if estimations. The first dummy captured SOCBs are as equally prone to distress as pri- whether any of the banks to which the firm is vate banks and, thanks to softer budget con- linked is an SOCB (yes = 1, otherwise 0). The straints, can issue debt or get equity injection second dummy captured whether a majority and continue lending even in distress, private of the linked banks are SOCBs (yes = 1, firms doing business primarily with SOCBs ­ otherwise 0). We report the estimation results will experience a smaller loss of access to in table 2B.7 using the former type of classifi- financing and unrealized investments over cation (dummy variable) because the results time. We try to shed some light on these mat- from the two dummies are not materially ters next. different. Our estimation results suggest that, on average, larger and older firms invest (grow Estimating the Effect of Banking with their fixed assets) more than smaller or SOCBs on Investment by Client Firms younger firms. Also, firms invest and then As a first step in our analysis, we linked firms gradually deplete (depreciate) investments to banks in the Prowess data set. This linking before investing again—hence, the negative of firms with banks is possible only for India. correlation with the lagged investment value. With these links established, we could merge Importantly, more successful firms with a our panel bank-level data set for India with higher growth of sales invest more. As for the the firm-level data set for India constructed links with SOCBs versus private banks, the in Melecky and Sharma (2020). As a result, story of firm investment needs to be unpacked we built a firm-level panel with key firm in two stages. 76   H IDDEN DEBT First, firms that move from banking with 0.0124 = 0.0466). However, SMEs that a private bank to banking with a SOCB switch to SOCBs invest significantly less than invest less on average—controlling for firm other firms (table 2B.7, column 4). The esti- characteristics and past investments (table mate is economically more significant than 2B.7, column 2). Even when switching to the effect of size, age, or sales growth. SOCBs, firms with higher growth of sales Moreover, successful SMEs with high sales invest more than other firms that switch to growth are held back even more by switching SOCBs. However, firms with high growth of to SOCBs in trying to realize their investment sales still invest less when switching to potential (see column 5 of table 2B.7, the tri- SOCBs rather than when banking with pri- ple interaction with PSB × Sales growth × vate banks. SME). This could suggest that SOCBs are Second, the big story emerges in the rela- particularly challenged by screening SME tion of SMEs to SOCBs. When we control for creditworthiness and potential for invest- SMEs,16 the results show a stark contrast ment. Future research could focus on whether between SMEs and larger firms. Namely, this result could be due to SOCBs not lending larger firms that switch to banking with enough to SMEs overall or the willingness of SOCBs invest more on average; this suggests SOCBs to lend to SMEs only for working that the earlier negative estimate was driven capital needs, on average. The result reflects by SMEs. Larger firms with higher growth of the anecdotal evidence about SOCB lending sales invest more than other larger firms that practices and credit underwriting. switch to SOCBs—even more so than average Anecdotal evidence suggests that SOCBs firms that switch to a private bank (0.0342 + focus more on meeting the lending quotas for BOX 2.1  Main Findings of the Overall Analysis State-owned banks, smaller banks, and banks However, the soft budget constraints impose sub- with a higher intermediation ratio of loans to stantial fiscal costs and erode market discipline. deposits—but still less than 100 percent—are This raises the question of whether this costly more prone to distress. The higher average vul- insurance and risk-absorbing function of SOCBs nerability of state-owned commercial banks pays off in terms of wider economic benefits, such (SOCBs) to distress may increase with the as sustained firm investment. share of state ownership, or at least it does in The type of bank ownership (public versus India. private) affects the investment of client firms, SOCBs adjust in distress differently than pri- with important effects on the economy. Namely, vate banks because of their soft budget con- larger firms that switch to banking with SOCBs straints. Weaker SOCBs enter distress more as opposed to banking mainly with private often than private banks, and when distressed, banks invest more than other firms. This is espe- they reduce lending more than healthy state cially true for larger firms with a higher growth banks. Although private banks enter distress less of sales. The opposite is true for SMEs. SMEs frequently, when they do, they reduce lending that switch to SOCBs invest significantly less much more than state banks in distress—and than other firms—especially successful SMEs therefore, much more than healthy SOCBs. with high sales growth. This finding may be In terms of financing, SOCBs enjoy softer bud- explained in part by the weak risk management get constraints, readily obtaining state equity and culture at state commercial banks—including debt support. The softer budget constraints, as their low capability to appraise SMEs’ credit- well as conditions of government recapitaliza- worthiness and screen individual SME invest- tion, enable SOCBs to sustain lending to clients ment projects for viability (Mishra, Prabhala, and their own investments in times of distress. and Rajan 2019). ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    77 the volume of extended credit than they focus on the quality of project screening and under- Proposed reforms call for better defining writing. These quotas are more easily met by SOCB’s social mandates, matching costly serving larger firms—including state-owned social mandates with earmarked subsidies, enterprises implicitly backed by a government moving to harder budget constraints, and guarantee—than opaque and risker SMEs. considering a role for wholesale funding While this anecdotal evidence may also apply to some private banks, private banks have in providing credit to the economy. stronger credit screening capabilities and can be more successful in reaching the higher on bank recapitalization. Figure 2.10 shows return-risk frontier that crediting SMEs how capital injections to distressed SOCBs offers. SOCBs are simply not as good at man- have soared recently in India. aging risks—especially on the credit side The effort to consolidate state-owned (Mishra, Prabhala, and Rajan 2019). banking in India is a welcome step—especially Main findings for the overall analysis are for smaller state banks with weak governance summarized in box 2.1. and the ability to efficiently intermediate mobilized deposits. Yet, even with substantial consolidation, further reforms of SOCBs may Policy Recommendations be needed. One question is whether state- SOCBs have a large footprint in South Asia, owned banks should remain retail lenders or especially in India. Their unique ability to whether they should intermediate the deposits reach out and mobilize deposits is not they mobilize through wholesale funding of matched by their ability to provide credit effi- private banks, as well as adequately regulated ciently to the economy. However, state own- and supervised nonbank credit institutions ership in banks can help shield local that could reach SMEs more efficiently and populations and microenterprises from spur productive local investment. shocks by, for example, providing contingent Another question is how to define the credit after natural disasters such as droughts, missing mandates for the many SOCBs to floods, and hurricanes (World Bank 2020). correctly set incentives, increase transparency, Historically, this positive role of state owner- and enhance financial accountability. One ship has come at the cost of more frequent proposal is to divide state commercial banks distresses at weaker state-owned banks and into purely commercial state banks that maxi- substantial—and increasing—­ fiscal outlays mize profit and SOCBs with a single or FIGURE 2.10  Capital Injections by the Indian Government to Distressed State-Owned Commercial Banks, FY2009–FY2020 1.0 0.8 0.6 Rs, billion 0.4 0.2 0 FY2009 FY2010 FY2011 FY2012 FY2013 FY2014 FY2015 FY2016 FY2017 FY2018 FY2019 FY2020 Source: Tables 2B.4 and 2B.5, in annex 2B; World Bank staff computations. Note: The bars depict the t statistics of the estimated adjustment coefficients. 78   H IDDEN DEBT limited number of well-defined social objec- measures—contributing to a long-term tives. On behalf of the central government, development objective of the government the state bank ownership unit should care- as featured in national financial devel- fully monitor the performance of both types opment strategies. Time horizon(s) for of banks through specific key performance performance/impact evaluation would indicators (KPIs), track closely the extra costs correspond to these objectives and strate- of fulfilling any social mandate, and enforce gies. Over prespecified horizons, impact steps to strengthen transparency and would be audited using a monitoring and accountability. evaluation framework established for this purpose. 1. Purely commercial state banks . Like This second type of state bank warrants their private sector counterparts, purely some special considerations. commercial state banks would focus on profitability—perhaps with greater 1. Subsidies and Incentives. The social emphasis on sustainability and a longer mandate could impose higher-than- profit-maximizing horizon.17 The main market operational costs (for example, mandate would be to raise revenue for to build new infrastructure to reach new government spending that could focus on customers in rural areas) and higher- socially beneficial areas, such as human than-market expected losses from loans capital development. The KPIs for these (cost of credit) and other financial ser- banks would be similar to those of pri- vices (for example, due to lower financial vate banks, but could cover and be moni- literacy, capabilities, and greater riskiness tored over longer horizons to ensure of the newly reached customers). There- greater sustainability. These banks would fore, the required subsidies (compensa- have relatively hard budget constraints tion) to cover the higher-than-market and be forced to adjust if they got into operational and credit costs would need distress, as private banks do. Manage- to be explicitly and thoroughly assessed, ment of these banks, including the board estimated, and provided on an annual of directors, would be accountable by basis in the central government budget losing autonomy quickly in the event of and included in the medium-term fiscal undue distress or mismanagement—first expenditure framework. The required to the state agency in charge of state level of subsidy would need to be moni- bank ownership and later to the agencies tored and adjusted once data on opera- charged with temporary administration tional costs and expected losses from and resolution. Purely commercial state fulfilling the social mandate have regu- banks, as well as all other state-owned larly been collected and analyzed. Once banks, must be properly supervised to the subsidies (regular fiscal provisions) ensure a level playing field with private for higher-than-market operational costs banks—including the treatment of lend- and credit costs are in place, state banks ing to nonbank financial institutions, could operate under a relatively hard both private and state-owned. budget constraint. 2. SOCBs with a social mandate(s). These 2. Governance and supervision. Manage- banks would have a broader mandate ment of these banks, including the board than raising state revenue/maximizing of directors, would be accountable for profit, such as advancing financial inclu- sustainable financial performance and sion in rural areas. Accordingly, their evaluated/audited impact by losing its incentive structure and management autonomy in a prompt corrective action would differ. The KPIs would encompass scheme involving replacement of direc- a combination of profitability and impact tors and management by the government ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    79 agency performing the ownership func- banks’ mandated focus on regions (such as tion for state banks; temporary adminis- focusing on extending credit to a few lagging tration by the supervisor (such as the cen- regions); or an asset class (such as lending to tral bank) to financially stabilize troubled opaque and risky micro, small, and medium banks; or the unwinding and closure of enterprises [MSMEs] or promoting access to the bank if its performance (on returns housing finance to the bottom 40 percent of and impact) proves unsustainable. Again, the population); or an industry (such as cofi- such state-owned banks must be properly nancing in infrastructure public-private part- and independently supervised to ensure nerships [PPPs]) (see chapter 1). These banks they remain sound and help create rather would have to hold more and diverse assets than distort the banking market. to pool across several types of concentration 3. Accounting for unexpected losses. If the risks and diversify them to the extent possi- social mandate involves taking idiosyn- ble. However, some national residual con- cratic risks that could be well diversified, centration risk can remain. The government no policy treatment of unexpected losses would have to reinsure the residual concen- would be required, in theory. However, tration risk using fiscal reserves and create higher-than-market unexpected losses fiscal space for contingent debt. To relieve (surprise variation of expected losses dur- the fiscal space of the reinsurance claims and ing some credit cycles/downturns) can costs, state commercial banks with plausibly arise because of bank social mandates. higher-than-market unexpected losses could Unexpected losses at state-owned banks seek reinsurance individually or as a group may occur only occasionally—in one or in the international markets or with interna- a few years during the credit cycle that tional financial institutions that can pool the can typically last five to seven years— residual national concentration risk and, to a and not necessarily during every credit large extent, diversify it away through global cycle. These higher-than-market unex- distribution. pected losses would require subsidies For infrastructure financing, for exam- (fiscal reserves) if assessed to be system- ple, more efficient financing models than atically attributable to the bank’s social SOCBs could exist. A major study by the mandate—as opposed to, for example, Asian Development Bank, UK Department corruption, fraud, or faulty gover- for International Development, Japan nance. The unexpected losses could be International Cooperation Agency, and assessed as systematic if they stemmed World Bank (ADB, DFID, JICA, and World from the inherently larger exposure Bank 2018; see also spotlight ES.1) pro- to, for instance, concentration or for- vides an extensive discussion of the financ- eign exchange risks that come with the ing options for infrastructure projects social mandate. The larger-than-market through public, corporate, and project ­ e xposures to these risks are triggered finance (and their hybrids) depending on occasionally by economy-wide shocks— the project characteristics. Here, probably such as local currency depreciation or the most efficient option is project bond supply-chain disruption—outside of the financing through local currency markets, bank’s control. with its ability to distribute the concentra- tion and foreign currency risks of large More specifically, some SOCBs are man- commercially viable projects, as well as dated to go beyond exposures to systemic mobilize financing with long maturity. This risk with which the market is comfortable option requires as one precondition the because their activities involve some inherent development of a deep and liquid govern- concentration risk and higher unexpected ment bond market in local currency—still losses. The latter can arise because of the something lacking in South Asia. 80   H IDDEN DEBT In contrast, for state commercial banks stabilization could also be large. Moreover, with social mandates such as financial inclu- because SOCBs tend to omit productive sion of MSMEs, lagging regions, or vulnera- SMEs from their lending—and the omission ble segments of the population, only limited is likely to be worse during downturns, alternative models could be available— when uncertainty increases—these alterna- including working with various (local) non- tive social costs are topped up with the cost bank credit institutions and microfinance of lost productivity growth in the medium organizations to reach the underserved, as to long term. well as specialized private banks (such as Management, performance, and fiscal back- banks focused on SMEs). But these alterna- ing . The KPIs for the management and tives are not likely to diversify the concentra- board of banks with risk-absorbing man- tion and correlation risks very efficiently. The dates would have to create and manage the efforts to fulfill such social mandates require tension between taking enough risk and a multi-pronged approach through pay- managing it sustainably from the perspec- ments, deposits, credit, contingency credit tives of effective pooling, reinsurance, and (for instance, after disasters; World Bank fiscal contingent liabilities. In principle, the 2020), and even insurance—but perhaps activities financed by these banks should most importantly through trust, relation- generate enough revenue—through risk- ships, and local expertise to meet goals and adjusted returns and/or future tax returns— set KPIs.18 Delivering financial education not to jeopardize fiscal sustainability. If the locally together with financial services could generation of such direct and indirect mon- be one way to build local trust, relationships, etary returns is not possible but the banks and expertise. can still generate meaningful socioeconomic Special considerations for countercyclical (non-monetary) benefits in the medium to lending. SOCBs are often praised for provid- long term, 19 the central government may ing countercyclical lending, which can be a create reserves to cover the unexpected cost useful direct tool to support macrofinancial of this essentially “social assistance” and, stabilization—especially in downturns. This for instance, made the fiscal council respon- objective of mitigating the downturns of sible for authorizing its release to banks. credit cycles also requires state commercial Similar arrangements can be applied to banks to assume losses (expected and unex- banks performing macrofinancial stabiliza- pected) higher than the market. Because tion, a function that may not generate suffi- state commercial banks are more likely to be cient direct or indirect monetary returns but distressed than private banks—and this like- may trigger implicit contingent liabilities for lihood increases with the share of state own- the central government. The skill mix of the ership—they will need to be recapitalized board of directors will need to reflect the for lending in downturns when nonperform- tensions that must be professionally man- ing loans for all banks rise systematically. aged and the knowledge base, experience, This recapitalization absorbs substantial and independence that such management fiscal resources (also captured in figure requires. 2.10) that, in fiscally constrained countries, In closing, SOCBs have a role to play in can crowd out other socially beneficial fis- South Asian markets, but it is unclear whether cal expenditures, such as on health care, the wider economic benefits can outweigh the education, or infrastructure—unless large fiscal costs (often triggered in the form reserves are created beforehand. Because of fiscal contingent liabilities). The successes the recapitalization requirements could be of SOCBs have included faster inclusion large, the alternative social costs of using of the population in the use of digital state-owned banks for macrofinancial payments and bank deposits, provision of ­ ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    81 fintech/bigtech platforms, 20 and bringing liquidity transactions as reported in the banks’ financial education to rural areas. The short- public financial accounts. falls involve costly recapitalizations because The average annual probability of distress of weak governance, operational inefficien- (PD) can be estimated as the average proba- cies, credit misallocation, and lack of risk bility of distress using historical data on iden- management that can trigger large contingent tified distress events: liabilities. The macrofinancial benefits can involve some stabilization of the credit cycle through 1 PDi = { ∑ N ∑T Di,n,t = 1 | Ii,n,t ≥ I i ;0 , T * N n =1 t =1 } countercyclical lending. But this benefit comes at a large cost of greater capital misal- (2A.1) location to less productive firms and large fiscal costs of recapitalization that is needed where Di,n,t is the distress 0/1 dummy variable to backstop lending in downturns and that and i = [private sector banks; public sector can crowd out other important public expen- banks]. ditures, such as on health care and educa- tion. The proposed reforms are urgently Examining the Distress Factors needed for SOCBs in South Asia and beyond to clearly generate net socioeconomic To assess whether certain bank characteristics benefits. could drive bank vulnerability to distress, we run a logit regression for Dn,t on bank charac- teristics (size; age; bank type, whether public Annex 2A. Methodology for sector or private); funding model of the bank Determining Bank Distress (the loan-to-deposit ratio, net foreign Identifying Distress Using Financial exchange exposure); and macrofinancial Soundness Indicators shocks (commodity price shocks, portfolio capital flows). All are included in the vector We define a distress event as the breach of a of control variables, Xi,n,t , together with year quantitative threshold. In principle, the thresh- fixed effects: old could be determined by an economic rela- tionship (identity), predicted/expected value, p(Di,n,t ) = α Xi,n,t + ε i,n,t .(2A.2) or even a practical rule of thumb. The thresh- 1 − p(Di,n,t ) old value I , together with an actual value of an indicator I, then help determine distance to dis- We avoid including the indicators, In,t or tress and generate a dummy variable, Di,n,t, their transformations that are used to identify identifying observed distress. distress: that is, Di,n,t. Including those would We identify distress at Indian public sector result in estimating a tautological relation- banks (PSBs) using selected indicators of finan- ship. By adding year fixed effects, we capture cial soundness. The main indicator of distress common time factors and any other relevant is if the interest coverage ratio (ICR) drops macroeconomic shocks. This approach also below 1. As robustness checks, we use the reduces the need to cluster errors. return on assets (ROA) dropping below zero; By running the logit regression, we are the bank capital adequacy ratio (CAR) against most interested in uncovering whether state- a threshold related to the minimum prudential owned commercial banks (SOCBs) are on requirement banks want to keep; and non-zero average more prone to distress than privately emergency liquidity assistance (ELA) provided owned banks—conditional on other factors, by the central bank to a commercial bank. For such as size, funding models, and governance ELA, we are missing data because it is difficult indicators. Here, domestically owned private to distinguish between regular and emergency banks (PVTBs) serve as the control group. 82   H IDDEN DEBT Adjustments in Distress and Loss Given Estimating the Impact of SOCB Distress Distress on Investments by Private Firms Let us define the loss given distress (LGD) as Our outcome variables of interest are the eco- the monetary loss due to all forced adjust- nomic impacts of SOCB distress on firm ments that the PSB in distress must perform financing and private investment. The impact to survive, restructure, or close. Therefore, of SOCB distress can vary by the type of compared with the traditional expected loss dominant adjustment that SOCBs undertake formula, our LGD is equal to the loss given in distress—and the size of the adjustment. distress multiplied by the exposure in distress. We utilize a reduced-form framework for The LGD is estimated relative to the control SOCB distress relative to private bank dis- group of similar private banks. tress. We run the following regression: The LGD can be estimated based on the monetary value of all the adjustments that FINn,t = βi Di,n,t + γ X n,t + ∈n,t , (2A.5) happen when a distress event occurs (Dn,t = 1). For PSBs, we focus on the following adjust- where FINn,t is firm lending (log-log growth in ment, ADJ|Dn,t = 1, through five categories of debt) and investment (log-log growth in fixed adjustment j: assets), respectively; and X are controls, includ- ing sector, year, and firm fixed effects. βi is our 1. Percent change in capital coefficient of interest, which is expected to be 2. Percent change in provisioning negative. That is, a firm linked to a bank that 3. Percent change in debt experiences distress will have greater problems 4. Percent change in lending in undertaking investment, other things remaining equal. If |βSOCB| > |βprivate|, distresses 5. Percent change in fixed assets (including of SOCBs are more harmful than distresses of sale of fixed assets). private banks. For instance, compared with The LGD for an individual PSB can be esti- private banks, SOCBs could be adjusting in mated as follows: times of distress mostly by reducing lending, while serving firms that do not have other LGD(i,n,t ) = ∑(Jj =1) w j ADJi, j,n,t | Di,n,t = 1, banking options (links to private banks). If (2A.3) |βSOB| ≅ |βPrivate| and β is significantly negative overall, distresses of private and public banks where ADJi,j,n,t is the monetary loss due to are equally harmful. Note that here SOCBs adjustment j of bank n of type i in distress at could be still more problematic if they experi- time t. Setting wj = 1 assumes that all adjust- ence distress more frequently than private ments in distress are equally important. banks. If |βSOB| < |βPrivate|, distress of SOCBs To estimate the adjustment size for each could be less harmful: for example, because category j for distressed SOCBs relative to SOCBs can sustain lending and avoid closing private banks, we run the following branches even if in distress—including due to regression: the soft budget constraints they could enjoy. Then, especially during systematic stresses— ADJ j,n,t = θ PSB[ PSBn *Dn,t − l ] + θ Dn,t − l such as during economic recessions, near- (2A.4) + FEn + CEt + n,t , financial crisis, or financial crisis episodes—SOCB presence in the banking sys- where θPSB is the parameter of interest. FEn tem could support the resilience of lending are bank fixed effects. CEt are common time through the cycle and private investments. But effects. We interact the distress dummy with the cost of this resilient lending could be borne the PSB dummy to estimate the difference in by higher taxes or more government debt. adjustment between distressed PSBs and The results of the regressions are presented PVTBs. The lag l takes values 0 and 1. in annexes 2B and 2C. ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    83 Annex 2B. Regression Tables: Probability of Distress for South Asian Banks and Adjustments to Distress, 2009–18 TABLE 2B.1  Probability of Distress for Indian Banks: Baseline Regression Results, 2009–18 (1) (2) (3) (4) (5) (6) PSB − Dummy 0.111*** 0.110*** 0.00659 0.00645 (3.44) (3.43) (0.65) (0.64) New PVTB − Dummy −0.0481 −0.0521 −0.00990 −0.00954 (−0.99) (−1.03) (−0.89) (−0.86) Foreign − Dummy −0.0645 −0.0632 −0.0326 −0.0324 (−1.46) (−1.42) (−1.43) (−1.42) SBI − Dummy 0.0762 0.0100 (1.29) (0.72) Govt shareholding >50% and <70% 0.116*** 0.0162 (3.88) (0.92) Govt shareholding >70% 0.134*** 0.0121 (3.95) (0.80) Bank size (log total assets) −0.0288** −0.0275** −0.0241*** −0.00982 −0.00984 −0.00924 (−2.88) (−2.63) (−3.32) (−1.40) (−1.39) (−1.45) Age (years) −0.000663 −0.000691 −0.000420 −0.000139 −0.000136 −0.0000945 (−1.93) (−1.94) (−1.44) (−1.01) (−0.99) (−0.72) Loan to deposit ratio (log) −0.125** −0.126** −0.133** −0.00535 −0.00523 −0.00902 (−2.77) (−2.73) (−2.80) (−0.59) (−0.58) (−0.70) FX liabilities to total liabilities (log) 0.00992 0.0101 0.00546 0.00481 0.00475 0.000718 (1.54) (1.57) (0.81) (1.37) (1.36) (0.32) NPL ratio (log) 0.0256 0.0254 0.0368* (1.95) (1.92) (2.39) Firm characteristics Average total assets (log) −0.00139 −0.00270 −0.00683 0.00176 0.00184 −0.000503 (−0.12) (−0.23) (−0.59) (0.52) (0.53) (−0.13) Average debt to equity (log) −0.00719 −0.00715 −0.00107 −0.00385 −0.00376 −0.00289 (−0.60) (−0.59) (−0.10) (−0.97) (−0.95) (−0.70) Average investment to assets (log) 0.0196 0.0183 0.0226 −0.0111 −0.0110 −0.00717 (0.83) (0.79) (0.88) (−1.05) (−1.05) (−0.64) Average return on assets 0.00187 0.00184 0.00125 0.000704 0.000701 0.000435 (1.59) (1.58) (1.39) (1.32) (1.32) (1.19) Observations 503 503 503 480 480 480 R-squared 0.349 0.349 0.335 0.601 0.601 0.570 Year time dummies Yes Yes Yes Yes Yes Yes Source: Kibuuka and Melecky 2020. Note: The estimations are performed on Prowess data. Bank ICR < 1; logit regressions with robust standard errors (marginal effects at the means). t statistics in parentheses. FX = foreign exchange; Govt = government; ICR = interest coverage ratio; NPL = nonperforming loans; PSB = public sector bank; PVTB = domestically owned private bank; SBI = State Bank of India. * p < 0.05, ** p < 0.01, *** p < 0.001. 84   H IDDEN DEBT TABLE 2B.2  Probability of Distress for Indian Banks: Robustness Test Using z-Score, 2009–18 (1) (2) (3) (4) (5) (6) PSB − Dummy 1.224** 0.729 1.236** 0.722 (2.89) (1.48) (2.90) (1.47) New PVTB − Dummy −1.651* −1.645** −1.512* −1.417* (−2.36) (−2.65) (−2.08) (−2.16) Foreign − Dummy −6.964*** −6.534*** −6.933*** −6.568*** (−8.45) (−8.59) (−8.40) (−8.57) SBI − Dummy 2.442** 2.342** (2.94) (3.09) Govt shareholding ≥50% and <70% 2.636*** 2.623*** (5.44) (4.92) Govt shareholding >70% 2.009*** 1.148* (4.93) (2.07) Bank size (log total assets) 0.170 0.0858 0.141 0.0122 0.865*** 0.630*** (1.00) (0.55) (0.79) (0.07) (5.07) (4.46) Age (years) 0.00583 −0.00171 0.00704 0.000101 0.0151* −0.00181 (0.79) (−0.28) (0.94) (0.02) (2.14) (−0.29) Loan to deposit ratio (log) −1.290 −2.226 −1.281 −2.213 −1.869** −3.526 (−1.84) (−1.30) (−1.83) (−1.29) (−2.82) (−1.84) FX liabilities to total liabilities (log) 0.218 0.0940 0.206 0.0963 −0.731*** −0.737*** (1.16) (0.64) (1.10) (0.66) (−5.25) (−6.10) NPL ratio (log) 0.878*** 0.889*** 0.860*** (5.07) (5.15) (4.47) Firm characteristics Average total assets (log) 0.672 0.741* 0.216 (1.87) (2.02) (0.51) Average debt to equity (log) −0.476* −0.457 0.0478 (−2.01) (−1.93) (0.19) Average investment to assets (log) 0.324 0.282 2.044* (0.39) (0.34) (2.02) Average return on assets −0.00195 −0.00233 −0.0135 (−0.44) (−0.52) (−1.34) Observations 538 468 538 468 538 468 R-squared 0.392 0.607 0.393 0.609 0.317 0.523 Year time dummies Yes Yes Yes Yes Yes Yes Sources: Kibuuka and Melecky 2020; Fitch Connect database. Note: t statistics in parentheses. FX = foreign exchange; Govt = government; NPL = nonperforming loans; PSB = public sector bank; PVTB = domestically owned private bank; SBI = State Bank of India. * p < 0.05, ** p < 0.01, *** p < 0.00. ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    85 TABLE 2B.3  Probability of Distress for South Asian Banks, 2009–18 (1) (2) (3) (4) (5) Bangladesh India Pakistan Sri Lanka Pooled PSB − Dummy 0.104 0.428** −0.0979 0.669 0.425*** (0.48) (3.14) (−0.68) (1.82) (5.25) Bank size (log total assets) −0.0766 −0.0736** −0.244*** 0.337*** −0.0762*** (−1.44) (−2.60) (−3.76) (3.42) (−3.35) Loan to deposit ratio (log) −1.390*** −0.225* −0.209* 0.540 −0.371*** (−4.09) (−2.18) (−2.04) (1.44) (−3.71) NPL ratio (log) 0.0491 −0.0229 0.0930 0.421*** 0.0766** (0.98) (−1.01) (1.54) (3.44) (2.64) Liquid assets to total assets ratio (log) −0.445*** −0.105** −0.287** −0.259 −0.278*** (−3.85) (−2.73) (−3.19) (−1.79) (−6.49) RWA to total assets (log) −0.319 −0.0751 −0.0262 0.314 −0.331* (−1.71) (−0.93) (−0.10) (0.62) (−2.21) Observations 204 243 153 86 686 R-squared 0.235 0.579 0.286 0.549 0.283 Year time dummies Yes Yes Yes Yes No Country time dummies No No No No Yes Sources: Kibuuka and Melecky 2020; Fitch Connect database. Note: Bank ICR < 1; logit regressions with robust standard errors (marginal effects at the means). t ­statistics in parentheses. ICR = interest coverage ratio; NPL = nonperforming loans; PSB = public sector bank; RWA = risk-weighted assets. * p < 0.05, ** p < 0.01, *** p < 0.001. TABLE 2B.4  Baseline Regressions: Adjustments Given Distress, 2009–18 a. Adjustment to Capital, Provisions, and Lending during Bank Distress Total capital Provisions Lending (1) (2) (1) (2) (1) (2) Distress: ICR < 1 −0.0210 −0.0290 −0.0486 −0.0783 −0.0661 −0.123*** (−0.64) (−0.82) (−0.26) (−0.40) (−1.03) (−3.93) Distress: ICR < 1 − Lagged 0.0427 0.105 0.00796 (0.72) (0.53) (0.21) Distress × PSB 0.255*** 0.251*** 0.206 0.330 −0.0175 0.0446 (4.77) (5.13) (0.90) (1.40) (−0.27) (1.19) Distress × PSB − Lagged −0.00198 −0.443 −0.0345 (−0.02) (−1.95) (−0.83) Constant 0.0648* 0.0645* −0.199 −0.199 0.197*** 0.199*** (2.36) (2.36) (−1.76) (−1.78) (6.91) (7.11) Observations 643 642 616 616 662 661 R-squared 0.0384 0.0394 0.0628 0.0647 0.133 0.138 Year time dummies Yes Yes Yes Yes Yes Yes Bank fixed effects Yes Yes Yes Yes Yes Yes Sample All banks All banks All banks All banks All banks All banks (continues next page) 86   H IDDEN DEBT TABLE 2B.4  Baseline Regressions: Adjustments Given Distress, 2009–18 (continued) b. Adjustment to Fixed Assets and Debt during Bank Distress Fixed assets Debt (1) (2) (1) (2) Distress: ICR < 1 −0.0236 −0.0215 −0.231 −0.247 (−0.57) (−0.54) (−1.92) (−1.70) Distress: ICR < 1 – Lagged −0.0435 0.0492 (−1.14) (0.24) Distress × PSB 0.0929 0.119* 0.182 0.256 (1.91) (2.43) (1.30) (1.53) Distress × PSB – Lagged −0.0547 −0.255 (−1.14) (−1.11) Constant 0.133*** 0.135*** 0.470*** 0.470*** (4.90) (4.94) (4.64) (4.63) Observations 662 661 604 604 R-squared 0.0204 0.0327 0.114 0.120 Year time dummies Yes Yes Yes Yes Bank fixed effects Yes Yes Yes Yes Sample All banks All banks All banks All banks Source: Kibuuka and Melecky 2020. Note: The estimations are performed on Prowess data. t statistics in parentheses. ICR = interest coverage ratio; PSB = public sector bank. * p < 0.05, ** p < 0.01, *** p < 0.001. TABLE 2B.5  Adjustments in Distress Compared with Private Banks Using Alternative Indicators of Distress, 2009–18 a. OLS Regressions with Robust Standard Errors Capital Provisions Debt ROA < 0 CRAR < 11% ROA < 0 CRAR < 11% ROA < 0 CRAR < 11% Distress 0.0233 −0.110 0.340* −0.213 −0.159 −0.105 (1.15) (−1.34) (2.31) (−0.97) (−0.86) (−0.35) Distress lagged −0.00977 0.0465 −0.454* 0.0539 −0.315 0.0676 (−0.55) (1.30) (−2.33) (0.32) (−1.65) (0.18) Distress × PSB 0.173*** 0.111 −0.0693 0.149 0.117 0.144 (4.94) (1.05) (−0.40) (0.68) (0.60) (0.47) Distress × PSB − Lagged 0.0387 −0.0674 0.0878 −0.0233 0.109 −0.110 (0.47) (−1.17) (0.42) (−0.12) (0.50) (−0.29) Constant 0.0456 0.0644* −0.184 0.0119 0.458*** 0.359*** (1.87) (2.54) (−1.92) (0.19) (4.74) (3.76) Observations 633 499 613 496 590 472 R-squared 0.0344 0.0293 0.0835 0.0396 0.121 0.134 Year time dummies Yes Yes Yes Yes Yes Yes Bank fixed effects Yes Yes Yes Yes Yes Yes Sample All banks All banks All banks All banks All banks All banks (continues next page) ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    87 TABLE 2B.5  Adjustments in Distress Compared with Private Banks Using Alternative Indicators of Distress, 2009–18 (continued) b. OLS Regressions with Robust Standard Errors Lending Fixed assets ROA < 0 CRAR < 11% ROA < 0 CRAR < 11% Distress 0.117 −0.0924* 0.0499 −0.0354 (0.51) (−2.24) (0.79) (−0.93) Distress lagged −0.228 −0.0782 −0.0790* −0.117* (−1.70) (−1.17) (−2.64) (−2.50) Distress × PSB −0.142 0.0657 0.0271 0.106 (−0.75) (1.39) (0.36) (1.98) Distress × PSB − Lagged 0.195 0.0540 0.0159 0.136* (1.23) (0.78) (0.41) (2.06) Constant 0.176*** 0.190*** 0.117*** 0.0747*** (6.43) (5.09) (4.66) (5.05) Observations 652 510 652 510 R-squared 0.0952 0.144 0.0487 0.0476 Year time dummies Yes Yes Yes Yes Bank fixed effects Yes Yes Yes Yes Sample All banks All banks All banks All banks Source: Kibuuka and Melecky 2020. Note: The estimations are performed on Prowess data. t statistics in parentheses. CRAR = capital to risk-weighted assets ratio; OLS = ordinary least squares; PSB = public sector bank; ROA = return on assets. * p < 0.05, ** p < 0.01, *** p < 0.001. TABLE 2B.6  Bank Adjustments in Distress across Bangladesh, India, Pakistan, and Sri Lanka, 2009–18 Dependent variable: Total capital Capital Provisions Lending Investment Debt Distress: ICR < 1 −0.0631 0.0466 0.0481 0.0437 −0.182 (−1.78) (0.62) (1.15) (1.07) (−0.78) Distress × PSB −0.0147 0.0159 −0.0488 −0.0659 −0.272 (−0.18) (0.10) (−0.92) (−0.88) (−0.37) Distress: ICR < 1 − (t+1) Lag 0.0648* −0.234* −0.179 −0.0747 0.00983 (2.32) (−2.35) (−1.93) (−1.43) (0.06) Distress × PSB − (t+1) Lag 0.224 0.251 0.171 0.140* −0.917 (1.54) (0.68) (1.80) (2.35) (−1.54) Observations 829 775 828 817 553 R-squared 0.331 0.430 0.535 0.371 0.349 Sources: Kibuuka and Melecky 2020; Fitch Connect database. Note: Pooled estimation for Bangladesh, India, Pakistan, and Sri Lanka. t statistics in parentheses. Regression includes year dummies/country-year dummies and fixed effects. ICR = interest coverage ratio; PSB = public sector bank. * p < 0.05, ** p < 0.01, *** p < 0.001. 88   H IDDEN DEBT TABLE 2B.7  Effect of Borrowing from State-Owned Commercial Banks on Investment by Client Firms, 2009–18   (1) (2) (3) (4) (5) Growth rate of gross fixed assets (t−1) −0.190*** −0.191*** −0.191*** −0.199*** −0.201*** (−18.41) (−18.52) (−18.51) (−19.37) (−19.57) Firm size (log total assets) 0.114*** 0.116*** 0.116*** 0.103*** 0.103*** (10.86) (10.82) (10.82) (9.84) (9.83) Age (years) 0.218*** 0.218*** 0.216** 0.218*** 0.221*** (3.30) (3.30) (3.28) (3.32) (3.36) Growth rate of sales 0.0355*** 0.0358*** 0.0275** 0.0259** 0.0260** (7.48) (7.50) (3.12) (2.93) (2.95) PSB −0.0272* −0.0280* 0.0342* 0.0329* (−1.97) (−2.04) (2.32) (2.24) PSB × Sales growth 0.0110 0.0124 0.0346** (1.04) (1.18) (2.98) PSB × SME −0.210*** −0.207*** (−10.33) (−10.22) PSB × Sales growth × SME −0.0491*** (−4.52) Firm fixed effects Yes Yes Yes Yes Yes Sector-year fixed effects Yes Yes Yes Yes Yes Firm ownership Private Private Private Private Private Observations 26,488 26,142 26,142 26,142 26,142 R-squared 0.384 0.386 0.386 0.395 0.397 Source: Kibuuka and Melecky 2020. Note: The estimations are performed on Prowess data. OLS regression with industry-time fixed effects and firm fixed effects. Standard errors clusters at the firm level, where PSB = 1 if at least one of the banks is a PSB, and 0 otherwise; SME = 1 if the firm is a SME based on India Chamber of Commerce definitions. t statistics in parentheses. OLS = ordinary least squares; PSB = ­public ­sector bank; SME = small and medium enterprise. * p < 0.05, ** p < 0.01, *** p < 0.001. ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    89 Annex 2C. Regression Tables for South Asian Scheduled Commercial Banks: Country Results, 2009–18 TABLE 2C.1  Pooled Data Set for South Asia: Summary Statistics for Scheduled Commercial Banks, 2009–18 Number of Standard 25th 50th 75th a. Public sector banks observations Mean deviation percentile percentile percentile Interest coverage ratio (ICR) 291 1.06 1.59 0.50 0.66 0.87 Total regulatory capital ratio (%) 280 15.95 13.67 11.02 12.65 14.55 Impaired loans (NPLs)/Gross loans (%) 256 10.51 11.30 2.56 6.36 15.13 Return on average assets (ROAA) (%) 276 0.57 1.42 0.22 0.67 1.12 Return on average equity (ROAE) (%) 276 5.87 18.05 3.47 9.34 15.43 Net interest margin (%) 276 3.2 1.7 2.2 2.7 3.6 z-score 292 −16.08 17.52 −19.84 −7.98 −5.91 Loans/Deposits (%) 297 116.81 170.66 69.15 77.06 87.50 Liquid assets/Total assets (%) 297 12.24 11.77 5.86 9.09 14.76 Risk-weighted assets/Total assets (%) 243 56.21 18.72 47.80 58.86 64.33 Gross loans (US$, million) 297 24,911.45 47,320.07 1,640.85 10,127.01 27,884.60 Total assets (US$, million) 297 40,315.98 76,324.62 3,328.45 17,489.58 44,920.93 Number of Standard 25th 50th 75th b. Private banks observations Mean deviation percentile percentile percentile Interest coverage ratio (ICR) 690 5.27 43.08 0.78 1.09 1.89 Total regulatory capital ratio (%) 691 25.20 31.67 12.52 14.99 21.55 Impaired loans (NPLs)/Gross loans (%) 622 6.44 10.09 1.77 4.19 7.18 Return on average assets (ROAA) (%) 649 1.30 1.41 0.75 1.19 1.83 Return on average equity (ROAE) (%) 649 11.49 15.81 6.23 11.21 16.39 Net interest margin (%) 648 4.4 1.8 3.4 4.2 5.1 z-score 694 −41.91 70.70 −41.46 −26.11 −15.64 Loans/Deposits (%) 732 468.23 4,578.58 72.19 85.85 101.14 Liquid assets/Total assets (%) 741 16.21 15.58 6.85 11.40 18.88 Risk-weighted assets/Total assets (%) 591 69.79 19.62 56.60 70.71 81.92 Gross loans (US$, million) 734 2,949.55 8,488.22 269.28 1,170.45 2,270.54 Total assets (US$, million) 741 5,388.02 13,907.40 475.59 1,882.30 3,976.16 Mean c. T-tests on means difference t statistic Interest coverage ratio (ICR) 4.207* (2.56) Total regulatory capital ratio (%) 9.242*** (6.35) Impaired loans (NPLs)/Gross loans (%) −4.069*** (−5.00) Return on average assets (ROAA) (%) 0.724*** (7.10) Return on average equity (ROAE) (%) 5.626*** (4.50) Net interest margin (%) 1.159*** (9.26) z-score −25.83*** (−8.99) Loans/Deposits (%) 351.4* (2.07) Liquid assets/Total assets (%) 3.972*** (4.46) Risk-weighted assets/Total assets (%) 13.58*** (9.39) Gross loans (US$, million) −21,961.9*** (−7.95) Total assets (US$, million) −34,928.0*** (−7.83) Source: Kibuuka and Melecky 2020. Note: The estimations are performed on Prowess data. NPLs = nonperforming loans. * p < 0.05, ** p < 0.01, *** p < 0.001. 90   H IDDEN DEBT TABLE 2C.2  Bangladesh: Summary Statistics for Scheduled Commercial Banks, 2009–18 Number of Standard 25th 50th 75th a. Public sector banks observations Mean deviation percentile percentile percentile Interest coverage ratio (ICR) 40 0.91 0.66 0.48 0.79 1.13 Total regulatory capital ratio (%) 35 9.03 12.39 7.27 10.10 10.63 Impaired loans (NPLs)/Gross loans (%) 41 27.30 14.67 16.73 24.94 35.09 Return on average assets (ROAA) (%) 38 −0.19 2.17 −0.38 0.29 1.10 Return on average equity (ROAE) (%) 38 −5.87 32.36 −9.81 3.63 9.54 Net interest margin (%) 38 2.4 1.3 1.6 2.4 3.3 z-score 40 −13.55 18.93 −15.95 −4.24 −3.10 Loans/Deposits (%) 44 70.16 23.33 60.67 65.26 76.26 Liquid assets/Total assets (%) 44 18.72 7.21 13.76 19.54 23.14 Risk-weighted assets/Total assets (%) 35 62.08 25.86 49.12 59.21 74.34 Gross loans (US$, million) 44 2,762.27 1,710.47 1,507.15 2,605.55 4,305.73 Total assets (US$, million) 44 5,484.82 3,905.07 2,140.28 4,616.16 8,236.91 Number of Standard 25th 50th 75th b. Private banks observations Mean deviation percentile percentile percentile Interest coverage ratio (ICR) 241 2.98 18.40 0.76 1.02 1.47 Total regulatory capital ratio (%) 233 21.85 26.02 11.42 12.71 18.59 Impaired loans (NPLs)/Gross loans (%) 234 6.20 11.08 2.51 4.30 5.73 Return on average assets (ROAA) (%) 227 1.44 1.54 0.79 1.10 1.81 Return on average equity (ROAE) (%) 227 15.22 20.75 8.53 11.63 16.45 Net interest margin (%) 227 4.5 1.9 3.5 4.3 5.2 z-score 245 −37.20 62.65 −37.61 −21.70 −15.54 Loans/Deposits (%) 262 261.88 874.96 79.37 86.55 98.98 Liquid assets/Total assets (%) 266 17.66 14.42 9.81 13.67 20.79 Risk-weighted assets/Total assets (%) 225 79.61 16.59 70.80 80.25 90.10 Gross loans (US$, million) 262 1,165.11 823.29 370.38 1,131.70 1,735.05 Total assets (US$, million) 266 1,684.12 1,168.89 512.35 1,664.22 2,589.33 Mean c. T-tests on means difference t statistic Interest coverage ratio (ICR) 2.071 (1.74) Total regulatory capital ratio (%) 12.82*** (4.75) Impaired loans (NPLs)/Gross loans (%) −21.10*** (−8.78) Return on average assets (ROAA) (%) 1.632*** (4.44) Return on average equity (ROAE) (%) 21.09*** (3.89) Net interest margin (%) 2.044*** (8.24) z-score −23.65*** (−4.73) Loans/Deposits (%) 191.7*** (3.54) Liquid assets/Total assets (%) −1.066 (−0.76) Risk-weighted assets/Total assets (%) 17.53*** (3.89) Gross loans (US$, million) −1,597.2*** (−6.08) Total assets (US$, million) −3,800.7*** (−6.41) Sources: Kibuuka and Melecky 2020; Fitch Connect database. Note: NPLs = nonperforming loans. * p < 0.05, ** p < 0.01, *** p < 0.001. ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    91 TABLE 2C.3 India: Summary Statistics for Scheduled Commercial Banks, 2009–18 Number of Standard 25th 50th 75th a. Public sector banks observations Mean deviation percentile percentile percentile Interest coverage ratio (ICR) 160 0.59 0.15 0.47 0.57 0.69 Total regulatory capital ratio (%) 159 12.14 1.36 11.08 12.23 13.08 Impaired loans (NPLs)/Gross loans (%) 159 6.43 5.94 2.21 4.10 9.65 Return on average assets (ROAA) (%) 159 0.33 0.85 0.16 0.54 0.87 Return on average equity (ROAE) (%) 159 5.82 15.05 3.17 9.41 15.55 Net interest margin (%) 159 2.5 0.5 2.2 2.5 2.8 z-score 160 −7.80 3.23 −8.29 −6.76 −5.80 Loans/Deposits (%) 160 78.24 8.31 73.06 77.63 83.04 Liquid assets/Total assets (%) 160 6.92 4.02 4.08 6.87 9.27 Risk-weighted assets/Total assets (%) 126 60.46 9.91 56.70 60.53 64.33 Gross loans (US$, million) 160 44,352.38 57,786.25 19,694.49 25,694.77 47,855.01 Total assets (US$, million) 160 71,209.07 93,511.33 31,785.23 42,153.50 75,204.03 Number of Standard 25th 50th 75th b. Private banks observations Mean deviation percentile percentile percentile Interest coverage ratio (ICR) 186 12.91 79.91 0.71 1.17 2.05 Total regulatory capital ratio (%) 181 31.95 49.19 13.71 15.38 19.09 Impaired loans (NPLs)/Gross loans (%) 163 4.36 11.23 0.98 1.80 4.62 Return on average assets (ROAA) (%) 176 1.12 1.09 0.70 1.24 1.74 Return on average equity (ROAE) (%) 176 8.23 9.34 3.58 9.61 14.16 Net interest margin (%) 175 4.0 1.5 3.0 3.8 4.8 z-score 186 −45.35 58.66 −45.04 −32.13 −17.85 Loans/Deposits (%) 185 1,325.66 9,009.76 71.39 84.67 102.42 Liquid assets/Total assets (%) 189 13.42 15.93 5.13 8.05 14.92 Risk-weighted assets/Total assets (%) 138 71.81 20.46 60.75 71.04 79.96 Gross loans (US$, million) 187 7,887.81 15,693.37 640.14 2,320.22 7,039.58 Total assets (US$, million) 189 14,164.31 25,061.59 1,166.56 4,663.29 19,520.47 Mean c. T-tests on means difference t statistic Interest coverage ratio (ICR) 12.32* (2.10) Total regulatory capital ratio (%) 19.80*** (5.41) Impaired loans (NPLs)/Gross loans (%) −2.074* (−2.08) Return on average assets (ROAA) (%) 0.787*** (7.41) Return on average equity (ROAE) (%) 2.412 (1.74) Net interest margin (%) 1.493*** (12.10) z-score −37.56*** (−8.72) Loans/Deposits (%) 1,247.4 (1.88) Liquid assets/Total assets (%) 6.501*** (5.41) Risk-weighted assets/Total assets (%) 11.36*** (5.81) Gross loans (US$, million) −36,464.6*** (−7.74) Total assets (US$, million) −57,044.8*** (−7.49) Source: Kibuuka and Melecky 2020. Note: The estimations are performed on Prowess data. NPLs = nonperforming loans. * p < 0.05, ** p < 0.01, *** p < 0.001. 92   H IDDEN DEBT TABLE 2C.4  Pakistan: Summary Statistics for Scheduled Commercial Banks, 2009–18 Number of Standard 25th 50th 75th a. Public sector banks observations Mean deviation percentile percentile percentile Interest coverage ratio (ICR) 37 1.33 1.04 0.77 1.06 1.47 Total regulatory capital ratio (%) 38 23.52 11.56 16.07 21.95 25.72 Impaired loans (NPLs)/Gross loans (%) 37 12.77 8.70 7.07 12.48 16.46 Return on average assets (ROAA) (%) 32 0.93 1.15 0.65 1.06 1.27 Return on average equity (ROAE) (%) 32 7.40 9.89 4.37 9.45 11.80 Net interest margin (%) 32 4.2 1.9 3.1 3.7 4.3 z-score 37 −35.25 25.54 −57.62 −29.91 −11.93 Loans/Deposits (%) 38 169.00 266.74 54.83 61.85 72.40 Liquid assets/Total assets (%) 38 11.95 6.06 8.10 11.36 13.87 Risk-weighted assets/Total assets (%) 38 51.40 18.66 39.15 43.50 59.68 Gross loans (US$, million) 38 1,941.80 2,475.47 343.99 973.91 2,386.87 Total assets (US$, million) 38 4,212.10 5,809.75 1,023.58 1,555.32 4,503.73 Number of Standard 25th 50th 75th b. Private banks observations Mean deviation percentile percentile percentile Interest coverage ratio (ICR) 123 1.20 0.78 0.72 0.98 1.40 Total regulatory capital ratio (%) 135 17.91 11.21 12.52 14.90 19.27 Impaired loans (NPLs)/Gross loans (%) 130 11.70 7.98 6.50 10.66 14.26 Return on average assets (ROAA) (%) 119 0.76 1.43 0.46 0.93 1.41 Return on average equity (ROAE) (%) 119 8.63 17.39 4.62 13.22 17.63 Net interest margin (%) 119 3.7 1.1 2.9 3.6 4.3 z-score 119 −21.56 13.95 −33.25 −20.18 −8.90 Loans/Deposits (%) 135 62.58 22.14 47.43 58.65 74.96 Liquid assets/Total assets (%) 135 12.82 14.22 6.33 8.85 13.13 Risk-weighted assets/Total assets (%) 135 52.17 12.71 43.82 50.70 60.10 Gross loans (US$, million) 135 1,922.54 1,812.03 639.35 1,491.07 2,659.02 Total assets (US$, million) 135 4,965.81 5,132.63 1,288.33 3,548.30 6,952.84 Mean c. T-tests on means difference t statistic Interest coverage ratio (ICR) −0.131 (−0.71) Total regulatory capital ratio (%) −5.615* (−2.66) Impaired loans (NPLs)/Gross loans (%) −1.071 (−0.67) Return on average assets (ROAA) (%) −0.177 (−0.73) Return on average equity (ROAE) (%) 1.229 (0.52) Net interest margin (%) −0.544 (−1.58) z-score 13.70** (3.12) Loans/Deposits (%) −106.4* (−2.46) Liquid assets/Total assets (%) 0.869 (0.55) Risk-weighted assets/Total assets (%) 0.767 (0.24) Gross loans (US$, million) −19.25 (−0.04) Total assets (US$, million) 753.7 (0.72) Sources: Kibuuka and Melecky 2020; Fitch Connect database. Note: NPLs = nonperforming loans. * p < 0.05, ** p < 0.01, *** p < 0.001. ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    93 TABLE 2C.5  Sri Lanka: Summary Statistics for Scheduled Commercial Banks, 2009–18 Number of Standard 25th 50th 75th a. Public sector banks observations Mean deviation percentile percentile percentile Interest coverage ratio (ICR) 54 2.38 3.18 0.66 0.86 1.52 Total regulatory capital ratio (%) 48 27.64 24.84 12.91 14.95 24.86 Impaired loans (NPLs)/Gross loans (%) 19 4.03 5.18 0.00 1.80 7.33 Return on average assets (ROAA) (%) 47 1.76 1.65 0.86 1.18 2.08 Return on average equity (ROAE) (%) 47 14.47 8.48 6.38 12.66 21.62 Net interest margin (%) 47 5.6 2.1 3.9 5.1 6.9 z-score 55 −29.13 17.20 −33.78 −26.18 −15.37 Loans/Deposits (%) 55 230.31 298.39 92.59 99.88 123.30 Liquid assets/Total assets (%) 55 22.74 20.37 9.09 16.11 28.59 Risk-weighted assets/Total assets (%) 44 43.52 24.32 24.29 47.13 60.81 Gross loans (US$, million) 55 1,945.31 2,688.73 128.11 424.57 5,076.80 Total assets (US$, million) 55 3,254.59 4,111.84 163.58 524.81 6,657.75 Number of Standard 25th 50th 75th b. Private banks observations Mean deviation percentile percentile percentile Interest coverage ratio (ICR) 140 2.62 2.63 0.97 1.32 3.49 Total regulatory capital ratio (%) 142 29.00 21.06 15.25 19.66 36.80 Impaired loans (NPLs)/Gross loans (%) 95 3.42 2.82 1.29 2.72 5.14 Return on average assets (ROAA) (%) 127 1.79 1.35 1.16 1.66 2.38 Return on average equity (ROAE) (%) 127 12.02 7.79 7.50 11.84 17.19 Net interest margin (%) 127 5.5 1.8 4.3 5.2 6.1 z-score 144 −62.28 109.98 −69.03 −37.11 −20.36 Loans/Deposits (%) 150 136.27 118.01 87.12 98.75 136.53 Liquid assets/Total assets (%) 151 20.20 17.06 7.13 14.25 28.71 Risk-weighted assets/Total assets (%) 93 68.63 15.56 61.34 67.71 77.58 Gross loans (US$, million) 150 834.30 1,096.94 69.29 330.63 1,242.97 Total assets (US$, million) 151 1,305.37 1,649.87 174.91 496.50 2,072.62 Mean  c. T-tests on means difference t statistic Interest coverage ratio (ICR) 0.237 (0.49) Total regulatory capital ratio (%) 1.368 (0.34) Impaired loans (NPLs)/Gross loans (%) −0.604 (−0.49) Return on average assets (ROAA) (%) 0.0305 (0.11) Return on average equity (ROAE) (%) −2.445 (−1.73) Net interest margin (%) −0.156 (−0.45) z-score −33.15*** (−3.51) Loans/Deposits (%) −94.04* (−2.27) Liquid assets/Total assets (%) −2.534 (−0.82) Risk-weighted assets/Total assets (%) 25.11*** (6.27) Gross loans (US$, million) −1,111.0** (−2.98) Total assets (US$, million) −1,949.2** (−3.42) Sources: Kibuuka and Melecky 2020; Fitch Connect database. Note: NPLs = nonperforming loans. * p < 0.05, ** p < 0.01, *** p < 0.001. 94   HIDDEN DEBT TABLE 2C.6  India: Average Characteristics of the Client Firms of Commercial Banks, 2009–18 Number of Standard 25th 50th 75th a. Public sector banks observations Mean deviation percentile percentile percentile Total assets (Rs, billion) 190 57.95 22.72 41.50 56.28 69.87 Debt to equity 190 5.11 8.00 2.40 3.11 4.54 Debt to assets 190 0.43 0.05 0.40 0.42 0.46 Investment to assets 190 0.48 0.07 0.43 0.47 0.52 Return on assets 190 −0.13 5.11 -0.99 0.27 1.36 Number of Standard 25th 50th 75th b. Private banks observations Mean deviation percentile percentile percentile Total assets (Rs, billion) 387 54.67 71.33 17.64 33.65 60.51 Debt to equity 377 2.83 8.35 0.96 1.59 2.44 Investment to assets 387 0.50 0.29 0.38 0.43 0.55 Return on assets 387 0.29 12.79 0.06 1.74 3.30 Mean  c. T-tests on means difference t statistic Total assets (Rs, billion) −3.28 (−0.82) Debt to equity −2.280** (−3.16) Investment to assets 0.0173 (1.12) Return on assets 0.423 (0.56) Source: Kibuuka and Melecky 2020. Note: The estimations are performed on Prowess data. Rs = Indian rupees. * p < 0.05, ** p < 0.01, *** p < 0.001. Notes typically confined to narrower mandates. The latter include supporting agricultural 1. Based on data from the Bank Regulation and activity in subregions, helping modernize Supervision Survey, World Bank. agriculture and boosting productivity, and 2. Some SOCBs operate as commercial banks the supporting small businesses and the set- while still implementing government pro- ting up of industries. The specialized banks grams, typically funded directly from the in both countries are inefficient and unprofit- budget, so some call them hybrid commercial able and have large pools of NPLs on their banks (Ferrari, Mare, and Skamnelos 2017). books. Only a much better capital position 3. While most of the prominent guidelines distinguishes Pakistan’s specialized banks on corporate governance issued by the from their counterparts in Bangladesh. Organisation for Economic Co-operation 6. A scheduled bank, in India, refers to a bank and Development (OECD) could be extended that is listed in the 2nd Schedule of the to SOCBs, these guidelines are not really tai- Reserve Bank of India Act, 1934. lored to the risk-managing business of hybrid 7. Small finance banks are a type of niche bank commercial banks. See https://www.oecd​ .org​ in India. Banks with a small finance bank /corporate/guidelines-corporate-­governance​ license can provide the basic banking services -soes.htm. of accepting deposits and extending lending. 4. The report uses the World Bank Country and 8. Priority sector lending is imposed by the RBI Lending Groups classifications to group on commercial banks to provide a specified countries by region and income groups. portion of bank lending to a few specific sec- 5. This chapter does not include analogous tors, such as agriculture (notably farm credit, statistics for specialized development banks ­ as well as agriculture infrastructure and ancil- in Bangladesh and Pakistan because they are lary services); micro, small, and medium not typical SOCBs. Although they may con- enterprises; export credit; education; housing; duct retail operations, their financing is much social infrastructure (such as hospitals, less market based, and lending operations are schools, and water and sanitation systems); ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    95 renewable energy; and weaker sectors (such as below 9 percent and have attempted to adjust small and marginal farmers, beneficiaries of way before that time. government-sponsored schemes, minority 16. The chapter uses the government of India’s communities, people with disabilities, and the definition for SMEs: annual turnover of no like). Priority sector lending also includes more than Rs 2.5 billion, or US$33.5 billion. credit lines provided to nonbank financial 17. Even here, it could be argued that any profit- institutions for on-lending to priority sectors. maximizing horizons longer than the market 9. While most banks have data on the key vari- would assume could benefit the banking sys- ables, resulting in about 700 observations in tem in terms of stability and thus generate each regression, data on foreign exposures social benefits. and dividends have a number of missing vari- 18. See, for example, the G20 Principles for ables, resulting in regressions with about 300 Financial Inclusion (https://www.afi-global.org​ and 500 observations, respectively. /sites/default/files/afi%20g20%20principles​ 10. In India, for instance, SOCBs are instructed .pdf); the International Labour Organization’s to keep their CRAR above 11 percent. Rural Policy Brief on “Empowering Rural 11. Old private banks are those that existed Communities through Financial Inclusion” when nationalization led to creating the cur- (https://www.ilo.org/wcmsp5/groups/public rent SOCBs. New private banks have been / ---ed​ ​ _ emp/documents/publication/wcms​ established more recently. See Mishra, _159004.pdf); and Barajas et al. (2020). Prabhala, and Rajan (2019) for details on the 19. This could include new jobs, connectivity to old and new private banks. markets, broadening social networks, and 12. See the Fitch Rating Action dated April 30, other aspects of social inclusion for which a 2020. clear monetary value is difficult to assign. 13. This finding dovetails with that of Sarkar and 20. An example is the integrated digital banking Sensarma (2010) that even partial privatiza- platform YONO offered by the State Bank of tion can help significantly improve the finan- India (https://www.sbiyono.sbi/wps/portal​ cial performance of SOCBs in India. /login). 14. For instance, Ashraf, Arshad, and Yan (2018) find that political pressure on state-owned banks is prevalent only in those countries References with weak political institutions. Strong polit- ADB, DFID, JICA, and World Bank (Asian ical institutions in the form of effective con- Development Bank, United Kingdom straints on policy change decisions by Department for International Development, incumbent government administrations and Japan International Cooperation Agency, and greater democratic accountability are helpful World Bank). 2018. The WEB of Transport in eliminating political pressure on state- Corridors in South Asia . Washington, DC: owned banks in developing countries. Also, World Bank. Richmond and others (2019) report one con- Ashraf B. N., S. Arshad, and L. Yan. 2018. “Do sistent feature across state-owned firms in Better Political Institutions Help in Reducing both the financial sector and real sector: Political Pressure on State-Owned Banks? overemployment relative to their private sec- Evidence from Developing Countries.” Journal tor counterparts. of Risk and Financial Management 11 (3): 15. SOCBs have been instructed by the Indian 1–18. government to maintain capital above 11 Ashraf, B. N., and Y. Shen. 2019. “Economic percent, even though the regulatory mini- Policy Uncertainty and Banks’ Loan Pricing.” mum is 9 percent. We opt for this higher Journal of Financial Stability 44: 100695. threshold because it shows greater sensitivity Atkinson, A. B., and J. E. Stiglitz. 1980. Lectures and richer adjustments than the threshold of on Public Economics. London: McGraw Hill. 9 percent. Anecdotal evidence suggests that B a n e r j e e , A . V. 1 9 9 7 . “ A T h e o r y o f banks typically “dress up” their regulatory Misgovernance.” Quarterly Journal of reporting between on-site supervision visits. Economics 112 (4): 1289–332. By the time they report a CRAR of 9 percent Barajas, A., T. Beck, M. Belhaj, and S. Ben Naceur. or less on audited statements and to off-site 2020. “Financial Inclusion: What Have We supervision, they actually are significantly Learned So Far? What Do We Have to Learn?” 96   H IDDEN DEBT IMF Working Paper WP/20/157, International Policy Research Working Paper 5729, World Monetary Fund, Washington, DC. Bank, Washington, DC. Bertay, A. C., A. Demirgüç-Kunt, and H. Huizinga. Hart, O. D., A. Shleifer, and R. Vishny. 1997. “The 2015. “Bank Ownership and Credit over the Proper Scope of Government: Theory and Business Cycle: Is Lending by State Banks Less Application to Prisons.” Quarterly Journal of Procyclical?” Journal of Banking & Finance Economics 112: 1127–62. 50: 326–39. IMF (International Monetary Fund). 2013. Calomiris, C. W., and S. H. Haber. 2014. Fragile “Changes in Bank Funding Patterns and by Design: The Political Origins of Banking Financial Stability.” Chapter 3 in Global Crises and Scarce Credit . Princeton, NJ: Financial Stability Report: Transition Princeton University Press. Challenges to Stability. Washington, DC: IMF. Cole, S. 2009. “Fixing Market Failures or Fixing Kibuuka, K., and M. Melecky. 2020. “State- Elections? Agricultural Credit in India.” Owned versus Private Banks in South Asia: American Economic Journal: Applied Agency Tensions, Distress Factors, and Real Economics 1(1): 219–50. Costs of Distress.” Background paper for Coleman, N., and L. Feler. 2015. “Bank Hidden Debt. World Bank, Washington, DC. Ownership, Lending, and Local Economic Kumbhakar, S., and S. Sarkar. 2003. “Deregulation, Performance during the 2008–2009 Financial Ownership, and Productivity Growth in the Crisis.” Journal of Monetary Economics Banking Industry: Evidence from India.” 71: 50–66. Journal of Money, Credit and Banking Cull, R., M. S. Martinez Peria, and J. Verrier. 35: 403–24. 2017. “Bank Ownership: Trends and Laeven, L., and R. Levine. 2009. “Bank Governance, Implications.” IMF Working Paper no. 17/60, Regulation and Risk Taking.” Journal of International Monetary Fund, Washington, Financial Economics 93 (2): 259–75. DC. Levy-Yeyati, E., A. Micco, and U. Panizza. 2007. de la Torre, A., J. C. Gozzi, and S. L. Schmukler. “A Reappraisal of State-Owned Banks.” 2007. “Stock Market Development under Economia 7 (2): 209–59. Globalization: Whither the Gains from Mazzucato, M., and C. C. R. Penna. 2016. Reforms?” Journal of Banking and Finance “Beyond Market Failures: The Market Creating 31 (6): 1731–54. and Shaping Roles of State Investment Banks.” de Luna-Martinez, J., and C. L. Vicente. 2012. Journal of Economic Policy Reform 19 (4): “Global Survey of Development Banks.” Policy 305–26. Research Working Paper 5969, World Bank, Melecky, M., and S. Sharma. 2020. “Hidden Washington, DC. Liabilities from State-Owned Enterprises in Duprey, T. 2015. “Do Publicly Owned Banks Lend South Asia.” Background paper for Hidden against the Wind?” International Journal of Debt , World Bank, Washington, DC. Central Banking 11 (2): 65–112. Unpublished. Ferrari, A., D. S. Mare, and I. Skamnelos. 2017. Micco, A., and U. Panizza. 2006. “Bank “State Ownership of Financial Institutions in Ownership and Lending Behavior.” Economics Europe and Central Asia.” Policy Research Letters 93 (2, November): 248–54. Working Paper WPS 8288, World Bank, Mishra, P., N. Prabhala, and R. G. Rajan. 2019. Washington, DC. “The Relationship Dilemma: Organizational Gerschenkron, A. 1962. Economic Backwardness Culture and the Adoption of Credit Scoring in Historical Perspective: A Book of Essays. Technology in Indian Banking.” Johns Cambridge, MA: Harvard University Press. Hopkins Carey Business School Research Gopalakrishnan, B., and S. Mohapatra. 2019. Paper no. 19-03, Johns Hopkins University, “Insolvency Regimes and Firms’ Default Risk Baltimore. under Economic Uncertainty and Shocks.” Perotti, E., and M. Vorage. 2010. “Bank MPRA Paper 96283, Munich Personal RePEc Ownership and Financial Stability.” Tinbergen Archive, University Library of Munich, Institute Discussion Paper TI 2010-022/2, Germany. Tinbergen Institute, Amsterdam. Gutierrez, E., H. P. Rudolf, T. Homa, and E. B. Richmond, C. J., D. Benedek, E. Cabezon, B. Beneit. 2011. “Development Banks: Role and Cegar, P. A. Dohlman, M. Hassine, B. Jajko, P. Mechanisms to Increase Their Efficiency.” Kopyrski, M. Markevych, J. A. Miniane, ST A TE - OWNED B A NKS VERSUS P RIV A TE B A NKS IN SOUT H A SI A    97 F. J. Parodi, G. Pula, J. Roaf, M. Song, M. Shleifer, A., and R. W. Vishny. 1994. “Politicians Sviderskaya, R. Turk Ariss, and S. Weber. 2019. and Firms.” Quarterly Journal of Economics “Reassessing the Role of State-Owned 109 (4): 995–1025. Enterprises in Central, Eastern and Southeastern Srinivasan, A., and A. Thampy. 2017. “The Effect Europe.” IMF Departmental Paper 19/11, of Relationships with Government-Owned European Department, International Monetary Banks on Cash Flow Constraints: Evidence Fund, Washington, DC. from India.” Journal of Corporate Finance Sarkar, S., and R. Sensarma. 2010. “Partial 46 (C): 361–73. Privatization and Bank Performance: Evidence Stiglitz, J. E. 1993. “The Role of the State in from India.” Journal of Financial Economic Financial Markets. World Bank Economic Policy 2 (4): 276–306. Review 7 (suppl. 1): 19–52. Shleifer, A. 1998. “State versus Private World Bank. 2020. South Asia Economic Focus, Ownership.” Journal of Economic Perspectives Spring 2020: The Cursed Blessing of Public 12: 133–50. Banks. Washington, DC: World Bank. South Asia’s State-Owned Enterprises: Surprise Liabilities 3 versus Positive Externalities S tate-owned enterprises (SOEs) in South possible positives of SOE operations, the Asia offer many important benefits. chapter searches for evidence that SOEs They provide public goods and help could provide strategic direction in their address market failures related to risky, long- industries when they undertake riskier long- term investments and natural monopolies. term investments, such as into research and However, because their operations and development (R&D). In other words, can liabilities are backed by government guaran- ­ SOE investments in R&D crowd in addi- tees, they also expose governments to large tional R&D investment of private firms in financial risks and potential (contingent) lia- the same industry? bilities. Using firm-level panel data from India, this chapter assesses whether SOEs are more prone to financial distress than compa- The Importance of Paying More rable private firms—and thus impose unfore- Attention to the Hidden Liabilities seen liabilities and expenditure needs on the of SOEs in South Asia governments. It further studies whether Nonfinancial state-owned enterprises (SOEs) SOEs’ financial distress relates to the persis- have a large footprint in South Asia. Total SOE tent underperformance and indebtedness of revenues amount to nearly 8 percent of GDP some SOEs or to the greater risks that SOEs in Sri Lanka, 12 percent in Pakistan, and confront compared to private firms. Drilling 19 percent in India (see table 3A.1).1 These deeper, the chapter tests alternative hypoth- shares are significant by international stan- eses for the underperformance of SOEs, dards, although some other countries—​ including weak corporate governance and ­ particularly formerly socialist countries of soft budget constraints in the form of both Eastern Europe and East Asia—have much debt and equity bailouts. To illustrate some larger SOE sectors. The total number of SOEs Note: This chapter draws on the background research paper: Melecky, M., S. Sharma, and D. Yang. 2020. “State- Owned Enterprises: The Distresses, Adjustments, and Fiscal Contingent Liabilities in South Asia.” Background paper for Hidden Debt. World Bank, Washington, DC. 99 100   H IDDEN DEBT exceeds 200 in Pakistan, 400 in Sri Lanka, and 1,300 in India. Although present in nearly all Many individual SOEs are persistent loss- sectors of the economy, they concentrate in the makers and financially unsustainable. energy, transport, utilities, and trading sectors. While the performance of SOEs should not Rationales for government involvement be judged solely on commercial terms, in SOEs. One major rationale for govern- these large and persistent SOE losses often ment ownership of firms in South Asia, as in other parts of the world, is the need to culminate in costly government bailouts. address market failures related to natural monopolies. Network industries in energy risk that coordination failures between pri- and transportation have high fixed costs, vate firms and skills providers could block often leading to their monopolization. In the supply of needed skills.3 The public sec- such cases, an unregulated private sector tor could play a role in breaking the dead- cannot be relied upon to produce goods and lock. Not surprisingly, many SOEs in South services efficiently and affordably. Public Asia have been early leaders in technical sector ownership could improve welfare if fields. For example, RITES India, the engi- the government’s capacity to regulate the pri- neering arm of the Indian Railways vate sector is limited (Stigler 1971; Peltzman Corporation, today provides infrastructure 1976; Dal Bo 2006). This could be why consulting services in India and abroad.4 South Asian SOEs are concentrated in net- South Asian SOEs also serve broader work sectors such as energy, transport, and developmental or public interest objectives. communication, where the potential for such The economic rationale for public ownership market failures is high. is less compelling in these situations. A case in Another major rationale for public own- point is the mobilization of SOEs to improve ership is that long-term and risky invest- the connectivity of remote or underserved ments in innovation are underfunded by areas. For example, India’s flagship BharatNet private investors because failures in financial Program, which aims to extend the reach of markets limit funding (Hall and Lerner the telecom network to remote and rural vil- 2010). In Europe, SOEs tend to invest more lages and is one of the largest rural connectiv- in R&D than private firms, particularly in ity projects of its kind in the world, is sustainable technologies with low commer- implemented in part by a state-owned enter- cial returns (Bortolotti, Fotak, and Wolfe prise, BBNL (Bharat Broadband Network 2019). Some SOEs in South Asia are under- Limited) (BBNL 2019). Pakistan Railways taking such investments. For example, the subsidizes select routes to provide mobility Solar Energy Corporation of India has been and connectivity to far-flung areas a pioneer in the commercialization of solar (Government of Pakistan 2016). power—an investment with high risk and How important are financial risks of potential positive spillovers.2 Our analysis of SOEs to governments? Although SOEs can firm-level data from India indicates that contribute to development objectives, they are SOEs on average spend more on R&D than a source of financial risk to South Asian gov- private firms do—and account for a dispro- ernments. Moreover, not all the risk is justi- portionate share of total R&D spending in fied for achieving the development objectives several industries. Moreover, the R&D activ- through state ownership. Owning firms ities of SOEs in South Asia have positive exposes government budgets and debt posi- spillovers on the performance of private tion to a host of external, macroeconomic, firms, regression analysis suggests. and sector-specific shocks that depend on the Relatedly, SOEs can complement the pri- industry profile of the SOE sector. To an vate sector by helping solve coordination extent, this risk is an unavoidable side effect problems. An example is a case of a strong of public ownership of firms. However, South complementarity between new technologies Asia’s SOE sectors regularly generate big and specialized skills in an industry, with the losses. For example, the SOE sectors of SOUT H A SI A ’ S ST A TE - OWNED ENTER P RISES   101 Pakistan and Sri Lanka generated net losses in loss-making SOEs—prone to distress—are two out of the three years between 2015 and present throughout the region. 2017, and India’s state public sector enter- Next, this study assesses the magnitude of prises (SPSEs), the SOEs owned by subna- the contingent liabilities arising from SOE tional governments, lost an amount equal to operations for the government fiscal stance 0.5 percent of GDP in 2017. (budget and debt positions). This requires For example, in 2014 the Sri Lankan gov- information on the government’s financial ernment had to inject SL Rs 123 billion commitments in case of SOE distress and the (approximately 1.2 percent of GDP) from the conditions under which they are triggered. budget into SOEs. 5 In the same year, the Such information is not easily available. Even Indian government approved a total assis- explicit government commitments, such as tance package of Rs 411 billion (approxi- guarantees on SOE debt, are not always well mately 0.3 percent of GDP) to revive 46 documented. The data available suggest that “sick” federal government–owned SOEs explicit commitments are sizable. For example, (Government of India 2014). SOE loans amounting to 1 percent of GDP were under government guarantee in Pakistan Many individual SOEs are persistent loss- in 2017. The implicit government commitment to cover SOE debt is even harder to quantify makers and financially unsustainable, often due to data limitations.6 This study presents culminating in costly government bailouts. approximate upper-bound estimates by assess- ing the total liabilities of SOEs in high likeli- Analyzing fiscal contingent liabilities hood of needing bailout funding. The from government ownership of SOEs. The upper-bound estimate is large. For example, main part of this chapter analyzes the fiscal the total liability of all chronically distressed contingent liabilities from government owner- Indian national-level CPSEs has ranged, since ship of firms (SOEs) in South Asia. The analy- 2008, between 3 percent and 5 percent of sis relies mainly on a detailed firm-level panel national GDP. This figure excludes subnational data set for India that has good coverage of SOES, the SPSEs, which are in a much worse firms that are majority privately owned as shape. Pakistan’s numbers are even more con- well the SOEs majority owned by the federal cerning. In the past five years, the total liabili- government (central public sector enterprises, ties of loss-making SOEs in Pakistan has or CPSEs). It is the only such data set avail- hovered around 12 percent of GDP. able for the South Asian countries. This firm- The proximate cause of these contingent level analysis is supplemented with more liabilities is the persistent financial underper- aggregate data from reports published by formance of SOEs. The Indian panel data set South Asian governments. shows that on average, CPSEs earn signifi- This analysis begins by assessing the inci- cantly less revenue per unit labor and per unit dence of SOE distress. A firm is defined to be capital than private firms. They also have sig- in distress when its earnings are not enough to nificantly higher debt-to-asset ratios than cover its interest payments. The interest cover- other firms. These findings are largely in line age ratio (ICR) is used to capture this relation- with the existing evidence base.7 ship. The firm-level data from India show that Explaining the underperformance of SOEs. Indian CPSEs are generally more likely to be in Why do SOEs underperform? One school of distress than comparable private firms. This is thought ascribes underperformance to an not because CPSEs are concentrated in particu- internal “agency problem” (Ehrlich et al. lar sectors where profit margins are lower, nor 1994): It is more difficult to align the incen- because they are engaged in inherently more tives of managers and owners in the public sec- risky activities. Yet CPSE distress tends to last tor because pay scales are more compressed, longer, often exceeding one year. While data job security is higher, and employee monitor- limitations preclude a similar analysis for other ing is less rigorous compared with the private countries, this study finds that persistent sector. Interestingly, South Asian governments 102   H IDDEN DEBT have been “corporatizing” SOEs to profession- losses and the debt-asset ratio among SOEs. alize their management and make it easier to This suggests that SOEs are more prone to monitor their performance by strengthening covering losses through additional loans. their corporate governance. For example, the Given that direct government loans are only a Indian government issued corporate gover- fraction of total SOE debt, SOEs must have nance guidelines for CPSEs in 2010 and now preferential access to bank loans. This opens a rates CPSEs on compliance with the guidelines question about the nexus between state-owned (Government of India 2010).8 We find that enterprises and state-owned banks, and more although there is a positive cross-sectional cor- broadly, the preference of even private banks relation between CPSEs’ corporate governance for lending to SOEs compared with similar pri- ratings and their commercial performance, vately owned companies. improvements in these ratings over time are Exploring the soft budget constraint hypoth- not significantly associated with improved per- esis. We explore the soft budget constraint formance. Therefore, insufficient evidence hypothesis further by comparing how CPSEs exists to conclude that corporatization has a and private firms in India adjust their assets and causal impact on SOE performance. liabilities when in financial distress or when Another, perhaps related, hypothesis is that experiencing a revenue shock. The growth of SOEs underperform private firms because they fixed assets declines significantly for private operate in a more constrained or distorted firms when they are in distress, but this relation- environment.9 For example, they could be ship is significantly weaker among CPSEs. under pressure to hire excess workers (Shleifer Similarly, the growth of debt and equity capital and Vishny 1994). Indeed, our analysis of SOE declines significantly for private firms when they performance measures suggests that CPSEs, on experience a negative shock, but does not for average, overemploy labor and capital. CPSEs. Correspondingly, the growth rate of Similarly, Baird et al. (2019) exploit a natural debt increases more for private firms when they experiment in India to show that excessive hir- experience a positive shock than for CPSEs. ing by SOEs has caused a high level of labor The lower sensitivity of SOE assets, debt, misallocation in the manufacturing sector. and equity to shocks is consistent with the SOE pricing decisions are also constrained. hypothesis that SOEs enjoy greater access to For example, Indian SOEs in industries such soft funding because governments implicitly as petroleum, electricity distribution, gas, and guarantee to bail them out. Banks can afford to fertilizers have had to charge below-cost discount shocks when assessing SOE creditwor- prices to subsidize consumers and farmers thiness if they believe there is an implicit govern- (Khanna 2012). In 2016, the electricity tariff ment guarantee on SOE debt. In China, implicit for agricultural use was about 31 percent of government guarantees on SOEs are found to the cost of supply, while the tariff for the resi- have reduced the sensitivity of SOE credit costs dential sector was 77 percent of the average to risk (Allen et al. 2017) and to make SOE debt cost of supply (Zhang 2019). Likewise, more attractive to financial markets (Jin, Wang, Pakistan Railways subsidizes select routes to and Zhang 2018). Credit rating agencies explic- provide mobility and connectivity to far-flung itly include the likelihood of government sup- areas (Government of Pakistan 2016). port and bailout when assessing the credit risks Another potential external factor is the of SOEs.10 Similarly, the lower sensitivity of much-studied soft budget constraint of SOEs— equity investment in SOEs to shocks could be the perception that SOEs have the implicit and due to the belief among private investors that unconditional support of the government their equity is implicitly insured. Similar issues (Kornai 1986). For example, SOEs might have arise in the context of public-private partner- access to softer loans with lower interest rates ships (see chapter 1). and looser conditions, as suggested by their Although this soft budget constraint is not persistently higher debt ratios. We find that, the only potential cause of SOE underperfor- compared with private firms, there is a stron- mance, it is likely to have been pivotal to the ger positive association between accumulated growth of unnecessary contingent liabilities SOUT H A SI A ’ S ST A TE - OWNED ENTER P RISES   103 from SOEs. Not only does it cause SOEs to various South Asian countries, such as its size, underperform by reducing market pressures performance, and liabilities. on SOE managers (Jensen 1986; Maskin and •  India. The main official data source for Xu 2001), but it also enables debt to build up Indian CPSEs are annual reports on CPSEs in loss-making SOEs. The constrained envi- published by the Department of Public ronment of SOEs could also temper the effec- Enterprises. Data on SOEs owned by the tiveness of internal SOE reforms. For example, state governments of India (state public Berkowitz, Ma, and Nishioka (2017) argue sector enterprises, SPSEs) are less easily that the apparent efficiency gains from SOE available. Our main data sources were corporatization in China were in fact due to a state-level SOE audit reports published by contemporaneous tightening of their operat- the Comptroller and Auditor General of ing environment. Bartel and Harrison (2005) India (CAGI). show that the effectiveness of partial SOE •  Pakistan . Data on Pakistani SOEs are privatization in Malaysia depended on exter- from the annual Federal Footprint: SOEs nal factors such as access to soft loans. Annual Report, published by the Ministry Policy lessons and implications. The most of Finance. immediate policy lesson of our analysis is that •  Sri Lanka. Sri Lankan data are from the the contingent liabilities from SOEs are non- annual SOE Performance Report, pub- transparent, and policy makers in South Asia lished by the Department of Public are not paying enough attention to them. Enterprises, and from the Annual Report Given the limitations of publicly available of the Ministry of Finance. These data data, it is difficult to quantify even SOEs’ were supplemented by a publicly available total liabilities or debt, much less their explicit database compiled from various govern- and implicit government commitments. ment reports by the independent think Governments must better assess and monitor tank Advocata Institute. the fiscal risks from SOEs, incorporate them •  Bangladesh. Bangladesh does not produce into their fiscal planning and debt manage- annual SOE reports. Our data for ment, and make funding provisions so that Bangladesh are based on the statistical SOE distress and rescue, when justified, does tables published in the annual Bangladesh not entail serious disruptions to critical pub- Economic Review, produced by the lic spending. Ministry of Finance. The deeper policy question is how to •  Bhutan. Data on Bhutanese SOEs are from ­ mitigate unnecessary contingent liabilities the State Enterprises Annual Report, pub- stemming from SOE operations. The evidence lished by the Ministry of Finance. presented in this chapter leads us to recom- mend a combination of internal reforms at Annex 3A lists these data sources and the the SOE level and external reforms in the country-specific definition/categorization of operating and broader controlling environ- SOEs used in this report. ment. These reforms are discussed in the final The data available in these official reports part of the chapter. are generally at an aggregate level. India, Pakistan, and Sri Lanka have been publish- ing SOE-level revenue and balance sheet Describing the Opaque and data in recent years. However, data for only Complex SOE Sector in South Asia a limited set of variables are available. For Using Data example, it is not always possible to measure value added and profits or to obtain SOE- Analyses Must Cope with a Lack of Data level information on government support. about South Asian SOEs India publishes firm-level data on CPSEs, This chapter relies mainly on publicly avail- but not SPSEs. This gap is worrying because able official reports and statistical tables to much of the debt and accumulated losses present stylized facts about the SOE sector in reside in SPSEs. 104   H IDDEN DEBT Given the limited availability of official are under the federal government (CPSEs) firm-level data, our firm-level analysis for and more than 1,000 SOEs under state gov- South Asia relies on an Indian firm-level data- ernments (SPSEs). Together, they generate base called Prowess, which is maintained by revenue equal to 19 percent of GDP. In the Center for Monitoring the Indian Bhutan, most strikingly, SOE revenues equal Economy (CMIE). Prowess contains detailed 38 percent of GDP.13 information on the balance sheet and perfor- Although economically significant, South mance of Indian firms, including a large share Asia’s SOE sector is not a global outlier in of India’s CPSEs. It is based on data reported terms of size. Many formerly socialist coun- by firms registered with the Registrar General tries still have a larger SOE sector than most of Companies. While Prowess is not fully rep- South Asian countries do. For example, resentative of the formal manufacturing sec- despite a major privatization drive in the pre- tor in India, it has good coverage of medium vious two decades, China still had more than and large firms. On average over the years, 150,000 SOEs in 2019; their total value Prowess has covered 80 percent to 90 percent added comprised about 20 percent of of the Indian CPSE sector by number and national output (Harrison et al. 2019). SOEs more than 95 percent of it by total revenue. It also have a major presence in much of is a panel data set, unlike the repeated cross- Central, Eastern, and Southern Europe, sectional Annual Survey of Industries. This accounting for more than 15 percent of value enables us to track CPSEs over time and look added in Belarus, Poland, and Russia. This at their debt dynamics. region has a total of 51,000 SOEs; Russia Prowess data consist of an unbalanced alone has more than 30,000 (Richmond et al. panel covering the period 1989–2018 with an 2019). uneven (growing) number of firms over time. South Asian SOEs are concentrated in To ensure that the sample used in the regres- energy, utilities, transport, and telecommunica- sion analysis stays constant across different tions. This concentration is most stark in the regressions with different outcome variables, case of Pakistan, where the energy and trans- we excluded observations that are missing val- port sectors together account for 95 percent of ues for any of the key variables needed for our SOE revenues (figure 3.2). For Sri Lanka, this analysis. The average sample size of this data share is 84 percent. While a precise breakdown set is about 12,000 firms per year.11 We further of SOE revenue or investment by sector is not cleaned the data set by replacing the values of available for Bangladesh and Indian SPSEs, outliers (those exceeding the 98th percentile) they too confirm to this pattern. In Bangladesh, with the value for the 98th percentile. the government has a monopoly on water and Table 3B.1, in annex 3B, presents summary sewage services and dominates the energy sec- statistics of key variables in Prowess for 2016. tor through the Bangladesh Oil, Gas, and CPSEs constitute about 1 percent of the Mineral Corporation (PETROBANGLA), Prowess sample. Bangladesh Power Development Board (PDB), and Bangladesh Petroleum Corporation (BPC) (World Bank 2019a). Two-thirds of the invest- The SOE Sector in South Asia Is Large ment of Indian SPSEs is concentrated in elec- and Complex tricity generation and distribution, with the South Asia has a sizable nonfinancial state- rest spread unevenly in manufacturing, finance, owned enterprise sector. Sri Lanka has 400 and infrastructure. nonfinancial SOEs, with the total revenue of SOEs are also present in manufacturing, the 42 largest “strategically important” nonfi- services, and other sectors, such as the pro- nancial SOEs equal to 8 percent of the GDP curement and distribution of agricultural com- (figure 3.1).12 Pakistan has more than 200 modities. Indian CPSEs are particularly such firms, and their revenue amounts to diverse; the manufacturing and services sector 12 percent of GDP. India has 331 SOEs that account for 62 percent and 20 percent of their SOUT H A SI A ’ S ST A TE - OWNED ENTER P RISES   105 FIGURE 3.1  Total Number and Average Revenue of South Asian State-Owned Enterprises, 2017 1,200 45 1,074 38 40 Average revenue as a percent of GDP 1,000 35 800 30 Number of SOEs 25 600 20 400 400 13 331 12 15 9 8 10 200 6 171 5 45 0 38 0 India CPSEs India SPSEs Pakistan Sri Lanka Bangladesh Bhutan Number (left scale) Revenue (right scale) Source: World Bank staff compilations based on data from government reports, various years (see details in annex 3A). Note: The total number of SOEs is as of 2017. Revenue is expressed as a percentage of GDP. Data are averaged over 2015–17 to smooth out annual ­fluctuations, except for Indian SPSEs and Sri Lanka, where data are available only for 2017. CPSEs = central public sector enterprises; SOEs = state-owned ­enterprises; SPSEs = state public sector enterprises. total revenue, respectively. CPSEs are also FIGURE 3.2  State-Owned Enterprise Revenue by Sector in India, present in mining, energy, and transport sec- Pakistan, and Sri Lanka, 2016–17 tors. Bangladesh and Sri Lanka SOEs have a 100 significant presence in the manufacturing, ser- 6 2 4 1 vices, energy, and transportation sectors. In 20 Percent of total SOE revenues 80 Bangladesh, for example, while many indus- 13 trial SOEs have been privatized, the state still 62 retains ownership of manufacturing compa- 60 nies such as the Bangladesh Jute Mills 89 Corporation, Bangladesh Steel & Engineering 40 Corporation, and Bangladesh Textile 64 20 Mills Corporation (BTMC). 20 The SOE sector is often an aggregate loss- 10 maker. The limited data on the commercial 7 0 1 performance of the SOE sector suggests that 0 India CPSEs Pakistan Sri Lanka the total profitability of the sector varies across countries and over time, with loss- Agriculture Energy Services and Trading Mining Manufacturing Transportation Other making years not uncommon (figure 3.3). In recent years, India’s CPSE sector and Source: World Bank staff compilations based on data from government reports, various years (see details in annex 3A). Bangladesh’s SOE sector have consistently Note: CPSEs = central public sector enterprises; SOE = state-owned enterprise. generated net profits. Pakistan’s SOE sector generated a net profit in 2014 and 2015, but a net loss in the next two years. Pakistan’s SOE Sri Lanka’s SOE sector generated losses in sector also shows a tendency toward rapidly 2015 and 2017. Indian SPSEs operated at a declining profitability in recent years, with its total loss in 2017. The losses can be large: net income dropping at an annual rate of Pakistan’s SOE sector and India’s SPSE sector 57 percent on average from 2014 to 2017. each lost about 0.6 percent of GDP in 2017. 106   H IDDEN DEBT FIGURE 3.3  Net Profit/Loss of South Asian State-Owned Enterprises, 2014–17 1.2 1.03 0.77 0.75 0.74 0.73 0.8 0.68 0.49 Percent of GDP 0.4 0.29 0.34 0.22 0.19 0 –0.21 –0.15 –0.4 –0.32 –0.52 –0.60 –0.8 2014 2015 2016 2017 India CPSEs Pakistan Bangladesh Sri Lanka India SPSEs Source: World Bank staff compilations based on data from government reports, various years (see details in annex 3A). Note: Net profit/loss for Bhutan was around 11 percent on average over 2014–17. CPSEs = central public sector enterprises; SOEs = state-owned enterprises; SPSEs = state public sector enterprises. Although the fraction of SOEs making a 35 percent across South Asian countries loss is sizable, a few large SOEs in specific figure 3.4). However, in every country stud- (­ sectors often account for a large share of the ied, the top 10 loss-making SOEs accounted total loss. The share of SOEs that reported a for more than 80 percent of the total SOE sec- loss in 2017 ranged from 24 percent to tor loss. The big loss-making SOEs were heavily concentrated in energy, utilities, trans- FIGURE 3.4  Share of State-Owned Enterprises That Reported a portation, and telecommunications. Loss in India, Pakistan, Sri Lanka, and Bangladesh, 2017 The persistence of chronic loss-makers among South Asian SOEs is also notable. We 40 do not have the data to track SOE-level prof- Percent of loss-making SOEs to total number of SOEs 35 its over time, except in the case of Pakistan 35 and India for CPSEs. In both countries, of the CPSEs that generated a loss in any one of the 30 29 five years preceding 2018, more than half had 26 generated losses in three of those five years. 25 24 24 About 20 percent had generated losses in all five years (figure 3.5). 20 South Asian SOEs receive significant financial inflows from their governments in 15 the form of equity injections, grants, and loans. Comprehensive data on government 10 financial outlays on SOEs are not easily available. We often had to combine snippets 5 of data from various sources to get a fuller picture. We are still not sure whether we have 0 the full picture: for example, there are no fig- India CPSEs India SPSEs Pakistan Sri Lanka Bangladesh ures on capital injection in Pakistan’s SOE Source: World Bank staff compilations based on government reports, various years (see details in reports, and we are not sure whether this is annex 3A). truly because the government has not injected Note: Data are for 2017. CPSEs = central public sector enterprises; SOEs = state-owned enterprises; SPSEs = state public sector enterprises. any capital into SOEs in recent years. SOUT H A SI A ’ S ST A TE - OWNED ENTER P RISES   107 Similarly, we could not obtain figures for FIGURE 3.5  Breakdown of Loss-Making State-Owned Enterprises government loans to Sri Lankan SOEs. Yet, in India and Pakistan by the Total Number of Years in Which They the available data suggest that government Made a Loss, 2012–17 outlays are substantial. For example, on 100 average during 2015–17, Pakistani SOEs 22 Percent of loss-making SOEs 27 received government loans and grants 80 amounting to 0.9 percent and 0.8 percent of 19 28 GDP, respectively (figure 3.6). Indian SOEs 60 received grants and subsidies amounting to 15 0.7 percent of GDP. 40 22 13 20 9 In every country studied, the top 10 loss- 25 18 making SOEs accounted for more than 0 80 percent of the total losses in the SOE sector. Indian CPSEs Pakistan In all 5 yrs In 4 of 5 yrs In 3 of 5 yrs In 2 of 5 yrs In 1 of 5 yrs Source: World Bank staff compilations based on data from Prowess and government reports for the Analyzing the Roots and Extent of five years preceding 2018, various years (see details in annex 3A). Hidden Liabilities in South Asian Note: CPSEs = central public sector enterprises; SOEs = state-owned enterprises. SOEs Does owning firms expose governments This section studies the risk of distress in to risks just like it would any owner—with SOEs using regression analysis, by contrasting the only difference being the size and sec- public and private ownership of firms while toral concentration of government owner- controlling for important firm characteristics. ship? The answer to this question has It then assesses the possible extent of contin- implications for how we think about the gent liabilities from SOEs that could affect the potential liabilities from SOEs. Thus, we central government fiscal stance—either explore the question by using the firm-level through the budget, debt, or combination of data from India and examining whether the two. CPSE distress looks similar to distress among private firms. CPSEs are more likely to be in distress SOEs’ Risk of Distress Must Be than similar private firms. Figure 3.7 plots Thoroughly Quantified the share of firms in distress in recent years. Because the government’s contingent liabili- Between 2004 and 2017, the incidence of dis- ties from SOEs originate in SOEs’ financial tress among CPSEs ranged between distress, we begin our analysis of hidden lia- 30 percent and 40 percent, compared to bilities by quantifying the risk of distress. This between 20 percent and 30 percent for non- part of the analysis is based on the Prowess SOEs. The gap in the rate of distress between panel data set of Indian firms. CPSEs and non-SOEs has been narrowing in Our main indicator of distress is based on recent years, but it is too early to say whether the interest coverage ratio (ICR): the ratio of this signals a more permanent improvement earnings (before interest and taxes) to inter- in the relative performance of CPSEs. Distress est payment. We define a firm as being in dis- among non-SOE Indian firms has been rising tress if earnings are not enough to cover consistently since 2010, and until 2015, also interest payments: that is, if the ICR is less rose for CPSEs. For reasons that are not obvi- than 1. This is a typical measure of distress ous, the distress rate among CPSEs featured in the financial literature. We explored other in the Prowess database dipped sharply measures, such as those based on debt and between 2015 and 2017. other liabilities, with qualitatively similar The “excess distress rate” of CPSEs—that results. is, the gap in the incidence of distress between 108   H IDDEN DEBT FIGURE 3.6  Average Annual Government Support for South Asian State-Owned Enterprises, 2015–17 1.0 0.9 0.8 0.7 Percent of GDP 0.5 0.5 0.3 0.2 0.2 0.2 0.1 0.1 0.01 0 n n n s n s s s s s s s n s s an an an ie an ie ie ie an ie io io io io io id id id id id ct ct ct ct Lo Lo Lo ct Lo Lo bs bs bs bs bs je je je je je l in l in l in l in su l in su su su su d ta ta ta ta d d d nd ta an an an an pi pi pi pi pi a Ca Ca Ca Ca Ca ts ts ts ts ts an an an an an Gr Gr Gr Gr Gr India SOEs India SPSEs Pakistan Sri Lanka Bangladesh Source: World Bank staff compilations based on data from government reports, averaged over 2015–17 (see details in annex 3A). Note: Indian SOEs include both CPSEs and SPSEs. CPSEs = central public sector enterprises; SOEs = state-owned enterprises; SPSEs = state public sector enterprises. FIGURE 3.7  Share of Distressed Firms in India, 1989–2017 60 50 Percent of all firms 40 30 20 10 0 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 CPSEs Non-SOEs Source: Melecky, Sharma, and Yang 2020. Note: Firms are considered distressed if their interest coverage ratio (ICR) is less than 1 in a given year. CPSEs = central public sector enterprises; SOEs = state-owned enterprises. CPSEs and non-SOEs—rises when we con- distressed non-SOEs has ranged between sider more persistent distress measures. For 5 and 13 percent. example, figure 3.8 presents the share of firms The excess distress among CPSEs is even with an ICR below 1 in the three consecutive higher after adjusting for differences in previous years. Based on this definition, the ­ attributes such as size and sector. Table 3B.2 share of persistently distressed CPSEs has presents estimates of the probability of dis- ranged between 10 percent and 20 percent in tress (ICR below 1) in the Prowess panel using recent years. The share of persistently ordinary least squares (OLS) regression. SOUT H A SI A ’ S ST A TE - OWNED ENTER P RISES   109 FIGURE 3.8  Share of Persistently Distressed Firms in India, 1991–2017 60 50 40 Percent 30 20 10 0 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 CPSEs Non-SOEs Source: Melecky, Sharma, and Yang 2020. Note: The excess distress rate is the gap in distress incidence between CPSEs and non-SOEs. The figure shows the percent of firms with an interest coverage ratio (ICR) of less than 1 in three consecutive previous years. CPSEs = central public sector enterprises; SOEs = state-owned enterprises. The “raw” excess distress rate in CPSE— example, controlling for size, age, and sector, that is, without adjusting for size and other CPSEs are 21 percentage points more likely attributes—is 14.5 percentage points to be in distress in two consecutive years ­ (column 1). This estimate is roughly consis- ­(column 5). tent with figure 3.7. Adding the size control makes a difference, with the CPSE gap in dis- The Likely Magnitude of Contingent tress incidence rising to 20.7 percentage (Potential) Liabilities from SOEs Must Be points. Larger firms tend to have lower dis- Established tress rates, and hence the CPSE gap becomes larger once we adjust for the fact that CPSEs The total liabilities of SOEs are quite large. As are larger than the average non-SOEs. While shown in figure 3.9, the total liabilities of this result does not imply a causal relation- SOEs in Sri Lanka exceeded 10 percent of ship between ownership and distress, it sug- GDP and were 20 percent of GDP for SOEs gests that the higher vulnerability of CPSEs in Pakistan and CPSEs in India in 2017. can be explained only by factors other than While the total liabilities of Indian SPSEs are size, age, and sector. not available for recent years, their total debt Next, we interact the CPSE indicator with (a component of total liabilities) amounted to broad sector dummies to examine whether 4 percent of GDP in 2017. If we assume that the excess distress rate of CPSEs is higher in their ratio of debt to total liabilities is the particular sectors (column 4). We find that same as that of Indian CPSEs, then their total relative to CPSEs in transport and services liabilities amount to 8 percent of GDP. The (the omitted sector dummy), those in the total SOE liabilities in Bangladesh are about manufacturing sector are more prone to 6 percent of GDP, reflecting the smaller size of excess distress, while those in the petroleum its SOE sector. industry are less prone. Government guarantees on SOE loans The patterns shown in table 3B.2 hold have turned a portion of these liabilities into true even when we use more persistent dis- an explicit contingent liability of South Asian tress measures (such as ICR being less than 1 governments. For example, in 2017, the stock in two or three consecutive years). For of SOE debt with a government guarantee 110   H IDDEN DEBT FIGURE 3.9  Total Liabilities and Debts for South Asian State-Owned Enterprises, 2017 30 25 25 21 20 Percent of GDP 15 15 11 10 8 7 6 6 4 5 0 India CPSEs India SPSEs Pakistan Sri Lanka Bangladesh Bhutan Liabilities Debt Source: Melecky, Sharma, and Yang 2020 (see details in annex 3A). Note: CPSEs = central public sector enterprises; SPSEs = state public sector enterprises. FIGURE 3.10  Outstanding Government Guarantees to State- official statistics, are comparatively low (less Owned Enterprises, 2015–17 than 0.1 percent of GDP). The implicit contingent liabilities from 5.18 5 SOE operations are harder to estimate because we do not know what portion of 4 SOE liabilities is implicitly guaranteed and Percent of GDP how the guarantee is triggered. Econometric 3 estimation of a country-level model of SOE bailouts is not possible because large bailout 2 1.72 events are too infrequent, and there are too 1.01 few counties in the region, while firm-level 1 data on government financial support to 0.01 SOEs—which could be used to estimate a 0 firm-level econometric model of bailouts—are India SPSEs Pakistan Sri Lanka Bangladesh not available.14 Source: Melecky, Sharma, and Yang 2020. We can get the upper-bound estimates of Note: The outstanding stock of government guarantees to SOEs in Sri Lanka is for the year 2019 and based on Note 33-A (Statement of Bank Guarantees Issued by Central Treasury) of the annual overall contingent liabilities by examining the report of the Ministry of Finance. SOEs = state-owned enterprises; SPSEs = state public sector total liabilities of SOEs in “distress” in a typi- enterprises. cal year and assuming that the loss to govern- ment is 100 percent of the distressed SOE’s behind it added up to 1 percent of GDP in liabilities. Figure 3.11 uses Prowess data to Pakistan (figure 3.10). Guarantee data on plot the total liabilities of Indian CPSEs at Indian CPSE loans are not available, but high risk of distress from 2000 to 2017. We loans to SPSEs amounting to 1.7 percent of use four alternative markers to identify CPSEs GDP were guaranteed by the Indian federal that are at high risk of distress in a given year: government in 2017. In Sri Lanka, the out- ICR less than 1 in the current year; a loss in standing stock of SOE debt with a govern- the current year; ICR less than 1 in the cur- ment guarantee amounted to approximately rent and previous two years; and a loss in 5 percent of GDP in 2019. The corresponding three out of the preceding five years. The last numbers for Bangladesh, as reported in two measures are more stringent. No matter SOUT H A SI A ’ S ST A TE - OWNED ENTER P RISES   111 FIGURE 3.11  Total Liabilities of Financially Distressed Central Public Sector Enterprises in India, 2000–17 6 5 4 Percent of GDP 3 2 1 0 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 Liabilities of loss-making CPSEs in the current year Liabilities of CPSEs with ICR < 1 in the current year Liabilities of CPSEs making a loss in 3 of 5 past years Liabilities of CPSEs with ICR < 1 in 3 consecutive years Source: Melecky, Sharma, and Yang 2020. Note: CPSEs = central public sector enterprises; ICR = interest coverage ratio. how the risk of distress is identified, the total Are SOEs Prone to Distress because They liabilities in the event of distress have been Operate in More Risky Markets? sizable in the recent years, although they have The excess distress rate of SOEs could reflect declined relative to the early 2000s. the higher risk that SOEs face and not neces- Although we cannot measure firm-level dis- sarily their inefficiency. Perhaps the develop- tress based on the ICR for other South Asian mental mandate of SOEs exposes them to countries, we can examine the total liabilities more types of risk or greater risk than other of loss-making SOEs using the limited SOE- observationally similar firms within the same level data available in Pakistan and Sri Lanka. industry. For example, SOEs could be leading This is depicted in figure 3.12. In Pakistan, the their industry in the exploration of new prod- total liabilities of loss-making SOEs have ucts, markets, and technologies, and thus may ranged between 12 percent and 18 percent of face abnormally high volatility in demand or GDP in recent years, a remarkably high production costs. Or SOEs engaged in the percentage. If we include only chronic loss-­ ­ procurement and distribution of food items makers—defined as SOEs that made a loss in and other essential commodities at controlled three out of the five past years—this number prices could be particularly vulnerable to remains between 8 percent and 12 percent of exchange rate and commodity price shocks. GDP. In Sri Lanka, the liabilities of loss-­ We examine this hypothesis by comparing making SOEs have hovered between 4 percent the volatility of firm-level sales across CPSEs and 5 percent of GDP. and non-SOEs in the Prowess panel. Following Comin and Philippon (2005), we measure the What Drives the Contingent volatility of a variable X (such as annual sales) Liabilities from SOEs? in firm i in a year t, Volatilityxit , as the stan- dard deviation of the annual growth rate of X This section analyzes the Prowess panel data over a 10-year period centered on the year t to better understand the drivers of SOE dis- (that is, between the years t − 4 and t + 5). tress and debt buildup. 112   H IDDEN DEBT FIGURE 3.12  Total Liabilities of Loss-Making State-Owned Enterprises in India, Pakistan, and Sri Lanka, 2005–17 20 18 16 15 14 12 Percent of GDP 12 12 8 5 5 5 4 4 4 5 4 4 3 3 3 2 2 2 2 1 0 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 India CPSEs Pakistan Sri Lanka Source: Melecky, Sharma, and Yang 2020 (see details in annex 3A). Note: CPSEs = central public sector enterprises. Indian CPSEs are not engaged in inher- Indian CPSEs overemploy labor and capi- ently more risky activities than private firms. tal. Controlling for size, age, and sector, the Overall, the results in table 3B.3 show that revenue-to-wage bill ratio for CPSEs is 85.8 CPSEs do not have significantly more volatile log points lower and their revenue-to-fixed- sales or profits than comparable non-SOEs.15 assets ratio is 21.5 log points lower (columns Adjusting for size is critical: even though the 2 and 4, respectively).16 Thus, CPSEs earn less raw volatility of CPSE sales is significantly per unit labor cost and per unit capital than lower (column 1), this difference disappears comparable private firms. CPSEs also have a when we include size as a control. higher debt-to-asset ratio than comparable Although individual SOEs do not face non-SOEs (column 6). more volatile conditions than individual non- We further compare CPSEs with other SOEs, the SOE sector as a whole could be firms in terms of revenue-based productivity more volatile because the shocks hitting SOEs measures: revenue total factor productivity are more correlated. But this hypothesis also (TFPR); and the marginal revenue products of is not confirmed: The volatility of aggregate capital (MRPK), labor (MRPL), and material CPSE sales has been similar to the aggregate inputs (MRPM). The estimates of TFPR, volatility of total sales of private firms in MRPK, MRPL, and MRPM are based on the the Prowess database in recent years. procedure outlined in Asker et al. (2014), which is based on a model in which firms produce differentiated products using a sim- Why Do SOEs Underperform ple (industry-specific), constant-return Cobb- Comparable Private Firms? Douglas production function and face a Using the same basic regression specification, demand curve that is constantly elastic. The we next show that SOEs commercially under- details of the estimation are presented in perform otherwise comparable private sector annex 3C. TFPR measures sales per unit firms. We regress indicators of performance inputs and should not be equated with physi- on the CPSE dummy and controls, such as cal total factor productivity (TFP). Differences size, age, and sector-year fixed effects. The in TFPR across firms could reflect distortions results are shown in table 3B.4. such as input adjustment costs, markups, and SOUT H A SI A ’ S ST A TE - OWNED ENTER P RISES   113 policy distortions. In a sense, TFPR is similar making optimal choices because too many to the revenue-to-input measures examined in external constraints and conflicting objectives table 3B.4. The difference is that it adjusts for are imposed on them, such as a government the usage of capital, labor, and raw material mandate leading to excessive hiring (Shleifer by employing industry-specific production and Vishny 1994). The results in tables 3B.4 function coefficients.17 and 3B.5 are more consistent with the latter Indian CPSEs are not concentrated in theory. In particular, they would seem to sectors in which profit margins are lower. ­ favor the hypothesis that SOEs are con- Table 3B.5 compares the revenue productivity strained from adjusting labor use. While across CPSEs and other firms.18 The regres- internal managerial problems can also lead to sions follow the same specifications as those inefficient input choices, they are unlikely to employed in tables 3B.3 and 3B.4. CPSEs cause a systematic and persistent overemploy- have a significantly higher TFPR than non- ment of an input. SOEs in the absence of other firm-level con- This interpretation is also consistent with trols (column 1). However, this gap becomes recent evidence from a natural experiment statistically insignificant upon introducing involving SOE privatization in India (Baird controls for size (column 2). Controlling for et al. 2019). This study finds that SOE priva- firm size makes a difference because larger tization improved the allocative efficiency of firms have significantly higher TFPR on aver- labor across firms by reducing the overem- age, perhaps reflecting higher price markups ployment of labor in the public sector and due to their greater market power. This sug- reallocating it to private firms with higher gests that CPSEs do not have lower TFPR marginal returns to labor. Finally, given that than comparably large private firms. we are measuring revenue productivity and Next, we consider the marginal revenue not physical productivity, the results could products of labor and capital. CPSEs have sig- also reflect mandates that prevent SOEs from nificantly lower MRPK and MRPL than non- adjusting prices in response to changes in SOEs (columns 4–8). The gap in MRPL is input costs (Khanna 2012). Some (implicit) particularly high (about 83 log points). These mandates could require the SOE to act as an results suggest that CPSEs use too much capi- insurance mechanism for volatile prices. tal and labor—especially the latter—that could be released and efficiently reallocated Do SOEs with Better Corporate to private firms with higher marginal returns Governance Perform Better? and capital and labor. Finally, we observe that CPSEs have a sig- Based on the idea that corporate governance nificantly higher marginal revenue product of reforms could improve SOE performance by intermediate inputs (columns 7 and 8). In a enabling better monitoring, greater opera- sense, CPSEs compensate for the overuse of tional autonomy, and reduced political inter- labor by “underusing” other inputs. For ference, “corporatization” and corporate example, SOEs could be using more manual governance reforms have taken root in South processes, which consume less power. Asian countries. For example, the Indian gov- It has long been argued that SOEs under- ernment has updated corporate governance perform due to internal management prob- guidelines for CPSEs and rates CPSEs on lems. It is harder to align the incentives of compliance with the guidelines (Government management and owners in the public sector of India 2010). The guidelines relate to the (see, for example, Ehrlich et al. 1994). The quality and independence of the board of compensation of managers tends to be less directors, audits, accounting standards, dis- strongly linked to the firm’s market perfor- closures, and risk management. mance in SOEs (Borisova, Salas, and To examine whether CPSEs with better Zagorchev 2019). A competing hypothesis is corporate governance perform better, we that SOEs managers are prevented from merged data on annual corporate 114   H IDDEN DEBT governance (CG) ratings of CPSEs into the still maintains a high degree of control over Prowess panel. The data are limited because corporate SOE boards (Fan, Morck, and not all CPSEs are rated, but there is some Yeung 2011). variation in the ratings across CPSEs and over time. 19 This variation allows us to Are Soft Loans and Soft Budgets the examine the association between the ratings Root of Contingent Liabilities? and performance within the subset of CPSEs that have been rated. Loans can be used to acquire productive Improvements in corporate governance assets, pay for working capital, or cover the must be complemented by broader reforms in occasional loss. The latter course of action is the governing environment around SOEs. not sustainable if financial markets are com- The results indicate that a higher corporate petitive and a firm continues to stack up governance rating is associated with better losses. To explore this mechanism, we exam- performance in the cross-section (table 3B.6). ine the correlation between the debt-to-asset Controlling for sector-year effects, a one- ratio and accumulated losses in the Prowess point improvement in the rating (which panel. ranges between 1 and 5) is associated with a Higher cumulative losses are associated 31.4-log point higher revenue-to-wage-bill with more indebtedness. Controlling for firm ratio (column 1). It is also associated with a fixed effects and sector-year fixed effects, the 5.6-percentage point reduction in the proba- estimated β in the specification in table 3B.7 bility of distress (ICR less than 1) (column 7) implies that a Rs 1 billion higher accumulated and a 0.07-point lower debt-to-asset ratio loss is associated with a 0.006 points higher (column 5). However, these correlations are debt-to-asset ratio (column 2). To put these not significant when controlling for firm fixed numbers in perspective, the standard devia- effects. Because it is possible that the corpo- tion of cumulative loss in 2016 is approxi- rate governance rating of CPSEs is correlated mately Rs 40 billion. The estimate thus with unobserved factors affecting firm perfor- implies that increasing cumulative loss by 1 mance, the sensitivity of the results to firm standard deviation increases the debt-to-asset fixed effects makes it hard to claim that this ratio by 0.24 points. The median debt-to- regression measures the impact of improved asset ratio was 0.36. Importantly, this associ- corporate governance. The weak correlation ation is significantly stronger among CPSEs. suggests that corporatization alone would not The estimated value of δ (column 4) implies solve the problems of SOE underperformance that the additional effect among CPSEs is and financial distress. 0.004 points. This suggests that CPSEs are The results call for more serious consider- more likely than other firms to use debt to ation of such reforms as part of package of cover losses. SOE reforms. The evidence base on corporati- We explore the soft loan hypothesis—and zation is limited and suggests that corporati- more broadly, the soft budget constraint zation needs to be complemented by even hypotheses—further by examining how the broader reforms of SOE governance and financial adjustment to distress and shocks related political economy issues to be effec- differ among CPSEs and non-SOEs. tive. Berkowitz, Ma, and Nishioka (2017) Soft loans and implicit guarantees distort argue that part of the positive impact of SOE the incentives of SOEs to monitor debt levels corporatization in China was due to external and act early to improve performance. The changes, such as reduced hiring pressures on results are presented in table 3B.8. Consider SOEs. Corporatization does not necessarily first what happens to the average firm on solve the problem of SOEs being captured by becoming financially distressed (denoted politicians or other insiders (Shleifer and “Enter distress” in the table) (columns 1, 3, Vishny 1994; Qian 1996). For example, there and 5). On average, firms that enter distress is evidence that the Chinese Communist Party have a significantly lower growth rate of SOUT H A SI A ’ S ST A TE - OWNED ENTER P RISES   115 paid-in capital (equity) and assets (columns 1 (columns 1 and 4). This result is reassuring and 5). This makes intuitive sense: firms in because in order to identify the differential distress are essentially drawing down their adjustment of CPSEs to shocks, we ideally assets. The relationship between distress and need shocks that are symmetric across CPSEs the growth rate of debt is positive, but the and other firms. However, controlling for level of statistical significance is marginal firm-level size, CPSEs have a higher probabil- (column 3). Further, this association is not ity of being in a negative shock than other robust to additional firm controls. firms (columns 2 and 5). More broadly, the The coefficient on the interaction of dis- statistically significant coefficients on firm- tress with the CPSE dummy is statistically sig- level variables like CPSE status, size, and age nificant and positive in the case of fixed indicate that the shocks as measured are not assets, implying that distress does not hamper entirely external to the firm (columns 3 CPSEs from acquiring fixed assets to the same and 6). extent as it does to non-SOEs. This result is The results on the differential adjustment consistent with the soft budget constraint to shocks are presented in table 3B.10. First, hypothesis. SOEs could also be required to considering the regressions without CPSE and expand they investment to stimulate the econ- shock term interactions (columns 1, 3, and 5), omy as the condition of recapitalization or the baseline pattern of adjustment to a shock other bailouts—as is the case for the state- makes intuitive sense. On average, firms fac- owned banks (see chapter 2). ing a negative (positive) shock have signifi- The interaction of distress with the CPSE cantly lower (higher) growth rates of debt, dummy is statistically insignificant in the case paid-in capital, and fixed assets. The most of paid-in capital or debt. This could be likely explanation is that the availability of because CPSEs adjust through other unob- funds to finance asset growth is sensitive to served channels, such as grants. Another pos- revenue shocks because banks and investors sibility is that our distress measure is too see revenue shocks as indicative of repayment permissive to be able to detect adjustment: as capacity. shown earlier, even 20 percent to 30 percent Next, we observe that the coefficients on of non-SOEs have an ICR of less than 1 in the interaction of the CPSE dummy with any given year. shocks have consistently a reverse sign to the To address these issues, we explore more corresponding baseline shock coefficients stringent measures of distress by using large (columns 2, 4, and 6) and, in most cases, are shocks to earnings. We first identify the 10th statistically significant. This suggests that and 90th percentile of revenue growth rates access to financing—whether through equity for every industry in the Prowess panel data. or debt—is significantly less sensitive to rev- A firm that experiences a revenue growth rate enue shocks for CPSEs than for the average above the 90th percentile of its historical firm and that CPSEs do not need to adjust to industry norm is classified as experiencing a shocks through assets to the same extent as positive “shock,” while a firm that experi- other firms. Another interpretation is that ences a revenue growth rate below the 10th private firms infer a negative shock as a sig- percentile of its industry norm is classified as nal of lower future returns and slow down experiencing a negative “shock.” Our their borrowing and investment, but SOEs assumption is that these unusual shocks to do not behave similarly—suggesting that the revenue reflect external market demand fac- latter’s decision making is less sensitive to tors.20 Results using similarly defined profit market signals. Overall, the results are con- shocks are similar. sistent with the soft budget constraint Table 3B.9 shows that the raw incidence of hypothesis, which states that the incentives negative shocks—that is, not controlling for of SOEs to monitor debt levels regularly and factors such as firm size—does not vary sig- act early to improve performance are nificantly across CPSEs and non-SOEs distorted. 116   H IDDEN DEBT The underperformance of and contigent In addition to the availability of panel data, liabilities stemming from SEOs may not justify the advantage of studying Indian CPSEs is blind and unconditional reduction in SOEs that they are relatively diverse and can be across the board. SOEs could still have impor- compared to private sector firms that resem- tant roles to play, including through their pos- ble them in attributes such as size, age, and sible spillovers on the industries in which their industry. If this were not the case, it would operate. We explore the existence of these posi- be difficult to distinguish between ownership tive spilovers in the South Asian context next. (that is, being state owned or private) and other attributes, such as size, age, and sector, when comparing the performance of firms. The SOE Sector Has a Role to Play The Indian CPSE sector accounts for in South Asia, Such as through Its only about 1 percent of the firms in the Long-Term Investment in R&D and Prowess database. Yet, in 2016, 4 of the top Positive Spillovers on Private Firms 20 Prowess firms in terms of total R&D This section explores a major rationale for expenditure were CPSEs. These CPSEs were supporting SOEs: that SOES can complement in aeronautics, heavy machinery, electronics, the private sector by undertaking risky invest- and oil and gas exploration. Figure 3.13 plots ments with long time horizons. Specifically, the industry-wise share of CPSEs in total this section examines the role of SOEs in industry fixed assets (x-axis) versus their undertaking R&D investments. It is well share in total industry R&D spending (y-axis) known that investments leading to innovation in 2016. In most of the industries in which are underfunded by private investors due to CPSEs have a significant presence (in terms of financial market failures (Hall and Lerner their share in total fixed assets), their share in 2010). SOEs can help plug this funding gap. R&D spending is disproportionately higher For example, in Europe, SOEs tend to invest than their share in fixed assets. Indeed, CPSEs more than private firms in R&D on sustain- account for more than 90 percent of total able technologies with low commercial industry R&D spending in industries such as returns (Bortolotti, Fotak, and Wolfe 2019). mining; shipbuilding; manufacturer of air- This analysis is based on Prowess, a panel craft, spacecraft, and related machinery; man- data set on Indian firms, including CPSEs. ufacture of structural metal products; electricity distribution; telecommunications; FIGURE 3.13  Central Public Sector Enterprise Share of Industry storage; and transport services. Gross Fixed Assets and Industry Research and Development Expenses in India, 2016 Table 3B.11 presents regressions compar- ing R&D spending in CPSEs and non-SOEs. 100 On average, CPSEs have significantly higher CPSE share of industry R&D expense (percent) R&D spending than other firms (column 1). 80 Note that this estimated R&D gap is robust to including sector-year fixed effects, implying that it is not driven by the concentration of 60 CPSEs in R&D–heavy industries. It is also robust to controlling for firm size and age 40 (columns 2 and 3), although it drops in mag- nitude upon doing so. In column 4, we inter- act the CPSE dummy with broad sector 20 dummies to observe that the positive R&D gap between CPSEs and other firms is largely driven by manufacturing. 0 20 40 60 80 100 The regressions presented in table 3B.12 CPSE share of industry gross fixed assets (percent) examine the relationship between R&D Source: Melecky, Sharma, and Yang 2020. spending in the public sector and productivity Note: CPSE = central public sector enterprise; R&D = research and development. SOUT H A SI A ’ S ST A TE - OWNED ENTER P RISES   117 in the private sector. This is to explore SOEs, let alone the contingent liabilities whether public sector R&D has any spillovers associated with them. on the private sector. Therefore, the fundamental policy message We estimate a positive relationship emerging from this chapter is that it is impor- between a firm’s own R&D stock and its rev- tant for governments to better assess and enue productivity (as measured by TFPR).21 monitor the fiscal risks from SOEs, incorpo- Further, an increase in the R&D stock in the rate them into their fiscal planning and debt CPSE’s own industry is also associated with management frameworks, and ensure that higher revenue per unit input: controlling for adequate provisions have been made for meet- firm fixed effects, the estimate of δ is positive ing triggered contingent liabilities without dis- and statistically significant at the 10 percent rupting public spending plans. For instance, level (column 2). This is not the case with the the government’s medium-term fiscal frame- private R&D stock. The estimate of δ is only work (MTFF) should incorporate these con- marginally sensitive to including controls for tingent liabilities by assessing SOE debt the share of the public sector in total industry trajectories and their sensitivity to shocks, revenue (column 3). keeping track of likely government commit- Given the potential endogeneity of R&D ments in case of distress (World Bank 2019b). spending, further research would be needed to better establish causation. But these pat- terns suggest that SOE R&D spending has Corporate governance guidelines positive spillovers on private firms in the same should be strengthened and enforced, industry. This is consistent with the idea that and more and better performance SOEs make long-term investments with posi- contracts should be adopted. tive externalities that would otherwise not be undertaken by the private sector. Any efforts to reduce the state’s direct presence in the It is also important to mitigate unnecessary economy by reducing SOE ownership could contingent liabilities from SOEs. The evidence thus start with a review to identify those presented in this chapter suggests that this will industries in which state presence could be entail combining internal, SOE-level reforms to beneficial in the long term, could be needed to improve their efficiency with external reforms create markets, or could expand reach in the to address the soft budget constraint on SOEs medium term and then exit, as opposed to and undue political intrusions into their opera- those industries in which state presence is tions. Internal reforms alone are unlikely to be hard to justify. enough because they seem to work only when SOEs operate in a truly competitive environ- Only a Combination of Internal ment (Bartel and Harrison 2005). Global les- and External Policy Reforms Can sons for the World Bank’s experience with SOE Help Better Manage Contingent reforms also suggest that efforts to improve Liabilities from SOEs in South Asia SOE financial performance entail working on several levers, many of which entail efforts to This chapter has shown that the contingent strengthen the broader governance environ- liabilities arising from SOEs in South Asia ment of SOEs (World Bank 2019b). can be large but difficult to precisely quan- tify due to their largely implicit and opaque Internal, SOE-Level Reforms: Improving nature and the lack of data. Governments in Corporate Governance and Performance South Asia do not track contingent liabilities Incentives from SOEs in a systematic manner. Hence, they are ill prepared if those liabilities are Corporate governance reforms that profession- triggered. In some cases, it is difficult to alize the boards of SOEs, increase their auton- quantify even the total liabilities and debt of omy, and improve financial reporting and 118   H IDDEN DEBT external audits can help improve SOE perfor- internal SOE reforms to address the issue of mance by reducing unnecessary costs arising contingent liabilities from SOEs may not be from poor employee or management effort, enough. misaligned incentives, and political interfer- ence. This recommendation is supported by External Reforms: Addressing the Soft the observed positive association between the Budget Constraint on SOEs quality of corporate governance and perfor- mance measures in Indian CPSEs (as shown in What markets believe about government this analysis) and in India SPSEs (Pargal and guarantees to SOEs matters. For instance, in Mayer 2014). Although South Asian countries 2015, the Baoding Tianwei Group became have pursued corporatization of SOEs in the first Chinese central government SOE to recent years, there is still room to strengthen default on its debt, shaking the market’s faith corporate governance guidelines and enforce in the implicit government guarantee behind their full implementation. For example, most SOEs. This reduction of implicit guarantees state-level power utilities in India comply with led to a decline in investment and net debt the basic corporate governance requirements issuance, an increase in cash holdings, and of the Companies Act, but not with the more reduced investment efficiency of SOEs in stringent guidelines that apply to CPSEs and China (Jin, Wang, and Zhang 2018). are recommended for state-owned enterprises Moreover, political influence on bank lend- (Pargal and Mayer 2014). The formal strin- ing to SOEs is a worrying issue in the region. gency of financial disclosure and audit require- The existence of a large state-owned commer- ments appears sound in Indian and Pakistan, cial bank (SOCB) sector in much of South but implementation is not well assessed Asia is notable in this regard because “soft” (OECD 2017; Naveed et al. 2018). loans of SOCBs to SOEs could be one channel Performance contracts could help address through which governments soften the SOE SOE underperformance by ameliorating budget constraints.22 However, even private agency problems. A performance contract sector banks are not immune to this problem. system typically defines SOE objectives and For example, in Pakistan, where banks are how they are to be assessed; monitors the largely privatized, preferential lending to achievement of objectives; and creates incen- politically connected firms has been estimated tives by linking management rewards to the to cause a significant level of GDP loss achievement of objectives. Elements of perfor- through misallocation of capital (Khwaja and mance contracting have already been adopted Mian 2008). Such banking issues are in South Asia. For example, India has intro- ­ discussed at length in chapter 2 of this report. duced a system under which CPSEs sign an Banking sector reform that increases competi- annual memorandum of understanding tion and makes banks less susceptible to polit- (MOU) with the responsible ministry and get ical influence could help reduce preferential rated on compliance with that MOU. India lending to SOEs and discipline their soft bud- has also introduced performance-based pay in get constraint. CPSEs. However, there are doubts about how However, reforms to improve the efficiency well this scheme is being implemented (Singh and competitiveness of the banking sector and Mishra 2013). and financial markets might not end soft Overall, we caution that the evidence on the loans to SOEs: banks will find SOE loans corporatization of SOEs is limited and has attractive as long as they believe them to be been mixed. Our findings on the positive asso- implicitly guaranteed by the government. ciation between CPSE corporate governance The most urgent and difficult policy issue, and performance are only suggestive. Similarly, therefore, is to address the soft budget con- evidence on the impacts of performance con- straint by making a credible commitment to tracts on SOEs is limited and does not point to not giving unconditional government support encouraging findings (see the survey in Smith to SOEs. The first step toward a credible and Trebilcock 2001). Relying solely on commitment is to make the objectives of each SOUT H A SI A ’ S ST A TE - OWNED ENTER P RISES   119 SOE clearer and more measurable. In addi- companies in return for a commitment to a tion to being central to any performance- charter of reform, which included better based management scheme in SOEs, this will metering and reduction in operational losses.23 allow governments to distinguish between Progress on achieving the charter is being losses incurred because of inefficiencies and monitored through a set of indicators. It is too losses incurred in efforts to meet socioeco- early to assess the success of the scheme, but nomic objectives. the commitment to reform that is required by Making the objectives of every SOE clearer UDAY is a good idea. It prevents the scheme and more measurable will also impose more from devolving into an unconditional bailout. discipline on the government itself, reducing It will be critical for the credibility of this the temptation to use SOEs as instruments of scheme to ensure that the charters are taken ever-shifting, short-term policy objectives seriously by the utilities and state govern- without due consideration to alternative ments. Of course, it would have been even instruments. The expectation of an implicit, better if the debt of the utilities had not been unconditional guarantee to SOEs could arise allowed to build up in the first place. because financial markets cannot always ascertain whether government support to an To be effective, reforms must signal SOE is justified by that SOE’s mandate, or is simply masking production inefficiency or a credible commitment to not ­giving mismanaged financial risks. Public sector unconditional support to SOEs. firms should be compensated for excess costs incurred in pursuit of explicit socioeconomic Governments must also define clear crite- mandates. They should not be compensated ria and methods for determining how to com- for costs arising from inefficiencies to sustain pensate SOEs for costs incurred to meet SOEs for political reasons—as is the case with development objectives. The design of the the frequent recapitalization of South Asian compensation scheme will depend on the public utility companies for losses due to objective of the SOE. For example, SOEs that unmetered electricity connections, uncollected provide subsidized goods or services should energy bills, electricity theft, fraud, and be compensated based on the gap between the payment evasion (Pargal and Mayer 2014; ­ price and the marginal cost of provision. The Zhang 2019). market price gap could be calculated by an To introduce greater financial discipline, independent body, such as a subgroup of the subsidies for providing public service at fiscal council. To the extent possible, grants to affordable (below-market) prices should be an SOE should be associated with the specific channeled through users, to the extent possi- subsidy programs that it is implementing and ble, rather than through the obscure financial kept separate from other balance sheet items. management systems of SOEs. Even better, this subsidy should be channeled Functions and business lines of SOEs through consumers/users of SOE services to should be well aligned with SOE objectives to install greater financial discipline. Similarly, control possible frivolous diversification and government loans to fund specific develop- the empire building tendencies of SOE man- mental investments should be earmarked agement or their higher-ups. as such. Governments can also better signal a credi- Governments should also adhere to their ble commitment to not giving unconditional own rules of SOE compensation and avoid support to SOEs by extending them regular postponing their obligations. For example, support based on previously specified criteria subsidies to utilities are not always paid and avoiding bailouts. In 2015, the govern- on time. During fiscal 2016, the difference ment of India announced a scheme (Ujwal between subsidies booked and subsidies DISCOM Assurance Yojana, UDAY) under received by state-owned power utilities in which state governments would take on the India was Rs 24 billion (Zhang 2019). debt of loss-making state-owned utility Had the government been paying this 120   H IDDEN DEBT compensation on time, there might not have consider setting up specialized and indepen- been a need for a scheme like UDAY. dent bodies to assess SOE performance at Governments should also avoid using circular the sector level or to ascertain whether the loans as substitutes for grants when that grant pu bl i c i nt erv enti on i s econ omi c al l y is justified on grounds of a needed social justified. subsidy. External accountability of government Governments must tie their hands against support to SOEs can be strengthened by mak- the possibility of providing unconditional ing data on SOE performance and financial support to SOEs by strengthening internal support more transparent and accessible . and external accountability. Internal checks Some South Asian countries—India, Pakistan, and balances should include strengthened and Sri Lanka—have been publishing SOE- oversight by the central government auditor level data in recent years. This is a welcome of government support to SOEs. Distributing development. However, as discussed in this the responsibility of regulation, oversight, chapter, the data are often patchy and not and policy making related to SOEs is also well documented, making it difficult to use important. Line ministries are often in them for analysis. Better data would not only charge of all three, potentially creating con- help assess and manage risks from SOEs, but flicting priorities. Given that line ministries also strengthen the credibility of government and finance ministries have limited capacity commitment to tightening the soft budget to monitor SOEs, governments could constraints for SOEs. Annex 3A. Sources of Data about South Asian SOEs Bangladesh Government of India, Ministry of Statistics and Programme Implementation. Various Government of Bangladesh, Finance Division, years. National Accounts Statistics. Ministry of Finance. Various years. Government of India. 2014. Annual Report: Bangladesh Economic Review. 2013–14. Government of India. 2018. Annual Report: Bhutan 2017–18. Government of Bhutan, Ministry of Finance. 2018. State Enterprises Annual Report. Pakistan Government of Pakistan, Ministry of Finance, India Implementation and Economic Reforms Unit. Various years. Federal Footprint: CMIE (Center for Monitoring the Indian SOE Annual Report. Economy). Various years. Prowess data set. Mumbai. Sri Lanka Government of India, Comptroller and Auditor General of India (CAGI). Various Advocata Institute (Sri Lanka). SOE database years. Report of the Comptroller and (https://www.research.advocata.org/soere​ Auditor General of India on Social, form/soe-data/). Economic, General, Revenue and General Government of Sri Lanka, Ministry of Sectors. Finance, Department of Public Enterprises. Government of India, Ministry of Heavy Various years. Performance Report. Industries and Public Enterprises, Government of Sri Lanka, Ministry of Department of Public Enterprises. Various Finance. 2019. Annual Report of the years. “Public Enterprises Survey.” Ministry of Finance. SOUT H A SI A ’ S ST A TE - OWNED ENTER P RISES   121 TABLE 3A.1  Definitions/Categorization of State-Owned Enterprises Used in This Report Country Definition/Categorization Bangladesh This report follows the classification of nonfinancial public enterprises used by the Finance Division of the Ministry of Finance and also as published in the Bangladesh Economic Review, which lists 45 nonfinancial public enterprises. India (CPSEs) Central public sector enterprises (CPSEs) are government companies in which more than 50 percent of the equity is held by the central government. (A “government company” is any company in which not less than 51 percent of the paid-up share capital is held by the central government; any state government or governments; or partly by the central government and partly by one or more state governments and includes a company that is a subsidiary company of such a government company.) Pakistan This report follows the categorization employed in the Government of Pakistan’s Federal Footprint: SOE Annual Report 2017, which lists 171 nonfinancial SOEs, including commercial and noncommercial public sector companies and federal authorities. Sri Lanka The Ministry of Finance broadly treats any state-controlled institution that is not a department or ministry as a state-owned enterprise (SOE). SOEs could be incorporated by an act of Parliament, or under the Companies Act. SOEs in Sri Lanka include statutory bodies, regulatory agencies, promotional institutions, educational institutions, public corporations, and limited companies. The report generally follows this broad definition of nonfinancial SOEs (which is also followed in the public SOE database published by the Advocata Institute). Annex 3B. Summary Statistics and Estimations for Indian Enterprises TABLE 3B.1  Summary Statistics of Prowess Data for Indian Central Public Sector Enterprises, 2016   N Mean SD p25 p50 p75 CPSEs 12,131 0.01 n.a. n.a. n.a. n.a. Total assets (Rs, million) 12,131 10,795.00 82,738.84 250.30 1,002.80 3,579.20 Sales (Rs, million) 12,131 4,161.56 9,552.77 209.60 929.50 3,103.80 Profit after tax (Rs, million) 12,131 57.19 518.73 –1.60 7.00 69.00 Cumulative loss (Rs, million) 12,131 –2,079.95 43,307.15 –384.00 –37.20 6.00 Interest expense (Rs, million) 12,131 175.17 470.05 3.80 21.50 95.70 R&D expense (Rs, million) 12,131 3.19 14.13 0.00 0.00 0.00 Debt (Rs, million) 12,131 2,208.86 5,984.42 55.90 276.70 1,199.40 Paid-in capital (Rs, million) 12,131 365.60 1,028.03 10.00 51.70 200.00 Gross fixed assets (Rs, million) 12,131 2,421.73 6,481.90 60.70 338.90 1,487.50 Debt-to-asset ratio 12,131 0.41 0.30 0.19 0.36 0.56 Interest coverage ratio (ICR) 12,131 27.71 3,017.92 0.71 1.59 4.00 Wages (Rs, million) 12,131 286.73 680.21 11.90 51.30 203.70 Size of firms (Rs, million) 12,131 6.35 1.86 5.13 6.39 7.55 Age of firms (years) 12,131 17.56 14.19 8.00 14.50 22.00 Corporate governance ratinga 97 4.00 1.75 4.00 5.00 5.00 Source: World Bank staff calculations based on Prowess data. Note: Currency unit: Indian rupee (Rs), million. In the last three columns, p25, p50, and p75 refer to 25th, 50th, and 75th percentiles, respectively. CPSEs = central public sector ­enterprises; R&D = research and development; SD = standard deviation. n.a. = not applicable. a. Corporate governance performance is rated on a five-point scale, from a low of 1 to a high of 5. 122   H IDDEN DEBT TABLE 3B.2  Probability of Indian Central Public Sector Enterprises Being Financially Distressed Distress: ICR < 1 for 2 consecutive Distress: ICR < 1 in a given year years (1) (2) (3) (4) (5) CPSEs 0.145*** 0.207*** 0.209*** 0.135*** 0.211*** (5.80) (8.80) (8.86) (3.73) (8.96) Size –0.0268*** –0.0265*** –0.0264*** –0.0167*** (–21.31) (–20.83) (–20.60) (–14.53) Age –0.000232 –0.000227 –0.000195 (–1.64) (–1.60) (–1.56) CPSE × Agriculture 0.0564 (0.80) CPSE × Mining –0.0380 (–0.37) CPSE × Petroleum –0.174* (–2.07) CPSE × Manufacturing 0.144**       (2.94)   Sector-year fixed effects Yes Yes Yes Yes Yes Observations 178,410 178,410 178,410 178,410 178,410 R-squared 0.0341 0.0442 0.0443 0.0448 0.0407 Source: Melecky, Sharma, and Yang 2020. Note: The table uses the following specification: Distressit = α + βCPSEi + γXi + θit + εit, where Distressit is an indicator for a firm i being in distress in year t. The specification controls for Xi, a set of time-invariant firm characteristics such as average size and age in the panel data set. It also controls for sectoral differences and sector-specific shocks though θit, a set of sector-year dummies. It is important to control for these factors because SOEs are larger than the typical non-SOEs and concentrate in certain sectors. For column 4, specification is modified to: Distressit = α + βCPSEi + γXi + δCPSEi ∗ Sectori + θit + εit. The standard errors are clustered by firm to account for serial-correlation in shocks. t statistics are in parentheses. . CPSEs = central public sector enterprises; ICR = interest coverage ratio. * p < 0.05, ** p < 0.01, *** p < 0.001. TABLE 3B.3  Volatility of Sales and Profit for Indian Central Public Sector Enterprises Sales volatility Profit volatility (1) (2) (3) (4) (5) (6) CPSEs –0.261** 0.0224 0.0538 –1.836*** –0.948* –0.844 (–2.71) (0.22) (0.54) (–4.10) (–2.08) (–1.84) Size –0.147*** –0.137*** –0.466*** –0.435*** (–10.58) (–9.88) (–7.20) (–6.70) Age –0.00681*** –0.0208** (–6.19) (–3.27) Sector-year fixed effects Yes Yes Yes Yes Yes Yes Observations 46,165 46,165 46,165 49,495 49,495 49,495 R-squared 0.0642 0.0777 0.0810 0.0170 0.0225 0.0236 Source: Melecky, Sharma, and Yang 2020. Note: The table uses a specification similar to that used to estimate distress incidence in table 3B.2. Volatility on the indicator for CPSE, controls such as size and age, and sector- year fixed effects are regressed: Volatilityxit = α + βCPSEi + γXi + θit + εit. Standard errors are clustered at the firm level. t statistics are in parentheses. CPSEs = central public sector ­enterprises. * p < 0.05, ** p < 0.01, *** p < 0.001. SOUT H A SI A ’ S ST A TE - OWNED ENTER P RISES   123 TABLE 3B.4  State Ownership and Financial Performance of Indian Central Public Sector Enterprises Log(revenue/wage) Log(revenue/fixed assets) Debt-to-asset ratio (1) (2) (3) (4) (5) (6) CPSEs 0.753*** –0.858*** –0.183* –0.215** 0.144*** 0.204*** (7.93) (–11.81) (–2.36) (–2.76) (4.51) (6.51) Size 0.700*** 0.0205*** –0.0187*** (158.22) (3.94) (–14.28) Age 0.000486 –0.00151** –0.00158*** (1.13) (–2.87) (–11.19) Sector-year fixed effects Yes Yes Yes Yes Yes Yes Observations 178,410 178,410 178,410 178,410 178,410 178,410 R-squared 0.132 0.602 0.182 0.183 0.0314 0.0475 Source: Melecky, Sharma, and Yang 2020. Note: Standard errors are clustered at the firm level. t statistics are in parentheses. CPSEs = central public sector enterprises. * p < 0.05, ** p < 0.01, *** p < 0.001. TABLE 3B.5  Productivity for Indian Central Public Sector Enterprises and Non-Central Public Sector Enterprises TFPR MRPK MRPL MRPM (1) (2) (3) (4) (5) (6) (7) (8) CPSEs 0.618*** –0.0156 –0.274** –0.313** –0.735*** –0.833*** 0.345*** 0.287*** (9.69) (–0.40) (–2.88) (–3.28) (–9.07) (–10.86) (4.69) (3.92) Size 0.254*** 0.0143* 0.107*** 0.00271 (96.30) (2.18) (23.35) (0.63) Age 0.00445*** 0.000620 –0.0141*** 0.00488*** (16.20) (1.03) (–30.79) (11.01) Sector-year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Observations 131,275 131,275 131,275 131,275 131,275 131,275 131,275 131,275 R-squared 0.160 0.464 0.258 0.259 0.0432 0.125 0.0770 0.0861 Source: Melecky, Sharma, and Yang 2020. Note: Standard errors are clustered at the firm level. t statistics are in parentheses. CPSEs = central public sector enterprises; MRPK = marginal revenue product of capital; MRPL = mar- ginal revenue product of labor; MRPM = marginal revenue product of material inputs; TFPR = revenue total factor p ­ roductivity. * p < 0.05, ** p < 0.01, *** p < 0.001. 124   H IDDEN DEBT TABLE 3B.6  Corporate Governance Ratings and Financial Performance of Indian Central Public Sector Enterprises Log(revenue/ Log(revenue/ wage) fixed assets) Debt-to-asset ratio ICR < 1 (1) (2) (3) (4) (5) (6) (7) (8) Corporate governance 0.314*** 0.00844 0.0843 0.0283 –0.0747*** 0.00403 –0.0565** 0.0163 rating (6.06) (0.26) (1.48) (1.31) (–3.76) (0.32) (–3.25) (0.91) Firm fixed effects No Yes No Yes No Yes No Yes Sector-year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Observations 398 398 398 398 398 398 398 398 R-squared 0.392 0.944 0.501 0.959 0.186 0.935 0.249 0.785 Source: Melecky, Sharma, and Yang 2020. Note: The table uses the following specification: Xit = α + βRatingit + γi + θit + εit. Here, Xit is a performance measure; Ratingit is the firm’s annual corporate governance rating; γi is a firm fixed effect; and θit are sector-year fixed effects. Standard errors are clustered at the firm level. t statistics are in parentheses. CPSEs = central public sector enterprises; ICR = i­nterest coverage ratio. * p < 0.05, ** p < 0.01, *** p < 0.001. TABLE 3B.7  Cumulative Losses and Debt-to-Asset Ratio of Indian Central Public Sector Enterprises Debt-to-asset ratio (1) (2) (3) (4) Cumulative loss 0.0146*** 0.00606*** 0.00746*** 0.0730*** (20.77) (11.59) (11.41) (11.16) CPSEs × Cumulative loss –0.00593*** 0.00425** (–7.00) (2.64) Firm fixed effects No Yes Yes Yes Sector-year fixed effects Yes Yes Yes Yes Size × Cumulative loss No No No Yes Observations 178,410 178,410 178,410 178,410 R-squared 0.0483 0.692 0.692 0.696 Source: Melecky, Sharma, and Yang 2020. Note: The table regresses the debt-to-asset ratio of firm i in year t on its cumulative loss, CLossit, firm fixed effects, γi , and sector-year fixed effects, θit, as follows: DebtAssetit α = βCLossit + δCPS Ei * CLossit + θXi * CLossit + γi + θit + εit. The coefficient on the interaction between the CPSE dummy and cumulative loss, δ, measures how the relationship between cumulative loss and the debt-to asset-ratio differs for CPSEs compared with the average firm. Knowing that CPSEs are larger than the average firm, the specification also controls for an interaction of size and other attributes with the cumulative loss variable. Cumulative losses are expressed in US$, billion. The debt-to-asset ratio is expressed as a percentage. Standard errors are clustered at the firm level. t statistics are in parentheses. CPSEs = central public sector enterprises. * p < 0.05, ** p < 0.01, *** p < 0.001. SOUT H A SI A ’ S ST A TE - OWNED ENTER P RISES   125 TABLE 3B.8  Distress and the Growth Rate of Paid-in Capital, Debt, and Fixed Assets of Indian Central Public Sector Enterprises Growth rate of paid-up Growth rate of gross fixed capital Growth rate of debt assets (1) (2) (3) (4) (5) (6) Enter distress –0.0190*** –0.0200* 0.0126* –0.0156 –0.00886** –0.0333** (–8.76) (–2.45) (2.48) (–0.86) (–3.05) (–3.15) CPSEs × Enter distress 0.0333 0.0232 0.0511* (1.66) (0.44) (2.33) Enter distress × Size 0.0000750 0.00455 0.00387* (0.06) (1.64) (2.22) Firm fixed effects Yes Yes Yes Yes Yes Yes Sector-year fixed effects Yes Yes Yes Yes Yes Yes Size × Enter distress No Yes No Yes No Yes Observations 178,410 178,410 178,410 178,410 178,410 178,410 R-squared 0.259 0.259 0.194 0.194 0.297 0.298 Source: Melecky, Sharma, and Yang 2020. Note: The basic regression is as follows: Yit = α + βShockit + δCPSEi * Shockit + θXi * Shockit + γi +θit + εit. The regressions include firm fixed effects, γi, and sector-year fixed effects, θit, as controls. γit is a measure of financial adjustment (the annual growth rate of debt, paid-up capital, or assets). Shockit is a measure of a firm-level shock. The regression is mainly interested in the interaction term coefficient, δ, which measures the differential adjustment of CPSEs to the shock. The regression first examines what happens when a firm enters distress. In this case, the “shock” measure is a dummy that is equal to 1 when a firm enters distress (switches from ICR of more than 1 in the previous year to ICR of less than 1 in the current year). These estimates should not be interpreted as measuring the causal impact of distress because distress is potentially endogenous to financial decisions. For example, depending on what the borrowed funds are used for, new debt could increase interest payments more ­rapidly than earnings. “Enter distress” refers to the point at which a firm becomes financially distressed based on ICR. Standard errors are clustered at the firm level. t statistics are in ­parentheses. CPSEs = central public sector enterprises. * p < 0.05, ** p < 0.01, *** p < 0.001. TABLE 3B.9  Probability of Negative Shocks to Sales and Profit for Indian Central Public Sector Enterprises Negative sales shock Negative profit shock (1) (2) (3) (4) (5) (6) CPSEs 0.00218 0.0380*** 0.0403*** –0.00191 0.00851* 0.00921* (0.32) (5.89) (6.28) (–0.47) (2.13) (2.30) Size –0.0156*** –0.0153*** –0.00454*** –0.00444*** (–35.34) (–34.47) (–14.55) (–14.10) Age –0.000287*** –0.0000901** (–6.58) (–2.70) Sector-year fixed effects Yes Yes Yes Yes Yes Yes Observations 178,410 178,410 178,410 178,410 178,410 178,410 R-squared 0.0111 0.0223 0.0227 0.00664 0.00787 0.00792 Source: Melecky, Sharma, and Yang 2020. Note: Standard errors are clustered at the firm level. t statistics are in parentheses. CPSEs = central public sector enterprises. * p < 0.05, ** p < 0.01, *** p < 0.001. 126   H IDDEN DEBT TABLE 3B.10  Sales Shock and the Growth Rate of Paid-up Capital, Debt, and Fixed Assets for Indian Central Public Sector Enterprises Growth rate of paid-up Growth rate of gross fixed capital Growth rate of debt assets (1) (2) (3) (4) (5) (6) Negative sales shock –0.0165*** –0.0209* –0.0538*** –0.0191 –0.0561*** –0.0324** (–6.49) (–2.25) (–8.38) (–0.87) (–18.32) (–3.00) Positive sales shock 0.0576*** 0.0309* 0.121*** 0.0586** 0.0898*** –0.00558 (17.90) (2.53) (21.58) (2.90) (24.38) (–0.41) CPSEs × Negative sales shock 0.0437* 0.118* 0.0578** (2.52) (2.47) (3.02) CPSEs × Positive sales shock –0.0349 –0.130** –0.110*** (–1.76) (–3.18) (–4.67) Negative sales shock × Size 0.000515 –0.00667 –0.00464* (0.33) (–1.80) (–2.52) Positive sales shock × Size 0.00442* 0.0105*** 0.0158*** (2.32) (3.34) (7.08) Firm fixed effects Yes Yes Yes Yes Yes Yes Sector-year fixed effects Yes Yes Yes Yes Yes Yes Size × Negative shock No Yes No Yes No Yes Size × Positive shock No Yes No Yes No Yes Observations 178,410 178,410 178,410 178,410 178,410 178,410 R-squared 0.263 0.263 0.199 0.199 0.307 0.308 Source: Melecky, Sharma, and Yang 2020. Note: Standard errors are clustered at the firm level. t statistics are in parentheses. CPSEs = central public sector enterprises. * p < 0.05, ** p < 0.01, *** p < 0.001. TABLE 3B.11  Research and Development Expenditure: Comparing Indian Central Public Sector Enterprises to Other Firms R&D expenditure (1) (2) (3) (4) CPSEs 7.491*** 2.896* 2.428 –0.804 (4.92) (2.16) (1.81) (–0.54) Size 2.003*** 1.935*** 1.939*** (20.51) (20.57) (20.62) Age 0.0598*** 0.0600*** (7.81) (7.83) CPSEs × Agriculture 1.554 (0.24) CPSEs × Mining –3.044 (–1.23) CPSEs × Petroleum 37.78 (1.84) CPSEs × Manufacturing 6.131*       (2.28) Sector-year fixed effects Yes Yes Yes Yes Observations 178,410 178,410 178,410 178,410 R-squared 0.0695 0.105 0.108 0.109 Source: Melecky, Sharma, and Yang 2020. Note: The table uses the following specification: R&Dit = α + βCPSEi + γXi + θit + εit. R&Dit measures R&D expenditure in a firm i in year t. The regression controls for Xi, a set of time-invariant firm characteristics such as average size and age in the panel data set. Firm size is measured by its total assets (in logs) averaged over the entire panel. The regression also controls for sectoral differences and sector-specific shocks through θit, a set of sector-year dummies. This is because SOEs are larger than the typical non-SOE and concentrate in certain sectors. The standard errors are clustered by firm to account for serial-correlation in shocks. t statistics are in parentheses. CPSEs = central public sector enterprises; R&D = research and development. * p < 0.05, ** p < 0.01, *** p < 0.001. SOUT H A SI A ’ S ST A TE - OWNED ENTER P RISES   127 TABLE 3B.12  Relationship between Public Sector Research and Development and Private Sector Performance in India Revenue total factor productivity (TFPR) (1) (2) (3) Log of R&D stock 0.102*** 0.0308*** 0.0289*** (18.15) (9.66) (9.42) Log of private R&D stock –0.0277* 0.00327 0.00778 (–2.51) (0.65) (1.54) Log of public R&D stock 0.000722 0.00926* 0.00806 (0.08) (1.99) (1.56) Private market share –1.358*** (–8.38) Public market share –1.426*** (–8.36) Firm fixed effects No Yes Yes Year fixed effects Yes Yes Yes Standard errors clustered at industry level Yes Yes Yes Observations 128,242 128,242 128,242 R-squared 0.118 0.832 0.835 Source: Melecky, Sharma, and Yang 2020. Note: The table uses the following basic specification: Xit = α + βR&D Stock it + γPublic R&D Stock it + γPrivate R&D Stock it + γi + θ it + εit. Xit is a revenue-based measure of the performance of firm i in year t. R&D Stockit is a measure of the firm’s own R&D stock. The R&D stock is measured as the sum of current and past R&D expenditures, assuming an annual depreciation of 20 percent. For details of measuring the stock of R&D and other intangible investment, please see Dutz et al. (2012). The variable Public R&D Stockit measures the total R&D stock of all CPSEs in the same 3-digit industry as firm i. The variable Public R&D Stockit measures the total R&D stock of all non-SOEs in the same 3-digit industry as firm i (excluding the firm itself). The variable γi is a firm fixed effect, and θit are year dummies. Because the public and private R&D stock variables are time-varying industry-level variables, they are collinear with industry-year dummies, and their effects cannot be identified separately from those of other industry-specific shocks. The regression is mainly interested in δ, which captures the relationship between the aggregate stock of R&D in public sector firms in firm i’s industry and firm i’s performance. Because the analysis is interested in spillovers to the private sector, the regression is estimated on the subsample of non–state-owned enterprises (SOEs) in Prowess. In a robustness check, controls were added for the share of the public sector and all private sector firms other than firm i in total industry revenue. This is because the public R&D stock variables could be correlated with the size of the public sector, with the latter potentially affecting the revenue of firm i through its effect on input and output prices. t statistics are in parentheses. R&D = research and development. * p < 0.05, ** p < 0.01, *** p < 0.001. Annex 3C. Productivity Estimation We follow the model and procedure employed Qit = Bit Pit−∈. (3C.2) in Asker et al. (2014). A firm i, in time t, pro- duces output Q it using the following Combining equations (3C.1) and (3C.2), technology: Asker et al. (2014) obtain the expression for the sales-generating production function: αK αL αM Qit = Ait K it Lit M it , (3C.1) βK βL βM Sit = Ωit K it Lit M it , (3C.3) where Kit is the capital input, Lit is the labor input, and Mit is raw materials. Ait is the where Ωit is a function of TFP and the para­ firm’s total factor productivity (TFP). meters of the demand curve, and The production function is assumed to  1 have constant returns to scale (CRS), imply- β X = α X  1 −  for input X ∈ {K, L, M}.   ing that αk + αL + αM = 1. The firm faces a demand curve with a con- Revenue total factor productivity (TFPR) stant elasticity: is defined as the log of Ωit. It can be inferred 128   H IDDEN DEBT from sales and input values using the sales Notes generating function in equation (3C.3) and 1. Unfortunately, data on another measure that taking logs: would be more appropriate than revenues for comparisons of the share of GDP—value TFPRit = log(S)it – βK log(K)it – added by the SOE sector—are not available βL log(L)it – βM log(M)it (3C.4) for most South Asian countries. 2. See the Solar Energy Corporation of India Ltd site: http://www.seci.co.in/. Similarly, the marginal revenue product of 3. The idea of coordination failures goes back an input X can be inferred as to Rosenstein-Rodan’s (1943) theory of a “big push” in industrialization. The technol- ogy and skills example is from Rodríguez- MRPXit = log(βX) + log(Sit) – log(Xit) Clare (2005). (3C.5) 4. See the Rail India Technical and Economic Service Ltd site: https://www.rites.com/index. To estimate the revenue production func- php?page=page&id=8&name=Profile& tion input coefficients (the βs), Asker et al. mid=8. (2014) use the first-order condition for the 5. See https://www.research.advocata.org/sri​ optimal use of a flexible input in a static -lankas-soes-burn-peoples-cash-burden​-­budge ts-undermine-national-savings/, which is based model with no frictions or distortions. The on data from the government of Sri Lanka, condition is that the revenue share of the Public Enterprises Department. input’s expenditure should equal its revenue 6. The notion of an implicit government guar- function coefficient: antee on SOE debt is not limited to South Asia (see, for example, European Commission 2016). Pit X it βX = . (3C.6) 7. For a review of the global evidence, see S it Shirley and Walsh (2001). For evidence on India, see Majumdar (1996). For evidence on Following the procedure in Asker et al. China, see Li, Lin, and Selover (2014) and (2014), we use the median expenditure Harrison et al. (2019). shares of labor and raw materials in every 8. Similarly, Naveed et al. (2018) suggest that at 3-digit industry to estimate βL and βM. The least in formal terms, SOEs in Pakistan fol- medians are calculated using Prowess data low clear corporate governance norms related to operational autonomy and finan- pooled over 2000 and 2016. We limit the cial reporting. sample to the same time period in our reve- 9. It is worth noting that the agency and envi- nue productivity regressions to avoid any ronmental hypotheses are not mutually potential biases arising from potential long- exclusive, nor are they necessarily indepen- term changes in these input coefficients due dent of each other. to demand or technological trends. We esti- 10. See, for example, S&P Global Ratings, 2017, mate βk as “SOE Shake-Up: China’s Support for Its Ailing Enterprises Will Become More  −1 Selective.” βK = − βL − β M , (3C.7) 11. Our results stay the same if for a given regres-  sion we include observations that might be missing other outcome variables but not where ∈ is set equal to 4 following Bloom those relevant to the regression in question. (2009). To compute TFPR and the marginal Further, when using Prowess to compute an revenue products of labor and capital, we aggregate statistic for the CPSE sector, we plug in the estimated input coefficients into include all observations (except those miss- equations (3C.4) and (3C.5). ing the statistic in question). SOUT H A SI A ’ S ST A TE - OWNED ENTER P RISES   129 12. Although Sri Lanka has 400 SOEs (including by using more exogenous drivers, such as subsidiaries), detailed financial data are export market shocks or input cost shocks. available for only 42 nonfinancial SOEs, 21. TFPR essentially measures sales per unit inputs. which have been identified by the govern- Further details about its estimation and inter- ment as strategically important state-owned pretation are discussed in this chapter. business enterprises (SOBEs), based on their 22. The lack of access to sufficiently disaggre- importance to the national economy and gated lending data, however, makes it diffi- size. cult to test whether SOCBs make softer loans 13. The numbers in this paragraph capture SOE to SOEs than private banks do in South Asia. revenues, not value added, as a share of GDP, 23. See UDAY’s official website at https://www​ and as such overstate the contribution of .uday.gov.in/home.php. SOEs to GDP. Official public sector reports of South Asian countries often tend to report SOE revenue but not value added. As a result, References we could not get comparable and up-to-date Allen, F., X. Gu, J. Qian, and Y. Qian. 2017. value added figures for SOEs across the set of “Implicit Guarantee and Shadow Banking: The five South Asian countries included in Case of Trust Products.” Working paper. figure 3.1. Asker, J., A. Collard-Wexler, and J. De Loecker. 14. Even when available, the data do not distin- 2014. “Dynamic Inputs and Resource (Mis) guish between capital investment, grants for Allocation.” Journal of Political Economy 122 developmental purposes, and grants or loans (5, October): 1013–63. to cover losses. Baird, M., A. V. Chari, S. Nataraj, A. Rothenberg, 15. The number of observations is lower than in S. Telhaj, and L. A. Winters. 2019. “The Public table 3B.2 because we cannot calculate Sector and the Misallocation of Labor: Evidence 10-year centered volatility for the last and from a Policy Experiment in India.” CEP first 5 years of the panel, and because only a Discussion Paper 1596, Centre for Economic subset of firms is observed continuously for Performance, London School of Economics and 10 years. Political Science. 16. Prowess does not report employment num- Bartel, A., and A. E. Harrison. 2005. “Ownership bers for most firms. We follow prior studies versus Environment: Disentangling the Sources using Prowess (such as Asker, Collard- of Public-Sector Inefficiency.” Review of Wexler, and De Loecker 2014) that have used Economics and Statistics 87 (1): 135–47. the wage bill instead of employment. BBNL (Bharat Broadband Network Limited). 17. See Hseih and Klenow (2009) for a discus- 2019. Annual Report, 2018–19. http://www​ sion on the interpretation of revenue-based . b b n l . n i c . i n / / a d m n i s / a d m i n / s h w i m g​ productivity measures. .aspx?ID=1342. 18. As explained in more detail in annex 3B, the Berkowitz, D., H. Ma, and S. Nishioka. 2017. TFPR calculation involves the estimation of “Recasting the Iron Rice Bowl: The Reform of 3-digit industry-specific input coefficients. To China’s State-Owned Enterprises.” Review of lessen the noise in these estimates, we only Economics and Statistics 99 (4): 735–47. include 3-digit industries that have at least Bloom, N. 2009. “The Impact of Uncertainty 10  firms in the panel in 2016. This reduces Shocks.” Econometrica 77 (3): 623–85. the sample size compared to the other Borisova, G., J. M. Salas, and A. Zagorchev. 2019. regressions. “CEO Compensation and Government 19. Eighty CPSEs were rated in 2010. This num- Ownership.” Corporate Governance: An ber increased to 161 by 2016. The median International Review 27 (2): 120–43. rating rose from 1.6 to 1.2 (on a five-point Bortolotti, B., V. Fotak, and B. Wolfe. 2019. scale where 5 is best and 1 is worst) between “Innovation at State-Owned Enterprises.” 2010 and 2017, while the standard deviation Working paper, Bocconi University, Milan. increased from 1.0 to 1.3. Comin, D., and T. Philippon. 2005. “The Rise in 20. We cannot claim that these shocks are truly Firm-Level Volatility: Causes and Consequences.” external to the firm because we do not know In NBER Macroeconomics Annual 2005, vol. their underlying causes. In future work, we 20, edited by Mark Gertler and Kenneth Rogoff, plan to instrument for these revenue shocks 229–312. Cambridge, MA: MIT Press. 130   H IDDEN DEBT Dal Bo, E. 2006. “Regulatory Capture: A Review.” Jensen, M. C. 1986. “Agency Costs of Free Oxford Review of Economic Policy 22 (2): C a s h F l o w, C o r p o r a t e F i n a n c e , a n d 203–25. Takeovers.” American Economic Review Dutz, M. A., S. Kannebley, M. Scarpelli, and 76: 323–29. S. Sharma. 2012. “Measuring Intangible Assets Jin, S., W. Wang, and Z. Zhang. 2018. “The Value in an Emerging Market Economy: An and Real Effects of Implicit Guarantees.” Application to Brazil.” Policy Research Working paper, Hong Kong University of Working Paper 6142, World Bank, Washington, Science and Technology. DC. Khanna, S. 2012. “State-Owned Enterprises in Ehrlich, I., G. Gallais-Hamonno, Z. Liu, and India: Restructuring and Growth.” Copenhagen R. Lutter. 1994. “Productivity Growth and Journal of Asian Studies 30 (2): 5–28. Firm Ownership: An Analytical and Empirical Khwaja, A. I., and A. Mian. 2008. “Tracing the Investigation.” Journal of Political Economy Impact of Bank Liquidity Shocks: Evidence 102 (October): 1006–38. from an Emerging Market.” American European Commission. 2016. “State-Owned Economic Review 98 (4): 1413–42. Enterprises in the EU: Lessons Learnt and Ways Kornai, J. 1986. “The Soft Budget Constraint.” Forwards in a Post Crisis Context.” Kyklos 39: 3–30. Institutional Paper 31, European Commission, Li, S., Y. C. Lin, and D. Selover. 2014. “Chinese Brussels. State-Owned Enterprises: Are They Inefficient?” Fan, J., R. Morck, and B. Yeung. 2011. Chinese Economy 47 (5–6): 81–115. “Capitalizing China.” NBER Working Paper Majumdar, S. K. 1996. “Assessing Comparative 17687, National Bureau of Economic Research, Efficiency of the State-Owned, Mixed, and Cambridge, MA. Private Sectors in Indian Industry.” Public Government of India, Ministry of Heavy Industries Choice 96: 1–24. and Public Enterprises. 2010. Corporate Maskin, E., and C. Xu. 2001. “Soft Budget Governance Guidelines for CPSEs. https://dpe​ Constraint Theories: From Centralization to .gov.in/sites/default/files/gcgcpse10.pdf. the Market.” Economics of Transition Government of India, Ministry of Heavy Industries 9: 1–27. and Public Enterprises. 2014. Annual Report: Melecky, M., S. Sharma, and D. Yang. 2020. 2013–14. Government of India. “State-Owned Enterprises in South Asia: The Government of India, Ministry of Heavy Industries Distresses, Adjustments, and Fiscal and Public Enterprises. 2018. Annual Report: Contingent Liabilities.” Background paper 2017–18. Government of India. for Hidden Debt. World Bank, Washington, Government of Pakistan. 2016. Federal Footprint: DC. SOE Annual Report 2015–16. Government of Naveed, S., Y. Salman, N. Jabeen, M. Zafar, I. Pakistan. Jadoon, and S. Irfan. 2018. “Governance and Hall, B. H., and J. Lerner. 2010. “The Financing of Management of SOEs in Pakistan.” Pakistan R&D and Innovation. In Handbook of the Economic and Social Review 56 (1, Summer): Economics of Innovation, vol. 1, edited by 47–66. Bronwyn H. Hall and Nathan Rosenberg, OECD (Organisation for Economic Co-operation 609–39. North-Holland, Amsterdam. and Development). 2017. Disclosure and Harrison, A., M. Meyer, P. Wang, L. Zhao, and M. Transparency in the State-Owned Enterprise Zhao. 2019. “Can a Tiger Change Its Stripes? Sector in Asia: Stocktaking of National Reform of Chinese State-Owned Enterprises in Practices. Paris: OECD. the Penumbra of the State.” NBER Working Pargal, S., and K. Mayer. 2014. Governance of Paper 25475, National Bureau of Economic Indian State Power Utilities: An Ongoing Research, Cambridge, MA. Journey. Directions in Development–Energy Hsieh, C-T, and P. J. Klenow. 2009. “Misallocation and Mining. Washington, DC: World Bank. and Manufacturing TFP in China and India.” Peltzman, S. 1976. “Toward a More General Quarterly Journal of Economics 124 Theory of Regulation.” Journal of Law and (4, November): 1403–48. Economics 19 (2): 245–48. SOUT H A SI A ’ S ST A TE - OWNED ENTER P RISES   131 Qian, Y. 1996. “Enterprise Reform in China: Singh, P., and R. K. Mishra. 2013. “Performance- Agency Problems and Political Control.” Related Pay in Central Public Sector Enterprises Economics of Transition 4 (2): 427–47. in India.” Indian Journal of Industrial Relations Richmond, C. J., D. Benedek, E. Cabezon, B. Cegar, 49 (2, October): 314–27. P. Dohlman, M. Hassine, B. Jajko, P. Kopyrski, Smith, A., and M. Trebilcock. 2001. “State-Owned M. Markevych, J. A. Miniane, F. J. Parodi, G. Enterprises in Less Developed Countries: Pula, J. Roaf, M. K. Song, M. Sviderskaya, R. Privatization and Alternative Reform Turk, and S. Weber. 2019. Reassessing the Role Strategies.” European Journal of Law and of State-Owned Enterprises in Central, Eastern, Economics 12: 217–52. and Southeastern Europe. Washington, DC: Stigler, G. J. 1971. “The Theory of Economic International Monetary Fund. Regulation.” Bell Journal of Economics and Rodríguez-Clare, A. 2005. “Coordination Failures, Management Science 2 (1): 3–21. Clusters, and Microeconomic Interventions,” World Bank. 2019a. “Non-financial State-Owned with comments by F. Rodríguez, R. Hausmann, Enterprises in Bangladesh: Improving and J. M. Benaventa. Economía 6 (1): 1–42. Corporate Governance and Financial Rosenstein-Rodan, P. N. 1943. “Problems of Performance.” World Bank, Washington, DC. Industrialisation of Eastern and South-Eastern Unpublished. Europe.” Economic Journal 53 (210/211): World Bank. 2019b. “Integrated State-Owned 202–11. Enterprises Framework (iSOEF) Guidance Shirley, M. M., and P. Walsh. 2001. “Public vs. Note Module 2: Assessing the Fiscal Private Ownership: The Current State of the Implications of SOE Reform.” World Bank, Debate.” Policy Research Working Paper 8288, Washington, DC. World Bank, Washington, DC. Zhang, F. 2019. In the Dark: How Much Do Shleifer, A., and R. W. Vishny. 1994. “Politicians Power Sector Distortions Cost India? South and Firms.” Quarterly Journal of Economics Asia Development Forum. Washington, DC: 109 (4, November): 995–1025. World Bank Group. Subnational Governments in South Asia: Balancing the 4 Fiscal Risks of Government Decentralization with the Returns F iscal decentralization is on the rise The chapter then quantifies contingent around the world, including in South l ­iabilities for India—the one country in the Asia. This chapter investigates subna- region with sufficient data—and examines tional fiscal risks and contingent liabilities in their impact on fiscal dynamics and the local South Asia, a region where many countries economy. India’s experience is illustrative for are already at high risk of debt distress.1 the rest of the region, especially for countries The chapter first examines the institu- such as Pakistan, where provincial borrowing tional frameworks across South Asia and has been expanding, and for Maldives and relates them to the likelihood of exposure Nepal, which have started to decentralize fis- to subnational fiscal risks. An important cal policy. finding is that central governments in most The chapter concludes by synthesizing the South Asian countries delegate extremely empirical findings, discussing the main drivers limited authority to subnational govern- of subnational fiscal risks in South Asia, and ments to borrow—thus minimizing subna- presenting recommendations to manage tional fiscal risks but also potentially them. foregoing the benefits of decentralized deci- sion making. The chapter next studies Pakistan, one of The Promise and Risks of Fiscal the two countries in South Asia whose insti- Decentralization in South Asia tutional framework allows the central gov- Fiscal decentralization holds great promise ernment to be exposed to significant for countries, but also carries many risks. subnational fiscal risk. The discussion high- On the one hand, because subnational lights how a lack of transparency in subna- ­ governments (SNGs) are closer to residents tional public debt statistics and guarantees and firms, they may have better information reduces policy makers’ accountability about spending needs and citizen demands and exposes the country to substantial (Oates 1972, 1999); stronger incentives due fiscal risks. to competition among jurisdictions (Keen Note: This chapter draws on the background research paper: Blum, F., and P. S. Yoong. 2020. “The Impact of Subnational Contingent Liability Realizations: Evidence from India.” Background paper for Hidden Debt. World Bank, Washington, DC. 133 134   H IDDEN DEBT and Marchand 1997); and greater exposure spending can crowd in more private invest- to higher accountability to make public ment and spur overall local economic activ- spending more efficient (Seabright 1996; ity and growth. For instance, countries with Persson, Roland, and Tabellini 2000). higher shares of decentralized expenditures In turn, the increased efficiency of public tend to experience higher rates of private FIGURE 4.1  The Relationship between Private Investment and Fiscal Decentralization a. Countries with a higher share of decentralized expenditures tend to receive more private investment 40 Gross fixed capital formation as percent of GDP 30 20 10 0 0 20 40 60 80 100 Percent of subnational expenditure in general government expenditure b. Indian states that have more revenue autonomy tend to generate more xed investment GJ 8 Gross fixed capital formation as percent of GDP 6 OR AS 4 AP MP TN GA BR HR KA MH 2 KL RJ UP SK PB WB HP ML JK 0 TR MN AR NL 60 70 80 90 100 Percent of untied resources in overall state receipts Sources: IMF 2019a, 2019b, 2019c, 2019d; OECD-UCLG 2019; RBI 2019b; World Bank staff calculations. Note: Panel a refers to 2017, but similar patterns hold for 2001–16. Untied resources refer to the sum of own-source revenues, tax devolution, and revenue deficit grants. AP = Andhra Pradesh; AR = Arunachal Pradesh; AS = Assam; BR = Bihar; GA = Goa; GJ = Gujarat; HP = Himachal Pradesh; HR = Haryana; JK = Jammu and Kashmir; KA = Karnataka; KL= Kerala; MH = Maharashtra; ML = Meghalaya; MN = Manipur; MP = Madhya Pradesh; NL = Nagaland; OR = Odisha; PB = Punjab; RJ = Rajasthan; SK = Sikkim; TN = Tamil Nadu; TR = Tripura; UP = Uttar Pradesh; WB = West Bengal. S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   135 investment (­ figure 4.1, panel a). In India, have no choice but to rescue the affected states with a greater share of revenue localities to reduce spillovers and limit the from decentralization—measured by the damage to their own sovereign creditwor- shares of untied resources in total state rev- thiness (Jenker and Liu 2014)—the so- enue—tend to g ­ enerate greater total invest- called soft budget constraint (Kornai ment (figure 4.1, panel b). 1986). These repeated bailouts can lead to On the other hand, fiscal decentraliza- persistent subnational fiscal problems, as tion can also carry risks. Theoretically, fis- illustrated by the experiences of Argentina cal risks can materialize independently of and Brazil. 2 In addition, SNGs that rely whether they were incurred by the central heavily on transfers from the central gov- government or SNGs. For instance, the pri- ernment may also be less motivated to mary sources of contingent liabilities in Sri raise revenues from their own sources Lanka, a highly centralized country, are (taxes and fees) because they do not fully explicit and implicit guarantees of the cen- internalize the social cost of local public tral government on the debt of state-owned expenditures—the so-called common pool enterprises (SOEs). These are estimated at problem (Oates 2005; Governatori and 7.1 percent of GDP in 2018 for the largest Yim 2012; Sow and Razafimahefa 2017). five SOEs alone. In India, many such SOEs Moral hazard, the soft budget constraint, in the electricity sector are owned by SNGs and the common pool problem have led to that implicitly guarantee their liabilities, subnational fiscal crises in both advanced shifting the risks to SNGs but not implying and emerging economies. Worldwide, esti- an ad hoc increase in risk. mates put the cost of unexpected shocks In practice, however, several factors fur- (surprises) to SNGs’ debt levels—referred to ther amplify the risks when fiscal policy is here as the realization of subnational con- delegated to SNGs. This amplification may tingent liabilities or triggered contingent occur because SNGs tend to lack the liabilities—at 3.7 percent of GDP on aver- capacity to monitor and manage these age over 1990–2014 and as much as risks. The capacity gaps particularly con- 15.1 percent of GDP in some SNGs in the cern implicit contingent liabilities—such as extreme (Bova et al. 2016). the likely default of an SOE owned by the Despite these risks, over the last 25 years, SNG—because they are more difficult to the share of subnational expenditure in gen- identify. Obtaining timely and high-quality eral government spending has increased sub- data on subnational finances and debt may stantially, in both advanced and emerging also be a challenge. Even in several mem- economies (figure 4.2, panel a). South Asia ber countries of the Organisation for has been no exception to this trend (­figure 4.2, Economic Co-operation and Development panel b). India and Pakistan, which have (OECD), such information is available given significant fiscal autonomy to their only with a lag, making it difficult for cen- states and provinces, respectively, for many tral governments to identify and react years, have continued to devolve a rising quickly to emerging fiscal risks (OECD share of public resources to SNGs. Nepal has 2018). been undergoing a substantial transformation Because SNGs often have explicit or to federalism since a new constitution was implicit guarantees from the central gov- finalized in 2015. Bhutan’s Twelfth Five-Year ernment to bail them out in case of finan- Plan (2018–23) proposes the decentralization cial distress, they may also engage in risky of significant decision-making power to its borrowing or run higher fiscal deficits— local bodies (Dzongkhags). Maldives is pur- that is, engage in a behavior called moral suing a greater decentralization of develop- hazard . When subnational fiscal risks ment and service delivery, including through materialize, central governments often hub islands. 136   H IDDEN DEBT FIGURE 4.2  Share of Subnational Expenditure in General Government Expenditure a. Decentralization has risen across the world 45 Percent of total expenditure 40 OECD 10 35 Selected advanced and emerging market economies 30 25 1990 1995 2000 2005 2010 Source: IMF 2014. Note: Percent of total expenditure is an unweighted average of the group. OECD 10 refers to the 10 largest members of the Organisation for Economic Co-operation and Development in terms of GDP. b. Decentralization has also risen across South Asia 60 50 Percent of total expenditure 40 30 20 10 0 11th plan 12th plan 2013–14 2018–19 2016–17 2018–19 2009–10 2018–19 Bhutan India Nepal Pakistan Sources: World Bank 2019a; World Bank staff calculations using data from various ministries of finance. The Unclear Extent of Subnational contracts, public expectations, and political Fiscal Liabilities and Rising Fiscal pressures). These risks may also be direct Risks in South Asia (predictable obligations that occur in any case) or contingent (obligations that are trig- From the perspective of a SNG, fiscal risks can emerge from many different sources. Like gered by a distinct, but uncertain event) central governments, SNGs may face liabili- (Polackova Brixi and Mody 2002). The ties that are explicit (specific obligations degree of fiscal risks depends on the design of defined by laws or contracts) or implicit (an fiscal federalism. In some countries, SNGs expected burden stemming from incomplete may not confront such a high level of risk S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   137 TABLE 4.1  Categorization of Fiscal Risks at the Subnational Level Direct Contingent (obligation in any event) (obligation if a particular event occurs) Explicit Debt (loans contracted, debt securities issued by the Guarantees for public and private sector (liability recognized subnational government) entities (such as firms, publicly owned banks) by a law or contract) Nondiscretionary or legally binding expenditures (such as civil service salaries, pensions) Implicit Social security schemes, future public pensions, Default of a public or private entity on (“moral” or expected future health care financing nonguaranteed obligations obligation) Natural disasters Recapitalization of publicly owned banks Source: Adapted from Polackova Brixi and Schick 2002. from implicit liabilities if these are funded pri- marily by the central government. Table 4.1 Subnational government borrowing does not categorizes sources of fiscal risk from the sub- occur in practice across most of South Asia. national viewpoint. Subnational explicit liabilities are not an immediate source of fiscal risk in most coun- elsewhere in the region do not raise debt tries. Across the region, only SNGs in India financing or issue guarantees (table 4.2). and Pakistan actively contract loans and issue In some cases—Afghanistan, Bangladesh, and guarantees. In India, subnational debt to some extent Bhutan—this is because most amounts to 24.8 percent of GDP,3 the highest SNGs are prohibited from borrowing, likely in among emerging market federations and the interest of minimizing fiscal risks. However, toward the high end of federalist countries in even when SNGs are legally authorized to bor- general (figure 4.3, panel a).4 By contrast, in row—as in the cases of Maldives, Nepal, and Pakistan, where provincial lending was much Sri Lanka—this does not occur in practice. more restricted before the approval of the There are several reasons why this may be the 2010 constitutional amendment, provincial case. First, the limited capacity to formulate debt is estimated to be 4 percent of GDP (see fiscal policy at the subnational level and the next section for details).5 This is toward the lack of creditworthiness make autonomous lower end of federal countries but higher than fiscal policy at the subnational level subna- ­ the average low- and lower middle-income tional borrowing risky and elevate borrowing country (1.9 percent of GDP).6 A key differ- costs. Second, in Maldives and Sri Lanka, the ence between India and Pakistan is in the high level of indebtedness of the central gov- composition of lending: while direct borrow- ernment makes it difficult for SNGs to justify ing from the market accounts for 54 percent borrowing independently. Both economies are of outstanding state liabilities in India today already at high risk of debt distress as they (figure 4.3, panel b), provincial debt in took on large amounts of non-concessional Pakistan mostly consists of external loans on- financing for capital investments in recent lent by the federal government. The move years. Third, decentralization is still nascent in toward market borrowing in India was in the Maldives and Nepal, and virtually nonexis- part due to the recommendations of the tent in Sri Lanka. In the case of Sri Lanka, pro- Twelfth Finance Commission of India for vincial councils were introduced with the greater fiscal autonomy, along with a higher passage of the constitution’s 13th amendment reliance on market discipline among states. (1987), but the central government maintains Subnational government borrowing does control over all policies and most of the plan- not occur in practice across most of South ning process (Kelly and Gunawardena 2016). Asia. Apart from India and Pakistan, SNGs As such, only one-tenth of the total budget is 138   H IDDEN DEBT FIGURE 4.3 India’s Subnational Debt in Comparison with Other Federations and Its Growing Access to Market Loans a. India’s subnational debt is the highest among emerging market federations 80 70 60 50 Percent 40 30 20 10 0 il eg o sia Pa a l a Au ia Be ia Sw ium nd Ge ia ny n da Fe stan n es ria Ar ica pa az in in c So atio ai d l r at ra He exi Un ma na ge st ay la r ov nt In Sp Ne Br Af lg St i st Au er k al Ca Ni ge r M r de h itz d M ut ite rz n ia d ss an Ru ia sn Bo Sources: For India, RBI 2019b; for Pakistan, World Bank staff estimates based on provincial debt bulletins; for other countries, OECD-UCLG 2019. b. Access of subnational governments to market loans has increased over the past decade 100 90 80 70 60 Percent 50 40 30 20 10 0 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 Market loans UDAY, power bonds Other internal debt (mostly NSSF) Loans from banks and FIs Loans and advances from the center Provident funds, etc. Other liabilities Source: RBI 2019b. Note: FIs = financial institutions; NSSF = National Small Savings Fund; UDAY = Ujwal DISCOM Assurance Yojana scheme. devolved to provincial and local authorities, South Asian countries suggests that subna- which have limited responsibilities for service tional fiscal risks could start to emerge as delivery and capital investment (table 4.3). the requisite legal framework and proce- Nonetheless, the push toward more dures evolve to enable SNGs to incur liabili- decentralized decision making in some ties on their own. S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   139 TABLE 4.2  Legal Authority for Subnational Governments in South Asia to Borrow and Issue Guarantees Country Are subnational governments allowed to borrow and issue guarantees? Subnational borrowing limits Afghanistan No. Municipalities are not allowed to borrow on their own. Cities can only borrow Not applicable from the central government. Bangladesh No. Local governments are not allowed to borrow. In the few instances in which the Not applicable national government on-lends internally borrowed funds to city corporations for capital spending, these become Treasury liabilities. Bhutan Yes. Article 22 of the 2008 constitution and the 2009 Local Government Act empower Not applicable local governments to “own assets and incur liabilities by borrowing on their own account subject to such limitations as may be provided for by law.” However, borrowing does not occur in practice. India Yes. Article 293(1) of the constitution empowers states to borrow domestically 3 percent of state GDP and issue guarantees within limits set by state legislation (as applicable). Central government consent is required if the state government is indebted to the center, which is currently the case for all states. Municipal corporations are also allowed to borrow with prior approval from the state government. Maldives Yes. The 2010 Decentralization Act and the 2013 Fiscal Responsibility Law permit local Up to one-third of the council’s income councils (city councils, island councils, and atoll councils) to borrow, obtain government of the previous financial year guarantees, or seek financing. However, borrowing does not occur in practice. Nepal Yes. With the passing of the 2015 constitution, states can now borrow and receive Suggested limit for provincial and local guarantees from the federal government, while local governments (municipalities government borrowing is 10 percent of and villages) can borrow and receive guarantees from federal and state governments. the sum of their share of revenues from Internal borrowing does not occur in practice because the requisite legal framework the value added tax (VAT) and excise and monetary instruments have not yet been finalized. taxes Pakistan Yes. Provinces are allowed to borrow and issue guarantees up to limits specified 0.85 percent of national GDP for by the National Economic Council. All foreign borrowing is on-lent by the central domestic borrowing government. Sri Lanka Yes. Provincial councils are allowed to borrow, as specified in the constitution, but the Not applicable legal framework for provincial borrowing has not yet been developed. Sources: Adapted from Ellis and Roberts 2016 and updated using World Bank 2019b for Bangladesh; IMF 2019b for Maldives; IMF 2019c for Nepal. TABLE 4.3  Vertical Imbalances between Subnational and Central Governments in South Asia SNG revenue as share of total SNG expenditure as share of GG revenue (percent) total GG expenditure (percent) Estimated fiscal gap (A) (B) (B − A) Bangladesh 3.8 8.2 4.4 Bhutan 0.8 26.0 25.2 India 52.8 59.0 6.2 Nepal 10.0 36.0 26.0 Pakistan 56.2 39.4 −16.8 Sri Lanka 4.6 10.0 5.4 Sources: World Bank 2019b; Bhutan Ministry of Finance and Royal Audit Authority (FY2016/17); RBI 2019b; Ministry of Finance, Pakistan (FY2018); IMF 2019c; Ministry of Finance, Sri Lanka (FY2018). Note: Vertical imbalance is defined as the difference between the share of SNG expenditure in total general government (GG) expenditure and the share of SNG revenue in total GG revenue. Consider Nepal, which recently transi- As a result, the SNGs have a relatively large tioned to a federalist system. Its 2015 consti- “fiscal gap”—that is, the difference between tution mandates substantial responsibility for the SNGs’ revenues from its own resources the provision of essential public services— and its own expenditures, expressed as a such as education, health care, and infrastruc- share in total general government (GG) ture—to provincial and especially local expenditures (table 4.3). Although transfers governments, which raise limited revenue. from the center help to finance this fiscal gap, 140   H IDDEN DEBT the demand for subnational borrowing could rules because the rules focus squarely on increase as provinces and local governments explicit liabilities—such as deficit and debt finance a greater share of functions that targets, and in some cases, direct contingent were fulfilled by the center. Similarly, the liabilities such as guarantees—but do not government of Maldives is also taking more deal with implicit contingent liabilities. concrete steps toward administrative and ­f iscal decentralization, 7 having recently Implicit contingent liabilities have served amended the 2010 Decentralization Act to set as an escape path from fiscal rules. aside a share of the state budget for local councils.8 While Maldives is much further The experience with fiscal responsibility from the prospect of subnational borrowing legislation at the central government level has due to its already elevated public debt, these been disappointing (table 4.4). Four countries local councils are legally authorized to bor- (India, Maldives, Pakistan, Sri Lanka) have row and could do so in the future. budget balance rules, but none of them have consistently adhered to the targets specified in Fiscal Responsibility Legislation the legislation. Pakistan’s public debt (com- and Subnational Fiscal Risks prising general government and SOE external debt) stood at 86.5 percent of GDP at the end Based on India’s experience, fiscal responsi- of June 2019—more than 20 percentage bility legislation can help minimize subna- points higher than the target specified in the tional fiscal risks, but only to some extent. Fiscal Responsibility and Debt Limitation Act In India, enactment of state-level fiscal (FRDLA) 2005. Similarly, in Maldives and Sri responsibility legislation began in the early Lanka, public debt ratios are much higher 2000s, following a period of deterioration than the targets. In Sri Lanka, the goals have in states’ fiscal indicators. Spurred on by regularly been modified and postponed, incentives for conditional debt restructuring diminishing the credibility of the fiscal and interest rate relief provided by the responsibility act. While there are many rea- Twelfth Finance Commission, 26 states sons why these rules do not “bite,” one expla- enacted fiscal responsibility legislation nation is that contrary to international best within two years, and all states had adopted practice, none of the fiscal rules implemented them by 2010. The introduction of fiscal by South Asian countries are formally responsibility legislation by these states cor- enforced or monitored by an independent relates with an improvement of fiscal indi- body (Lledó et al. 2017). In addition, only cators: while all 17 general category states Maldives’ and Pakistan’s fiscal responsibility ran revenue deficits in the early 1990s,9 12 legislation have well-specified escape clauses recorded revenue surpluses in 2012–13, and for the case of natural disasters or national almost all states lowered their fiscal deficits security emergencies.10 (RBI 2015). Most states managed to turn Given that most SNGs in South Asia do their high primary deficits into surpluses, not incur liabilities of their own accord, the and the ratio of state debt to gross state remainder of this chapter focuses on Pakistan domestic product mostly declined. However, and India, the only two countries in the region most of these positive effects disappear once where SNGs actively borrow. resources transferred by the central govern- ment are excluded from the current deficit (Simone and Topalova 2009), suggesting Subnational Debt, Data, and that fiscal rules do not encourage indepen- Transparency: Lessons from dent and lasting efforts by states to become Pakistan fiscally sustainable. Moreover, as our analy- SNGs in Pakistan have significant autonomy sis will show, implicit contingent liabilities to borrow. The constitution enables provinces have served as an escape path from fiscal to borrow both domestically and from abroad S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   141 TABLE 4.4  Long-Established Fiscal Rules for the Central Government of Several South Asian Countries Country (year) Budget deficit target Debt target Treasury guarantees India (2003) 3% of GDP by 2008; suspended in 2009, None 0.5% of GDP; requires reinstated in 2013 with a deadline of a statement of explicit 2017–18; in 2016, FRBM committee contingent liabilities in the recommended targeting 3% by FY2020, medium-term expenditure 2.8% by FY2021, and 2.5% by 2023 fiscal framework Sri Lanka (2003) Less than 5% of GDP by 2006 onward Less than 85% by 2006 and less than 60% by Less than 4.5% of GDP; end-2013 amended to 7.5% in 2013 and 10% in 2016 2013 amendment: less than 80% by end-2013 and less than 60% by 2020 Pakistan (2005) Balanced (current) budget by 2008 and Debt ratio to be reduced to 60% by 2013; New guarantees to be limited surplus thereafter; 2016 amendment provision maintained by the 2016 amendment to 2% of GDP imposes limit of 4% of GDP from until FY2017/18, after which it sets out a FY2017/18 to FY2019/20 and 3.5% of GDP transition path toward reducing debt to 50% thereafter of GDP Maldives (2013) Not exceeding 3.5% of GDP by end-2016 Debt (including guarantees) ratio to be reduced No specified limits, except and maintained at that level thereafter; to 60% by the end of 2016; for 2017–22, required that it “should not exceed the primary balance to be in surplus by level of debt to be determined by the minister amount allocated for loans end 2016 of finance or guarantees in the national budget” Source: Adapted from Lledó et al. 2017. Note: FRBM = Fiscal Responsibility and Budget Management Act (2003, India). (on-lent through the federal government) and National Economic Council.13 Given that the to issue guarantees within limits (if any) council is primarily composed of provincial imposed by provincial legislature. Provinces policy makers, this amendment may further also borrow from commercial banks, includ- induce provincial borrowing (Refaqat 2015). ing those that are state owned, primarily to Indeed, the National Economic Council finance purchases of commodities such as raised the cumulative domestic borrowing wheat and sugar (termed “commodity limit for provinces from 0.5 percent of GDP operations”). to 0.85 percent of GDP (approximately PRs Although official figures are not available, 323 billion) in FY2017.14 we estimate Pakistan’s subnational debt to be Nonetheless, the bulk of provincial debt at about 4.0 percent of GDP or PRs 1.56 tril- consists of foreign multilateral/bilateral loans lion as of the end of June 2019.11 The rela- contracted by the federal government, rather tively low level of subnational debt in than domestic borrowing.15 Most of this debt Pakistan is in part due to the existence of a is highly concessionary, with long maturities hard-budget constraint: prior to passage of and fixed interest rates (figure 4.4, panel a). the 18th constitutional amendment of 2010, Although provincial governments decide on provinces were not authorized to raise new the interest rate and disbursement mode of loans if they still held outstanding debt to the each loan, the choice of currency is made by federal government.12 Since all foreign bor- the Economic Affairs division of the federal rowing is on-lent by the federal government government (Manoel et al. 2012). Because to the provincial governments, this meant most of the debt is denominated in foreign that provinces were effectively prohibited currency, primarily in US dollars, provinces from borrowing without explicit consent. are exposed to exchange rate risks (figure 4.4, Since 2010, however, provinces have been panel b). In the event of a significant deprecia- authorized to raise financing and extend tion of the Pakistan rupee with respect to the guarantees within limits specified by the US dollar, increases in outstanding debt stock 142   H IDDEN DEBT FIGURE 4.4  Composition, Currency Denomination, and Interest Rate Structure of Provincial Debt, Pakistan, 2019 a. Provincial debt mostly consists of external debt, b. Most of the debt stock has xed interest rates denominated primarily in US dollars 100 1 1 100 10 16 18 22 17 8 28 27 80 1 2 0 80 Percent of total debt Percent of total debt 3 1 60 60 40 81 81 81 40 78 83 67 73 20 20 0 0 Punjab Sindh KP Balochistan Punjab Sindh KP USD PRs JPY Other Fixed Variable Sources: Subnational debt bulletins and Government of Balochistan 2019. Note: Data are as of end-June 2019. KP = Khyber Pakhtunkhwa. JPY = Japanese yen; PRs = Pakistan rupees; USD = US dollars. and the costs of debt servicing could cause fis- since December 2016, thus limiting the cal stress to provinces, which rely solely on opportunity for time series analysis. Khyber foreign exchange management by the federal Pakhtunkhwa and Sindh publish a single debt government to mitigate such risks (Manoel bulletin on their Finance Department web- et al. 2012). A 25.5 percent currency depreci- sites outlining the composition, currency, and ation in FY2019 illustrates this risk. It caused creditor of the most recent debt stock. Budget the outstanding debt stock for Punjab, Sindh, documents for Sindh have more detailed and Khyber Pakhtunkhwa to jump by information, but only as far back as FY2015. 37.7 percent, 25.0 percent, and 20.5 percent, Balochistan occasionally publishes debt data respectively, year-on-year by the end of June in a white paper on the budget (available for 2019.16 FY2020, FY2015, and FY2010). Poor recording of debt, guarantees, and A second source of debt information is the contingent liabilities elevates the exposure to internal finance account of each province, future fiscal shocks. There is no unified, cen- maintained by the province’s Auditor General trally audited time series of provinces’ debt (AG). These accounts are sent to the AG of levels on an individual basis (published by Pakistan but are not centrally audited or pub- each province’s Finance Department) or on an licly available. For this report, we have aggregate basis (published by the Ministry of obtained these records for all Pakistani prov- Finance or the State Bank of Pakistan). inces, ranging from FY2008 to FY2018.17 In Instead, there are two different sources of general, these provincial debt accounts are provincial debt data. The first source consists constructed based on historical balances, to of debt bulletins, which are published by all which payments and receipts to the province’s provinces except Balochistan. However, debt account at the State Bank of Pakistan are information varies in coverage and does not added. There are some differences among allow for long time series analysis. Punjab’s provinces in how the debt data are accounted Finance Department publishes the most com- for in these reports. For instance, the reports prehensive information, offering a composi- for Sindh do not indicate any buildup of debt tional analysis of debt by origin, currency, stock due to commodity financing, while creditor, and risk, but it has done so only those for Punjab do. Nevertheless, they are S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   143 broadly consistent in their classification of compares figures for Punjab from the two debt over a 10-year horizon. Unfortunately, sources and highlights that while the trajecto- detailed analysis has proven futile due to sev- ries are comparable, the AG public debt eral issues with the data quality. accounts show a lower level of indebtedness. First, the AG accounts data are inconsis- Similar patterns are observed for Sindh and tent with data published in the debt bulletins Khyber Pakhtunkhwa, while the debt levels (figure 4.5, panel a).18 Figure 4.5, panel b are similar for Balochistan. FIGURE 4.5  Discrepancies and Understatements in the Accounting for Provincial Debt, Pakistan a. There are signi cant discrepancies between data in the debt bulletins and the Auditor General accounts for several provinces 800,000 700,000 600,000 500,000 PRs, million 400,000 300,000 200,000 100,000 0 Balochistan Punjab Khyber Pakhtunkhwa Sindh Debt bulletins Auditor General accounts Sources: Subnational debt bulletins and Auditor General public debt accounts. Note: Data refer to end-June 2018. b. Auditor General accounts for Punjab tend to understate the level of debt 800,000 700,000 600,000 500,000 PRs, million 400,000 300,000 200,000 100,000 0 8 9 0 1 2 3 4 5 6 7 8 /0 /0 /1 /1 /1 /1 /1 /1 /1 /1 /1 07 08 09 10 11 12 13 14 15 16 17 20 20 20 20 20 20 20 20 20 20 20 Debt bulletins Fiscal accounts Sources: Subnational debt bulletin and Auditor General public debt accounts for Punjab. 144   H IDDEN DEBT Such discrepancies, in turn, may be due to Second, public debt levels—as reported in two reasons: first, the AG debt reports com- these AG accounts—turn negative for pile data on a cash-accounting basis and do Balochistan in FY2011 and for Khyber not report or account for the depreciation Pakhtunkhwa in FY2017 (figure 4.6, panel a). impact in foreign currency loans; and second, One possible explanation is that some bor- not all loans are reported in AG debt reports rowings were disbursed into other accounts due to discrepancies in accounting practices, but serviced through the State Bank of which leads to omission of certain loans and Pakistan account, leading to an underestima- direct third-party payments. tion of the debt stock.19 FIGURE 4.6  Public Debt Levels for Balochistan and Khyber Pakhtunkhwa, 2007/08–2017/18 a. Negative public debt levels suggest underestimation of the debt stock for Balochistan and Khyber Pakhtunkhwa 60,000 50,000 40,000 30,000 PRs, million 20,000 10,000 0 –10,000 –20,000 8 9 0 1 2 3 4 5 6 7 8 /0 /0 /1 /1 /1 /1 /1 /1 /1 /1 /1 07 08 09 10 11 12 13 14 15 16 17 20 20 20 20 20 20 20 20 20 20 20 Balochistan Khyber Pakhtunkhwa b. Fiscal de cits and the debt trajectory do not track each other well from FY2008 to FY2018 2.0 Percentage point change in debt-to-GDP ratio 1.5 1.0 0.5 0 –0.5 2008 2010 2012 2014 2016 2018 Fiscal year Stock-flow adjustment Fiscal deficit Debt increase Sources: Provincial Auditor General public debt accounts; World Bank staff calculations. S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   145 Finally, there is no apparent relationship General Provident Fund (GPF), available for between fiscal deficits and the debt trajectory government employees, is an emerging fiscal using the AG accounts debt data. Figure 4.6 risk. In Sindh, it is expected that the unfunded (panel b) disaggregates the year-on-year GPF liability will more than double, from PRs change in the debt-to-GDP ratio of aggregate 100 billion in FY2014 to PRs 228 billion by provincial debt into changes in the fiscal defi- 2030, posing significant risk to the sustain- cits and the stock-flow adjustment (SFA)— ability of public finances.22 The governments with the latter measured as the residual. The of Khyber Pakhtunkhwa and Balochistan figure highlights that most of the variation in similarly have their own pension and provi- debt levels cannot be explained by the fiscal dent investment funds, but had not yet deficit. assessed the size of unfunded liabilities at the Apart from public debt, there is also no time of writing. 23 In the case of Khyber regular reporting of risks that may arise from Pakhtunkhwa, the provident fund is an exclu- guarantees and contingent liabilities at the sive liability of the government because provincial level. Provinces do not systemati- employee contributions are not collected. cally record the amount of guarantees and let- Fiscal risks also emanate from the power ters of comfort provided, yet experience sector. Although most of the guarantees are shows that contingent liability shocks can provided by the federal government, provin- exert long-term effects on provincial finances. cial governments also play a role in financing For example, when the Bank of Punjab suf- infrastructure investments in their respective fered some PRs 16.8 billion in losses due to jurisdictions. Out of the PRs 75 billion in nonperforming loans in FY2008, the govern- guarantees issued by the government of ment of Punjab—which owned 51 percent of Punjab, for example, PRs 70 billion accrues the Bank of Punjab at the time—made capital to the power sector. These guarantees come in injections equivalent to PRs 10 billion in the form of (1) credit guarantees of loans FY2010 and PRs 7 billion in FY2011. issued by special purpose vehicles for the con- Subsequently, in FY2015 and FY2017, the struction of power plants and (2) commit- government issued two letters of comfort ment to financial support in the case of totaling PRs 14.2 billion to the State Bank of project cost overruns. While these guarantees Pakistan to guarantee the provisioning are part and parcel of financing much-needed requirement against an agreed amount of capital investments—and do not result in nonperforming loans. Even though the guar- financial outflows unless they are called24— antees have matured and have not been trig- delays in the implementation of such projects gered, budgeting for such large contingent could pose financial liabilities for the provin- liabilities can crowd out public spending on cial government.25 Recording and disclosing more important and immediate development them regularly would help both the provincial priorities. It is unclear whether other provin- and federal governments better manage cial governments have also lent support to potential fiscal risks. their respective commercial banks,20 but simi- lar shocks cannot be ruled out in the future. Unfunded pension liabilities are also a sig- Estimating Contingent Liability nificant source of implicit contingent liabili- Shocks, Adjustment Costs, and ties for provinces. In Punjab, the government Mitigating Factors Using Data for estimates that unfunded accrued pension lia- India bilities stood at PRs 3.8 trillion as of the end Among South Asian nations, India has the of June 2016.21 Although the Punjab govern- longest history and the richest sources of data ment created the Punjab Pension Fund to par- available to analyze subnational fiscal risks. tially fund future pension liabilities, the gap These data make it possible to implement an between the fund’s total assets and projected econometric framework that estimates (1) the liabilities remain significant. Similarly, the probability of contingent liability shocks; 146   H IDDEN DEBT (2) the adjustments that occur after the marginal premium over central government shocks; and (3) their impact on relevant eco- securities, and yields vary by state, because nomic outcomes such as investment. To this there is no explicit central government guar- end, this section first lays out the institutional antee on state borrowing. However, the varia- background for subnational borrowing in tion in yields across states is limited and only India before discussing the methodology, marginally reflects states’ fiscal situation, data, and results of an econometric analysis ­ partially owing to the wide-spread perception of contingent liability shocks. that state securities enjoy an implicit guaran- tee by the central government. RBI also man- ages the borrowing and, through an Institutional Background automated debit mechanism, ensures repay- Indian states enjoy fiscal autonomy to incur ment of states’ liabilities. liabilities, either directly domestically or In 2018, states borrowed primarily from through on-lending of external borrowing by private markets (figure 4.7). In addition, the central government. The Indian institu- 25.6 percent of subnational debt was owed to tional framework regulates subnational bor- pension, savings, and other funds. States also rowing through three mechanisms. First, the borrow from state-owned banks and enter- Finance Commission, a constitutional body prises, such as the State Bank of India and the primarily tasked with determining the distri- National Bank for Agriculture and Rural bution of central funds to states, incentivizes Development, which accounted for slightly fiscal responsibility through the intergovern- less than 10 percent of total borrowing in mental transfer system. For instance, the 2018. Loans from the central government, Thirteenth Finance Commission proposed a which include states’ external borrowing, subnational debt relief scheme for subna- accounted for 3.8 percent of total borrowing tional loans from the central government that in 2018. was extended to states that had reduced their Historically, the development of state debt fiscal deficit. Second, states’ fiscal position and fiscal risks over the last two decades can and debt levels are also regulated as part of be broadly divided into three subperiods. The state-level fiscal responsibility laws, following first, from the late 1990s to about 2004, was the passing of the central Fiscal Responsibility a period of fiscal slippage. In this period, the and Budget Management (FRBM) Act in absence of regulation and central oversight 2003. These laws typically limit fiscal deficits meant that states exposed themselves to sig- to less than 3 percent of GDP and in most nificant contingent and noncontingent liabili- cases prescribe an overall subnational debt ties, resulting in fiscal deficits and rising debt ceiling. Third, SNGs require approval from (figure 4.8, panel a). Fiscal pressure was com- the central government to incur liabilities pounded through the issuance of power whenever they are indebted with the central bonds by state governments, which increased government—which, in practice, applies to all liabilities by 22.8 percent in FY2004. With states. concerns about fiscal risks mounting, the Indian states borrow through so-called c entral Fiscal Responsibility and Budget ­ state development loans, which are dated Management Act was passed in 2003, and the securities issued by state governments. State Twelfth Finance Commission initiated incen- development loans are auctioned through the tives schemes for subnational fiscal responsi- Reserve Bank of India (RBI) on a weekly bility in 2004, which resulted in a second basis, with the issuing states providing the period of gradual consolidation, until about details of envisioned terms and conditions for 2012. More recently, states’ debt has started their borrowing prior to the auction. The RBI increasing again, from about Rs 18 trillion also issues notifications in leading newspapers (measured at 2011 prices) in 2012 to Rs 30 before the auction to assist in marketing. trillion in 2018 (figure 4.8, panel a), or about State development loans are valued at a 25 percent of GDP. Jharkhand and Nagaland S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   147 FIGURE 4.7 Indian States’ Sources of Borrowing, 2018 60 Percent of total borrowing 50 40 30 20 10 0 on nd nt d nc es ris d un , y f nd ce an rp ne SF s AY la nc an rb a ds er ) es d nc fu es NS te w UD he ion m ns ba va lo en te-o ge ve fro Loa et ad et ot sat in er k nd ta (n nd nt res ar n sa s pe M a nk rom co ds, sit m es un Co po ba f nc g tf De va in en w ad rro id ov Bo Source: World Bank staff calculations using data from the Reserve Bank of India. Pr Note: NSSF = National Small Savings Fund; UDAY = Ujwal DISCOM Assurance Yojana scheme. FIGURE 4.8  Aggregate Subnational Debt and UDAY Debt, India a. Subnational debt has tripled in real terms b. Debt from contingent liabilities, such over the last two decades in India as UDAY, has been a key contributor to rising subnational debt 3,500 250 6 3,000 5 Rs, 10 billion, at 2011 prices 200 Percent of total debt 2,500 4 Rs, 10 billion 150 2,000 3 1,500 100 2 1,000 50 500 1 0 0 0 19 1 19 3 19 5 19 7 20 9 20 1 20 3 20 5 20 7 20 9 20 1 20 3 20 5 20 7 19 14 15 16 17 18 E) E) 9 9 9 9 9 0 0 0 0 0 1 1 1 1 (R (B 20 20 20 20 20 19 19 20 20 20 Nominal (left axis) As % of total debt (right axis) Source: World Bank staff calculations using data from the Reserve Bank of India. Note: BE = budget estimate; RE = revised estimate; UDAY = Ujwal DISCOM Assurance Yojana debt relief scheme. have the two highest debt levels, at 49 percent The debt dynamics can be driven by direct and 44 percent of GDP, respectively, whereas and contingent liabilities. A key example of a Assam carries the lowest debt burden, at contingent liability is the Ujwal DISCOM 17 percent of GDP. Assurance Yojana (UDAY) scheme. As part of 148   H IDDEN DEBT this scheme, state governments could take on The realization of guarantees presents up to 75 percent of electricity distribution com- another source of contingent liabilities. Indian panies’ debt starting in FY2016 and repay lend- states issue various guarantees, including for ers through the issuance of new bonds. This loans incurred as part of public-private part- acquisition of debt occurred “below the line” nerships (PPPs) or for the borrowing of public and thus was not recorded as an expenditure. sector enterprises. Outstanding guarantees The participation in UDAY was voluntary, but have decreased in recent years, from a peak of to date 32 states and union territories have 8 percent of GDP in FY2014 to less than joined the scheme. At the end of FY2018, states 3 percent in FY2018 (figure 4.9, panel a). had incurred Rs 30.7 billion in debt related to To manage such guarantees properly and to UDAY, accounting for about 5 percent of states’ buffer their potential impact on the budget, figure 4.8, panel b). total debt stock (­ state governments can invest in a Guarantee FIGURE 4.9  Outstanding Guarantees and Investment in India’s Guarantee Redemption Fund a. States’ exposure to guarantees has decreased in recent years 9 Outstanding guarantees (% of GDP) 8 7 6 Percent of GDP 5 4 3 2 1 0 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 /0 /0 /0 /0 /0 /0 /0 /1 /1 /1 /1 /1 /1 /1 /1 /1 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 Fiscal year b. Bu ers against the realization of outstanding guarantees remain low Investment in the Guarantee Redemption Fund (% of outstanding guarantees) 80 Percent of outstanding guarantees 70 60 50 40 30 20 10 0 a l nd a na h sh ya d a t ur m a m ga ra ish ur Go an es an ra sa ip de la ga la ja en ip ad ry kh izo Od ha an As ga Gu ra n Tr Ha tB Pr ra la eg M aP M Na Te es ta ra M hy Ut W dh ad An M Source: World Bank staff calculations using data from the Reserve Bank of India. S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   149 Redemption Fund (GRF), which is maintained states’ GDP and gross fixed capital formation by the RBI. As of June 2018, only 15 of 29 was taken from the Reserve Bank of India’s states had invested in the fund. Among those “Handbook of Statistics on the Indian that had invested, the buffers accounted for Economy” (RBI 2015). 16.4 percent of guarantees outstanding The analysis of subnational debt develop- figure 4.9, panel b). States can also buffer for (­ ments begins with a decomposition of unex- the repayment of liabilities by paying into the pected shocks into those occurring to the Consolidated Sinking Fund (CSF), also main- budget (“above the line”) and those occurring tained by RBI. Contributions to the CSF are through the SFA (“below the line”) (see annex higher than those to the GRF and stood at an 4A, on methodology). Budgetary (above-the- aggregate of Rs 1,025 billion at the end of June line) shocks are defined as deviations in the 2018, equivalent to 0.71 percent of GDP. realized fiscal deficit from the budgeted fiscal Contributions to the GRF are earmarked to deficit. In contrast, the SFA (below-the-line) cover payments from the invocation of guaran- shock captures all changes in the debt stock tees, whereas the CSF aims at covering liabili- that are not explained by the fiscal deficit. ties from market-based borrowing. Investments A positive SFA can arise for two reasons: first, in both funds can act as collateral to avail of a because of below-the-line acquisitions of liabil- Special Drawing Facility at the RBI at favor- ities and assets (such as due to triggered con- able borrowing rates, which acts as an incen- tingent liabilities); and second, because of tive to invest in the funds. changes to the valuation of the existing debt Indian states face multiple sources of con- stock. Changing valuations can arise, for tingent liabilities. A central question is how the instance, because of movements in the triggering of such potential liabilities affects exchange rate if debt is denominated in a for- the local economy, both directly and through eign currency or because of changes to interest adjustments made by state governments. To rates. While not modeled here explicitly, statis- address this question, we performed an tical discrepancies can also be responsible for economic analysis to quantify the impact of ­ changes to the SFA. historic contingent liability realization at the Figure 4.10 highlights the distribution of subnational level on the real economy, provide the budgetary and SFA shocks through box evidence on how state and provincial govern- ments adjust to them, and identify mitigating FIGURE 4.10  Distribution of Above-the-Line and Below-the-Line factors. Shocks to Subnational Debt, India 4 Data and Methodology To study contingent liabilities at the subna- Shock value, percent of GDP tional level, we constructed a panel data set of 2 Indian states covering a total of 29 states from 1991 to 2018. The primary data source for this investigation was the Reserve Bank of 0 India’s State Finances data set (RBI 2015, 2019b). We complemented this data with vari- −2 ous other sources: information on budgeted expenditure (by line item), adoption of fiscal rules, and fiscal transparency measures were −4 obtained directly from states’ finance depart- ments websites and were hard coded into a comprehensive data set. Detailed information Above-the-line shocks Below-the-line shocks on fiscal transfers by type was obtained from Source: Blum and Yoong 2020. the Finance Commission reports. Data on Note: The figure excludes outside values. 150   H IDDEN DEBT plots. The horizontal line in the middle of the contingent liabilities materializes below figure represents the median of the pooled dis- the line (such as the UDAY debt relief tribution of shocks, while the box represents scheme). Given the prevalence of cash the range between the 75th and 25th percen- accounting, this means that this definition tile. The whiskers show the lower and upper captures contingent liabilities that do not adjacent values, defined as 1.5 times the inter- go through the budget and thus involve, quartile range. Figure 4.10 shows that most of for instance, bailouts of SOEs or pension the budgetary shocks are negative. Indeed, the funds if governments take over their debt, median budgetary shock is negative and the but not through the payment of subsidies. 75th percentile of shocks lies just marginally Similarly, this definition captures the real- above zero, highlighting that states underspent ization of debt guarantees, but not of price their budget in many years. Underspending guarantees. occurs primarily on the capital side, with bud- get execution rates averaging 77.3 percent between FY2010 and FY2018. By contrast, Results: Examining the SFA shocks occur both on the positive and the Occurrence of Contingent negative side (right part of the figure). Liability Shocks Figure 4.10 shows that more than 50 percent We identify contingent liability shocks as of SFA shocks are p ­ ositive and thus unexpect- unexpected shocks to the SFA series, stan- edly increase subnational debt. Taken together, dardized by the baseline debt stock. ­igure 4.10 illustrate that subna- the plots in f Applying the Kalman filter to the series, we tional debt shocks in India occur primarily classify observations for which the residual below the line. exceeds the predicted mean by more than 1 Based on this analysis, and consistent standard deviation as the contingent liabil- with the literature (see, for example, ity shock (annex 4B). Figure 4.11 highlights Bova et al. 2016), this chapter identifies the frequency at which such shocks occur contingent liability shocks as positive over time. The identified shocks occurred spikes in the SFA series. This approach is most frequently in Manipur, Meghalaya, suitable because a significant share of Mizoram, Nagaland, and Tripura, with five events occurring in each state between 1991 FIGURE 4.11  Notable Subnational Shocks in India, 1990–2020 and 2018. Uttar Pradesh (at no event) as well as Andhra Pradesh and Madhya 40 Pradesh (at one event) have experienced the least frequent occurrence over the same Percent of states experiencing distress period. The average frequency of contin- 30 gent liability shocks—as we identify such shocks—is 11.4 percent. The spikes visible in the figure dovetail with 20 the policy narratives and anecdotal evidence on the realizations of contingent liabilities: The 1990s are generally associated with significant 10 fiscal profligacy by the Indian states, culminat- ing in debt crises and resulting in the gradual adoption of fiscal responsibility legislation in 0 the 2000s.26 In addition, during this time, bor- 1990 2000 2010 2020 rowings by state-owned public enterprises were Fiscal year removed from the ceiling for state borrowings, leading to a buildup of debt in the SOE sector Source: Blum and Yoong 2020. Note: The relatively sharp changes in 2007–08 occurred during the global financial crisis. The other and associated government bailouts. spikes in 2000 and thereafter can be related to specific policy events. Furthermore, in 1999, the National Small S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   151 Savings Fund (NSSF) was established, which Such shocks have direct budgetary impacts invested significantly in state securities. A spe- because they raise borrowing costs and limit fis- cial feature of NSSF borrowing is that it is cal space, thus requiring a fiscal policy adjust- unrelated to states’ borrowing requirements: ment. This adjustment can involve reducing because the NSSF is a savings device, states are expenditure or increasing revenue. SNGs may allocated available funding based on a sharing also receive assistance from the center, either quota between the central government and the through increased transfers or by receiving states. As a result of the introduction, the share loans. of NSSF borrowing spiked significantly, from This section assesses econometrically how 0 percent in FY1999 to 31 percent the year Indian states adjust to the realization of con- after, providing a possible explanation for the tingent liabilities (unexpected shocks to significant spike visible in 2000 (Rangarajan SNGs’ debt levels). To this end, we estimate a and Prasad 2012). The spikes in the later 2000s difference-in-difference regression, comparing and 2010s can also be linked to specific policy states before and after a contingent liability events. In FY2008, India conducted the shock to states without a shock in the same Agricultural Debt Waiver and Debt Relief year. Figure 4.12 presents the estimated effects Scheme to the tune of Rs 600 billion, which of SFA shocks on the key fiscal outcome vari- increased public liabilities. From FY2015 ables we consider (see also table 4C.1, in onward, states started adopting the UDAY annex 4C, for the regression tables). The scheme, in which states take on the liabilities of error bars present 90 percent confidence indebted power distribution companies. Finally, intervals calculated using the standard errors in FY2018, eight states provided farm loan from the regression estimation. waivers amounting to 0.32 percent of GDP. Figure 4.12 suggests that debt increases mechanically in response to the contingent liability shock. This increase persists for one How Do State Governments Adjust? additional year, with debt increasing by Contingent liability realizations are defined 4.3 percent in the year after the shock, but as unexpected shocks to SNGs’ debt levels. dissipates thereafter. Governments adjust FIGURE 4.12  Estimated Fiscal Adjustments by Indian Subnational Governments to Contingent Liability Shock 20 9.6 2.6 5.1 8.2 Percentage point change compared to 5.5 4.3 3.7 10 2.9 0.5 0.3 0 treatment group –0.1 –10 –2.9 –2.5 –0.3 –3.8 –20 –30 –21.1 –40 Debt Total Capital Revenue Total Tax Non-tax Fiscal deficit expenditure expenditure expenditure revenue revenue revenue Year of shock Year after shock Source: Blum and Yoong 2020. 152   H IDDEN DEBT contemporaneously to a contingent liability Analyzing this question helps uncover shock by reducing expenditure by a statisti- whether the states face hard or soft budget cally insignificant 2.9 percent, with expendi- constraints—a factor that is likely to deter- ture reverting to trend in the years after. The mine fiscal behavior in the longer term. expenditure adjustment is reflected in both Hence, we estimate the effect of contingent capital and revenue expenditure. The esti- liability realizations on three types of assis- mates further suggest that governments are tance provided by the central government— unable to increase revenue contemporane- loans, grants, and tax devolution—again ously when a contingent liability shock using the difference-in-difference framework. occurs, but that revenue increases in the sub- The estimated treatment effects are presented sequent year by 8.2 percent. This effect is in figure 4.13. driven by an increase in tax revenue. Because Our results suggest that states do indeed of the declining expenditure and only margin- receive assistance from the central govern- ally positive response or no response of reve- ment when a contingent liability shock nue in the year of the shock, the fiscal deficit occurs. This assistance occurs through the decreases by 20 percent. However, the deficit provision of loans from the central govern- remains unaffected in subsequent years. ment and increased tax devolution. Loans Taken together, the results suggest that the from the central government comprise either realization of a contingent liability leads external project borrowing—which is typi- state governments to (1) reduce expenditure, cally fixed multiple years in advance—or loan split approximately equally between capital assistance to the states. Our estimates suggest and revenue expenditure, and (2) increase that loans from the center rise by more than 9 revenue through taxes in the subsequent year. percent in the year of a contingent liability While Indian states enjoy significant fiscal shock and the year after, likely driven by loan autonomy, they receive financial support from assistance from the center to the states. the central government. A central question is Grants and tax devolution are assigned by thus whether the central government provides the Finance Commission for five-year peri- financial assistance when contingent liability ods. Grants are assigned in nominal value. shocks occur or whether states are required to Tax devolution is determined based on a manage the fiscal impact independently. f ormula that considers factors such as ­ FIGURE 4.13  Estimated Assistance from the Indian Central Government to Subnational Governments Hit by Contingent Liability Shocks 20 9.6 9.9 9.2 Percentage point change compared 4.6 6.4 10 3.5 2.0 to treatment group 0 –10 –2.9 –20 –30 –40 Loan from center Grants Tax devolution Interest to center Year of shock Year after shock Source: Blum and Yoong 2020; see also table 4C.2 in annex 4C. S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   153 population, income, and the gap between the economy. It does so by reestimating the previ- state’s income and that of the state with the ous difference-in-difference specification using highest income. These factors, and the gross fixed capital formation (GFCF) (in logs) weights assigned to them, vary among in a given state and year as the outcome Finance Commissions. Our estimates suggest variable. that central government support through Figure 4.14 reports the results by plotting grants is not affected by the shock. By con- the estimated treatment effect estimated for trast, tax devolution received from the central five years before and after the occurrence of government increases by 10 percent in the contingent liability shock. GFCF in the state year after a contingent liability shock. falls significantly in the year of a contingent Why does tax devolution respond but liability shock, continues to decline in the grants do not? This remains somewhat a puz- year after, and remains significantly below zle. One possible explanation is that grants the trend for three years after the event. Then, are fixed in nominal terms by the Finance it gradually returns to the trend. Reassuringly, Commission and are often earmarked and the figure does not a identify significantly committed to specific projects. Therefore, diverging trends between affected and unaf- they provide a limited leeway for responses. fected states before the shock, suggesting that By contrast, the central government allegedly the observed divergence in trends is indeed enjoys flexibility in the timing of the payout driven by the contingent liability shock. of tax devolution. This flexibility provides a To quantify the impacts highlighted in mechanism to counter fiscal shocks at the f igure 4.14, the coefficient estimates in ­ subnational level. Taken together, the evi- table 4C.3, column 1, in annex 4C, confirm dence is consistent with an interpretation that that GFCF falls significantly below its trend states enjoy rather soft budget constraints following a contingent liability shock. The that partially buffer the impact of realizations capital formation experiences a maximum of contingent liabilities. FIGURE 4.14  Decreases in Indian Subnational Governments’ Gross What Are the Economic Costs of Fixed Capital Formation following Contingent Liability Shocks Adjustments to Contingent Liability Shocks? 0.4 Do debt shocks at the subnational level affect Percentage di erence between 0.2 local investments and dampen local economic treatment and control development? Such negative spillovers could occur for various reasons. For one, debt 0 expenditure—​ shocks can reduce public capital ­ as we have shown. This reduction, in turn, −0.2 decreases public capital formation as well as private investment that relies on the execution −0.4 of public investment (such as connective infra- structure) and that is typically “crowded in” −0.6 by public investment. In addition, contingent −5 −4 −3 −2 −1 0 1 2 3 4 5 liability shocks can dampen local investments Years relative to distress event indirectly—for instance, by raising the tax burden, and thus discouraging private capital Source: Blum and Yoong 2020. Note: The figure plots the coefficient estimates βs of the following regression: formation, or by reducing the viability of 5 investment projects, firm creditworthiness, Log  ( GFCF )it = α 0 + ∑β s CLit + s + γ j + µt + ε it . and local lending by banks. For this reason, s =−5 Each coefficient measures the relative value of gross fixed capital formation (GFCF, in logs) s years this section investigates the costs of subna- before and after a contingent liability shock, compared to states with no such shock. The x-axis tional contingent liability shocks for the local plots the time relative to the shock. The y-axis plots the values for the corresponding βs. 154   H IDDEN DEBT reduction of 32.2 percent in the year after the likelihood of contingent liability shocks shock and then returns gradually to trend. around elections. We chose elections because they, arguably, shape the main incentives of policy makers. Elections may influence the What Factors Can Explain and Mitigate timing of when contingent liabilities are real- Contingent Liability Shocks? ized. For instance, policy makers may be To guide policy, a central question is which prone to adopt a more lenient fiscal policy in factors explain contingent liability shocks and the run-up to an election. They can take on how these factors can be reformed to mitigate debt of state-owned enterprises to secure jobs the occurrence and repercussions of these in the short term. Alternatively, policy makers shocks. To investigate this question, we focus may instead delay the shock until after elec- on five factors: political incentives during tions because the adjustments required in elections, transparency, legal frameworks and response to a contingent liability realization fiscal rules, markets, and fiscal capacity and and the impact on the local economy may intergovernmental frameworks. cause negative political fallout. In our data analysis, we focus on state Political Incentives Related to Elections legislative assembly (Vidhan Sabha) elections To identify approaches to mitigating fiscal because they largely determine the state-level shocks, it is important to understand whether governments in India—which hold authority policy makers can influence the timing of over fiscal policy. The econometric analysis when these shocks occur and whether the provides evidence of the interrelationship shocks are affected by political incentives. To between elections and fiscal policy. address these questions, we examine the Figure 4.15 shows that the likelihood of a contingent liability shock increases FIGURE 4.15  Occurrence of Contingent Liability Shocks around significantly in the year before an election, ­ Indian State Legislative Assembly Elections peaks in the year of and after the election, and then gradually reverts to the trend. This 0.4 thus provides direct evidence that contingent Likelihood of distress (relative to control) liability realizations, as defined here, respond 0.3 to the political incentives provided by elections. 0.2 Transparency In addition to elections, increasing transpar- ency is an alternative measure to hold policy 0.1 makers accountable and align their incentives with fiscal responsibility. To assess the effect 0 of transparency measures on contingent lia- bility realizations, we use the gradual adop- tion of debt transparency measures across −0.1 Indian states as a case study (figure 4.16). −3 −2 −1 0 1 2 3 4 Such measures range from the publication of Time relative to election (years) debt and guarantee data in the annual finan- Source: Blum and Yoong 2020. cial statements and budgets to the publication Note: The figure plots the coefficient estimates βs of the following regression: of dedicated reports analyzing debt and 3 CL Shockit = α 0 + ∑β s Electionit + s + γ j + µt + ε it . (in some cases) outstanding guarantees. s =−3 To systematically assess the effectiveness of Each coefficient measures the relative likelihood of a contingent liability (CL) shock occurring s years increased transparency, we collected informa- before and after an election, compared to states with no such election. The x-axis plots the time relative to election, with –1 denoting the year before an election and 1 denoting the year after, for tion on states’ publication of debt and guar- instance. The y-axis plots the values for the corresponding βs. antee-related information from the websites S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   155 of states’ finance departments. We differenti- FIGURE 4.16  Adoption of Transparency Measures and Fiscal Rules ated those states publishing information on by Indian States, 2001–19 debt and guarantees in their annual financial 30 statements from those states publishing dedi- cated debt and guarantee reports. Our data 25 reveal that by the end of FY2019, a total of 22 of the 29 states in the sample had pub- Number of states 20 lished information on debt and guarantees at least once. 15 To estimate the effect of these policy mea- sures on the likelihood of contingent liability 10 shocks, we ran difference-in-difference panel regressions that exploit the staggered publica- 5 tion of reports. The regressions compare the change in the likelihood of a contingent liabil- 0 ity shock in states that have recently increased 20 1 20 2 20 3 20 4 20 5 20 6 20 7 20 8 20 9 20 0 20 1 20 2 20 3 20 4 20 5 20 6 20 7 20 8 19 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 20 transparency to the change in states that have Started publishing debt report Introduced fiscal rules not increased transparency. This approach estimates the effect of policy measures under Source: Blum and Yoong 2020. the assumption that the trajectory of contin- gent liability shocks in adopting and non- FiGURE 4.17 Occurrence of Contingent Liability Shocks around the adopting states would have been similar had Publication of Debt Reports in india the policy measures not been implemented. A 0.3 visualization of the coefficient estimates is Effect on contingent liability realization shown in figure 4.17. Detailed coefficient esti- mates are reported in table 4C.4, in annex C. 0.2 Our estimates suggest that the publica- tion of debt transparency reports does not 0.1 reduce the likelihood of contingent liability shocks initially, but does so with a lag of two years (figure 4.17). This effect persists 0 in subsequent years. Therefore, increased transparency through the publication of −0.1 debt-related information takes time to become effective but, once it does, it per- manently reduces the likelihood of −0.2 shocks.27 Our findings are consistent with −1 0 1 2 3 other empirical work that finds a negative Time relative to publication of debt report (years) correlation between transparency and the Source: Blum and Yoong 2020. stock-flow adjustments in a cross-country analysis (Weber 2012). Starting in FY2002, Indian states gradually adopted fiscal responsibility legislation that The Legal Framework and Fiscal Rules curbed the fiscal deficit at 3 percent and the In addition to increased accountability and revenue deficit at 0 percent, and in some altered political incentives, the likelihood of cases, imposed restrictions on debt and guar- contingent liability realizations can also be antees. The adoption of subnational fiscal affected by legislative changes that tie policy rules increased rapidly in FY2006, and most makers’ hands. To study the effectiveness of states had adopted fiscal rules by FY2011 such measures, we focus on the adoption of figure 4.16). We compiled data from states’ (­ subnational fiscal rules in Indian states. finance department websites on the adoption 156   H IDDEN DEBT of fiscal rules and used a similar strategy to control for state and year fixed effects. The compare the change in the likelihood of con- results suggest that interest rates paid do not tingent liability shocks for states that had respond to the breaching of the fiscal deficit recently adopted a fiscal rule to those who rule by Indian states (figure 4.19). Future had not. Detailed coefficient estimates are research could firm this preliminary finding reported in table 4C.4. by using actual current rates on Indian states’ Our estimates suggest that the likelihood debt—either newly issued, debt with floating of a contingent liability shock diminishes sig- rates, or debt refinancing. nificantly in the year before and the current year in which a fiscal rule is adopted Fiscal Capacity and the Intergovernmental (­ figure 4.18). Although this estimate is statis- Framework tically significant, results also suggest that it is The occurrence and impact of contingent lia- only temporary, because our estimates do not bility shocks can also depend on the states’ detect a significant effect of the fiscal rule capacity to buffer shocks. For example, some adoption on the occurrence of contingent lia- states have lower potential for generating bility shocks in subsequent years. their own revenue, or they depend more on transfers to fund their spending than other Markets states. Our evidence on fiscal rules is consistent with To examine this hypothesis, we distin- the perception that fiscal rules by themselves guish between special and general category are unlikely to be a major driver of fiscal con- states in India. India’s National Development solidation.28 This perception is shared by the Council has designated 11 states as special markets. To validate this market perception, category states, owing to their hilly and dif- we estimated a panel regression of interest ficult terrain, low population density or size- calculated rates paid by states in a given year—­ able share of tribal population, strategic as annual interest payments divided by the location along borders with neighboring debt stock—on an indicator of whether a countries, economic and infrastructural state breached the fiscal deficit rule of backwardness, and/or nonviable nature of 3 percent of GDP. In the estimation, we state finances (PRS India 2013). This classifi- cation has two practical implications. First, FIGURE 4.18  Occurrence of Contingent Liability Shocks around the given their unique circumstances, special cat- Enactment of Fiscal Rule egory status can be considered a proxy for low fiscal capacity, thus making special cat- 0.2 egory states more prone to contingent liabil- E ect on contingent liability realization ity shocks and reducing the likelihood that 0.1 they are able to mitigate the impact on the real economy. Second, special category states have his- 0 torically received favorable terms in the allo- cation of central funding. Until India’s −0.1 Planning Commission was abolished, its funding allocation favored special category states by reserving 30 percent of total funds −0.2 for them and distributing funds at a ratio of grants to loans of 90 percent to 10 percent, −0.3 compared to 30 percent to 70 percent for gen- eral category states. This favorable treatment −2 −1 0 1 2 3 ended with the replacement of the Planning Time relative to fiscal rule enactment (years) Commission by the National Institution for Source: Blum and Yoong 2020. Transforming India (NITI Aayog) in 2015. S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   157 While central funds transferred through the FIGURE 4.19  Breaching the Fiscal Rule: The Effect on Interest Rates Finance Commission do not explicitly con- Paid by Indian States sider special category status when allocating 0.4 Effect of breaching fiscal rule on interest rate funds, grants assigned by the Finance Commission implicitly primarily benefit spe- cial category states. As such, revenue deficit grants accrue primarily to special category 0.2 states. Taken together, special category states are thus states that (1) have comparatively weak fiscal capacity to buffer shocks and (2) face softer budget constraints than general category states. 0 To understand whether low fiscal capacity impacts contingent liability shocks, we repeat the analysis and expand it by differen- tiating between special and general category −0.2 states. The estimation results suggest that −2 −1 0 1 2 special and general category states experi- Time relative to breaching rule (years) ence contingent liability shocks at compara- Source: Blum and Yoong 2020. ble ­frequencies—13 percent of the time for special category states and 10 percent for general category states. However, low fiscal reports reduces the likelihood of an imme- capacity in special category states implies diate contingent liability shock in the spe- that contingent liability shocks have a more cial category states, compared to a adverse impact on the local economy of spe- two-year lag in general category states. cial category states than general category Taken together, the combination of lower states. Namely, in general category states, fiscal capacity and softer budget con- GFCF contracts by only 2.5 percent in the straints can increase the impact of contin- year after a contingent liability shock—an gent liability shocks on the local economy. impact not statistically distinguishable from However, the evidence also illustrates how zero. By contrast, in special category states, basic institutional measures could effec- GFCF contracts by more than 60 percent in tively mitigate contingent liability shocks the year after a contingent liability shock. in subnational states with initially lower Therefore, lower fiscal capacity can leave fiscal capacity and softer budget limited space for subnational governments to constraints. buffer contingent liability shocks and can Six Takeaways. In sum, our quantitative expose the local economy to a more adverse research into the fiscal stance of Indian states economic impact of contingent liability yields six takeaways: shocks. Because special category states enjoy Contingent liability shocks—as measured 1.  softer budget constraints, measures that in this report—occur about 11 percent of enhance policy makers’ incentives and the time. harden their budget constraints could be T he contingent liability shock triggers fis- 2.  more effective. For instance, in special cat- cal adjustments through reduced expendi- egory states, the introduction of fiscal rules ture and increased revenue. decreases the likelihood of a contingent S tates enjoy relatively soft budget con- 3.  liability shock in the years after their intro- straints because contingent liability duction, unlike in general category states. shocks trigger support from the center. Moreover, increased transparency through Contingent liability shocks have material 4.  the publication of debt and guarantee impact on local economic development 158   H IDDEN DEBT because they reduce local investments for multiple years. The accrual of contingent liabilities is a policy S hocks do not have purely external 5.  decision that is shaped by the incentives sources or origin. Instead, they respond to political incentives, can be mitigated of local policy makers and their abilities through increased transparency, and their to manage subnational fiscal risks. impact depends on states’ fiscal capacity. F iscal rules and markets currently do 6.  local policy makers and their abilities to man- not impose sufficient discipline on states’ age subnational fiscal risks. Broadly, our anal- finances to mitigate contingent liability ysis has focused on four factors that influence shocks. fiscal risks. The first is transparency, which, in an electoral system, is crucial to hold policy makers accountable. The second is a legal Improved Transparency and framework, including fiscal rules—either Fiscal Rules, the Disciplining self-imposed or imposed by the central gov- ­ Role of Markets, and Better ernment—that limits the ability of subnational Intergovernmental Frameworks decision makers to accrue excessive liabilities. Are Needed to Achieve Better The third is market pricing, which ensures Subnational Fiscal Outcomes in that the debt financing cost incurred by SNGs South Asia is commensurate with the subnational fiscal risk. The fourth is fiscal capacity and its reflec- This chapter has reviewed the exposures to tion in the intergovernmental framework. subnational fiscal risk across South Asia and Based on these considerations, the discussion has provided new evidence on the adverse that follows proposes policy recommenda- effects of contingent liability shocks on fiscal tions for governments in South Asia to achieve and economic outcomes in India. greater fiscal discipline. It shows that contingent liability shocks occur relatively frequently, trigger fiscal adjust- Policy Recommendations ments, and are influenced by policy makers’ incentives as shaped by the prospects of elec- Transparency tion, transparency, fiscal rules, existing fiscal The effect of transparency measures on the space, and the softness of budget constraints. It management of subnational fiscal risks may also shows that contingent liability shocks sig- be slower but more significant and persis- nificantly affect local economic development: tently positive than the other factors. triggered contingent liabilities reduce invest- Gradually, the Indian states adopted measures ment in Indian states for up to four years after to improve transparency and public informa- the shock and thus dampen local economic tion on subnational debt and contingent lia- activity. India’s experience is illustrative for the bilities. The positive effect of these measures rest of the region, especially for countries such took time to materialize, but when they did, as Pakistan, where provincial borrowing has the effects appeared significant and lasting. been expanding, and for Maldives and Nepal, There is no reason why India’s positive expe- which have started to decentralize fiscal policy. rience with fiscal transparency at the subna- Our analysis suggests some pathways to tional level could not be replicated more mitigate contingent liability shocks and the widely in South Asia and beyond. associated negative spillovers. For the path- To increase fiscal transparency across South ways to be effective, policy makers must Asia, central and subnational governments understand that the realizations of contingent could undertake three measures. A first step liabilities are rarely exogenous events. The would be the adoption of accounting stan- accrual of contingent liabilities is a policy dards that highlight contingent liability risks decision that is shaped by the incentives of when they accrue, not when they materialize, S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   159 to allow for adequate budgeting and decision government transfers more resources to the making. This would require moving from the state through tax devolution. This bailout cash-based standards prevalent in South Asia hurts the central government’s finances. It toward accrual accounting. Second, SNGs reinforces the states’ perception of the “soft should collect and consolidate information on budget constraint” that exists in federalist debt and other sources of fiscal risks in a single systems and reduces the incentives for states entity at the subnational level.29 This unit to address underlying sources of fiscal risk. could be a specialized debt management unit India has had a mixed experience with within the finance department that fulfills a subnational fiscal rules. However, other coun- back-office function for the entire subnational tries in the region should not automatically government. While many Indian states and discard this policy tool. For instance, Pakistan Pakistani provinces have established debt man- already has a legal limit on domestic borrow- agement offices, information on debt and ing; however, it does not consider that most sources of fiscal risks remain scattered across of its provincial debt comes through external institutions. Centralizing them into one entity loans that are on-lent from the central gov- would enable the production of consolidated, ernment. This practice makes the limit irrele- whole-of-government financial statements. vant. Before Pakistan can start adopting Third, this information could then be audited, recommendations concerning fiscal rules and publicized, and analyzed by an independent other debt limitations, more comprehensive national agency to ensure its consistency and and timely data collection on provincial accuracy. This agency could be an independent finances is needed (see box 4.1). fiscal council (see discussion that follows). International experience suggests that fiscal While data on state finances in India are rules are most effective when they help rein- more comprehensive and easier to access force a political commitment to fiscal responsi- than those of Pakistan, there are still gaps in bility. To that end, establishing state-level reporting that hinder greater fiscal transpar- institutions and strengthening central-level ency. There are no consolidated financial institutions could improve the implementation statements for the whole of government, of fiscal responsibility legislation in India. while audited accounts at both the central Specifically, the recommendation by the and state level take about 10 to 12 months Fiscal Responsibility and Budget Management to be produced. Moreover, there are no con- Act (FRBM) Committee for the central gov- solidated accounts on state-owned enter- ernment to set up an independent fiscal coun- prises. The quality of accounting standards cil could also cover Indian states. This council is also uneven across government levels, would be responsible for ensuring compliance making it difficult to consolidate informa- of states with the fiscal rule and examining tion across government jurisdictions. justifications for deviating from expected fis- Publishing an integrated, total public debt cal targets—rather than arbitrarily evoking database that includes explicit and implicit “escape clauses” in existing legislation. The guarantees would help the states and the independent fiscal council could have powers central government identify threats to fiscal to punish fiscal laxity and reward subnational sustainability in a more systematic and fiscal discipline. timely manner. In countries with strong public sector governance and high institutional capacity, ­ Legal Framework and Fiscal Rules subnational fiscal councils could also be a Although many countries have imposed limits solution. For example, in Iceland, the on subnational borrowing through fiscal Municipal Fiscal Oversight Committee rules, our analysis has shown that these are (MFOC) has the power to impose sanctions not always effective in limiting fiscal shocks. on municipalities that breach fiscal rules. Part of the problem is moral hazard. When While fiscally responsible municipalities have states are in fiscal distress, the central greater autonomy, municipalities with 160   H IDDEN DEBT BOX 4.1  Recommendations for Improving Fiscal Reporting and Transparency in Pakistan Given Pakistan’s high debt ratio and general fi ­ scal databases should be institutionalized and aligned stress, more attention to monitoring and disclos- with the federal Debt Management and Financial ing subnational fiscal risks is especially war- Analysis System (DMFAS). ranted. While provincial debt may be under Pakistani provinces should also endeavor to control now, the lack of transparency elevates identify and report on contingent liabilities by ­ fiscal risks in the near and medium terms. reporting on the number and total amount of In addition to the general recommendations guarantees explicitly issued to both private and presented in this chapter, the federal government public enterprises on a regular basis. Some discus- or the Controller General of Accounts could sion of implicit obligations (such as those embed- establish some standards about what is consid- ded in contracts for public-private partnerships) ered subnational public debt and mandate a for- should also be included, at least qualitatively, in mat against which all provinces must report debt provincial budget documents. The governments of stocks. Provincial debt bulletins should provide a Punjab and Sindh do this to some extent in their clearer breakdown of domestic and external debt latest white papers on their respective budgets, but figures, especially on (1) commodity financing; do not undertake systematic evaluations of such (2) debt to the federal government; (3) guarantees implicit contingent liabilities. Continuing to by type/sector of institution; and (4) type of credi- improve the functioning of provincial debt man- tors. Costs of debt should also be made explicit agement offices and the coherence of debt manage- along with information on redemption schedules. ment strategies would help provinces build the To improve transparency, reporting should be capacity to undertake such an assessment, and in made public on the websites of the province’s doing so minimize the likelihood of unexpected finance departments and as part of the Debt contingent liability shocks in the future.a Policy Coordination Office’s (DPCO) publica- tions. Eventually, standardized provincial debt a. All provinces except Balochistan have established debt management units. excessive debt must obtain approval from the have limits on outstanding risk-weighted MFOC for all major revenue, expenditure, guarantees, in accordance with their fiscal and borrowing decisions. The MFOC can responsibility legislation, it is important that also withhold transfers and recommend to the the states (1) calculate these risk weights minister of local government that a munici- accurately; (2) undertake debt sustainability pality have its fiscal powers vested in a finan- analysis on total public and publicly guaran- cial management board. teed debt; and (3) place limits on guarantees As subnational fiscal autonomy grows, in relation to their credit risks. countries should consider instituting robust guarantee management frameworks and poli- Market Pricing cies. This involves adopting the necessary leg- Markets play an important role in influencing islation for the issuance of guarantees, clear the fiscal behavior of subnational govern- procedures for assessing and monitoring such ments. When access to debt markets and bor- guarantees, and ideally, a register of subna- rowing costs reflect the likelihood of fiscal tional guarantees maintained at the subna- stress, policy makers have an incentive to tional level and/or central level. In India, state sustain fiscal discipline (de Groot, Holm- ­ governments impose guarantee fees varying Hadulla, and Leiner-Killinger 2015). There is from 0.5 percent to 2 percent of the total evidence, however, that this mechanism does guarantee amount, but this is often waived in not function effectively. In India, yields vary practice (RBI 2019a). Although several states too little across states to reflect any fiscal S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   161 metrics (Saggar and Adki 2017). Our analysis lowers risk premiums. As values of the ratio confirms that the response of borrowing costs increase, the risk premiums decrease more to the breaches of fiscal rules is limited strongly because a larger ratio increases the (figure 4.19). Similar findings in Canada and ­ likelihood of the SNG receiving a bailout. Germany have been linked to explicit and Enhancing market signals could involve implicit promises of central bailouts the following measures (IMF 2018): (Schuknecht, Von Hagen, and Wolswijk 2009; Booth, Georgopoulos, and Hejazi 2007). •  Improving the quality, coverage, and time- By contrast, evidence for the United liness of data on states’ fiscal health . States—whose 11th constitutional amend- For instance, in Brazil, SNG fiscal data on ment prohibits subnational bailouts—shows deficits and debt are available on a quar- a link between fiscal deficits and borrowing terly basis and disaggregated below the costs for states. It reinforces the evidence state level. States should also endeavor to that explicit or implicit support from the provide more complete information on center matters (Bayoumi, Goldstein, and explicit and implicit contingent liabilities. Woglom 1995). •  Adopting “no bailout” clauses in intergov- Allowing markets to function in South ernmental arrangements. In Spain, SNGs Asia requires the development of competitive that miss fiscal targets may forfeit parts of sovereign bond and debt markets that accu- their fiscal autonomy and are required to rately price risks. Market signals for subna- submit restructuring plans. Adopting similar tional borrowing in India and Pakistan are measures in Indian states could make subna- significantly distorted because of (1) implicit tional fiscal responsibility laws more credi- guarantees of the central government on sub- ble and catalyze reforms toward greater national debt; (2) on-lending of external debt transparency and market discipline. by the central government that provides an •  Lowering the statutory liquidity ratio explicit credit risk guarantee; and (3) in the requirement to further liberalize financial case of India, the joint auctioning of state markets and improve bond market liquid- securities by the RBI that pools subnational ity. More liquid markets are more efficient fiscal risks across states. at pricing because of higher trades that mix In addition, markets do not price the (non-) the preferences of diverse market partici- transparency in the risk premiums of govern- pants and, in turn, help discover correct ment bond spreads. Bernoth and Wolff (2008) prices more effectively. show that in addition to the level of indebted- ness, the increase in the “creative” part of fis- Fiscal Capacity and the Intergovernmental cal policy and accounting is punished by Framework markets across the European Union at the The intergovernmental fiscal framework must national level. Creative accounting should address a tension between providing incen- increase risk premiums because if a country is tives and insurance to the subnational govern- nontransparent, financial markets take gim- ments. On the one hand, higher reliance on mickry as a “tip of the iceberg” signal. One central transfers can weaken accountability concerning result of Bernoth and Wolff’s anal- and impose a soft budget constraint, leading ysis is that the disciplining force of markets to inefficient spending and fiscal unsustain- may not work in a monetary union. This dove- ability. In India, for instance, states that tails with our results for the subnational bond depend more on central government revenues market in India and weakens the hope in mar- tend to have higher SNG debt (figure 4.20). ket forces to help discipline the fiscal affairs of This is consistent with the notion of a widely Indian states. Moreover, Heppke-Falk and documented “flypaper effect,” which suggests Wolff (2008) find evidence of investor moral that greater central grant allocation hazard even in the German subnational bond attracts greater SNG ­expenditure—sometimes market in which the larger interest payments- above the sustainable level.30 On the other to-revenue ratio of SNGs counterintuitively hand, fiscal capacity at the subnational level is 162   H IDDEN DEBT FIGURE 4.20  Reliance of Indian States on Central Government the budget. In Argentina, following a subna- Revenues and Share of State Debt, 2017 tional pension crisis, the central government absorbed subnational pension liabilities but 50 JK combined these with reforms to the pension system to support fiscal sustainability going State debt, percent of state GDP MZ forward. MN 40 NL WB Annex 4A. Methodology HP PB UP A quantitative study of contingent liabilities ML requires a systematic measurement of their 30 RJ KL BR GA effects should they materialize. To this end, it TR HR AR is useful to conceptually define shocks to the SK debt level as deviations from expectations, GJ MP defined as follows: 20 TN AP OR AS MH KA Debtt − E(Debtt ) = Debtt − (Debtt−1 +FDPlan) 20 40 60 80 100 = (FD − FDPlan ) + SFAt , Percent of revenue from center (4A.1) Source: World Bank staff calculations using data from the Reserve Bank of India. Note: AP = Andhra Pradesh; AR = Arunachal Pradesh; AS = Assam; BR = Bihar; GA = Goa; GJ = Gujarat; HP = Himachal Pradesh; HR = Haryana; JK = Jammu and Kashmir; KA = Karnataka; where Debtt is the actual subnational debt in KL = Kerala; MH = Maharashtra; ML = Meghalaya; MN = Manipur; MP = Madhya Pradesh; year t; E(Debtt ) is the expected debt level in MZ = Mizoram; NL = Nagaland; OR = Odisha; PB = Punjab; RJ = Rajasthan; SK = Sikkim; TN = Tamil Nadu; TR = ­Tripura; UP = Uttar Pradesh; WB = West Bengal. year t; FD is the fiscal deficit; FDPlan is the planned fiscal deficit; and SFA is the stock- typically limited, while the central govern- flow adjustment. ment can collect sizable revenues through Equation (4A.1) defines the expected debt direct and indirect taxes. This means that level as the amount of liabilities accumulated SNGs have limited buffers when they are hit if the budgeted fiscal deficit had been real- by idiosyncratic shocks. A soft budget con- ized and any debt accumulation had straint and central government support are occurred through the budget. The equation then needed to prevent a large fiscal adjust- also highlights that the actual debt level at ment that can negatively affect the local econ- time t can differ from its expected value for omy.31 Such support occurs frequently, both two reasons: first, because the realized fiscal in India, as documented in this chapter, and deficit, FD, can differ from the planned fiscal around the world (see, for example, Cordes et deficit FD Plan ; and second, because of al. 2014).32 changes that do not go through the budget— Countries have taken different paths to that is, changes that happen “below the line” resolving the trade-off between incentives and in deficit accounting. The latter is often insurance. The paths typically contain a mix referred to as the stock-flow adjustment between providing central bailouts in (SFA), defined as the residual of the year-on- exceptional cases while maintaining control ­ year change in debt after accounting for the of subnational financing in normal years. For fiscal deficit: instance, after the bailouts of regional govern- ments, Spain passed a budget stability law Debtt − Debtt − 1 = FDt + SFAt.(4A.2) that enhanced transparency and monitoring of subnational budgets through an early The SFA thus captures all changes in the warning system. The country also set fiscal debt stock that are not explained by the fiscal rules that contained sanction and enforce- deficit. To identify the factors that can drive ment mechanisms. Colombia adopted a for- the SFA, it can be useful to consider the mal bankruptcy procedure for SNGs to ­ government debt stock in year t as consisting address debt solvency issues without affecting of a single representative bond, Bt , priced S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   163 at rate pt  . Net of the fiscal deficit, the debt Annex 4B. The Kalman Filter stock then evolves as follows: The purpose of filtering is to extract useful SFAt = (pt Bt − pt−1 Bt−1) information from a signal, removing the = [pt (Bt − Bt−1) + (pt − pt−1) Bt−1]. noise. The Kalman filter is the best known of (4A.3) these filtering methods. It is a recursive algo- rithm that estimates unknown variables using Equation (4A.3) highlights that the SFA imperfect measurements of these variables. In can arise for two reasons: first, because of the our application, the unknown (state) variable below-the-line acquisition of liabilities (and that we are trying to estimate is the underly- assets), holding their valuation constant; and ing level of the stock-flow adjustment (or second, because of changes to the valuation of other public finance series) after we have fil- the existing debt stock. Changing valuations tered out the noise from expected expendi- can arise, for instance, because of movements tures or debt waivers. in the exchange rate if debt is denominated in In order to estimate this latent variable, we a foreign currency or because of changes to must model how we believe it behaves. Since interest rates. While not modeled here explic- we are using time-series data, we focus on itly, statistical discrepancies can also be modeling our series as autoregressive inte- responsible for changes to the SFA.33 grated moving average (ARIMA) processes, Consistent with the literature (see, for as they are highly flexible. To select the example, Bova et al. 2016), this chapter iden- ARIMA model that best fits our data series tifies contingent liability shocks using the SFA for each subnational region, we implement because a significant share of contingent lia- the Hyndman-Khandakar algorithm. This bilities materializes “below the line” (such as algorithm selects the model that minimizes the UDAY debt relief scheme in India). the Akaike information criterion. “Above-the-line” contingent liability shocks Kalman defined his filter using state- to the fiscal deficit, such as relief expenditures space methods, which simplifies implemen- related to natural disasters, are rare in Indian tation in discrete time. Therefore, we states. Given the prevalence of cash account- rewrite the best ARIMA model for each ing, this means that this definition captures subnational entity in its corresponding contingent l ­iabilities that do not go through state-space form and estimate this model the budget—for instance, bailouts of state- using the square-root filter to numerically owned enterprises or pension funds if govern- implement the Kalman filter recursions (De ments take over their debt—but not through Jong 1991; Durbin and Koopman 2001, the payment of subsidies. Similarly, this defi- sec. 6.3). nition captures the realization of debt guaran- When the model is not stationary, the tees, but not of price guarantees. ­ filter is augmented as described by De Jong To identify unexpected shocks in the SFA, (1991), De Jong and Chu-Chun-Lin (1994), we apply a Kalman filter to the series. and Durbin and Koopman (2001, sec. 5.7). Conceptually, the Kalman filter predicts an We then estimate the parameters of this expected value of the series for the next linear state-space model by maximum likeli- period given its historic trajectory. Annex 4B hood. The Kalman filter is used to construct provides a detailed description of the statisti- the log likelihood, assuming normality and cal methodology. A contingent liability shock stationarity. in this application is then defined as a data Once we have these parameter estimates, observation that sufficiently exceeds the pre- we estimate the underlying states at each time dicted expectation. More specifically, outliers period using previous information from the are defined by standardizing the Kalman filter data series. The data series is predicted by residuals and classifying any observation that plugging in the estimated states. The residuals lies more than 1 standard deviation above the are then calculated as the differences between mean as a contingent liability (shock). the predicted and the realized data series. 164   H IDDEN DEBT Annex 4C. Regression Tables TABLE 4C.1  Effect of Contingent Liability Realizations on Fiscal Variables (2) (3) (4) (5) (6) (7) (8) (1) Total Capital Revenue Total Tax Non-tax Fiscal Variables Debt expenditure expenditure expenditure revenue revenue revenue deficit CL shock 0.0550*** −0.0294 −0.0380 −0.0252 −0.00344 0.0290 0.0373 −0.211** (0.0185) (0.0192) (0.0505) (0.0220) (0.0333) (0.0283) (0.0350) (0.0877) CL shock (t−1) 0.0426** 0.00536 0.0511 −0.000616 0.0821** 0.0960** 0.00263 0.0258 (0.0170) (0.0173) (0.0502) (0.0184) (0.0327) (0.0358) (0.0315) (0.0769) CL shock (t−2) 0.00610 0.0166 0.0366 0.0150 0.0366 0.00839 −0.0220 −0.0360 (0.0181) (0.0126) (0.0474) (0.0150) (0.0361) (0.0383) (0.0328) (0.0862) Observations 644 619 619 619 644 619 644 613 R-squared 0.994 0.994 0.953 0.994 0.981 0.990 0.955 0.916 Source: Blum and Yoong 2020. Note: The table estimates the following regression: Yit = β0 + β1CL Shockit + β2CLShockit−1 + β3CLShockit−2 + γi + μt + εit. Yit denotes the outcome variable of interest, measured in logs. This specification allows for persistent effects of contingent liability shocks by including two lags of the independent variable. The coefficients of interest are β1 , β2 , and β3, which measure the effect of the contingent liability shock on the outcome variable. These coefficients can be interpreted as causal under the assumption that the trajectory of the outcome variable had been similar in affected and nonaffected states in the absence of the shock. CL = contingent liability. Standard errors clustered at the state level in parentheses. All outcome variables are in logs. *** p < 0.01, ** p < 0.05, *p < 0.1. TABLE 4C.2  Effect of Contingent Liability Realizations on Assistance from the Central Government (1) (2) (3) (4) Variables Loan from center Grants Tax devolution Interest to center CL shock 0.0985*** 0.0354 0.0200 0.0458 (0.0346) (0.0258) (0.0330) (0.0385) CL shock (t−1) 0.0916** −0.0286 0.0956** 0.0640* (0.0355) (0.0323) (0.0376) (0.0327) CL shock (t−2) 0.0491 −0.0249 −0.0136 0.126** (0.0386) (0.0348) (0.0464) (0.0503) Observations 644 619 619 588 R-squared 0.969 0.966 0.981 0.956 Source: Blum and Yoong 2020. Note: Standard errors clustered at the state level are in parentheses. All outcome variables are in logs. CL = contingent liability. *** p < 0.01, ** p < 0.05, * p < 0.1. S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   165 TABLE 4C.3  Effects of Contingent Liability Realizations on Gross Fixed Capital Formation Outcome: log(GFCF) Heterogeneity variables (1) (2) (3) Variables All states Special category state High capital expenditure CL shock −0.196 0.0613 0.362 (0.146) (0.124) (0.272) CL shock (t−1) −0.322** −0.0257 0.130 (0.140) (0.0925) (0.205) CL shock (t−2) −0.236 0.133 0.0693 (0.160) (0.0973) (0.192) CL shock × Heterogeneity variable −0.612* −3.419 (0.302) (2.156) CL shock (t−1) × Heterogeneity variable −0.594*** −2.731* (0.210) (1.454) CL shock (t−2) × Heterogeneity variable −0.843*** −1.730 (0.262) (1.496) Observations 532 532 532 R-squared 0.944 0.948 0.946 Source: Blum and Yoong 2020. Note: CL = contingent liability; GFCF = gross fixed capital formation. *** p < 0.01, ** p < 0.05, * p < 0.1. TABLE 4C.4  Effect of Adoption and Transparency of Fiscal Rules on the Likelihood of Contingent Liability Shocksa Outcome: Realization of a contingent liability at time t (1) (2) (3) (4) (5) (6) Exogenous Fiscal Debt Guarantee Fiscal Debt Guarantee variable rule report report rule report report t −0.104** 0.0220 −0.0550 −0.0962 0.107 −0.0253 (0.0453) (0.0814) (0.0682) (0.0597) (0.123) (0.114) t−1 −0.0424 −0.0747 0.0121 0.0179 −0.0175 0.119 (0.0713) (0.0654) (0.0997) (0.0967) (0.0935) (0.147) t−2 0.0295 −0.133*** −0.107*** 0.0663 −0.115** −0.0909** (0.0925) (0.0391) (0.0327) (0.121) (0.0470) (0.0402) t−3 0.0478 −0.117*** −0.0830*** 0.0299 −0.126*** −0.0892** (0.103) (0.0274) (0.0252) (0.123) (0.0313) (0.0335) t × SCS −0.0152 −0.219* −0.0772 (0.135) (0.125) (0.125) (t−1) × SCS −0.172* −0.158 −0.270* (0.0925) (0.0978) (0.150) (t−2) × SCS −0.118 −0.0465 −0.0412 (0.148) (0.0719) (0.0631) (t−3) × SCS 0.0268 0.0224 0.0157 (0.166) (0.0414) (0.0391) Observations 650 650 650 650 650 650 R-squared 0.084 0.087 0.083 0.087 0.091 0.087 Sources: RBI 2019a; Blum and Yoong 2020. Note: The table presents the results of the following panel regressions: CL Shockit = β0 + β1Xit + β2 Xit−1 + β3Xit−2 + β4Xit−3 + δi + μt + εit. The outcome variable is an indicator variable that takes the value 1 for state-year observations in which a contingent liability realization was identified, using the methodology outlined previously. Xit denotes the regressor of interest. Depending on the specification, it either takes the value 1 in years in which states adopted fiscal rules or 0 otherwise, or when states began the publication of debt or guarantee reports. State and year fixed effects are further included in the regression. As such, this specification can be interpreted as a difference-in-difference design for a linear probability model, in which the coefficients of interest ( β1 to β4) estimate the change in probability of a contingent liability shock occurring following the intervention, controlling for state and year specific developments. Standard errors are clustered at the state level in parentheses. SCS = special category state. a. This analysis focuses on the publication of specialized debt and guarantee reports. The independent variable thus takes the value 0 for instances when debt- and guarantee- related information was only published in annual financial statements. *** p < 0.01, ** p < 0.05, * p < 0.1. 166   H IDDEN DEBT Notes 14. This limit varies for each province according to its share of the national population. For 1. The most recent joint World Bank–IMF debt Punjab and Khyber Pakhtunkhwa, for sustainability analyses for Afghanistan (IMF instance, borrowing limits are equal to PRs 2020a), Maldives (IMF 2020c), Pakistan 143 billion and PRs 44 billion, respectively. (IMF 2020b), and Sri Lanka (IMF 2019a) 15. In Khyber Pakhtunkhwa and Balochistan as of assess these countries as being at high risk of the end of June 2019, all of the debt is on-lent debt distress. by the central government. This arrangement 2. Brazil’s federal government bailed out exists in part because these state governments ­subnational governments in 1989, 1993, and do not have a debt management strategy and 1997–2000 (Manoel, Garson, and Mora 2013). hence cannot borrow on their own. 3. Data are for the end of 2018 (RBI 2019a). 16. According to the latest Punjab debt bulletin, 4. China’s subnational debt-to-GDP is similarly 90 percent of the net year-on-year growth in high (estimated at 20.6 percent), but it has a outstanding debt stock was due to rupee unitary system. depreciation in the previous six months. 5. Data are for the end of June 2019 and are esti- 17. The Pakistani fiscal year runs from July 1 to mated by World Bank staff for this study. For June 30. more details, see the next section on Pakistan. 18. There are also discrepancies between debt 6. OECD-UCLG (2019, figure 7.3). Data are stock recorded by the Ministry of Finance and for 2016. the provincial financial statements in some 7. See Maldives (n.d.). cases. In Balochistan, for example, the latter 8. See https://presidency.gov.mv/Press/Article​ understates the debt stock quite substantially. /22833. 19. This finding is based on discussions with staff 9. Eleven states have been designated special at the Office of the Auditor General of category states because of their unique cir- Pakistan. cumstances, such economic and infrastruc- 20. The government of Khyber Pakhtunkhwa, tural backwardness, and nonviable nature of for example, owns 70 percent of the Bank of state finances. The remaining 17 states are Khyber and would likely intervene in the case general category states. of a crisis. 10. Pakistan’s fiscal rule can also be suspended if 21. This is a lower-bound estimate given that social and poverty-reducing expenditures fall actual growth in salaries, pensions, and num- below 4.5 percent of GDP or if health and ber of employees has been 3 percentage points education spending fail to double in terms of to 4 percentage points higher than the assumed percent of GDP over a 10-year period. annual increases in the actuarial valuations. 11. This estimate uses data from the latest available See Government of Punjab (2019, 55–56). debt bulletin of each province except for 22. World Bank (2017b). Balochistan, for which data come from 23. See Government of Khyber Pakhtunkhwa Government of Balochistan (2019). (2019). 12. Article 167, Clause (3) states, “A Province 24. In Sindh, for example, the government may not, without the consent of the Federal ­provided a sovereign guarantee of $700 mil- Government, raise any loan if there is still lion to the Thar coal power plant, which was outstanding any part of a loan made to the inaugurated in April 2019. Province by the Federal Government, or in 25. S e e  h t t p s : / / t r i b u n e . c o m . p k / s t o r y​ respect of which guarantee has been given by /2101536/2-pll-sngpl-tussle-threatens-derail​ the Federal Government; and consent under -1200mw​-project/. this clause may be granted subject to such 26. See also Government of India (2017). conditions, if any, as the Federal Government 27. Guarantee reports induce a similar effect may think fit to impose.” (see table 4C.4, col. 3). 13. Article 167, Clause (4) states, “A Province may 28. See Government of India (2018). raise domestic or international loan or give 29. This contrasts with today’s practice in guarantees on the security of the Provincial Pakistani provinces. See, for example, World Consolidated Fund within such limits and sub- Bank (2017a). ject to such conditions as may be specified by 30. See Baicker (2005); Lutz (2010); Litschig and the National Economic Council.” Morrison (2013); Cascio, Gordon, and Reber S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   167 (2013); Lundqvist (2015); Liu and Ma European Fiscal Union: Lessons from the (2016). Experience of Fiscal Federations, edited by C. 31. The one exception is Pakistan, where the Cottarelli and M. Guerguil. London: Routledge. division of resources is such that fiscal space d e G r o o t , O . , F. H o l m - H a d u l l a , a n d lies with the provinces and not with the cen- N. ­Leiner-Killinger. 2015. “Cost of Borrowing tral government. Shocks and Fiscal Adjustment.” Journal of 32. One interesting international example comes International Money and Finance 59 (C): from Spain, where in 2012 the ­central 23–48. ­government provided loans and guarantees De Jong, P. 1991. “The Diffuse Kalman Filter.” to multiple regions to service their debt Annals of Statistics 19 (2): 1073–83. ­during the Eurozone crisis, thus preempting a De Jong, P., and S. Chu-Chun-Lin. 1994. further deterioration of the fiscal crisis while “Stationary and Non-Stationary State Space weakening fiscal incentives going forward. Models.” Journal of Time Series Analysis 15 (2): 33. For an alternative decomposition of the debt 151–66. trajectory, see Giavazzi and Missale (2004). Durbin, J., and S. J. Koopman. 2001. Time Series Analysis by State-Space Methods . Oxford, United Kingdom: Oxford University Press. References Ellis, P., and M. Roberts. 2016. Leveraging Baicker, K. 2005. “Extensive or Intensive Urbanization in South Asia: Managing Spatial Generosity? The Price and Income Effects of Transformation for Prosperity and Livability. Federal Grants.” Review of Economics and Washington, DC: World Bank. Statistics 87 (2): 371–84. Giavazzi, F., and A. Missale. 2004. “Public Debt Bayoumi, T., M. Goldstein, and G. Woglom. 1995. Management in Brazil.” NBER Working Paper “Do Credit Markets Discipline Sovereign 10394, National Bureau of Economic Research, Borrowers? Evidence from US States.” Journal Cambridge, MA. of Money, Credit and Banking 27 (4): Governatori, M., and D. Yim. 2012. “Fiscal 1046–59. Decentralization and Fiscal Outcomes.” Bernoth, K., and G. B. Wolff. 2008. “Fool the Economic Papers 468, European Commission, Markets? Creative Accounting, Fiscal Brussels. Transparency and Sovereign Risk Premia.” Government of Balochistan. 2019. Balochistan Scottish Journal of Political Economy 55 (4): Budget White Paper 2019–2020. Government 465–87. of Balochistan, Pakistan. Blum, F., and P. S. Yoong. 2020. “The Impact of Government of India, Ministry of Finance, Subnational Contingent Liability Realizations: Department of Economic Affairs, Economic Evidence from India.” Background Division. 2017. “Economic Survey 2016–17.” paper for  Hidden Debt . World Bank, Government of India. Washington, DC. Government of India, Ministry of Finance, Booth, L., G. Georgopoulos, and W. Hejazi. 2007. Department of Economic Affairs, Economic “What Drives Provincial-Canada Yield Division. 2018. Economic Survey, 2017–18. Spreads?” Canadian Journal of Economics 40 Government of India. (3): 1008–32. Government of Khyber Pakhtunkhwa. 2019. “One Bova, E., M. Ruiz-Arranz, F. Toscani, and H. Elif Khyber Pakhtunkhwa.” White Paper 2019/20 Ture. 2016. “The Fiscal Costs of Contingent Budget, June. Government of Khyber Liabilities: A New Dataset.” IMF Working Pakhtunkhwa, Pakistan. Paper WP/16/14, International Monetary Fund, Government of Maldives. n.d. Strategic Action Washington, DC. Plan 2019–2023. President’s Office, Republic of Cascio, E. U., N. Gordon, and S. Reber. 2013. Maldives. “Local Responses to Federal Grants: Evidence Government of Punjab. 2019. White Paper on the from the Introduction of Title I in the South.” Budget FY 2019–20. Government of Punjab, American Economic Journal: Economic Policy India, June. 5 (3): 126–59. Heppke-Falk, K. H., and G. B. Wolff. 2008. Cordes, T., M. Guerguil, L. Jaramillo, M. Moreno- “Moral Hazard and Bail-Out in Fiscal Badia, and S. Ylaoutinen. 2014. “Subnational Federations: Evidence for the German Länder.” Fiscal Crises.” Chapter 6 in Designing a Kyklos 61 (3): 425–46. 168   H IDDEN DEBT IMF (International Monetary Fund). 2014. “Public Keen, M., and M. Marchand. 1997. “Fiscal Expenditure Reform: Making Difficult Competition and the Pattern of Public Choices.” Chapter 2 in Fiscal Monitor April Spending.” Journal of Public Economics 66 (1): 2014. Washington, DC: IMF. 33–53. IMF (International Monetary Fund). 2018. “Fiscal Kelly, R., and A. S. Gunawardena. 2016. Discipline in Indian States: Market-Based and “Financing Sustainable Urbanization in Sri Other Options.” Selected Issues, IMF Country Lanka.” Chapter 7 in The Sri Lankan Report no. 18/255, International Monetary Economy: Charting A New Course, edited by Fund, Washington, DC, August. P. Athukorala, E. Ginting, H. Hill, and IMF (International Monetary Fund). 2019a. U. Kumar. Manila: Asian Development Bank. “Sixth Review under the Extended Arrangement Kornai, J. 1986. “The Soft Budget Constraint.” under the Extended Fund Facility and Requests Kyklos 39 (1): 3–30. for Waiver of Nonobservance and Modification Litschig, S., and K. M. Morrison. 2013. “The of Performance Criterion–Press Release; Staff Impact of Intergovernmental Transfers on Report; and Statement by the Alternate Education Outcomes and Poverty Reduction.” Executive Director for Sri Lanka.” International American Economic Journal: Applied Monetary Fund, Washington, DC. Economics 5 (4): 206–40. IMF (International Monetary Fund). 2019b. Liu, C., and G. Ma. 2016. “Taxation without “Maldives: Technical Assistance Report–Public Representation: Local Fiscal Response to Investment Management Assessment.” Country Intergovernmental Transfers in China.” Report no. 19/102, International Monetary International Tax and Public Finance 23 (5): Fund, Washington, DC. 854–74. IMF (International Monetary Fund). 2019c. Lledó, V., S. Yoon, X. Fang, S. Mbaye, and Y. Kim. “Selected Issues: Nepal.” CR/Issues/2019/02/15​ 2017. “Fiscal Rules at a Glance.” Background / N e p a l - 2 0 1 8 - A r t i c l e - I V- C o n s u l t a t i o n . Note, International Monetary Fund, International Monetary Fund, Washington, DC. Washington, DC. IMF (International Monetary Fund). 2019d. Lundqvist, H. 2015. “Granting Public or Private “Estimating the Stock of Public Capital in 170 Consumption? Effects of Grants on Local Countries. August 2019 Update.” International Public Spending and Income Taxes.” Monetary Fund, Washington, DC. International Tax and Public Finance 22 (1): IMF (International Monetary Fund). 2020a. 41–72. “Islamic Republic of Afghanistan: Request for Lutz, B. 2010. “Taxation with Representation: Disbursement under the Rapid Credit Facility– Intergovernmental Grants in a Plebiscite Press Release; Staff Report; and Statement by Democracy.” Review of Economics and the Executive Director for the Islamic Republic Statistics 92 (2): 316–32. of Afghanistan.” International Monetary Fund, Manoel, A., S. Garson, and M. Mora. 2013. Washington, DC, April 17. “Brazil: The Subnational Debt Restructuring IMF (International Monetary Fund). 2020b. of the 1990s–Origins, Conditions and “Pakistan: Request for Purchase under the Results.” In Until Debt Do Us Part: Rapid Financing Instrument–Press Release; Subnational Debt, Insolvency, and Markets, Staff Report; and Statement by the Executive edited by O. Canuto and L. Liu, 33–79. Director for Pakistan.” International Monetary Washington, DC: World Bank. Fund, Washington, DC, April 17. Manoel, A., S. Refaqat, H. Onder, and M. Ashraf. IMF (International Monetary Fund). 2020c. 2012. “Punjab Debt Sustainability Analysis.” “Maldives: Request for Disbursement under the World Bank Policy Paper Series on Pakistan Rapid Credit Facility–Press Release; Staff 13/12, World Bank, Washington, DC, October. Report; and Statement by the Executive Oates, W. E. 1972. Fiscal Federalism. New York: Director for Maldives.” International Monetary Harcourt Brace Jovanovich. Fund, Washington, DC, April 23. Oates, W. E. 1999. “An Essay on Fiscal Jenker, E., and Z. Lu. 2014. “Sub-National Federalism.” Journal of Economic Literature Credit Risk and Sovereign Bailouts–Who 37 (2): 1120–49. Pays the Premium?” IMF Working Paper Oates, W. E. 2005. “Toward A Second Generation WP/14/20, International Monetary Fund, Theory of Fiscal Federalism.” International Tax Washington, DC. and Public Finance 12 (4): 349–73. S u b n a t i o n a l G o v e r n m e n t s i n S o u t h A s i a   169 OECD (Organisation for Economic Co-operation RBI (Reserve Bank of India). 2019b. “Handbook and Development). 2018. Subnational of Statistics on Indian States.” Reserve Bank of Governments in OECD Countries: Key  Data, India. 2018 edition. https://www.oecd.org/regional​ Saggar, S., and M. Adki. 2017. “State Government /Subnational-governments-in-OECD-Countries​ Yield Spreads–Do Fiscal Metrics Matter?” -Key-Data-2018.pdf Mint Street Memo no. 08, Reserve Bank of OECD-UCLG (Organisation for Economic India. Co-operation and Development–United Cities Schuknecht, L., J. Von Hagen, and G. Wolswijk. and Local Governments). 2019. “2019 Report 2009. “Government Risk Premiums in the of the World Observatory on Subnational Bond Market: EMU and Canada.” European Government Finance and Investment—Key Journal of Political Economy 25 (3): 371–84. Findings.” OECD-UCLG. Seabright, P. 1996. “Accountability and Persson, T., G. Roland, and G. Tabellini. 2000. Decentralisation in Government: An Incomplete “Comparative Politics and Public Finance.” Contracts Model.” European Economic Journal of Political Economy 108: 1121–61. Review 40 (1): 61–89. Polackova Brixi, H., and A. Mody. 2002. “Dealing Simone, A., and P. Topalova. 2009. “India’s with Fiscal Risk: An Overview.” Chapter 1 in Experience with Fiscal Rules: An Evaluation Government at Risk: Contingent Liabilities and and the Way Forward.” IMF Working Paper Fiscal Risk, edited by H. Polackova Brixi and 09/175, International Monetary Fund, A. Schick, 21–58. Washington, DC: World Washington, DC. Bank; New York: Oxford University Press. Sow, M., and I. Razafimahefa. 2017. “Fiscal Polackova Brixi, H., and A. Schick, eds. 2002. Decentralization and Fiscal Policy Performance.” Government at Risk: Contingent Liabilities and IMF Working Paper, 17/64, International Fiscal Risk. Washington, DC: World Bank; Monetary Fund, Washington, DC. New York: Oxford University Press. Weber, A. 2012. “Stock-Flow Adjustments and PRS India. 2013. “Special Category Status and Fiscal Transparency: A Cross-Country Centre-State Finances.” PRS India, New Delhi. Comparison.” IMF Working Paper 12/39, Rangarajan, C., and A. Prasad. 2012. “Managing International Monetary Fund, Washington, State Debt and Ensuring Solvency: The Indian DC. Experience.” Policy Research Working Paper World Bank. 2017a. “Khyber Pakhtunkhwa 6039, World Bank, Washington, DC. Province, Pakistan: Public Financial Refaqat, S. 2015. “Fiscal Decentralization in Management Assessment Report.” World Bank, Pakistan: A Forgotten Agenda.” The World Washington, DC. Bank Policy Paper Series on Pakistan. World World Bank. 2017b. “Sindh Public Expenditure Bank, Islamabad, June. Review.” World Bank, Washington, DC. RBI (Reserve Bank of India). 2015. “Handbook of World Bank. 2019a. South Asia Economic Focus: Statistics on Indian States.” Reserve Bank of Rethinking Decentralization. Washington, DC: India. World Bank. RBI (Reserve Bank of India). 2019a. “State World Bank. 2019b. “Local Government Public Finances: A Study of Budgets.” Reserve Bank of Financial Management Assessment for India. Bangladesh.” World Bank, Dhaka. ECO-AUDIT Environmental Benefits Statement The World Bank Group is committed to reducing its environmental footprint. In support of this commitment, we leverage electronic publishing options and print-on-demand technology, which is located in regional hubs worldwide. Together, these initiatives enable print runs to be lowered and shipping distances decreased, resulting in reduced paper consumption, chemical use, greenhouse gas emissions, and waste. Our books are printed on Forest Stewardship Council (FSC)–certified paper, with a minimum of 10 percent recycled content. The fiber in our book paper is either unbleached or bleached using totally chlorine-free (TCF), processed chlorine–free (PCF), or enhanced elemental chlorine–free (EECF) processes. More information about the Bank’s environmental philosophy can be found at http://www.worldbank.org/corporateresponsibility. South Asia Development Matters Hidden Debt Solutions to Avert the Next Financial Crisis in South Asia The COVID-19 crisis, which has sent economies in South Asia and around the world into a deep recession, has highlighted South Asia’s rising debt levels and sizable hidden liabilities. State-owned enterprises, state-owned commercial banks, and public-private partnerships have been at the center of the rising debt wave and the latest pandemic response. Historically, South Asia has relied on these direct public interventions more than other regions. The interventions have helped governments tackle key development challenges and rapidly deliver relief measures during crises. However, because of their inefficiencies and weak governance, the interventions are also a significant source of public indebtedness and macrofinancial risks. Hidden Debt examines the trade-off between tackling development challenges through direct state presence in the market and avoiding unsustainable debt due to economic inefficiencies of such off–balance sheet operations, which greatly leverage public capital. The study recommends a reform agenda based on the four interrelated principles of purpose, incentives, transparency, and accountability (PITA). The reforms can mitigate the risks that off–balance sheet operations will become the source of the next financial crisis in South Asia. www.worldbank.org/southasiahiddendebt ISBN 978-1-4648-1667-3 SKU 211667