DIREC TIONS IN DE VELOPMENT Agriculture and Rural Development Risk Modeling for Appraising Named Peril Index Insurance Products A Guide for Practitioners Shadreck Mapfumo, Huybert Groenendaal, and Chloe Dugger Risk Modeling for Appraising Named Peril Index Insurance Products Direc tions in De velopment Agriculture and Rural Development Risk Modeling for Appraising Named Peril Index Insurance Products A Guide for Practitioners Shadreck Mapfumo, Huybert Groenendaal, and Chloe Dugger © 2017 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 20 19 18 17 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 of the data included in this work. 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 considered to be a limitation upon or waiver of the privileges and immunities 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: Mapfumo, Shadreck, Huybert Groenendaal, and Chloe Dugger. 2017. Risk Modeling for Appraising Named Peril Index Insurance Products: A Guide for Practitioners. Directions in Development. Washington, DC: World Bank. doi:10.1596/978-1-4648-1048-0. 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 reuse a component of the work, it is your responsibility to determine whether permission is needed for that reuse 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): 978-1-4648-1048-0 ISBN (electronic): 978-1-4648-1049-7 DOI: 10.1596/978-1-4648-1048-0 Cover photo: Adzope, a village in the eastern part of Côte d’Ivoire. © Ami Vitale / World Bank. Further permission required for reuse. Cover design: Debra Naylor, Naylor Design, Inc. Library of Congress Cataloging-in-Publication Data has been requested. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Contents Foreword xvii Acknowledgments xix About the Authors xxi Endorsements xxiii Abbreviations xxv Chapter 1 Introduction 1 1.1 Guide Overview 3 1.2 The Case Example 4 Bibliography 5 Chapter 2 Critical Concepts in Named Peril Index Insurance 7 2.1 Why Is Insurance Useful for Smallholder Farmers? 7 2.2 What Is Named Peril Index Insurance? 9 2.3 Who Are the Main Stakeholders in the Risk Transfer Process? 11 2.4 How Are Named Peril Index Insurance Products Developed? 15 Note 18 Bibliography 18 PART 1 Decision Tools for Insurance Managers 19 Chapter 3 Prefeasibility Study 21 3.1 Introduction 21 3.2 Outline of Emerging Managerial and Process Controls 22 Note 28 Bibliography 28 Chapter 4 Product Design and Evaluation—The Base Index 29 4.1 Introduction 29 4.2 Basis Risk and the Implied Deductible 30 4.3 Steps in Product Design and Evaluation 34 Bibliography 47 Risk Modeling for Appraising Named Peril Index Insurance Products   v   http://dx.doi.org/10.1596/978-1-4648-1048-0 vi Contents Chapter 5 Product Pricing—The Base Index 49 5.1 Introduction 49 5.2 Outline of Emerging Managerial and Process Controls 50 Bibliography 63 Chapter 6 Product Evaluation—The Redesigned Index 65 6.1 Introduction 65 6.2 Outline of Emerging Managerial and Process Controls 68 6.3 Step 1: Determine Key Model Inputs and Assumptions 68 6.4 Step 2: Evaluate Key Managerial Decision Metrics 70 6.5 Step 3: Document and Communicate the Product Options and Business Decision 72 Bibliography 73 Chapter 7 Detailed Market Analysis 75 7.1 Introduction 75 7.2 Outline of Emerging Managerial and Process Controls 76 Note 82 Bibliography 82 Chapter 8 Value of Index Insurance to a Financier 83 8.1 Introduction 83 8.2 Outline of Emerging Managerial and Process Controls 85 Part 1 Conclusion 90 Notes 90 Bibliography 91 PART 2 Probabilistic Modeling for Insurance Analysts 93 Chapter 9 How to Use Part 2 95 9.1 Introduction 95 9.2 Website 96 9.3 Overview of Key Assumptions and Limitations of This Guide’s Models 96 9.4 Monte Carlo Software Tools 97 Note 98 Bibliography 98 Chapter 10 Fundamentals of Probabilistic Modeling 99 10.1 The Case for Probabilistic Modeling in Index Insurance 100 10.2 Key Building Blocks for Probabilistic Modeling 107 10.3 Key Outputs for Probabilistic Modeling 130 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Contents vii 10.4 A Reminder on How to Use the Models in This Guide 134 Notes 135 Bibliography 137 Chapter 11 Evaluating the Base Index 139 11.1 Background and Objectives 139 11.2 Model Inputs 140 11.3 Model Computations 145 11.4 Model Outputs 165 11.5 Alternative Modeling Approach: Retrospective Analysis 167 Bibliography 170 Chapter 12 Pricing the Base Index 171 12.1 Background and Objectives 171 12.2 Model Inputs 171 12.3 Model Computations 175 12.4 Model Outputs 208 Note 209 Bibliography 210 Chapter 13 Evaluating the Redesigned Index 211 13.1 Background and Objectives 211 13.2 Model Inputs 212 13.3 Model Computations 216 13.4 Model Outputs 228 13.5 Alternative Modeling Approach: Retrospective Analysis 229 Bibliography 229 Chapter 14 Detailed Market Analysis 231 14.1 Background and Objectives 231 14.2 Model Inputs 231 14.3  Model Computations 235 14.4  Model Outputs 247 Bibliography 248 Chapter 15 Value of Index Insurance 251 15.1  Background and Objectives 251 15.2  Model Inputs 252 15.3  Model Computations 254 15.4  Model Outputs 263 Bibliography 264 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 viii Contents Chapter 16 Alternative Probabilistic Modeling Approaches 267 16.1  Overview of Alternative Approaches 267 16.2  Approach 1: Simulating the Payouts Directly 268 16.3  Approach 2: Simulating the Index 270 16.4  Approach 3: Simulating the Weather 271 16.5  Which Model to Use? 271 Notes 272 Bibliography 272 Chapter 17 Conclusion 273 Glossary 275 Boxes 1.1 Promoting High Standards of Professional Behavior 3 3.1 Changes in Risk Conditions over Time 23 10.1 A Brief History of Monte Carlo Simulation 105 10.2 When to Use Monte Carlo Simulation 106 11.1 Overview of Calculations for the Base Index Product Evaluation Metrics 166 12.1 Summary of Key Formulas Used in the Chapter 198 12.2 Overview of Calculations for the Equitable Premiums Metrics 207 13.1 Overview of Calculations for the Redesigned Index Product Evaluation Metrics 228 14.1 Overview of Calculations for the Detailed Market Analysis Metrics 247 15.1 Overview of Calculations for the Value of Index Insurance Metrics 263 Case Example Boxes 1CB.1 Excellence Insurance Background 5 2CB.1 Smallholder Agriculture and Household Finance in Mapfumoland 8 2CB.2 Insured Units and Proxies for Mass Bank Product 10 2CB.3 Excellence Insurance Staffing and Resources 15 2CB.4 Base Index and Redesigned Index Triggers 16 2CB.5 Specific Years Comparison for Base Index and Redesigned Index 17 3CB.1 Excellence Insurance’s Prefeasibility Study for Mapfumoland 22 3CB.2 Summary of Key Points from the Research Plus Prefeasibility Study on the Mapfumoland Market 25 3CB.3 Excellence Insurance Technical Evaluation of Prefeasibility Study 27 4CB.1 Overview of the Base Index for Mass Bank 30 4CB.2 Categorical Classification of Past Damages for Areas A to J, 1984–2013 35 4CB.3 Subject Specialist Information for Base Index Product Design 36 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Contents ix 4CB.4 Term Sheet for Base Index 38 4CB.5 Calculation of Historical Index Values for the Base Index 40 4CB.6 Calculation of Historical Payouts for the Base Index 41 4CB.7 Historical Payouts for the Base Index 41 4CB.8 Excellence Insurance Product Evaluation Guidelines for the Base Index 43 4CB.9 Product Evaluation Decision Metrics for the Base Index 44 5CB.1 Mass Bank Loan Portfolio 50 5CB.2 Pricing Model Inputs for the Base Index—No Reinsurance 51 5CB.3 Excellence Insurance Risk Management Committee Guidelines for Index Product Pricing 54 5CB.4 Product Model Outputs for Base Index—No Reinsurance 55 5CB.5 Pricing Decisions for Base Index—No Reinsurance 55 5CB.6 Pricing Model Inputs for Base Index—Proportional Reinsurance Only 56 5CB.7 Pricing Model Outputs for Base Index—Proportional Reinsurance Only 57 5CB.8 Pricing Decisions for Base Index—Proportional Reinsurance Only 57 5CB.9 Pricing Model Inputs for Base Index—Proportional and Nonproportional Reinsurance 59 5CB.10 Pricing Model Outputs for Base Index—Proportional and Nonproportional Reinsurance 60 5CB.11 Pricing Decisions for Mass Bank Base Index—Proportional and Nonproportional Reinsurance 61 5CB.12 Equitable Premiums for the Base Index 62 6CB.1 Term Sheet for Redesigned Index 65 6CB.2 Historical Payouts for Redesigned Index 67 6CB.3 Product Performance Model Inputs for the Redesigned Index 69 6CB.4 Product Performance Model Outputs for the Redesigned Index 71 6CB.5 Outcome of Mass Bank Negotiations 72 7CB.1 The Effect of Client Liquidity on Excellence Insurance’s Premium Volumes 76 7CB.2 Market Analysis Model Inputs for Prototypes 77 7CB.3 Excellence Insurance Risk Management Committee Guidelines for Market Segments 80 7CB.4 Market Analysis Outputs for Mapfumoland Market Segments 80 7CB.5 Managerial and Actuarial Market Analysis Decisions for the Mapfumoland Market Segments—Redesigned Prototypes 81 8CB.1 Approaching Buyer Goods 84 8CB.2 Value of Index Insurance Model Inputs for Buyer Goods 86 8CB.3 Buyer Goods Guidelines for Value of Index Insurance 88 8CB.4 Value of Index Insurance Model Outputs for Buyer Goods 89 8CB.5 Managerial and Actuarial Value of Index Insurance Decisions for Buyer Goods—Redesigned Prototype 1 90 11CB.1 Inputs—Step 1 141 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 x Contents 11CB.2 Inputs—Step 2 141 11CB.3 Inputs—Step 3 141 11CB.4 Inputs—Step 4 142 11CB.5 Inputs—Step 5 143 11CB.6 Inputs—Step 6 144 11CB.7 Inputs—Step 7 145 11CB.8 Computations—Step 8 147 11CB.9 Computations—Step 10 148 11CB.10 Computations—Step 11 149 11CB.11 Computations—Step 13 150 11CB.12 Computations—Step 14 152 11CB.13 Computations—Step 15 153 11CB.14 Computations—Steps 16–17 153 11CB.15 Computations—Step 18 154 11CB.16 Determining Area Level Scenario Frequency and Severity Values 155 11CB.17 Computations—Steps 19–23 156 11CB.18 Computations—Steps 24–29 157 11CB.19 Computations—Steps 30–35 158 11CB.20 Computations—Steps 36–37 160 11CB.21 Computations—Step 38 161 11CB.22 Computations—Steps 39 and 40 164 11CB.23 Computations—Step 41 164 11CB.24 Computations—Steps 42–44 165 11CB.25 Outputs 167 11CB.26 Review of Base Index Performance for Historical Events with Greater than 50 Percent Damage Level 168 11CB.27 Review of Base Index Payouts of at Least 30 Percent 168 11CB.28 Calculation of Risk Metrics 169 12CB.1 Inputs—Step 1 172 12CB.2 Inputs—Step 2 173 12CB.3 Inputs—Step 3 174 12CB.4 Inputs—Step 4 174 12CB.5 Inputs—Step 6 175 12CB.6 Inputs—Step 7 175 12CB.7 Computations—Steps 8–12 177 12CB.8 Computations—Steps 13 and 14 179 12CB.9 Computations—Step 15 181 12CB.10 Computations—Step 16 184 12CB.11 Computations—Step 17 186 12CB.12 Computations—Step 18 189 12CB.13 Computations—Step 19 192 12CB.14 Computations—Step 20 192 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Contents xi 12CB.15 Computations—Step 21 193 12CB.16 Computations—Steps 25 and 26 196 12CB.17 Computations—Step 27 197 12CB.18 Computations—Step 31 201 12CB.19 Computations—Steps 32–34 201 12CB.20 Computations—Step 35 204 12CB.21 Computations—Steps 36 and 37 206 12CB.22 Computations—Step 38 207 12CB.23 Outputs 209 13CB.1 Inputs—Step 1 213 13CB.2 Inputs—Step 2 213 13CB.3 Inputs—Step 3 213 13CB.4 Inputs—Step 4 214 13CB.5 Inputs—Step 5 214 13CB.6 Inputs—Step 6 215 13CB.7 Computations—Step 7 217 13CB.8 Computations—Step 9 218 13CB.9 Computations—Steps 10–14 219 13CB.10 Computations—Steps 15–19 220 13CB.11 Computations—Steps 20–25 222 13CB.12 Computations—Steps 26 and 27 223 13CB.13 Computations—Step 28 224 13CB.14 Computations—Steps 29 and 30 226 13CB.15 Computations—Step 31 227 13CB.16 Outputs 228 14CB.1 Inputs—Step 1 232 14CB.2 Inputs—Step 2 234 14CB.3 Computations—Step 3 236 14CB.4 Computations—Step 4 238 14CB.5 Computations—Step 5 241 14CB.6 Computations—Step 6 244 14CB.7 Computations—Step 7 246 14CB.8 Outputs 248 15CB.1 Inputs—Steps 1–5 253 15CB.2 Inputs—Step 6 253 15CB.3 Computations—Step 8 255 15CB.4 Computations—Steps 10–16 256 15CB.5 Computations—Steps 17–23 257 15CB.6 Computations—Steps 24–26 260 15CB.7 Computations—Steps 27–29 261 15CB.8 Computations—Step 30 262 15CB.9 Evaluating the Relevance of Insurance to a Specific Financier 262 15CB.10 Outputs 264 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 xii Contents Figures 2.1 Individual as Policyholder and Insured Party 12 2.2 Aggregator as Policyholder and Insured Party 12 2.3 Aggregator as Policyholder (Agent) on Behalf of the Insured Party 12 2.4 Product Design and Data Processing Internal to Insurer 13 2.5 Product Design and Data Processing Provided by One External Firm 13 2.6 Product Design and Data Processing Provided by Two Separate External Firms 14 2.7 Product Design and Data Processing Provided by External Firm, with Actuarial Analyst Internal to Insurer 14 4.1 Insurer and Insured Party Basis Risk 30 4.2 Redesigned Index Implied Deductible 32 4.3 Product Design Basis Risk versus the Redesigned Index Implied Deductible 32 4.4 Identifying Product Design Basis Risk versus the Implied Deductible 33 4.5 Base Index Design and Evaluation Process 35 4CB.3.1 Base Index Trigger and Exit for Trigger 2 38 5.1 Portfolio Product Pricing Managerial Decision Process—No Reinsurance 51 5.2 Portfolio Product Pricing Managerial Decision Process—Proportional Reinsurance Only 56 5.3 Portfolio Product Pricing Managerial Decision Process— Proportional and Nonproportional Reinsurance 58 5CB.9.1 Base Index Proportional and Nonproportional Reinsurance 59 5.4 Equitable Premium Pricing 61 6.1 Redesigned Index Evaluation Managerial Decision Process 69 7.1 Detailed Market Analysis Managerial Decision Process 77 8.1 Value of Index Insurance Managerial Decision Process 85 10.1 Skewing and Bounding in Distributions 108 10.2 Probability Density Chart of a Discrete Probability Distribution 109 10.3 Probability Density Chart of a Continuous Probability Distribution 110 10.4 Cumulative Probability Chart of a Discrete Probability Distribution 112 10.5 Cumulative Probability Chart of a Continuous Probability Distribution 113 10.6 Cumulative Probability Chart of a Continuous Distribution with the P10 Displayed 113 10.7 Probability Density Chart of Continuous Distribution with the P10 Displayed 114 10.8 Probability Density Charts of Two Beta Distributions 118 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Contents xiii 10.9 PERT (30, 50, 100) 120 10.10 Frequency Graph of Historical Payout Amounts, Together with the Maximum Likelihood Estimation of the Fitted Beta Distribution 122 10.11 Frequency Graph of Historical Payout Amounts, Together with the Maximum Likelihood Estimates of the Fitted Beta Distribution and Alternative Fits Based on Parameter Uncertainty 124 10.12 Parameter Uncertainty for Alpha and Beta 124 10.13 Tight vs. Broad Parameter Uncertainty 125 10.14 Generating a Random Value from a Gamma (5, 1) Distribution 128 10.15 Histogram Plot of Next Year’s Total Payout Amount 130 11.1 Generating Scenario Payout Ratios 151 11.2 Generating Expected Return Periods for Inventory Damage and the Base Index 159 11.3 Generating Probability of Having No Insured Party Basis Risk Event and Expected Insured Party Basis Risk Amount 162 11.4 Generating Historical Years with the Largest Insured Party Basis Risk Ratios 163 12.1 Generating Expected Losses and Required Capital 178 12.2 Generating Expected Combined Ratios and Profit Margins 180 12.3 Generating Probability of Fund Ruin 183 12.4 Generating Potential Economic Value Added 186 12.5 Generating Sharpe Ratios 188 12.6 Generating Product Pricing Decision Metrics (Proportional Reinsurance Only) 191 12.7 Generating Product Pricing Decision Metrics (Proportional and Nonproportional Reinsurance) 194 12.8 Generating APPIU and RCPIU for Each Area 200 12.9 Generating EROC for Each Area 203 12.10 Generating Equitable Premium Rates for Each Area 205 13.1 Generating Return Periods for the Base and Redesigned Indexes 223 13.2 Generating Probability of No Implied Deductible Event and Expected Implied Deductible Amounts 225 13.3 Generating Historical Years with Largest Implied Deductible Amount 226 14.1 Generating Expected Premium Incomes 235 14.2 Generating Expected Losses and Required Capital 237 14.3 Generating Expected Combined Ratios and Profit Margins 240 14.4 Generating Projected Values for Economic Value Added (EVA) Metrics 243 14.5 Generating Sharpe Ratios 245 15.1 Generating the Value of Index Insurance Decision Metrics 259 16.1 Overview of Three Approaches to Simulating Payout Ratios 268 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 xiv Contents Tables 2.1 Key Differences between Indemnity Insurance and Index-Based Agricultural Insurance 9 4CB.9.1 Base Index Product Evaluation Summary 45 5.1 Template for Risk Management Committee Guidelines on Index Product Pricing 53 5CB.9.1 Reinsurance Terms 60 7.1 Template for Risk Management Committee Guidelines on Market Segments 79 8.1 Template for Client Guidelines for Value of Index Insurance and Net NPL Decision Metrics 88 11.1 Summary of Model Components for Evaluating the Base Index 140 11.2 Model Inputs 140 11.3 Model Computations 146 11.4 Model Computations 146 11.5 Model Computations 149 11.6 Model Computations 151 11.7 Model Computations 155 11.8 Model Computations 156 11.9 Model Computations 158 11.10 Model Computations 159 11.11 Model Outputs 166 11.12 Retrospective Classification Matrix for the Base Index 169 12.1 Summary of Model Components for Pricing the Base Index 172 12.2 Model Inputs 172 12.3 Model Computations 176 12.4 Model Computations 176 12.5 Model Computations 177 12.6 Model Computations 190 12.7 Model Computations 193 12.8 Model Computations 199 12.9 Model Outputs 209 13.1 Summary of Model Components for Evaluating the Redesigned Index 212 13.2 Model Inputs 212 13.3 Model Computations 216 13.4 Model Computations 217 13.5 Model Computations 219 13.6 Model Computations 220 13.7 Model Computations 221 13.8 Model Computations 222 14.1 Summary of Model Components for the Detailed Market Analysis 232 14.2 Model Inputs 232 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Contents xv 14.3 Model Computations 235 14.4 Model Outputs 248 15.1 Summary of Model Components for the Value of Index Insurance 252 15.2 Model Inputs 252 15.3 Model Computations 254 15.4 Model Computations 254 15.5 Model Computations 256 15.6 Model Computations 257 15.7 Model Computations 258 15.8 Model Outputs 263 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Foreword This guide was written to introduce a wider audience of insurers to index insurance as a risk management tool for agriculture. Index insurance, which is a ­ relatively recent innovation, has exciting potential for addressing the need for agricultural insurance in developing economies. Even more significant, index insurance is a tool for achieving broader financial inclusion and for increasing investment in smart agricultural technologies in the regions of the world that need it most. Index insurance thus has the potential to contribute to increased agricultural sustainability and improved food security. Our experience working on index insurance in more than 20 developing countries has demonstrated its value as a tool for reaching a huge but largely unserved market segment: small, semi-commercial farmers and the array of ser- vice providers that they use. Small farmers can be an attractive customer base for more and more insurers—especially those in markets whose traditional life, prop- erty, and vehicle market shares are spoken for. Developing business lines involving index insurance is not without challenges. For a successful index insurance market to develop, important prerequisites— such as the availability of historical data, product design capabilities, and distribu- tion channels—must be in place. To safeguard the smooth development of index markets, insurers must rigorously evaluate the quality of the products they offer and must take extra care to ensure that distributors and policyholders fully understand the benefits and the limits of the purchased coverage. Without these extra steps to ensure responsible insurance practices, insurers can damage the implementation and potential of index insurance in the market. In an increasingly competitive insurance market, creative product develop- ment and imaginative business strategies are becoming the norm. This guide will help emerging market insurers who seek to stay on the cutting edge to success- fully penetrate new market segments. Ceyla Pazarbasioglu Senior Director Finance and Markets Global Practice The World Bank Group Risk Modeling for Appraising Named Peril Index Insurance Products   xvii   http://dx.doi.org/10.1596/978-1-4648-1048-0 Acknowledgments The authors would like to thank numerous colleagues and peers who have provided invaluable contributions to this guide. The authors note that any errors, ­ either of omission or commission, in this guide are entirely their own. The views presented herein represent those of the authors and are not an official position of the World Bank Group. We thank Utako Saoshiro of the World Bank Group for her input, ideas, and assistance on the content, text, and diagrams throughout this guide, and Dr. Kurt Rinehart of EpiX Analytics for his valuable input to chapter 10. We also thank Dr. Yong Bum Cho, Professor Gary Venter, and Dr. Fan Yang of the AIG Model Validation Group for their review of the models and text in this guide. We would like to acknowledge Dr. Erwann Michel Kerjan of the Wharton School of the University of Pennsylvania for his in-depth feedback on the drafts. We also acknowledge Dr. Francisco Zagmutt of EpiX Analytics for his critical review of chapter 10. Over the past 12 years, the authors benefited from interactions with the ­following people: Carlos Arce, Mae Joy Armada, Erin Bryla, Martin Reto Buehler, Eric Chapola, Daniel Clarke, Julie Dana, Conrad De Jesus, Christopher Ereso, Gilles Galludec, Hanna Joy Germinal, Rose Goslinga, Ulrich Hess, Joseph Kakweza, Vijay Kalavakonda, Harini Kannan, Juliet Kyokunda, Geric Laude, Gift Livata, Peter Maina, William Martirez, Michael Mbaka, Agrotosh Mookerjee, Ronald Ngwira, Fredrick Odhiambo, Sharon Onyango, Gary Reusche, Rhoda Rubaiza, James Sharpe, Andrea Stoppa, Charles Stutley, Joanna Syroka, Daniele Torriani, Christina Ulardic, Anita Untario, Panos Varangis, Duncan Warren, Simon Young, and Andiry Zaripov. Many thanks go to David Crush, Alejandro Alvarez de la Campa, and Samuel Munzele Maimbo of the World Bank Group for their support of this project and to Lucille Gavera, Bonny Jennings, and Anna Koblanck for making this guide a user-friendly resource. We would also like to acknowledge our other colleagues, clients, and peers at the World Bank Group and EpiX Analytics for the many useful ideas and con- structive conversations. Finally, we greatly thank our families for their support, understanding, and encouragement during the process of writing and finalizing this guide. Risk Modeling for Appraising Named Peril Index Insurance Products   xix   http://dx.doi.org/10.1596/978-1-4648-1048-0 About the Authors Shadreck Mapfumo is senior financial sector specialist in the Finance and Markets Global Practice of the World Bank Group. He provides agricultural and index-based insurance advisory services to insurance companies and intermediar- ies in Africa, the Caribbean, and Asia. Shadreck has a master’s degree in actuarial studies from the Australian National University, he is a member of the Institute of Actuaries of Australia, and he holds the Associate in Reinsurance and Associate of the Chartered Insurance Institute (U.K.) designations. He has worked in index insurance for more than 12 years. Huybert Groenendaal is managing partner of EpiX Analytics LLC, a consulting firm specializing in probabilistic modeling and decision modeling. Huybert leads consulting engagements worldwide in a broad range of fields and industries, from pharmaceuticals and oil and gas to manufacturing, insurance, health, and epide- miology. Huybert holds a PhD from Wageningen University (the Netherlands) and an MBA in finance from the Wharton School of Business, University of Pennsylvania. Huybert is an affiliate faculty member at Colorado State University. Chloe Dugger is a partner market executive at Vitality Group, the international expansion arm of the South African insurer Discovery Limited. Prior to this position, Chloe was an operations officer in the Finance and Markets Global ­ Practice of the World Bank Group, where she managed advisory projects for companies looking to scale up agricultural and index-based insurance across Africa. Chloe has a master’s degree in development studies from the University of Oxford. Risk Modeling for Appraising Named Peril Index Insurance Products   xxi   http://dx.doi.org/10.1596/978-1-4648-1048-0 Endorsements “This guide, written by leaders in the field, is poised to become a go-to reference for anyone interested in index insurance. It is packed with useful information for both practitioners and regulators as well as individuals who simply want to learn more about this complex subject. As such, the Microinsurance Working Group of the International Actuarial Association recommends this guide as a very timely contribution to inclusive insurance literature.” —Nigel Bowman, Chairman, Microinsurance Working Group, International Actuarial Association “It is with great interest that I have read Risk Modeling for Appraising Named Peril Index Insurance Products. The authors provide a very useful guide for practitio- ners and academics alike. They show the great potential of index insurance and describe the critical success factors without ignoring trust-related and technologi- cal hurdles or market dynamics. This makes the guide an indispensable read for anybody interested in this domain, and I cannot applaud enough the World Bank Group for sponsoring this important ­ initiative.” —Eckart Roth, Chief Risk Officer, Peak Reinsurance Company Limited, Hong Kong SAR, China “This book is an important step forward to harness the power of insurance to create value and bring hope to underinsured societies around the world, an important step toward creating more vibrant and sustainable economies in these critical and developing areas. This book provides a comprehensive and practical guide to microinsurance and its pricing implementation. It will be a business enabler within the insurance industry and development and governmental ­ sectors, and a model for entrepreneurs in adjacent and unrelated areas to bet- ter understand economic opportunity outside the already industrialized and consumer-based economies.” —Sean C. Keenan, Senior Managing Director of ­ Model Risk Management, AIG, United States “An instant classic! This guide will become the reference point against which the development and quality of index insurance will be measured, and the founda- tion for many further efforts and publications to come. An extremely useful go-to aid for general insurance staff, be they managers, sales force, outreach officers, or those with underwriting or claims handling functions, as well as ­ Risk Modeling for Appraising Named Peril Index Insurance Products   xxiii   http://dx.doi.org/10.1596/978-1-4648-1048-0 xxiv Endorsements the actuarial, product design, and development specialists. All will find the necessary information and tools to venture into the exciting new world of index ­ ­insurance.” —​Martin Buehler, Principal Insurance Officer, International Finance Corporation, United States “The importance of microinsurance for small entrepreneurs and their finan- cial partners is gaining recognition. This guide gives a comprehensive introduc- tion to the risk management and quantitative analysis an institution would want to undertake to offer this insurance in a sound, sustainable manner.” —Gary Venter, President of the Gary Venter Company and Actuary in Residence at Columbia University, United States “This guide will help insurance managers and actuaries navigate the challenging yet exciting journey of index insurance development. With the steps and consid- erations clearly laid out for interested stakeholders, this material should help accelerate the evolution of index insurance.” —​ Geric Laude, President and Chief Executive Officer, CARD Pioneer Microinsurance National Capital Region, the Philippines “This guide provides a necessary reference document on index insurance. It uniquely blends the qualitative and quantitative aspects of the subject, making it suitable for specialists and newcomers. It makes extensive use of case studies, ensuring engagement with the reader.” —Corneille Karekezi, Group Managing Director and Chief Executive Officer, African Reinsurance Company, Nigeria “This comprehensive guide promises to be an invaluable guide for actuaries, actuarial analysts, and insurance managers faced with the task of modeling index insurance products. Packed with useful tools, step-by-step guidance, and accessible explanations of applied statistics, this guide is a timely contribution to not only the world of impact insurance but also the financial inclusion agenda, supporting the goals of agricultural sustainability and food security.” — Lisa Morgan, Technical Officer, International Labour Organization’s Impact Insurance Facility, Switzerland “This is an important book written by experts who know what they’re talking about. The book is clear, educational, and consistent. The authors show how to implement index insurance step by step in a very transparent way. Without a doubt, it is the most prominent book I have come across in this area. From that perspective, the book fills a big vacuum.” —Auguste Mpacko Priso, Head of Microinsurance Working Group, Institute of Actuaries of France, France Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Abbreviations ACRE Agriculture and Climate Risk Enterprise APPIU average payout per insured unit ARC2 African Rainfall Climatology, version 2 (NOAA) AUS average unit size CIRAD Centre de coopération internationale en recherche agronomique pour le développement (French agricultural research and inter- national cooperation organization) CVaR conditional value at risk EROC expected return on capital EVA economic value added FEWS NET Famine Early Warning Systems Network (USAID) GOF goodness of fit MFI microfinance institution MLE maximum likelihood estimation NGO nongovernmental organization NPL nonperforming loan NPR nonproportional reinsurance PERT project evaluation and review techniques PR proportional reinsurance RCPIU required capital per insured unit SAR special administrative region TVaR tail value at risk VaR value at risk A full glossary of terms follows at the end of the book. Terms included in the glossary appear in bold type upon first use in the book. Risk Modeling for Appraising Named Peril Index Insurance Products   xxv   http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 1 Introduction The main audience for this guide is the managers and actuarial analysts of insurance companies in developing countries interested in developing and ­ ­evaluating named peril index insurance product lines. However, a number of other stakeholders could also find this guide useful, including • Farmer organizations, financial institutions, and agriculture value chain actors and investors evaluating the potential benefits and risks of index insurance policies. ­ onsumer • Insurance regulators assessing insurance products for client value and c protection purposes. • Students interested in quantitative risk analysis and probabilistic modeling. Named peril index insurance is a financial instrument for transferring risk from individuals or groups to international risk carriers. Index insurance products trig- ger compensation to the insured party based on the deviations of a proxy such as rainfall, temperature, or humidity that is highly correlated with a named peril such as drought or excess rain. In turn, these named perils correlate with financial losses for the insured party, for example, decreased yield for a crop affected by drought, or the death of livestock. As a relatively new instrument in developing countries, named peril index insurance products are often designed by specialized product design teams external to the insurer that underwrites the risk. Some examples of firms that provide these product design services include ACRE Africa, Columbia University’s International Research Institute for Climate and Society, MicroEnsure, PlaNet Guarantee, EARS Earth Environment Monitoring BV, and CIRAD (Centre de coopération internationale en recherche agronomique). Index insurance product design is a very technical and labor-intensive, and thus expensive, process. Premium volumes generated from these products are often still too low to support a complete product design team at an insurer; Risk Modeling for Appraising Named Peril Index Insurance Products   1   http://dx.doi.org/10.1596/978-1-4648-1048-0 2 Introduction however, many external product design firms have accessed donor resources to partly fund their costs. In many cases, insurers in developing countries are minimally involved in the setting of contract triggers, or the product review and refinement process. Concrete metrics and statistics that explain the product design team’s reasons for recommending a specific product structure can help insurance managers make sound business decisions regarding what products to offer. For example, insurers are sometimes asked to change a product’s premium rate and coverage level to meet policyholder price expectations. Clear informa- tion on the implications of such a change on the insurer’s profit objectives and risk tolerance will ensure that the manager makes an informed business decision. Without clear tools for evaluating these changes in product structure and price, insurance managers risk engaging in blind underwriting that goes against their business objectives. However, with deeper involvement in product design and evaluation, insurers can develop the best contract wording for their market, wording that clearly explains issues like basis risk and implied deductibles to policyhold- ers. They can also apply innovations and experiences from other classes of business in their market, contributing insights that product design specialists ­ may lack. The main objectives of this guide are to • Promote informed business decision making among insurance companies by providing them with effective tools for evaluating named peril index insurance business opportunities and products; • Support the improvement of named peril index insurance product offerings through structured and transparent collaboration and communication between insurers, product design teams, and policyholders; • Encourage more insurance companies to write index-based insurance policies that protect against key risks and improve access to finance among the unbanked and underbanked market in developing countries; • Improve the technical capacity of insurance companies in quantitative risk analysis of index insurance products and named peril index insurance pricing analytics; and • Encourage practices in the index insurance industry that are in the best ­ interests of various stakeholders (insured parties, policyholders, insurers, reinsurers, and regulators) and build confidence in the products offered (see box 1.1). Index insurance is commonly perceived to be complicated and difficult to evalu- ate. This is one reason index insurance products have not yet achieved high penetration in developing countries, despite their clear potential to improve the risk management options for vulnerable populations. This guide attempts to close the knowledge gap to grow this important market and provide protection to more low-income customers. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Introduction 3 Box 1.1 Promoting High Standards of Professional Behavior The guide encourages practices in the index insurance industry that are in the best interests of various stakeholders (insured parties, policyholders, insurers, reinsurers, and regulators) and build confidence in the products offered. As such, it supports actuaries and other professionals involved in designing and pricing index-based insurance products in following the principles outlined in the Actuaries’ Code of the Institute and Faculty of Actuaries, United Kingdom. The main principles of the code that the guide promotes are as follow: Principle 1: Integrity—Members will act honestly and with the highest standards of integrity. Section 1.3 Members will be honest and truthful in promoting their business services. The guide provides tools to support practitioners in being clear and honest about the workings, accuracy, and value of products developed, as well as about the implications of changing various parameters within the products offered to fit the needs of different stakeholders. Principle 2: Competence and Care—Members will perform their professional duties competently and with care. Section 2.2 Members will not act unless they have an appropriate level of relevant ­knowledge and skill. Section 2.7 Members will keep their competence up to date. The guide encourages practitioners to continue identifying and developing the best techniques to apply in the course of their work. Principle 5: Communication—Members will communicate effectively and meet all applicable reporting standards. ­ Section 5.1 Members will ensure that their communication, whether written or oral, is clear and timely, and that their method of communication is appropriate. Section 5.3 Members will take such steps as are sufficient and available to them to ensure that any communication with which they are associated is accurate and not misleading, and contains sufficient information to enable its subject matter to be put in proper context. A central goal of the guide is to promote clear communication of the features of ­ products developed to facilitate informed decision making by new and existing buyers of index-based insurance products, insurers, reinsurers, regulators charged with approving new products, and other affected stakeholders. 1.1  Guide Overview Part 1 of this guide provides a summary of the insights and decisions required for the insurer to make an informed decision to launch and expand an index insurance business line. Part 1 explains each key decision the insurance manager makes at each step in the product design, evaluation, and pricing Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 4 Introduction process, and the information the insurance manager and actuarial analyst need from the product design team and other sources to make these decisions. Insurance managers are the primary audience for part 1. Part 2 of this guide provides a step-by-step guide to calculating the d ­ ecision metrics used by the insurance manager in part 1. These metrics are calculated using probabilistic modeling that provides insights into risks related to the index insurance product. Probabilistic models generate thousands of ­ possible future scenarios based on historical risk patterns, potential changes in those risk patterns over time, other relevant information, and uncertainty m ­ easures. The models use these thousands of scenarios to provide an understanding of spe- cific elements of an index insurance product that are important for the insur- ance manager’s decision making. Actuarial analysts are the primary ­ audience for part 2. This book complements the work on product reliability outlined in Morsink, Clarke, and Mapfumo (2016), which looks at the same issues of product quality but from a client value perspective. Mathematically, the approaches that they present, and the ones that we explain and promote in this book, end up being very similar. It is also important to point out what this guide does not cover. • First, the key objective of this guide is to provide a framework for approaching the assessment of index insurance. The book does not cover the comparison of index insurance with other potential risk mitigation products or the assess- ment of if and when index insurance is appropriate for a given situation. While some of the methods and analyses presented in this book can be useful in help- ing address these issues, the scope of this guide is limited to the appraisal of the index insurance product. • Second, the book mainly focuses on the assessment of retail index insurance products and does not show examples of how to appraise sovereign index insurance products. The overall approaches and metrics discussed in this guide can, however, also be applied to sovereign index insurance. • Third, the guide does not include assessment examples with indices that are based on area average loss or damage. Instead it focuses on indirect indices such as temperature and precipitation. Nevertheless, the main principles discussed in the guide do apply to other types of indices. ­ 1.2 The Case Example Throughout this guide, we will refer to a concrete example of a product design and evaluation process—our case example (case example box 1CB.1). This example uses hypothetical data that are based on our experience developing index insurance products in more than 20 developing countries. Wherever a reader sees a box labeled “Case Example,” he or she will find new information on this hypothetical example. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Introduction 5 Case Example Box 1CB.1 Excellence Insurance Background Excellence Insurance was founded in 2008 in Mapfumoland. Mapfumoland is a lower-­ middle-income country with a population of 50 million people. Since its launch, Excellence has earned a reputation for innovation and a focus on bottom-of-the-pyramid, low-income consumers. The Mapfumoland market overall has low insurance penetration and a large unbanked and uninsured population. As part of its strategy to expand its customer base into the low-income population, Excellence is considering launching a named peril index insurance product line. Excellence has already been approached by Mass Bank, a commer- cial bank with a significant portfolio of loans to smallholder farmers, about developing an index product to protect its portfolio from defaults after drought. Bibliography Banks, E., ed. 2002. Weather Risk Management: Markets, Products, and Applications. Basingstoke: Palgrave. Dick, W., A. Stoppa, J. Anderson, E. Coleman, and F. Rispoli. 2011. Weather Index–Based Insurance in Agricultural Development: A Technical Guide. Rome: IFAD. Morsink, K., D. Clarke, and S. Mapfumo. 2016. “How to Measure Whether Index Insurance Provides Reliable Protection.” Policy Research Working Paper 7744, World Bank, Washington, DC. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 2 Critical Concepts in Named Peril Index Insurance 2.1  Why Is Insurance Useful for Smallholder Farmers?1 The purpose of insurance is to transfer a specific type of risk from an individual or a group to a third party capable of handling the financial impact of the loss. Most risks can be classified as either high frequency/low impact risks, or low frequency/high impact risks. High frequency/low impact risks have a short return period and are usually retained by the party concerned and managed through risk mitigation strategies, such as regular doctor’s visits or the use of smoke detectors in the home. However, low frequency/high impact events, such as death, a major medical emergency, or the destruction of valuable assets, can require the transfer of such risks to a well-capitalized third party that can absorb part or all of the financial impact. Insurance is one of the most common tools for transferring this type of risk. Insurers are able to take on this risk because they pool a large number of dif- ferent risks, thereby diversifying and reducing their overall exposure. This risk pooling is most effective when the insured risks are relatively independent, which means the risk events will not all occur at the same time. For example, when one health insurance policyholder undergoes an expensive procedure to address chronic heart disease, other policyholders will not all require the same procedure at the same time. Another reason that insurers can take on risk is that they buy reinsurance for some of this risk. Reinsurance is a form of insurance for insurance companies that transfers a portion of the exposure to reinsurance providers. It is important to note that insurance transfers only the monetary value of the residual risk that is not managed by the insured party’s implementation of necessary risk mitigation measures. Farmers that use good farming practices undertake risk mitigation strategies such as the use of drought-resistant seeds Risk Modeling for Appraising Named Peril Index Insurance Products   7   http://dx.doi.org/10.1596/978-1-4648-1048-0 8 Critical Concepts in Named Peril Index Insurance Case Example Box 2CB.1 Smallholder Agriculture and Household Finance in Mapfumoland Rose Jituboh is a Mapfumoland farmer who lives in Bwanje, an area prone to short dry spells; she farms half a hectare of land. Agricultural production is her household’s main source of income, but she and her family members also take on odd jobs at construction sites and other farms in her local area. The family also receives remittances from an older daughter in the nearest city. Like her neighbors, in the past Rose has used saved seeds to plant her maize crop each year. In the past decade, though, she and her neighbors have lost large parts of their har- vests when dry spells hit during the germination or flowering phases of the crop cycle. Rose remembers 2007 and 2010 as particularly bad years for dry spells. This year, Rose applied for a loan from Mass Bank for $80 to buy drought-resistant seeds and fertilizer. With these improved inputs, she will be less likely to lose her harvest if a dry spell occurs this season. and appropriate fertilizer against high frequency/low impact risk events like dry spells. These risk mitigation measures are part of why a high frequency event—like a dry spell in a specific location—can have a relatively low impact (case example box 2CB.1). Insurance products transfer the residual risk of low frequency/high impact risk events to insurance companies. The insured party regularly pays a small amount for protection against the devastating effects of a rare but very severe event, such as a major drought or earthquake, against which it is very difficult to implement successful risk mitigation measures. Although smallholder farmers can benefit from risk transfer through insur- ance products, the specific type of risks they face makes index insurance a prom- ising tool for this population. Index insurance solves a major problem for insurers wishing to cover low frequency/high impact events that affect many insured parties at the same time, often in logistically challenging situations and typically with relatively low insured amounts per insured party. Many risks, such as fire, accident, or death, affect insured parties indepen- dently. In these cases, insurance companies find it operationally and financially feasible to visit each affected party and assess his or her level of damage to determine the claim payment. This type of insurance is called indemnity insurance. For example, when a car covered by car insurance is in an accident, the insurance company sends an adjuster to evaluate the damage to the vehicle. Based on the adjuster’s evaluation, the insurance company will make a claim payment for the estimated cost to repair or replace the car. This evalu- ation of actual losses is expensive and can often take significant time and resources to complete. Table 2.1 summarizes the main differences between indemnity and index- based agriculture insurance from the perspective of the insurer. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Critical Concepts in Named Peril Index Insurance 9 Table 2.1  Key Differences between Indemnity Insurance and Index-Based Agricultural Insurance Indemnity insurance products (multi- peril crop insurance) Named peril index insurance products Coverage • Most perils that affect agriculture • Only perils specified in the contract production (for example, hail or drought) except for exclusions specified in the contract Underwriting and product design • Historical inventory damage data for • Historical hazard data (for example, requirements the individual farmer or for a time series for meteorological data) population representative of the • Historical inventory damage data farmer’s experience • Agronomic data • Farmer location • Location of the measurement point (for • Farmer acreage example, weather station or satellite pixel) Underwriting and product design • High because of requirement for • High because of technical capacity costs farmer-level yield data needed Target market • Large and medium commercial • Governments farmers • Smallholder farmers • Agribusinesses • Input suppliers • Financial institutions • Nongovernmental organizations Contract monitoring activities • Yield measured at the end of the • Real-time hazard data used to monitor season the contract throughout the season Loss assessment • Completed for each farmer • No field assessments • Semi-objective process • Transparent and objective evaluation using real-time hazard data Risk of adverse selection • High • Low Risk of moral hazard • High • Low Basis risk • Low • Moderate to high 2.2  What Is Named Peril Index Insurance? When risks affect a large population all at the same time—called covariant risks—and often in difficult on-the-ground circumstances, assessing the losses of each individual insured party that is affected is not feasible. For example, a major typhoon might affect tens of thousands of insured parties at the same time. The insurer will not have the resources to assess each claim individually in a short period even in the best conditions. Damage to infrastructure caused by the typhoon will make such assessments even less feasible. In the case of small rural farmers in developing countries, a loss event such as a major drought will similarly affect large numbers of farmers. Smallholder land- holdings of fewer than two hectares would require an extremely high number of assessments, even over a relatively small area or region. Furthermore, each farmer will insure a relatively small value—for example $100—making the potential revenue per insured unit very small. These factors make indemnity insurance for smallholder farmers operationally and financially unattractive for the insurer. For covariant risks in logistically challenging environments, index insurance offers Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 10 Critical Concepts in Named Peril Index Insurance an efficient mechanism for providing coverage without relying on individual assessments for claim processing. An important assumption of index insurance products is that insured units within a given geographical area have similar characteristics, and the effect of the deviation in the proxy is similar for all insured units. When a claim is triggered for a specific area, all insured units are compensated at the same payout rate, usually a percentage of the sum insured. Individual payouts are calculated automatically based on deviations in a proxy, such as the cumulative amount of rainfall during a specific period, or the wind speed of a typhoon. Neither the number of individuals affected nor the on-the- ground conditions affect the claim process. It is important to note that there will be situations in which an insured party experiences a loss attributable to a hazard event but does not receive a payout. The index product will only pay out for hazard events that are specifically cov- ered by the policy—those for which the proxy(ies) meet the specified triggers. An important element of index insurance product structuring is when the proxy triggers a payout. In some cases, proxies can trigger a payout for asset pro- tection rather than for replacement. For example, in Kenya index-based livestock insurance products trigger a payout to pastoralists for the purchase of animal feed when pasture levels begin to decrease because of drought. An important feature is that the payout comes as the pasture is decreasing, not when it has disappeared. This way, the pastoralists can purchase feed to keep their animals alive rather than using a payout to replace animals that have died. Named peril index insurance is a relatively new financial instrument for trans- ferring risk from individuals or groups to international risk carriers. Although the instrument has been used for many years in developed countries such as Belgium, France, Germany, Italy, Switzerland, the United Kingdom, and the United States, its use in the developing world is fairly recent. See case example box 2CB.2. Case Example Box 2CB.2 Insured Units and Proxies for Mass Bank Product Mass Bank provides loans of between $75 and $160 to farmers like Rose Jituboh for the purchase of improved inputs. Most farmers use the loans to purchase drought-resistant seeds and fertilizer. Like Rose, Mass Bank’s customers live within 20 kilometers of 10 ground- based weather stations—the geographical areas for the index insurance product. Excellence Insurance is developing an index product for Mass Bank that covers maize crops against both dry spells like the ones experienced recently in Rose’s area and more serious droughts. The loans from Mass Bank are the insured units. The product uses two proxies: cumulative rainfall in millimeters during the flowering period and the number of consecutive dry days during the entire cover period. Based on the deviations in these prox- ies, the product makes payouts of between 0 percent and 100 percent of the sum insured (the value of the loan). For Rose’s $80 loan, the 100 percent payout will equal $80. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Critical Concepts in Named Peril Index Insurance 11 2.3  Who Are the Main Stakeholders in the Risk Transfer Process? The key stakeholders in the risk transfer process are the regulator, the insured party, the policyholder, the insurer, the product design team, the data provider, and the reinsurer, each of whom are defined below. Regulator—The regulator approves the issuing of the product in the market and also determines and implements consumer protection rules. Insured party—The insured party is the individual or firm that transfers away the unwanted residual risk. The insured party can be an individual farmer or a small or medium enterprise, or it can be the same organization that is the policyholder. Policyholder (the client)—As a market segment, smallholder farmers consti- tute a large number of insured units, each with small insured values, which makes issuing policies to each smallholder farmer operationally and financially unattractive for insurers. Working with a single policyholder organization, that is, an aggregator, such as an input supplier, a microfinance institution, a coop- erative, or a commercial bank, provides insurers with a less expensive way of ­ reaching smallholder farmers. A single policy is issued to the policyholder that covers all the insured units. In some cases, the aggregator will be both the poli- cyholder and the insured party, such as when a commercial bank insures its own portfolio of loans. In other cases, the aggregator is the policyholder, but the policy specifies that the insured parties are the individual smallholder farmers. In the latter case, the aggregator is acting as an agent of the insurer and is there- fore remunerated through an agreed-on commission structure. Insurer—The insurer underwrites the risk. The insurer is the party legally responsible for the liabilities arising from the policy. The insurer issues the policy, collects premiums, reinsures part of the portfolio, and settles claims arising from the policy. If the product does not perform as expected by the end users, the finan- cial and reputational risks fall on the insurer. For this reason it is critical that the insurer fully understand the features of each named peril product it underwrites. See figures 2.1–2.3 for the potential configurations of policyholders and insured parties. Product design team—The product design team possesses specialized skills in developing named peril index products. Often, this team is part of an insurance intermediary, but it can also be made up of members of the insurer’s internal staff. In the interest of the long-term sustainability of an index product line, we recommend that insurers and other key stakeholders work toward developing this product design capacity locally. If this capacity is not initially available locally, insurers can hire international resources to design index products and build local capacity. Remaining wholly dependent on international resources for product design services can be difficult and often financially imprudent. In the long run, local resources tend to produce better-designed products because of their under- standing of the local environment. For agriculture index insurance products in particular, local agronomists play a crucial role in designing a Base Index that takes into account all the critical aspects of the local crop and soil characteristics. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 12 Critical Concepts in Named Peril Index Insurance Figure 2.1 Individual as Policyholder and Insured Party Regulator Pays out in event Pays out in event of covered loss of covered loss Reinsurer Insurer Individual as policyholder Pays premium Pays premium and insured for risk transfer for risk transfer party Figure 2.2  Aggregator as Policyholder and Insured Party Pays out in event Pays out in event of covered loss of covered loss Reinsurer Insurer Aggregator as Pays premium Pays premium policyholder for risk transfer for risk transfer and insured party Figure 2.3  Aggregator as Policyholder (Agent) on Behalf of the Insured Party Pays agent commission and pays out in event Pays out in event of covered loss to of covered loss insured party Reinsurer Insurer Aggregator as Pays premium Sends premiums policyholder for risk transfer from insured party (agent) for risk transfer Pays premium, Transfers payout usually bundled with to insured party loan or inputs from received from aggregator insurer for covered loss Insured party (farmers, borrowers) Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Critical Concepts in Named Peril Index Insurance 13 Data processing team—Index insurance products require real-time hazard data for claim processing. These data can come from publicly or privately owned weather stations, remote sensing equipment, or satellites, and often must be pro- cessed and converted into a suitable format for analysis by the insurer. Many firms that provide product design services have also developed capabilities in processing data for insurance purposes. Data provider—Depending on the country, data providers are public agencies, private firms, or a combination of the two. Data are collected through ground- based or satellite instruments. The data provider supplies the historical data needed for product design and pricing, and real-time data for claim settlement. See figures 2.4–2.7 for various configurations of product design and data pro- cessing and provision. Figure 2.4 Product Design and Data Processing Internal to Insurer Insurance manager Obtain and pay for data from data providers Insurer Data providers Actuarial analyst (public, as part of product private) design team Product Data design processing team team Figure 2.5 Product Design and Data Processing Provided by One External Firm Insurance Pays intermediary with manager both product design and data processing capabilities Insurer Intermediary Actuarial analyst as part of product design team Product Data design processing team team Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 14 Critical Concepts in Named Peril Index Insurance Figure 2.6 Product Design and Data Processing Provided by Two Separate External Firms Pays intermediary 1 for product design Intermediary 1 Actuarial analyst Insurance as part of product manager design team Insurer Product design team Intermediary 2 Pays intermediary 2 for data processing Data processing team Figure 2.7 Product Design and Data Processing Provided by External Firm, with Actuarial Analyst Internal to Insurer Insurance Pays intermediary with manager both product design and data processing capabilities Insurer Intermediary Actuarial analyst as part of insurer Product Data design processing team team Reinsurer—The reinsurer, the insurer of insurers, accepts all or a portion of the risk underwritten by the insurer. Because of the covariant nature of the risks insured using index insurance, a significant portion of the risk should be rein- sured on the international market. Reinsurance protects the solvency of the local insurance company and also provides a foreign exchange inflow when a major loss event occurs, which can benefit both the insurer and the national economy. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Critical Concepts in Named Peril Index Insurance 15 Throughout this guide, most of the attention is placed on the interactions between two key roles: the insurance manager and the actuarial analyst. The insurance manager is the staff member of the insurer charged with decision making regarding the insurer’s index insurance product line. The actuarial analyst uses analytics and risk modeling to provide the insurance manager with metrics for evaluating index insurance business opportunities and prod- ucts. The actuarial analyst can be a part of an internal or external product design team, or a member of the insurer’s staff who is charged with analyzing information provided by an external product design team. See case example box 2CB.3. Case Example Box 2CB.3 Excellence Insurance Staffing and Resources At Excellence Insurance, management has assigned a promising actuary on staff, Lindiwe Maneli, to serve as the actuarial analyst for the index insurance product line, and an experi- enced executive, Ghassimu Sow, to the role of insurance manager. Excellence is consider- ing hiring the Mapfumoland specialist insurance intermediary firm Hazard Analytics to provide the product design and data processing services. 2.4 How Are Named Peril Index Insurance Products Developed? Index insurance contracts are designed using historical hazard and inventory damage data to trigger payouts at specific frequency and severity levels. The index insurance product design process typically occurs in two phases. In the first phase, the product design team develops a product based mainly on input from local subject specialists (for example, agronomists), and evaluates (chapter 4) and prices (chapter 5) this initial product. This guide calls this product the Base Index. The Base Index is designed with the goal of providing maximum transfer of the risk of the named peril. It provides the highest level of coverage possible against damage to the farmer’s inventory. The Base Index triggers a payment when the proxy’s behavior indicates that any damage to inventory—no matter how small—is expected. A major challenge for Base Index design is basis risk—the difference between the payout triggered by the index insurance product and the actual losses expe- rienced by the insured party that are attributable to the named peril. Insured party basis risk describes the scenario in which the payout amount is less than the farmer’s actual losses attributable to the named peril. In this case, the farmer experiences an economic loss from the named peril but is not adequately com- pensated by the claim payout. Insurer basis risk describes the scenario in which the payout is greater than the actual losses the insured party experiences from the named peril. In this case, the insurer suffers an economic loss because of unnecessary claim payments. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 16 Critical Concepts in Named Peril Index Insurance Two types of basis risk cause these outcomes for the insured party and the insurer: product design basis risk and geographical basis risk. An example of product design basis risk is provided here. Imagine a very simple index insurance policy that stipulates that a payout will occur when less than 100 millimeters of rain falls during the entire growing season. Imagine that 105 millimeters of rain fell during the period, but all in the last week of the season. In this example, farmers will likely experience losses because of the very dry season overall, but the policy will not pay out. Product design basis risk is discussed, evaluated, and quantified in detail in chapter 4 of this guide. An example of the other type of basis risk, geographical basis risk, occurs when a farmer’s field is so far away from the location where the proxy is mea- sured that the conditions in her field do not match the proxy measurement. This type of basis risk is not discussed or quantified in this guide. With the advent of satellite products, insurers can reduce geographical basis risk by using multiple, precise measurement locations. Because it provides such a high level of coverage, the Base Index is also very expensive, and many policyholders will request a lower price—and lower coverage—product. However, it is extremely important that the insurer always ­ produce a Base Index to explain to the policyholder the difference between complete coverage—that provided by the Base Index—and the coverage pro- vided by other product options. Without this explicit comparison, policyholders often fall into the trap of expecting complete coverage even when they have purchased a lower coverage, less expensive product. In some cases, the policyholder will purchase the Base Index. More often, however, the Base Index will cost more than the policyholder is initially willing or able to pay. This is when the second phase of the product design begins. The product design team must now use input from the policyholder on price to rede- sign and improve the product with new parameters so that the cost of the prod- uct decreases (case example box 2CB.4). This second product is called the Redesigned Index. The product design team also evaluates the Redesigned Index (chapter 6), just like the Base Index. The trigger levels for the Redesigned Index proxies embody a specific implied deductible, which is the difference in coverage between the Base Index and the Redesigned Index. The deductible is the amount of residual risk that is carried by Case Example Box 2CB.4  Base Index and Redesigned Index Triggers Excellence Insurance’s Base Index for Mass Bank is designed to pay out in seasons with less than 100 millimeters of cumulative rainfall (the trigger). The Redesigned Index for Mass Bank is designed to pay out in seasons that have less than 75 millimeters of cumulative rainfall. The Redesigned Index is cheaper than the Base Index, but in any season during which rainfall measures between 75 millimeters and 100 millimeters, the Redesigned Index will not pay out. If Mass Bank purchases the Redesigned Index, the bank retains more drought risk—the implied deductible. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Critical Concepts in Named Peril Index Insurance 17 the policyholder and not transferred to the insurance company by virtue of pur- chasing the Redesigned instead of the Base Index. The less residual risk that is transferred to the insurance company, the lower the cost of the index insurance product and the higher the deductible. Products with lower deductibles will be more expensive. Policyholder understanding of the levels of triggers and the amount of the implied deductible is paramount to successful implementation of index insur- ance. If policyholders do not understand these factors, they will have incorrect expectations of when the product will pay out. For example, a policyholder with an incomplete understanding of a product designed to cover catastrophic drought may expect to receive a payout after a short dry spell. One way in which the insurance provider can ensure policyholder under- standing of the index product is by explaining the product’s behavior with refer- ence to previous experience in the insured area, as demonstrated in case example box 2CB.5. Because named peril index insurance is relatively new, many people believe that an index that does not trigger when losses are experienced on the ground is always caused by product design basis risk. In many cases, however, the Base Index—which would have paid out for most losses—was too expen- sive and the policyholder selected the Redesigned Index and so is responsible for the implied deductible. It is critical to understand and distinguish between these two situations in which a low (or no) payout occurs despite significant loss of inventory. The tools in this guide provide and explain quantitative and probabilistic tools and techniques that insurance managers can use to evaluate and communi- cate the characteristics, future behavior, and value of index insurance products. The processes suggested for the insurer’s review of these products are critical strategies for practicing responsible finance and for the long-term sustainability Case Example Box 2CB.5 Specific Years Comparison for Base Index and Redesigned Index The Excellence Insurance manager, Ghassimu Sow, uses examples from 2010 and 2007 to explain the difference in coverage for the Base Index and the Redesigned Index to Mass Bank. Farmers in Rose Jituboh’s and other areas remember these years as having very bad dry spells that affected their crops and harvests. Ghassimu explains that in 2010, the Base Index would have paid out 61 percent of the sum insured for Area H. For a farmer with a loan of $80, the payout would have been $49. The Redesigned Index, however, would have paid out only 11 percent—$9 for a $80 loan. In 2007 in Area D, the Base Index would have paid out 40 percent of the sum insured ($24), while the Redesigned Index would not have paid out at all. These concrete numbers help the Mass Bank managers understand that these are very different products. If they select the less expensive Redesigned Index, they will receive less coverage against drought and dry spells. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 18 Critical Concepts in Named Peril Index Insurance of index insurance markets. Specifically, well-informed and educated providers, buyers, and users of index insurance will help further develop and sustain index insurance markets. This guide provides methods for meeting consumer protection responsibilities such as providing transparent services and treating policyholders fairly. Failure to implement responsible insurance principles will lead to reputational challenges for the product, the insurer, and the market as whole, which in turn will lead to low product sales. Although the framework and tools presented in this guide do allow for a much better understanding of index insurance, it is important to note that the concepts are very difficult. A thorough assessment of a product does allow for a clear explanation of the product’s characteristics, but many insured parties will still find it difficult to fully understand the product. This guide does not cover how best to communicate index insurance concepts to insured parties or confirm their understanding. In many cases, the distribution channels and aggregators, who normally act as policyholders, will have a critical role in ensuring the insured party’s understanding of the product characteristics. Note 1. Named peril index insurance can also be useful for other stakeholders, such as micro, small, and medium enterprises engaged in nonfarming activities that are nonetheless exposed to weather risks. For example, microentrepreneurs in some coastal areas are vulnerable to typhoons or hurricanes that can damage or destroy their inventory. This guide focuses on index insurance for farming-related activities, but it is important to remember that the tools discussed here can also be applied to other types of insured parties. Bibliography Banks, E., ed. 2002. Weather Risk Management: Markets, Products, and Applications. Basingstoke: Palgrave. Dick, W., A. Stoppa, J. Anderson, E. Coleman, and F. Rispoli. 2011. Weather Index-Based Insurance in Agricultural Development: A Technical Guide. Rome: IFAD. Mahul, O., V. Niraj, and D. Clarke. 2012. “Improving Farmers’ Access to Agricultural Insurance in India.” Policy Research Working Paper 5987, World Bank, Washington, DC. Morsink, K., D. Clarke, and S. Mapfumo. 2016. “How to Measure Whether Index Insurance Provides Reliable Protection.” Policy Research Working Paper 7744, World Bank, Washington, DC. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 P A RT 1 Decision Tools for Insurance Managers The chapters of part 1 provide a summary of the insights and assessments required for the insurer to make an informed decision to launch and expand an index insurance business line. Chapter 3 explains the process of completing a prefeasibility study, which establishes the presence—or lack thereof—of key prerequisites for launching an index insurance business line in a new market. Chapters 4 through 6 cover the pilot phase of launching an index insurance business line, during which the insurer works with a small selection of policy- holders to design an initial product offering. These chapters detail the process of designing, evaluating, and pricing the Base Index, and evaluating the Redesigned Index. Chapter 7 provides a description of a detailed market analysis to be completed following the pilot phase. The market analysis uses the insurer’s experience dur- ing the pilot phase to provide an understanding of the market’s potential for a commercial index insurance business line. Finally, chapter 8 explains how to determine the value of index insurance to a financier that provides loans to small farmers. Given a particular financier’s historical default rates and projected portfolio, the chapter illustrates to what degree index insurance can protect the financier against nonperforming loans caused by the named peril. Part 2 of the guide (chapters 9 through 15) explains the quantitative models* that produce the metrics and results used in part 1. * In part 2 of the guide, Monte Carlo simulation models are used (and explained in detail) to determine the expected outcomes of an index insurance product, as well as the risks inherent in such a product. Risk Modeling for Appraising Named Peril Index Insurance Products   19   http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 3 Prefeasibility Study 3.1 Introduction The purpose of the prefeasibility study is to determine whether the market possesses the basic prerequisites for the design and introduction of named peril ­ index insurance products. The ability to develop, refine, and scale up named peril index insurance products opens opportunities for risk carriers (insurers and ­ reinsurers) to reach large rural populations of potential customers. In the past few years, insurers have participated in pilot projects aimed at providing proof of concept for commercial index insurance products. With the focus on implement- ing pilots, relatively little energy was spent on analysis of the prerequisites for the successful expansion of coverage with these products. Experience to date from various pilot project studies has shown that reaching commercially viable volumes for named peril index insurance in a given market requires the presence of several key resources. This chapter discusses how the insurance manager should assess the prerequisites for a specific market. The insurance manager must evaluate each prerequisite for every target area and for the market as a whole, and only proceed with the product if the prerequisites exist in enough target areas to provide sustainable business volumes. The key questions that the prefeasibility study answers follow: • Are potential policyholders interested in buying this product? • Is a pool of subject specialists available to assist with product design? • Are historical hazard data series available with which to design and price products? • Are data providers able to provide real-time or near real-time hazard data for claim settlement during each risk period? • Are qualitative and quantitative inventory damage data for product design and product evaluation available? • Are local or international product design capabilities available? • Are distribution channels available through which the product can be sold effectively? Risk Modeling for Appraising Named Peril Index Insurance Products   21   http://dx.doi.org/10.1596/978-1-4648-1048-0 22 Prefeasibility Study • Are reinsurers willing to offer the necessary reinsurance capacity? • Has regulatory approval been granted to underwrite this product? • Are direct or indirect subsidies available? These 10 key points are discussed in more detail in the following section, and examples are provided in case example boxes 3CB.1–3CB.3. concrete ­ Case Example Box 3CB.1 Excellence Insurance’s Prefeasibility Study for Mapfumoland For Excellence Insurance the market is a specific country—Mapfumoland. Agriculture is a major part of the Mapfumoland economy, and more than two-thirds of the population engage in agricultural production. Most of these households pursue subsistence farming, but a growing segment—about 25 percent—engage in semi-commercial farming, using improved inputs and selling a portion of their harvests. Excellence hired a consulting firm, Research Plus, to complete a prefeasibility study for index insurance in Mapfumoland. The consultants submitted a detailed report on the avail- ability of historical weather data, ground and satellite real-time data providers, historical inventory damage data, local and international product design capabilities, distribution channels, reinsurance capacity, and the local regulatory position on weather index insur- ance. Lindiwe Maneli, the actuarial analyst, is studying the report and will summarize the findings before discussing them with the insurance manager, Ghassimu Sow. 3.2 Outline of Emerging Managerial and Process Controls STEP 1:  Summarize the Status of Prerequisites for the Product Design and Risk Transfer Process For each target area within the potential market, each of the following basic prerequisites should be in place and available to the insurer: • Potential policyholders: A sustainable index insurance market is one with either a large number of potential insured parties or a small number of players with very large portfolios to be insured. The insurer must be convinced that the market has sufficient demand for the product. • Subject specialists: Robust product structures are usually developed with ­ assistance from subject specialists such as agronomists and hydrologists. It is therefore important to make sure that the product development team has access to this expertise as it designs the products. • Historical hazard data: Most reinsurers require between 20 and 30 years of historical data—such as daily or dekadal rainfall data—to perform product pricing (box 3.1). This information is required for several stages within the product design, product evaluation, and product pricing processes. If such information is not available, designing a robust product will not be feasible. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Prefeasibility Study 23 Box 3.1 Changes in Risk Conditions over Time The use of historical hazard and inventory damage data for product design, evaluation, and pricing makes an implicit assumption that past risk conditions (for example, weather patterns) will continue into the future. Sometimes, however, these conditions change over time. Section  16.1 briefly continues this discussion. For now, two main types of changes in risk conditions are considered: ­ • Changes in the proxy due to changes in climate: If weather events become more severe, we can expect more severe damage than is observed in the historical data. • Changes in the degree of damage to inventory: In some cases, weather will not change, but the same events will cause significantly more or less damage. For example, environmental degradation or rapid urbanization may increase losses relative to previous, similar events. Conversely, new drainage systems or drought-resistant plant varieties may decrease losses relative to previous events of the same nature. When insurers design, evaluate, and price index insurance products, changing risk ­conditions must be accounted for as accurately as possible using qualitative and quantitative methods. • Real-time claim settlement hazard data: The intended use of an index insurance product is to offer prompt claim settlement during the risk period, at the end of the risk period, or both so that policyholders can have access to funds as soon as possible after the hazard occurs. If claims cannot be settled promptly because hazard data are not available, a major purpose of developing an index product is defeated. As a result, the data provider must be able to provide real-time or near real-time data throughout the entire risk period. In this way, ­ all key stakeholders can monitor the index parameters throughout the season. In addition, the insurer and reinsurer must have the necessary information to ensure they are holding appropriate liquid resources to make the required pay- ments within the agreed-on claim settlement period. Even with excellent his- torical hazard data, the insurer should not proceed with an index product if real-time hazard data are not readily available.1 • Historical inventory damage data: The product design team needs detailed qualitative data, quantitative data, or both on how the indexed peril has affected the insured parties in the past. Written records of historical yields are not available for most smallholder farmers. In these cases the product design team relies on farmers’ recollections along with information from local experts, government, and international sources, such as FEWS NET (USAID’s Famine Early Warning Systems Network), to rank the level of crop damage caused by the named peril in each year and geographical area. This process is termed qualitative classification of past damages, and is discussed in detail in section 4.3. Sometimes, the product design team has access to quantitative data, such as recorded yearly yields or loan write-offs. Using the available qualitative and quantitative data, the product design team will Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 24 Prefeasibility Study ­ valuate the Base Index against the information on inventory damage from e previous years. The available data must be substantial and accurate enough to support the product design process. • Product design capabilities: High-quality product design capabilities must be available to the insurer, either internally or externally. The product design team should be able to both design and statistically evaluate the performance of the products. As discussed in chapter 2, we recommend that insurers work toward developing this product design capacity locally in the long run. • Clear distribution channels: Given the small sum insured per farmer, selling index-based insurance to individual farmers is usually uneconomic. Most successful index schemes use distributors such as agribusinesses, financial ­ institutions, cooperatives, or other institutions that act as the aggregator and ­ policyholder on behalf of groups of farmers or other low-income individuals. The use of aggregators leads to low administrative costs for underwriting and claim settlement. Before investing heavily in the development of named peril index insurance, the insurer should identify clear distribution channels. In most cases, named peril index insurance is bundled with other services such as access to finance. Understanding the underlying service in which the farmer is interested is critically important to the success of index distribution. The insurer must evaluate the value chain for each crop to be insured. Farmers in a poorly organized value chain will likely be blocked from accessing financing to pay for farming inputs as well as insurance premiums. The insurer should also pay close attention to issues of market liquidity and the cost of finance when evaluating value chains because these will affect the potential market size for named peril index insurance. • Reinsurance capacity: Named peril index insurance is normally used to transfer covariant risks that can affect a whole country or region at the same time. As a result, most of the risk is transferred to international financial markets instead of being kept locally. Therefore, before offering named peril index insur- ance, the insurer should make sure it has access to sufficient reinsurance capacity. As long as volumes are high and data are of good quality, reinsurance capacity is usually accessible. Reinsurance prices are, however, sensitive to market condi- tions and sometimes volatile. For example, a major disaster in Asia can increase reinsurance renewal prices worldwide. For this reason, ­ insurers should consider the potential for reinsurance price increases when evaluating products. • Regulatory approval: In many developing countries, index insurance is not specifically regulated but is included under the “miscellaneous” class. This clas- ­ sification is common during pilot index insurance projects that are supported by multilateral organizations that provide quality control and a degree of self-regulation. However, as the product line matures, comprehensive regula- ­ tion is needed to ensure the functioning of the market and proper treatment Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Prefeasibility Study 25 of policyholders. Regulators’ understanding and approval of index products are critical for scaling up index insurance product lines. • Premium subsidies: The availability of direct or indirect subsidies is not a ­ prerequisite, but can considerably support the development, scalability, and viability of index insurance products, especially during the early stages of the product life cycle. However, it is important to consider whether these subsidies will be in place for the short or long term. If for the short term, the insurer will have to determine whether the target market will be willing and able to pay higher premiums once the subsidies end. Based on research into the prerequisites for index insurance in the market under consideration, the insurance manager and actuarial analyst evaluate the relative strength of each prerequisite for the target areas and the overall market. If all prereq- uisites are in place for the overall market, the insurance manager identifies the mar- ket as a priority for launching the pilot phase. If many prerequisites are missing from the market, the insurance manager should consider ­ waiting to develop an index insurance product line until conditions improve and more prerequisites are met. Although this list of prerequisites is not exhaustive, our experience working on index insurance in more than 20 developing countries suggests that these are the critical elements for scaling up index insurance business lines. ­ In addition to the prerequisites discussed above, the insurance company in its due diligence process should consider other important factors, including, for example, the following: • In what ways does the firm have a comparative advantage in this market (is it already doing business there; is it able to leverage experience and expertise)? • Is the cost-benefit analysis for this market superior to that for other investment or business development opportunities the firm may have? Because the above points are not unique to the evaluation of an index ­ insurance product, they are not discussed in further detail in this guide. Case Example Box 3CB.2 Summary of Key Points from the Research Plus Prefeasibility Study on the Mapfumoland Market Prerequisite Key points from the Research Plus prefeasibility study Potential policyholders • More than 500,000 smallholder farmers work with the five distribution channels that have expressed interest in the index product. • The rural bank and the agribusiness Buyer Goods are also interested in purchasing an index product to protect their agrifinance and input advance portfolios. Subject specialists • In each area, a number of local extension officers, specialists from agribusinesses and suppliers, and employees of research institutions work closely with smallholder farmers. • The report provides a list of three to five recommended subject specialists for each area. These specialists helped the Research Plus consultant develop qualitative classifications of past damages. box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 26 Prefeasibility Study Case Example Box 3CB.2  Summary of Key Points from the Research Plus Prefeasibility Study on the Mapfumoland Market (continued) Prerequisite Key points from the Research Plus prefeasibility study Historical hazard data • The Mapfumoland meteorological department operates 100 weather stations, which have recorded 30 years of good quality daily historical rainfall data. Of these weather stations, 50 have also recorded 20 years of daily temperature, humidity, and wind speed data. The data can be accessed for a nominal fee. • ARC2 daily rainfall satellite data are available from 1983 at a pixel size of 10 kilometers by 10 kilometers. Real-time claim • Of the 100 meteorological department weather stations, 80 are fully functional and settlement hazard can provide real-time data. data • ARC2 daily rainfall satellite data are also available and can be accessed for free. Historical inventory • Research Plus worked with selected subject specialists in each area to develop damage data area-specific categorical classifications of past damages. • Substantial qualitative information is available from FEWS NET, local government agencies, farmers, and local agribusiness firms. Product design • Two Mapfumoland specialist insurance intermediaries offer product design services capabilities and charge a fair service fee. Hazard Analytics has the stronger reputation in the local and international market. • Several international product design firms can also be hired to build internal capacity at Excellence. • The report recommends outsourcing the product design function to Hazard Analytics. Distribution channels • Five distribution channels have expressed interest in bundling named peril index insurance with existing services provided to maize farmers: a rural bank, a microfinance institution, a seed company, the agribusiness Buyer Goods, and a nongovernmental organization. • The maize value chain is well organized. The government purchases 50 percent of yields for the national grain reserve, and several local and national input suppliers cooperate with financial institutions to provide inputs on credit. Reinsurance capacity • All five reinsurance companies currently working with Excellence have expressed interest in supporting this class of business. Regulatory approval • The regulator has agreed that the index product may be launched, but has requested sample policy documents. Premium subsidies • Premium subsidies are currently not available. Note: ARC2 = African Rainfall Climatology, version 2; FEWS NET = Famine Early Warning Systems Network. STEP 2:  Evaluate, Document, and Communicate the Business Decision At this stage, the insurance manager documents the presence or absence of each of the basic prerequisites in the market. Based on the status of the prerequisites, the manager decides whether it is worth the effort for the insurer to pursue a pilot phase. The more prerequisites that are in place, the more confident the manager can be in starting the product design process. Because this is a subjective decision, the manager may want to specify a minimum number of prerequisites that each market must have in place before recommending a pilot phase. STEP 3:  Plan For and Resource the Pilot Phase If sufficient prerequisites are in place, we recommend that the insurer launch a pilot phase by working with a few potential policyholders to design, evaluate, Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Prefeasibility Study 27 Case Example Box 3CB.3 Excellence Insurance Technical Evaluation of Prefeasibility Study Insurance manager(s) Ghassimu Sow Actuarial analyst(s) Lindiwe Maneli Technical evaluation of the prefeasibility study YES NO Are potential policyholders interested in buying this product? X Is a sufficient pool of subject specialists available to assist with product design? X Are sufficient historical hazard data series available to design and price products? X Are data providers able to provide real-time or near real-time hazard data for X claim settlement during each risk period? Are sufficient qualitative or quantitative inventory damage data for product X design and product evaluation available? Are sufficient local or international product design capabilities available? X Are distribution channels available through which the product can be sold X effectively? Are reinsurers willing to offer the necessary reinsurance capacity? X Has regulatory approval been granted to underwrite this product? X Are premium subsidies available X Total 9 1 Final decision Should the company initiate a pilot project? X launch, and monitor the performance of several products before moving to the commercial phase in which the insurer offers the products to the wider market. Pilot testing allows the insurer to evaluate whether named peril index insurance is the right product for the target market and risk in question. In some situations, other risk management products provide better solutions than index insurance. In other situations, index insurance is best combined with other agriculture insurance and risk management solutions to provide a hybrid product. These nuances are most effectively uncovered during the pilot phase. The insurer must dedicate sufficient resources to the pilot phase because it is a critical part of market research. The activities undertaken during the pilot phase are similar to those implemented during the commercial phase; the only difference is the scale at ­ which each activity is undertaken, that is, with one or two policyholders in the pilot phase rather than with many policyholders in the commercial phase. Chapters 4, 5, and 6 explain in detail how to undertake a pilot with one policy- holder. Chapters 7 and 8 expand the discussion to rolling out the product line to the broader market. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 28 Prefeasibility Study Note 1. In many developing countries, weather stations do not meet the international standards necessary for their data to be used for insurance claim settlement. Installing ­ more and better quality weather stations can improve product quality by reducing geographical basis risk for products based on weather station data. Better weather station data can also help calibrate satellite-based data that are now widely used for product design. Bibliography Clarke, D. 2012. Weather-Based Crop Insurance in India. Washington, DC: World Bank. Dick, W., A. Stoppa, J. Anderson, E. Coleman, and F. Rispoli. 2011. Weather Index–Based Insurance in Agricultural Development: A Technical Guide. Rome: IFAD. Mahul, O. V. Niraj, and D. Clarke. 2012. “Improving Farmers’ Access to Agricultural Insurance in India.” Policy Research Working Paper 5987, World Bank, Washington, DC. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 4 Product Design and Evaluation— The Base Index 4.1 Introduction Now that the insurer has verified the presence of the prerequisites for index product design in the chosen market (chapter 3), the pilot phase begins with the design and evaluation of the Base Index for one or two policyholders. The insurer will follow the same steps for product design and evaluation during the later commercial phase when it takes on a larger number of policyholders. As discussed in chapter 3, product design is often outsourced to specialist firms. However, because the insurance company is ultimately responsible for the perfor- mance of the product, the insurer must fully understand product parameters and performance, including product design basis risk. Underwriting index insurance products without a solid understanding of the product can cause capital flight in the medium to long term because of unexpectedly high claims or basis risk events. Increasing the understanding of product performance and behavior by both the insurer and the policyholder is therefore critically important for the long-term sustainability of index insurance. To help bridge the knowledge gap between product design teams and insurers, this chapter highlights key points for discussion that promote transparency in the product design and risk transfer process. The Base Index product design process, in particular the data required for the process, is discussed. In addition, the pro- cess for evaluating the level of product design basis risk for the Base Index, a critical process in making index insurance business decisions, is explained. The key managerial questions answered during Base Index product design and evaluation are the following: • How does the frequency of the Base Index’s projected payouts compare to the frequency of actual inventory damage events? • What is the Base Index’s level of insured party basis risk? Specifically, – How frequently will damage to the insured party’s inventory that is caused by the named peril exceed the payouts provided by the Base Index? – What will be the magnitude of uncompensated inventory damage? Risk Modeling for Appraising Named Peril Index Insurance Products   29   http://dx.doi.org/10.1596/978-1-4648-1048-0 30 Product Design and Evaluation—The Base Index • What is the Base Index’s level of insurer basis risk? Specifically, – How frequently will payouts exceed the actual damage to the insured party’s inventory that is caused by the named peril? – What will be the magnitude of claims that exceed the actual inventory damage? This chapter discusses these key questions in detail and provides concrete examples in the case example boxes. See case example box 4CB.1 to start. Case Example Box 4CB.1 Overview of the Base Index for Mass Bank The Hazard Analytics product design team contracted by Excellence Insurance is developing a Base Index for Mass Bank. Mass Bank is new to lending to small farmers and is one of the few Mapfumoland financial institutions that does so. Although Mass Bank is excited to be a first mover in this market segment, it is still cautious about the many risks associated with agricultural lending. Mass Bank recently partnered with a local agricultural college to offer extension services to its customers to manage some production risks. Even so, because Mass Bank’s customers practice rain-fed agriculture, it is very concerned about the risk of loan defaults following a severe drought. 4.2  Basis Risk and the Implied Deductible Because named peril index insurance is relatively new, many people believe that basis risk is always the cause of an index not triggering when losses caused by the named peril are experienced on the ground. In reality, the term “basis risk” applies to a narrower set of scenarios related to the performance of the Base Index. Basis risk is defined as the difference between the payout triggered by the Base Index and the actual losses attributable to the named peril. This differ- ence can be positive or negative. Insurer basis risk describes when the payout is greater than actual losses—the insurer suffers economic losses caused by unnecessary payments. Insured party basis risk describes when the payout is smaller than the actual losses—the insured party suffers an economic loss from the named peril but the contract does not provide adequate compensa- tion (figure 4.1). Figure 4.1 Insurer and Insured Party Basis Risk a. Insurer basis risk b. Insured party basis risk Actual loss to Payout triggered Payout less than Payout triggered insured party by Base Index actual loss by Base Index Payout exceeds Actual loss to actual loss insured party Payout in event Payout in event of covered loss of covered loss Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Design and Evaluation—The Base Index 31 4.2.1  Product Design Basis Risk Product design basis risk results from the inability of the Base Index to ever per- fectly reflect the reality on the ground because its payouts reflect average losses, not the losses of the specific insured party. Product design basis risk is inherent in all named peril index insurance products; therefore, the insurer’s focus should be on reducing instead of eliminating it. Once a product has been marketed for a number of years and reliable quan- titative inventory damage data have been collected, these data can be used to refine the index product. Over time, the product design team will have increas- ingly reliable information with which to analyze product design basis risk. This chapter provides a detailed process for using this type of information to evaluate the Base Index’s product design basis risk, starting in the pilot phase. It is important to note that changes in risk conditions over time must be addressed regularly as part of evaluating product design basis risk. Changes in the behavior of the proxy will lead to changes in the payouts triggered by the index insurance product, while changes in the degree of damage to inventory will cause changes in the actual losses attributable to the named peril. These elements are central to determining the magnitude of basis risk for the product. 4.2.2  The Redesigned Index’s Implied Deductible Index product design typically occurs in two phases. In the first phase, the product design team develops a Base Index using input from local subject spe- cialists, and evaluates and prices it. The Base Index provides the highest level of coverage possible against damage to the farmer’s inventory caused by the named peril. The Base Index triggers a payment when the proxy’s behavior indicates that any damage to inventory—no matter how small—is expected. The Base Index is used as a point of reference for discussing product options with policyholders, and is extremely important for ensuring that the policy- holder fully understands the product purchased. It is also extremely important that the insurer always produce a Base Index to explain to the policyholder the difference between complete coverage—that provided by the Base Index—and the coverage provided by other product options. Without this explicit compari- son, policyholders often fall into the trap of expecting complete coverage even when they have purchased a lower coverage, less expensive product. If the Base Index meets the policyholder’s expectations, it is the final product purchased by the policyholder. In many cases, however, the Base Index will cost more than what the policyholder is initially willing or able to pay, so the second phase of product design begins. The product design team uses input from the policyholder on price to design the Redesigned Index with new parameters so that the premium decreases. The implied deductible is the difference in coverage between the Base Index and the Redesigned Index. It is the amount of risk that the policyholder chooses to retain and not transfer to the insurance company. The Base Index is designed to transfer as close to all of the risk from the named peril as possible from the Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 32 Product Design and Evaluation—The Base Index Figure 4.2 Redesigned Index Implied Deductible a. Base Index b. Redesigned Index Risk of insured party to be transferred with Redesigned Risk of insured party Index Redesigned to be transferred Index product: Insurer Insurer Portion of risk less coverage, with Base Index Base Index less expensive of insured party product: retained = more coverage, Redesigned Index more expensive implied deductible Figure 4.3 Product Design Basis Risk versus the Redesigned Index Implied Deductible Redesigned Index Historical (or Base Index payout payout (adjusted to current) inventory (based on expert meet policyholder damage data input) willingness to pay) Product design basis Redesigned Index risk implied deductible policyholder to the insurance company. The Redesigned Index, on the other hand, transfers less of the named peril risk from the policyholder to the insurance company (figures 4.2 and 4.3). Of course, the insurer must make sure that the policyholder fully understands the implications of choosing the Redesigned Index. The best tools for explaining the differences between the Base Index and the Redesigned Index are discussed in chapter 6. 4.2.3  Identifying Examples of Product Design Basis Risk versus the Implied Deductible Do not forget that the Redesigned Index is derived from the Base Index. If the Base Index suffers from product design basis risk, the Redesigned Index will too. So, when a policyholder experiences losses caused by a named peril but does not receive a payout from the index product that is equal to the damages, the insurer and other stakeholders should always try to determine whether the cause is product design basis risk or the implied deductible of the Redesigned Index. In these cases, we suggest that the insurer calculate the payout values for the risk period for both the Base Index and the Redesigned Index. If the Base Index triggers but the Redesigned Index does not, the implied deductible is the reason for the difference. If neither the Base Index nor the Redesigned Index triggers, in spite of observed losses caused by the named peril on the ground, product design basis risk is the cause of the problem. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Design and Evaluation—The Base Index 33 For example, imagine a very simple Base Index for drought and dry spell cov- erage that provides a payout of 100 percent of the sum insured when the cumu- lative rainfall for the period is less than 50 millimeters (exit) and no payout when the cumulative rainfall is more than 100 millimeters (trigger). It provides pay- outs equal to 2 percent of the sum insured per millimeter of rain below the trigger. The Redesigned Index also provides a payout of 100 percent of the sum insured when the cumulative rainfall is less than 50 millimeters (exit), but it provides no payout when the cumulative rainfall is more than 75 millimeters (trigger). It provides payouts equal to 4 percent of the sum insured per millimeter of rain below the trigger. As seen in figure 4.4, when cumulative rainfall is 50 millimeters or less, both the Base Index and the Redesigned Index will pay out 100 percent of the sum insured. When cumulative rainfall is 100 millimeters or higher, both pay out nothing. For cumulative rainfall between 50 millimeters and 100 millimeters, the Redesigned Index always pays less than the Base Index. This difference in payout is the implied deductible. Take the case in which cumulative rainfall for the period is 80 millimeters. The Base Index payout will be 40 percent of the sum insured. This is the amount of damage to the insured farmers’ inventory that we expect to see from drought and dry spells. However, the Redesigned Index will provide no payout. The dif- ference in payouts here—40 percent of the sum insured—is the value of the implied deductible. The farmer experiences a loss of 40 percent but receives no payout, because this is the value of the risk retained when the insured party selected the less expensive, lower coverage product. Now take the case of cumulative rainfall of 105 millimeters. The Base Index provides no payout for rainfall at this level, meaning that no damage to the farmers’ inventory is expected from drought. But imagine that, although rainfall was at this relatively high level, temperatures were extremely high, Figure 4.4 Identifying Product Design Basis Risk versus the Implied Deductible 120 100 80 mm 105 mm 80 Payout (%) 60 40 20 Implied deductible of 40% 0 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 110 Cumulative rainfall (millimeters) Redesigned Index Base Index Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 34 Product Design and Evaluation—The Base Index resulting in drought conditions that caused damage to the insured farmers’ crops. The damage to the farmers’ inventory was caused by the named peril— drought—but the proxy used for the index was not sufficient to predict this damage. An additional proxy—temperature—is required to account for this scenario. This is an example of product design basis risk. 4.3 Steps in Product Design and Evaluation This section discusses the steps in the Base Index design and evaluation process, including evaluating the Base Index for product design basis risk (see summary in figure 4.5). Chapter 11 provides a step-by-step guide to using the probabilistic models on which the decision metrics in this section are based. Product pricing is also an important part of evaluating the Base Index, and is discussed in chapter 5. 4.3.1  STEP 1: Collect Historical Hazard Data To design the Base Index, the product design team collects historical hazard data. These data are used for designing and evaluating the Base Index in Step 7. Historical hazard data are available in many forms, for example, daily rainfall, daily temperature, and daily wind speed. The historical behavior of specific hazards helps determine the triggers for the Base Index. For example, to design a product that triggers based on average temperature and cumulative rainfall during a spe- cific period of the crop cycle, the product design team will need historical daily temperature and rainfall data. 4.3.2  STEP 2: Collect and Summarize Historical Inventory Damage Data Historical inventory damage data are often very scarce for the low-income market. Therefore, product design teams often produce a categorical classifica- tion of past damages for each year and geographical area. In this process, the product design team relies on farmers’ recollections and information from local experts as well as government and international sources to categorize the level of crop damage caused by the named peril in each year and geographical area (case example box 4CB.2). Of course, these semi-quantitative data may be biased (for example, recall bias), so the analysis and results based on these data should be interpreted and used with care. Once a product has been marketed for a number of years and reliable quan- titative inventory damage data have been collected, these data can be used to refine the index product. Over time, the product design team will have more reliable information with which to analyze product design basis risk. 4.3.3  STEP 3: Collect Relevant Information from Subject Specialists and Policyholder Successful index development relies heavily on inputs from subject specialists such as agronomists, hydrologists, and seismologists who provide information that helps the product design team set index triggers and payout rates (case example box 4CB.3). These specialists—and the prospective policyholder—can Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Design and Evaluation—The Base Index 35 Figure 4.5  Base Index Design and Evaluation Process Step 1: Collect historical hazard data Step 4: Design Base Index term sheet Historical hazard data, for example, Summarize terms of contract and rainfall, wind speed, temperature features of Base Index Step 2: Collect and summarize historical Step 5: Calculate historical inventory damage data index values If quantitative data not available then categorical classifications of past Using historical hazard data collected damages, for example, via surveys/ interviews Step 3: Collect relevant information Step 6: Calculate historical from subject specialists and payout policyholder From agronomists, hydrologists, Compare historical index values seismologists, and policyholders to set to payout schedule in term sheet, index triggers, critical times for crop determine historical payout per year cycle, payout rates, coverage period and area Step 7: Evaluate Base Index product design basis risk Metrics for product design Metrics for the insured party Metrics for insurer product basis risk product design basis risk design basis risk Projected return periods for Probability that the Base Index Probability that the Base Index inventory damage will not experience an insured will not experience an insurer Projected return periods for party basis risk event basis risk event the Base Index Expected value and TVaR for Expected value and TVaR for Return period ratios insured party basis risk as a insurer basis risk as a percentage percentage of the portfolio value of the portfolio value Historical years with largest Historical years with largest insured party basis risk ratios insured party basis risk ratios Step 8: Document/communicate business decision Present basis risk evaluation of Base Index to policyholder If policyholder not satisfied, change structure of product to improve basis risk evaluation of Base Index If policyholder satisfied, move forward with the pricing Base Index (chapter 5) Note: TVaR = tail value at risk. Case Example Box 4CB.2 Categorical Classification of Past Damages for Areas A to J, 1984–2013 The Hazard Analytics product design team uses historical rainfall data from the local meteorological service. The data show 30 years of daily rainfall for all 10 of the geographical areas needed. Based on information from FEWS NET and interviews with local government agencies, the local agricultural college, farmers like Rose Jituboh, and agribusiness firms, the Hazard Analytics product box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 36 Product Design and Evaluation—The Base Index Case Example Box 4CB.2  Categorical Classification of Past Damages for Areas A to J, 1984–2013 (continued) design team classified each year in each area as either good or bad according to inventory damage due to drought risk. Each good year was assigned a value of zero. The team rated each bad year on a scale of 1 to 5, where 5 corresponded to the highest damages from drought and 1 to mild damages from drought. Key 0 Good year 1 1–20 percent loss 2 21–40 percent loss 3 41–60 percent loss 4 61–80 percent loss 5 81–100 percent loss Note: FEWS NET = Famine Early Warning Systems Network. Case Example Box 4CB.3  Subject Specialist Information for Base Index Product Design The Hazard Analytics product design team hires a team of local agronomists identified in the prefeasibility report with experience working with maize farmers in the 10 geographical areas of interest. The agrono- mists design a product with two triggers. Having these two triggers is important because the dry days measure the spread of rainfall while the cumulative triggers measure the quantity. For a crop to do well, sufficient rainfall that is spread out across the season is required. box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Design and Evaluation—The Base Index 37 Case Example Box 4CB.3  Subject Specialist Information for Base Index Product Design (continued) Trigger 1: Consecutive Dry Days The agronomists report that local maize crops need a long duration of regular rainfall to grow, so they recommend looking at the number of consecutive dry days in a growing season and its effect on the crop’s health. The number of consecutive dry days, defined as the number of days with less than 2.5 millimeters of rain during the crop cycle, will be the first proxy used in the Base Index design (Trigger 1). The agronomists explain that there are two important thresholds for consecutive dry days. The first is the number of consecutive dry days after which the crop will suffer from water stress, which is 20 days. The second is the number of consecutive dry days after which the crop will die, which is 40 days, especially if the dry period occurs during the flowering stage in the crop cycle. Based on the information from the agronomists, the product design team designs the Base Index so that it pays out 2.5 percent of the sum insured for each consecutive dry day above the first thresh- old (20 days). After 21 consecutive dry days, the payout will be 2.5 percent of the sum insured. By the 40th consecutive dry day—the day the crop will die—the total payout will reach 50 percent of the sum insured. The payout will also be 50 percent for any number of consecutive dry days greater than 40. Forty consecutive dry days is called the exit for Trigger 1, because this number (and above) receives the maximum payout of 50 percent. Trigger 2: Cumulative Rainfall Because maize is especially vulnerable to dry weather during the flowering stage of the crop cycle, cumula- tive rainfall (in millimeters) during the flowering period will be the second proxy used in the product design (Trigger 2). The agronomists identify the threshold for cumulative rainfall during the flowering period as 100 millimeters of rain, which is the cumulative amount of rain that must fall during the flowering period for the maize to flower successfully. Any cumulative amount below this threshold will result in the loss of the crop. Based on the information from the agronomists, the Hazard Analytics product design team designs the Base Index so that it will pay out 2 percent of the sum insured for every millimeter less than 100 millimeters of cumulative rain during the flowering period. So, if a cumulative total of 99 millimeters of rain falls during the flowering period, the payout will be 2 percent of the sum insured. If a cumulative total of 50 millimeters or less falls during the flowering period, the payout will be 100 percent of the sum insured (figure 4CB.3.1). The exit for Trigger 2 is 50 millimeters, because this amount (and lower) receives the maximum payout of 100 percent. Final Payout The product design team decides that the final payout will be the greater of the payout for Trigger 1 or Trigger 2. box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 38 Product Design and Evaluation—The Base Index Case Example Box 4CB.3  Subject Specialist Information for Base Index Product Design (continued) Figure 4CB.3.1  Base Index Trigger and Exit for Trigger 2 150 Claim trigger: If cumulative rainfall is less than 100 millimeters, excellence pays Cumulative 2.0% of sum insured (mass bank farmer’s loan, that is, $100) for every rainfall (millimeters) millimeter below 100 millimeters as proxy for 100 drought or dry spell As an example, if cumulative rainfall is 80 millimeters, then 2.0% of $100 mass bank farmer loan x 20 millimeters = $40 insurance payout 50 Exit: If cumulative rainfall is less than 50 millimeters, then 100% of total sum insured (that is, $100) is paid to the mass bank farmer July 2 August September 2 Flowering period of maize crop Case Example Box 4CB.4 Term Sheet for Base Index Insured areas Areas A to I Area J Participating measurement Stations A to I Station J stations Target crops Maize Maize Type of insurance cover Weather index insurance that pays out Weather index insurance that pays out a a defined percentage of the total sum defined percentage of the total sum insured when the following events insured when the following events occur at participating measurement occur at participating measurement stations during the total cover period: stations during the total cover period: • A specified number of consecutive • A specified number of consecutive dry days OR dry days OR • Total rainfall less than a specified level. • Total rainfall less than a specified level. These measures approximate weather These measures approximate weather conditions that may cause inventory conditions that may cause inventory damage for the policyholder that, as a damage for the policyholder that, as a result, cause losses for the insured. result, cause losses for the insured. Total contract period June 20–September 17, inclusive June 20–September 17, inclusive Maximum payout The greater of Trigger 1 payout or The greater of Trigger 1 payout or Trigger 2 payout (see below), up to Trigger 2 payout (see below), up to 100 percent of the total sum insured 100 percent of the total sum insured Maximum specified distance 20 kilometer radius 20 kilometer radius Total sum insured Total loan portfolio Total loan portfolio box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Design and Evaluation—The Base Index 39 Case Example Box 4CB.4  Term Sheet for Base Index (continued) Insured areas Areas A to I Area J Claim Trigger 1 Payout trigger event definition Number of consecutive dry days Number of consecutive dry days Days immediately following one Days immediately following one another, where the total rainfall another, where the total rainfall recorded on each day is recorded on each day is 2.5 millimeters or less. Recorded 2.5 millimeters or less. Recorded rainfall is that taken from the rainfall is that taken from the participating measurement stations participating measurement stations during the cover period for Trigger 1. during the cover period for Trigger 1. The longest consecutive dry day The longest consecutive dry day period during the cover period is the period during the cover period is the index value, which is evaluated index value, which is evaluated against the payout schedule. against the payout schedule. Cover period June 20–September 17, inclusive June 20–September 17, inclusive Payout schedule Trigger 20 consecutive dry Trigger 20 consecutive dry days days Payout rate 2.5 percent per dry Payout rate 2.5 percent per dry day above the day above the trigger trigger Exit 40 consecutive dry Exit 50 consecutive dry days days Number of payments allowed Only one payment is allowed for this Only one payment is allowed for this trigger. trigger. Timing of payment Payments due according to the Payments due according to the definitions above may be made at the definitions above may be made at the end of the total cover period, that is, end of the total cover period, that is, after September 17. after September 17. Claim Trigger 2 Payout trigger event definition Total rainfall for flowering period Total rainfall for flowering period Total millimeters of rainfall recorded Total millimeters of rainfall recorded during the flowering period. during the flowering period. Recorded rainfall is that taken from Recorded rainfall is that taken from the participating measurement the participating measurement stations. For the contract period, stations. For the contract period, cumulative rainfall is obtained by cumulative rainfall is obtained by summing daily amounts over the summing daily amounts over the contract period. The resulting amount contract period. The resulting amount is the index value, which is evaluated is the index value, which is evaluated against the payout schedule. against the payout schedule. Cover period July 25–September 2, inclusive July 25–September 2, inclusive Payout schedule Trigger 100 millimeters Trigger 60 millimeters Payout rate 2 percent per Payout rate 2 percent per millimeter millimeter below trigger below trigger Exit 50 millimeters Exit 10 millimeters Number of payments allowed Only one payment is allowed for this Only one payment is allowed for this trigger. trigger. Timing of payment Payments due according to the Payments due according to the definitions above may be made at the definitions above may be made at the end of the total cover period, that is, end of the total cover period, that is, after September 17. after September 17. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 40 Product Design and Evaluation—The Base Index also provide detailed information for the coverage period, especially critical times during the crop cycle. 4.3.4  STEP 4: Design Base Index Term Sheet Based on information from the subject specialists and the prospective policy- holder, the product design team writes the Base Index term sheet. The term sheet summarizes the terms of the contract and the features of the Base Index. 4.3.5  STEP 5: Calculate Historical Index Values Now the product design team will begin evaluating the Base Index by calculating historical index values from the historical hazard data they have already collected (case example box 4CB.5). This process demonstrates how the product would have performed if it had been in the market during past growing seasons. In this step, the product design team calculates the historical index value for each proxy, year, and geographical area. 4.3.6  STEP 6: Calculate Historical Payouts Now the product design team compares each historical index value to the payout schedule in the term sheet to determine the historical payouts for each year and geographical area (case example box 4CB.6). The historical payouts are estimates Case Example Box 4CB.5 Calculation of Historical Index Values for the Base Index For each year and geographical area, the Hazard Analytics product design team follows the steps below. Calculate the historical index values for Proxy 1 (number of consecutive dry days) • Define cover period as June 20 to September 17 for each year. • Define dry day as a day with 2.5 millimeters or less of rain. • Define dry day trigger (level above which payment starts) as 20 days. • Define dry day exit (level at or above which full payout is triggered) as 40 days. • Classify each day as either dry (less than or equal to 2.5 millimeters of rain) or wet (more than millimeters of rain). 2.5 ­ • Find the longest stretch of consecutive dry days. The number of days in this period is the index value for Proxy 1. Calculate the historical index values for Proxy 2 (cumulative rainfall during flowering period) • Define the flowering period as July 25 to September 2 for each year. • Define cumulative rainfall trigger (level below which payment starts) as 100 millimeters. • Define cumulative rainfall exit (level at and below which full payout is triggered) as 50 millimeters. • Add up the cumulative rainfall amount for each day during the flowering period. This value is the index value for Proxy 2. Note: In many product design cases, a daily cap is included in the product design so that any amount above such a limit is not included in the cumulative total. In addition, the contract start date is often dynamic and not fixed, which means the contract starts once a certain condition has been met, such as having received a total rainfall amount of more than 50 millimeters within a three-day period. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Design and Evaluation—The Base Index 41 Case Example Box 4CB.6 Calculation of Historical Payouts for the Base Index For each year and geographical area, the Hazard Analytics product design team follows the steps below. • Calculate the historical payout for Proxy 1: Compare the historical index value for Proxy 1 to the payout schedule in the term sheet (case example box 4CB.4) to find the histori- cal payout for Proxy 1. • Calculate the historical payout for Proxy 2: Compare the historical index value for Proxy 2 to the payout schedule in the term sheet (case example box 4CB.4) to find the histori- cal payout for Proxy 2. • Determine the final payout: Select the higher payout between that for Proxy 1 and that for Proxy 2. This value is the final payout for the specific year and geographical area. See results in case example box 4CB.7. Case Example Box 4CB.7 Historical Payouts for the Base Index of what the Base Index would have paid out to the policyholder if the contract had been in place during previous seasons. These are used in Step 7 to evaluate product design basis risk and in chapter 5 for Base Index product pricing. 4.3.7  STEP 7: Evaluate Base Index Product Design Basis Risk The actuarial analyst now uses the historical payouts and the historical inventory damage data provided by the product design team to evaluate the Base Index’s product design basis risk. Using the probabilistic model detailed in chapter 12, the actuarial analyst calculates the following metrics to provide a quantitative description of the index’s basis risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 42 Product Design and Evaluation—The Base Index 4.3.7.1  Metrics for Product Design Basis Risk • The projected return period for inventory damage is the frequency at which inventory damage caused by the named peril occurs at specific damage levels (for example, damage to 10 percent of the inventory, 30 percent of the inventory, 50 percent of the inventory, and 70 percent of the inventory). For example, if two droughts that damage 10 percent of crops are observed in a 20-year period, the return period for drought at this damage level is 1 in 10 years. • The projected return period for the Base Index is the frequency at which the Base Index makes a payout at specific payout levels (for example, payouts of 75 percent of the sum insured, 50 percent of the sum insured, 25 percent of the sum insured, and 5 percent of the sum insured). • The return period ratio is calculated as the ratio of the projected return period for inventory damage to the projected return period for the Base Index at spe- cific damage and payout levels. When the return period ratio is equal to 1, the Base Index triggers a payout at the same frequency as the occurrence of actual inventory damage caused by the named peril. When the ratio is greater than 1, the Base Index triggers a payout more frequently than the occurrence of inven- tory damage caused by the named peril (insurer basis risk). The greater the value of the ratio, the greater the amount of insurer basis risk. When the ratio is between 0 and 1, the Base Index triggers payouts less frequently than the occurrence of actual inventory damage caused by the named peril—farmers experience damage from the named peril but the contract does not trigger a payout (insured party basis risk). The greater the value of the ratio, the smaller the amount of insured party basis risk. 4.3.7.2  Metrics for Insured Party Product Design Basis Risk • The probability that the Base Index will not experience an insured party basis risk event is the likelihood that the product will either trigger a payout when there is inventory damage caused by the named peril, or trigger no payout when there is no inventory damage due to the named peril. • The expected value and tail value at risk (TVaR) for insured party basis risk as a percentage of the portfolio value are the magnitude of underpayments to the policyholder due to product design basis risk expressed as a percentage of the total portfolio value for all areas covered by the product. • Historical years with largest insured party basis risk ratios lists historical years when the Base Index would have triggered insufficient compensation to the insured party because of product design basis risk. 4.3.7.3  Metrics for Insurer Product Design Basis Risk • The probability that the Base Index will not experience an insurer basis risk event is the likelihood that the product will not trigger a payout when there is no inventory damage due to the named peril. • The projected value and TVaR for insurer basis risk as a percentage of the portfolio are the magnitude of overpayments to the policyholder due to product design Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Design and Evaluation—The Base Index 43 basis risk expressed as a percentage of the total portfolio value for all areas covered by the product. • Historical years with largest insurer basis risk ratios lists historical years when the Base Index would have triggered excessive compensation to the insured party because of product design basis risk. The insurance manager, the actuarial analyst, and the product design team review the values of these metrics and evaluate them against a set of guidelines developed by the insurer’s risk management committee, with input from international reinsurance brokers or international reinsurers where necessary (case example boxes 4CB.8 and 4CB.9). The risk management committee guidelines should indicate the acceptable range of values for each metric. These agreed-on guidelines are critical for managing the insurer’s reputational risk, which is linked to the product’s quality. When a product’s metrics fall outside of this range, the insurance manager should request that the product design team review and improve the Base Index structure. In cases in which the Base Index meets all the requirements outlined in the evaluation guidelines, the insurer moves on to the next step: policyholder engage- ment. However, if the Base Index does not meet the requirements, the product design team must review the product to identify changes to the structure that will improve the Base Index’s basis risk evaluation. These changes should still align with the recommendations provided by the subject specialists. Case Example Box 4CB.8 Excellence Insurance Product Evaluation Guidelines for the Base Index Risk committee guidelines for index products Decision metric Insured party basis risk Insurer basis risk Projected return period for inventory Must be as close as possible to each other for each area, damage and the Base Index especially for damage or payout levels of 50 percent and 70 percent Return period ratio Must be between 0.7 and 1.2 for each area and damage or payout level Probability that the Base Index will Must be greater than Same as insured party basis not experience a basis risk event 75 percent for each area risk Expected value for basis risk as a Must be less than 5 percent Same as insured party basis percentage of the portfolio value risk TVaR at 95 percent for basis risk as a Must be less than 20 percent Same as insured party basis percentage of the portfolio value risk Final decision Present Base Index to policyholder Restructure index for specific areas Consider alternative solutions (non-index) Note: TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 44 Product Design and Evaluation—The Base Index Case Example Box 4CB.9 Product Evaluation Decision Metrics for the Base Index The product evaluation model outputs above give the Excellence Insurance team important insights into different characteristics of the Base Index. With regard to insured party basis risk, Lindiwe, the actuarial analyst, observes the following: • The projected return period for inventory damage and the Base Index: At the 70 percent damage level, Bwanje (Area B) has an inventory damage return period of 1 in 40 years, but a Base Index return period of 1 in 30 years. • The return period ratio: Insured party basis risk is likely to occur in Areas C, D, F, G, H, and I at the 70 percent damage-to-payout rate (catastrophic level). The ratios for Areas C and F, however, are close enough to 1 that we can disregard them. Areas G, H, and I, however, have return period ratios that are too low to meet the risk committee’s criteria (below 70 percent). If the other metrics for these areas also do not meet the criteria, the insurance manager will ask the product design team to improve the Base Index structure for these areas. If the return period ratios and other metrics do not improve with the changes in structure, then index insurance may not be a good risk management tool for the risk in these areas. • The probability that the Base Index will not experience an insured party basis risk event: Except for Area I (74 percent), the probabilities for each area are greater than 75 percent. In other words, for each of the nine other areas, there is at least a 75 percent probability that no insured party basis risk event will occur in the next risk period. Since Area I failed to meet the risk committee criteria for both the return period ratio and this metric, the product design team should revisit the structure for this area. box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Design and Evaluation—The Base Index 45 Case Example Box 4CB.9  Product Evaluation Decision Metrics for the Base Index (continued) • The expected value and TVaR for insured party basis risk as a percentage of the portfolio value: On average, we expect this portfolio to suffer a 3 percent insured party basis risk loss, but the insured party basis risk for a 1-in-20-year event could be as high as 16 percent of the portfolio (TVaR). • Historical years with largest insured party basis risk ratios: The most recent year in which the Base Index would have provided an insufficient payout because of product design basis risk was 2011 (Area A). In this year, the Base Index would have triggered only a 3 percent payout, but the underlying data for this year (not shown in the table) show that the actual loss to the policyholder would have been 30 percent. With regard to the insurer basis risk, Lindiwe observes the following: • The return period ratio: Insurer basis risk is projected in Areas A, B, E, and J at the 70 percent damage-to- payout level. However, since Area E’s ratio of 1.08 is very close to 1, we can disregard it. Only Areas B and J have ratios greater than the risk committee’s guideline of 1.2. • The probability that the Base Index will not experience an insurer basis risk event: The probability for Area J (72 percent) is less than 75 percent—the risk committee’s cut-off. • The expected value and TVaR for insurer basis risk as a percentage of the portfolio value: The projected value  and TVaR of insurer product design basis risk are 2 percent and 13 percent of portfolio value, respectively, both of which are within the committee guidelines. • Historical years with largest insurer basis risk ratios: In 1997, the policyholder would have lost about 50 percent in Area H, but the Base Index would have paid 65 percent. Based on this analysis, Lindiwe completes the product evaluation for the Base Index as in case example box table 4CB.9.1. Lindiwe and her manager, Ghassimu, determine that the index needs restructuring for Areas B, I, and J. Case Example Box Table 4CB.9.1 Base Index Product Evaluation Summary Basis risk evaluation for the Base Index Decision metric Insured party basis risk Insurer basis risk Projected return period for the inventory Requirement is not fulfilled for Bwanje (Area B; 1 in 40 years versus 1 damage and the Base Index in 30 years at 70 percent damage and payout levels) Return period ratio Requirement is not fulfilled for Areas B, G, H, I, and J at the 70 percent damage-to-payout level Probability that the Base Index will not Requirement fulfilled except for Requirement fulfilled except for experience a basis risk event Area I Area J Projected value for basis risk as a percentage Requirement fulfilled Requirement fulfilled of the portfolio value TVaR at 95 percent for risk as a percentage of Requirement fulfilled Requirement fulfilled the portfolio value Decision Present Base Index to policyholder Restructure index for specific ¸ Areas B, I, and J areas Consider alternative solutions (non-index) Ghassimu notes that, before launching a product based on this Base Index, Excellence Insurance will need to decide how it will manage insured party basis risk events. Excellence will also inform the regulator box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 46 Product Design and Evaluation—The Base Index Case Example Box 4CB.9  Product Evaluation Decision Metrics for the Base Index (continued) of its strategy for managing basis risk events so that it can be evaluated against consumer protection guidelines. As discussed in part 2 of this guide, an important consideration when interpreting the results for product design basis risk is what data and assumptions were used. For example, in this case example 30 years of historical data are used to estimate the return period ratio. Only some of those years experienced weather that would have triggered payout amounts. Therefore, the return-period ratios ­ are based on relatively few observations (fewer than 30). As a result, it is not certain that the very low return period ratio for Bwanje (Area B) is definitely an indication of product value. Instead, the data may include outliers that are causing this result, and in fact the product may work very well for Bwanje. All of the product evaluation results need to be interpreted taking into account the data and assumptions used. Note: TVaR = tail value at risk. 4.3.8  STEP 8: Document and Communicate Business Decision Once the insurer has a Base Index that meets its internal guidelines for evaluating basis risk, the insurance manager and actuarial analyst present the Base Index to the policyholder and explain the basis risk evaluation. They should explain each metric so that the policyholder clearly understands the Base Index’s strengths and weaknesses. Even though the Base Index meets the insurer’s internal guidelines for basis risk, the policyholder may not be satisfied with the product. In this case, the product design team must review the product to identify changes to the struc- ture (for example, the addition of new proxies) that will improve the Base Index’s basis risk evaluation. The sample of historical insured party basis risk events can be especially help- ful for this conversation. For each area, the insurance manager can refer to the year in which an insured party basis risk event occurred and compare the histori- cal damage level with the payout triggered by the Base Index. The insurance manager must explain that, had the product been in place in that year, the pay- out would have been less than the value of the damage caused by the named peril because of product design basis risk. Only once the insurer presents a Base Index that meets the policyholder’s expectations for basis risk can the product design team move on to pricing the Base Index (chapter 5). As highlighted earlier, basis risk can never be eliminated, only minimized. It is important for the insurer to clearly explain to the policyholder how the basis risk will be managed. Because of limited data points per geographical area, area-spe- cific metrics will be subject to higher uncertainty and may be biased; therefore, we advise using them with extra care. Critical decisions should be based on portfolio- level metrics and statistics that will have less uncertainty. This caveat applies to all metrics covered in this and all subsequent chapters. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Design and Evaluation—The Base Index 47 Bibliography Crouhy, M., D. Galai, and R. Mark. 2006. The Essentials of Risk Management. New York: McGraw-Hill. Dick, W., A. Stoppa, J. Anderson, E. Coleman, and F. Rispoli. 2011. Weather Index–Based Insurance in Agricultural Development: A Technical Guide. Rome: IFAD. Lam, J. 2003. Enterprise Risk Management: From Incentives to Controls. Hoboken, NJ: Wiley. Morsink, K., D. Clarke, and S. Mapfumo. 2016. “How to Measure Whether Index Insurance Provides Reliable Protection.” Policy Research Working Paper 7744, World Bank, Washington, DC. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 5 Product Pricing—The Base Index 5.1 Introduction At this stage in the pilot phase, the insurer has designed and evaluated the Base Index (chapter 4). Now the insurance manager needs to determine the price for the Base Index. Many policyholders who have risk exposure in mul- tiple geographical areas will want to purchase an insurance product with a single premium rate across the different areas (case example box 5CB.1). These products, called portfolio-priced products, must account for the risk profiles in each area, the correlations in risk between all the areas, and the value insured in each area. This chapter explains the process for determining the price for a portfolio- priced Base Index under three scenarios: (1) the policy is not reinsured, (2) the policy is reinsured through proportional reinsurance only, and (3) the policy is reinsured through a combination of nonproportional reinsurance and propor- tional reinsurance. Each geographical area covered by a portfolio-priced product in reality has a different risk profile corresponding to a different premium rate. Therefore, the policyholder may want to know the specific premium rates for each geographical area in the portfolio, called the equitable premium rates. The policyholder may find these risk ratings useful in making future decisions about lending in specific geographical areas. This chapter also explains the process for providing the ­equitable premium rate for each geographical area. The key managerial questions answered during Base Index product pricing are the following: • What portfolio-priced premium rate for the Base Index meets the profit ­ objectives and risk tolerance of the insurer when –– The policy has no reinsurance? –– The policy is reinsured through proportional reinsurance? –– The policy is reinsured through nonproportional reinsurance, or a combina- tion of nonproportional and proportional reinsurance? Risk Modeling for Appraising Named Peril Index Insurance Products   49   http://dx.doi.org/10.1596/978-1-4648-1048-0 50 Product Pricing—The Base Index Case Example Box 5CB.1 Mass Bank Loan Portfolio Mass Bank has made loans to maize farmers for the purchase of inputs in 10 geographical areas, including Bwanje (Area B). The farmers will repay the loans with their earnings from crop sales at the end of the season. The total loan amount for each area is the sum of the loans to each farmer in that area. These total loan amounts vary from $140,160 in Area A to $2,252,250 in Area E. Mass Bank has requested a product to cover the farmers’ crops against drought, because drought damage is a main reason that farmers fail to repay the loans. Mass Bank would like the product to have one premium rate for all 10 geographical areas. • What profit margins, combined ratios, and economic value added could the insurance firm expect under different premium rates? • What is the equitable premium rate (that is, the risk-based premium rate) for each geographical area that makes up the policyholder’s portfolio? 5.2 Outline of Emerging Managerial and Process Controls To address the key questions related to product pricing, we recommend the decision-making processes described below and summarized in figure 5.1. Chapter 12 provides a step-by-step guide to using the probabilistic models that produce the values for the decision metrics discussed. 5.2.1  Portfolio Product Pricing—No Reinsurance 5.2.1.1  STEP 1: Determine Key Model Inputs and Assumptions Before the modeling and pricing analysis process begins, the insurance manager and the analyst agree on the inputs into the model for the specific product. These inputs are assumptions based on data from the prospective policyholder, data from the insurer, and data from the product design team (case example box 5CB.2). The insurance manager and the analyst determine the following inputs: Internal data from the policyholder • Number of insured units per area • Average sum insured per unit per area ($) Internal data from the insurer • Starting fund value ($) • Expense loading (as a percentage of premiums) • Target profit margin (percent) • Required return on capital (percent) Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Pricing—The Base Index 51 Figure 5.1 Portfolio Product Pricing Managerial Decision Process—No Reinsurance Step 1: Determine key model inputs and assumptions (the policy has no reinsurance) Data from policyholder Data from insurer Data from product design team • Starting fund value ($) • Number of insured units per area • Total expense costs (as % of • Average sum insured per unit per premiums) • Historical payouts (section 4.3) area ($) • Target profit margin (%) • Required return on capital (%) Step 2: Evaluate key managerial decision metrics according to risk management guidelines Value creation and protection Risk tolerance Risk appetite • Probability of fund ruin • Economic value added • Probability of negative profit • TVaR of projected losses • Sharpe ratio • Combined ratio • Probability of profit below target profit margin Step 3: Document and communicate business decision • If premium rate does not meet risk management guideline criteria, evaluate premium rates using proportional reinsurance. Repeat step 1 and 2 with additional data. Note: TVaR = tail value at risk. Case Example Box 5CB.2 Pricing Model Inputs for the Base Index—No Reinsurance Internal data from Mass Bank Mass Bank provides information on the number of loans in each geographical area (­ number of insured units) and the average loan size (average sum insured per unit) as shown in the table in this box. The total loan portfolio—the total loan book that has been approved and for which individual contracts have been signed between Mass Bank and the individual farmers—is $8 million. Internal data from Excellence Insurance The starting fund value refers to the policyholder funds for this class that are available at the start of the season and can be used to pay claims. The value of this fund is equal to the total premiums received during all previous periods less claims paid from this fund during all previous periods. Excellence’s accounting department reports that the starting fund value is $50,000. Because this is the first time the product will be offered and no premiums have yet been collected, the starting fund is the amount that Excellence has decided to invest in the new product. The Excellence accounting department estimates that the acquisition, general, and administrative costs related to all activities for this product (expense costs) are equal to 15 percent of the gross premium. Excellence management has indicated that a profit margin of 10 percent is required to meet ­ box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 52 Product Pricing—The Base Index Case Example Box 5CB.2  Pricing Model Inputs for the Base Index—No Reinsurance (continued) profit objectives. Management also reports that the required return on capital—the return that share- holders require to keep their capital in this business line—is 5 percent. The required return on capital is equal to the difference between the expected return to shareholders from the business line and the return currently realized from the liquid assets in which the capital funds are invested. Data from the product design team The historical payouts are provided in case example box 4CB.7. Ghassimu, the insurance manager, decides to evaluate portfolio premium rates between 3 percent and 12 percent. Lindiwe, the actuarial analyst, uses these inputs to calculate the values of the different pricing-related metrics for the Base Index for the Mass Bank portfolio. Data from the product design team • Historical payouts (section 4.3) The insurance manager also selects the portfolio premium rates to evaluate. After the analysis and modeling process, the insurance manager must ensure that the inputs used in the modeling were indeed the agreed-on ­ values. In cases in which the analyst has deviated from these values—for example, by reducing the expense cost or profit margin to produce a lower price—he or she must provide full explanations for the changes and provide sources for the values used. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Pricing—The Base Index 53 5.2.1.2  STEP 2: Evaluate Key Managerial Decision Metrics Based on the agreed-on inputs, the actuarial analyst produces several metrics to be used by the insurance manager to better understand the product’s perfor- mance under a variety of premium rates. The values of these metrics will help the insurance manager identify the premium rates for the product that will meet the insurer’s profit objectives and risk tolerance. For each portfolio premium rate, the actuarial analyst calculates the following metrics: • Projected losses • Projected combined ratio (percent) • Projected loss ratio (percent) • Projected profit margin (percent) • Probability of fund ruin (percent) • Probability of negative profit (percent) • Probability of profit below target profit margin (percent) • Economic value added (percent; shareholder value) • Sharpe ratio. The insurance manager evaluates these metrics against guidelines set by the insurer’s risk committee (see template in table 5.1 and sample guidelines in case example box 5CB.3), which should indicate the acceptable range of values for each metric. When a product’s metrics fall outside of this range, the insurance manager and analyst should consider either a higher premium rate or the effect of reinsurance options. When evaluating each portfolio premium rate, the insurance manager follows a hierarchical evaluation structure in which the manager first evaluates a rate by looking at the value creation and protection measures. If these are satisfied, Table 5.1 Template for Risk Management Committee Guidelines on Index Product Pricing Decision metrics Risk management committee guidelines 1. Value creation and protection   Economic value added   Sharpe ratio   Combined ratio (projected loss ratio + total expense costs) Indicative decision 2. Risk tolerance   Probability of fund ruin   Probability of negative profit   Probability of profit below target profit margin Indicative decision 3. Risk appetite   TVaR of projected losses Note: TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 54 Product Pricing—The Base Index Case Example Box 5CB.3 Excellence Insurance Risk Management Committee Guidelines for Index Product Pricing Decision metrics Risk management committee guidelines 1. Value creation and protection   Economic value added Must be greater than 0 percent unless supporting business and reputational benefits can offset the loss   Sharpe ratio Must be greater than 0 unless supporting business and reputational benefits can offset the loss   Combined ratio (projected loss ratio + total Must be less than 100 percent expense costs) Indicative decision 2. Risk tolerance   Probability of fund ruin Must be less than 2 percent   Probability of negative profit Must be less than 25 percent   Probability of profit below target profit margin Must be less than 25 percent Indicative decision 3. Risk appetite   TVaR of projected losses TVaR net reinsurance must be less than $200,000 Note: TVaR = tail value at risk. the manager can then look at the risk tolerance measures before finally ­ evaluating the risk appetite measures (see case example boxes 5CB.4 and 5CB.5). When a premium rate fails the higher criteria, it should be eliminated from consideration unless it can be shown that the policyholder is also bringing in supporting business that has positive value creation and protection and has an overall ­ portfolio that is profitable to the insurer. ­ In general, a good portfolio premium rate will have a low probability of fund ruin, a low probability of negative profit, and a low probability of profit below the target profit margin. In addition, it will have a positive economic value added. It is important to note that this model does not take into account that with higher premium rates, the demand for the product may be lower. The level of demand will not affect the projected loss ratio or projected profit ratio, which are relative values, but it will affect the projected payouts, which is an absolute value. Client price sensitivity is taken up in chapter 14. 5.2.1.3  STEP 3: Document and Communicate the Business Decision The insurance manager uses the risk committee guidelines for index product pricing to document the hierarchical evaluation of the premium rates. If a pre- mium rate or rates meet all the criteria, the insurance manager lists the pre- mium rates and explains the projected impact on the insurer’s profit margins Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Pricing—The Base Index 55 Case Example Box 5CB.4 Product Model Outputs for Base Index—No Reinsurance Note: EVA = economic value added; TVaR = tail value at risk. Case Example Box 5CB.5 Pricing Decisions for Base Index—No Reinsurance On the basis of the company guidelines, Ghassimu and Lindiwe agree on the following: Managerial and actuarial decisions (indicate Decision metrics minimum acceptable premium rate) 1. Value creation and protection   Economic value added 11 percent   Sharpe ratio 10 percent   Combined ratio (projected loss ratio + total 10 percent expense costs) Indicative decision 10 percent 2. Risk tolerance   Probability of fund ruin Not met by any rate considered   Probability of negative profit Not met by any rate considered   Probability of profit below target profit margin Not met by any rate considered Indicative decision Consider reinsurance options 3. Risk appetite   TVaR of projected losses Consider reinsurance options Final decision Write Do not write Consider next reinsurance scenario ¸ or other risk management tools Note: TVaR = tail value at risk. and risk exposure. If the premium rates do not meet the criteria, the insurance manager and the actuarial analyst move on to evaluate the same premium rates under the second scenario—the policy is reinsured through proportional reinsurance only. 5.2.2  Portfolio Product Pricing—Proportional Reinsurance Only To address the key questions related to product pricing with only propor- tional reinsurance, we recommend the decision-making process summarized in figure 5.2. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 56 Product Pricing—The Base Index Figure 5.2 Portfolio Product Pricing Managerial Decision Process—Proportional Reinsurance Only Step 1: Determine key model inputs and assumptions (policy with proportional reinsurance) Data from policyholder Data from insurer Data from product design team Same as in figure 5.1 plus • Same as in figure 5.1 • Same as in figure 5.1 • Percentage ceded to reinsurer Step 2: Evalute key managerial decision metrics according to risk mangement guidelines set Value creation and protection Risk tolerance Risk appetite • Same as in figure 5.1 • Same as in figure 5.1 • Same as in figure 5.1 Step 3: Document and communicate business decision • If premium rate does not meet risk management guideline criteria, evaluate premium rates using proportional and nonproportional reinsurance. 5.2.2.1  STEP 1: Determine Key Model Inputs and Assumptions The insurance manager and the analyst use the same inputs into the model as for the analysis with no reinsurance (section 5.2.1), but add percentage ceded to the reinsurer as an input parameter (case example box 5CB.6). This guide does not provide a discussion of the mechanism for determin- ing the parameters for reinsurance arrangements, whether for proportional or nonproportional reinsurance. For further information on this topic, see Cass et al. (1997). 5.2.2.2  STEP 2: Evaluate Key Managerial Decision Metrics Based on the agreed-on inputs, the actuarial analyst again produces the product pricing metrics (case example box 5CB.7). The insurance manager evaluates these outputs against the insurer’s profit objectives and risk appetite as in ­ section 5.2.1 (case example box 5CB.8). Case Example Box 5CB.6 Pricing Model Inputs for Base Index—Proportional Reinsurance Only Ghassimu and Lindiwe decide to evaluate a proportional reinsurance policy with percent of the insured portfolio ceded to the reinsurer. 80 ­ Internal data from the insurer • Percentage ceded to the reinsurer: 80 percent All other inputs are the same as in case example box 5CB.2. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Pricing—The Base Index 57 Case Example Box 5CB.7 Pricing Model Outputs for Base Index—Proportional Reinsurance Only Note: EVA = economic value added; TVaR = tail value at risk. Case Example Box 5CB.8 Pricing Decisions for Base Index—Proportional Reinsurance Only Managerial and actuarial decisions (indicate Decision metrics minimum acceptable premium rate) 1. Value creation and protection   Economic value added 11 percent   Sharpe ratio 10 percent   Combined ratio (projected loss ratio + total 10 percent expense costs) Indicative decision 10 percent 2. Risk tolerance   Probability of fund ruin Not met by any rate considered   Probability of negative profit Not met by any rate considered   Probability of profit below target profit margin Not met by any rate considered Indicative decision Consider more reinsurance options 3. Risk appetite   TVaR of projected losses Consider combined proportional and nonproportional reinsurance options Final decision Write Do not write Consider next reinsurance scenario ¸ or other risk management tools Note: TVaR = tail value at risk. Because the insurer receives a commission that only covers its costs (that is, expense loading) for this product, adding proportional reinsurance does not change the metrics for value creation and risk tolerance. The reinsurer’s fortunes follow those of the insurer. However, the absolute value of the TVaR is proportionately reduced. 5.2.2.3  STEP 3: Document and Communicate the Business Decision The insurance manager uses the risk committee guidelines for index product ­ pricing to document the hierarchical evaluation of the premium rates. If a pre- mium rate or rates meet all the criteria, the insurance manager lists the pre- mium rates and explains the projected impact on the insurer’s profit margins and risk exposure. If the premium rates do not meet the criteria, the insurance manager and the actuarial analyst move on to evaluate premium rates using Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 58 Product Pricing—The Base Index nonproportional reinsurance, or a combination of nonproportional and propor- tional reinsurance. 5.2.3  Portfolio Product Pricing—Proportional and Nonproportional Reinsurance To address the key questions related to product pricing with proportional and nonproportional reinsurance, we recommend the decision-making process ­summarized in figure 5.3. 5.2.3.1  STEP 1: Determine Key Model Inputs and Assumptions The insurance manager and the analyst use the same inputs for the pricing model as for the analysis with proportional reinsurance only (section 5.2.2), with the addition of the following input parameters (case example box 5CB.9): Internal data from the insurer • Amount retained under nonproportional treaty ($) • Aggregate loss limit under nonproportional treaty ($) • Percentage carried by the reinsurer under nonproportional treaty (percent) • Estimated nonproportional reinsurance premium rate (percent) Figure 5.3 Portfolio Product Pricing Managerial Decision Process—Proportional and Nonproportional Reinsurance Step 1: Determine key model inputs and assumptions (policy with proportional and nonproportional reinsurance) Data from policyholder Data from insurer Data from product design team Same as in figure 5.2 plus Percentage ceded to reinsurer Amount retained under nonproportional treaty ($) • Same as in figure 5.2 Aggregate loss limit under • Same as in figure 5.2 nonproportional treaty Percentage carried by reinsurer under nonproportional treaty Estimated nonproportional reinsurance premium rate (%) Step2: Evaluate key managerial decision metrics according to risk management guidelines set Value creation and protection Risk tolerance Risk appetite • Same as in figure 5.2 • Same as in figure 5.2 • Same as in figure 5.2 Step 3: Document and communicate business decision • Index product to be recommended only if premium rate meets all risk management guideline criteria Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Pricing—The Base Index 59 For a discussion of how to determine reinsurance agreement parameters, we again refer the reader to Cass et al. (1997). 5.2.3.2  STEP 2: Evaluate Key Managerial Decision Metrics Based on the agreed-on inputs, the actuarial analyst again produces the ­ product pricing metrics. The insurance manager evaluates these outputs against Case Example Box 5CB.9 Pricing Model Inputs for Base Index—Proportional and Nonproportional Reinsurance Ghassimu and Lindiwe decide to evaluate the Base Index with pricing that combines the proportional reinsurance policy discussed in section 5.2.2 with a nonproportional reinsurance policy with the terms discussed below. Internal data from the insurer • Amount retained under nonproportional treaty ($): Excellence Insurance will retain the first  $85,000 of claim amounts per season. • Aggregate loss limit under nonproportional treaty ($): Any loss that exceeds $709,000 will be covered by Excellence. • Percentage carried by the reinsurer under nonproportional treaty: The reinsurer will pay 90 percent of the losses for any loss greater than $85,000 but less than $709,000. • Estimated nonproportional reinsurance premium rate (percent): The reinsurance will cost 5 percent of all premiums. See figure 5CB.9.1 and table 5CB.9.1. Figure 5CB.9.1  Base Index Proportional and Nonproportional Reinsurance 80% Portfolio risk ceded to reinsurer Portfolio reinsurance of 80% Total portfolio Above $709,000 retained by Excellence + $709,000 Nonproportional 10% 90% reinsurance claims claims paid by paid by reinsurer 20% Portfolio Excellence $85,000 risk retained by Below $85,000 retained by Excellence Excellence box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 60 Product Pricing—The Base Index Case Example Box 5CB.9  Pricing Model Inputs for Base Index—Proportional and Nonproportional Reinsurance (continued) Table 5CB.9.1 Reinsurance Terms All the other inputs are the same as for case example box 5CB.2 and case example box 5CB.6. the insurer’s profit objectives and risk appetite as in sections 5.2.1 and 5.2.2 (case example boxes 5CB.10 and 5CB.11). 5.2.3.3  STEP 3: Document and Communicate the Business Decision The insurance manager uses the risk management committee guidelines for index product pricing to document the hierarchical evaluation of the premium rates. If a premium rate or rates meet all the criteria, the insurance manager lists the premium rates and explains the projected impact on the insurer’s profit margins and risk exposure. If the premium rates do not meet the criteria, the insurance manager and the actuarial analyst should not recommend proceeding with the index product, or alternatively recommend that the product design team redesign the Base Index. 5.2.4 Equitable Premium Pricing As discussed in section 5.1, each geographic area covered by the portfolio-priced index product has a different risk profile (for example, less rain or more extreme maximum temperatures), which corresponds with a different premium rate. The premium rates that are specific to each area in the portfolio are called equitable premium rates (figure 5.4). Case Example Box 5CB.10 Pricing Model Outputs for Base Index—Proportional and Nonproportional Reinsurance Note: EVA = economic value added; TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Pricing—The Base Index 61 Case Example Box 5CB.11 Pricing Decisions for Mass Bank Base Index—Proportional and Nonproportional Reinsurance Actuarial and managerial decisions (indicate Decision metrics minimum acceptable premium rate) 1. Value creation and protection   Economic value added 6 percent   Sharpe ratio 6 percent   Combined ratio (projected loss ratio + total 6 percent expense costs) Indicative decision 6 percent 2. Risk tolerance   Probability of fund ruin 7 percent   Probability of negative profit 9 percent   Probability of profit below target profit margin 10 percent Indicative decision 10 percent 3. Risk appetite   TVaR of projected losses Requirement fulfilled Final decision Write ¸¸ 10 percent and above Do not write Consider other risk management tools Note: TVaR = tail value at risk. Figure 5.4 Equitable Premium Pricing Step 1: Determine key model inputs and assumptions (equitable premium pricing) Data from policyholder Data from insurer Data from product design team • Same as in figure 5.2 • Same as in figure 5.2 • Same as in figure 5.2 Step 2: Evaluate key managerial decision metrics according to risk mangement guidelines set Value creation and protection Risk tolerance Risk appetite Equitable • Same as in figure 5.2 • Same as in figure 5.2 • Same as in figure 5.2 + premium rate for each geographic area Step 3: Document and communicate business decision • Share rates with policyholder to use as reference when making related lending decisions Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 62 Product Pricing—The Base Index 5.2.4.1  STEP 1: Determine Key Model Inputs and Assumptions The insurance manager and the analyst use the same inputs into the equitable premium model as for the analysis with proportional reinsurance only section 5.2.2). No additional inputs are needed because the insurance manager (­ and actuarial analyst will just be identifying the premium rates specific to each geographical area. The client may find these risk ratings useful in making future decisions about lending in these areas. It is important to note that the final commercial price for each geographic area will not be equal to the pure equitable premium rates because the insurer adds expense and profit loading. Often clients expect to be charged a product’s pure risk premium rate. However, this approach is not possible because the insurer has to meet the costs of running the insurance fund (such as staff salaries and ­ rentals). The attractiveness of the insurance product to prospective clients is affected by the size of these extra loadings. For example, a potential policyholder would need to be very risk averse to accept loadings as high as 50 percent. 5.2.4.2  STEP 2: Evaluate Key Managerial Decision Metrics Based on the agreed-on inputs, the actuarial analyst produces equitable premi- ums for each geographical area (case example box 5CB.12). In this case, the goal of the analysis is not to find one overall premium rate that can be applied to the total portfolio of geographical areas, but to find the equitable premium for each area that takes into account each area’s specific characteristics and risks. It is important to note that the equitable premium is for the area, not for individual insured units. No attempt is made to calculate equitable premiums for the insured unit because index insurance is based on area averages. All policy- holders in a geographical area pay a single premium rate. Individual insured units within the geographical area may in fact have different risk profiles, but are all considered to be one homogeneous class. Also, the equitable premium rates are based on historical hazard or inventory damage data, so the results can be influenced by, for example, several recent years of high losses in a particular area, even though this area has the same actual risk profile as an adjacent one. Testing the significance of the differences in equitable premium rates for areas in this scenario is important, but beyond the scope of this guide. Case Example Box 5CB.12 Equitable Premiums for the Base Index Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Pricing—The Base Index 63 5.2.4.3  STEP 3: Document and Communicate the Business Decision At this stage, the insurance manager documents the equitable premium rates for each area covered by the product. The manager shares these rates with the ­ policyholder so that the policyholder can review them and keep them as a point of reference when making related lending decisions. Bibliography Cass, M. R., P. R. Kensicki, G. S. Patrik, and R. C. Reinarz. 1997. Reinsurance Practices. 2nd ed. Malvern, PA: Insurance Institute of America. Clarke, D., O. Mahul, and N. Verma. 2012. “Index Based Crop Insurance Product Design and Ratemaking: The Case of Modified NAIS in India.” Policy Research Working Paper 5985, World Bank, Washington, DC. Crouhy, M., D. Galai, and R. Mark. 2006. The Essentials of Risk Management. New York: McGraw-Hill. Harrison, C. M. 2004. Reinsurance Principles and Practices. Malvern, PA: American Institute for Chartered Property Casualty Underwriters/Insurance Institute of America. Lam, J. 2003. Enterprise Risk Management: From Incentives to Controls. Hoboken, NJ: Wiley. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 6 Product Evaluation—The Redesigned Index 6.1 Introduction At this stage in the pilot phase, the insurer has designed and evaluated the Base Index (chapter 4) and determined its price under different reinsurance arrange- ments (chapter 5). The Base Index provides the highest possible level of coverage against damage to the farmer’s inventory and is used as a point of reference for discussing product options with policyholders. Based on feedback from the policyholder on price, the insurer now designs the Redesigned Index (case example box 6CB.1) and calculates its historical payouts using the same process as in section 4.3 on the Base Index (case example box 6CB.2). Case Example Box 6CB.1 Term Sheet for Redesigned Index Based on the product pricing analysis, the Base Index has a minimum acceptable premium of 10 percent with proportional and nonproportional reinsurance. However, Mass Bank is only willing to pay for a prod- uct with a premium of 4 percent or less. Ghassimu and Lindiwe instruct the Hazard Analytics product design team to formulate a Redesigned Index with triggers, exits, and payout rates that provide less cover- age than those for the Base Index. Insured Areas A to I Area J Participating Stations A to I Station J measurement stations Target crops Maize Maize Type of insurance cover Weather index insurance that pays out a Weather index insurance that pays out a defined percentage of the total sum defined percentage of the total sum insured when the following events insured when the following events occur at participating measurement occur at participating measurement stations during the total cover period: stations during the total cover period: • A specified number of consecutive dry • A specified number of consecutive dry days OR days OR • Total rainfall less than a specified level. • Total rainfall less than a specified level. box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products   65   http://dx.doi.org/10.1596/978-1-4648-1048-0 66 Product Evaluation—The Redesigned Index Case Example Box 6CB.1  Term Sheet for Redesigned Index (continued) Insured Areas A to I Area J These measures approximate weather These measures approximate weather conditions that may cause inventory conditions that may cause inventory damage for the policyholder that, as a damage for the policyholder that, as a result, cause losses for the insured. result, cause losses for the insured. Total contract period June 20–September 17, inclusive June 20–September 17, inclusive Maximum payout The greater of Trigger 1 payout or Trigger 2 The greater of Trigger 1 payout or Trigger 2 payout (see below), up to 100 percent payout (see below), up to 100 percent of of the total sum insured the total sum insured Maximum specified 20 kilometer radius 20 kilometer radius distance Total sum insured Total loan portfolio Total loan portfolio Claim Trigger 1 Payout event definition Number of consecutive dry days Number of consecutive dry days Days immediately following one another Days immediately following one another in which the total rainfall recorded on in which the total rainfall recorded on each day is 2.5 millimeters or less. each day is 2.5 millimeters or less. Recorded rainfall is taken from the Recorded rainfall is taken from the participating measurement stations participating measurement station during the cover period for Trigger 1. during the cover period for Trigger 1. The longest consecutive dry day period The longest consecutive dry day period during the cover period is the index during the cover period is the index value, which is evaluated against the value, which is evaluated against the payout schedule. payout schedule. Cover period June 20–September 17, inclusive June 20–September 17, inclusive Payout schedule Trigger 25 consecutive days Trigger 25 consecutive days Payout rate 2.5 percent per dry day Payout rate 2.5 percent per dry day above the trigger above the trigger Exit 45 consecutive dry days Exit 45 consecutive dry days Number of payments Only one payment is allowed for this Only one payment is allowed for this allowed trigger. trigger. Timing of payment Payments due according to the definitions Payments due according to the definitions above may be made at the end of the above may be made at the end of the total contract period, that is, after total contract period, that is, after September 17. September 17. Claim Trigger 2 Payout event definition Total rainfall for flowering period Total rainfall for flowering period Total millimeters of rainfall recorded Total millimeters of rainfall recorded during the flowering period. Recorded during the flowering period. Recorded rainfall is that from the participating rainfall is taken from the participating measurement stations. For the contract measurement station. For the contract period, cumulative rainfall is obtained period, cumulative rainfall is obtained by summing daily amounts over the by summing daily amounts over the contract period. The resulting amount is contract period. The resulting amount is the index value, which is evaluated the index value, which is evaluated against the payout schedule. against the payout schedule. Cover period July 25–September 2, inclusive July 25–September 2, inclusive box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Evaluation—The Redesigned Index 67 Case Example Box 6CB.1  Term Sheet for Redesigned Index (continued) Insured Areas A to I Area J Payout schedule Trigger 75 millimeters Trigger 55 millimeters Payout rate 2 percent per millimeter Payout rate 2 percent per millimeter below trigger below trigger Exit 25 millimeters Exit 5 millimeters Number of payments Only one payment is allowed for this Only one payment is allowed for this allowed trigger. trigger. Timing of payment Payments due according to the definitions Payments due according to the definitions above may be made at the end of the above may be made at the end of the total contract period, that is, after total contract period, that is, after September 17. September 17. Case Example Box 6CB.2 Historical Payouts for Redesigned Index Lindiwe points out to Ghassimu that 1986 in Area C provides a good example of the Redesigned Index’s lower level of coverage compared with the Base Index. The historical payout value for 1986 in Area C is 10 percent for the Redesigned Index, or $8 for a sum insured of $80. For the Base Index, the historical payout value is 22.5 percent, or $18. The actuarial analyst and the insurance manager price the Redesigned Index using the same process used for the Base Index, described in chapter 5. Because the Redesigned Index provides less coverage, its premium will be lower than that for the Base Index. Now the insurance manager has two products to discuss with the client: the Base Index and the Redesigned Index. This chapter explains how to determine Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 68 Product Evaluation—The Redesigned Index and explain to the client the most important differences in level of coverage between the Base Index and the Redesigned Index. It is important to note that the three main elements of the Redesigned Index—the product design, the price, and the level of coverage—are, of course, all highly interrelated. A product design team could devise many different Redesigned Indexes based on the same Base Index. Each of these Redesigned Indexes would provide a different level of coverage and have a different price. Changing any one of these three elements—product design, product price, or product coverage—affects the other two. The key managerial questions answered during Redesigned Index product performance evaluation are the following: • What level of coverage is provided by the Redesigned Index compared with the Base Index? –– How often does the Resigned Index pay out compared with the Base Index (that is, what is the Redesigned Index’s return period compared with that of the Base Index)? –– How much less does the Redesigned Index pay compared with the Base Index (that is, what is the size of the implied deductible for the Redesigned Index)? –– In what percentage of years is there no implied deductible for the Redesigned Index? –– What are the largest differences in historical payouts between the Redesigned Index and the Base Index? • In years with especially high losses, will the Redesigned Index provide payouts? Recall that the Base Index and the Redesigned Index that is developed from it have the same level of product design basis risk. If the policyholder has ques- tions about product design basis risk for the Redesigned Index, the insurance manager should repeat the explanations covered in section 4.3 on the Base Index. 6.2 Outline of Emerging Managerial and Process Controls To address the key questions related to Redesigned Index evaluation, we recom- mend the decision-making processes described below and summarized in ­ figure 6.1. Chapter 13 provides a step-by-step guide to using the probabilistic models that produce the Redesigned Index product evaluation metrics discussed in this section. 6.3 Step 1: Determine Key Model Inputs and Assumptions Before the modeling process begins, the insurance manager and the actuarial analyst agree on the inputs into the model. These inputs are assumptions based on data from the prospective policyholder, data from the insurer, and data from the product design team (case example box 6CB.3). Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Evaluation—The Redesigned Index 69 Figure 6.1 Redesigned Index Evaluation Managerial Decision Process Step 1: Determine key model inputs and assumptions Data from product design team Data from insurer Data from policyholder • Historical payouts for Base Index • Confidence interval • “Significant payout” threshold • Historical payout for Redesigned Index Step 2: Evaluate and summarize key managerial decision metrics (output of chapter 13 ) Implied deductible metrics for use in explaining effects of choosing Redesigned Index over Base Index • Projected return periods for Base Index • Projected return periods for Redesigned Index • Return period ratios • Percentage of years when there is no implied deductible • Expected and TVaR for the Redesigned Index implied deductible as a percentage of the portfolio value • Largest differences in historical payouts between Base Index and Redesigned Index Step 3: Document and communicate business decision—explain impact of selecting Redesigned Index vs. Base Index Note: TVaR = tail value at risk. Case Example Box 6CB.3 Product Performance Model Inputs for the Redesigned Index Internal data from Mass Bank Mass Bank tells Ghassimu that their threshold for a significant payout is 10 percent of the sum insured. Internal data from Excellence Insurance Ghassimu and Lindiwe select a 90 percent prediction interval by setting the low at the 5th percentile and the high at the 95th percentile. This allows Excellence Insurance to understand the range of a number of key outcomes (for example, profit margin) that are likely to be observed over the next risk period with a 90 percent confidence level. Data from the product design team The product design team provides the historical payouts for the Base Index and the Redesigned Index. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 70 Product Evaluation—The Redesigned Index Internal data from the policyholder • Threshold for a significant payout. Based on our experience from pilot ­ studies, policyholders generally consider a payout that is 1.5–2 times the gross premium rate significant, that is, if the premium rate is 10 percent of the insured amount, any payout above 15–20 percent of the sum insured is “significant.” Internal data from the insurer • Prediction interval Data from the product design team • Historical payouts for the Base Index (section 4.3) • Historical payouts for the Redesigned Index (calculated as in section 4.3) 6.4 Step 2: Evaluate Key Managerial Decision Metrics Based on the agreed-on inputs, the actuarial analyst produces the values of the metrics to be used by the insurance manager to better understand the Redesigned Index’s performance compared with the Base Index. The metrics will help the insurance manager explain these differences to the policyholder so that the policyholder can make an informed decision about what product to purchase. The actuarial analyst calculates the values of the following metrics for each geographical area (case example box 6CB.4): Implied deductible metrics • Projected return periods for the Base Index • Projected return periods for the Redesigned Index • Return period ratios • Percentage of years when there is no implied deductible • Projected value and tail value at risk (TVaR) for the Redesigned Index implied deductible as a percentage of the portfolio • Largest differences in historical payouts between Base Index and Redesigned Index Because the Redesigned Index is in response to the policyholder’s request for a lower premium, the policyholder is the best party to evaluate the above metrics. As long as the Redesigned Index provides meaningful coverage to the policyholder in catastrophic years and the policyholder is clear on the limits of the coverage, the risk management committee does not need to set internal guidelines for acceptable return periods. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Evaluation—The Redesigned Index 71 Case Example Box 6CB.4 Product Performance Model Outputs for the Redesigned Index Note: TvaR = tail value at risk. • Comparison of projected return periods—Base Index versus Redesigned Index: For Bwanje (Area B), the Base Index pays a claim that is greater than 10 percent once in seven years, whereas the Redesigned Index pays such a claim once in 13 years. This means that Mass Bank will retain more risk with the Redesigned Index than with the Base Index. • Percentage of years with no implied deductible: Looking at Bwanje, 80 percent of the time the payouts for both the Base and Redesigned Indexes are equivalent. The Base Index provides higher payouts than the Redesigned Index 20 percent of the time. In other words, once in every five years, the Base Index is expected to trigger a higher payout than the Redesigned Index. The higher the percentage of years with no implied deductible, the closer the coverage of the Redesigned Index to the Base Index. • Projected implied deductible amount: If it chooses the Redesigned Index, Mass Bank will retain additional risk valued at 4 percent of the portfolio value that would be transferred to Excellence Insurance if it instead chose the Base Index. Once in every 20 years, the Redesigned Index’s implied deductible could be as high as 16 percent (tail value at risk, or TVaR). In other words, in these extreme years, losses equal to 16 percent of the portfolio value will be absorbed by Mass Bank (Redesigned Index) instead of trans- ferred to Excellence (Base Index). box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 72 Product Evaluation—The Redesigned Index Case Example Box 6CB.4  Product Performance Model Outputs for the Redesigned Index (continued) • Largest differences in historical payouts between Base Index and Redesigned Index: If both products had been in place in the past, the Redesigned Index’s largest implied deductible would have occurred in 1996 when the Base Index would have triggered a 79 percent payout, but the Redesigned Index would have triggered only a 29 percent payout. By choosing the Redesigned Index, Mass Bank would have retained an additional 50 percent in risk costs that would have been transferred away by the Base Index. 6.5 Step 3: Document and Communicate the Product Options and Business Decision The insurance manager and the actuarial analyst summarize the key managerial decision metrics for the Redesigned Index (case example box 6CB.5). The insur- ance manager must then clearly explain to the prospective policyholder and other key stakeholders the impact of selecting the Redesigned Index over the Base Index. The message to the policyholder must be unambiguous: by selecting the Redesigned Index with the lower premium, the policyholder is choosing a product with a lower level of coverage. The insurance manager should demonstrate the Redesigned Index’s lower level of coverage by explaining the metrics for the historical payouts, the return periods, and the implied deductible, and by providing specific examples of years in which the Redesigned Index produces different historical payouts than the Base Index. After a thorough discussion of the options, the policyholder decides whether to purchase the Redesigned Index. The policyholder may instead decide to pur- chase the Base Index at the higher premium level. Either way, the insurance manager documents the main discussion points and the product selected, and includes these as an appendix to the policy document. In many cases, however, the policyholder will purchase neither the Redesigned Index nor the Base Index. Instead, the policyholder will ask the insurance man- ager to provide a new product with a different premium rate—one that is more expensive than the Redesigned Index but still less expensive than the Base Index. Once this new Redesigned Index is designed and priced, the insurance manager and the actuarial analyst will repeat the managerial decision process from this section with the new product. Case Example Box 6CB.5 Outcome of Mass Bank Negotiations After in-depth discussions with Excellence Insurance, Mass Bank decides to purchase the Redesigned Index for the coming season. The team at Excellence Insurance celebrates the breakthrough. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Product Evaluation—The Redesigned Index 73 Bibliography Brehm, P. J. 2007. Enterprise Risk Analysis for Property & Liability Insurance Companies: A Practical Guide to Standard Models and Emerging Solutions. New York: Guy Carpenter. Crouhy, M., D. Galai, and R. Mark. 2006. The Essentials of Risk Management. New York: McGraw-Hill. Lam, J. 2003. Enterprise Risk Management: From Incentives to Controls. Hoboken, NJ: Wiley. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 7 Detailed Market Analysis 7.1 Introduction After completing the pilot phase (chapters 4, 5, and 6), the insurer now has the necessary information with which to complete a detailed analysis of the broader market for index insurance. Even with the necessary prerequisites in place (chapter 3) and a successfully implemented pilot, a market may still lack impor- tant characteristics for a named peril index insurance business line to be profit- able and sustainable. The key managerial questions that are addressed during the detailed market analysis are the following: • Which market segments (rural banks, microfinance institutions [MFIs], seed companies, agribusinesses, and nongovernmental organizations [NGOs]) pro- vide the highest projected volumes and profit for the investment of the insur- er’s resources and should therefore be prioritized? • For which market segments should the insurer pursue a full business case? • Which premium rates for each prioritized market segment meet the target policyholders’ price requirements and the insurer’s profit and risk profile? For the market analysis, the insurer designs and prices a Base Index and a set of Redesigned Indexes to test the coverage and price combinations preferred by spe- cific market segments. The insurer can use information gleaned from the pilot phase—such as characteristics of agricultural lending portfolios and typical yields for specific areas—as well as additional research to design these prototype products. The design, evaluation, and pricing process for these products is the same as for the pilot phase, but these indexes are now referred to as prototype products. The Base and Redesigned Prototypes must meet the risk management committee’s guidelines for product quality (basis risk metrics for the Base Prototype and prod- uct performance metrics for the Redesigned Prototypes) and product pricing. The insurer shares the products with a number of key players in each market segment, clearly indicating the return periods, premium rates, basis risk evalua- tion (Base Prototype), and implied deductibles (Redesigned Prototypes). Based Risk Modeling for Appraising Named Peril Index Insurance Products   75   http://dx.doi.org/10.1596/978-1-4648-1048-0 76 Detailed Market Analysis on feedback from these discussions, the insurer identifies the preferred products, if any, for each market segment. Next, the insurer uses information about each market segment to determine whether they represent sufficient potential business volumes to meet the insur- er’s profit objectives and risk tolerance. One issue that the insurer should keep in mind is the effect of liquidity on aggregators’ ability to pay for insurance (case example box 7CB.1). Most named peril index product sales occur through bundling with financial products such as input finance. To estimate future sales in this case, the insurer needs to fully understand and evaluate the availability of liquid resources for both input financ- ing and premium payments. A lack of liquidity in a market can prevent the expansion of products like named peril index insurance. 7.2 Outline of Emerging Managerial and Process Controls To address the key questions related to the detailed market analysis, we recom- mend the decision-making processes described below and summarized in figure 7.1. Chapter 14 provides a step-by-step guide to using the probabilistic models that produce the decision metrics discussed. Case Example Box 7CB.1 The Effect of Client Liquidity on Excellence Insurance’s Premium Volumes Progressive Agriculture has been operating a contract-farming business in Bwanje for 10 years. Progressive supplies inputs to smallholder farmers, provides them with extension services, and buys their maize produce at the end of the season. Most farmers in this pro- gram are very happy with Progressive Agriculture because it pays prices that are above market rates. In the past, Progressive Agriculture has used its own resources to provide input loans to the farmers, but with its expanding reach it no longer has the financial capac- ity to provide the necessary loans. Progressive approaches ABC Bank, a competitor to Mass Bank, to provide financing for its input program. ABC Bank agrees to provide $10 million in financing to Progressive Agriculture. The loan amounts per farmer differ by area. If repayment rates are high in the first year, ABC Bank will increase the total financing by 20 percent every year, in line with Progressive Agriculture’s expansion plans. In the past, farmers’ repayment rates to Progressive Agriculture have always been higher than 90 percent, except in those years when farmers were affected by drought. ABC Bank insists on making drought insurance part of the financing package. Progressive Agriculture purchases an index product from Excellence Insurance to cover the $10 million in financing, greatly increasing Excellence’s index insurance premium volumes. It is important to note that Excellence Insurance’s business volume from Progressive Agriculture is dependent on ABC Bank or another bank continuing to finance this program. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Detailed Market Analysis 77 Figure 7.1  Detailed Market Analysis Managerial Decision Process Step 1: Determine key model inputs and assumptions Data from insurer External research on the market • Target loss ratio (%) • Number of firms by market segment • Target profit margin (%) • Modal portfolio size by market segment ($) • Required return on capital (%) • Most popular prototype for each market segment • Risk-free rate (%) • Estimated uptake by prototype and market segment • Expense loading (%) (number of firms) Step 2: Evaluate key managerial decision metrics according to risk management guidelines (output of chapter 14) Value creation/protection Risk tolerance Growth target • Economic value added • Probability of negative profit • Expected premium volume • Sharpe ratio (%) • Probability of profits below target • Combined ratio profit margin Step 3: Document and communicate business decision • Identify priority market segments, target premium rates for each segment, projected impact on insurer profit margin and risk exposure 7.2.1  STEP 1: Determine Key Model Inputs and Assumptions Before the modeling and pricing analysis process begins, the insurance manager and the analyst agree on the inputs into the model for the specific product. These inputs are assumptions based on data from the insurer and external research on the specific market. To perform this detailed market analysis, the insurance manager and the ana- lyst determine the following inputs (case example box 7CB.2): Internal data from the insurer • Target loss ratio (percent): This is based on the loss ratio of a successful prod- uct from the pilot phase. The specific values will be the 25th percentile (mini- mum), projected value (expected), and the 75th percentile (maximum) of the loss ratio for this product. • Target profit margin (percent) Case Example Box 7CB.2 Market Analysis Model Inputs for Prototypes Excellence Insurance was pleased with the results of the pilot phase and got the go-ahead from its board to do a detailed market analysis for a commercial launch. Based on the experience during the pilot phase, Ghassimu and Lindiwe work with the Hazard Analytics product design team to design three prototype products—the Base Prototype, which has a premium rate of 10 percent; the Redesigned Prototype 1, which has a premium rate of 6 percent; and the Redesigned Prototype 2, which has a premium rate of 4 percent. box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 78 Detailed Market Analysis Case Example Box 7CB.2  Market Analysis Model Inputs for Prototypes (continued) Ghassimu and Lindiwe determine the inputs in the “insurer assumptions” table for the detailed market anal- ysis for the Mapfumoland market segments. They use the loss ratio for the Redesigned Index launched during the pilot phase as the target loss ratio for this analysis. The information under “external research on the market” comes from a detailed value chain analysis that Excellence commissioned from an international ­consulting firm. Internal data from Excellence Insurance Note: MFI = microfinance institution; NGO = nongovernmental organization. External research on the market Note: MFI = microfinance institution; NGO = nongovernmental organization. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Detailed Market Analysis 79 • Required return on capital (percent) per risk period • Risk-free rate (percent) per risk period • Expense loading by market segment and portfolio size (percent) External research on the market • Number of firms in market by size and market segment • Modal portfolio size by firm and market segment • Premium rates for each product type for each market segment and size • Most popular prototype for each market segment and size • Minimum, most likely, and maximum uptake for the most popular prototype for each market segment and size 7.2.2  STEP 2: Evaluate Key Managerial Decision Metrics Instead of spreading its resources too thinly, it is important for the insurance company to focus its resources on market segments that have the highest expected premium income and satisfactory profit potential, particularly because availability of reinsurance capacity for named peril index insurance is critical. Reinsurers are interested in supporting classes of business for which there is suf- ficient volume and profit potential. Prioritizing segments to focus on is crucial to the insurer’s success in launching an index insurance business line. We suggest that the insurer prioritize the market segments into different tiers to first focus resources on the most attractive market opportunities (table 7.1; case example boxes 7CB.3–7CB.5). For each market segment, the insurance manager evaluates the preferred prototype for that segment against guidelines set by the insurer’s risk committee,1 which should indicate the acceptable range of values for each tier and metric. Based on this analysis, the insurance manager categorizes each market seg- ment into a respective tier, with Tier 1 segments receiving first priority for prod- uct launch. Some market segments will fail to meet the insurer’s minimum profit Table 7.1 Template for Risk Management Committee Guidelines on Market Segments Decision metrics Risk management committee guidelines 1. Growth target Expected premium income   Tier 1   Tier 2   Tier 3 Do not qualify 2. Value creation and protection Economic value added Sharpe ratio Combined ratio table continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 80 Detailed Market Analysis Table 7.1  Template for Risk Management Committee Guidelines on Market Segments (continued) Decision metrics Risk management committee guidelines 3. Risk tolerance Probability of negative profit Probability of profits below target profit margin 4. Overall performance Total expected premium income across all qualifying segments Case Example Box 7CB.3 Excellence Insurance Risk Management Committee Guidelines for Market Segments Decision metrics Risk management committee guidelines 1. Growth target Expected premium income   Tier 1 ≥$1million   Tier 2 ≥$500,000 and <$1million   Tier 3 ≥$250,000 and <$500,000 Do not qualify <$250,000 2. Value creation and protection Economic value added >0 percent Sharpe ratio >0 percent Combined ratio <100 percent 3. Risk tolerance Probability of negative profit <50 percent Probability of profits below target profit margin <50 percent 4. Overall performance Total expected premium income across all >$3 million qualifying segments Case Example Box 7CB.4 Market Analysis Outputs for Mapfumoland Market Segments Note: EVA = economic value added; MFI = microfinance institution; NGO = nongovernmental organization. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Case Example Box 7CB.5 Managerial and Actuarial Market Analysis Decisions for the Mapfumoland Market Segments—Redesigned Prototypes Qualifying Value creation and protection Risk tolerance premium income Projected premium Is the probability Is the probability income for each Is EVA Is Sharpe ratio Is combined of negative of profits below segment meeting all Projected criterion criterion ratio criterion profit criterion target profit value creation and premium satisfied? satisfied? satisfied? satisfied? margin satisfied? risk tolerance Tier Firm size Market segment income (YES/NO) (YES/NO) (YES/NO) (YES/NO) (YES/NO) criteria Tier 1 Large Agribusinesses $1,772,010 YES YES YES YES YES $1,772,010 Medium Agribusinesses $1,800,990 YES YES YES YES YES $1,800,990 Medium Seed companies $1,150,680 YES YES YES YES YES $1,150,680 Tier 2 Large NGOs $499,500 YES YES YES YES YES $499,500 Large Seed companies $798,680 YES YES YES YES YES $798,680 Small Agribusinesses $600,840 YES YES YES YES NO 0 Tier 3 Large Rural banks $310,020 YES YES YES YES YES $310,020 Medium MFIs $300,000 YES YES YES YES YES $300,000 Medium Rural banks $450,150 YES YES YES YES NO 0 Small Rural banks $339,984 YES YES YES YES YES $339,984 Total projected premium $6,971,864 Final decision Pursue business ¸ opportunity Defer investment in the product until market conditions improve Name of actuarial analyst Signature of actuarial analyst Name of insurance manager Signature of insurance manager 81 Note: EVA = economic value added; MFI = microfinance institution; NGO = nongovernmental organization. 82 Detailed Market Analysis and risk appetite guidelines, and should not be selected for investment. In some cases, no market segments will meet the guidelines, meaning that the overall market does not warrant further investment of firm resources. 7.2.3  STEP 3: Document and Communicate Business Decision At this stage, the insurance manager documents the market segments identified for prioritization (Tiers 1–3). Note 1. The model automatically calculates the decision metrics for the favorite prototype product for each segment. In the book’s case example, this will be the Base Prototype for small NGOs, medium NGOs, and large NGOs; the Redesigned Prototype 1 for small rural banks, small MFIs, small agribusinesses, medium rural banks, medium MFIs, medium agribusinesses, large rural banks, large MFIs, and large agribusiness; and the Redesigned Prototype 2 for small seed companies, medium seed companies, and large seed companies. Bibliography Brehm, P. J. 2007. Enterprise Risk Analysis for Property & Liability Insurance Companies: A Practical Guide to Standard Models and Emerging Solutions. New York: Guy Carpenter. Crouhy, M., D. Galai, and R. Mark. 2006. The Essentials of Risk Management. New York: McGraw-Hill. Lam, J. 2003. Enterprise Risk Management: From Incentives to Controls. Hoboken, NJ: Wiley. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 8 Value of Index Insurance to a Financier 8.1 Introduction Based on the results of the detailed market analysis (chapter 7), the insurer has identified the market segments it will target with specific prototype products. The next step is to attract the major players in these segments as policyholders through sales and marketing activities. This chapter looks at offering index products to a specific market segment: financiers such as commercial banks, microfinance ­ institutions, and agribusinesses that provide financing to smallholder farmers. This market segment is important for providers of index insurance because financiers’ loan books are often large, and when insured as a whole portfolio, can provide significant premium volumes. Using the tools in this chapter, the insurer can identify the market players for which the index products are most valuable. ­ The key managerial questions addressed in this chapter are as follow: • Does the named peril affect the nonperforming loan or default rates of the prospective policyholder—a financier lending to small farmers? • What is the maximum amount a rational financier will be willing to pay to protect its capital by purchasing named peril index insurance? • Is the index product commercially attractive to the financier (that is, are the costs lower than the forecast expected benefits)? Our discussions with financiers in developing countries indicate that they have limited access to capital resources and have particular difficulty raising capital after a disaster has affected their portfolio. These financiers note that in ­ the past, donor organizations helped some financiers that lent to the low-income market recapitalize, but donor funds are now less available. As a result, these financiers require new ways to protect their capital while also growing their ­lending portfolios. The standard risk mitigation strategy used by financiers to protect their port- folios is general provisioning—setting aside capital to cover the income that will Risk Modeling for Appraising Named Peril Index Insurance Products   83   http://dx.doi.org/10.1596/978-1-4648-1048-0 84 Value of Index Insurance to a Financier be lost from restructuring (extending the term of) or writing off (forgiving) loans. These restructured or written-off loans are called nonperforming loans (NPLs). This risk mitigation strategy involves a cost to the financier in the form of the capital set aside (opportunity cost) and the expenses incurred in attempting to recover the debt. Lending to smallholder farmers, however, is a relatively risky enterprise and can require strategies beyond general provisioning because of covariant risks. Perils such as droughts, floods, and tropical cyclones can result in the total loss of a smallholder farmer’s crop or livestock. If the farmer’s income comes mainly from farming activities, these loss events can lead directly to loan defaults. If a financier has lent to a large number of such farmers, the impact of the risk event on its loan portfolio—increased NPLs—can be significant. In this case, an afford- able risk transfer solution that protects the financier’s loan portfolio from adverse weather events may be an attractive product for the financier. However, if the farmer’s income comes mainly from nonfarming activities that are not affected by the same weather perils, the farmer may still be able to pay back the loans received. Such farmers generally pay back their loans weekly or biweekly rather than at the end of the season. As a result, a poor agricultural yield caused by weather events will not necessarily affect the overall default rate for the portfolio of a financier that has lent to a large number of this type of farmer. These farmers—and the financier’s portfolio—will not be very sensitive to weather perils. In this case, a financier may not be interested in a risk transfer solution for adverse weather events. To effectively sell an index product to a financier, the insurer must first deter- mine whether there is a clear link between the specific named peril and the financier’s default rates. A necessary first step in making this determination is gaining access to the financier’s historical records on defaults—both restructured and written-off loans (case example box 8CB.1). The actuarial analyst will use these data to determine the extent to which named peril index insurance can reduce the financier’s losses during years with bad weather. Finally, the actuarial Case Example Box 8CB.1  Approaching Buyer Goods Ghassimu and Lindiwe present the results of the detailed market analysis to the Excellence board and receive approval to prioritize the large agribusiness market segment with its Redesigned Prototype 1 product, which has a premium rate of 6 percent. Excellence Insurance is now targeting Buyer Goods, a leading cotton buyer and processor, as a new client for index products. Buyer Goods provides its contract farmers with $24 million worth of inputs at the start of each growing season, and recoups the cost from each farmer at the end of the season when purchasing the cotton harvest. After an initial meeting with Ghassimu and Lindiwe, Buyer Goods has agreed to ­ provide  Excellence with detailed information about its provision of in-kind advances of farming inputs to farmers in 10 geographical areas over the past 10 years, including ­repayment rates. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Value of Index Insurance to a Financier 85 analyst will identify the maximum price the financier will be willing to pay for this reduction in losses—the value of the index insurance. 8.2 Outline of Emerging Managerial and Process Controls To address the key questions related to the value of index insurance for a finan- cier, we recommend the decision-making processes described below, summarized in figure 8.1. Chapter 15 provides a step-by-step guide to using the probabilistic models that produce the decision metrics discussed. 8.2.1  STEP 1: Determine Key Model Inputs and Assumptions Before the analysis and modeling process begins, the insurance manager and the analyst agree on the inputs into the model for the prospective policyholder, the financier. These inputs are assumptions based on the internal data of the financier and data from the product design team. The insurance manager and the analyst determine the following inputs (case example box 8CB.2): Internal data from the policyholder (financier) • Target maximum annual default rate (percent; this provides an indication of the financier’s risk tolerance) • Financier’s cost of capital (percent) Figure 8.1 Value of Index Insurance Managerial Decision Process Step 1: Determine key model inputs and assumptions Data from policyholder (financier) Data from product design team • Target annual maximum default rate (%) • Cost of capital • Debt recovery expense (%) • Historical payout ratios for the Redesigned Index • Historical default rate values by geographic area (%) • Distribution of loans by geographic area (%) • Prediction interval (%) Step 2: Evaluate key managerial decision metrics for financier’s portfolio guidelines Value creation and protection Risk tolerance Risk appetite • Probability of default rate greater • Value of insurance • Probable maximum loss (TVaR) than target value Step 3: Document and communicate business decision • Document degree to which index product transfers default risk, and price the index product • Propose product to financier if price is close to value of insurance Note: TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 86 Value of Index Insurance to a Financier Case Example Box 8CB.2 Value of Index Insurance Model Inputs for Buyer Goods Ghassimu and Lindiwe decide on the inputs below for the value of index insurance analysis for Buyer Goods. • Debt recovery expense (percent of loan amount; costs incurred by financier to try to recover debt) • Historic default rates by geographic area (percent; restructures and write-offs) • Distribution of loans by geographic area (percent) • Prediction interval (percent) Data from the product design team • Historical payout ratios for the prototype product. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Value of Index Insurance to a Financier 87 8.2.2  STEP 2: Evaluate Key Managerial Decision Metrics To understand and explain the benefits of the index insurance product for the financier, the actuarial analyst evaluates the financier’s portfolio under two scenarios: • The financier’s portfolio has no insurance coverage (gross default rate) • The financier’s portfolio is covered with the insurer’s named peril index insurance prototype (net default rate) ­ By comparing the two situations, the insurance manager can determine the value of the index insurance coverage to the financier. The actuarial analyst calculates the following metrics for the financier’s portfolio: Gross default rate (without index insurance coverage) • Probability of default rate greater than target value • Expected default rate and probable maximum loss (tail value at risk [TVaR]) • Projected cost of gross default risk Net default rate (with index insurance coverage) • Probability of default rate greater than target value • Expected default rate and probable maximum loss (TVaR) • Projected cost of net default risk Value of insurance • Value of index insurance to financier (the difference between the cost of the gross default risk and the net default risk) We recommend that the insurance manager work with the client to produce guidelines for evaluating the value of index insurance and the net default rate decision metrics before the insurer begins the analysis of the value of insurance (table 8.1). The financier should base the guidelines on its own risk management policies. Using the metric “probability of net default rate greater than target value,” the insurance manager determines the degree to which the named peril affects the financier’s defaults. Here, lower output values mean that the named peril has a higher impact on the defaults. Using the metric “net default rate probable maximum loss,” the insurance manager determines whether insurance lowers defaults in the years with the worst named peril events to a level that is acceptable to the financier’s manage- ment. If the net TVaR is higher than the financier’s target default rate, the NPLs or defaults may be caused by factors other than those captured by the index. Alternatively, the index structure may need to be improved to better capture losses for the financier. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 88 Value of Index Insurance to a Financier Table 8.1 Template for Client Guidelines for Value of Index Insurance and Net NPL Decision Metrics Decision metrics Client’s guidelines 1. Value creation and protection   Value of insurance 2. Risk tolerance   Probability of net default rate greater than target value 3. Risk appetite   Net default rate probable maximum loss (TVaR) Decision Index insurance is a good solution for default risk Index insurance is not a good solution for default risk Note: NPL = nonperforming loans; TVaR = tail value at risk. Using the metric “value of insurance,” the insurance manager determines the approximate amount that the financier will be willing to pay for the named peril index insurance product. In most situations, the financier will not be willing to pay a premium rate that is much higher than the value of the insurance calcu- lated by the model.1 However, to arrive at the final premium for the product, the insurer will need to include expense and profit loading in the value of insurance. See case example boxes 8CB.3–8CB.5. 8.2.3  STEP 3: Document and Communicate the Business Decision At this point, the insurance manager documents the degree to which the named peril index product transfers the financier’s default risk and the premium rate for the product that will make it commercially attractive to the financier. Now the insurance manager returns to the pricing process completed in chapter 5. If the pricing for the prototype product is less than or equal to the Case Example Box 8CB.3  Buyer Goods Guidelines for Value of Index Insurance Decision metrics Client’s guidelines 1. Value creation and protection   Value of insurance Greater than 2 percent 2. Risk tolerance   Probability of net default rate greater than target value Less than 5 percent 3. Risk appetite   Net default rate probable maximum loss (TVaR) Less than 5 percent Decision Index insurance is a good solution for default risk Index insurance is not a good solution for default risk Note: TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Value of Index Insurance to a Financier 89 Case Example Box 8CB.4 Value of Index Insurance Model Outputs for Buyer Goods Lindiwe notes that with no insurance coverage (gross default rate), Buyer Goods’ probability of a default rate greater than the target is 59 percent, the projected default rate for a 1-in-20-year event (tail value at risk) is 7.81 percent, and the projected cost of retaining the gross default risk is 5.73 percent. With coverage using Redesigned Prototype 1 (net default rate), the probability of a default rate greater than the target value declines to 0 percent, the projected default rate for a 1-in-20-year event declines to 3.44 percent, and the projected cost of the retained default risk is 3.08 percent. The value of index insurance in this case is the difference between the cost of the gross default risk and the net default risk—2.65 percent. A premium rate of about 3 percent should be acceptable to Buyer Goods. However, the Redesigned Prototype 1 has a premium rate of 6 percent. It is not likely that Buyer Goods will be willing to purchase the product at this premium rate. Excellence Insurance may need to restructure the product and go through the evaluation process again until a good balance between coverage and a cost that is acceptable to the client is achieved. value of the insurance, the insurance manager can use the value-of-insurance metrics to present the product to the financier.2 Alternatively, the insurance manager may determine that it is not feasible to offer the product at the neces- sary rate and inform the financier of this finding. Another option is to seek collaboration with a donor organization to provide funding to cover a portion of Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 90 Value of Index Insurance to a Financier Case Example Box 8CB.5 Managerial and Actuarial Value of Index Insurance Decisions for Buyer Goods—Redesigned Prototype 1 Decision metrics Actuarial and managerial analysis 1. Value creation and protection   Value of insurance 2.65 percent 2. Risk tolerance   Probability of NPL value greater than target value 0 percent 3. Risk appetite   Net default rate probable maximum loss (TVaR) 3.44 percent Decision Index insurance is a good ¸ premium is too high) (but solution for default risk Index insurance is not a good solution for default risk Note: NPL = nonperforming loan; TVaR = tail value at risk. the cost of the index product so that it is more attractive to the financier. Through this intervention, the donor organization would help provide coverage against extreme weather events to low-income producers, but would not have to pay for all future damages. This intervention would also help develop the index insurance industry by making the products accessible to smallholder farmers. Part 1 Conclusion Part 1 of this guide provides insurance managers with a guide to index ­ insurance business line development and decision making. Launching an index insurance business line is an innovative approach to reaching new market segments in the agriculture sector. Small and semi-commercial farmers and the many service providers with whom they engage—financial institutions, input suppliers, and agribusinesses—are a large and mostly untapped market. Using the tools provided in this guide, insurers will be able to prudently navigate this new market. Notes 1. In some situations, financiers may be willing to pay more, for example, if they expect the impact of perils to increase over time, have a very high cost of capital, are other- wise very risk adverse, or will use the product as part of a customer loyalty program. 2. In some cases, the financier may still be interested in purchasing the index product even though the price for the insurance is greater than the value of insurance, for example, because of very high cost of capital for the financier, or because the financier wants to use the product in a customer loyalty program. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Value of Index Insurance to a Financier 91 Bibliography Brehm, P. J. 2007. Enterprise Risk Analysis for Property & Liability Insurance Companies: A Practical Guide to Standard Models and Emerging Solutions. New York: Guy Carpenter. Crouhy, M., D. Galai, and R. Mark. 2006. The Essentials of Risk Management. New York: McGraw-Hill. Lam, J. 2003. Enterprise Risk Management: From Incentives to Controls. Hoboken, NJ: Wiley. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 P A RT 2 Probabilistic Modeling for Insurance Analysts Part 1 of this guide (chapters 3 through 8) provides a description of the insights and decisions required for the insurer to make an informed decision to launch and expand an index insurance business line. Chapters 11 through 15 in part 2 of this guide provide a step-by-step guide to using the probabilistic models that can be used to calculate the decision metrics discussed in part 1. Chapter 9 gives an overview of how to use part 2 of the guide, including an overview of its models and helpful Monte Carlo software tools. Chapter 10 provides the reader with an overview of the main terms and tech- niques that are used to perform probabilistic analysis. The goal of this chapter is to help the reader understand the remaining chapters in part 2. Chapter 10 starts with an explanation of probabilistic analysis, the use of Monte Carlo simulation for probabilistic analysis, and the main building blocks of Monte Carlo simulation models. The chapter also discusses a variety of probability distributions, the ­ correlation of different variables, and how to incorporate expert opinion into probabilistic models. Anyone who already has significant expertise in probabilis- tic modeling and Monte Carlo simulation can skip chapter 10. Chapters 11 through 13 explain the probabilistic modeling for the pilot phase of launching an index insurance business line, which includes evaluating the Base Index for product design basis risk, pricing the Base Index, and evaluat- ing the Redesigned Index. The probabilistic calculations for each chapter include guidance on implementing the analysis in the Excel files provided on the guide ­website (https://www.indexinsuranceforum.org/). Chapter 14 details the modeling for identifying and prioritizing the market segments that have the highest projected volumes and profit compared with the investment of the insurer’s resources. Chapter 15 explains how to determine the extent to which index insurance can reduce a financier’s losses during years with high default rates caused by the Risk Modeling for Appraising Named Peril Index Insurance Products   93   http://dx.doi.org/10.1596/978-1-4648-1048-0 94 Probabilistic Modeling for Insurance Analysts  named peril as well as the maximum price the financier may be willing to pay for this reduction in losses (that is, the value of the index insurance). The insurer can use this analysis in marketing the product to specific clients. Finally, all probabilistic models have inherent assumptions, and those ­ presented in this guide are no exception. Therefore, chapter 16 explains the models’ key assumptions and discusses alternative modeling approaches that analysts can also use with index insurance products. In general, we recommend that the reader approach chapters 11–15 sequen- tially, but each chapter can also be read independently. Each chapter mentions any overlap or interdependencies between models in the relevant chapters. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 9 How to Use Part 2 9.1 Introduction When reading this guide, as well as when developing or using any probabilistic model, always be critical of what assumptions are made in the use of the data, the analysis, and the development of the model. The main, simplifying assump- tion in a model should be well articulated to all stakeholders so that they are aware of the assumptions and can decide whether the model framework needs refining. Typically, such refinements involve making changes to the existing model rather than building a new model. The probabilistic approach used for the modeling throughout this guide is not the only one available. Especially when it comes to simulating payout ratios and inventory damage ratios, alternative, retrospective approaches to modeling index insurance products can help analysts understand pricing, basis risk, and other characteristics. No single modeling approach is always the best for index insurance because of important differences across cases, including the following: • Different situations, with different dynamics and types of uncertainties. For example, in certain regions the weather patterns may change rapidly over time, while in others they may be more stable. As a result, the model may or may not need to include dynamics and parameters to reflect such changing weather. • Differences in data availability: In some regions very little or no historical loss data may be available, while in other regions reliable data may be available because the insurer has already provided coverage there for many years. Depending on the availability of reliable data, a probabilistic model could be built differently and be more or less complex. • Different modeling capabilities: Different insurers will have different capabilities in developing and using probabilistic models. Depending on the capabilities, more or less complex probabilistic models will be appropriate. • Different decisions: Depending on the specific decision to be made, different probabilistic models are needed. For example, within the approach in this Risk Modeling for Appraising Named Peril Index Insurance Products   95   http://dx.doi.org/10.1596/978-1-4648-1048-0 96 How to Use Part 2 guide a number of different models are used that relate to product pricing, evaluating basis risk, and changes in coverage levels with price. Each of these questions requires different models, although in some situations it may possible to develop one comprehensive probabilistic model to support all be ­ decision points at once. ­ The form, scope, and complexity of probabilistic models depend on many fac- tors. However, we believe that a simple model with clearly understood assump- tions and limitations is often better than an extremely complex model that is more difficult to handle. Still, keep in mind that a model that does not fully capture the key building blocks or important dynamics can be misleading if used for decision making under the belief that it is comprehensive when in fact it is not. Chapter 16 discusses in detail the main assumptions underlying the models presented in this guide and explains a few alternative probabilistic modeling approaches. Example models of these alternative probabilistic approaches are available online at http://www.indexinsuranceforum.org. 9.2 Website The website http://www.indexinsuranceforum.org provides links to important supplementary materials for this guide, including the following: • Excel files for the models discussed in chapters 11–15 • References to relevant papers and books • Links to index insurance–related websites. We recommend that all readers visit the guide website and download the various Excel files before reading chapters 11–15. ­ 9.3 Overview of Key Assumptions and Limitations of This Guide’s Models The probabilistic models used in this guide rely on specific assumptions that have important limitations. First, the models assume that the index insurance product that is evaluated is the only product that the insurer offers. For example, when calculating the Sharpe ratio (an important risk metric explained in chapter 10), only the risks and returns of the specific index insurance product are included. In reality, an insurer should not consider the Sharpe ratio of the product in isolation, but should also consider how the index insurance product will affect its overall Sharpe ratio, which includes the products it already offers. Analyzing only the index insurance product in the absence of the insurer’s entire portfolio also affects the calculation of other metrics discussed in the guide, such as the prob- ­ ertain ability of ruin and the probability that the profit margin will be less than a c target set by the insurer. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 How to Use Part 2 97 Second, the models consider only a one-year (or one-season) time frame. In other words, when estimating metrics such as the capital required or the prob- ability of ruin, the models only consider these risks over a one-year horizon. Insurers, however, should also look at these metrics over multiple years (for example, a three- or five-year horizon) to obtain a clearer picture of the longer- term performance, risks, and profitability of the product. Third, the models assume that the historical patterns related to the index are not changing significantly over time. Such changes might include an increase in the frequency of drought in a specific area or a decrease in the severity of drought in the same area. Such changes can be incorporated in a probabilistic analysis (see chapter 16), but to prevent the analysis from becoming too complex this dynamic is not incorporated. As an example of an alternative model, for sovereign programs, the World Bank Group would commission a catastrophe risk model that combines historical data ­ istorical-based with physical science to get a more accurate estimate than just a h approach. The approach of combining historical data with physical science is also taken in many mature agriculture insurance markets (for example, the AIR WORLDWIDE model in China). However, to ensure that the reader will under- stand the modeling approach, appreciate its limitations, and develop a good foundation for building or refining models that are appropriate (and potentially more complex) for any given situation, the models are kept relatively simple in this guide. Chapter 16 provides a more in-depth look at the various assumptions, limita- tions, and alternative approaches to the probabilistic models that are presented in this guide. 9.4 Monte Carlo Software Tools The models discussed in this guide and on the website use Monte Carlo1 simulation models and were developed using both Microsoft Excel and a ­ ­ commercial Excel add-in called ModelRisk. However, many different commer- cial and open-source software tools can be used to build and run Monte Carlo simulation models. 9.4.1 @RISKTM @RISK is a commercial Excel add-in for conducting Monte Carlo simulations. Using a graphical interface (point and click), users can assign distributions to variables, perform simulations, and display and inspect results. More information is available at http://www.palisade.com. 9.4.2  Crystal Ball ­ imulations. Crystal Ball is a commercial Excel add-in for performing Monte Carlo s The Crystal Ball interface is similar in appearance and functionality to @RISK. More information is available at http://www.oracle.com/us/products/­ applications​ /­crystalball/overview/index.html. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 98 How to Use Part 2 9.4.3 ModelRiskTM ModelRisk is a commercial Excel add-in for performing Monte Carlo s ­ imulations. Using a graphical interface (point and click), users can assign distributions to variables, perform simulations, and display and inspect results. More information is available at http://www.vosesoftware.com. 9.4.4 RiskSolver RiskSolver is a commercial risk analysis add-in for Excel that is built around a set of optimization functions. The RiskSolver interface is similar in appearance and functionality to @RISK and Crystal Ball but in addition draws on a wide range of optimization capabilities. The RiskSolver tools also work in online spreadsheets and with application program interfaces. More information is ­ available at http://www.solver.com. 9.4.5 R R is a free, open-source statistical analysis software system that is easily down- loaded and installed, and is operated using the R programming language to ­ perform mathematical and statistical functions. It is well suited to a wide range of statistical analysis and simulation. The language is very flexible and users can design custom functions to perform analyses, as well as download and install function packages designed by others for specific problems. A host of functions have been developed for risk analysis in R. The biggest hurdle for most new users is that because R is a general tool and does not have built-in automated outputs and sensitivity analysis features in a simple point-and-click interface, modeling must be performed by entering and running a series of commands. More infor- mation is available at http://www.r-project.org/. Note 1. Chapter 10 discusses Monte Carlo simulations in detail. For the moment, the reader can interpret it as a technique for performing a probabilistic analysis. Bibliography Bolker, B. M. 2008. Ecological Models and Data in R. Princeton, NJ: Princeton University Press. http://ms.mcmaster.ca/~bolker/emdbook/. ModelAssist. “A Free and Comprehensive Quantitative Risk Analysis Training and Reference Software.” http://www.epixanalytics.com/ModelAssist.html. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 10 Fundamentals of Probabilistic Modeling The objective of this chapter is to provide a brief and accessible introduction to probabilistic modeling, in particular Monte Carlo simulation. Readers who are already familiar with these concepts should skip to chapter 11. Probabilistic modeling is a wide and evolving field; many excellent books and other resources are available for learning about different aspects of probabilistic modeling. A few good books and resources include the following: • Bolker, B. M. 2008. Ecological Models and Data in R. Princeton, NJ: Princeton University Press. http://ms.mcmaster.ca/~bolker/emdbook/. • Cherubini, U., E. Luciano, and W. Vecchiato. 2004. Copula Methods in Finance. Hoboken, NJ: John Wiley & Sons. • Embrechts, P., F. Lindskog, and A. McNeil. 2003. “Modeling Dependence with Copulas and Applications to Risk Management.” In Handbook of Heavy Tailed Distributions in Finance, edited by S. T. Rachev, 329–84. Amsterdam: Elsevier. • Forbes, C., and M. Evans. 2010. Statistical Distributions. 4th ed. Hoboken, NJ: Wiley. http://www.wiley.com/WileyCDA/WileyTitle/productCd-0470390638​ .html. • Gelman, A. 2013. Bayesian Data Analysis. 3rd ed. Boca Raton, FL: Chapman & Hall/CRC. http://www.stat.columbia.edu/~gelman/book/. • Jewson, S., and A. Brix. 2005. Weather Derivative Valuation: The Meteorological, Statistical, Financial, and Mathematical Foundations. Cambridge: Cambridge University Press. • Law, A. M., and W. D. Kelton. 2006. Simulation Modeling and Analysis. 4th ed. New York: McGraw-Hill. • ModelAssist. “A Free and Comprehensive Quantitative Risk Analysis Training and Reference Software.” http://www.epixanalytics.com/ModelAssist.html. This chapter was written with key contributions from Dr. Kurt Rinehart, Risk and Statistical Consultant, EpiX Analytics LLC. Risk Modeling for Appraising Named Peril Index Insurance Products   99   http://dx.doi.org/10.1596/978-1-4648-1048-0 100 Fundamentals of Probabilistic Modeling • Yan, J. 2006. “Multivariate Modeling with Copulas and Engineering Applications.” In Springer Handbook of Engineering Statistics, edited by H. Pham, 973–90. London: Springer-Verlag. When discussing specific topics or techniques, we will occasionally refer the reader to these titles for more in-depth study. 10.1 The Case for Probabilistic Modeling in Index Insurance Will there be a drought next year? How many typhoons might occur? How likely is it that an index insurance product will pay out if there is a large drought or typhoon? How much capital should the insurer have on hand to cover a season of potentially high claims? When assessing an index insurance product, there are typically a number of different uncertainties about the future and the index product’s characteristics. How can we make sense of all of these uncertainties and understand the likely performance of a new index insurance product in a region of the world where weather may vary from year to year? Probabilistic modeling (or quantitative risk analysis1) can help us take into account a large variety of risks and uncertainties to develop an understanding of what to expect on average,2 as well as for the best case and worst case scenarios. As a result, probabilistic modeling is very helpful for the evaluation of index insurance products. The objective of this chapter is to describe the fundamentals of probabilistic modeling, with special emphasis on its application in the field of index insurance. In the early years of index insurance development, instead of probabilistic analysis, practitioners working in developing countries favored using the burn analysis contract valuation method. This method uses historical payout ratios as the inputs for calculating the statistics for product evaluation and pricing. The advantages of using this method are its simplicity and the client’s ability to easily relate the premium charged to previous experience. However, the burn analysis method has two main drawbacks. First, it assumes that future experience will be similar to past experience, which is not always the case in reality. If an extreme event has not occurred in the past, the burn analysis method will not produce pricing that accounts for extreme events. This limitation in the analysis results in challenges for most perils that are covered with index insurance because extreme events can and often do occur. The insurer needs to have set aside sufficient capital to manage this risk of extreme events. Second, the historical sample used for the burn analysis contract valuation method is usually very small, likely biasing the results. More information on burn analysis and other alternative methods of valuing contracts can be found in Jewson and Brix (2005). What is probabilistic modeling? A quick definition is that probabilistic model- ing is a quantitative modeling approach based on the theory of probability that provides a prediction of a range of possible outcomes with their accompanying probabilities. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 101 We consider probabilistic modeling the most appropriate and flexible method for evaluating, pricing, and understanding index insurance products because, when applied correctly, it can provide significant quantitative insights into the product and take into account the different types of risks for index insurance products. A critically important characteristic of a good probabilistic analysis is that it predicts the full range of possible future scenarios and does not under- estimate the risks. Probabilistic analysis typically forecasts a large set of possible outcomes that go beyond what has occurred in the recent past. In other words, probabilistic analysis considers scenarios that have not yet occurred. In addition, probabilistic analysis can take into account new circumstances that have not been observed in the past. For example, if a country is about to invest in flood mitigation measures such as flood walls, the probabilistic analysis can incorporate the impact of these measures on the probability and severity of flood claims. One disadvantage of the probabilistic modeling approach is that a probabilis- tic model needs to be carefully developed, a process that relies on numerous assumptions and significant input data. Obtaining valuable insights from proba- bilistic modeling therefore requires an understanding and appreciation of the mathematical techniques, main assumptions, and different data sources upon which the results rely. 10.1.1  What Is a Model? Before we discuss the probabilistic part of probabilistic modeling, let us first discuss the idea behind a model. A model is a simplified representation of reality that can help us better understand how a system works and can support more informed decision making about that system. In our daily lives, we typically focus on the most relevant factors of a complex reality or system to gain insight and make sound decisions. For example, the geographical maps that we use to find our way in the real world are “models” of the earth’s surface. There is no way to represent every rock and tree and the exact curvature of the earth at every point, but we do not really need those details to find our way. What we need is a way to represent distances and directions accurately so that we can find our way from place to place reliably. The world is full of models. A map is a model of a location. A musical score is a model of the sound of a symphony. An architectural drawing is a model of a house or other building. A company’s financial statement is a model of the finan- cial health of the company. Models come in many forms, fields, and applications. Models can also be mathematical representations of certain situations or systems. For example, in economics, models are used to provide an understanding of how changes in sup- ply or demand influence a country’s economic output. In operations research, models are used to help us understand how much inventory a retailer or manu- facturer should carry. In atmospheric science, climate models are used to simu- late the interactions of the atmosphere, oceans, land, and ice to determine the potential influence of carbon on future climates. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 102 Fundamentals of Probabilistic Modeling What unites these models is that, by definition, they all include simplifica- tions, selections of what is and is not included, and assumptions. The best models strike just the right balance between simplicity and realistic behavior. Including too much detail in a model results in longer development and run times and may obscure the insight we hope to gain. The best models make realistic assumptions and document them clearly for the users of the model. Some assumptions are implicit in the mathematical tools themselves, while others are necessary to simplify the problem so that an effi- cient, useful model can be built. In the end, whatever type of model is developed of whatever complexity, it is critically important that the assumptions and limita- tions of the model are clear so that decisions based on model outputs take them into account. We emphasize this point because all too often decision makers that use models to inform and support their decisions come to “believe” in the model outputs without much critical review. Models also often contain submodels. Probabilistic models typically com- prise a mathematical description of a system in which multiple components of the system are uncertain or affected by chance. For example, an insurance product may cover multiple regions where the future weather of each of these regions will be uncertain and will vary. A model for this index insurance product brings together many submodels for all of the constituent variables to account for the future weather in each area. Another way to think about it is to start at the high level and work to the lower levels. At the high level, we intend to model insurance payout amounts for all covered units in all regions. To create such a model, we have to create submodels for drought frequency and drought severity for each of the regions. Each of these sub- models, in turn, includes models for their parameters, such as the probability of a drought in any given year. A probabilistic model can include a large number of submodels. 10.1.2  Deterministic versus Probabilistic Models Before we get into how probabilistic models work and how to interpret their results, let us look at what they are not: they are not deterministic. In elementary school, we all learned that 2 + 2 = 4. This is an example of a deterministic calculation that provides a simple, invariable answer. Adding 2 and 2 always gives 4, and multiplying any number by 2 always gives twice that num- ber. There is no gray area here. We know this outcome exactly and precisely. With deterministic models, the assumption is that the inputs and output are perfectly known. These models can be very helpful for examining some situations, but they do not provide insight into how likely some outcomes are to occur, that is, how much uncertainty is associated with a particular outcome.3 An example of a deterministic analysis for index insurance would be to calcu- late how often (and how much) an index product would have paid out over the past 30 years, and use these data to evaluate the product’s performance. If this product’s highest payment in the past 30 years was 40 percent of the insured amount, will the maximum possible payout for next year also be 40 percent? Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 103 Even when we have data from 30 years, it is possible for future payouts to be higher than any that have occurred in the past. But how likely is it that future payouts will be higher? And what if the insurance company has policies in mul- tiple regions with different weather profiles? In contrast to deterministic models, probabilistic models take into account variability and uncertainty, both in the inputs and in the outputs. When we have a coin and ask ourselves how many times it will show tails in the next 10 tosses, there is no single answer because the results are affected by chance. Any answer between 0 and 10 is possible, and each of these outcomes has its own probability. Determining the outcomes of coin tossing and the accompanying probabilities is a problem a probabilistic model can help us with. The outcome in this case is conceptually very different from the outcome of a deterministic model. The output of the deterministic model is fully determined by the parameter values, which are assumed to be known, and the initial conditions. In contrast, probabi- listic models incorporate the fact that we are typically uncertain about what the future may hold. Therefore, the output of a probabilistic model shows the range of possible outcomes and includes the probabilities for each of the different outcomes. Two specific types of uncertainty are important when quantifying risk using a probabilistic model: 1. Variability (also called aleatory uncertainty, secondary uncertainty, stochastic variability, or interindividual variability): This uncertainty results from chance (that is, randomness) and is a function of the system being modeled. Two examples are the variability in the cumulative rainfall or the number of hours of sunshine per year. 2. Parameter uncertainty (also called epistemic uncertainty, primary uncer- tainty, or fundamental uncertainty) is a characteristic of measurement inac- curacy or the analyst’s incomplete understanding of the phenomena (that is, level of ignorance). For example, we may be uncertain about the true average rainfall level in a particular area or the true annual probability of a claim. Unlike variability, parameter uncertainty decreases as we collect more data or expand our knowledge. We discuss this type of uncertainty more in section 10.2.2.2. Both types of uncertainty need to be included in probabilistic models for index insurance because both affect the model’s forecast of the product’s perfor- mance, profitability, and so on. For example, the year-to-year variability in the payout ratios, driven by variability in annual weather, must be taken into account to understand risk metrics such as the potential magnitude of annual payouts in certain years. In addition, our epistemic uncertainty (that is, ignorance or lack of knowledge) about the true average annual rainfall in a certain area may be con- siderable, especially if we have limited historical weather data or weather pat- terns are changing, and needs to be taken into account. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 104 Fundamentals of Probabilistic Modeling 10.1.3  What Is a Probability Distribution? Evaluating and pricing of index insurance products require an analysis and appraisal of what the future may hold. None of us knows exactly what will happen in the future, but we can assume that some results are more probable than others. In a classic demonstration of probability, if a fair coin (a coin with an equal probability of landing heads or tails for each throw) is flipped 100 times, it is extremely unlikely that the result will be 100 heads and 0 tails. It is much more likely that the result will be closer to equal amounts of heads and tails. This outcome can be demonstrated by actually flipping a coin 100 times and summarizing the results. A probability distribution is a mathematical expression of the chances of observing various outcomes from a specific situation or experiment. One type of probability distribution, a binomial distribution, allows us to build a simple, quantitative risk model of the coin flip problem. Binomial distributions show the results of multiple events or trials that each have an outcome that can take one of two values (for example, success or failure).4 Suppose you are asked to predict the number of times a coin will land head side up out of 10 flips. The result could be any number from 0 to 10, but not all of these outcomes are equally likely. With this model, we can identify the most likely outcome (5 heads) and the least likely outcomes (0 heads and 10 heads). We can also determine the probability of seeing at least 5 heads, more than 8 heads, either 5 or 6 heads, and so forth. In an insurance context, a probability distribution can compute figures such as the average annual claims for a specific product or the probability that claims will exceed a specific amount. Because probability distributions provide an understanding of the probability of various outcomes, they are central to index insurance product evaluation and pricing. They provide a way to gain an under- standing of what might happen, how large the risk might be, and how much required capital the insurer needs. There are hundreds of different probability distributions5 and most probabi- listic models combine many different types to represent various uncertainties. The choice of probability distribution depends on the type of uncertainty. In fact, one of the fundamental tasks of probability modeling is to bring together the appropriate probability distributions to stand in for the most important elements of the problem. 10.1.4  What Is Monte Carlo Simulation? Real-life problems that we wish to model often cannot be summarized as one simple mathematical equation or one probability distribution. For example, sup- pose that for an index insurance product, we need to forecast the probability that a given area will experience drought during the next season or year. One way of looking at this is that we have two possible outcomes: drought or no drought. Because a Bernoulli distribution6 can represent the result of a single event that has two possible outcomes, we can easily use it to solve this problem. However, the Bernoulli distribution only tells us whether drought will or will not occur. It tells us nothing about the severity of the drought. To model drought severity, which will determine the payout amount for our index insurance product, we Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 105 need another distribution (see section 10.2.1.2). Not surprisingly, the more com- plexity in the model (for example, more probability distributions), the more difficult the calculations become. What if we also want to estimate total payouts for the following year for 10 areas, each with weather patterns that are different but also somewhat related because the areas are close together? The model for estimating the total payouts will need to include even more distributions now. At some point, calculating the probabilities for complex, real-life problems becomes mathematically too diffi- cult using probability calculations. This is where Monte Carlo simulation pro- vides a solution. Monte Carlo simulation (or probabilistic simulation, or sometimes just simu- lation; box 10.1) is a way of running many “experiments” and then looking at a summary of the observed outcomes. It is a computerized mathematical tech- nique for generating a range of possible outcomes that also provides the associ- ated probabilities. In Monte Carlo simulation, we use a computer to “roll the dice” according to the probability distributions in our model and then report back on the outcome. It is a formal way of looking at many possible scenarios arising from a system with various risks and uncertainties and summarizing the results (box 10.2). We often use the term “draw” to mean a random value resulting from a given probability distribution. This evokes the image of drawing a card from a random- ized deck. So a random draw is a realized value that results from a random process. Each scenario or iteration of the Monte Carlo simulation depends on random draws from all the distributions from which the probability model is built. Some of the scenarios will be very unlikely, but they must all be possible. The model results will not properly reflect the spectrum of possible outcomes if they include outcomes that could never actually occur. If the model outputs include impossible results, the model is not built properly. Box 10.1  A Brief History of Monte Carlo Simulation Monte Carlo simulation was developed in the 1940s at Los Alamos National Laboratory by Stanislaw Ulam and John von Neumann. The technique was instrumental in solving some of the difficult analytics required for the Manhattan Project, the research program that devel- oped the first nuclear weapons. Because the Manhattan Project was secret at the time, the method was given the code name “Monte Carlo,” after the Monte Carlo Casino in Monaco. In the 1950s, Monte Carlo simulations started to also be used in physics, chemistry, and operations research, and in 1964 David B. Hertz introduced Monte Carlo methods to finance with his Harvard Business Review article “Risk Analysis in Capital Investment.” Nowadays, with the availability of relatively inexpensive computing power, Monte Carlo simulation is used in a great number of fields, ranging from engineering, biology, and medi- cine, to business and finance. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 106 Fundamentals of Probabilistic Modeling Box 10.2  When to Use Monte Carlo Simulation Imagine that you live in a country where 30 percent of the population is allergic to a certain food item. What is the probability that four randomly selected individuals will have the allergy? Most people will answer that this is 0.3 × 0.3 × 0.3 × 0.3 = 0.0081 or slightly less than 1 percent. This is a probability calculation. The probability could also be determined by Monte Carlo simulation. To do so, we would put together a probability model with four distributions, each representing one of the four people. If we then simulate this Monte Carlo model 10,000 times, in about 81 of the iterations all four individuals would have the allergy. In other words, based on the results of the 10,000 Monte Carlo iterations, we could state that the probability of randomly selecting four allergic people would be approximately 81/10,000 = 0.0081. So, why would we ever want to perform a Monte Carlo simulation, which gives approxi- mate answers, rather than a probability calculation that gives an exact answer? The reason is that when probability models get more complex (for example, more distributions, or intricate relationships between distributions), using probability calculations to calculate the answer becomes too difficult. In Monte Carlo simulation, we use the computer to simulate thousands of scenarios instead of doing the calculations ourselves. For example, what if we do not exactly know the proportion of patients with the food allergy? Or what if we use a diagnostic test that is not 100 percent accurate? These additional uncertainties can make the probability calcula- tions too difficult, but including them in a Monte Carlo simulation is relatively easy. Monte Carlo simulation relies on the law of large numbers. The principle is that if we simu- late the Monte Carlo model many times, the result will be close to an exact answer. In the food allergy example, if we simulate the model many times the answer will be extremely close to the 0.0081 probability. In summary, if possible, manual probability calculations in a probabilistic model are pre- ferred to Monte Carlo simulation because the answer is exact. However, in most index insur- ance models the probability models will be too difficult to solve with probability calculation, and Monte Carlo simulation is the best approach. In this guide, Monte Carlo simulation is used for all probabilistic models. How many scenarios should be run for a Monte Carlo simulation? There is no universally correct answer, but we recommend running at least 10,000 scenarios. With fewer scenarios, the resulting statistics, especially the statistics in the tails of the distributions, such as the 90th, 95th, or 99th percentile, become less reliable. Running more scenarios is never incorrect because it simply gets closer to the true answer. However, more scenarios will take more time while the computer runs the model. The marginal benefit of doing more scenarios after 30,000 or 100,000 becomes lower and lower. Depending on the complexity of the model and the software platform used for the simulation, 10,000 scenarios may take a few seconds (for a sim- ple model) or several hours (for more complex models). Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 107 Let’s take our index insurance example from above. With a Monte Carlo model, the computer simulates the drought/no drought frequency variable by taking a random draw from a binomial distribution. At the same time, the com- puter takes a random draw from a distribution for the payout amount associated with the drought (severity). For each Monte Carlo scenario, the value for the payout amount will be the random draw from the distribution when drought occurs or when no drought occurs. Each scenario represents one possible out- come for the system we are modeling. If we now run many scenarios, say 10,000 or more, and summarize the results, we end up with a distribution for the total drought-related payout for the index product for the following year. Monte Carlo simulates the aggregate distribution of the frequency (drought/no drought) and severity distributions (payout amount). From this aggregate distribution we can quantify many risk-related aspects of the index product. For example, if 95 percent of the scenarios from the Monte Carlo simulation produce values of less than $75,000, we can say there is a 95 percent chance that the total payout amount for the following year will be $75,000 or less. Put differently, there is only a 5 percent chance of the total pay- out amount exceeding $75,000. As you can see, a great advantage of Monte Carlo simulation is that the com- puter runs thousands of possible scenarios very quickly, typically in a matter of seconds or minutes. Doing this manually would take days or longer, and would not show us how likely it is that different scenarios will occur. But, before running a Monte Carlo simulation model, it first must be built. The next section addresses three critical building blocks of Monte Carlo models. 10.2  Key Building Blocks for Probabilistic Modeling Building sound Monte Carlo simulation models requires careful consideration of the three main building blocks of probabilistic models: • Use of appropriate probability distributions • Correct use of the input data for these distributions • Proper accounting for the associations and relationships between variables Some probabilistic models also include a fourth element, time series, which combine probability distributions and relationships between sequential time steps. For the index insurance models covered within this book, however, the three building blocks above are the most important. 10.2.1  Probability Distributions Many different distributions can be used in a probabilistic risk model. Selecting the best distribution to apply in a specific situation is partly dictated by science and partly by art. The first thing to consider when picking a distribution is the nature of the variable that will be modeled. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 108 Fundamentals of Probabilistic Modeling First, all distributions represent either discrete variables, which can only take integer values, or continuous variables, which can take any value. Counts of events are discrete (for example, number of typhoons per year), while rainfall and temperature are continuous because they can be measured to any numerical precision. Second, many distributions are either right- (also called positive) or left- (also called negative) skewed (see figure 10.1), which means that the distribution is asymmetric, with one of the tails extending further than the other (see section 10.2.1.1.4 for more on tails). Skewed distributions are suitable for some variables and not others. The distribution of individual incomes is a classic example of a skewed distribution because the distribution peaks at relatively low values and then the right tail extends far to the right to account for the relatively few, extremely high incomes.7 Third, many distributions are bounded at specific values, meaning that all the values must be above, below, or both above and below a certain value. An income distribution, for example, is bounded on the left by zero because income is a positive value. However, a distribution for a probability or proportion will be bounded on the left at zero and on the right at 1. Fourth, some distributions are designed for stochastic, or random, processes. Recall the binomial distribution discussed earlier in this chapter. This distribution should automatically come to mind whenever you consider a series of events, each of which has two possible outcomes. The Poisson distribution is ideally suited to some either/or processes. Poisson distributions represent random, discrete events Figure 10.1 Skewing and Bounding in Distributions 0.035 0.030 0.025 Probability density 0.020 0.015 0.010 0.005 0 10 20 30 40 50 60 70 80 90 100 Values of random variable Right skewed, left bounded distribution Left skewed, right bounded distribution Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 109 that occur at some underlying rate, for example, counts of customer visits, auto- mobile accidents, typhoons, and many other “count” variables (see sections 10.2.1.2 and 10.2.1.2.3). Special distributions such as the PERT distribution8 have been specifically created for cases lacking hard data on the parameter to model, for which expert opinion therefore could be used (see section 10.2.1.2). This section first takes a detailed look at how to interpret a probability ­ distribution and then introduces common distributions that will be used in this guide. For a more complete and comprehensive discussion of probability distributions, we encourage the reader to consult the references listed at the start of this chapter. 10.2.1.1  How to Interpret Probability Distributions Probability density charts are represented on an x-y axis. The horizontal axis, conventionally called the x-axis, shows the random variable, the thing that we want to know or predict. This variable can be anything: net profits, crop yield, number of home runs by a certain baseball player, number of typhoons in the coming year, total insurance claim value in a calendar year, and so on. The vertical axis (y-axis) shows the probability for the x-axis variable. Probabilities can only take values from 0 to 1, so the y-axis is scaled from 0 to 1 and the total probabilities across all x-axis values also add up to 1.9 Three concepts will be important for interpreting probability distributions: cumulative probabilities, percentiles, and distribution tails. 10.2.1.1.1 Discrete and Continuous Probabilities. As discussed in paragraph 10.2.1, a key characteristic of a probability distribution is whether it rep- resents a discrete or continuous variable.10 Figure 10.2 shows a discrete Figure 10.2 Probability Density Chart of a Discrete Probability Distribution 0.20 0.18 0.16 0.14 0.12 Probability 0.10 0.08 0.06 0.04 0.02 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Value of random variable (number of typhoons) Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 110 Fundamentals of Probabilistic Modeling Figure 10.3 Probability Density Chart of a Continuous Probability Distribution 0.025 0.020 0.015 Probability 0.010 0.005 0 50 100 150 200 250 Values of random variable probability distribution for number of typhoons. The typhoon distribution is a discrete probability distribution, formally called a probability mass function, because the count of typhoons per year can only be an integer. There is no such thing as 1.5 typhoons. Figure 10.3 shows that the probability of the number of typhoons being equal to 2 is about 0.08, or an 8 percent probability. According to the model that generated this distribution, there is an 8 percent chance of two typhoons occurring in the next year. Continuous probability distributions are formally called probability density functions, and one is shown in figure 10.3. Note that there are no bars as in figure 10.2, just a smooth line. This smooth line indicates that a probability for any value on the x-axis can be found, for example, at x = 50, x = 50.01, x = 50.1, and so on. Of course, the continuous variable has limitless precision, so really when we say x = 50, we mean x = 50.00000000…. There are an infinite number of possible values of X, so there is essentially an infinitely small (that is, 0) prob- ability that x is exactly 50. The probability density function, on the other hand, returns the density for a continuous distribution. Because continuous variables can take any number of decimal points, it does not make sense to talk about an exact probability of observing a value. Instead we use the more abstract concept of the density, which is proportional to, but not the same as, the probability of observing a certain value. The formulas for the probability mass function (discrete probability distribution) and the probability density function (continuous probability distribution) are both expressed as f (x), with f being the probability function, and x being the value evaluated. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 111 The Bernoulli distribution, for example, is a probability mass function expressed as f (x) = p x (1−p)1−x, where p is the probability of success for a Bernoulli trial. For the less mathematically inclined readers, probability functions can be con- fusing, so this guide provides explanations of the concepts behind each distribu- tion. More detailed mathematics for the distributions used and discussed in this guide are covered in the additional texts recommended at the start of this chapter. 10.2.1.1.2 Cumulative Probabilities. Both continuous and discrete probability distributions can represent the probability of X being within a certain range. To find the probability of the number of typhoons in the discrete probability distri- bution in figure 10.2 being equal to any value up to and including 2, we add the probabilities for the x-axis values 0 (0.01), 1 (0.03), and 2 (0.08). Adding these up gives a value close to 0.12, or 12 percent. In other words, there is a 12 percent chance of two or fewer typhoons occurring in the next year. The conventional way of expressing this problem is as follows: Probability(X ≤ 2) = 0.12. In this notation, X represents the random variable. A lower case x is used to indicate a specific instance of that variable, so the more general form of the rela- tionship above is as follows: Probability(X ≤ x) = …. Probability(X ≤ x) is a very common method of interpreting probability dis- tributions, as is probability(X > x). Another term for both of these probabilities is cumulative probability because they are the accumulation (that is, the sum) of all probabilities of X up to the threshold of x (X ≤ x) or beyond the threshold of x (X > x). We can show cumulative probabilities in cumulative distributions. Whereas the distributions seen so far show the probabilities associated with individual values of the random variable, the cumulative distribution for Probability(X ≤ x) shows the probability of all values up to and including a certain value. The cumu- lative distribution is built by adding the probability for each value of the variable to all the preceding values. Once this is done for every variable value, we end up with a cumulative probability distribution (or cumulative distribution function). Notice that the cumulative probability distribution in figure 10.4 is discrete. The variable value moves in discrete steps of 1. The variable cannot take a value of 1.5 so there is no unique probability associated with 1.5. The cumulative probability of the variable being less than or equal to 1.5 is exactly the same as Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 112 Fundamentals of Probabilistic Modeling Figure 10.4 Cumulative Probability Chart of a Discrete Probability Distribution 1.0 0.9 0.8 Cumulative probability 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Values of random variable the cumulative probability of the variable being less than or equal to 1, hence the flat lines at each step in the cumulative probability distribution. The cumulative probability up to a given x value is computed for continuous distributions in a comparable way. The only difference is that where we could add up the probabilities in the discrete distribution, with a continuous distribu- tion, we integrate the area under the curve to determine the cumulative probability. Once we do this for every variable value, we get a continuous cumu- lative probability distribution (cumulative distribution function). The cumulative probability distribution makes it much easier to find the kinds of probability measures that are of interest in risk and insurance modeling. For example, in figure 10.5 we can readily see that there is about a 65 percent chance of values less than or equal to 50, and about a 90 percent chance of values less than or equal to 100. 10.2.1.1.3 Percentiles. One type of probability measure that we often use with continuous distributions is the percentile. When we use a cumulative probability distribution, the percentiles are the points along the x-axis that correspond to certain cumulative probabilities. For example, the 10th percentile (P10) is the value of x such that the Probability(X ≤ x) = 0.10. In figure 10.6 the vertical line represents the P10. The line touches the distribution at the y-axis value of 0.1. The corresponding x-axis value is 138. From this we know that the P10 is approximately 138. There is a 10 percent probability that a random x will be less than or equal to 138 and there is a 90 percent chance that a random x will be greater than 138. We can also find percentiles for a continuous distribution using a probability density chart (as opposed to a cumulative probability chart). With a continuous probability density chart, the probability is proportional to the area under the Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 113 Figure 10.5 Cumulative Probability Chart of a Continuous Probability Distribution 1.0 0.8 Cumulative probability 0.6 0.4 0.2 0 50 100 150 200 250 300 Values of random variable Figure 10.6 Cumulative Probability Chart of a Continuous Distribution with the P10 Displayed 1.0 0.9 0.8 0.7 Cumulative probability 0.6 0.5 0.4 0.3 0.2 0.1 0 100 138 200 300 400 500 600 Values of random varible Note: P10 = 10th percentile. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 114 Fundamentals of Probabilistic Modeling Figure 10.7 Probability Density Chart of Continuous Distribution with the P10 Displayed 0.006 0.005 Probability density 0.004 0.003 0.002 0.001 0 100 138 200 300 400 500 600 Values of random variable Note: P10 = 10 percentile. th curve. As illustrated by the vertical line in figure 10.7, the P10 divides the area under the curve such that 10 percent of the area under the curve is to the left and 90 percent of the area under the curve is to the right of 138. In general, probability distributions rather than cumulative probability distri- butions are used throughout this guide because most people find them easier to interpret and they are the default in most modeling software. When it comes to finding specific percentile values we rely on the modeling software to provide the correct figure. When we show probability distributions we generally leave out any numbering along the x-axis but label the percentiles and other key metrics to orient the viewer to the scale of the distribution. 10.2.1.1.4  Distribution Tails. Figure 10.7 also illustrates a term that is common in probabilistic modeling: tail. When we look at this distribution, we see a peak on the left side and a long, extending tail on the right, in the positive direction. This distribution is positively skewed. We can readily tell that low values of x are more likely than high values. As a result, the P50 is far to the left. Although the curve extends far to the right, there is very little area under the curve on the right. For the distribution shown in figure 10.7, 50 percent of the area falls to the left of x = 228. The long right tail tells us that there is still some chance of getting values of x higher than 500, even if it is very unlikely. This is the nature of long- tailed, positively skewed distributions. They can show us low-likelihood, extreme-value situations. These are very important for risk assesments of index insurance products. Although less relevant for index insurance, we want to briefly mention the importance of fat-tailed distributions. These distributions have one or both tails that are both long (x-values very far from the median) and fat (x-values on the Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 115 tail have a higher probability of occurring than for thinner tailed distributions). These distributions can be helpful for modeling inventory damage and can model very large (or very small) damages. The Pareto distribution has the fattest tail of all probability distributions. The models used and discussed in this guide do not use the Pareto distribu- tion to simulate inventory damage levels (for example, a dollar amount of losses). Instead, the models in this guide simulate the payouts of the index insurance product as a percentage of the sum insured, which does not corre- spond exactly to the damage experienced on the ground. Since the payout ratio is a percentage, and percentage payouts can only range from 0 percent to 100 percent, this guide uses, as a default, the beta distribution to represent the payout ratio (which ranges from 0 to 1; see section 10.2.1.2.4). 10.2.1.2  Selecting Probability Distributions This section explains a select group of probability distributions that are most helpful when modeling index insurance products. These specific distributions were selected using the same four considerations discussed at the start of section 10.2.1: • Type of variable (discrete or continuous): One of the uncertainties in our models, for example, is whether the product will pay out or not pay out in the next season. Because there are only two potential outcomes for this variable (­payout/ no payout) we selected a Bernoulli distribution that represents such discrete situations. • Right or left skew of the distribution: For example, a left-skewed distribution may be appropriate for representing the annual rainfall in a particular area if there are many years with similar amounts of rainfall but some years with extremely low rainfall. • Bounding of the distribution: One of the uncertainties is what the size of the payouts should be, which is a percentage of the amount insured. Because such percentages can only range from 0 percent to 100 percent, a beta distribution (which is bounded at 0 and 1) was selected to represent this variable. • Specific design of the distribution for stochastic processes: Poisson distributions, for example, represent random, discrete events that occur at some underlying rate. Poisson distributions are often used to model events such as accidents. Not all distributions are discussed in this section and for those that are dis- cussed, the descriptions are fairly brief. We again encourage the reader to refer to the additional texts recommended at the start of this chapter to become more familiar with the different probability distributions that are available to represent uncertainties in Monte Carlo simulation models. 10.2.1.2.1  Bernoulli Distribution. The Bernoulli distribution models whether a single trial results in success or failure, for example, winning a coin toss (or not), winning the lottery next week (or not), or defaulting on a loan in the next year Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 116 Fundamentals of Probabilistic Modeling (or not).11 It models a single event that has an outcome that can take one of two values, which are usually denoted as 1 for a success and 0 for a failure. The outcome of the single event is governed by the probability of success, which is conventionally represented by the letter p, which is the only parameter for this simple distribution. f (x) = px (1−p)1−x For a fair coin, the probability of success ( p) is 0.5, or 50 percent, but for winning the lottery the probability of success is much smaller, for example, 0.00000000001. When modeling an index insurance product, we can use the Bernoulli distribution to simulate payouts for the next year in a specific area. The probability of a payout is often not equal to 50 percent but may be 15 percent, for example. In this case, the Bernoulli distribution within our Monte Carlo simu- lation model will generate a 1 in 15 percent of the scenarios and a 0 in 85 percent of the scenarios. The Bernoulli distribution can be related, or linked, to other distributions in the model. For example, a second distribution can provide the actual payout amount in the case in which a payout occurs (see section 10.2.1.2.4.2). Finally, a Bernoulli distribution is the same as a binomial distribution (see the next paragraph) when in the binomial distribution the number of events (or trials) is set at one. 10.2.1.2.2 Binomial Distribution. The binomial distribution is one of the most commonly used discrete distributions and it represents the outcome of multiple Bernoulli events, or trials. The best known example is the multiple coin toss problem discussed in section 10.1.3. The binomial distribution shows the distri- bution of the number of successes out of a certain number of trials along with the corresponding probability. For example, if an employee of a health agency is randomly testing people in a country for a disease with a prevalence rate of 8 percent, the number of people in that survey who do have the disease will follow a binomial distribution. Or if a bank has outstanding loans and the annual default rate is 2 percent, the number of defaults during the next year will also follow a binomial distribution. The parameters for the binomial distribution are the number of trials and the probability of success. Conventionally, these parameters are denoted as n and p, respectively. The binomial distribution in a Monte Carlo model samples how many successes (conventionally called s) there will be from the n trials.12 The binomial distribution can also help with an analysis of index insurance products. For example, in any given year or season an index insurance policy can result in a payout without any inventory damage caused by the named peril occurring (an example of insurer basis risk). Suppose we have reason to believe that the probability of such an event is 3 percent per year and remains the same over the next five years. We can use a binomial distribution to represent the probability distribution of the number of insurer basis risk events over the next Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 117 five years. The number of years is the number of trials (n) and 3 percent is the probability of success (p).13 It is important to note that if we only want to model the probability of an insurer basis risk event in the next year, then we use a binomial distribution with just one trial, which is equivalent to a Bernoulli distribution. In this guide, all of the models focus on forecasting claims for the next season or year, so we use Bernoulli distributions (one trial), not binomial distributions (a series of trials). 10.2.1.2.3  Poisson Distribution. The Poisson distribution is also a discrete distribu- tion, but unlike the Bernoulli and binomial distributions there is no probability of a trial succeeding or not. Instead, the Poisson distribution represents the count of independent events that happen according to a certain rate per unit of exposure. Units of exposure are often measures of time, so we might look at the number of automobile accidents in a specific city per year. The unit of exposure here is one year. The Poisson distribution can estimate how many accidents in the city might occur over the next five years. Units of exposure can be any continuum, such as volume, mass, or area. For example, we might look at the number of automobile accidents per intersection, where one intersection during one month of time is the unit of exposure. During this unit of exposure, a number of accidents can happen that can be simulated by the Poisson distribution. The two parameters for a Poisson distribution are the rate (average number of events) per unit of exposure (lambda), and the exposure quantity (t).14 In the car accident example, the rate per unit of exposure is 400 accidents per year (lambda) and the exposure quantity is five years (t). The Poisson distribution forecasts the number of events that could actually happen, called alpha. Poisson event data are commonly referred to as count data. The Poisson dis- tribution is useful for index insurance modeling of counts such as the number of typhoons occurring during the next year. 10.2.1.2.4  Beta Distribution. The beta distribution is the first continuous distri- bution discussed in this guide (the Bernoulli, binomial, and Poisson are all discrete). It is bounded by 0 on the left and 1 on the right. In other words, the beta distribution can only generate values greater than 0 and less than 1. The beta distribution generally has two uses: (1) to model the uncertainty in a probability or proportion and (2) to model continuous variables that range from 0 to 1. 10.2.1.2.4.1  Use of the Beta Distribution #1: Modeling the uncertainty in a prob- ability or proportion. The beta distribution models uncertainty in the true value of a probability or proportion.15 Imagine that 10 cars are randomly selected in a large city and 3 of them are blue. How many cars in the city could now be esti- mated to be blue? A simple estimate is 30 percent, but with so few observations we should not overstate our certainty that 30 percent is the true proportion of blue cars in the city. With only 10 observations, actual values of 20 percent or 40 percent could also easily result in the same data (3 blue cars out of 10). Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 118 Fundamentals of Probabilistic Modeling This uncertainty about the probability or proportion can be very large when few data are available, which is fairly common for index insurance applications for which often only 10 to 30 years of historical data exist. The beta distribution provides a distribution of possible values for the probability that an index prod- uct will pay out during the next season. The parameters of the beta distribution are conventionally represented by the Greek letters alpha and beta. Previous researchers have worked out various means of translating the data we usually work with into proper values of alpha and beta.16 For example, we can estimate the probability of success (p) using a beta distribution with alpha = s + 1 and beta = n − s + 1, where s is the number of successes and n is the number of trials. In the example of the 3 blue cars out of 10 randomly selected cars, our uncer- tainty about the actual proportion of blue cars in the city can be described by the blue line in figure 10.8. The peak of this distribution (the mode) is at 0.3 on the x-axis, which you probably suspected given that you saw 3 blue cars out of 10 cars. However, the distribution shows considerable uncertainty around what the true proportion is, and that even values as low as 0.10 and as high as 0.70 are possible. On the other hand, if we had seen 30 blue cars out of 100 randomly selected cars, the orange line beta distribution would have represented our uncer- tainty about the proportion of blue cars. In this case the uncertainty is quite a bit less because the estimate of the proportion was based on more data. Turning to index insurance applications, suppose that we are again modeling insurer basis risk events, that is, situations in which the index product provides a payout despite there being no inventory damage attributable to the named peril. We have 10 years of historical data (n) and during those 10 years, an insurer Figure 10.8 Probability Density Charts of Two Beta Distributions 10 9 8 7 Probability density 6 5 4 3 2 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Values of random variable Beta (3+1, 10–3+1) distribution Beta (30+1, 100–30+1) distribution Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 119 basis risk event occurred in one year (s). A simple, first estimate of the probability of an insurer basis risk event is 10 percent (s/n). However, given our limited data set, we represent the uncertainty about the true probability as a beta distribution with alpha = 2 (that is, s + 1 = 2) and beta = 10 (that is, n − s + 1 = 10). If we do not use the beta distribution we ignore the uncertainty in the probability and can underestimate the overall risk. 10.2.1.2.4.2 Use of the Beta Distribution #2: Modeling continuous variables ranging from 0 to 1. The second potential use of the beta distribution is for modeling a continuous variable that can only vary from 0 to a maximum of 1. ­ An example is the payout (as a percentage of the insured amount) that an index insurance product may produce next year. In fact, several of the example models that come with this guide use the beta d ­ istribution to simulate the payout ratios for cases in which there is a payout greater than 0 percent. The technique for determining the input parameters of the beta distribution based on the historical payout ratio for specific index products, called fitting distributions to data, is covered in section 10.2.2.1. 10.2.1.2.5  Gamma Distribution. The gamma distribution is another continuous distribution, but this one is bounded on the left by 0 and positively skewed with- out a right bound, meaning the right tail extends far to the right. A common application of the gamma distribution is in modeling the rate of Poisson events per unit of exposure, or the time required for a specific number of Poisson- distributed events to occur. In the insurance industry the gamma distribution is used to represent potential payout amounts in currency terms, for example, $400, rather than as a percentage of the sum insured, for example, 35 percent. The left- side bound at 0 corresponds to modeling payouts because only positive payout amounts are sensible. The long right-side tail represents the general behavior that, although most payouts will be of smaller amounts, a small number of payouts may be extremely large. Gamma distributions can also be used to represent trig- ger values, for example, 44 millimeters of rain during a crop season, because many triggers such as rainfall and degree hours can only have positive values. There are different parameterizations of the gamma distribution, meaning that it can be calculated from functions that take different input parameters. A com- mon parameterization requires two parameters: shape and scale. These are denoted by the Greek letters alpha and beta, respectively. The alpha parameter drives the shape of the distribution while the beta parameter drives its spread (scale). In practice, the larger the shape parameter (alpha), the more the gamma distribu- tion tends to look like the normal distribution. With smaller values of alpha, the distribution looks like a much skewed distribution called the exponential. The scale parameter (beta) is directly proportional to its standard deviation. The bigger the scale parameter, the greater the variability in the distribution samples. A neat feature of the gamma distribution is that its mean is simply alpha times beta. This calculation is a quick way to check that the parameter values make sense compared with the historical data. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 120 Fundamentals of Probabilistic Modeling As noted above, the gamma distribution can model payout amounts for index insurance when these are modeled in currency terms. The gamma distribution can also be used to model trigger values. However, the example models within this guide use historical payouts as a percentage of the insured amount (payout ratios). When handled in this fashion, the payouts are bounded by 0 and 1, with 1 being 100 percent. Thus, we use a beta distribution to model payouts in this guide (see chapter 16). 10.2.1.2.6  PERT Distribution. The PERT distribution is also a continuous distri- bution. It has three parameters: the minimum, most likely (mode), and maximum. This distribution is particularly useful for modeling experts’ opinions about a quantity, such as the market size for a new product or the potential market share a new entrant can capture. For example, when an insurance company plans to launch a new index product, there may be uncertainty about the number of bank branches that will purchase the product. Experts might indicate that they think the most likely number of bank branches that will purchase the product will be 50, but could be as low as 30 or as high as 100. Figure 10.9 shows the distribution for the PERT (minimum = 30, most likely = 50, maximum = 100), also written as PERT (30, 50, 100). An important characteristic of the PERT distribution is that all values between the minimum and maximum are possible because it is a continuous distribution. For the bank branch example, we need to use Excel’s ROUND function to make sure we do not get outcomes such as 45.7 branches enrolling, instead of 46. As figure 10.9 also shows, values around the most likely value are sampled often while values close to the minimum of 30 and maximum of 100 are less likely to be sampled. Figure 10.9  PERT (30, 50, 100) 0.035 0.030 0.025 Probability density 0.020 0.015 0.010 0.005 0 20 40 60 80 100 120 Number of new enrollees Note: PERT = program evaluation and review techniques. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 121 10.2.2  Distribution Parameters Each of the distributions discussed, and every probability distribution, has certain parameters. The Bernoulli distribution has one (p). The binomial distribution has two (n and p) as does the Poisson (lambda and t), the beta (alpha and beta), and the normal (mean and standard deviation). Where do the values for these parameters come from? In general, probability distributions are parameterized in two ways: expert opinion elicitation and distri- bution fitting. Expert opinion elicitation involves consulting the opinions of experts to quantify uncertainty.17 Alternatively, historical data can be used to estimate model parameters using distribution fitting. Distribution fitting is discussed in the next section. For further information on choosing parameters based on historical data or expert opinion the reader is encouraged to read the additional texts recom- mended at the start of this chapter, specifically Gelman (2013), Bolker (2008), and Law and Kelton (2006). 10.2.2.1  Distribution Fitting Distribution fitting is used to represent historical data in the model to forecast future behavior. For example, the amount of money that the next customer in a store will spend may be modeled using the spending amounts of the last 100 people. In this case, we must have reason to believe that the historical spending data are credible for estimating the spending of the next person. In the index insurance models in this guide, distribution fitting is used to iden- tify the appropriate value of the distribution parameters for the index product payout ratios in the next year. In the case example, the payout is a proportion of the insured value—severe droughts result in 100 percent payouts and less severe droughts result in payouts of smaller percentages. The payout ratios are values from 0 (0 percent) to 1 (100 percent) and are continuous. The beta distribution, which also ranges from 0 to 1 and is continuous, is a reasonable choice. To use the beta distribution, we first need to estimate the values of the beta parameters alpha and beta. To fit a distribution to data, historical data on the phenomenon of interest are needed. The process of fitting a distribution to the observed data is similar to trying on a number of different sizes and types of shirts to find the one that fits the best. For the case example, the historical data on payouts can be represented by a frequency graph (also called a histogram) of the different historical payout amounts that have occurred in the period covered by the historical data, as in figure 10.10. Conceptually, the distribution can be manually fitted by choosing values for the beta parameters, for example, alpha = 1 and beta = 2, and using them to plot the curve of the corresponding beta distribution over the data (the blue histo- gram bars). Then the process can be repeated with different parameter values for the beta distribution (for example, alpha = 0.5, beta = 3) to see if the resulting curve matches the data better or worse than the first curve. If this process is repeated with different parameter values, a set of values that best mimicked the Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 122 Fundamentals of Probabilistic Modeling Figure 10.10  Frequency Graph of Historical Payout Amounts, Together with the Maximum Likelihood Estimation of the Fitted Beta Distribution 3.5 3.0 2.5 Probability density 2.0 1.5 1.0 0.5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Historical payout ratio pattern would eventually emerge. Figure 10.10 is a beta distribution with alpha = 0.65 and beta = 1.07, represented by the orange curve.18 To fit the distribution, a judgment must be made of which curve (which parameters of the beta distribution) best approximate the data. There are dif­ ferent ways of calculating and judging the closeness of the curve to the data. Closeness is commonly determined by calculating the (joint) likelihood that the observed data came from the fitted distribution. Then a combination of param- eters can be found that maximizes the chances that all the data points observed came from the fitted distribution and parameters. This process is called maxi- mum likelihood estimation (MLE) and parameters from fitted distributions are called MLEs. Risk modeling software packages include automatic procedures to perform MLE fitting very quickly. Multiple distributions may be appropriate for representing a given situation in a model. For example, some people prefer the gamma distribution to the lognor- mal distribution (which is another continuous distribution with a long tail to the right) for representing payouts. If there is no specific reason for choosing gamma over lognormal, the fits from both distributions can be compared using goodness of fit (GOF) statistical criteria. A number of GOF statistics, such as the Anderson-Darling statistic and the Kolmogorov-Smirnov statistics, can help with selecting the best fitted distribution. However, these statistics do not take into account the complexity of candidate probability distributions (the number of parameters of the different distributions) Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 123 that could be fitted to the data. In contrast, GOF statistics based on information criteria, such as the Akaike information criterion and the Bayesian information criterion, do penalize distributions for extra parameters.19 For readers interested in learning more about information criteria, we recommend Gelman (2013) and Bolker (2008). Different GOF statistics may return different conclusions about which distribu- tion has the best fit. As we discussed in previous sections, it is therefore always necessary to also consider the key characteristics of the distribution (the type of variable, skewness, bounding, design for stochastic processes) as well as whether the ­ distribution is commonly applied in the specific field (that is, industry practices). 10.2.2.2  Parameter Uncertainty As discussed above, parameter uncertainty is the measure of the incompleteness of our knowledge (also called our lack of confidence or ignorance) about the true value of a parameter that cannot be readily observed. When a parameter is esti- mated, an idea of what the parameter might be is obtained, but the confidence around this estimate will depend on the amount of data that is available. With only a few observations, the uncertainty about the parameter is relatively high because many different values could result in those same observations. Let us consider two examples in index insurance in which parameter uncertainty is relevant: ­ First, recall that when the distribution was fitted to the data in figure 10.10 MLEs were used to determine that the best fitting beta distribution parameters were alpha = 0.65 and beta = 1.07. However, the same data could also have come from a beta distribution with alpha = 0.59 and beta = 1.18 or alpha = 0.55 and beta = 0.8. In other words, although alpha = 0.65 and beta = 1.07 are the MLEs of the beta distribution for this data, the exact values of both of these parameters are still uncertain. Figure 10.11, where the best fit (based on MLEs) is the orange line and alternative fits are in green, shows this parameter uncertainty. There are actu- ally an infinite number of alternative fits possible, but these other potential fits will be similar to the green lines. The data we have observed could have come from any of these alternative beta distributions. Figure 10.12 shows the uncer- tainty in the values of both the alpha and beta parameters of the fitted beta distribution. The uncertainty distributions of both of the variables are centered around the MLE of alpha and beta, which are equal to 0.65 and 1.07, respec- tively. Alternative values for the variables are, however, still possible. For example, even though the MLE of alpha is 0.65, its value could actually be 1, or an even higher value. For the second example of parameter uncertainty, suppose we are modeling the chances of a drought occurring in the next year but do not know the prob- ability of drought in a given year. Historical data might tell us that one drought year (s = 2) occurred in the past 10 years (n = 10). The best estimate of the probability of a drought occurring next year is 20 percent (s/n), which is also the MLE for the p parameter in a binomial distribution. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 124 Fundamentals of Probabilistic Modeling Figure 10.11  Frequency Graph of Historical Payout Amounts, Together with the Maximum Likelihood Estimates of the Fitted Beta Distribution (Orange) and Alternative Fits Based on Parameter Uncertainty 3.5 3.0 2.5 Probability density 2.0 1.5 1.0 0.5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Historical payout radio Figure 10.12  Parameter Uncertainty for Alpha and Beta 0.20 0.18 0.16 0.14 Probability density 0.12 0.10 0.08 0.06 0.04 0.02 0 0.5 1.0 1.5 2.0 2.5 Values of parameter Alpha Beta Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 125 With this information, we can model an uncertainty distribution for the parameter p with a beta distribution (that is, a continuous distribution that is bounded at 0 and 1). Using what is known in Bayesian statistics as a conjugate and uninformed prior,20 we can then use the trials and successes in our histori- cal data (trials, n, is the number of years of data, and successes, s, is the num- ber of years in which a drought occurred) to compute the two parameters of the beta distribution as alpha = s + 1 and beta = n − s + 1. Using these two parameters within the beta distribution will provide the uncertainty distribu- tion for the probability of a drought in the next year (the uncertainty around the true probability, p). When the data are numerous, this distribution of parameter uncertainty will be relatively tight, which indicates that we have more confidence around what the true parameters are. It will be focused on a narrow range of possible values. When the data are few, the distribution will be broader, which indicates that we have less confidence about the true value. This is illustrated in figure 10.13 where the uncertainty is shown around the true probability p for two cases: (1) 10 samples with 2 successes (for example, 2 years of drought out of 10 years) and (2) 100 samples with 20 successes. The modeling of the parameter uncertainty with the beta distribution is based on historical data and assumes that future patterns will be similar to those in the past. As a result, parameter uncertainty does not account for possible changes in the systems themselves over time (see section 9.3 and chapter 16). System changes relevant to index insurance may include climate change (which could change the frequency and severity of payouts) or increased resilience of a farmer’s crop or livestock to weather risk due to breeding or management practices (which could Figure 10.13  Tight vs. Broad Parameter Uncertainty 12 10 Probability density 8 6 4 2 0 0.1 0.2 0.3 0.4 0.5 0.6 Possible values of the annual probability of a drought 2 out of 10 20 out of 100 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 126 Fundamentals of Probabilistic Modeling change the actual losses, and therefore the amount of product design basis risk of the index product). As can be seen in figure 10.13, when 2 drought years out of 10 years have been observed there is considerably more uncertainty around the true probability of a drought next year than when 100 years of historical data are available. However, in this example, one of the assumptions is that the annual probability of a drought does not change over time, which may or may not be a valid assumption. If the probability does change (for example, increase or decrease) over time, the distribution might be modified using time series methods. If the drivers that cause droughts can be better understood, we may be able to better predict them and therefore reduce parameter uncertainty because we could explain why droughts have happened in the past (or not) and when they will likely happen again (Jewson and Brix 2005). Such unveiling and understanding of drivers is complicated, but would help index insurance with better pricing. Whenever any parameter is estimated from historical data, especially in situa- tions in which only limited data are available, parameter uncertainty can be important. The field of statistics (both classical statistics and Bayesian statistics) provides a variety of tools for estimating the relevant parameter uncertainty. In the index insurance models in this guide, the probability of a payout per year per region includes parameter uncertainty. Parameter uncertainty in esti- mating the probability distribution of historical payouts could also have been included. However, it was not included because parameter uncertainty consid- erably slows down the speed of the Monte Carlo simulation models.21 Therefore, if you use Monte Carlo models with limited amounts of historical data we encourage you to include parameter uncertainty. Alternatively, you can run the model twice, once with and once without parameter uncertainty, to determine whether including parameter uncertainty significantly affects the model’s results and interpretations. 10.2.3  Modeling Relationships between Variables So far this chapter has assumed that the probabilistic variables in the model are independent or uncorrelated. The price of oil and the number of wins of a profes- sional sports team are uncorrelated. Knowing the price of oil will give one no advantage in predicting one’s favorite team’s wins. Variables can also be correlated. A positive correlation means that as the value of one variable increases, the value of the other variable also tends to go up. Among children, age and height are positively correlated. As children age, they get taller and if we know a child’s age we can make a more accurate guess as to the child’s height. Negative correlation means the values of the two variables move in opposite directions. As one increases, the other tends to decrease, and vice versa. A statistical association between two or more variables is also called a correlation or a statistical relationship. It is important to note that correlation does not imply causation. Although two variables may be correlated, one does not necessarily cause the behavior of the other. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 127 For example, imagine a single variable, Variable A, that represents an expert’s estimate (and his or her uncertainty) of the number of bank branches in Region A that will purchase index insurance policies next year. Let us assume sales in Region A are expected to range between 30 and 100, with the most likely number of sales being 50. We can model Variable A with a PERT distribution that takes into account the most likely value for A (50) as well as the minimum (30) and maximum values for A (100). The most likely value is the peak of the PERT distribution and the minimum and maximum are the tails. If we draw values for A from this distribution, most will be around the most likely value but we will occasionally draw values close to the tails of the distribution. Now imagine that we also offer index insurance in another nearby region, Region B. We can add a second variable in our model, Variable B, representing the number of bank branches in Region B that will purchase index insurance policies next year. In this case, we can use a PERT distribution for B with parameter values appropriate to Region B. Let’s assume sales in Region B are expected to range between 50 and 200, with the most likely number of sales being 100. Given the shapes of both PERT distributions, values closer to the maximum for Variable A and Variable B are relatively rare, and it is even rarer that we will draw high values for both variables in the same scenario of a Monte Carlo simulation. This assumption holds true as long as the two variables are uncorrelated (indepen- dent of each other). If, however, Variables A and B are strongly positively corre- lated, then when Variable A is high, Variable B will also tend to be high, and vice versa. Another way to say this is that when A and B are correlated, if sales in Region A are high (close to the estimated maximum of 100), we also expect high sales in Region B (closer to the estimated maximum of 200). Even though it is rare to see a high value for Variable A, when we do, we are also likely to see a high value for Variable B. The same is true for extremely low values. As can be seen, the positive correlation between the variables, in this case the number of bank branches that purchase the product, can increase the risk to the insurance company because it increases the likelihood of extreme values, both positive and negative. Analysts commonly describe correlations between variables in a probabilistic model using Spearman’s rank order correlation coefficient, which is a metric that can vary between −1 and +1 and indicates the strength and direction of the correlation. Values closer to +1 or −1 indicate stronger positive or negative correlation, respectively. Negative values indicate negative correlation (the vari- ables move in opposite directions), and positive values indicate positive correla- tion (the variables move in the same direction). One characteristic of the rank order correlation coefficient is that it assumes that the relationship between the variables is linear and the same throughout the range of the variables (for example, no wedge-shaped correlations). An alternative and more flexible way to represent correlations in a probabilistic model is to use copulas, which are used in the models in this guide. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 128 Fundamentals of Probabilistic Modeling 10.2.3.1  Understanding Copulas To understand copulas, it is helpful to first work through an example of randomly sampling from a probability distribution. This method uses percentiles. Remember that if a student takes an academic test in school and her score is on the 90th percentile, her score is greater than 90 percent of the scores from the rest of the students in her class and also lower than 10 percent of the scores from the rest of the students. If her actual test score was 42 out of 50, that means that 90 percent of her class received lower scores than 42, and 10 percent received higher scores.22 For any percentile x between 0 and 1 (remember that 1 is equal to 100 percent), we can find the corresponding test score, the score that is greater than x percent of the scores. This is also true of any probability distribution. Any value in the distribution can be related to the percentile, which is the percentage of the distri- bution that the specific value exceeds. How are percentiles used to sample from a probability distribution? First, a number between 0 and 1 is randomly selected to serve as a percentile. The ran- dom number is drawn from a uniform distribution, which means that all values from 0 to 1 are equally likely to be drawn.23 Second, the value that corresponds to that percentile in the probability distribution of interest is selected. For exam- ple, we randomly select 0.8 as the percentile from the gamma distribution with alpha = 5 and beta = 1 in figure 10.14. The corresponding value for the 80th percentile of the gamma distribution is 6.72. Now this same method of sampling probability distributions can be applied to the situation of correlated variables within a probabilistic model. Let us return to our earlier example of Variable A and Variable B, which represent sales of index insurance products in two different regions, and use the method above to demonstrate how the value of one affects the other. First, we will look at the situ- ation in which they are completely independent (zero correlation). Assume that they also both have PERT distributions. To start, we randomly draw a percentile Figure 10.14  Generating a Random Value from a Gamma (5, 1) Distribution 1.0 0.8 Cumulative probability 0.6 0.4 0.2 0 6.72 5 10 15 Value of random variable Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 129 from a uniform distribution bounded by 0 and 1, for example 0.8 (the 80th percentile), and select the corresponding value for A from its beta distribution, 66 in our example. Next, we draw a value for Variable B. Because Variables A and B are independent, the value we drew for Variable A has no impact on what we will draw for Variable B. For Variable B, we again randomly draw a percentile and select the corresponding value. We now have two random, independent values for A and B. Second, let us take the situation in which Variable A and Variable B are strongly, positively correlated. In this case, if we randomly draw the 80th percentile for Variable A, we will also draw a large percentile for Variable B. In other words, the correlation between the variables restricts the range of possible percentiles for the second variable. The stronger the correlation between the variables, the narrower the range of possible percentiles for the second distribution becomes. The weaker the correlation between the variables, the wider the range of possible percentiles for the second distribution becomes. In our example, Variable A and Variable B are strongly and positively corre- lated, and we have randomly drawn the 80th percentile (representing sales of 66) for Variable A. For Variable B, we will restrict the likely percentiles that can be picked, for example between the 75th and 85th percentiles. If A and B had been weakly correlated, the range of likely percentiles might stretch from the 65th to the 95th percentile. A copula operates similarly to the example described above. It essentially restricts the percentile ranges of samples from the two distributions so that the resulting samples have the proper degree and direction of correlation. Correlations can exist not just between two variables (where there is one relationship) but also between multiple variables in a model. Fortunately, copulas have the ability to correlate many more than just two distributions. When multiple distributions are correlated, the numbers of relationships increase rapidly. For example, when there are three distributions, there are three relationships (between A and B, between A and C, and between B and C). When there are four distributions there are six relationships (A-B, A-C, A-D, B-C, B-D, C-D, or 3 + 2 + 1). When we are interested in the historical payout amounts for 10 nearby geographical areas, there will be a total of 45 relationships to consider (9 + 8 + …. + 1).24 Like probability distributions that can be fitted to data, copulas can also be fitted to data. When a copula is fitted to historical data, the strength of each of the relationships is based on the strength of the relationship that is displayed in the historical data. An advantage of fitting copulas to data is that it allows us to quickly and conveniently reflect relationships between variables based on histori- cal patterns. A disadvantage is that, if relationships change over time, basing the correlations on historical data may not be valid. The mathematics of both fitting and simulating copulas can be fairly complex, but fitting copulas and simulating from copulas is simple when using standard risk modeling software packages. For readers who are interested in learning more about the mathematics and applications of copulas, we recommend Cherubini, Luciano, and Vecchiato (2004; see the texts at the start of this chapter). Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 130 Fundamentals of Probabilistic Modeling 10.3  Key Outputs for Probabilistic Modeling So far this chapter has looked at how and why we use probabilistic modeling as a tool to better understand index insurance products. It has also discussed how a Monte Carlo simulation based on a probability model results in an outcome that is itself also a probability distribution. The probabilistic models do not deliver only a simple number, but provide a more comprehensive and realistic view of the results. So what are some appropriate ways to summarize, interpret, and communicate the results from a probabilistic model? 10.3.1  General Metrics Used in Probabilistic Modeling As an example, consider the histogram in figure 10.15 showing the hypothetical results from a model that estimates next year’s total payout amount across 10 regions. What does this histogram represent at the elementary level? Remember that when a Monte Carlo simulation model is run, at least 10,000 scenarios are gener- ated that each represents a possible future outcome. Of course, we do not want to present a decision maker with 10,000 possible answers to the question “What will next year’s claim amount be?” This is where the histogram comes in, because it provides a convenient overview of all the 10,000 possible outcomes of the Monte Carlo simulation. Many different metrics can help describe different aspects of this histogram, but we use four key metrics to interpret and summarize a histogram: • The mean • The spread, using percentiles such as the P5 and P95 Figure 10.15  Histogram Plot of Next Year’s Total Payout Amount 5% 90% 5% 0.12 5% Mean (X = 1.73) 95% (x = 10.72) 0.10 P95 = $10.72 million P5 = $1.73 0.08 million Mean of all values = $5 million Probability 0.06 Mean of all values above the P95 = $13.77 million (TVaR-95) 0.04 0.02 P99 = $15.58 million 0 5 10 15 20 25 30 Total payout amount (US$ million) Note: P = percentile, for example, P5 = 5th percentile; TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 131 • The tail value at risk (TVaR) • The probability of exceeding (or being less than) a certain threshold value The mean of the outcome distribution (also called the expected value) tells us what the expected outcome is given all of the uncertainties and risks. It is the value one expects to get on average if an experiment is run many times. Going back to figure 10.15, the expected value of the total payout distribution is the average annual payout the insurer can expect to make over many years, which in this example is approximately US$5 million. However, the mean outcome does not tell us anything about the uncertainty surrounding next year’s payout amount. A common measure used to indicate the amount of uncertainty or randomness is the standard deviation.25 A commonly used rule of thumb for understanding the standard deviation is that the mean plus or minus twice the standard deviation contains 95 percent of the range of out- comes. However, this rule only applies for normally (Gaussian) distributed vari- ables. Therefore, this rule is not appropriate for the above distribution of annual payouts, nor for most resulting distributions that are used for index insurance. Percentiles provide an easier way to describe the uncertainty around the expected outcome (that is, the spread of distributions). As discussed in section 10.2.3.1, any value in the distribution can be related to the percentile, which is the percentage of the distribution that the specific percentile value exceeds. Two common percentile values reported for Monte Carlo results are the P5 and the P95.26 The P5 value will be on the left side of the distribution because the P5 value is greater than only 5 percent of values. The P95 is on the right side of the distribution because it is greater than 95 percent of the values. The P50 corre- sponds to the median, the value that separates the lower half of the values from the upper half. Another way to think of percentiles is in terms of frequency. Again, the P95 value is greater than or equal to 95 percent of a distribution. This means that only 1 out of every 20 scenarios (5 percent of the values) resulted in values greater than the percentile amount. Figure 10.15 shows that the P95 for the payout amount distribution is $10.72 million, meaning that only 500 out of the 10,000 scenarios in the Monte Carlo model have claims that exceeded $10.72 million. We can also say that we expect to see values above the P95 only once every 20 years (500/10,000 = 1/20). Of course, such high payout amounts can occur multiple years in a row (each year having a 5 percent chance), but 1 in 20 years is the conventional way to describe and operationalize a probability of 5 percent. The P95 of a payout distribution is also known as the value at risk at the 95th percentile, or the VaR-95. A VaR value is always related to a certain time period, one year in our example. Therefore, when interpreting a VaR value it is important to always understand both the specific percentile it is reported at (P95 in our example) and the time period. VaR values for payout amounts with either higher percentiles or longer time periods will logically be higher. The tail value at risk (TVaR), also called the conditional value at risk (CVaR), is another important and related metric for describing future possible Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 132 Fundamentals of Probabilistic Modeling payouts for index insurance. The TVaR is the expected value over a restricted portion of a distribution. In figure 10.15, the TVaR-95 is the expected value of the upper 5 percent of the distribution, that is, the mean of payouts that are greater than or equal to the P95 value (the VaR-95). In this example, the TVaR- 95 is $13.77 million, which tells us that once in every 20 years we can expect an annual payout of this amount. As the histogram shows, payout values higher than $13.77 million are also possible. For example, the P99 of the payout dis- tribution is $15.58 million, and logically the TVaR-99 will be even higher. The fourth and final key metric this guide uses for summarizing probability distributions is the probability that a result is greater (or less) than a specific threshold. We can use any target value and identify the corresponding probability of the outcome being either lower or higher than this target value. For example, an insurer reviewing the payout distribution in figure 10.15 might want to know the probability that the payout amount in any given year will be greater than $15 million (which in this example is about 1.2 percent). Or the insurer might want to identify the probability of fund ruin, where the target value is the specific threshold of cash flows that, if not reached, will result in the ruin of the insurer’s business. Similar to finding the probability of an outcome greater than or less than a certain value, the probability of the result falling between two values can also be found by subtracting the percentile of the smaller value from the percentile of the larger. A classic application of this approach is to subtract the P2.5 from the P97.5 to get the values that bound the middle 95 percent of the distribution. This method is in essence the concept used to calculate 95 percent confidence intervals. 10.3.2  Outputs Used in Index Insurance Applications So far this chapter has discussed four key metrics for summarizing an outcome from a probabilistic analysis. This section discusses five output metrics that are specifically used in financial and insurance applications: • Probability of negative profits • Required capital • Economic value added (EVA) • Sharpe ratio • Return period These output metrics are used in the example models explained in chapters 11 through 15 of this guide. First, the probability of negative profits represents the probability that the insurer will experience a loss. Typically, the probability of a loss per week is dif- ferent from the probability of a loss per year, so this metric also needs to be defined over a certain period. For the models in this guide, the probability is defined per year, but the probability can also be calculated over a multiple-year period. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 133 In a Monte Carlo simulation model, the probability of a negative profit is calculated by simulating the annual profits and determining in what per- centage of the scenarios the profits are negative. For example, if, after run- ning 10,000 scenarios, 1,894 scenarios resulted in losses and the remainder in profit, the annual probability of a negative profit is estimated to be about 19 percent. Second, required capital indicates how much capital must be held by the insurer to be confident (within a certain confidence level, for example, 95 per- cent) that the payouts and expenses can all be paid from the premium income and the required capital. For example, if the required capital at a 95 percent confidence level is $100,000, then in 19 out of 20 years (95 percent) the insurer will have more than sufficient capital to cover all payouts from premium income and the $100,000 in required capital. However, in 1 out of 20 years, the insurer will have to pay out all of its capital to cover the payouts. In other words, the required capital is a TVaR value (often based on the 95th percentile or higher) for how much capital the insurer needs to pay for claims within a certain period. Two more index insurance–related outputs of a Monte Carlo simulation are discussed below: economic value added (EVA) and the Sharpe ratio. Both have required capital as one of their inputs. Because required capital is determined using a Monte Carlo simulation, two separate and sequential Monte Carlo simulations must be run to calculate these secondary metrics. The initial simu- lation estimates the required capital amount, given the variability in annual payouts by the insurer. In the example models, the output for this first Monte Carlo simulation is recorded as the “first simulation required capital.” The next simulation calculates the secondary metrics using the value for “first simula- tion required capital.” Wherever we use required capital to calculate a second- ary metric in the models in this guide we indicate that two separate and sequential Monte Carlo simulations need to be performed. Third, EVA is an estimate of the profits that the insurer can expect from the index insurance product in excess of the required rate of return. It is typically expressed as a percentage of the required capital, as follows: EVA = (Profits − Cost of capital)/Required capital; or more precisely, (Net premium income − Payout amounts − Required capital EVA =  × Required return on capital)/Required capital. Because the EVA depends on the payout amounts, which can vary greatly from year to year, the EVA can also vary greatly. Therefore, it is useful to report not only the expected EVA, but also the P5 and P95 of the EVA. Fourth, the Sharpe ratio, also called the reward-to-variability ratio, can be used to gain an understanding of the performance of an investment by adjusting for Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 134 Fundamentals of Probabilistic Modeling its risks. More precisely, the Sharpe ratio is the excess in return per unit of risk, and is typically defined as follows: (Expected return on capital − Risk-free rate)/ Sharpe ratio =  Standard deviation of the return on capital. Because the Sharpe ratio adjusts (that is, divides) the excess return (the return on the required capital that the insurer sets aside to cover potential payouts) by the standard deviation of the return on capital, investments with different risk profiles can be compared to understand which provides more excess return per unit of risk. Fifth, the return period provides an indication of how frequently certain events can be expected to occur. It represents the average period until the next event happens over the long term. For example, the return period for an index insurance product payout of at least 50 percent of the insured amount might be 14.6 years. In this case, we expect that over the long term the product will pay out at least 50 percent once every 14.6 years. Of course, because this is an aver- age metric, the product may pay out at least 50 percent more often, or there may be long periods in which it does not pay out at all because of year-to-year variability. 10.4  A Reminder on How to Use the Models in This Guide It is important to keep in mind that probabilistic and deterministic model building is not a simple, automatic process. Every model, by definition, is a simplification of reality with assumptions, strengths, weaknesses, and limitations. Different indi- viduals may end up with different models for the same problem because of differ- ences in judgment, experience, and preferences. This chapter provides some guidelines for choosing probability distributions for a given quantity. For the examples within this guide, we recommend using a beta distribution for representing and modeling payout ratios. However, the beta distribution cannot always be assumed to be the most appropriate distribution for this variable. When modeling payout ratios (which vary between 0 and 1), a Poisson distribution should never be used because a Poisson generates only discrete values such as 0, 1, 2, and so forth. However, depending on the data, a beta, gamma, or lognormal distribution could be appropriate.27 Examining GOF statistics can help an analyst decide, but the type of variable, the skewness, bounding, and any special purpose of the distribution should also be considered in determining which distribution best represents the variable. Sometimes models can be “wrong” because they violate basic principles of probability modeling (for example, using distributions in which the left tails ­ extend below zero to simulate potential payouts), but there can often be several ways to build a correct model. Nevertheless, building accurate and useful Monte Carlo simulation models is not solely an “art.” Building models requires expertise and knowledge of the different building blocks. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 135 The intent of this guide is to make you, the reader, an informed user of the probabilistic framework. Our models are not the ultimate, definitive models for index insurance. Rather than encouraging you to take up our models, add your data, and apply them blindly, we hope to help you develop the general skills necessary for developing, critiquing, and interpreting probabilistic mod- els for index insurance modeling. You may very well need to adapt, rebuild, or enhance our models to suit your unique circumstances, data, and decision questions. Finally, we want to again stress that clear and transparent communication of the assumptions behind any model you use or build is critical. Any simulation model you use to evaluate an index insurance product will have inherent assumptions, and you must clearly communicate to all relevant stakeholders these underlying hypotheses, assumptions, and potential limitations. Without this communication and understanding, stakeholders may not obtain the potential benefits (i.e., more informed decisions) of a probabilistic analysis of the index insurance product. Notes 1. To be precise, a quantitative risk analysis does not have to be probabilistic, although many quantitative risk analyses use probabilistic modeling. 2. We define the term “on average,” which is also referred to as “the mean,” later in this chapter, but the reader can think of it as the average number of dots that one would get if one rolled a dice many times. 3. As discussed in section 10.3, numerous quantitative metrics can be used to describe the amount of uncertainty we may have about what could happen in the future. 4. The distribution for a single one of these events or trials is a Bernoulli distribution (see section 10.2.1.2.1). 5. See the references at the start of this chapter for more complete lists of probability distributions. 6. See section 10.2.1.2.1 for more information on the Bernoulli distribution. 7. Depending on the country, income distributions tend to be more or less skewed. The degree to which the income distribution is skewed is often summarized with a statistic called the Gini coefficient, which measures inequality. See http://data.worldbank.org​ /indicator/SI.POV.GINI. 8. The name PERT is an acronym for project evaluation and review techniques, a project management system for which the PERT distribution was developed to include uncer- tainty in project timelines. The PERT distribution is discussed in more detail later in this chapter. 9. This actually is only the case for discrete distributions and not for continuous distribu- tions, as seen in the next section. 10. If a theoretically discrete variable takes extremely large values, we can treat it as continuous. For example, people disagree as to whether money is continuous or dis- crete, but risk models generally deal with very large values of money as continuous. Money at very high values behaves as if it were a continuous variable because the discrete steps (cents) are insignificantly small compared with the total values that Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 136 Fundamentals of Probabilistic Modeling might be modeled. If millions of dollars are modeled, the values appear continuous. When values closer to one dollar are modeled, the values appear discrete. 11. Note that “success” does not always mean a “good” outcome. It just means the out- come that we are focusing on for our model. If we want to model the failure of a certain kind of factory equipment, our success, the thing we are modeling, may be the failure of the machine. 12. Some texts call each Monte Carlo scenario or iteration a trial, which can be confusing when you discuss Bernoulli distribution trials. This guide uses the term “scenario” for Monte Carlo iterations. 13. Remember, “success” is the occurrence of the outcome we are analyzing, in this case an insurer basis risk event. A basis risk event is not a positive outcome, but it is the one we are investigating. 14. There is only one input parameter in a Poisson distribution, which is the multiplica- tion of lambda and t. In other words, the single rate input, lambda, can be normalized to any t exposure unit. If, for example, the lambda rate is 2 accidents per month, this rate can be normalized to 24 accidents per year, assuming that all calendar months have a rate of accidents of 2 per month. 15. When the beta distribution represents a probability, and the beta distribution is a parameter in another distribution (for example, the probability is used in a Bernoulli distribution), we call the fact that the probability itself is a distribution “parameter uncertainty.” See section 10.2.2.2. 16. In fact, this research and the mathematics behind the beta distribution are part of Bayesian statistics. Bayesian statistics is a large and important field of statistics that is worth studying further but falls outside the scope of this guide. 17. Methods to elicit expert opinion are beyond the scope of this guide, but O’Hagan et al. (2006) and EFSA Guidance (2014) provide a thorough guide to expert elicita- tion methods. Available from http://eu.wiley.com/WileyCDA/WileyTitle/productCd​ -0470029994.html and http://www.efsa.europa.eu/en/efsajournal/pub/3734, respectively. 18. This process of iterative fitting is not necessary for a number of distributions for which maximum likelihood estimations (MLEs) are available, such as the gamma, normal, Poisson, binomial, and exponential distributions. However, conceptually we can still think of distribution fitting this way for these distributions. 19. Information criteria are commonly used in statistics and are based on the MLE of each of the distributions and also take into account the number of parameters within each distribution, to avoid overfitting the data to a distribution with many parameters. 20. For more information about Bayesian analysis and priors, see Gelman (2013). 21. Monte Carlo models that include parameter uncertainty are typically slower because during each iteration, the parameter values need to be reestimated, which commonly takes considerable computing power. 22. Of course, it could be that some students had a score of exactly 42. In situations in which percentiles describe continuous variables, however, this is not an issue because with continuous distributions the probability of an outcome of exactly 42 is zero. 23. A uniform distribution is a probability distribution in which all the values within its range have equal probability. 24. Often such multi-to-multi variable relationships are summarized in a correlation matrix that describes the correlations between all variables. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Fundamentals of Probabilistic Modeling 137 25. “Variance” is also often used, which is the standard deviation squared. 26. The P2.5 and P97.5 are also common, since the range of values between both percen- tiles provides the 95 percent confidence level. 27. The lognormal is another commonly used continuous distribution. Bibliography Bolker, B. M. 2008. Ecological Models and Data in R. Princeton, NJ: Princeton University Press. http://ms.mcmaster.ca/~bolker/emdbook/. Brehm, P. J. 2007. Enterprise Risk Analysis for Property & Liability Insurance Companies: A Practical Guide to Standard Models and Emerging Solutions. New York: Guy Carpenter. Cherubini, U., E. Luciano, and W. Vecchiato. 2004. Copula Methods in Finance. Hoboken, NJ: John Wiley & Sons. EFSA (European Food Safety Authority). 2014. “Guidance on Expert Knowledge Elicitation in Food and Feed Safety Risk Assessment.” EFSA Journal 12 (6): 3734. Embrechts, P., F. Lindskog, and A. McNeil. 2003. “Modelling Dependence with Copulas and Applications to Risk Management.” In Handbook of Heavy Tailed Distributions in Finance, edited by S. T. Rachev, 329–84. Amsterdam: Elsevier. Forbes, C., and M. Evans. 2010. Statistical Distributions. 4th ed. Hoboken, NJ: Wiley. http://www​.wiley.com/WileyCDA/WileyTitle/productCd-0470390638.html. Gelman, A. 2013. Bayesian Data Analysis. 3rd ed. Boca Raton, FL: Chapman & Hall/CRC. http://www.stat.columbia.edu/~gelman/book/. Grossi, P., H. Kunreuther, and C. C. Patel. 2005. Catastrophe Modeling: A New Approach to Managing Risk. New York: Springer Science Business Media. Hertz, D. B. 1964. “Risk Analysis in Capital Investment.” Harvard Business Review. 1964: 95–106. Jewson, S., and A. Brix. 2005. Weather Derivative Valuation: The Meteorological, Statistical, Financial, and Mathematical Foundations. Cambridge: Cambridge University Press. Law, A. M., and W. D. Kelton. 2006. Simulation Modeling and Analysis. 4th ed. New York: McGraw-Hill. Lehman, D. E., H. Groenendaal, and G. Nolder. 2012. Practical Spreadsheet Risk Modeling for Management. Boca Raton, FL: Chapman & Hall/CRC. ModelAssist. A Free and Comprehensive Quantitative Risk Analysis Training and Reference Software. http://www.epixanalytics.com/ModelAssist.html. O’Hagan, A., C. E. Buck, A. Daneshkhah, J. R. Eiser, P. H. Garthwaite, D. J. Jenkinson, J. E. Oakley, and T. Rakow. 2006. Uncertain Judgements: Eliciting Experts’ Probabilities. Hoboken, NJ: John Wiley & Sons. Ragsdale, C. T. 2001. Spreadsheet Modeling and Decision Analysis: A Practical Introduction to Management Science. Cincinnati, OH: Southwestern College. Tang, A., and E. A. Valdez. 2009. “Economic Capital and the Aggregation of Risks Using Copulas.” University of New South Wales, Sydney, Australia. Yan, J. 2006. “Multivariate Modeling with Copulas and Engineering Applications.” In Springer Handbook of Engineering Statistics, edited by H. Pham, 973–90. London: Springer-Verlag. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 11 Evaluating the Base Index 11.1  Background and Objectives Chapter 4 explained the data required for the key managerial questions for Base Index product design and evaluation during the pilot phase of launching an index insurance business line. It described eight steps in the analysis and decision- making process and provided interpretations of each of the decision metrics (figure 4.5). The critical issue to remember about the Base Index is that because it provides such a high level of coverage, it is also very expensive and many policyholders will request a lower price—and lower coverage—product. However, it is extremely important that the insurer always produce a Base Index to explain to the policy- holder the difference between complete coverage—that provided by the Base Index—and the coverage provided by other product options. Without this explicit comparison, policyholders often fall into the trap of expecting complete coverage even when they have purchased a lower coverage, less expensive product. This chapter provides a step-by-step guide to using the probabilistic models that produce the decision metrics discussed in chapter 4 for evaluating the Base Index for product design basis risk. Using the Base Index’s historical payout and historical inventory damage ratios for selected geographical areas within a market, the model simulates four key scenario parameters: payout ratios (Steps ­ 14–18), inventory damage ratios (Steps 19–23), insured party basis risk ratios (Steps 24–29), and insurer basis risk ratios (Steps 30–35). The model then uses these four groups of parameters to calculate key basis risk decision metrics, including return periods, return period ratios, the probabil- ity of no basis risk event occurring, and the magnitude of the expected basis risk events (Steps 36–44). These metrics allow the insurer to determine whether a particular Base Index needs improvement or if index insurance is not an appro- priate risk management instrument for the prospective client. At the end of this chapter, section 11.5 provides a brief discussion of how retrospective analysis can also be used to evaluate the Base Index for product design basis risk. Risk Modeling for Appraising Named Peril Index Insurance Products   139   http://dx.doi.org/10.1596/978-1-4648-1048-0 140 Evaluating the Base Index Table 11.1  Summary of Model Components for Evaluating the Base Index Model component Section Excel sheet label Steps Description Model inputs 11.2 MI_11.2_MODEL Steps 1–7 User-defined assumptions, relevant portfolio INPUTS information, Base Index historical payout ratios, and historical inventory damage ratios are entered for all areas. Model computations 11.3.1 MC_11.3.1&.2_ Steps 8–10 Calculation of historical insured party basis DERIVED_INPUTS risk ratios for each area. These derived inputs are used in Steps 24–29. 11.3.2 MC_11.3.1&.2_ Steps 11–13 Calculation of historical insurer basis risk DERIVED_INPUTS ratios for each area. These derived inputs are used in Steps 30–35. 11.3.3 MC_11.3.3_BI_ Steps 14–18 Simulation of scenario payout ratios for SCENARIOS each area 11.3.4 MC_11.3.4_DR_ Steps 19–23 Simulation of scenario inventory damage SCENARIOS ratios for each area 11.3.5 MC_11.3.5_INSD Steps 24–29 Simulation of scenario insured party basis PARTY BASIS RISK risk amounts 11.3.6 MC_11.3.6_INSR Steps 30–35 Simulation of scenario insurer basis risk BASIS RISK amounts 11.3.7 MC_11.3.7_DECISION Steps 36–44 Calculation of product evaluation decision METRICS metrics Model output 11.4 MO_11.4_MODEL None Summary of product evaluation decision OUTPUTS metrics Table 11.1 provides a summary of the model components along with a guide to the sections in this chapter and the worksheet names in the accompanying Excel files. This chapter uses the same case example of a product design and evalua- tion process as in part 1. Wherever a box is labeled “case example,” screen shots of the model inputs, computations, or outputs for the case example are provided. 11.2  Model Inputs For the Base Index product evaluation, the analyst starts by specifying the model inputs agreed upon with the insurance manager for the Base Index product evaluation (table 11.2). Table 11.2  Model Inputs Model component Section Excel sheet label Steps Description Model inputs 11.2 MI_11.2_MODEL Steps 1–7 User defined assumptions, relevant INPUTS portfolio information, Base Index historical payout ratios, and historical inventory damage ratios are entered for all areas. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Base Index 141 11.2.1  Significant Payout and Damage Levels (Step 1) The magnitude of the payout or damage that is being considered is among the user-defined assumptions used by the model. For the case example, the ana- lyst specifies 10 percent inventory damage as a mild loss, 30 percent as a mild- to-medium loss, 50 percent as a medium loss, and 70 percent as a severe loss (case example box 11CB.1). Case Example Box 11CB.1  Inputs—Step 1 11.2.2  Prediction Interval (Step 2) In Step 2, the analyst specifies the prediction interval based on the insurer’s desired level of accuracy. The upper limit of the interval is used in calculat- ing capital requirements. For example, if the insurer wants to hold capital at 99 percent tail value at risk (TVaR) (the payout amount for a 1-in-100 year event), the upper limit should be set at 99 percent. In the case example, the insurance manager and the analyst specify the upper limit as 95 percent (case example box 11CB.2). The prediction interval between the 5th (low) and 95th percentile is often called the 90 percent prediction interval. Case Example Box 11CB.2  Inputs—Step 2 11.2.3  Total Sum Insured per Area (Step 3) The total sum insured per area (case example box 11CB.3) will be used in simu- lating insured party and insurer basis risk amounts (during Steps 24–35). Case Example Box 11CB.3  Inputs—Step 3 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 142 Evaluating the Base Index 11.2.4  Historical Payout Ratios (Step 4) Historical payout ratios for the Base Index (case example box 11CB.4) will be used for determining scenario payout ratios (Steps 14–18), basis risk amounts (Steps 24–35), and return periods (Steps 36–44). Case Example Box 11CB.4  Inputs—Step 4 11.2.5  Independent Historical Inventory Damage Data (Step 5) Chapter 4 discussed the process for collecting independent historical inven- tory damage data and completing a qualitative classification of past damages for product design and evaluation. Based on data from multiple sources, the product design team rates the level of crop damage caused by the named peril—in each year for each geographical area—from 1 to 5, where 5 is the damages from drought and 1 represents mild damages from drought highest ­ (case example box 11CB.5). In years with a rating of 1, farmers experienced up to a 20 percent loss in yields. In years with a rating of 2 the yield loss was 21 to 40 percent. In years with a rating of 3 the loss was 41 to 60 percent, and so on. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Base Index 143 Case Example Box 11CB.5  Inputs—Step 5 11.2.6  Historical Inventory Damage Ratios (Step 6) The independent historical inventory damage data classifications are provided as categorical data, as shown in Step 5. The analyst then converts them into damage percentages using the midpoints of the damage ranges for each category (case example box 11CB.6). For example, a damage classification of 1 corresponds to the damage ratio 10 percent (midpoint between 0 and 20 percent), a damage classification of 2 corresponds to the damage ratio 30 percent (midpoint between 21 and 40 percent), and so on. Note that in cases in which actual historical inventory damage ratios are avail- able, the analyst should of course use these values for Step 6. Later sections in this chapter provide estimates for the magnitude of insurer and insured party basis risk for the insurer to use in evaluating the quality of the Base Index. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 144 Evaluating the Base Index Case Example Box 11CB.6  Inputs—Step 6 Let’s look at Area B from the case example. We can see from Step 6 that this area had three years with inventory damage events (30 percent in 1986, 30 percent in 1989, and 10 percent in 2003). But we know that the Base Index triggered five payouts (case example box 11CB.4). The Base Index triggered at least two unnecessary payouts. A closer look shows that of the five years in which payouts were triggered, only one (1989) corresponds to a year in which inventory damage occurred. So, the index actually triggered four unnecessary payouts from the insurer to the insured party, all of which are examples of insurer basis risk. That still leaves two years in which inventory damage occurred but no payout would have been trig- gered (1986 and 2003). In these cases, the insured party would have experienced inventory damage but received no payout, both examples of insured party basis risk. 11.2.7  Nonzero Historical Payout and Inventory Damage Ratios (Step 7) In this step the analyst manually records all the nonzero values for the historical payout ratios and historical inventory damage ratios from Steps 4 to 6 (case example box 11CB.7). These inputs will be used in the simulation of payout ratios (Steps 14–18) and inventory damage ratios (Steps 19–23). Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Base Index 145 Case Example Box 11CB.7  Inputs—Step 7 11.3  Model Computations The model completes seven sequential sets of computations for the Base Index product evaluation, starting with calculating the derived inputs—historical basis risk ratios (Steps 8–13)—then simulating four key scenario parameters (Steps 14–35), and finally determining the product evaluation decision metrics (Steps 36–44) (table 11.3). Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 146 Evaluating the Base Index Table 11.3  Model Computations Model component Section Excel sheet label Steps Description Model computations 11.3.1 MC_11.3.1&.2_ Steps 8–10 Calculation of historical insured party basis DERIVED INPUTS risk ratios for each area. These derived inputs are used in section 11.3.5 (Steps 24–29). 11.3.2 MC_11.3.1&.2_ Steps 11–13 Calculation of historical insurer basis risk DERIVED INPUTS ratios for each area. These derived inputs are used in section 11.3.6 (Steps 30–35). 11.3.3 MC_11.3.3_BI_ Steps 14–18 Simulation of scenario payouts for each area SCENARIOS 11.3.4 MC_11.3.4_DR_ Steps 19–23 Simulation of scenario inventory damage SCENARIOS ratios for each area 11.3.5 MC_11.3.5_INSD Steps 24–29 Simulation of scenario insured party basis PARTY BASIS RISK risk amounts 11.3.6 MC_11.3.6_INSR Steps 30–35 Simulation of scenario insurer basis risk BASIS RISK amounts 11.3.7 MC_11.3.7_DECISION Steps 36–44 Calculation of product evaluation METRICS decision metrics 11.3.1  Calculation of Historical Insured Party Basis Risk Ratios (Steps 8–10) Table 11.4  Model Computations Model component Section Excel sheet label Steps Description Model 11.3.1 MC_11.3.1&.2_ Steps 8–10 Calculation of historical insured Computations DERIVED INPUTS party basis risk ratios for each area. These derived inputs are used in section 11.3.5 (Steps 24–29). 11.3.1.1 Overview Insured party basis risk describes the scenario in which the payout amount is less than the farmer’s actual losses caused by the named peril. In this case the farmer experiences an economic loss from the named peril but is not adequately com- pensated by the claim payout. The insured party basis risk ratio (table 11.4) for the Base Index is calculated as follows (case example box 11CB.8): Historical insured party = Max (0, [Historical inventory − Historical payout ratio]). basis risk ratio damage ratio Step 6 Step 4 Any time that the historical payout ratio is larger than the historical inventory damage ratio, the insured party basis risk is zero. In this situation, insurer basis risk will be greater than zero because the insurer will have paid out more than the actual losses (see section 11.3.2). Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Base Index 147 11.3.1.2  Implementation in Excel (MC_11.3.1&.2_Derived Inputs) Case Example Box 11CB.8  Computations—Step 8 The table shows that three insured party basis risk events occurred in Area B in 1986, 1989, and 2003. For example, for 1986 the historical insured party basis risk ratio is Historical insured party basis risk ratio = Max (0, [Historical inventory damage ratio − Historical payout ratio]) = Max (0, 30% − 0%) = 30%. The same calculation is used for all 30 years and all 10 regions. In Step 9 (no case example box provided) the model reorders the historical insured party basis risk ratios from the most recent year at the top to the least recent year at the bottom. The sorted years will be the key inputs for the model’s determination of historical years with the largest insured party basis risk ratios in Step 41. In Step 10, the analyst manually combines all the nonzero basis risk ratios from Step 8 into one list (case example box 11CB.9). In Step 24 the model will use only these values to fit a beta probability distribution for the simulation of insured party basis risk ratios. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 148 Evaluating the Base Index Case Example Box 11CB.9  Computations—Step 10 11.3.2  Calculation of Historical Insurer Basis Risk Ratios (Steps 11–13) 11.3.2.1 Overview Insurer basis risk describes the scenario in which the payout is greater than the actual losses the insured party experiences from the named peril (table 11.5). In this case the insurer suffers an economic loss because of unnecessary claims payments. The calculation for the historical insurer basis risk ratio for the Base Index is as follows: Historical insurer = Max (0, [Historical payout − Historical inventory basis risk ratio ratio ­ damage ratio]). Step 6 Step 4 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Base Index 149 Table 11.5  Model Computations Model component Section Excel sheet label Steps Description Model computations 11.3.2 MC_11.3.1&.2_ Steps 11–13 Calculation of historical DERIVED INPUTS insurer basis risk ratios for each area. These derived inputs are used in section 11.3.6 (Steps 30–35). The case example results are shown in case example box 11CB.10. Any time the historical payout ratio is smaller than the historical inventory damage ratio, the insurer basis risk is zero. 11.3.2.2  Implementation in Excel (MC_11.3.1&.2_Derived Inputs) Case Example Box 11CB.10  Computations—Step 11 Step 11 shows that the Base Index generates four insurer basis risk events in Area B during the 30-year period. Step 12 (no case example box provided) is similar to Step 9. The model reorders the insurer basis risk ratios from the most recent year to the least recent. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 150 Evaluating the Base Index Case Example Box 11CB.11  Computations—Step 13 In Step 13, the analyst manually combines all the nonzero basis risk ratios from Step 11 into one list (case example box 11CB.11). These will be used in Step 30 to fit a beta probability distribution for the simulation of insurer basis risk ratios. 11.3.3  Simulation of Scenario Payout Ratios (Steps 14–18) 11.3.3.1 Overview Based on the historical payout ratios (Step 4), the model simulates the scenario payout ratios (table 11.6) for the Base Index using estimates for three stochastic elements: • Frequency distribution (Step 14): This distribution describes the frequency of payouts. Because most index insurance products will only pay out once a year (or season) and not multiple times, a Bernoulli distribution is the most appropri- ate for the frequency distribution. The frequency distribution describes the prob- ability of a payout for each area based on the historical frequency of payouts. • Severity distribution (Step 15): If a payout is made within a certain area, the percentage of the insured amount that needs to be paid out can vary widely. Some years the payout may be only 10 percent of the sum insured, while Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Base Index 151 Table 11.6  Model Computations Model component Section Excel sheet label Steps Description Model computations 11.3.3 MC_11.3.3_BI_ Steps 14–18 Simulation of scenario payout ratios for SCENARIOS each area in other years with more severe weather the payout may be closer to 100 ­percent. The severity distribution describes the variability in payout ratios for each area based on the historical severity of payouts. • Correlation function (Steps 16–17): The occurrence of payouts in nearby areas or regions is typically codependent because of weather patterns. Severe weather and high payouts in one area often coincide with severe weather and high payouts in an adjoining area. The distribution—a copula in this case— describes the degree of correlation between payout ratios for each area based on the historical correlation of payouts. Figure 11.1 summarizes the process for simulating scenario payout ratios from historical payout ratios for each area. Figure 11.1  Generating Scenario Payout Ratios Historical payout ratios (Step 4) Frequency distribution Correlation function Severity distribution (Step 14) (Steps 16 and 17) (Step 15) Scenario payout ratio (Step 18) Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 152 Evaluating the Base Index Once the model estimates the frequency, severity, and correlation for each area, it simulates the payout ratios per area. 11.3.3.2  Implementation in Excel (MC_11.3.3_BI_Scenarios) Case example boxes 11CB.12–11CB.14 show the simulation of scenario payout ratios for the case example. Case Example Box 11CB.12  Computations—Step 14 Looking at Area B from the case example, case example box 11CB.4 shows that this area had five years with payouts greater than zero (payouts of 15 percent in 1989, 20 percent in 1997, 2.5 percent in 2009, 7.5 percent in 2011, and 5 percent in 2012) and hence 25 years with no payouts. Without taking uncer- tainty into account, the p parameter for this area would be the proportion of years with payouts to the total number of historical years. Annual probability of a payout = Number of historical years with payouts/Total number of historical years = 5/30 = 16.7% distribution is used to simu- However, because we do need to take uncertainty into account, the beta ­ late the uncertainty in the annual probability of payouts. Annual probability of a payout = (p) ~ beta distribution (alpha, beta) = (p) ~ beta distribution (6, 26) This simulation for Area B is shown in the Step 14 table as Beta (6;26), which is a beta distribution with alpha = 6 and beta = 26. In Step 14, the model fits a beta distribution to the nonzero historical payout to estimate the annual probability of a payout that is greater than zero. This estimated probability is the parameter p, which is used in the Bernoulli distribu- tion (frequency distribution). With a large amount of data, this parameter could be estimated as a propor- tion of successes (number of years with a payment) to trials (total number of years). Number of historical years with payouts/ Annual probability of a payout =  Total number of historical years Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Base Index 153 However, because this sample is very small, we need to take uncertainty into account by estimating p for each scenario using a beta distribution. Thus, the annual probability of a payout is simulated as follows: Annual probability of a payout = (p) ~ beta distribution (alpha, beta), where alpha = [Number of historical years with payouts > 0] + 1 beta = Number of historical years with no payouts + 1. Case Example Box 11CB.13  Computations—Step 15 In the case example, the number of nonzero historical payout ratios is very low for each area (between two and six). Therefore, instead of using only the data points for each area, the model assumes that the severity distribution for each area is the same as that for all areas. In other words, the case example model is assuming homogeneity of payout ratios for all of the areas. For each area the model fits a beta distribu- tion using all 54 nonzero historical payout ratios (all 54 data points of all 10 areas). Each severity scenario that will be generated during the simulation process will come from this beta distribution for all nonzero payout ratios observed across all years in all areas. This simulation for Area B is shown in the Step 15 table as Beta (0.65;1.07). In Step 15, the model fits a beta distribution to the nonzero historical ­ ayout ratios to estimate payout severity (case example box 11CB.13). Other p probability distributions can be used, but the beta distribution, which has a minimum of 0 and a maximum of 1, tends to fit data that ranges from 0 to 1 ­ (that is, 100 percent). Case Example Box 11CB.14  Computations—Steps 16–17 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 154 Evaluating the Base Index Steps 16–17 fit a copula to the historical payout ratios to estimate the correla- tion between annual payouts in each area (case example box 11CB.14). In Step 16A the model determines the best-fitting copula for the historical payout ratios. Different copulas can be used in this step. In quantitative finance the two commonly used copulas are the normal (or Gaussian) copula and the t copula (see chapter 10 and Cherubini, Luciano, and Vecchiato [2004]). However, because of the long-tail nature of risks insured through named peril index insur- ance, the use of the normal copula would not be appropriate. We instead do recommend using either the t copula (see chapter 10) or other copulas with the ability to capture tail dependence (see Cherubini, Luciano, and Vecchiato [2004] for more details about tail dependence and the difference between alternative copulas). In Step 16B, the model estimates the parameters of the copula from the historical payout ratio data. Finally, in Step 17, the model simulates the copulas. It is important to note that the greater the correlation between the areas, the greater the total amount that the insurer may have to pay out in a season for all areas together. In other words, higher correlation causes higher risk to the insurer (and possibly reinsurer). Potentially higher payouts caused by highly correlated exposure will mean that the insurer must hold more required capital, which involves greater cost. In these circumstances, finding areas with less correlated exposure or obtaining additional reinsurance can help the insurer reduce the costs of the required capital. Case Example Box 11CB.15  Computations—Step 18 For Area B, the aggregate annual payout ratio is Aggregate annual payout ratio = ~ Frequency (0 or 1) × Severity = ~ Bernoulli (p) × Beta = 1 × 0.43 = 43 percent. In Step 18, the simulation incorporates all three stochastic elements discussed above (case example box 11CB.15): • The frequency of payouts for each area (Step 14) • The severity of payouts for each area (Step 15) • The correlation between payouts in all areas (Steps 16–17) Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Base Index 155 Step 18 combines all three elements to simulate annual payout ratios for each of the areas. The model generates a copula that represents the correlation between and among all the different areas (Step 17). Values picked from this copula will determine the frequency value for each area (case example box 11CB.16). Case Example Box 11CB.16  Determining Area Level Scenario Frequency and Severity Values For Area B, Step 17 picked the 81st percentile from the copula, so the model selects the 81st percentile value from the frequency distribution (Bernoulli) for Area B (Step 14). This value is 1, indicating a payout is expected to occur. The model also generates a severity value for Area B from the severity distribution (beta distribution), which equals 0.43 in the example (Step 15). Note that only the frequency value, and not the severity value, is determined by the copula. Step 18 combines the frequency and severity distributions into an aggregate payout ratio simulation for each area. Aggregate annual payout ratio = ~ Frequency (0 or 1) × Severity = ~ Bernoulli (p) × Beta Step 14 Step 15 The scenario payout ratios for each area are calculated in the same way. Note that whenever the frequency distribution (Bernoulli distribution) simu- lates a 0, meaning there is no payout expected, we ignore the value of the sever- ity distribution (beta distribution). This makes sense conceptually because if no payout occurs then the payout’s magnitude does not matter. Mathematically, we can see that multiplying the frequency value (zero) by any severity value will produce a payout ratio of zero. Taking this a bit further, if we know that the payout ratio for an area is zero, then we also know that the frequency (Bernoulli) distribution generated a zero. 11.3.4  Simulation of Scenario Inventory Damage Ratios (Steps 19–23) 11.3.4.1 Overview This set of model computations simulates the scenario inventory damage ratios for the Base Index (table 11.7). Later in this chapter, the model compares the scenario inventory damage ratios to the scenario payout ratios (Steps 14–18) to evaluate the basis risk of the Base Index. Table 11.7  Model Computations Model component Section Excel sheet label Steps Description Model computations 11.3.4 MC_11.3.4_DR_ Steps 19–23 Simulation of scenario inventory SCENARIOS damage ratios for each area Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 156 Evaluating the Base Index 11.3.4.2  Implementation in Excel (MC_11.3.4_DR_Scenarios) The steps for simulating the scenario inventory damage ratios (Step 19–23; case example box 11CB.17) are similar to those for simulating the payout ratios (Steps 14–18) but use different model inputs. Instead of using nonzero historical payout ratios, the model uses nonzero historical inventory damage ratios. Here the focus is on simulating inventory damage caused by the named perils rather than on simulating the payouts triggered by the Base Index. Case Example Box 11CB.17  Computations—Steps 19–23 11.3.5  Simulation of Scenario Insured Party Basis Risk Amounts (Steps 24–29) 11.3.5.1 Overview The simulation of scenario insured party basis risk amounts quantifies the insured party basis risk losses for the Base Index (table 11.8). Table 11.8  Model Computations Model component Section Excel sheet label Steps Description Model computations 11.3.5 MC_11.3.5_INSD Steps 24–29 Simulation of scenario insured party basis PARTY BASIS RISK risk amounts 11.3.5.2  Implementation in Excel (MC_11.3.5_Insd Party Basis Risk) The process for simulating scenario insured party basis risk amounts is similar to that for simulating scenario payout ratios (Steps 14–18) but with two key differences. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Base Index 157 First, the model’s main inputs are the nonzero historical insured party basis risk ratios (Step 10) instead of the nonzero historical payout ratios. Second, Steps 28 and 29 simulate monetary amounts for insured party basis risk rather than just a ratio. Case example box 11CB.18 shows the simulation of scenario insured party basis risk amounts for the case example. Case Example Box 11CB.18  Computations—Steps 24–29 In the case example, the scenario insured party basis risk amounts for Areas E ($148,846), H ($186,609), and I ($24,774) sum to the total insured party basis risk amount of $360,229. In Step 28, the model calculates insured party basis risk amounts for each area by multiplying the insured party basis risk ratios (Steps 8–10) by the total sum insured per area (Step 3). In Step 29, the model sums the basis risk amounts for all the areas to give a scenario total basis risk amount. The greater the frequency and severity of insured party basis risk events (that is, false negatives), the greater the amount of insured party basis risk. 11.3.6  Simulation of Scenario Insurer Basis Risk Amounts (Steps 30–35) 11.3.6.1 Overview The simulation of scenario insurer basis risk amounts quantifies the insurer basis risk losses for the Base Index (table 11.9). Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 158 Evaluating the Base Index Table 11.9  Model Computations Model component Section Excel sheet label Steps Description Model computations 11.3.6 MC_11.3.6_INSR BASIS RISK Steps 30–35 Simulation of scenario insurer basis risk amounts 11.3.6.2  Implementation in Excel (MC_11.3.6_Insurer Basis Risk) The steps for simulating the scenario insurer basis risk amounts (case example box 11CB.19) are similar to those for simulating insured party basis risk amounts (Steps 24–29) but use different model inputs. The input values are the nonzero historical insurer basis risk ratios (Step 13). The reader is referred back to Steps 24–29 for further details on the modeling. Case Example Box 11CB.19  Computations—Steps 30–35 In the case example, the basis risk amounts for Areas B ($52,322), D ($20,786), and F ($27,799) sum to the total insurer basis risk amount of $100,907. 11.3.7  Calculation of Product Evaluation Decision Metrics (Steps 36–44) At this point the model has simulated four key scenario parameters (payout ratios, inventory damage ratios, insured party basis risk amounts, and insurer basis risk amounts) for evaluating the Base Index. Based on these parameters, the model then calculates a number of important metrics that help gauge the level of basis risk inherent in the Base Index (table 11.10). Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Base Index 159 Table 11.10  Model Computations Model component Section Excel sheet label Steps Description Model computations 11.3.7 MC_11.3.7_DECISION METRICS Steps 36–44 Calculation of product evaluation decision metrics 11.3.7.1  Expected Return Periods for Inventory Damage and the Base Index 11.3.7.1.1 Overview. The expected return period for inventory damage is the expected frequency at which inventory damage caused by the named peril occurs at specific damage levels (for example, damage to 10 percent of the inven- tory, 30 percent of the inventory, 50 percent of the inventory, and 70 percent of the inventory). The expected return period for the Base Index is the frequency at which the Base Index makes a payout at specific payout levels. Figure 11.2 provides an overview of how the model estimates the expected return periods for both inventory damage and the Base Index. Figure 11.2  Generating Expected Return Periods for Inventory Damage and the Base Index Scenario damage or payout ratios (Steps 18 and 23) Significant damage or payout level (Step 1) Run at least 10,000 Monte Carlo scenarios Exceedance probability at defined damage or payout (Steps 36 and 37) Projected return period at defined damage or payout levels (Steps 36 and 37) Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 160 Evaluating the Base Index 11.3.7.1.2  Implementation in Excel. Case example box 11CB.20 shows the simu- lation of expected return periods for inventory damage events and Base Index payouts. Case Example Box 11CB.20  Computations—Steps 36–37 For Area B in the case example, inventory damage was greater than 10 percent (mild damage) in 1,100 of the 10,000 Monte Carlo scenarios (not shown box steps, the model automatically counts these). Number of scenarios with inventory damage > significant damage level/ Exceedance probability =  Total number of scenarios = 1,100/10,000 = 11 percent At the 10 percent damage level the exceedance probability is 11 percent for Area B. In other words, there is an 11 percent probability that the inventory damage level will be greater than 10 percent in Area B. In the case example, the inventory damage return period for Area B at the 10 percent damage level is nine. Return period = 1/Exceedance probability = 1/11 percent =9 In Area B, the next 10 percent or greater damage level is expected to occur in nine years. In other words, we expect that once in every nine years, the inventory damage level in Area B will be greater than 10 percent. In Step 36, at least 10,000 scenario inventory damage ratios (Monte Carlo simulations) are generated for each significant damage level (Step 1) and area. The proportion of the 10,000 scenarios with damage ratios greater than each significant damage level (10 percent, 30 percent, 50 percent, and 70 percent) is the exceedance probability:  umber of scenarios with inventory damage > significant damage level/ Exceedance probability = N Total number of scenarios. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Base Index 161 Also in Step 36, the model calculates the inventory damage return period for each damage level and area as Return period = 1/Exceedance probability. In Step 37, the model calculates the exceedance probabilities and expected return periods for Base Index payouts for each of the four significant payout levels and areas using the same equations as used in Step 36. 11.3.7.2  Return Period Ratio 11.3.7.2.1 Overview. The return period ratio shows the level of insurer or insured party basis risk for each significant damage or payout level and area. When this ratio is equal to 1, the Base Index triggers a payout at the same frequency as the occurrence of actual inventory damage events. When the ratio is greater than 1, the Base Index triggers a payout more frequently than the occurrence of insured events (insurer basis risk). When the ratio is between 0 and 1, the Base Index triggers payouts less frequently than the occurrence of actual insured events ­ (insured party basis risk). 11.3.7.2.2 Implementation in Excel (MC_11.3.7_Decison Metrics). The return period ratio metric is calculated as Return period ratio = Inventory damage return period/Base Index return period. Case example box 11CB.21 shows the calculation of the return period ratios. Case Example Box 11CB.21  Computations—Step 38 In the case example, the inventory damage return period at the 10 percent damage level was nine years for Area B. The Base Index return period at the 10 percent damage level was seven years for Area B. Inventory damage attributable to the named peril occurs once in every nine years, but the Base Index pays once in every seven years. The index is paying more frequently than is necessary, thus leading to insurer basis risk. The return period ratio is 1.25. Return period ratio = Inventory damage return period/Base Index return period = 9.0507/7.2366 = 1.25 The Base Index return period ratio is greater than 1, which confirms the presence of insurer basis risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 162 Evaluating the Base Index 11.3.7.3  Insured Party Basis Risk Statistics 11.3.7.3.1 Overview. When the return period ratio is between 0 and 1, the Base Index is triggering payouts less frequently than actual inventory damage events, indicating the presence of insured party basis risk. However, the return period ratio does not tell us whether the Base Index is triggering in the right years and for the right amounts. Even with a return period ratio of 1, the Base Index may still have insured party basis risk. This section explains the calculation of additional statistics that further describe the Base Index’s insured party basis risk. Figure 11.3 provides an overview of how the model simulates the probability of having no insured party basis risk events and the expected insured party basis risk amount (Steps 39–40). Figure 11.4 provides an overview of how the model determines the historical years with the largest insured party basis risk ratios (Step 41). Figure 11.3  Generating Probability of Having No Insured Party Basis Risk Event and Expected Insured Party Basis Risk Amount Scenario insured party basis risk amounts (Step 29) Run at least 10,000 Monte Carlo scenarios Prediction interval (Step 2) Insured party basis risk amount Probability that the Base • Lower Index will not experience an • Expected insured party basis risk event • Upper (Step 39) • TVaR (Step 40) Note: TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Base Index 163 Figure 11.4  Generating Historical Years with the Largest Insured Party Basis Risk Ratios Historical inventory Base Index damage ratios historical payout ratios (Step 6) (Step 4) Historical insured party basis risk ratios (Steps 8 and 9) Historical years with the largest insured party basis risk ratios • Year • Historical basis risk ratio • Historical payout ratio • Historical inventory damage ratio (Step 41) 11.3.7.3.2  Implementation in Excel (MC_11.3.7_Decison Metrics). In Step 39 (case example box 11CB.22), the model runs at least 10,000 scenarios for each area based on the scenario insured party basis risk amounts and calculates the propor- tion of the scenarios in which the insured party basis risk amount was zero. This value indicates the percentage of years when no insured party basis risk events are expected for each area. Put differently, this figure is the probability that no insured party basis risk event will occur during the next risk period for each of the areas. In Step 40, the model uses the same 10,000 scenarios to determine the expected amount of insured party basis risk for the portfolio (all geographical areas). This amount is reflected in currency terms as well as a percentage of the total sum insured. Based on the prediction interval selected in Step 2, the model also calculates the appropriate percentile and TVaR values. These values indicate the expected magnitude of the Base Index’s insured party basis risk. Note that when the percentage of years with no insured party basis risk is higher, the mag- nitude of the basis risk is lower and vice versa. These insured party basis risk metrics also provide a good starting point for an insurer that is pricing the Base Index, as discussed in detail in chapter 12. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 164 Evaluating the Base Index Case Example Box 11CB.22  Computations—Steps 39 and 40 Step 39 shows that Area B has an 88 percent chance of having no insured party basis risk event in the next risk period. Step 40 shows that the expected insured party basis risk amount is $276,655, which is 3 percent of the ­portfolio’s total sum insured. The prediction interval for the case example is 90 percent (Step 2), so the model also shows the 5th and 95th percentiles and the TVaR 95 percent. The TVaR 95 percent tells us that for a 1-in-20 year event, the insured party basis risk amount for all the areas is expected to be as high as $1,310,579. Note: TVaR = tail value at risk. In addition to the insured party basis risk metrics discussed above, memorable years in which the Base Index would have failed to trigger or triggered inade- quate payouts will be of interest to the prospective policyholder. A product that fails to trigger in years that are considered catastrophic has low client value and should not be promoted. Case example box 11CB.23 shows the calculation of the historical years with the largest insured party basis risk ratios. In Step 9, the model reordered the insured party basis risk ratios from the most recent to least recent year for each area. In Step 41, the model now selects the years with the largest insured party basis risk ratios. In areas where this value—the largest basis risk ratio—is repeated across multiple years, the model selects the most recent of these years. The most recent year events are chosen because prospective policyholders are more likely to remember these than older events. Next, the model selects each year’s corresponding historical payout ratio and historical inventory damage ratio. Case Example Box 11CB.23  Computations—Step 41 For Area B in the case example, the largest historical basis risk ratio was 30 percent, which occurred in 1986. In that year, if the Base Index contract had been in place, the insured party would have suffered inventory damage of about 30 percent but the index would not have triggered. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Base Index 165 The insurance manager can use these years as examples when explaining the limitations of the coverage provided by the Base Index to the policyholder. 11.3.7.4  Insurer Basis Risk Statistics 11.3.7.4.1 Overview. Next the model calculates the metrics for insurer basis risk. These metrics provide more detail about the amount of payouts the insurer can expect as a result of insurer basis risk (case example box 11CB.24). Case Example Box 11CB.24  Computations—Steps 42–44 For the case example, there is an 84 percent probability of Area B having no insurer basis risk event in the next risk period. For the whole portfolio, the expected insurer basis risk value is $159,939 (that is, 2 percent of the sum insured for the portfolio). The tail value at risk indicates that the insurer basis risk is expected to be $1,009,740 (13 percent of portfolio value) once every 20 years. The largest insurer basis risk amount in Area B—20 percent of the insured amount—occurred most recently in 1997 when the Base Index would have triggered a 20 percent payout despite no inventory damage caused by the named peril. Note: TVaR = tail value at risk. 11.3.7.4.2 Implementation in Excel (MC_11.3.7_DECISION METRICS). The steps for calculating the insurer basis risk statistics are similar to those for calcu- lating the insured party basis risk statistics (Steps 39–41) but use different model inputs. Instead of using scenario insured party basis risk amounts, the model now uses scenario insurer basis risk amounts. Please refer to Steps 39–41 for a detailed explanation of the process. 11.4  Model Outputs The model output sheet summarizes the product evaluation decision metrics (table 11.11) for the Base Index produced in Steps 36–44, including the following: • Inventory damage return periods for each area • Base Index return periods for each area • Return period ratios for each area Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 166 Evaluating the Base Index Table 11.11  Model Outputs Model component Section Excel sheet label Steps Description Model outputs 11.4 MO_11.4_MODEL OUTPUTS None Summary of product evaluation decision metrics. Box 11.1  Overview of Calculations for the Base Index Product Evaluation Metrics Scenario metrics (one Monte Carlo scenario) • Historical insured party basis risk ratio = Max (0, [Historical inventory damage ratio − Historical payout ratio]) • Historical insurer basis risk ratio = Max (0, [Historical payout ratio − Historical inventory dam- age ratio]) • Annual probability of a payout = Number of historical years with payouts/Total number of historical years • Annual probability of a payout = (p) ~ beta distribution (alpha, beta) • Size of basis risk ~ beta distribution, based on fit to empirical data • Correlation of basis risk ~ t copula, based on fit to empirical data, where alpha = [Number of historical years with payouts > 0] + 1 beta = Number of historical years with no payouts + 1 • Aggregate annual payout ratio per area ~ Frequency (0 or 1) × Severity ~ Bernoulli (p) × beta Metrics based on at least 10,000 Monte Carlo scenarios • Exceedance probability = (Number of scenarios with inventory damage > significant damage level)/Total number of scenarios • Return period = 1/Exceedance probability • Return period ratio = Inventory damage return period/Base Index return period • Probability that no insured party basis risk event will occur during the next risk period = Number of scenarios where basis risk = 0/Total number of scenarios • Probability that the Base Index will not experience an insured party or insurer basis risk event in the next risk period for each area • Expected amount of insured party and insurer basis risk for the portfolio • Historical years with largest insured party and insurer basis risk events for each area The insurance manager uses these metrics in chapter 4 to answer the key managerial questions for evaluating the Base Index (see box 11.1). See case example box 11CB.25 for outputs. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Base Index 167 Case Example Box 11CB.25  Outputs Note: TVaR = tail value at risk. 11.5 Alternative Modeling Approach: Retrospective Analysis This chapter provides a step-by-step guide to using probabilistic models to evalu- ate the Base Index for product design basis risk. Chapter 16 discusses two addi- tional probabilistic modeling approaches for calculating these metrics. This section briefly describes a different type of nonprobabilistic analysis that is also used in index insurance product design and evaluation: retrospective analysis. A retrospective analysis can be used to evaluate the Base Index for basis risk. The key inputs to the retrospective analysis are the historical payout ratios and historical inventory damage ratios also used in the probabilistic approach (Steps 4 and 6). In the probabilistic approach, these inputs were used to simulate projected future values for both ratios, which were then compared to evaluate the basis risk of the Base Index. The retrospective approach does not require simulation of any projected future values. Instead, historical inventory damage ratios are simply compared to historical payout ratios. The analysis is based only on historical values. The model tests the predictive power of the Base Index in retrospect. First, the model identifies the years for each area where the inventory damage ratio and payout ratios were both high, meaning that the Base Index correctly triggered a high payout that corresponded to high inventory damage caused by the named peril (case example box 11CB.26). Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 168 Evaluating the Base Index Case Example Box 11CB.26 Review of Base Index Performance for Historical Events with Greater than 50 Percent Damage Level For the case example, we set the level for a high inventory damage ratio as 50 percent and greater and the level for a high payout ratio as 30 percent and greater. Looking back at the historical inven- tory ­damage ratios in Step 6, we see that out of the 300 data points (that is, 10 areas for 30 years), 23 have damage ratios of at least 50 percent. The total number of historical events with damage ratios greater than 50 percent is 23 for the case example. When these 23 data points are compared with the corresponding points for the historical payout ratios, we see that 22 of them also triggered payouts of at least 30 percent. These are the years for which the Base Index correctly triggered a high payout that corresponded to high inventory damage caused by the named peril. Only one historical event with a high damage level—Area I in 1984—triggered a payout of less than 30 percent. Second, the model identifies the years for each area in which the inventory damage ratio and payout ratio were both low, meaning that the Base Index correctly triggered a low payout that corresponded to low inventory damage ­ caused by the named peril (case example box 11CB.27). The results of this analysis can be shown in a classification matrix, as in table 11.12. Using the classification matrix, the basis risk metrics can be calculated for the Base Index as in case example box 11CB.28. Based on these metrics from the retrospective analysis, the insurer can con- clude whether the Base Index’s level of product design basis risk is acceptable. A good source of advice on acceptable levels for each metric is international reinsurers that have supported the writing of index products in different markets around the world. Case Example Box 11CB.27 Review of Base Index Payouts of at Least 30 Percent Looking back at the historical payout ratios for the case example in Step 4, out of the 300 data points, 25 have payout ratios of at least 30 percent. These are the total number of years with historical payouts greater than 30 percent. When these 25 data points are compared with the corresponding points for historical inventory dam- age ratios, we see that 22 of them also have inventory damage ratios greater than 50 percent. These are the years for which the Base Index correctly triggered a low payout that corresponded to low inventory damage caused by the named peril. Three high payouts—for Area G in 1998, Area J in 2006, and Area D in 2007—corresponded with historical events with damage levels less than 50 percent. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Base Index 169 Table 11.12 Retrospective Classification Matrix for the Base Index Historical payout ratios of Historical payout ratios of at least 30 percent less than 30 percent Total Historical events with damage ratios of at least 50 percent 22 1 23 Historical events with damage ratios of less than 50 percent 3 274 277 Total 25 275 300 Case Example Box 11CB.28  Calculation of Risk Metrics Probability of Base Index triggering correctly = [Number of high historical payout ratios that correspond to high historical damage ratios + number of low historical payout ratios that correspond to low historical damage ratios] /Total data points = [22 + 274]/300 = 99 percent Probability of Base Index triggering insufficient payout when inventory damage occurs (insured party basis risk) = Number of low historical payout ratios that correspond to high historical damage ratios /Total number of historical events with high damage ratios = 1/23 = 4 percent Probability of Base Index triggering payout unnecessarily (insurer basis risk) = Number of high historical payout ratios that correspond to low historical damage ratios /Total number of historical events with low damage ratios = 3/277 = 1 percent Inventory damage return period = Total number of data points/Total number of historical events with high damage ratios = 300/23 = 13 years Base Index return period = Total number of data points/Total number of high historical payout ratios = 300/25 = 12 years From the above calculations of the metrics, we can conclude that if the Base Index had been in place during the past 30 years, it would have triggered payouts correctly 99 percent of the time. In 4 percent of cases, it would have triggered insufficient payouts when the policyholder experienced inventory damage of more than 50 percent from the named peril (low insured party basis risk). The insurer would have made a payout of at least 30 percent when the inventory damage was less than 50 percent in only 1 percent of cases (suggesting a fairly low insurer basis risk). Furthermore, the inventory damage return period and the Base Index return period are very similar, which confirms an overall high level of accuracy. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 170 Evaluating the Base Index Because of limited data per area, the retrospective approach is best applied to a whole portfolio rather than to individual areas. Unfortunately, this portfolio- level approach does not provide the information necessary to improve the index structure for specific areas. A clear limitation of the retrospective approach is that it only considers how the Base Index would have performed during the past 30 years. Although this approach provides some limited insight into the risks associated with index insur- ance products, probabilistic modeling provides far more. For example, a retro- spective analysis cannot estimate the TVaR metric of basis risk that probabilistic models do. Bibliography Brehm, P. J. 2007. Enterprise Risk Analysis for Property & Liability Insurance Companies: A Practical Guide to Standard Models and Emerging Solutions. New York: Guy Carpenter. Cherubini, U., E. Luciano, and W. Vecchiato. 2004. Copula Methods in Finance. Hoboken, NJ: John Wiley & Sons. Crouhy, M., D. Galai, and R. Mark. 2006. The Essentials of Risk Management. New York: McGraw-Hill. Embrechts, P., F. Lindskog, and A. McNeil. 2003. “Modelling Dependence with Copulas and Applications to Risk Management.” In Handbook of Heavy Tailed Distributions in Finance, edited by S. T. Rachev, 329–84. Amsterdam: Elsevier. Grossi, P., H. Kunreuther, and C. C. Patel. 2005. Catastrophe Modeling: A New Approach to Managing Risk. New York: Springer Science Business Media. Lam, J. 2003. Enterprise Risk Management: From Incentives to Controls. Hoboken, NJ: Wiley. Law, A. M., and W. D. Kelton. 2006. Simulation Modeling and Analysis. 4th ed. New York: McGraw-Hill. Lehman, D. E., H. Groenendaal, and G. Nolder. 2012. Practical Spreadsheet Risk Modeling for Management. Boca Raton, FL: Chapman & Hall/CRC. Morsink, K., D. Clarke, and S. Mapfumo. 2016. “How to Measure Whether Index Insurance Provides Reliable Protection.” Policy Research Working Paper 7744, World Bank, Washington, DC. Ragsdale, C. T. 2001. Spreadsheet Modeling and Decision Analysis: A Practical Introduction to Management Science. Cincinnati, OH: Southwestern College. Tang, A., and E. A. Valdez. 2009. “Economic Capital and the Aggregation of Risks Using Copulas.” University of New South Wales, Sydney, Australia. Yan, J. 2006. “Multivariate Modeling with Copulas and Engineering Applications.” In Springer Handbook of Engineering Statistics, edited by H. Pham, 973–90. London: Springer-Verlag. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 12 Pricing the Base Index 12.1  Background and Objectives Chapter 5 explained the key managerial questions for Base Index product pricing during the pilot phase of launching an index insurance business line. It explained a series of steps for determining the price for a portfolio-priced Base Index under three situations: • The policy is not reinsured • The policy is reinsured through proportional reinsurance only • The policy is reinsured through a combination of nonproportional reinsurance and proportional reinsurance This chapter provides a step-by-step guide to using the probabilistic models that produce the decision metrics discussed in chapter 5. Using the Base Index’s historical payout ratios for the portfolio, the model simulates the scenario port­ folio payout amount (steps 8–12) and then estimates decision metrics for different Base Index portfolio-priced premium rates with no reinsurance (Steps 13–18), with proportional reinsurance only (Steps 19–24), and with proportional and nonproportional reinsurance (Steps 25–30). Based on these metrics the insurer can determine the portfolio-priced premium rate for the Base Index that best meets the profit objectives and risk tolerance of the insurer. In addition to providing metrics for portfolio pricing the Base Index, the model in this chapter also calculates the equitable premiums for each of the geographical areas (Steps 31–38). The insurer will repeat the pricing process with any later Redesigned Indexes or prototype products. Table 12.1 provides a summary of the model components along with a guide to the sections in this chapter and the worksheets in the accompanying Excel files. 12.2  Model Inputs The analyst starts by specifying the model inputs agreed upon with the insurance manager for pricing the Base Index (table 12.2). Risk Modeling for Appraising Named Peril Index Insurance Products   171   http://dx.doi.org/10.1596/978-1-4648-1048-0 172 Pricing the Base Index Table 12.1  Summary of Model Components for Pricing the Base Index Model component Section Excel sheet label Steps Description Model inputs 12.2 MI_12.2_MODEL Steps 1–7 User-defined assumptions, relevant INPUTS portfolio and insurer information, historical payout ratios, and reinsurance terms are entered. Model computations 12.3.1 MC_12.3.1__PAYOUT_ Steps 8–12 Simulation of scenario payout amounts SCENARIOS 12.3.2 MC_12.3.2_NO Steps 13–18 Calculation of product pricing decision REINSURANCE metrics for no reinsurance 12.3.3 MC_12.3.3_PR Steps 19–24 Calculation of product pricing decision REINSURANCE metrics for proportional reinsurance only 12.3.4 MC_12.3.4_PR & NP Steps 25–30 Calculation of product pricing decision REINSURANCE metrics for proportional and nonproportional reinsurance 12.3.5 MC_12.3.5_EQUITABLE Steps 31–38 Calculation of equitable premium metrics PREMIUMS for each geographical area Model outputs 12.4 MO_12.5_MODEL None Summary of pricing decision metrics for no OUTPUT reinsurance, proportional reinsurance only, and proportional and nonproportional reinsurance, plus equitable premium rates for each geographical area Table 12.2  Model Inputs Model component Section Excel sheet label Steps Description Model inputs 12.2 MI_12.2_MODEL INPUTS Steps 1–7 User defined assumptions, relevant portfolio and insurer information, historical payout ratios, and reinsurance terms are entered. 12.2.1  Exposed Units (Step 1) The portfolio pricing depends on the total sum insured per insured area, known as the exposed units per area. This input is calculated from the number of insured units and the average unit size (AUS; the average sum insured per unit) for each area (case example box 12CB.1). In cases in which the policyholder has high uncertainty about the average unit size per area, this uncertainty can be specified as a probability distribution. The most appropriate distribution for this operation is a project evaluation and review techniques (PERT) distribution for which the input parameters are the minimum, most likely, and maximum values (see chapter 10). Case Example Box 12CB.1  Inputs—Step 1 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 173 12.2.2  Internal Insurer Assumptions (Step 2) The analyst next specifies inputs based on internal insurer data (case example box 12CB.2). • Total sum insured ($): Total for all geographical areas. • Starting fund value ($): The accumulated net premiums from previous risk periods and any start-up funds for the index insurance business line. • Expense loading (as a percentage of premiums): Selling, general, and adminis- trative costs. These costs will differ from company to company. For a new pro­ duct line, the insurer can use rates from a comparable class in its portfolio or from data collected from other companies writing the same class of business. Reinsurers may also give some guidance based on international experience. • Target profit margin (%): The profit margin that the insurer is targeting for the business line. • Required return on capital (%): The return that the insurer’s shareholders require to keep their capital in this business line. Please note that this is an effective rate per risk period and not per year. • Risk-free rate (%): The interest an investor would expect from a risk-free investment. Typically, the cost of the interest rate on a three-month U.S. Treasury bill is used as a proxy. • Prediction interval (%). 12.2.3  Premium Rates (Step 3) The analyst next inputs the portfolio gross premium rates to be evaluated by the model (case example box 12CB.3). The expense costs (as specified in Step 2) will be subtracted before arriving at net premium rates. Case Example Box 12CB.2  Inputs—Step 2 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 174 Pricing the Base Index Case Example Box 12CB.3  Inputs—Step 3 12.2.4  Historical Payout Ratios (Step 4) The historical payout ratios for the Base Index (case example box 12CB.4) will be used in the simulation of payout ratios in steps 8–12. 12.2.5  Nonzero Historical Payout Ratios (Step 5) In this step (no case example box provided) the analyst manually records all the nonzero values for the historical payout ratios from Step 4. These inputs will be used in the simulation of payout ratios (Steps 8–12). Case Example Box 12CB.4  Inputs—Step 4 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 175 12.2.6  Proportional Reinsurance Terms (Step 6) The analyst specifies the percentage of the risk that will be ceded to the rein- surer under a potential proportional reinsurance arrangement (case example box 12CB.5). Case Example Box 12CB.5  Inputs—Step 6 12.2.7  Nonproportional Reinsurance Terms (Step 7) The analyst specifies several parameters for a potential nonproportional reinsur- ance arrangement (case example box 12CB.6). The treaty retention is the amount of claim exposure that the insurer will retain. None of the payout amounts less than the treaty retention amount speci- fied by the user will be covered by the reinsurer. The aggregate loss limit is the upper limit of exposure that the reinsurer will cover. For any claims between the treaty retention and the aggregate loss limit, the reinsurer will cover a percentage of the losses (the percentage carried by the reinsurer under nonproportional treaty). Case Example Box 12CB.6  Inputs—Step 7 Finally, the reinsurer will charge the insurer a reinsurance premium that is a percentage of the retained premium income. 12.3  Model Computations The model completes five sets of computations for pricing the Base Index, start- ing with simulating the key scenario parameters—payout amounts for the Base Index (Steps 8–12)—then producing product pricing decision metrics for the Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 176 Pricing the Base Index portfolio-priced premiums under the three reinsurance scenarios (Steps 13–30), and finally determining the equitable premiums for each geographical area (Steps 31–38) (see table 12.3). Remember that the insurer will repeat these pricing computations with any later Redesigned Indexes or prototype products. 12.3.1  Simulation of Scenario Portfolio Payout Amount (Steps 8–12) 12.3.1.1 Overview Based on the historical payout ratios (Step 4), the model simulates the scenario payout ratios for the Base Index using estimates for three stochastic elements: frequency, severity, and correlation. 12.3.1.2  Implementation in Excel (MC_12.3.1__PAYOUT_SCENARIOS) Steps 8–12 (table 12.4) for simulating the scenario portfolio payout amount are similar to Steps 14–18 in section 11.3.3 but with one key difference. In chapter 11, the model calculates only the payout ratios. In this chapter, how- ever, Steps 11 and 12 calculate monetary amounts for the payouts by area and for the total portfolio rather than just ratios. The reader is referred back to section 11.3.3 for further details on the modeling. Table 12.3  Model Computations Model component Section Excel sheet label Steps Description Model computations 12.3.1 MC_12.3.1__PAYOUT_ Steps 8–12 Simulation of scenario SCENARIOS payout amounts 12.3.2 MC_12.3.2_NO Steps 13–18 Calculation of product REINSURANCE pricing decision metrics for no reinsurance 12.3.3 MC_12.3.3_PR Steps 19–24 Calculation of product REINSURANCE pricing decision metrics for proportional reinsurance only 12.3.4 MC_12.3.4_PR & NP Steps 25–30 Calculation of product REINSURANCE pricing decision metrics for proportional and nonproportional reinsurance 12.3.5 MC_12.3.5_EQUITABLE Steps 31–38 Calculation of equitable PREMIUMS premium metrics for each geographical area Table 12.4  Model Computations Model component Section Excel sheet label Steps Description Model computations 12.3.1 MC_12.3.1__PAYOUT_ Steps 8–12 Simulation of scenario SCENARIOS payout amounts Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 177 Case example box 12CB.7 shows the simulation of the scenario portfolio payout amount for the case example. 12.3.2  Calculation of Product Pricing Decision Metrics—No Reinsurance (Steps 13–18) At this point the model has simulated the scenario portfolio payout amounts. Based on these scenario payouts, the model now calculates decision metrics that estimate the financial results for the insurer under a number of different pre- mium rates and in three reinsurance situations (table 12.5). First, the model addresses the scenario in which the Base Index is not reinsured. Case Example Box 12CB.7  Computations—Steps 8–12 Table 12.5  Model Computations Model component Section Excel sheet label Steps Description Model computations 12.3.2 MC_12.3.2_NO Steps 13–18 Calculation of product pricing REINSURANCE decision metrics for no reinsurance Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 178 Pricing the Base Index 12.3.2.1  Expected Losses and Required Capital 12.3.2.1.1 Overview. Figure 12.1 provides an overview of how the model simu- lates the expected losses and required capital based on the scenario portfolio payout amount (Step 12). The portfolio payout amount is the same as losses for the insurer because with no reinsurance the insurer will need to pay all claims. 12.3.2.1.2 Implementation in Excel (MC_12.3.2_NO REINSURANCE). Case example box 12CB.8 shows the steps in which the model estimates the expected portfolio losses and required capital. In Step 13, the model inserts the value for the scenario portfolio payout amount from Step 12 as the scenario losses. Remember, because no reinsurer is involved, the portfolio payout amount is the same as the losses for the insurer. In Step 14, the model generates at least 10,000 scenario loss amounts (Monte Carlo scenarios) for the portfolio and determines the expected losses for the next risk period. Based on the prediction interval selected in Step 2, the model also calculates the appropriate percentile and tail value at risk (TVaR) values. These values indicate the expected magnitude of the insurer’s losses. Figure 12.1  Generating Expected Losses and Required Capital Scenario portfolio losses (Step 12) Run at least 10,000 Monte Carlo scenarios Prediction interval (Step 2) Projected portfolio losses Required capital • Lower (Step 14) • Expected • Upper • Tail value at risk (Step 14) Note: TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 179 Case Example Box 12CB.8  Computations—Steps 13 and 14 In the case example, the expected losses for the next risk period are $619,287. The prediction interval for the case example is 90 percent, so the model also shows the 5th and 95th percentiles and the tail value at risk (TVaR) 95 percent. The TVaR 95 percent tells us that for a 1-in-20 year event, the losses are expected to be as high as $2,488,867. In the case example, the required capital is $1,869,580. Required capital = TVaR losses – Expected losses = $2,488,867 – $619,287 = $1,869,580 The insurer should keep $1,869,580 in reserve (as required capital) to stay solvent in case of a 1-in-20 year event (TVaR 95 percent). Also in Step 14, the model uses the same 10,000 scenarios to calculate the required capital for the portfolio. The required capital is the amount of capital that the insurer will need to keep in reserve to be sure that it can make the pay- outs for extreme events, defined as the total claim that is expected once every 20 years (that is, TVaR). Required capital = TVaR losses − Expected losses Remember, the model—and its calculation of required capital (see section 10.3.2)— assumes that the index insurance product that is evaluated is the only product that the insurer offers (see section 9.3). In reality, the insurer would most likely have several lines of business, and the capital allocated to each busi- ness line will be a function of the overall capital required for the whole firm. The guide follows this simplistic approach because each company will have unique business and asset compositions, and each market will have different regula- tory requirements. Following a monoline insurance approach allows us to dem- onstrate the principles underlying probabilistic modeling without introducing too much complexity. 12.3.2.2  Combined Ratios and Profit Margins 12.3.2.2.1 Overview. In this section the model compares combined ratios and profit margins across different premium rates; later sections compare them Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 180 Pricing the Base Index across alternative reinsurance situations. Profit margins and combined ratios are useful in both product development and financial reporting. Profit margin measures the percentage of the insurance premium that the insurer actually keeps in earnings. The calculation of the combined ratio and profit margin is shown in figure 12.2. Figure 12.2  Generating Expected Combined Ratios and Profit Margins Total sum Premium rates insured (Step 3) (Step 2) Scenario gross premium income (Step 15) Scenario losses (Step 13) Expense loading (Step 2) Scenario loss ratio (Step 15) Scenario combined ratio (Step 15) Total profit Prediction margin Run at least 10,000 interval Monte Carlo scenarios (Step 2) (Step 2) Projected profit margins Projected • Lower combined ratio • Expected • Upper • Lower • TVaR • Expected • Probability of negative profit • Upper • Probability of profit (Step 15) below target (Step 15) Note: TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 181 12.3.2.2.2  Implementation in Excel (MC_12.3.2_NO REINSURANCE).  In Step 15 (case example box 12CB.9), the model first calculates the scenario gross pre- mium income for each premium rate (Step 3) and the total portfolio sum insured (Step 2). Scenario gross premium income = Premium rate × Total portfolio sum insured (Step 3) (Step 2) Case Example Box 12CB.9  Computations—Step 15 For the case example, the gross premium income for the 10 percent premium rate is $800,000. Scenario gross premium income = Premium rate × Total portfolio sum insured = 10 percent × 8,000,000 = $800,000 The scenario loss ratio for the 10 percent premium rate is 62.5 percent (not shown in Step 15 table). Scenario loss ratio = Scenario losses/Scenario gross premium income = 500,000/800,000 = 62.5 percent The scenario combined ratio for the 10 percent premium rate is 77.5 percent. Scenario combined ratio = Scenario loss ratio + Expense loading = 62.5 percent + 15 percent = 77.5 percent For the case example, the scenario profit margin for the 10 percent premium rate is 22.5 percent. Scenario profit margin = 100 percent − Scenario combined ratio = 100 percent − 77.5 percent = 22.5 percent For the 10 percent premium rate in the case example, the probability of a negative profit is 34 percent. Probability of a negative profit = ~ 3,400 of the 10,000 Monte Carlo scenarios = 34 percent For the 10 percent premium rate in the case example, the probability of a profit of less than the target profit margin is 38 percent. ­ Probability of a profit below the target profit margin = ~ 3,800 of the 10,000 Monte Carlo scenarios = 38 percent Note: TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 182 Pricing the Base Index Also in Step 15, the model calculates the scenario loss ratio for each premium rate using the scenario losses (Step 13) and the gross premium income. Scenario loss ratio = Scenario losses/Scenario gross premium income (Step 13) Next, Step 15 calculates the scenario combined ratio for each premium rate using the scenario loss ratio and expense loading. Scenario combined ratio = Scenario loss ratio + Expense loading (Step 2) Next, Step 15 calculates the scenario profit margin for each premium rate from the scenario combined ratio. Scenario profit margin = 100 percent − Scenario combined ratio (Step 15) At this point, the model generates at least 10,000 Monte Carlo combined ratio scenarios for each of the 10 possible premium rates and determines the expected combined ratio for the next risk period (92 percent for the 10 percent premium rate in the case example). Based on the prediction interval selected in Step 2, the model also calculates the appropriate percentile and TVaR values for the combined ratio. Using the same 10,000 scenarios, the model next calculates the profit margins for each of the 10 premium rates and determines the expected profit margin for the next risk period (8 percent for the 10 percent premium rate in the case example), along with the appropriate percentiles. Finally, the model calculates the probability of a negative profit and the prob- ability of a profit of less than the target profit margin (Step 2) for each of the premium rates. percent of Monte Carlo scenarios in which the Probability of a negative profit =  scenario profits are lower than $0 Probability of a profit below   = percent of Monte Carlo scenarios in which the the target profit margin scenario profits are lower than the target profit margin (Step 2) 12.3.2.3  Probability of Fund Ruin 12.3.2.3.1 Overview. The probability of fund ruin (figure 12.3) is another important metric for the insurer when considering pricing for the Base Index. This probability indicates the likelihood that the capital fund available to cover the product risk will be exhausted over a specified time frame. The model con- siders a one-year or one-growing-season period. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 183 Figure 12.3  Generating Probability of Fund Ruin Premium rates Total portfolio Expense sum insured loading (Step 3) (Step 2) (Step 2) Starting fund value (Step 2) Scenario net premium income (Step 16) Scenario total funds at risk Scenario losses (Step 16) (Step 13) Scenario net fund position (Step 16) Run at least 10,000 Monte Carlo scenarios Probability of fund ruin (Step 16) Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 184 Pricing the Base Index Implementation in Excel (MC_12.3.2_NO REINSURANCE) In Step 16 (case example box 12CB.10), the model calculates the net premium rate from the gross premium rate and the expense loading. Scenario net premium rate = Scenario gross premium rate × (1 – Expense loading) (Step 3) (Step 2) Case Example Box 12CB.10  Computations—Step 16 For the case example, the scenario net premium rate for the 10 percent gross premium rate is 8.5 percent: Scenario net premium rate = Scenario gross premium rate × (1–Expense loading) = 10 percent × (1 – 0.15) = 8.5 percent The scenario net premium income for the 10 percent gross premium rate is $680,000 (not shown in Step 16 table). Scenario net premium income = Scenario net premium rate × Total sum insured = 8.5 percent × 8,000,000 = $680,000 In the case example, the total funds at risk for the 10 percent premium rate are $730,000. Scenario total funds at risk = Scenario net premium income + Starting fund value = 680,000 + 50,000 = $730,000 For the case example, the scenario net fund position for the 10 percent premium rate is $230,000. Scenario net fund position = Total funds at risk – Scenario losses = 730,000 – 500,000 = $230,000 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 185 The net premium rates will be used in Step 35 when estimating the capital return. Also in Step 16, the model calculates the scenario net premium income for each premium rate from the net premium rates and the total portfolio sum insured. Scenario net premium income = Scenario net premium rate × Total sum insured (Step 3) (Step 3) (Step 2) Also in Step 16, the model calculates the total funds at risk for each pre- mium rate from the net premium income and the starting fund value (Step 2). This value is the maximum amount of exposure that the insurer is willing to take during a risk period. Scenario total funds at risk = Scenario net premium income + Starting fund value (Step 2) Step 16 next calculates the scenario net fund position for each premium rate from the scenario losses (Step 13) and the total funds at risk. Scenario net fund position = Scenario total funds at risk − Scenario losses (Step 13) At this point, the model generates at least 10,000 scenario net fund positions for each premium rate. The proportion of the 10,000 scenarios with fund posi- tions less than zero (32 percent for the 10 percent premium rate in the case example) is the probability of fund ruin. 12.3.2.4  Economic Value Added 12.3.2.4.1 Overview. Another metric for the insurer to evaluate when deciding premium rates is the economic value added (EVA). EVA is the profit earned by the firm minus the cost of financing the firm’s capital (figure 12.4). The insurer behind the risk-taking activity establishes the capital requirement. For the EVA to be positive, the profit from the index insurance product will need to more than cover the costs of the capital required for issuing the index insurance prod- uct. Specifically, EVA is the difference between the value derived from selling the product (the premium income) and the cost of doing so, including expenses, potential payouts, and financing costs of the required capital. In this guide, the EVA is expressed as a percentage of required capital. Implementation in Excel (MC_12.3.2_NO REINSURANCE) In Step 17 (case example box 12CB.11), the model calculates the net premium income as in Step 16. Next, the model calculates the scenario capital charge. Scenario capital charge = Required capital × Required return on capital (Step 14) (Step 2) Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 186 Pricing the Base Index Figure 12.4  Generating Potential Economic Value Added Required capital Required return on Scenario net premium income (1st simulation) capital (Step 16) (Step 14) (Step 12) Scenario capital charge (Step 17) Scenario projected losses (Step 13) Scenario economic valued added (Step 17) Prediction interval (Step 3) Run at least 10,000 Monte Carlo scenarios Projected economic valued added • Lower • Expected • Upper (Step 17) Case Example Box 12CB.11  Computations—Step 17 box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 187 Case Example Box 12CB.11  Computations—Step 17 (continued) The capital charge for the 10 percent premium rate is $93,479 in the case example. Scenario capital charge = Required capital × Required return on capital = $1,869,580 × 5 percent = $93,479 In the case example, the EVA for the 10 percent premium rate is 4.6 percent. Scenario EVA = (Scenario net premium income − Scenario losses − Scenario capital charge)/Required capital = ($680,000 − $500,000 − $93,479)/$1,869,580 = 4.6 percent Note: EVA = economic value added. If the insurer were free to invest the required capital it would have expected to generate gains equal to the capital charge. But because the insurer must hold the required capital, it incurs the capital charge as an opportunity cost. Next, Step 17 calculates the scenario EVA for each premium. Scenario EVA = (Scenario net premium – Scenario losses – Scenario capital charge)/ income Required capital (Step 16) (Step 13) (Step 14) The EVA indicates the profit earned by the product less the cost of financing the required capital, as a percentage of the capital required. Based on these calculations, the model generates at least 10,000 scenario EVA results for each premium rate and determines the expected EVA for the next risk period (–2 percent for the 10 percent premium rate in the case example), along with the appropriate percentile values for the EVA. 12.3.2.5  The Sharpe Ratio 12.3.2.5.1 Overview. The Sharpe ratio (figure 12.5) provides the insurer with an estimate of how much risk is being taken to get a certain return. Imagine that the insurer has two investment alternatives, both with a 15 percent expected return. The Sharpe ratios for these investments will clearly show if one has a higher risk than the other. A negative Sharpe ratio indicates an investment with an expected negative return per unit of risk assumed, while a positive Sharpe ratio indicates an investment with an expected positive return per unit of risk assumed. Implementation in Excel (MC_12.3.2_NO REINSURANCE) In Step 18 (case example box 12CB.12), the model first calculates scenario return on capital for each premium rate from the net premium income (Step 16), the scenario losses (Step 13), and the required capital (Step 14). Scenario return on capital = (Scenario net premium income − Scenario losses)/ (Step 16) (Step 13) Required capital (Step 14) Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 188 Pricing the Base Index Figure 12.5  Generating Sharpe Ratios Scenario net Required capital Scenario losses premium income (1st simulation) (Step 16) (Step 13) (Step 14) Scenario return on captial (Step 18) Run the model for at least 10,000 Monte Carlo scenarios Risk-free rate (Step 12) Projected return on capital • Expected • Standard deviation (Step 18) Sharpe ratios (Step 18) Based on these calculations, the model generates at least 10,000 Monte Carlo scenario returns on capital for each premium rate and determines the expected return on capital for the next risk period (3 percent for the 10 percent premium rate in the case example) as well as the standard deviation of the return on capi- tal (36 percent). Note that for all premium levels considered, the standard deviation of the return on capital is the same. The standard deviations are all the same because Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 189 Case Example Box 12CB.12  Computations—Step 18 The scenario return on capital for the 10 percent premium rate is 9.6 percent (rounded up to 10 percent) for the case example. Scenario return on capital = (Scenario net premium income – Scenario losses)/Required capital = ($680,000 − $500,000)/$1,869,580 = 9.6 percent For the case example, the Sharpe ratio is 0.027 (rounded up to 3 percent) for the 10 percent premium rate. Although positive, this ratio is less than those for the 11 percent and 12 percent premium rates because of the lower expected return. Sharpe ratio = (Expected return on capital – Risk-free rate)/Standard deviation of the expected return on capital = (3 percent − 2 percent)/36 percent = 2.7 percent the actual premium levels are assumed not to change the risk of the insured units. Although the expected returns will increase with higher premiums, the standard deviation (that is, the spread around the expected returns) does not change. Next, the model calculates the Sharpe ratio from the expected return on capital, the risk-free rate (Step 2), and the standard deviation for the return on capital. Standard deviation of Sharpe ratio = (Expected return on capital − Risk-free rate)/ the return on capital (Step 2) A positive Sharpe ratio indicates an investment with an expected positive return per unit of risk assumed. Premium rates with higher Sharpe ratios are Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 190 Pricing the Base Index preferred because the higher the Sharpe ratio, the greater the expected return on the capital invested relative to the amount of risk taken. These metrics are also calculated for situations with proportional and nonpro- portional insurance in the following sections. 12.3.3  Calculation of Product Pricing Decision Metrics—Proportional Reinsurance Only (Steps 19–24) 12.3.3.1 Overview The preceding sections discussed the generation of pricing decision metrics for the Base Index for the scenario with no reinsurance. Now we will review the same process for the scenario with proportional reinsurance only (table 12.6 and figure 12.6). The objective of this process is to evaluate the effect of proportional reinsur- ance arrangements. Reinsurance can reduce the premium rates of an index insur- ance product because the reinsurance firm may need to set aside less required capital than the insurer would have because the reinsurer’s portfolio is highly diversified. In addition, the reinsurer may have a lower cost of capital or other operational costs that are lower. There are two main differences between the current proportional reinsurance only and the previously evaluated no reinsurance situations: First, the scenario losses (Step 13) will be reduced by the percentage ceded to the reinsurer (Step 6; “net PR” means “net proportional reinsurance”). Scenario retained claims (net PR) = Scenario losses × (1 − percent ceded to reinsurer) (Step 13) (Step 6) Second, the insurer passes along a portion of the premium income with the risk ceded to the reinsurer. Scenario gross premium income (net PR) = Scenario gross premium income × (Step 15) (1 − percent ceded to reinsurer) (Step 6) This guide assumes that the insurer makes no profit on the ceded risk. Therefore, the percentage of the insured amount that is ceded to the reinsurer represents a reduction in the sum insured. This conservative approach recognizes that reinsur- ance market prices are very volatile with more favorable terms when the market is soft (that is, high liquidity) than when it is hard (that is, low liquidity).1 Table 12.6  Model Computations Model component Section Excel sheet label Steps Description Model computations 12.3.3 MC_12.3.3_PR Steps 19–24 Calculation of product pricing REINSURANCE decision metrics for proportional reinsurance only Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 191 Figure 12.6  Generating Product Pricing Decision Metrics (Proportional Reinsurance Only) Scenario Percentage ceded to Scenario gross losses reinsurer premium income (Step 13) (Step 6) (Step 15) Expense Scenario retained claims (net PR) Scenario gross premium loading (Step 19) income (net PR) (Step 21) (Step 2) Scenario combined ratio (net PR) (Step 21) Run at least 10,000 Run at least 10,000 Monte Carlo scenarios Monte Carlo scenarios Prediction interval (Step 2) Projected combined ratio and profit margin Required capital Projected retained • Lower (net PR) claims (net PR) • Expected • Lower • Upper (Step 20) • Expected • TVaR • Upper • Probability of negative profit • TVaR • Probability of negative profit below margin target (Step 20) (Step 21) Note: PR = proportional reinsurance; TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 192 Pricing the Base Index 12.3.3.2  Implementation in Excel (MC_12.3.3.1_PR REINSURANCE) The modeling process for the proportional reinsurance only scenario is very simi- lar to that for the no reinsurance situation (Steps 13–18) but uses scenario retained claims net of proportional reinsurance in place of scenario losses. In Step 19 the model calculates the scenario retained claims (net PR) (case example box 12CB.13) from the scenario losses and the percentage ceded to the insurer. These are the claim payouts for which the insurer is responsible. In Step 20 (case example box 12CB.14), the model generates at least 10,000 scenario retained claims (net PR) for the portfolio and determines the expected Case Example Box 12CB.13  Computations—Step 19 For the case example, with 80 percent of the risk ceded to the reinsurer (step 6), the scenario losses for the portfolio are reduced from $500,000 to $100,000. Scenario retained claims (net PR) = Scenario losses × (1 − percent ceded to reinsurer) = 500,000 × (1 − 80 percent) = $100,000 Note: PR = proportional reinsurance. Case Example Box 12CB.14  Computations—Step 20 For the case example, with proportional reinsurance (PR) the expected retained claims for the next risk period are reduced to $123,857 (from $619,287) and the required capital to $373,915 (from $1,869,580). Note that the required capital is therefore reduced by 80 percent compared with the previously evalu- ated no reinsurance situation, which is exactly the percentage ceded to the proportional reinsurer. Required capital (net PR) = TVaR retained claims (net PR) − Expected retained claims (net PR) = $497,772 − $123,857 = $373,915 Note: TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 193 retained claims (net PR) for the next risk period. Based on the prediction interval selected in Step 2, the model also calculates the appropriate percentile and TVaR values of the retained claims. Also in Step 20, the model uses the same 10,000 scenarios to calculate the required capital (net PR). Required capital (net PR) = TVaR retained claims (net PR) − Expected retained claims (net PR) In Step 21, the model simulates the combined ratios and profit margins for each premium rate (case example box 12CB.15). The process is similar to Step 15, but uses gross premium income (net PR) rather than gross premium income. The remaining steps for determining the pricing decision metrics for the pro- portional reinsurance situation (Steps 22–24) are similar to the no reinsurance situation (Steps 16–18) but use the scenario losses (net PR) and gross premium income (net PR) formulas where appropriate. Case Example Box 12CB.15  Computations—Step 21 For the case example, the gross premium income (net PR) for the 10 percent premium rate is $160,000. Gross premium income (net PR) = Gross premium income × (1 − Percentage ceded to reinsurer) = $800,000 × (1 − 80 percent) = $160,000 Note: PR = proportional reinsurance; TVaR = tail value at risk. 12.3.4  Calculation of Product Pricing Decision Metrics—Proportional and Nonproportional Reinsurance (Steps 25–30) 12.3.4.1 Overview In this section the model calculates the product pricing metrics for the scenario in which the Base Index has both proportional and nonproportional reinsurance (table 12.7 and figure 12.7). Table 12.7  Model Computations Model component Section Excel sheet label Steps Description Model computations 12.3.4 MC_12.3.4_PR & NP Steps 25–30 Calculation of product pricing REINSURANCE decision metrics for proportional and nonproportional reinsurance Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 194 Pricing the Base Index Figure 12.7  Generating Product Pricing Decision Metrics (Proportional and Nonproportional Reinsurance) Scenario retained Percentage ceded Percentage carried by Scenario gross claims (net PR) to reinsurer reinsurer under premium income nonproportional (net PR and treaty NPR) (Step 25) (Step 7) (Step 6) (Step 27) Expense loading (Step 2) Scenario retained claims Scenario combined ratio (net PR and NPR) (net PR and NPR) (Step 26A) (Step 27) Protection interval Run at least 10,000 Monte Run at least 10,000 Monte Carlo scenarios (Step 2) Carlo scenarios Required capital (net PR and NPR) Projected combined (Step 26) ratio and profit margin (net PR and NPR) Projected retained • Lower claims (net PR and NPR) • Expected • Lower • Upper • Expected • TVaR • Upper • Probability of negative profit • TVaR • Probability of profit below target (Step 26B) (Step 27) Note: NPR = nonproportional reinsurance; PR = proportional reinsurance; TVaR = tail value at risk. There are two main differences between the proportional and nonpropor- tional reinsurance situation and the no reinsurance situation: First, the scenario losses and gross premium income are again reduced by the (proportional reinsurance) percentage that is ceded to the reinsurance company (Step 6). In addition, the reinsurance company will pay a set percent- age of the nonceded claims between the treaty retention and the aggregate loss limit (Step 7). When the losses are less than the treaty retention, the reinsurance policy will not pay the insurer anything on the nonceded claims. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 195 For nonceded claims that are between the treaty retention and the aggregate loss limit, the insurer’s payout will be as follows (NPR = nonproportional reinsurance): Scenario reinsurance (Scenario losses Percent carried by the payout (net PR = net PR − Treaty × reinsurer under and NPR) retention) nonproportional treaty. (Step 19) (Step 7) (Step 7) For any nonceded losses that are higher than the aggregate loss limit, the rein- surance will pay out nothing. The insurer’s net retained claims will be Scenario retained claims (net PR and NPR) = Scenario retained claims net PR − (Step 19) Reinsurance payout net PR and NPR. Second, the nonproportional reinsurance will have direct costs to the insurer, specifically, the reinsurance premium rate, which is typically specified as a per- centage of the total premium amount. This cost will influence the insurer’s profit and loss metrics directly. Gross premium income = [Gross premium × (1 − Percent ceded to × (1 − Expense (net PR and NPR) income ­ reinsurer)] loading) (Step 15) (Step 6) (Step 2) − nonproportional reinsurance premium rate] 12.3.4.2  Implementation in Excel (MC_12.3.4_PR & NP REINSURANCE) The modeling process for the proportional and nonproportional reinsurance situ- ation is again very similar to that for the no reinsurance situation (Steps 13–18) but uses scenario retained claims net of proportional and nonproportional rein- surance in place of scenario losses. In Step 25 (case example box 12CB.16), the model calculates the scenario retained claims (net PR) from the scenario payout amount (Step 12) and the percentage ceded to the insurer (Step 6). This process (and the resulting value) is identical to Step 19 for the proportional reinsurance only scenario. In Step 26A, the model calculates the scenario retained claims (net PR and NPR). Scenario reinsurance (Scenario retained Percent carried by the payout (net PR = claims net PR − × reinsurer under and NPR) Treaty retention) nonproportional treaty (Step 19) (Step 7) (Step 12) Scenario retained claims = Scenario retained − Scenario reinsurance (net PR and NPR) claims net PR payout (Step 19) Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 196 Pricing the Base Index Case Example Box 12CB.16  Computations—Steps 25 and 26 In the case example, the scenario reinsurance payout for the portfolio is $13,500 and the scenario retained claims (net PR and NPR) are $86,500. Scenario reinsurance payout (net PR and NPR) = (Scenario retained claims net PR − Treaty retention) × Percent carried by the reinsurer under nonproportional treaty = (100,000 − 85,000) × 90 percent = $13,500 Scenario retained claims (net PR and NPR) = Scenario retained claims net PR − Scenario reinsurance payout = 100,000 − 13,500 = $86,500 For the case example, with both proportional and nonproportional reinsurance the expected retained claims for the next risk period are reduced to $62,861 (from $123,857 with proportional reinsur- ance only in Step 20) and the required capital to $65,263 (from $373,915 in Step 20). The reason for this large reduction in the required capital is that whenever claims are higher than the treaty retention, the reinsurer starts to cover the costs of a percentage of the payouts. This can greatly reduce the risk and the TVaR, and therefore the required capital for the insurer. Required capital (net PR and NPR) = TVaR retained claims net PR and NPR − Expected retained claims net PR and NPR = 128,124 − 62,861 = $65,263 Note: NPR = nonproportional reinsurance; PR = proportional reinsurance; TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 197 In Step 26B, the model generates at least 10,000 scenario retained claims (net PR and NPR) for the portfolio and determines the expected retained claims (net PR and NPR) for the next risk period. Based on the prediction interval selected in Step 2, the model also calculates the appropriate percentile and TVaR values of the retained claims. Also in Step 26B, the model uses the same 10,000 scenarios to calculate the required capital (net PR and NPR). Required capital = TVaR retained claims − Expected retained (net PR and NPR) (net PR and NPR) claims (net PR and NPR) In Step 27, the model simulates the combined ratio and profit margin for each premium rate (case example box 12CB.17). The process is the same as for Step 15, but uses gross premium income (net PR and NPR) in place of gross premium income. The remaining steps for determining the pricing decision metrics for the proportional and nonproportional reinsurance situation (Steps 28–30) are the same as for the no reinsurance situation (Steps 16–18), but use the scenario retained claims and gross premium income that are net of PR and NPR instead. See box 12.1 for a recap of the formulas used in the preceding text. Case Example Box 12CB.17  Computations—Step 27 For the case example, the gross premium income (net PR and NPR) for the 10 percent premium rate is $128,000. Scenario gross premium income (net PR and NPR) = [Scenario gross premium income × (1 − Percent ceded to reinsurer)] × (1 − Expense loading − Nonproportional reinsurance premium rate) = [$800,000 × (1 − 80 percent)] × (1 − 15 percent − 5 percent) = $128,000 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 198 Pricing the Base Index Box 12.1  Summary of Key Formulas Used in the Chapter Scenario metrics (one Monte Carlo scenario) • Retained claims (net PR) = Losses × (1 − Percent ceded to reinsurer) • Reinsurance payout (net PR and NPR) = [Retained claims (net PR) − Treaty retention] × Percent carried by the reinsurer under proportional treaty • Retained claims (net PR and NPR) = Retained claims (net PR) − Reinsurance payout • Gross premium income = Premium rate × Total portfolio sum insured • Gross premium income (net PR) = Gross premium income × (1 − Percent ceded to reinsurer) • Gross premium income (net PR and NPR) = [Gross premium income × (1 − Percent ceded to reinsurer)] × (1 − Expense loading − Nonproportional reinsurance premium rate) • Loss ratio = Losses/Gross premium income • Combined ratio = Loss ratio + Expense loading • Profit margin = 1 − Combined ratio • Net premium rate = Gross premium rate × (1 − Expense loading) • Net premium income = Net premium rate × Total sum insured • Total funds at risk = Net premium income + Starting fund value • Net fund position = Total funds at risk − Losses Metrics based on at least 10,000 Monte Carlo scenarios • Required capital = TVaR losses − Expected losses • Required capital (net PR) = TVaR retained claims (net PR) − Expected retained claims (net PR) • Required capital (net PR and NPR) = TVaR retained claims (net PR and NPR) − Expected retained claims (net PR and NPR) • Probability of a negative profit = Percentage of scenarios in which the scenario profits are less than $0 • Probability of a profit below the target profit margin = Percentage of scenarios in which the scenario profits are less than the target profit margin • Probability of fund ruin = Percentage of scenarios in which the scenario net fund positions are less than $0 • EVA = (Net premium income − Losses − Capital charge)/Required capital • Return on capital = (Net premium income − Losses)/Required capital Note: EVA = economic value added; NPR = nonproportional reinsurance; PR = proportional reinsurance; TVaR = tail value at risk. 12.3.5  Calculation of Equitable Premiums Metrics (Steps 31–38) The preceding sections discussed the generation of pricing decision metrics for the Base Index to help the insurer select the best portfolio-priced premium rate for the entire portfolio of geographical areas. However, each geographical area covered by a portfolio-priced index product in reality has a different risk profile Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 199 (for example, less rain or more extreme maximum temperatures), which corre- sponds with a different premium rate. The premium rates that are specific to each area in the portfolio are called equitable premium rates. An equitable premium for each area takes into account the relative riskiness of each area. Areas that are riskier require the insurance company to have more capital available on hand and should therefore have higher premiums than the less risky areas. To calculate the equitable premium rate for each area (table 12.8), the model first simulates the payout amounts by area as in Steps 8–11. Recall that in Step 12, the model sums the payout amounts for all areas to produce the total port- folio payout amount. However, because we are now looking at individual areas, this additional step is not needed. Instead, we will use the payout amounts for each area. In this guide, equitable premiums are calculated as follows: {[ APPIU + EROC × ( RCPIU )]/AUS } Equitable premium rate = , (1− Total expense loading ) where APPIU = average payout per insured unit, EROC = expected return on capital, RCPIU = required capital per insured unit, and AUS = average unit size (Step 1). To make the logic easier to follow, we will first explain how the APPIU, RCPIU, and EROC are calculated before explaining the calculation of the equi- table premium rate for each area. 12.3.5.1  Components APPIU and RCPIU 12.3.5.1.1 Overview. Figure 12.8 provides an overview of how the model gener- ates the APPIU and RCPIU from the scenario payout amounts by area. 12.3.5.1.2  Implementation in Excel (MC_12.3.5_EQUITABLE PREMIUMS).  In Step 31, the model first generates at least 10,000 scenario payout amounts for each area and determines the average (expected) payout amount for each Table 12.8  Model Computations Model component Section Excel sheet label Steps Description Model computations 12.3.5 MC_12.3.5_EQUITABLE Steps 31–38 Calculation of equitable PREMIUMS premium metrics for each geographical area Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 200 Pricing the Base Index Figure 12.8  Generating APPIU and RCPIU for Each Area Scenario payout amount for area (Step 11) Prediction interval (Step 2) Run at least Monte Carlo 10,000 scenarios TVaR payout amount for area Average total payout for area (Step 31) (Step 31) Total stand-alone required capital Stand-alone required capital for area (Step 32) (Step 32) Area beta (Step 33) Number of Required exposed units capital for area for portfolio (Step 1) (Step 14) Required capital for area (Step 34) RCPIU APPIU for area for area (Step 34) (Step 31) Note: APPIU = average payout per insured unit; RCPIU = required capital per insured unit; TVaR = tail value at risk. (case example box 12CB.18). Based on the prediction interval selected in Step 2, the model also calculates the appropriate TVaR of the payout amount for each of the areas. The expected payout per area and the TVaR of the payout per area are both used in Step 32 to calculate an area’s stand-alone required capital. Also in Step 31, the model calculates the APPIU for each area. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 201 Case Example Box 12CB.18  Computations—Step 31 For Area B in the case example, the APPIU is $5.46. APPIU for area = Average payout amount for area/Number of exposed units for area = 20,598/3,773 = $5.46 Note: APPIU = average payout per insured unit.  verage payout amount for area/ APPIU for area = A Number of insured units for area (Step 1) In Step 32 (case example box 12CB.19), the model calculates the stand-alone required capital for each area. Case Example Box 12CB.19  Computations—Steps 32–34 For the case example, the stand-alone required capital for Area B is $198,800. Stand-alone required capital for area = TVaR payout amount for area − Average payout amount for area = 219,398 − 20,598 = $198,800 The portfolio stand-alone required capital for the portfolio is $5,597,927 81,290 + 198,800 + 779,630 + 308,511 + 1,621,737 + 202,522 + 667,255 + 844,747 + Portfolio stand-alone capital =  307,278 + 586,156 = $5,597,927 box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 202 Pricing the Base Index Case Example Box 12CB.19  Computations—Steps 32–34 (continued) For the case example, the beta for Area B is 3.5513 percent (rounded to 3.6 percent in the Step 33 table). Area beta = Stand-alone required capital for area/Total stand-alone required capital for portfolio = 198,800/5,597,927 = 3.5513 percent For the case example the required capital for Area B is $66,395. Required capital for area = Required capital for portfolio × Area beta = 1,869,580 × 3.5513 percent = $66,395 The RCPIU for Area B is $17.60. RCPIU for area = Required capital for area/Number of exposed units for area = $66,395/3,773 = $17.60 Note: RCPIU = required capital per insured unit. Stand-alone required = TVaR payout amount − Expected payout amount capital for area for area for area (Step 31) (Step 31) Also in Step 32, the model calculates the total stand-alone required capital for the portfolio by summing the stand-alone required capital for all of the geo- graphical areas. The total stand-alone required capital for the portfolio will always be higher than the required capital under portfolio pricing (Step 14). With portfolio pric- ing, the required capital is lower because of diversification. Step 33 calculates the beta for each area. Total stand-alone required capital Area beta = Stand-alone required capital for area/ for portfolio (Step 32) (Step 32) This beta value represents the equitable proportion of the portfolio’s total stand-alone required capital that is allocated to each area in light of its relative riskiness. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 203 Step 34 calculates the required capital for each area. Required capital for area = Required capital for portfolio × Area beta (Step 14) (Step 33) Also in Step 34, the model calculates the RCPIU for each area. RCPIU for area = Required capital for area/Number of exposed units for area (Step 1) 12.3.5.2  Component EROC 12.3.5.2.1 Overview. Another important input to calculation of equitable pre- miums per area is the EROC, the expected return on capital. Figure 12.9 Figure 12.9  Generating EROC for Each Area Net premium rate Total sum insured Expected for portfolio for portfolio portfolio losses (Step 16) (Step 2) (Step 14) Net premium income for portfolio (Step 16) Required capital for portfolio (Step 14) Expected return on capital (EROC) for portfolio (Step 35) Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 204 Pricing the Base Index provides an overview of how the model generates the EROC from the scenario payout amounts by area. Case Example Box 12CB.20  Computations—Step 35 For the case example, the expected return on capital (EROC) for the portfolio at the 10 percent premium rate is 3.2474 percent (rounded up to 3 percent). EROC for portfolio = (Net premium income for portfolio − Expected portfolio losses)/Required capital for portfolio = (680,000 − 619,287)/1,869,580 = 3.2474 percent Using the 10 percent premium rate increases shareholder value for the insurer because the EROC is positive. 12.3.5.2.2  Implementation in Excel (MC_12.3.5_EQUITABLE PREMIUMS).  In Step 35, the model calculates the EROC (case example box 12CB.20). EROC for portfolio = (Net premium income – Expected portfolio/Required capital for portfolio losses) for portfolio (Step 16) (Step 14) (Step 14) 12.3.5.3 Equitable Premium Rates 12.3.5.3.1 Overview. Figure 12.10 provides an overview of how the model gener- ates the equitable premium rates for each area from the APPIU, RCPIU, EROC, and AUS. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 205 Figure 12.10  Generating Equitable Premium Rates for Each Area EROC for RCPIU for area APPIU for area AUS for area portfolio (Step 35) (Step 34) (Step 31) (Step 1) Equitable pure risk premium rate for area (Step 37) Expense loading (Step 2) Final equitable pure risk premium rate for area (Step 38) Note: APPIU = average payout per unit insured; AUS = average unit size; EROC = expected return on capital; RCPIU = required capital per insured unit. 12.3.5.3.2  Implementation in Excel (MC_12.3.5_EQUITABLE PREMIUMS).  In Step 36A (case example box 12CB.21), the model calculates the expected return on capital per unit for each area and portfolio-priced premium rate. EROC per unit for area = RCPIU for area × EROC for portfolio (Step 34) (Step 35) In Step 36B the model then calculates the equitable pure risk premium income per unit for each area and portfolio-priced premium rate. Equitable pure risk premium = APPIU for area + EROC per unit for area income per unit for area (Step 31) (Step 36A) Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 206 Pricing the Base Index Case Example Box 12CB.21  Computations—Steps 36 and 37 In the case example, the expected return on capital (EROC) per unit for Area B at the 10 percent portfolio- priced premium rate is $0.57. EROC per unit for area = RCPIU for area × EROC for portfolio = $17.60 × 3.2474 percent = $0.57 The equitable pure risk premium income per unit for Area B at the 10 percent portfolio-priced premium rate is $6.03. Equitable pure risk premium income per unit for area = APPIU for area + EROC per unit for area = 5.46 + 0.57 = $6.03 In the case example, the equitable pure risk premium rate for Area B at the 10 percent portfolio-priced premium rate is 8.04 percent. Equitable pure risk premium rate for area = Equitable pure risk premium income per unit for area/AUS for area = 6.03/75 = 8.04 percent Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 207 In Step 37, the model calculates the equitable pure risk premium for each area and portfolio-priced premium rate. Equitable pure risk premium = Equitable pure risk premium/AUS for area rate for area income per unit for area (Step 36B) (Step 1) In Step 38 (case example box 12CB.22), the model calculates the final equi- table premium rate for each area and each portfolio-priced premium rate. Final equitable premium = Equitable pure risk/(100 percent − Expense loading) rate for area premium rate for area (Step 37) (Step 2) The same process is followed in calculating equitable rates at different portfo- lio premium levels and for each geographical area, A to J. This result means that Area B is less risky than the average of the other areas. See box 12.2 for a summary of the calculations illustrated in figure 12.10. Case Example Box 12CB.22  Computations—Step 38 In the case example, the final equitable premium rate for Area B at the 10 percent portfolio-priced pre- mium rate is 9.46 percent. Final equitable premium rate for area = Equitable pure risk premium rate for area/(100 percent − Expense loading) = 8.04 percent/(100 percent − 15 percent) = 9.46 percent In this case the insurer charges a 10 percent premium rate for the product across all areas of the portfo- lio, but the equitable premium rate for Area B is actually 9.46 percent. Box 12.2  Overview of Calculations for the Equitable Premiums Metrics Metrics based on at least 10,000 Monte Carlo scenarios {[ APPIU + EROC × ( RCPIU )]/AUS } Equitable premium rate = , (1− Total expense loading) box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 208 Pricing the Base Index Box 12.2  Overview of Calculations for the Equitable Premiums Metrics (continued) where APPIU = average payout per insured unit, EROC = expected return on capital, RCPIU = required capital per insured unit, and AUS = average unit size. • APPIU for area = Average payout amount for area/Number of insured units for area • Stand-alone required capital for area = TVaR payout amount for area − Expected payout amount for area • Area beta = Stand-alone required capital for area/Total stand-alone required capital for portfolio • Required capital for area = Required capital for portfolio × Area beta • RCPIU for area = Required capital for area/Number of insured units for area • EROC for portfolio = (Net premium income for portfolio − Expected portfolio losses)/ Required capital for portfolio • EROC per unit for area = RCPIU for area × EROC for portfolio • Equitable pure risk premium income per unit for area = APPIU for area + EROC per unit for area • Equitable pure risk premium rate for area = Equitable pure risk premium income per unit for area/AUS for area • Final equitable premium rate for area = Equitable pure risk premium rate for area/(100 per- cent − Expense loading) Note: TVaR = tail value at risk. 12.4  Model Outputs The model output sheet (table 12.9 and case example box 12CB.23) summa- rizes the product pricing metrics for the Base Index that were calculated in Steps 8–38. These include the following for each portfolio-priced premium rate under the no reinsurance, proportional reinsurance only, and proportional reinsurance and nonproportional reinsurance situations: • Losses • Required capital • Combined ratios • Profit margins • Probability of fund ruin • EVA • Sharpe ratio For each geographical area and at each portfolio-priced premium rate: • Equitable premium rates Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Pricing the Base Index 209 Table 12.9  Model Outputs Model component Section Excel sheet label Steps Description Model outputs 12.4 MO_12.4_MODEL None Summary of pricing decision metrics for no OUTPUTS reinsurance, proportional reinsurance only, and proportional and nonproportional reinsurance, and equitable premium rates for each geographical area Case Example Box 12CB.23  Outputs Note: EVA = economic value added; TVaR = tail value at risk. The insurance manager uses these metrics in chapter 5 to answer the key managerial questions for pricing the Base Index. The insurer will produce these same pricing metrics for any later Redesigned Indexes or prototype products by repeating the same pricing process. Note 1. In reality insurers and reinsurers may have different costs of capital, for example, because of different degrees of risk and diversification. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 210 Pricing the Base Index Bibliography Brehm, P. J. 2007. Enterprise Risk Analysis for Property & Liability Insurance Companies: A Practical Guide to Standard Models and Emerging Solutions. New York: Guy Carpenter. Cherubini, U., E. Luciano, and W. Vecchiato. 2004. Copula Methods in Finance. Hoboken, NJ: John Wiley & Sons. Crouhy, M., D. Galai, and R. Mark. 2006. The Essentials of Risk Management. New York: McGraw-Hill. Embrechts, P., F. Lindskog, and A. McNeil. 2003. “Modelling Dependence with Copulas and Applications to Risk Management.” In Handbook of Heavy Tailed Distributions in Finance, edited by S. T. Rachev, 329–84. Amsterdam: Elsevier. Grossi, P., H. Kunreuther, and C. C. Patel. 2005. Catastrophe Modeling: A New Approach to Managing Risk. New York: Springer Science Business Media. Harrison, C. M. 2004. Reinsurance Principles and Practices. Malvern, PA: American Institute for Chartered Property Casualty Underwriters/Insurance Institute of America. Lam, J. 2003. Enterprise Risk Management: From Incentives to Controls. Hoboken, NJ: Wiley. Law, A. M., and W. D. Kelton. 2006. Simulation Modeling and Analysis. 4th ed. New York: McGraw-Hill. Lehman, D. E., H. Groenendaal, and G. Nolder. 2012. Practical Spreadsheet Risk Modeling for Management. Boca Raton, FL: Chapman & Hall/CRC. Ragsdale, C. T. 2001. Spreadsheet Modeling and Decision Analysis: A Practical Introduction to Management Science. Cincinnati, OH: Southwestern College. Tang, A., and E. A. Valdez. 2009. “Economic Capital and the Aggregation of Risks Using Copulas.” University of New South Wales, Sydney, Australia. Yan, J. 2006. “Multivariate Modeling with Copulas and Engineering Applications.” In Springer Handbook of Engineering Statistics, edited by H. Pham, 973–90. London: Springer-Verlag. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 13 Evaluating the Redesigned Index 13.1  Background and Objectives Chapter 6 explains the key managerial questions for the evaluation of the Redesigned Index during the pilot phase of launching an index insurance busi- ness line. It outlines a series of steps for determining and explaining the differ- ences in the level of coverage provided by the Base and the Redesigned Indexes. As also discussed in earlier chapters, an objective of a Redesigned Index is typically to provide the policyholder with a lower-cost alternative to the Base Index. While the Base Index provides the highest level of coverage possible against inventory damage caused by the named peril, the Redesigned Index pro- vides a lower level of coverage and so has a lower cost. The implied deductible is the difference in coverage between the Base Index and the Redesigned Index. It is the amount of risk that the policyholder chooses to retain and not transfer to the insurance company. It is extremely important that the insurer always produce a Base Index to explain to the policyholder the difference between complete coverage—that provided by the Base Index—and the coverage provided by the Redesigned Index. Without this explicit comparison, policyholders often fall into the trap of expecting complete coverage even when they have purchased a less expensive product that provides lower coverage. This chapter provides a step-by-step guide to using the probabilistic models that produce the decision metrics discussed in chapter 6. The model simulates three key scenario parameters: • The payout ratios for the Base Index (Steps 10–14) • The payout ratios for the Redesigned Index (Steps 15–19) • The implied deductible amounts (Steps 20–25) The model then uses these three parameters to calculate key Redesigned Index product evaluation decision metrics, including return periods, return period ratios, Risk Modeling for Appraising Named Peril Index Insurance Products   211   http://dx.doi.org/10.1596/978-1-4648-1048-0 212 Evaluating the Redesigned Index Table 13.1  Summary of Model Components for Evaluating the Redesigned Index Model component Section Excel sheet label Steps Description Model input 13.2 MI_13.2_MODEL INPUTS Steps 1–6 User-defined assumptions, relevant portfolio information, and Base Index and Redesigned Index historical payout ratios are entered for all areas. Model computations 13.3.1 MC_13.3.1_DERIVED INPUTS Steps 7–9 Calculation of historical implied deductible ratios. These derived inputs are used for Steps 20–25. Model computations 13.3.2 MC_13.3.2_BI_SCENARIOS Steps 10–14 Simulation of scenario payout ratios for the Base Index for each area Model computations 13.3.3 MC_13.3.3_RI_SCENARIOS Steps 15–19 Simulation of scenario payout ratios for the Redesigned Index for each area Model computations 13.3.4 MC_13.3.4_IMPL-DED_ Steps 20–25 Simulation of scenario implied SCENARIOS deductible amounts for each area and for the portfolio Model computations 13.3.5 MC_13.3.5_DECISION METRICS Steps 26–31 Calculation of product evaluation decision metrics Model output 13.4 MO_13.4_MODEL OUTPUT None Summary of product evaluation decision metrics the probability of no implied deductible event occurring, and the magnitude of the expected implied deductibles (Steps 26–31). These metrics allow the insurer to understand and clearly explain to the policyholder the differences in level of cover- age between the Redesigned Index and the Base Index. The insurer will repeat the product evaluation process with any later prototype products. Table 13.1 provides a summary of the model components along with a guide to the sections in this chapter and the worksheets in the accompanying Excel files. Section 11.5 provides a brief discussion of how retrospective analysis can also be used to evaluate the Redesigned Index. 13.2  Model Inputs The analyst starts by specifying the model inputs (table 13.2) agreed upon with the insurance manager for the evaluation of the Redesigned Index. 13.2.1  Significant Payout Levels (Step 1) The significant payout level inputs are the same as those specified in section 11.2.1. However, in this section the purpose of the inputs is to facilitate the Table 13.2  Model Inputs Model component Section Excel sheet label Steps Description Model input 13.2 MI_13.2_MODEL INPUTS Steps 1–6 User defined assumptions, relevant portfolio information, and Base Index and Redesigned Index historical payout ratios are entered for all areas. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Redesigned Index 213 generation and comparison of return periods for both the Base Index and the Redesigned Indexes. Four damage levels are evaluated (case example box 13CB.1) because an index may provide insufficient coverage at mild and mild-to-medium impact levels but sufficient coverage for higher damage levels and vice versa. For a poli- cyholder that is most concerned with covering medium-to-severe damage, the ability of the Redesigned Index to make appropriate payments at the higher damage levels will be most relevant for evaluating the product. Case Example Box 13CB.1  Inputs—Step 1 13.2.2  Prediction Interval (Step 2) The prediction interval inputs (case example box 13CB.2) are the same as those specified in section 11.2.2 and will be used for the implied deductible metrics for the Redesigned Index in Steps 26–31. Case Example Box 13CB.2  Inputs—Step 2 13.2.3 Total Sum Insured per Area (Step 3) The total sum insured per area (case example box 13CB.3) will be used for gen- erating the implied deductible amounts in Steps 20–25. Case Example Box 13CB.3  Inputs—Step 3 13.2.4  Base Index Historical Payout Ratios (Step 4) The Base Index historical payout ratios (case example box 13CB.4) will be used for simulating return periods for the Base Index (Steps 10–14) and simulating implied deductible amounts for the Redesigned Index (Steps 20–25). Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 214 Evaluating the Redesigned Index Case Example Box 13CB.4  Inputs—Step 4 13.2.5  Redesigned Index Historical Payout Ratios (Step 5) The Redesigned Index historical payout ratios (case example box 13CB.5) will be used for determining the implied deductible amounts (Steps 20–25) and the expected return period for the Redesigned Index (Steps 26 and 27). 13.2.6  Nonzero Historical Payout Ratios (Step 6) In Step 6 (case example box 13CB.6) the analyst manually records all the non- zero values for the historical payout ratios from Steps 4 and 5. These inputs will Case Example Box 13CB.5  Inputs—Step 5 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Redesigned Index 215 Case Example Box 13CB.6  Inputs—Step 6 For the case example, about half of the payouts for the Base Index (25 out of 54) would not have been made with the Redesigned Index in place. This should not come as a surprise, given that the Redesigned Index provides less coverage and is less expensive. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 216 Evaluating the Redesigned Index be used in the simulation of scenario payout ratios for the Base Index (Steps 10–14) and the Redesigned Index (Steps 15–19). Note that the two columns of figures cannot be compared row-by-row and should be considered as two separate tables because the product parameters are different. However, we can make some broad conclusions based on the figures. 13.3  Model Computations The model completes five sets of computations for evaluating the Redesigned Index (table 13.3), starting with calculating the derived inputs—historical implied deductible ratios (Steps 7–9)—then simulating the three key scenario parameters (Steps 10–25), and finally producing the product evaluation decision metrics (Steps 26–31). Remember, the insurer will repeat the product evaluation computations for any later prototype products. Table 13.3  Model Computations Model component Section Excel sheet label Steps Description Model computations 13.3.1 MC_13.3.1_DERIVED INPUTS Steps 7–9 Calculation of historical implied deductible ratios. These derived inputs are used for Steps 20–25. 13.3.2 MC_13.3.2_BI_SCENARIOS Steps 10–14 Simulation of scenario payout ratios for the Base Index for each area 13.3.3 MC_13.3.3_RI_SCENARIOS Steps 15–19 Simulation of scenario payout ratios for the Redesigned Index for each area 13.3.4 MC_13.3.4_IMPL-DED_SCENARIOS Steps 20–25 Simulation of scenario implied deductible amounts for each area and for the portfolio 13.3.5 MC_13.3.5_DECISION METRICS Steps 26–31 Calculation of product evaluation decision metrics 13.3.1  Calculation of Historical Implied Deductible Ratios (Steps 7–9) 13.3.1.1 Overview The historical implied deductible ratio (table 13.4) is the difference between the historical payout ratios for the Base and Redesigned Indexes. Historical implied = Max (0, Base Index historical − Redesigned Index historical deductible ratio payout ratio payout ratio) (Step 4) (Step 5) The implied deductible ratio can only be zero or positive. It is the reduction in payouts that results from redesigning the Base Index. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Redesigned Index 217 Table 13.4  Model Computations Model component Section Excel sheet label Steps Description Model computations 13.3.1 MC_13.3.1_DERIVED INPUTS Steps 7–9 Calculation of historical implied deductible ratios. These derived inputs are used for Steps 20–25. 13.3.1.2  Implementation in Excel (MC_13.3.1_Derived Inputs) In Step 7, the model calculates the implied deductible ratio for each year and for each area (case example box 13CB.7). In Step 8 (no case example box), the model reorders the historical implied deductible ratios for all areas from Step 7, from the most recent year at the top to the least recent year at the bottom. The model will use these derived inputs in Step 31 when it selects the historical years with the largest implied deductibles. In Step 9 (case example box 13CB.8), the analyst manually combines all of the nonzero historical implied deductible ratios from Step 7 into one column. Case Example Box 13CB.7  Computations—Step 7 For the case example, there are five historical years in Area B that have positive implied deductible ratios. These years are 1989 (12.5 percent), 1997 (12.5 percent), 2009 (2.5 percent), 2011 (7.5 percent), and 2012 (5 percent). In these five historical years, the insured party would have assumed additional risk as a result of choosing the Redesigned Index rather than the Base Index. For example, in 1997, the implied deductible was 12.5 percent. Historical implied deductible ratio = Max (0, Base Index historical payout ratio − Redesigned Index historical payout ratio) = Max (0, 20 percent − 7.5 percent) = 12.5 percent Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 218 Evaluating the Redesigned Index Case Example Box 13CB.8  Computations—Step 9 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Redesigned Index 219 The model will use these derived inputs in Steps 20–25 for the simulation of implied deductible amounts. 13.3.2  Simulation of Scenario Payout Ratios for the Base Index (Steps 10–14) 13.3.2.1 Overview The model’s simulation of the scenario payout ratios for the Base Index for each area (table 13.5) incorporates three stochastic elements (frequency, severity, and correlation) just as in section 11.3.3. Table 13.5  Model Computations Model component Section Excel sheet label Steps Description Model computations 13.3.2 MC_13.3.2_BI_ Steps 10–14 Simulation of scenario SCENARIOS payout ratios for the Base Index for each area 13.3.2.2  Implementation in Excel (MC_13.3.2_BI_SCENARIOS) Steps 10–14 for simulating the scenario payout ratios for the Base Index (case example box 13CB.9) are exactly the same as Steps 14–18 in section 11.3.3. The reader is referred back to section 11.3.3 for further details on the modeling. Case Example Box 13CB.9  Computations—Steps 10–14 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 220 Evaluating the Redesigned Index 13.3.3  Simulation of Scenario Payout Ratios for the Redesigned Index (Steps 15–19) 13.3.3.1 Overview The purpose of determining the scenario payout ratios for the Redesigned Index (table 13.6) is to compare them to those for the Base Index. In this way, the model evaluates and quantifies the implied deductible assumed by the insured party when selecting the Redesigned Index. Table 13.6  Model Computations Model component Section Excel sheet label Steps Description Model computations 13.3.3 MC_13.3.3_RI_ Steps Simulation of scenario payout SCENARIOS 15–19 ratios for the Redesigned Index for each area 13.3.3.2  Implementation in Excel (MC_13.3.3_RI_SCENARIOS) Steps 15–19 for simulating the scenario payout ratios for the Redesigned Index (case example box 13CB.10) are similar to Steps 10–14 but use the historical payout ratios for the Redesigned Index (Step 5) as inputs. The reader is again referred back to section 11.3.3 for further details on the modeling. Case Example Box 13CB.10  Computations—Steps 15–19 box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Redesigned Index 221 Case Example Box 13CB.10  Computations—Steps 15–19 (continued) Comparing the results of Steps 10 and 15 for the case example, we can see that the frequency of payouts for the Redesigned Index is less than for the Base Index. These results make sense because the Redesigned Index provides a lower level of coverage. With the Redesigned Index, the insured party retains more risk than with the Base Index. In other words, some inventory damage caused by named peril events that would be covered by the Base Index will not trigger for the Redesigned Index. For example, in Area B, the Base Index triggers five payouts but the Redesigned Index triggers only two. The missing three payouts are part of the Redesigned Index’s implied deductible. On seeing the lower level of coverage provided by the Redesigned Index, the insured party may decide to purchase an index product that provides more coverage, such as the more expensive Base Index. 13.3.4  Simulation of Scenario Implied Deductible Amounts (Steps 20–25) 13.3.4.1 Overview The main objective of generating the scenario portfolio implied deductible amount (table 13.7) is to quantify the Redesigned Index’s implied deductible. Table 13.7  Model Computations Model component Section Excel sheet label Steps Description Model computations 13.3.4 MC_13.3.4_IMPL-DED_ Steps 20–25 Simulation of scenario SCENARIOS implied deductible amounts for each area and for the portfolio 13.3.4.2  Implementation in Excel (MC_13.3.4_IMPL_DED_SCENARIOS) Steps 20–25 for simulating the scenario portfolio implied deductible amount (case example box 13CB.11) are similar to Steps 10–14. However, there are two main differences between the calculations: First, these steps use the historical implied deductible ratios for the Redesigned Index as inputs (Step 9), rather than the historical payout ratios for the Base Index. Second, Steps 24 and 25 calculate monetary amounts for the payouts by area and for the total portfolio rather than just ratios. The reader is referred back to section 11.3.3 for further details on the modeling. 13.3.5  Calculation of Product Evaluation Decision Metrics (Steps 26–31) At this point the model has simulated three key scenario parameters (payout ratios for the Base Index and the Redesigned Index as well as the implied deduct- ible amounts) for the evaluation of the Redesigned Index. Based on these three parameters, the model now calculates a number of important metrics that help determine the level of coverage provided by the Redesigned Index compared with the Base Index (table 13.8). Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 222 Evaluating the Redesigned Index Case Example Box 13CB.11  Computations—Steps 20–25 For the case example, the scenario portfolio implied deductible amount is $611,718. This amount is the sum of the implied deductible amounts for Areas E ($603,966) and F ($7,751). Table 13.8  Model Computations Model component Section Excel sheet label Steps Description Model computations 13.3.5 MC_13.3.5_DECISION Steps 26–31 Calculation of product METRICS evaluation decision metrics 13.3.5.1  Expected Return Periods for the Base and Redesigned Indexes 13.3.5.1.1 Overview. Figure 13.1 provides an overview of how the model gener- ates the expected return periods for the Base and Redesigned Indexes, which is similar to the process explained in section 11.3.7.1. 13.3.5.1.2  Implementation in Excel (MC_13.3.5_DECISION METRICS). Step 26, in which the return period is simulated for the Base Index (case example box 13CB.12), is exactly the same as Step 37 in section 11.3.7.1. The reader is referred to section 11.3.7.1 for further details on the modeling. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Redesigned Index 223 Figure 13.1  Generating Return Periods for the Base and Redesigned Indexes Base and Redesigned Index scenario payout ratios (Steps 14 and 19) Significant payout levels (Step 1) Run at least 10,000 Monte Carlo scenarios Exceedance probability at significant payout levels for Base and Redesigned Indexes (Steps 26 and 27) Return period at significant payout levels for Base and Redesigned Indexes (Steps 26 and 27) Case Example Box 13CB.12  Computations—Steps 26 and 27 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 224 Evaluating the Redesigned Index In Step 27, the model calculates the exceedance probabilities and expected return periods for Redesigned Index payouts for each significant payout level and area (case example box 13CB.12). 13.3.5.2  Return Period Ratio (Step 28) 13.3.5.2.1 Overview. The return period ratio shows the level of implied deduct- ible for specific payout levels for each of the areas in the portfolio. When this ratio is equal to 1, the Redesigned Index triggers a payout at the same expected frequency as the Base Index. When the ratio is between 0 and 1, the Redesigned Index triggers payouts less frequently than the Base Index (implied deductible). We expect this outcome because the lower cost Redesigned Index provides less coverage than the Base Index. 13.3.5.2.2  Implementation in Excel (MC_13.3.5_DECISION METRICS). Step 28 is similar to Step 38 in section 11.3.7.2 but uses the return periods for the Base Index and Redesigned Index (case example box 13CB.13). In Step 28, the model calculates the return period ratio for each area and payout level. Return period ratio = Expected return period/Expected return period for for Base Index Redesigned Index (Step 26) (Step 27) Case Example Box 13CB.13  Computations—Step 28 13.3.5.3  Implied Deductible Statistics (Steps 29–31) 13.3.5.3.1 Overview. When the return period ratio is between 0 and 1, the Redesigned Index is triggering payouts at a lower expected frequency than the Base Index, indicating an implied deductible. We expect this outcome because the lower cost Redesigned Index provides less coverage than the Base Index. However, the return period ratio does not tell us whether the Redesigned Index is triggering in the right years and for the right amounts. This section explains the calculation of three additional metrics that further describe the Redesigned Index’s implied deductible: the probability of having no implied deductible amounts in the next risk period, the expected amount of the implied deductible, and historical years with the largest implied deductible amounts. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Redesigned Index 225 Figure 13.2  Generating Probability of No Implied Deductible Event and Expected Implied Deductible Amounts Scenario implied deductible amount for area (Step 24) Run at least 10,000 Monte Carlo scenarios Prediction interval (Step 4) Projected implied deductible amount Percentage of years with • Lower no implied deductible • Expected (Step 29) • Upper • TVaR (Step 30) Note: TVaR = tail value at risk. Figure 13.2 provides an overview of how the model generates the first two metrics: the probability of having no implied deductible and the amount of the implied deductible. Figure 13.3 provides an overview of how the model generates the historical years with the largest implied deductible amounts. 13.3.5.3.2 Implementation in Excel (MC_13.3.5_DECISION METRICS). Steps 29–31 are similar to Steps 39–41 in section 11.3.7.2, but use the sce- nario implied deductible amounts (Step 24) instead of the insured party basis risk amounts. In Step 29, the model generates at least 10,000 scenarios for each area based on the scenario implied deductible amounts and calculates the pro­ portion of the scenarios in which the implied deductible was zero (case example box 13CB.14). This figure indicates the percentage of years in which no implied deductible is expected for each area. Expressed differently, Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 226 Evaluating the Redesigned Index Figure 13.3  Generating Historical Years with Largest Implied Deductible Amount Base Index Redesigned Index historical historical payout ratios payout ratios (Step 4) (Step 5) Historical implied deductible ratios (Steps 7 and 8) Historical years with the largest implied deductible ratios • Year • Implied deductible ratio • Base Index historical payout ratio • Redesigned Index historical payout ratio (Step 31) Case Example Box 13CB.14  Computations—Steps 29 and 30 For the case example, in Area B there is an 80 percent chance of having no implied deductible in the next risk period. Because we know that higher percentages of years with no implied deductible correspond with a lower magnitude of implied deductible, we can tell that the implied deductible amount for Area A (91 percent of years with no implied deductible) will be the lowest of all the areas, and the amount for Area J (55 percent) will likely be the highest. The Redesigned Index affects the different areas in different ways. box continues next page Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Redesigned Index 227 Case Example Box 13CB.14  Computations—Steps 29 and 30 (continued) For the portfolio as a whole, the expected implied deductible amount is $309,342, which is 4 percent of  the total portfolio sum insured. This figure means that in the next risk period, the policyholder can expect to miss out on $309,342 in payouts that it would have received from the Base Index. For a 1-in-20-year event, the implied deductible amount is expected to be as high as $1,283,575 (that is, TVaR 95 percent of the implied deductible). Note: TVaR = tail value at risk. this figure is the probability that no implied deductible will occur during the next risk period. In Step 30, the model uses the same 10,000 scenarios to determine the expected amount of the implied deductible for the portfolio (all geographical areas). This amount is reflected in currency terms and as a percentage of the total sum insured. Based on the prediction interval selected in Step 2, the model also calculates the appropriate percentile and the tail value at risk. The values of these metrics indicate the expected magnitude of the Redesigned Index’s implied deductible. Note that when the percentage of years with no implied deductible for an area is higher, the magnitude of the implied deductible is lower and vice versa. In addition to the metrics related to the implied deductible amounts, memo- rable years in which the Redesigned Index would have failed to trigger or would have triggered payouts smaller than those for the Base Index will be of interest to the prospective policyholder. In Step 8, the model reordered the historical implied deductible ratios from the most recent to least recent year for each area. In Step 31, the model now selects the years with the largest implied deduct- ible ratios (case example box 13CB.15). In areas where this value—the largest implied deductible ratio—is repeated across multiple years, the model selects the most recent of these years. The most recent year events are chosen because pro- spective policyholders are more likely to remember these than older events. Next, the model selects each year’s corresponding Base Index and Redesigned Index payout ratios. The insurance manager will use these years as examples when explaining to the policyholder the limitations of the coverage provided by the Redesigned Index in comparison to the Base Index. Case Example Box 13CB.15  Computations—Step 31 For Area B in the case example, the largest historical implied deductible ratio was 12.5 percent, which occurred in 1989. In that year, the Base Index would have triggered a 15 percent payout but the Redesigned Index would have triggered only a 2.5 percent payout. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 228 Evaluating the Redesigned Index 13.4  Model Outputs The model output sheet summarizes the product evaluation decision metrics (box 13.1; case example box 13CB.16) for the Redesigned Index produced in Steps 7–31. These include the following: • Base Index and Redesigned Index return periods for each area • Return period ratios for each area • Probability that the Redesigned Index will have no implied deductible in the next risk period for each area • The expected amount of the implied deductible for the portfolio • Historical years with largest implied deductible events for each area Box 13.1  Overview of Calculations for the Redesigned Index Product Evaluation Metrics Derived inputs • Historical implied deductible ratio = Max (0, Base Index historical payout ratio − Redesigned Index historical payout ratio) Metrics based on at least 10,000 Monte Carlo scenarios • Return period ratio = Expected return period for Base Index/Expected return period for Redesigned Index Case Example Box 13CB.16  Outputs Note: TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Evaluating the Redesigned Index 229 The insurance manager uses these metrics in chapter 6 to answer the key managerial questions for evaluating the Redesigned Index. The insurer will produce these same outputs for any later prototype products by repeating the same product evaluation process. 13.5 Alternative Modeling Approach: Retrospective Analysis Section 11.5 detailed how to use retrospective analysis to evaluate the Base Index for basis risk. A retrospective approach can also be used to evaluate the Redesigned Index’s implied deductible. In this case the retrospective approach compares the Base Index historical payout ratios (Step 4) and the Redesigned Index historical payout ratios (Step 5). All of the analysis is based only on historical values. The reader is referred to section 11.5 for further details on the modeling. Bibliography Brehm, P. J. 2007. Enterprise Risk Analysis for Property & Liability Insurance Companies: A Practical Guide to Standard Models and Emerging Solutions. New York: Guy Carpenter. Cherubini, U., E. Luciano, and W. Vecchiato. 2004. Copula Methods in Finance. Hoboken, NJ: John Wiley & Sons. Crouhy, M., D. Galai, and R. Mark. 2006. The Essentials of Risk Management. New York: McGraw-Hill. Embrechts, P., F. Lindskog, and A. McNeil. 2003. “Modelling Dependence with Copulas and Applications to Risk Management.” In Handbook of Heavy Tailed Distributions in Finance, edited by S. T. Rachev, 329–84. Amsterdam: Elsevier. Grossi, P., H. Kunreuther, and C. C. Patel. 2005. Catastrophe Modeling: A New Approach to Managing Risk. New York: Springer Science Business Media. Lam, J. 2003. Enterprise Risk Management: From Incentives to Controls. Hoboken, NJ: Wiley. Law, A. M., and W. D. Kelton. 2006. Simulation Modeling and Analysis. 4th ed. New York: McGraw-Hill. Lehman, D. E., H. Groenendaal, and G. Nolder. 2012. Practical Spreadsheet Risk Modeling for Management. Boca Raton, FL: Chapman & Hall/CRC. Morsink, K., D. Clarke, and S. Mapfumo. 2016. “How to Measure Whether Index Insurance Provides Reliable Protection.” Policy Research Working Paper 7744, World Bank, Washington, DC. Ragsdale, C. T. 2001. Spreadsheet Modeling and Decision Analysis: A Practical Introduction to Management Science. Cincinnati, OH: Southwestern College. Tang, A., and E. A. Valdez. 2009. “Economic Capital and the Aggregation of Risks Using Copulas.” University of New South Wales, Sydney, Australia. Yan, J. 2006. “Multivariate Modelling with Copulas and Engineering Applications.” In Springer Handbook of Engineering Statistics, edited by H. Pham, 973–90. London: Springer-Verlag. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 14 Detailed Market Analysis 14.1  Background and Objectives Chapter 7 explained the key managerial questions for a detailed analysis of the broader market for index insurance beyond the pilot phase of launching an index insurance business line. For the market analysis, the insurer designs and prices a Base Index and a set of Redesigned Indexes—prototype products—which are evaluated (chapters 4 and 6) and priced (chapter 5) using the same process as in the pilot phase. The objective of the detailed market analysis is to identify the specific market segments that provide the highest expected volumes and profit for the investment of the insurer’s resources, as well as to identify the product coverage and price combinations preferred by these market segments. This chapter provides a step-by-step guide to using the probabilistic models that produce the decision metrics for the market analysis discussed in chapter 7. Table 14.1 provides a summary of the model components along with a guide to the sections in this chapter and the worksheets in the accompanying Excel files. 14.2  Model Inputs The analyst starts by specifying the model inputs agreed upon with the insurance manager for the detailed market analysis (table 14.2). 14.2.1  Internal Insurer Assumptions (Step 1) The analyst first specifies inputs based on the following internal insurer data (case example box 14CB.1): • Target loss ratio (percentage): The target loss ratio (minimum, most likely, and maximum) can be based on expert opinion (for example, the insurer’s experi- ence from other areas or regions) or on an area-specific analysis like that for the equitable premium rates in section 12.3.5. In addition, the target loss ratio may be part of the insurer’s overall risk appetite strategy. In general, we do not advise using the minimum or maximum loss ratio from the pilot phase because Risk Modeling for Appraising Named Peril Index Insurance Products   231   http://dx.doi.org/10.1596/978-1-4648-1048-0 232 Detailed Market Analysis Table 14.1  Summary of Model Components for the Detailed Market Analysis Model component Section Excel sheet label Steps Description Model input 14.2 MI_14.2_MODEL Steps 1–2 Internal insurer assumptions INPUTS and data from external market research are entered for all areas. Model computations 14.3 MC_14.3_MODEL_ Steps 3–7 Calculation of detailed market COMPUTATIONS analysis decision metrics Model outputs 14.4 MO_14.4_MODEL None Summary of detailed market OUTPUTS analysis decision metrics Table 14.2  Model Inputs Model component Section Excel sheet label Steps Description Model input 14.2 MI_14.2_MODEL INPUTS Steps 1–2 Internal insurer assumptions and data from external market research are entered for all areas. Case Example Box 14CB.1  Inputs—Step 1 Note: MFI = microfinance institution; NGO = nongovernmental organization. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Detailed Market Analysis 233 this is an annual loss ratio, which will typically have high variability. In this chapter the objective is to examine the viability of an index insurance product, which requires looking at the loss ratio over a longer period than one year to understand the trend. As a starting point, we recommend using the 25th and 75th percentile of the pilot phase loss ratio as the minimum and maximum for the detailed market analysis, implying a decision horizon of four to five years. • Target profit margin (percentage) (section 12.2.2). • Required return on capital (section 12.2.2). • Risk-free rate (section 12.2.2). • Prediction interval: Remember, the upper limit of the interval is used in calcu- lating the capital requirements. For example, if the insurer wants to hold capi- tal at 99 percent tail value at risk (TVaR; the payout amount for a 1-in-100 year event), the upper limit should be set at 99 percent. In the case example, the insurance manager and the analyst specify the upper limit as the 95th percentile. • Expense loading for each market segment and firm size. The insurer can use its experience during the pilot phase to set these values. Reinsurers may also give some guidance based on international experience. 14.2.2  External Research on the Market (Step 2) The remaining inputs are based on in-depth research on the market, for example, obtained from a specialist research firm (case example box 14CB.2). • Number of firms in market by size and market segment. • Modal portfolio size by firm size and market segment. The modal portfolio size is the average insurable amount per firm, which is specified for each market segment and firm size. The analyst must estimate the most likely minimum and maximum average insurable amount per firm. These input parameters will be used for project evaluation and review techniques (PERT) distributions to represent uncertainty about the average insurable amount per firm. • Premium rates for each prototype product. • Most popular prototype for each market segment and size. • Number of firms that are expected to purchase the most popular prototype by firm size and market segment. The analyst must estimate the most likely, minimum, and maximum number of firms. These input parameters will be used for PERT distributions to represent uncertainty about the number of firms that will purchase the product. The above list includes data that are generally needed to estimate market and market segment demand. If the available market and demand data come in a different format, the probabilistic model will have to be adjusted to take into account this alternative data. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 234 Detailed Market Analysis Case Example Box 14CB.2  Inputs—Step 2 In the case example, the insurer hires an international consulting firm to complete a detailed value chain study and stakeholder interviews on the preferred prototype option for each market segment by size of firm. The Base Prototype provides the highest level of coverage (and highest premium rate at 10 percent), and is the most popular option for nongovernmental organizations of all sizes. The Redesigned Prototype 1 provides coverage for events with at least mild-to-medium severity levels (6 percent premium rate) and is the most popular option for rural banks, microfinance institutions, and agribusinesses of all sizes. The Redesigned Prototype 2 covers only the most severe events (4 percent premium rate) and is the most popular option for seed companies of all sizes. Note: MFI = microfinance institution; NGO = nongovernmental organization. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Detailed Market Analysis 235 14.3  Model Computations The model completes one set of computations to produce the detailed market analysis decision metrics (table 14.3). 14.3.1  Premium Incomes (Step 3) 14.3.1.1  Overview Figure 14.1 provides an overview of how the model generates the expected premium income for the most popular prototype for each market segment and ­ firm size. Table 14.3  Model Computations Model component Section Excel sheet label Steps Description Model 14.3 MC_14.3_MODEL Steps 3–7 Calculation of detailed market computations COMPUTATIONS analysis decision metrics Figure 14.1  Generating Expected Premium Incomes Most Modal Scenario number popular portfolio of firms prototype size purchasing (Step 2) (Step 2) (Step 3) Premium rate (Step 2) Scenario premium income (Step 3) Prediction interval (Step 1) Run at least 10,000 Monte Carlo scenarios Projected premium income • Lower • Expected • Upper (Step 3) Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 236 Detailed Market Analysis 14.3.1.2  Implementation in Excel (MC_14_MODEL_COMPUTATIONS) In Step 3, the model determines the scenario number of firms purchasing the prototype product for each market segment and firm size using a PERT distri- bution (case example box 14CB.3). The PERT distribution accounts for uncertainty about the number of firms that will actually purchase the index product. PERT(Minimum number of firms, Most likely Scenario number of firms purchasing =  number of firms, Maximum number of firms) (Step 2) Also in Step 3, the model calculates the premium income for the most popu- lar prototype option for each market segment and firm size, Scenario premium = Modal portfolio × Premium rate × Scenario of firms purchasing size number (Step 2) (Step 2) (Step 3) Case Example Box 14CB.3  Computations—Step 3 In the case example, the scenario number of medium seed companies purchasing is three. Scenario number of firms purchasing = PERT(Minimum number of firms, Most likely number of firms, Maximum number of firms) = PERT(2, 4, 5) =3 The scenario premium income for medium seed companies is $900,000. Scenario premium income = Modal portfolio size × Premium rate × Scenario number of firms purchasing = $7,500,000 × 4 percent × 3 = $900,000 Note: MFI = microfinance institution; NGO = nongovernmental organization. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Detailed Market Analysis 237 At this point, Step 3 generates at least 10,000 scenario premium income amounts for each market segment and firm size and determines the expected premium income for each ($1,150,680 for medium seed companies in the case example). Based on the prediction interval selected in Step 1, the model also calculates the appropriate percentile values. 14.3.2  Expected Losses and Required Capital (Step 4) 14.3.2.1  Overview Figure 14.2 provides an overview of how the model generates the expected losses and required capital for the most popular prototype for each market seg- ment and firm size. Figure 14.2  Generating Expected Losses and Required Capital Scenario Scenario premium loss income ratio (Step 3) (Step 1) Scenario losses (Step 4) Prediction interval (Step 1) Run at least 10,000 Monte Carlo scenarios Projected losses • Lower Required capital • Expected (Step 4) • Upper • TVaR (Step 4) Note: TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 238 Detailed Market Analysis 14.3.2.2  Implementation in Excel (MC_14.3_MODEL COMPUTATIONS) Step 4 is similar to Steps 13 and 14 in section 12.3.2.1, but uses scenario loss ratios to calculate the scenario losses rather than historical payout ratios. In Step 4, the model determines the scenario loss ratio for each market ­ segment and firm size using a PERT distribution (case example box 14CB.4). Scenario loss = PERT(Minimum target, Most likely target, Maximum target ratio loss ratio loss ratio loss ratio) (Step 1) (Step 1) (Step 1) Case Example Box 14CB.4  Computations—Step 4 In the case example, the scenario loss ratio for medium seed companies is 74 percent. Scenario loss ratio = PERT(Minimum target loss ratio, Most likely target loss ratio, Maximum target loss ratio) = PERT(13 percent, 79 percent, 14 percent) = 73.5787 percent The scenario losses for medium seed companies are $662,208. Scenario losses = Scenario premium income × Scenario loss ratio = 900,000 × 73.5787 percent = $662,208 In the case example, the required capital for medium seed companies is $546,473. Required capital = TVaR losses − Expected losses = $1,398,334 − $851,861 = $546,473 The insurer should keep $546,473 in reserve to stay solvent in case of a 1-in-20 year event (TVaR 95 percent). Note: MFI = microfinance institution; NGO = nongovernmental organization; TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Detailed Market Analysis 239 Also in Step 4, the model calculates the scenario losses for each market seg- ment and firm size. Scenario losses = Scenario premium income × Scenario loss ratio (Step 3) At this point, Step 4 generates at least 10,000 scenario losses for each market segment and firm size and determines the expected losses for each ($851,861 for medium seed companies). Based on the prediction interval selected in Step 1, the model also calculates the appropriate percentile and TVaR values. Finally, Step 4 uses the same 10,000 scenarios to calculate the required capital for the portfolio. Required capital = TVaR losses − Expected losses 14.3.3  Combined Ratios and Profit Margins (Step 5) 14.3.3.1 Overview Figure 14.3 provides an overview of how the model generates the combined ratio and profit margin for the most popular prototype for each market segment and firm size. 14.3.3.2  Implementation in Excel (MC_14.3_MODEL_COMPUTATIONS) Step 5 is similar to Step 15 in section 12.3.2.2, but uses the scenario loss ratio based on the PERT distribution (Step 4) to calculate the scenario combined ratio. In Step 5 (case example box 14CB.5), the model first calculates the scenario combined ratio for each market segment and firm size. Scenario combined ratio = Scenario loss ratio + Expense loading (Step 4) (Step 1) Also in Step 5, the model calculates the scenario profit margin for each mar- ket segment and firm size. Scenario profit margin = 100 percent − Scenario combined ratio At this point, Step 5 generates at least 10,000 scenario combined ratios for each market segment and firm size and determines the expected combined ratio for each (89 percent for medium seed companies in the case example). Based on the prediction interval selected in Step 1, the model also calculates the appropri- ate percentile values. Step 5 also generates at least 10,000 scenario profit margins for each market segment and firm size and determines the expected profit margin (11 percent for Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 240 Detailed Market Analysis Figure 14.3  Generating Expected Combined Ratios and Profit Margins Scenario Expense loss ratio loading (Step 4) (Step 1) Target profit margin (Step 1) Scenario combined ratio Scenario profit margin (Step 5) (Step 5) Run at least 10,000 Monte Carlo scenarios Prediction interval (Step 1) Projected profit margins Projected combined ratio • Lower • Expected • Lower • Upper • Expected • TVaR • Upper • Probability of negative profit • Probability of profit (Step 5) below target (Step 5) Note: TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Detailed Market Analysis 241 Case Example Box 14CB.5  Computations—Step 5 For the case example, the scenario combined ratio for medium seed companies is 89 percent. Scenario combined ratio = Scenario loss ratio + Expense loading = 74 percent + 15 percent = 89 percent The scenario profit margin for medium seed companies is 11 percent. Scenario profit margin = 100 percent − Scenario combined ratio = 100 percent − 89 percent = 11 percent The probability of a negative profit is 31 percent. Probability of a negative profit = Number of scenarios with profit < 0 /Total number of scenarios = 3,100/10,000 = 31 percent The probability of a profit below the target profit margin is 50 percent. Probability of a profit below the target profit margin = Number of scenarios with profit margin < Target profit margin/ Total number of scenarios (Step 1) = 5,000/10,000 = 50 percent Note: MFI = microfinance institution; NGO = nongovernmental organization. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 242 Detailed Market Analysis medium seed companies in the case example) and appropriate percentile values for each. Finally, Step 5 calculates the probability of a negative profit and the probabil- ity of a profit below the target profit margin (Step 1) for each market segment and firm size. Total number Probability of a negative profit = Number of scenarios with profit < 0 / of scenarios Probability of a profit below = Number of scenarios < Target profit margin/ the target profit margin with profit margin Total number of scenarios (Step 1) 14.3.4  Economic Value Added (Step 6) 14.3.4.1 Overview Economic value added (EVA) measures the flow of economic value created from a business, taking into account the costs of the firm’s capital. EVA is the difference between the value derived from selling the product and the cost of doing so. The reader is referred to section 12.3.2.4 for more detail on EVA. Figure 14.4 provides an overview of how the model generates the EVA for the most popular prototype for each market segment and firm size. 14.3.4.2  Implementation in Excel (MC_14.3_MODEL COMPUTATIONS) In Step 6 (case example box 14CB.6), the model calculates the capital charge and expense amount for each of the market segments. Scenario capital charge = Scenario required capital × Required return on capital (Step 4) (Step 1) Scenario expense amount = Expense loading × Scenario premium income (Step 1) (Step 3) Also in Step 6, the model calculates the scenario EVA. (Scenario Scenario Scenario = premium − Scenario − expense − Capital charge)/ EVA income losses amount Required capital (Step 3) (Step 4) (Step 4) At this point, Step 6 generates at least 10,000 scenario EVA results for each market segment and firm size and determines the expected EVA for each (18 percent for medium seed companies in the case example). Based on the prediction interval selected in Step 1, the model also calculates the appropriate percentile values of the EVA. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Detailed Market Analysis 243 Figure 14.4  Generating Projected Values for Economic Value Added (EVA) Metrics Required capital Required return on Scenario premium Expense (1st simulation) capital income loading (Step 4) (Step 1) (Step 3) (Step 1) Scenario capital charge Scenario expense amount (Step 6) (Step 6) Scenario losses (Step 4) Scenario economic value added (Step 6) Prediction Run at least 10,000 interval Monte Carlo scenarios (Step 1) Projected economic value added • Lower • Expected • Upper (Step 6) Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 244 Detailed Market Analysis Case Example Box 14CB.6  Computations—Step 6 In the case example, the scenario capital charge for medium seed companies is $27,324. Scenario capital charge = Scenario required capital × Required return on capital = $546,473 × 5 percent = $27,324 The scenario expense amount for medium seed companies is $135,000. Scenario expense amount = Expense loading × Scenario premium income = 15 percent × 900,000 = $135,000 The scenario EVA for medium seed companies is 30 percent. Scenario EVA = (Scenario premium income − Scenario losses − Scenario expense amount − Scenario capital charge)/ Required capital = (900,000 − 572,817 − 135,000 − 27,324)/546,473 = 30 percent Note: MFI = microfinance institution; NGO = nongovernmental organization. 14.3.5  Sharpe Ratio (Step 7) 14.3.5.1 Overview Figure 14.5 provides an overview of how the model generates the Sharpe ratios for only the most popular prototype for each market segment and firm size. 14.3.5.2  Implementation in Excel (MC_14.3_MODEL COMPUTATIONS) Step 7 is similar to Step 18 in section 12.3.2.5, but includes the scenario expense amount in the calculation of the scenario return on capital. In Step 7 (case example box 14CB.7), the model calculates the scenario return on capital for each market segment and firm size. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Detailed Market Analysis 245 Figure 14.5  Generating Sharpe Ratios Scenario premium Scenario Scenario expense Required capital income losses amount (1st simulation) (Step 3) (Step 4) (Step 6) (Step 4) Scenario return on capital (Step 7) Run at least 10,000 Monte Carlo scenarios Projected return on capital • Expected • Standard deviation Risk-free (Step 7) rate (Step 1) Sharpe ratio (Step 7) Scenario return = (Scenario premium − Scenario − Scenario expense/Required on capital income losses amount) capital (Step 3) (Step 4) (Step 6) (Step 4) At this point, the model generates at least 10,000 Monte Carlo scenario returns on capital for each market segment and firm size and determines the expected return on capital for each (23 percent for medium seed companies in the case example) as well as the standard deviation of the return on capital (40 percent). Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 246 Detailed Market Analysis Case Example Box 14CB.7  Computations—Step 7 The scenario return on capital for medium seed companies is 35 percent for the case example. Scenario return on capital = (Scenario premium income − Scenario losses − Scenario expense amount)/Required capital = (900,000 − 572,817 − 135,000)/546,473 = 35 percent For the case example, medium seed companies have a Sharpe ratio of 0.53. Sharpe ratio = (Expected return on capital − Risk-free rate)/Standard deviation of expected return on capital = (23 percent − 2 percent)/40 percent = 0.53 Note: MFI = microfinance institution; NGO = nongovernmental organization. Finally, the model calculates the Sharpe ratio. Sharpe ratio = (Expected return − Risk-free)/Standard deviation of expected on capital rate return on capital (Step 1) A positive Sharpe ratio indicates an investment with an expected positive return per unit of risk assumed. Premium rates with higher Sharpe ratios are preferred because the higher the Sharpe ratio, the greater the expected return on the capital invested relative to the amount of risk taken. As a reminder, all metrics within this guide, including the Sharpe ratio, are calculated for the specific index product and do not take into account other products the insurer has in the market. In general, the insurer should also evalu- ate how the Sharpe ratio of its overall product portfolio is affected by the new index product. See box 14.1 for a summary of the market analysis calculations. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Detailed Market Analysis 247 Box 14.1  Overview of Calculations for the Detailed Market Analysis Metrics Scenario metrics (one Monte Carlo scenario) • Number of firms purchasing = PERT(Minimum number of firms, Most likely number of firms, Maximum number firms) • Premium income = Modal portfolio size × Premium rate × Scenario number of firms purchasing • Loss ratio = PERT(Minimum target loss ratio, Most likely target loss ratio, Maximum target loss ratio) • Scenario losses = Scenario premium income × Scenario loss ratio • Scenario combined ratio = Scenario loss ratio + Expense loading • Scenario profit margin = 100 percent − Scenario combined ratio • Capital charge = Scenario required capital × Required return on capital • Scenario expense amount = Expense loading × Scenario premium income • Scenario EVA = (Scenario premium income − Scenario losses − Scenario expense amount − Capital charge)/Required capital • Scenario return on capital = (Scenario premium income − Scenario losses − Scenario expense amount)/Required capital Metrics based on at least 10,000 Monte Carlo scenarios • Premium income • Losses • Required capital = TVaR losses − Expected losses • Combined ratio • Profit margin • Probability of a negative profit = Number of scenarios with profit < 0/Total number of scenarios • Probability of profit below the target profit margin = Number of scenarios with profit margin < target profit margin/Total number of scenarios • Economic value added • Return on capital • Sharpe ratio = (Expected return on capital − Risk-free rate)/Standard deviation of return on capital Note: EVA = economic value added; PERT = project evaluation and review techniques; TVaR = tail value at risk. 14.4  Model Outputs (MO_14.4_MODEL OUTPUTS) The model output sheet (case example box 14CB.8) summarizes the detailed market analysis decision metrics for the Redesigned Index produced in Steps 3–7 (table 14.4). These include the following: • Premium income • Projected losses Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 248 Detailed Market Analysis • Projected combined ratio • Projected profit margins • Probability of negative profit • Probability of profit below target • EVA • Sharpe ratios Case Example Box 14CB.8  Outputs Note: EVA = economic value added; MFI = microfinance institution; NGO = nongovernmental organization. Table 14.4  Model Outputs Model component Section Excel sheet label Steps Description Model outputs 14.4 MO_14.4_MODEL None Summary of detailed market OUTPUTS analysis decision metrics The insurance manager uses these metrics in chapter 7 to answer the key managerial questions for a detailed analysis of the broader market for index insurance beyond the pilot phase. Bibliography Brehm, P. J. 2007. Enterprise Risk Analysis for Property & Liability Insurance Companies: A Practical Guide to Standard Models and Emerging Solutions. New York: Guy Carpenter. Cherubini, U., E. Luciano, and W. Vecchiato. 2004. Copula Methods in Finance. Hoboken, NJ: John Wiley & Sons. Crouhy, M., D. Galai, and R. Mark. 2006. The Essentials of Risk Management. New York: McGraw-Hill. Embrechts, P., F. Lindskog, and A. McNeil. 2003. “Modelling Dependence with Copulas and Applications to Risk Management.” In Handbook of Heavy Tailed Distributions in Finance, edited by S. T. Rachev, 329–84. Amsterdam: Elsevier. Grossi, P., H. Kunreuther, and C. C. Patel. 2005. Catastrophe Modeling: A New Approach to Managing Risk. New York: Springer Science Business Media. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Detailed Market Analysis 249 Lam, J. 2003. Enterprise Risk Management: From Incentives to Controls. Hoboken, NJ: Wiley. Law, A. M., and W. D. Kelton. 2006. Simulation Modeling and Analysis. 4th ed. New York: McGraw-Hill. Lehman, D. E., H. Groenendaal, and G. Nolder. 2012. Practical Spreadsheet Risk Modeling for Management. Boca Raton, FL: Chapman & Hall/CRC. Ragsdale, C. T. 2001. Spreadsheet Modeling and Decision Analysis: A Practical Introduction to Management Science. Cincinnati, OH: Southwestern College. Tang, A., and E. A. Valdez. 2009. “Economic Capital and the Aggregation of Risks Using Copulas.” University of New South Wales, Sydney, Australia. Yan, J. 2006. “Multivariate Modelling with Copulas and Engineering Applications.” In Springer Handbook of Engineering Statistics, edited by H. Pham, 973–90. London: Springer-Verlag. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 15 Value of Index Insurance 15.1  Background and Objectives Chapter 8 explained the key managerial questions for evaluating the value of index insurance for a specific market segment—financial service providers such as microfinance institutions, commercial banks, and agribusinesses that provide financing to smallholder farmers. The objective of the analysis is to determine the extent to which the named peril index insurance prototype product can reduce the service provider’s losses during years with high defaults, as well as the maximum price the service provider will be willing to pay for this reduction in losses (that is, the value of the index insurance). This chapter provides a step-by-step guide (table 15.1) to using the probabi- listic models that produce the decision metrics discussed in chapter 8. The model simulates two key scenario parameters: the gross default rate (Steps 10–16) and net default rate for the prototype product (Steps 17–23). The model then uses these parameters to calculate the value of index insur- ance decision metrics (Steps 24–30). These metrics allow the insurer to clearly explain to the policyholder the degree to which the named peril affects the default rates and the relative benefits of purchasing the index insurance proto- type product. The model in this chapter relies on two key assumptions: First, the quality of the analysis depends on the reliability of the historical default rate data. The calculation of the value of index insurance will not be reliable if the default data are not reliable (for example, missing data points in certain geographical areas or years). However, the analysis can still be directionally useful as long as the insurer and the financial service provider both understand and appreciate the limitations of the data. Second, the analysis assumes that the effect of the named peril on the financial service provider’s portfolio is felt at the end of the risk period, for example, an agricultural loan that is payable as a bullet payment at the end of the season. This assumption is violated if loan repayments are made weekly or monthly, and the peril is one such as typhoons over a six-month period. In these cases, Risk Modeling for Appraising Named Peril Index Insurance Products   251   http://dx.doi.org/10.1596/978-1-4648-1048-0 252 Value of Index Insurance the default data should relate to repayment soon after the occurrence of the event (for example, 30 days or 60 days after a typhoon), and the data processing becomes more complicated. Such analysis is beyond the scope of this guide. 15.2  Model Inputs The analyst starts by specifying the model inputs agreed upon with the insurance manager (table 15.2) for the value of index insurance analysis. 15.2.1  Data from the Policyholder (Steps 1–5) The analyst first specifies inputs based on internal data obtained from the pro- spective policyholder (case example box 15CB.1): • Target maximum annual default rate (percent). This metric provides an indi- cation of the financial service provider’s risk tolerance. • The financial service provider’s cost of capital (percent). • Debt recovery expense (percentage of nonperforming loans). These are the costs incurred by the financier to try to recover debt. • Prediction interval (percent). • Historical default rates by geographic area (restructures and write-offs). Ideally, more than 10 years of data for each area should be available. • Distribution of loans by geographic area (percent). Table 15.1  Summary of Model Components for the Value of Index Insurance Model component Section Excel sheet label Steps Description Model input 15.2 MI_15.2_MODEL Steps 1–7 Data from the prospective policyholder and INPUTS historical payout ratios for the prototype product are entered. Model computations 15.3.1 MC_15.3.1_DERIVED Steps 8–9 Calculation of historical net default rates for INPUTS each area. These derived inputs are used for Steps 24–30. 15.3.2 MC_15.3.2_GROSS Steps 10–16 Simulation of scenario gross default rates for NPL SCENARIOS each area and the portfolio. 15.3.3 MC_15.3.3_NET NPL Steps 17–23 Simulation of scenario net default rates for SCENARIOS each area and the portfolio. 15.3.4 MC_15.3.4_DECISION Steps 24–30 Calculation of value of index insurance METRICS decision metrics. Model outputs 15.4 MO_15.4_MODEL None Summary of value of index insurance OUTPUTS decision metrics. Table 15.2  Model Inputs Model component Section Excel sheet label Steps Description Model input 15.2 MI_15.2_Model Inputs Steps 1–7 Data from the prospective policyholder and historical payout ratios for the prototype product are entered. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Value of Index Insurance 253 Case Example Box 15CB.1  Inputs—Steps 1–5 In the case example, the prospective policyholder discussed in this chapter is a large agribusiness interested in Redesigned Prototype 1 (6 percent premium rate). The agribusiness provides the insurer with 10 years of default data for 10 geographical areas. 15.2.2  Historical Payout Ratios for the Prototype Product (Step 6) The historical payout ratios are those for the prototype product the insurer will evaluate (case example box 15CB.2). 15.2.3  Nonzero Gross Default Rates (Step 7) In Step 7 (no case example box), the analyst manually records all the nonzero values for gross default rates from Step 4. These figures will be used in Step 10 to fit a probability distribution to the nonzero historical gross default rates. Case Example Box 15CB.2  Inputs—Step 6 In the case example, the insurer uses the historical payout ratios for Redesigned Prototype 1. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 254 Value of Index Insurance 15.3  Model Computations The model completes four main sets of computations (table 15.3) to analyze the value of index insurance for the prototype product, starting with calculating the derived inputs—net default rates (Steps 8–9)—then simulating the two key scenario parameters (Steps 10–23), and finally producing the value of index insurance decision metrics (Steps 24–30). 15.3.1  Calculation of Historical Net Default Rates (Steps 8–9) 15.3.1.1 Overview The historical net default rate (table 15.4) is the amount of default risk that the policyholder would have retained if the prototype policy had been in place in the past. It is the default risk that is not linked to the named peril and so not covered by the policy. It is calculated as follows: Historical net default rate = Max (0,[Historical gross default rate − Historical payout ratio]). Step 4 Step 6 Insurance is meant to indemnify losses and not enrich the insured. The value of the historical net default rate cannot be less than zero because that would be an enrichment of the insured. Table 15.3  Model Computations Model component Section Excel sheet label Steps Description Model computations 15.3.1 MC_15.3.1_DERIVED_INPUTS Steps 8–9 Calculation of historical net default rates for each area. These derived inputs are used for Steps 24–30. Model computations 15.3.2 MC_15.3.2_GROSS NPL Steps 10–16 Simulation of scenario gross default SCENARIOS rates for each area and the portfolio Model computations 15.3.3 MC_15.3.3_NET NPL Steps 17–23 Simulation of scenario net default rates SCENARIOS for each area and the portfolio Model computations 15.3.4 MC_15.3.4_DECISION Steps 24–30 Calculation of value of index insurance METRICS decision metrics Table 15.4  Model Computations Model component Section Excel sheet label Steps Description Model computations 15.3.1 MC_15.3.1_DERIVED INPUTS Steps 8–9 Calculation of historical net default rates for each area. These derived inputs are used for Steps 24–30. 15.3.1.2 Implementation in Excel (MC_15.3.1_DERIVED INPUTS) In Step 8 (case example box 15CB.3), the model calculates the historical net default rate for each area and year. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Value of Index Insurance 255 Case Example Box 15CB.3  Computations—Step 8 For Area B in the case example, the historical net default rate is 3 percent for 2009. Historical net default rate = Max(0,[Historical gross default rate − Historical payout ratio]) = Max(0,[3 percent − 0 percent]) = 3 percent In Step 9 (no case example box), the analyst manually records all the nonzero values for historical net default rates from Step 8. These figures will be used in Step 17 to fit a probability distribution to the nonzero historical net default rates. 15.3.2  Simulation of Scenario Gross Default Rates (Steps 10–16) 15.3.2.1 Overview The purpose of determining the scenario gross default rates (table 15.5), which illustrate the situation in which the policyholder has no index insurance cover- age, is to compare them with the net default rates, which describe the situation in which the policyholder has purchased the prototype index product. 15.3.2.2  Implementation in Excel (MC_15.3.2_GROSS NPL SCENARIOS) Steps 10–16 (case example box 15CB.4) for simulating the scenario gross default rates are similar to Steps 14–18 in section 11.3.3. However, there are two main differences between the calculations. First, Steps 10–16 use the historical nonzero gross default rates from the policyholder (Step 7), rather than the historical pay- out ratios for the Base Index. Second, Step 16 calculates a weighted average of the default rates for all areas using the distribution of loans by geographic area (Step 5) as weights. Scenario Scenario gross Scenario gross … Scenario gross gross default rate for Area default rate for Area default rate for Area = + + portfolio A × Distribution of B × Distribution of N × Distribution default rate loans for Area A loans for Area B of loans for Area N (Step 15) (Step 15) (Step 15) The reader is referred back to section 11.3.3 for further detail on the modeling. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 256 Value of Index Insurance Table 15.5  Model Computations Model component Section Excel sheet label Steps Description Model computations 15.3.2 MC_15.3.2_GROSS Steps 10–16 Simulation of scenario gross NPL SCENARIOS default rates for each area and the portfolio Case Example Box 15CB.4  Computations—Steps 10–16 In the case example the scenario gross default rate is 3.67 percent. Scenario portfolio gross default rate = Scenario gross default rate for Area A × Distribution of loans for Area A + Scenario gross default rate for Area B × Distribution of loans for Area B + … Scenario gross default rate for Area J × Distribution of loans for Area J = 3.6 percent ×10 percent + 1.0 percent ×10 percent + 11.2 percent ×10 percent + 2.6 percent × 10 percent + 0.0 percent × 10 percent + 7.2 percent × 10 percent + 4.7 percent × 10 percent + 2.5 percent × 10 percent + 0.9 percent ×10 percent + 2.9 percent ×10 percent = 3.67 percent Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Value of Index Insurance 257 15.3.3  Simulation of Scenario Net Default Rates (Steps 17–23) 15.3.3.1 Overview The scenario net default rates (table 15.6) provide information on the policy- holder’s defaults in the situation in which it is covered by the prototype index product. These are compared with the gross default rates for the situation with- out coverage to determine the value of the index insurance. 15.3.3.2  Implementation in Excel (MC_15.3.3_NET NPL SCENARIOS) Steps 17–23 for simulating the scenario net default rates (case example box 15CB.5) are similar to Steps 10–16 but use the historical nonzero net default rates (Step 9) as inputs. The reader is referred back to section 11.3.3 for further detail on the modeling. Table 15.6  Model Computations Model component Section Excel sheet label Steps Description Model computations 15.3.3 MC_15.3.3_NET NPL Steps 17–23 Simulation of scenario net SCENARIOS default rates for each area and the portfolio Case Example Box 15CB.5  Computations—Steps 17–23 Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 258 Value of Index Insurance 15.3.4  Calculation of Value of Index Insurance Decision Metrics (Steps 24–30) At this point the model has simulated two key scenario parameters (gross and net default rates for the prototype product) for the analysis of the value of index insurance. Based on these parameters, the model now calculates the metrics that help determine the value of the index insurance to the prospective policyholder (table 15.7). 15.3.4.1 Overview Figure 15.1 provides an overview of how the model generates the cost of the gross default risk, the net default risk, and the value of index insurance. 15.3.4.2  Implementation in Excel (MC_15.3.4_DECISION METRICS) In Step 24 (case example box 15CB.6), the model generates at least 10,000 scenario gross portfolio default rates and calculates the proportion of these that are greater than the target maximum default rate (Step 1). This figure represents the probability of the policyholder’s portfolio having a default rate greater than the target. In Step 25, the model uses the same 10,000 scenarios to determine the expected gross portfolio default rate. Based on the prediction interval selected in Step 3, the model also calculates the appropriate percentile and tail value at risk (TVaR) values. Step 26 calculates the required capital and the cost of the gross portfolio default risk. Required capital = TVaR gross portfolio − Expected gross portfolio (gross default rate) default rate default rate (Step 25) (Step 25) Cost of gross portfolio = (Expected gross portfolio + Cost of capital × Required capital)/ default risk default rate (1 − Debt recovery expense) (Step 25) (Step 2) (Step 2) The required capital is included in the cost of the default risk because the financial service provider must reserve capital to remain solvent following an extreme event (TVaR). The debt recovery expense is included in the cost of default risk because the financial service provider incurs this cost while trying to recover delinquent loans. In Step 27 (case example box 15CB.7), the model generates at least 10,000 scenario net portfolio default rates and calculates the probability of the policy- holder’s portfolio having a default rate greater than the target (Step 1) if the financial service provider has purchased the prototype product. Table 15.7  Model Computations Model component Section Excel sheet label Steps Description Model computations 15.3.4 MC_15.3.4_DECISION Steps 24–30 Calculation of value of index METRICS insurance decision metrics Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Value of Index Insurance 259 Figure 15.1  Generating the Value of Index Insurance Decision Metrics Scenario gross and net portfolio default rate (Steps 16 and 23) Prediction interval (Step 3) Run at least 10,000 Monte Carlo scenarios Projected gross and net portfolio Required capital default rate (gross and net default • Lower rate) • Expected (Steps 26 and 29) • Upper • TVaR (Steps 25 and 28) Debt Target maximum Cost of capital recovery expense default rate (Step 2) (Step 2) (Step 2) Cost of gross and net Probability of gross and portfolio default risk net portfolio default value of index insurance rate greater than target (Steps 25 and 28) (Steps 24 and 27) Note: TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 260 Value of Index Insurance Case Example Box 15CB.6  Computations—Steps 24–26 For the case example, the probability of a default greater than the target maximum (4 percent) in the case in which the large agribusiness does not have index insurance coverage is 59 percent (gross default rate). The expected gross portfolio default rate is 4.42 percent of the total portfolio value. For a 1-in-20 year event, the gross default rate is expected to be 7.81 percent (TVaR). The required capital for this portfolio is 3.39 percent of the total portfolio value (not shown). Required capital (gross default rate) = TVaR gross portfolio default rate − Expected gross portfolio default rate = 7.81 percent − 4.42 percent = 3.39 percent The cost to the agribusiness of the gross portfolio default risk is 5.73 percent of the total portfolio value. = (Expected gross portfolio default rate + Cost of capital Cost of gross portfolio default risk  × Required capital)/(1 − Debt recovery expense) = (4.42 percent + 5 percent × 3.39 percent)/(1 − 0.2) = 5.73 percent Even though the expected portfolio gross default rate is 4.42 percent per year, the actual costs to the agribusiness due to defaults are 5.73 percent because the business incurs expenses for debt recovery and must reserve capital in case of an extreme event (TVaR). Note: TVaR = tail value at risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Value of Index Insurance 261 Case Example Box 15CB.7  Computations—Steps 27–29 For the case example, the probability of a default greater than the target maximum (4 percent) in the case in which the large agribusiness purchases the Redesigned Prototype 1 is 0 percent, a significant reduction from the situation without index coverage (59 percent). The expected net portfolio default rate is 2.41 percent of the total portfolio value versus 4.42 percent without insurance. The cost to the large agribusiness of the net portfolio default risk is 3.08 percent, down from 5.73 percent. The Redesigned Prototype 1 significantly reduces the default rate and the cost of the default risk for the agribusiness. The required capital (net default rate) is 1.03 percent (not shown in illustration of steps). Required capital (net default rate) = TVaR net portfolio default rate − Expected net portfolio default rate = 3.44 percent − 2.41 percent = 1.03 percent The cost of the net portfolio default risk is 3.08 percent. (Expected net portfolio default rate + Cost of capital Cost of net portfolio default risk =  × Required capital)/(100 percent − Debt recovery expense) = (2.41 percent + 5 percent × 1.03 percent)/(100 percent − 20 percent) = 3.08 percent Note: TVaR = tail value at risk. In Step 28, the model uses the same 10,000 scenarios to determine the expected net portfolio default rate if the financial service provider is covered by the prototype product. Based on the prediction interval selected in Step 3, the model also calculates the appropriate percentile and TVaR values. Step 29 calculates the required capital and the cost of the net portfolio default risk. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 262 Value of Index Insurance Required capital = TVaR net portfolio − Expected net portfolio (net default rate) default rate default rate (Step 28) (Step 28) Cost of net portfolio = (Expected net portfolio + Cost of capital × Required capital)/ default risk default rate (100 percent − Debt recovery expense) (Step 25) (Step 2) (Step 2) In Step 30 (case example box 15CB.8), the model calculates the value of index insurance. Cost of gross portfolio default risk − Value of index insurance =  Cost of net portfolio default risk Case Example Box 15CB.8  Computations—Step 30 In other words, the value of the index insurance is the difference between the cost of default risk without index insurance and the cost of default risk with the index insurance policy in place. Based on the results of this valuation of index insurance, the financial service provider will likely not be willing to pay a premium rate that is much higher than the value of insurance metric calculated by the model. However, to arrive at the final premium for the product, the insurer will need to load the value of insurance metric with expenses and profits (case example box 15CB.9). See box 15.1 for a summary of the value of insurance metrics. Case Example Box 15CB.9 Evaluating the Relevance of Insurance to a Specific Financier In the case example, the value of index insurance for the large agribusiness is 2.65 percent of the total portfolio value. A premium rate of up to about 3 percent should be acceptable to the agribusiness. Remember that the premium rate for the Redesigned Prototype 1 is 6 percent. At this point, the insurance manager may return to the pricing process completed in chapter 5. If the pricing for the pro- totype product is close to the value of the insurance, the insurance manager can use the value of insurance metrics to offer the product to the agribusiness. Alternatively, the insurance manager may determine that it is not feasible to offer the product at the required rate and inform the agribusiness of the results from the analysis. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Value of Index Insurance 263 Box 15.1  Overview of Calculations for the Value of Index Insurance Metrics Derived inputs • Historical net default rate = Max[0,(Historical gross default rate − Historical payout ratio)] Scenario metrics (one Monte Carlo scenario) • Scenario gross portfolio default rate = Scenario gross default rate for Area A × Distribution of loans for Area A + Scenario gross default rate for Area B × Distribution of loans for Area B + …Scenario gross default rate for Area N × Distribution of loans for Area N Metrics based on at least 10,000 Monte Carlo scenarios • Probability of the policyholder’s portfolio having a default rate greater than the target = Number of scenarios in which default rate > target/Total number of scenarios • Required capital = TVaR default rate − Expected default rate • Cost of default risk = (Expected default rate + Cost of capital × Required capital)/(1 Debt recovery expense) • Value of index insurance = Cost of gross portfolio default risk − Cost of net portfolio default risk Note: TVaR = tail value at risk. 15.4  Model Outputs (MO_15.4_MODEL OUTPUTS) The model output sheet (table 15.8 and case example box 15CB.10) summa- rizes the value of insurance analysis decision metrics for a specific financial insti- tution’s lending portfolio. These include the following: For both gross and net portfolio default rates • Probability of default rate greater than target • Expected default rate • TVaR default rate • Cost of default risk For the financial service provider’s portfolio • Value of index insurance The insurance manager uses these metrics in chapter 8 to answer the key managerial questions about the value of index insurance to a financier. Table 15.8  Model Outputs Model component Section Excel sheet label Steps Description Model outputs 15.4 MO_15.4_MODEL None Summary of value of index OUTPUTS insurance decision metrics Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 264 Value of Index Insurance Case Example Box 15CB.10  Outputs Bibliography Brehm, P. J. 2007. Enterprise Risk Analysis for Property & Liability Insurance Companies: A Practical Guide to Standard Models and Emerging Solutions. New York: Guy Carpenter. Cherubini, U., E. Luciano, and W. Vecchiato. 2004. Copula Methods in Finance. Hoboken, NJ: John Wiley & Sons. Crouhy, M., D. Galai, and R. Mark. 2006. The Essentials of Risk Management. New York: McGraw-Hill. Embrechts, P., F. Lindskog, and A. McNeil. 2003. “Modelling Dependence with Copulas and Applications to Risk Management.” In Handbook of Heavy Tailed Distributions in Finance, edited by S. T. Rachev, 329–84. Amsterdam: Elsevier. Grossi, P., H. Kunreuther, and C. C. Patel. 2005. Catastrophe Modeling: A New Approach to Managing Risk. New York: Springer Science Business Media. Lam, J. 2003. Enterprise Risk Management: From Incentives to Controls. Hoboken, NJ: Wiley. Law, A. M., and W. D. Kelton. 2006. Simulation Modeling and Analysis. 4th ed. New York: McGraw-Hill. Lehman, D. E., H. Groenendaal, and G. Nolder. 2012. Practical Spreadsheet Risk Modeling for Management. Boca Raton, FL: Chapman & Hall/CRC. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Value of Index Insurance 265 Ragsdale, C. T. 2001. Spreadsheet Modeling and Decision Analysis: A Practical Introduction to Management Science. Cincinnati, OH: Southwestern College. Tang, A., and E. A. Valdez. 2009. “Economic Capital and the Aggregation of Risks Using Copulas.” University of New South Wales, Sydney, Australia. Yan, J. 2006. “Multivariate Modelling with Copulas and Engineering Applications.” In Springer Handbook of Engineering Statistics, edited by H. Pham, 973–90. London: Springer-Verlag. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 16 Alternative Probabilistic Modeling Approaches 16.1  Overview of Alternative Approaches As seen in chapters 11 through 15, the simulation of payout ratios is a key ele- ment of many of the model computations related to index insurance in this guide. This fact reflects the modeling solution that we found most appropriate for the set of index insurance problems addressed here. As with any model selec- tion, the approach presented in this guide makes a number of assumptions and has a number of limitations. Chapter 9 outlines three assumptions and limitations to the models in this guide. It is assumed that the models • Evaluate the index product in isolation from the insurer’s other products. • Consider only a one-year time horizon. • Assume no changes in the underlying system over time.1 As we have said before and stress again, understanding the assumptions and limitations of any model is important, and that goes for these models as well. This chapter looks in depth at three modeling approaches for simulating payout ratios: • Simulating the payout amounts directly • Simulating the index that drives the payouts • Simulating the weather that drives the index values The first approach is the one used to simulate payout amounts throughout this guide and the remaining two are alternative approaches. Figure 16.1 provides an overview of the three approaches. It is important to keep in mind the following three points: • Approach One is the simplest of the three approaches because it simulates product payouts directly based on historical data. Risk Modeling for Appraising Named Peril Index Insurance Products   267   http://dx.doi.org/10.1596/978-1-4648-1048-0 268 Alternative Probabilistic Modeling Approaches Figure 16.1  Overview of Three Approaches to Simulating Payout Ratios Weather Simulating the Simulating the Simulating the index weather weather product payout product over time index amounts terms Approach #1 Approach #2 Approach #3 • Approach Three is the most complex because it uses a model to simulate many aspects of the weather. Based on this general weather model, this approach simulates the specific index, then combines it with the product terms, and finally simulates the payouts. • Approaches Two and Three both explicitly take into account the product terms in the model simulation. Approach One only takes into account the product terms through the historical payout data that are used. 16.2 Approach 1: Simulating the Payouts Directly Before discussing some of the key assumptions and limitations of this framework, let us summarize the different steps of how total payout amounts are projected for a number of regions. First, the model uses a Bernoulli distribution to simulate whether there will be a payout greater than 0 percent for each year in each area. This Bernoulli distribution can be seen as a frequency distribution,2 except that in this case the frequency can only be 0 or 1. This frequency distribution is then combined with a distribution to describe the actual payout as a percentage of the insured amount. The parameter uncertainty for the probability of a payout per region per year is modeled using a beta distribution. Second, the model simulates the payouts as percentages of the insured amount (payout ratios).3 The payout amount as a percentage of the insured amount can also be called the severity. In the case example, we used a beta distribution that we fit to all payout ratios greater than 0 percent for the past ­ 30 years of data. The historical data tell us what payouts would have occurred in each of the past 30 years if the index product had been in effect.4 In summary, we are able to consult the historical data and determine the index values for each year and then use the index values combined with the index product terms to determine the payout ratios for each year. The beta distribution is a sensible Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Alternative Probabilistic Modeling Approaches 269 choice for the severity distribution because we simulate claim severity as a per- centage (ranging from 0 percent to 100 percent). Third, the model takes into account relationships between the areas by using a copula. It first fits a copula to 30 years of historical payout ratios to estimate the strength of the codependency between areas. In other words, the copula determines whether there is a relationship between payouts occurring in one area and payouts occurring in other areas. We used a t copula, which is suitable in this case because it can accommodate different strengths of relationships between different areas (for example, areas close to each other may be more cor- related than areas that are farther apart).5 This approach of simulating payouts directly has a number of important char- acteristics and includes some important assumptions. As explained above, the model is not actually simulating the weather (such as rainfall), nor is it simulating the weather index (for example, drawing from a distribution of index trigger values). Instead the model directly simulates the uncertainty around the actual payout amounts. An important advantage of this approach is its simplicity and the relative ease of explaining and understanding its results. However, this approach has a number of limitations. In addition to those discussed in detail in chapter 9, the limitations of Approach One also include the following: • The probability of a payout being greater than 0 percent (frequency distribu- tion) is estimated per area (and therefore can vary between areas), but is assumed to be constant over time (that is, not increasing or decreasing over time). For example, if the model estimates the probability of a payout for Area A as 0.06, this value is assumed to have remained the same over time and to stay the same in the future. • The severity distribution is calculated for each area but is based on the histori- cal payout ratios of all areas because only a small number of historical payouts are available per area (that is, a payout does not occur in each year). In other words, the model assumes that the severity of payouts does not vary between areas. However, the model does not take into account that payout ratios in some areas could tend to be much higher than in others. Because it is based on historical data, the severity distribution also does not take into account changes over time. The model does not take into account that payouts may have been increasing (or decreasing) over time, or that payouts could be expected to increase or decrease in the future. • With regard to correlation, the model assumes that the occurrence of payouts is correlated among areas. In other words, if one area has a payout, other nearby areas will likely also have payouts. It is important to note that even though the occurrence of payouts is correlated between areas, the payout ratios are not. • Weather, and therefore the indexes used in a weather-based index insurance product, may go through multiyear cycles of, for example, dry and wet years. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 270 Alternative Probabilistic Modeling Approaches Dry years may be followed by more dry years, and vice versa. Such temporal relationships are not taken into account in the model. The model assumes that any data for the past 30 years are predictive, and more recent data are not more predictive than data from 25 to 30 years ago. In summary, Approach One takes into account some important risks and the relationships between areas, but also makes a number of assumptions that may or may not be valid, depending on the situation. It is important that the analyst, as well as the managers using the results, be aware of these main assumptions and limitations. 16.3 Approach 2: Simulating the Index An alternative to simulating the payout amounts directly is for the model to simulate the index or indexes directly, combine them with the product terms, and then simulate the payout amounts. For example, if the index is the amount of cumulative rainfall over a three- month period, with Approach Two the model first simulates the cumulative rainfall per three-month period. Second, based on the simulated rainfall (for example, 30 millimeters of cumulative rainfall), the model applies the prod- uct terms, which indicate that 30 millimeters of rainfall results in a payout of 20 ­percent of the insured amount. In reality, index insurance products can be based on multiple triggers, such as rainfall, cumulative hours or days of sunshine, temperature, and so forth. Using Approach Two, these indexes can be modeled individually (taking into account relevant correlations) based on historical weather records. It is important to note that this approach accounts for relationships between years. The model can take into account that there may be long- or short-term trends in weather patterns. For example, if last year was especially dry, we might be more inclined to predict that next year will be dry. On the other hand, some areas could have been getting wetter or drier or hotter over the past decade, so we might want to account for that in our simulations of possible future events. ­ Implementation of this approach requires accounting for not only year-to-year correlations in different weather metrics, but also correlations between metrics across different areas. These considerations contribute substantially to the com- plexity of the final model and the time, effort, and data needed to build it. This complexity can also slow down model run time. Some advantages of Approach Two include the following: • Generally, more data are available for modeling the index. When using Approach One to fit the nonzero payout ratio data to distributions, we can only use the years in which payouts would have been made. With Approach Two, we can use all the historical weather data to simulate the index, and then expose it to the index product terms. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Alternative Probabilistic Modeling Approaches 271 • It is easier to take into account that in different areas there are both different frequencies and different severities of payouts. Simulating the actual index for each area can provide a more precise and accurate reflection of the relation- ships between weather patterns across areas than the copulas used to relate payouts in different areas in Approach One. • Temporal relationships (that is, index changes over time) can be reflected more precisely by using a time series approach to forecasting next year’s indexes. This factor can help reflect both gradually changing weather patterns and mul- tiyear cycles in which weather patterns are related between sequential years.6 Although these advantages make Approach Two attractive, keep in mind that with this more comprehensive model comes the need for more data and more assumptions. As the model’s complexity increases, the number of assumptions and variables typically do too. Depending on the available data, the experience of the modeling team, the availability of already existing models, the time frame available for building the model, and other issues, the user must evaluate whether the added complexity is likely to result in a better result in the end. 16.4 Approach 3: Simulating the Weather Approach Three is even more comprehensive and complex than Approach Two. The first step in this approach is to build a more general and comprehensive weather system model of which simulated weather hazard data are the output. Such a model might include multiyear oscillations in weather patterns (for example, El Niño), directional trends in temperature and precipitation, and other large-scale drivers of weather patterns. It would also need to account for correla- tions among areas and weather metrics. Next, the model uses the simulated weather hazard data to determine the simulated trigger values for each area, which are then combined with the product terms to produce the payout ratios. The key here is that the calculation of trigger values and payout ratios depends on simulated hazard data, not on historical hazard data. As one can imagine, while such a model may be more flexible and take into account more weather-related factors, it could be a considerable challenge to develop. Weather models of this type do, however, exist, so they might not need to be built from scratch. These models would still require much effort to learn and to adapt their parameters for use in a probabilistic payout model. Approach Three therefore would be more challenging to pursue. We include it here to illustrate the range of different frameworks that might be used for this index insurance problem. 16.5  Which Model to Use? Every probabilistic (and deterministic) model is by definition a simplification of reality. The key when developing and using probabilistic models for index insur- ance is to build and use models that incorporate correct and valid inputs and Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 272 Alternative Probabilistic Modeling Approaches assumptions that can ideally be supported by empirical data. It is also of utmost importance to clearly communicate to all relevant stakeholders the main assumptions and limitations of the model. In many cases, analysts start off thinking that they need very “realistic” mod- els to capture the behavior of the real world. However, in our experience it is best to start with the simplest model that fulfills all the needed functions and uses valid assumptions. Only then should analysts add more complexity as necessity dictates. Our choice to use Approach One in this guide is in line with our belief that a relatively simple model for which the assumptions and limita- tions can be clearly understood will be more useful than an extremely complex model that is difficult to understand and explain to decision makers and ­consumers alike. Notes 1. By incorporating a time series approach into the modeling, our approach can account for changes over time. This addition can help reflect both gradually changing weather patterns and multiyear cycles in which weather patterns are related between sequen- tial years. However, we have not included this element in the modeling for this guide to reduce complexity and model running time. 2. In modeling losses, a frequency distribution is often used to model the number of loss events. 3. The uncertainty in the actual loss amount (if an event occurs) is also known as second- ary uncertainty. 4. The calculations of these historical payout ratios are described in detail in chapter 11. 5. The model fit a t copula to the historical data with the assumption that the correla- tions between any pair of areas could be unique, as opposed to assuming the same correlation for all pairs of areas. We found this approach more realistic than assuming that all areas were related with the same strength. Goodness-of-fit statistics for the copula actually favored the single correlation approach, but we considered this to be a function of the sparse data for fitting and chose to use the multiple correlation form anyway. This is therefore a case in which our situational understanding and judgment overrode a strictly quantitative method and goodness-of-fit statistic for the copula fitting. 6. With Approach One, some trend parameters can be estimated and included in the model to reflect changes over time. However, given the limited amount of data used in Approach One, such a time series approach will be more challenging. Bibliography Banks, E. 2002. Weather Risk Management: Markets, Products, and Applications. Basingstoke: Palgrave. Jewson, S., A. Brix, and C. Ziehmann. 2007. Weather Derivative Valuation: The Meteorology, Statistics, Financial and Mathematical Foundations. New York: Cambridge University Press. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Chapter 17 Conclusion This guide covers a lot of ground. Part 1 provides a summary of the insights and decisions required for the insurer to make an informed decision to launch and expand an index insurance business line. Part 2 provides a step-by-step guide to calculating the decision metrics that can be used by the insurance manager. One of our main goals for this guide is to support the improvement of named peril index insurance product offerings through structured and transparent collabora- tion and communication between insurers, product design teams, and policy- holders. With this in mind we would like to leave the reader with two reminders of best practice for using the tools in this guide: First, remember that because it provides such a high level of coverage, the Base Index is also very expensive and many policyholders will request a lower price—and lower coverage—product. It is extremely important that the insurer always produce a Base Index to explain to the policyholder the difference between complete coverage—that provided by the Base Index—and the cover- age provided by other product options. Without this explicit comparison, policy- holders often fall into the trap of expecting complete coverage even when they have purchased a lower coverage, less expensive product. Second, when using the models in this guide, as well as when developing or using any probabilistic model, always be critical of the assumptions that are made in the use of the data, the analysis, and the development of the model. The main, simplifying assumption in a model should be well articulated to all stakeholders so that they are aware of the assumptions and can decide whether the model framework needs refining. These two strategies, and the additional guidance provided in this guide, are ­ onsumer critical for practicing responsible finance. They will help insurers meet c protection responsibilities such as providing transparent services and treating policyholders fairly. Failure to implement responsible insurance principles will lead to reputational challenges for the product, the insurer, and the market as a whole. By implementing the tools provided in this guide, insurers can help ensure the healthy, sustainable, and responsible development of index insurance markets. Risk Modeling for Appraising Named Peril Index Insurance Products   273   http://dx.doi.org/10.1596/978-1-4648-1048-0 Glossary probabilistic Actuarial analyst The individual (or team) responsible for performing ­ modeling and generating decision metrics for consideration by the insurance manager. Adverse selection A situation in which sellers have information that buyers do not about some aspect of product quality; in insurance, a situation in which the people who have insurance are more likely to make a claim than the average population used by the insurers to set their rates. Agent An individual or entity that is authorized to represent one or more insur- ance companies to sell insurance. Aggregate distribution A distribution that takes into account the frequency of an event happening, and the severity or impact of the event. Aggregator An entity that accumulates risk exposures of several insured parties within a given geographical area and transfers them to an insurer. Usually acts as the policyholder. Aleatory uncertainty Inherent randomness associated with a future loss or payout; this uncertainty cannot be reduced by the collection of additional data. Also called randomness, variability, stochastic uncertainty, or irreducible uncertainty. Base Index An index structure that is designed to exhibit the highest correlation between payouts and inventory losses caused by the insured peril and hence provide the highest level of coverage possible against damage to the insured inventory. As soon as the proxy’s behavior starts deviating from its normal level (as defined by subject specialists), the Base Index triggers a payment. Basis risk The imperfect correlation between the actual losses suffered by an entity or individual and the payments received from a risk transfer instrument designed to cover these losses. In other words, the risk that index measure- ments of the loss will be different from actual individual losses. Basis risk ratio Basis risk expressed as a percentage of the sum insured. Bernoulli distribution A discrete probability distribution of a random variable that takes values of either 1 or 0. The probability of a 1 is often called the probability of success. Risk Modeling for Appraising Named Peril Index Insurance Products   275   http://dx.doi.org/10.1596/978-1-4648-1048-0 276 Glossary Beta distribution A flexible, continuous, and bounded probability distribution described by two shape parameters. It is commonly used when the range of the random variable is known. Binomial distribution A discrete distribution that is often used to simulate the number of successful outcomes from a certain number of trials in which the probability of success for each of these trials is the same. Burn analysis contract valuation method An actuarial approach to estimating the premium rate based on the historical performance of the contract. Capacity Total limit of liability that a company or the insurance and reinsurance industry can assume, according to generally accepted criteria of solvency. Categorical classification of past damages See qualitative classification of past damages. Claims Payment for losses covered by insurance. Combined ratio A metric of profitability for an insurer that indicates whether an insurer has made an underwriting loss or gain. It is defined as the proportion of claims paid (or payable) plus administrative and operating expenses (A&O) to premiums earned. A combined loss ratio greater than 1 (or 100 percent) indicates that the premiums collected from the insured are not sufficient to pay the claim (indemnity) and cover A&O expenses. In this case, the insurer faces an underwriting loss. Conditional value at risk (CVaR) See tail value at risk. Continuous probability distribution A probability distribution that describes a set of uninterrupted values over a range. In contrast to the discrete probability distribution, the continuous distribution assumes there are an infinite number of possible values. Continuous variable A variable that can take any value within its range. Contract monitoring Process during the risk period whereby the proxy is con- tinuously evaluated against the contract payout schedule. Copula A distribution used to describe (or simulate) the dependence between two or more random variables. From a mathematical perspective, a copula is a multivariate probability distribution for which the marginal probability distribution of each variable is a uniform distribution. Correlation The relationship or interdependence between two or more variables. Cost of capital Return that shareholders require to keep their capital in a certain investment or line of business. See also required return on capital. Covariant risks Risks that are likely to affect many individuals or households at the same time, for example, drought that affects adjacent areas during the same growing season. Cumulative distribution function A function that gives the probability that the random variable X is less than or equal to x, for every value of x. All random variables (discrete and continuous) have a cumulative distribution function. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Glossary 277 Cumulative probability The probability that the random variable X is less than or equal to x. Damage level The amount of damage (expressed as a percentage of the inven- tory value) for a single insured unit. A 30 percent damage level means that 30 percent of the inventory value was damaged during a particular period (for example, one year). Data processing Operations on data, often done with the use of a computer, to retrieve or transform information. Data provider The party responsible for supplying historical and real-time claim settlement data to the parties to an index structure. Debt recovery expense The cost incurred by the lender in its attempt to recover part or all of the outstanding debt by the defaulting party. Deductible The proportion (or amount) of an insured loss that the policyholder agrees to pay or bear before any recovery from the insurer. Default rate The probability per unit of time (often per year) that a borrower will fail to make payments on a loan as required. Default risk The risk that a borrower will not meet contractual obligations, such as interest payments or principal repayment on a loan, when they are due. Dekadal rainfall Rainfall accumulated over the period of the 1st to 10th day, 11th to 20th day, or 21st to final day of a calendar month. The third dekad of a month will be 10 days for a month with 30 days, 11 days for a month with 31 days, and either 8 or 9 days for February. Derived inputs Values used in the model that are arrived at through a structured manipulation of specific input values. Deterministic model A model in which every set of variable states is uniquely determined by parameters in the model and by sets of previous states of these variables; therefore, a deterministic model always performs the same way for a given set of initial conditions. (In contrast, see probabilistic model.) Discrete probability distribution A probability distribution that describes dis- tinct values, usually integers, with no intermediate values. In contrast, a continuous distribution assumes there are an infinite number of possible values. Discrete variable A variable that can take only a distinct value such as 0, 1, 2, 3, 4,... or 0, 1/3, 2/3,.... Distribution fitting The fitting of a probability distribution to data of a random variable. For example, we could fit a continuous distribution to data on annual historical payouts with the goal of forecasting the frequency of occurrence of different magnitudes of payouts. Economic value added A measure of how much value a company (or business unit) creates. It is calculated by subtracting the cost of capital from the operat- ing profits. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 278 Glossary Epistemic uncertainty The lack of knowledge associated with a random variable, for example, future annual payout ratios. Epistemic uncertainty can be reduced by the collection of additional data. Also called statistical uncertainty or parameter uncertainty. Equitable premium rate The premium rate that each area should be charged to accumulate the total premium suggested for the portfolio should the insurer decide to avoid cross-subsidization and charge fair premiums for each area. Exceedance probability The chance or probability per certain period of an event (for example, a rainfall or flooding event) occurring that is equal to or larger than a certain threshold. Exit The threshold amount of the index below or above which the maxi- mum payout will be paid. For example, 50 millimeters of rainfall is the exit for a contract if at 50 millimeters or less of rain the maximum amount of 100 percent pays out. Expected loss The sum of the probabilities of each insured event multiplied by the estimated amount (in currency) of the loss for each of these events. Expected value The anticipated value or outcome of a certain event. In probabil- ity analysis, the expected value can be calculated by multiplying each possible outcome by its respective probability, then summing those values. Expense costs Costs of doing business, such as administration costs, commissions, and other overhead costs, that must be included in the premium to allow the insurer to continue providing insurance services. Expense loading Expense costs expressed as a percentage of gross premium. Exposed units Units that are likely to be affected by the insured perils during the risk period, when included in the portfolio that defines the index. Fair coin A coin with an equal probability of landing heads or tails for each throw (that is, 50 percent probability for each side). Fat-tailed distribution A distribution that belongs to the family of heavy-tailed statistical distributions. Fat-tailed distributions have heavier (fatter) tails than the normal distribution. Financier A person or entity that controls the use and lending of large amounts of money. Frequency The number of times a value recurs in a group interval. Geographical basis risk Basis risk that is caused because of the distance between the insured unit and the measurement location. The measurement location refers to the coordinates where measurements of the proxy are recorded. Goodness of fit A set of mathematical tests performed to find the best fit between a standard probability distribution and a data set. Historical hazard data Data on weather parameters (for example, millimeters of rain during a growing season) or recorded classifications or intensities of natu- ral disasters (for example, Category I and II typhoons). Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Glossary 279 Historical inventory damage data Data on historical damages to inventory suf- fered by the insured party as a result of the insured peril. Historical payouts Calculated data on how much an index insurance product would have paid out in the past based on historical index data such as daily rainfall. Hybrid product (wii + ayii, or wii + indemnity) A product that combines ele- ments from two or more of the following: weather index insurance (wii), area yield index insurance (ayii), and traditional indemnity insurance. Implied deductible The difference in the amount of risk that is covered by the Base Index and the Redesigned Index. Typically the Redesigned Index has lower costs (and therefore lower coverage of the policyholders’ risk); there- fore, the implied deductible is the reduction of risk coverage resulting from the lower premium cost. Indemnity insurance Type of insurance that seeks to compensate the insured party such that the party regains exactly the same financial position as before the occurrence of the loss event. The validity and magnitude of the loss is usu- ally determined by inspection of the damaged inventory by a licensed loss assessor. Independent risk Risks for which there is no relation between the results of one and those of the other(s). In other words, if one independent risk event occurs, the probability or the impact of the other risk occurring does not change because there is no relationship between the risks. Insurance intermediary An individual or entity that acts as either an insurance agent or broker in facilitating risk transfer from an insured party to selected insurers. Insurance manager The staff member of the insurer charged with decision mak- ing regarding the insurer’s index insurance product line. Insurance regulator Government agency responsible for approving the issu- ing of insurance products, monitoring company solvency, and implement- ing consumer protection rules. Insured party The individual or entity that transfers away the unwanted residual risk. The insured party can be an individual, a farmer, or small or medium enterprise, or it can be the same organization that is the policyholder. Insured party basis risk The risk that the Base Index’s measurement of the insured party’s loss will be lower than actual inventory damage. In other words, a case in which the insured unit suffers a loss that is greater than the payment triggered by the Base Index. Insured unit An agreed-on measure of the inventory (for example, input cost for an acre of land) that is the subject of the insurance coverage. Insurer The entity that underwrites the risk; the party legally responsible for the liabilities arising from the insurance policy (up to the limit and minus the deductible). Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 280 Glossary Insurer basis risk The risk that the Base Index’s measurement of the insured party’s loss will be higher than actual inventory damage. In other words, the insured unit suffers a smaller loss than the payment triggered by the Base Index. Inventory damage ratio Damage to inventory as a percentage of the total value of the inventory. For example, an inventory damage ratio of 25 percent for inven- tory worth $100 indicates that damages to the inventory amounted to $25. Iteration Within a Monte Carlo simulation model, an iteration (also often called a trial or simulation) is one calculation of the Monte Carlo model that uses a random sample of each of the probability distributions within the model, result- ing in one possible outcome of the model. An iteration can also be seen as a possible future scenario. With Monte Carlo simulation models, typically at least 10,000 iterations are used to estimate the range of possible outcomes. Law of large numbers As the risk pool increases and losses are independent, the actual loss approaches the expected loss. Liquidity Having sufficient cash or liquid assets to meet day-to-day operating needs. Loss assessment Determination of the extent of damage resulting from the occurrence of an insured peril and the settlement of the claim. Mean One of several measures of the location of a distribution. For a data set, the mean is the arithmetic average of all values. For a probability distri- bution, the mean is the sum of all possible values weighted by their proba- bility. It is also equivalent to the balance point of the distribution. Modal portfolio The most common portfolio size. For example, if looking at rural banks, the modal portfolio would be the dollar value of outstanding or disbursed loans that is common among those entities. Monoline insurance A single class of insurance business. Monte Carlo simulation A computer-based method of analysis developed in the 1940s that uses statistical sampling techniques in obtaining a probabilistic approximation to the solution of a mathematical equation or model. It is a method of calculating the probability of an event using values randomly selected from sets of data, repeating the process many times, and deriving the probability from the distributions of the aggregated data. Moral hazard A condition that increases the likelihood that a person will inten- tionally cause or exaggerate a loss. Also, careless behavior caused by the pres- ence of insurance that increases the expected claims filed by policyholders. Named peril index insurance An index insurance structure that is meant to pro- tect the insured party against the effects of specific perils such as drought, excess rain, or typhoon. Net fund position The financial status of an insurance fund after paying for claims triggered during the season. The fund is made up of premiums accu- mulated from previous seasons, current season premiums, and any other funds that management may decide to allocate to the fund. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Glossary 281 Net premium income Gross premium less expenses. Nonperforming loans Loans that are not up to date with scheduled repayments. Nonproportional reinsurance A type of reinsurance in which an insurer pays a premium to a reinsurer so that the reinsurer will cover all or a proportion of the losses above a certain threshold. Parameter uncertainty See epistemic uncertainty. Parameterization Selection of the parameters and values of those parameters within a model. Parameterization of a probability distribution means selecting the values that describe the distribution. Probability distributions can often be parameterized different ways, for example, a PERT distribution can be described by a minimum, mostly likely, and maximum, but it can also be described by the 10th percentile, 50th percentile, and 90th percentile. Pareto distribution A continuous probability distribution with the longest tail of all probability distributions. The Pareto distribution was originally used to model demographics such as income distributions. Payout level In this guide, the level of payout of an index insurance product, expressed as a percentage of the sum insured. For example, if the payout level over a one-year period was 75 percent and the sum insured was $1,000, then the payout of the policy would be $750. Percentile Values that divide a sample of data into 100 groups containing (as far as possible) equal numbers of observations. For example, 30 percent of the data values lie below the 30th percentile. PERT (project evaluation and review techniques) distribution A continuous and bounded distribution that is often used to model expert opinion. The PERT distribution requires the same three parameters as the triangular distribution: the minimum, most likely, and maximum. Poisson distribution A discrete distribution that models the number of occur- rences of an event in a period t with an expected rate of “lambda” events per period t when the time between successive events follows a Poisson process. Policyholder The party in whose name an insurance policy is issued. Portfolio-priced product A product with a single premium rate across different areas rather than equitable premium rates for each area. This single pre- mium rate must take into account the risk profiles in each of the individual areas, the correlations in risk between all the areas, and the value insured in each area. Prediction interval The estimate of the interval within which future observation of a certain metric or outcome will fall, with a certain probability. For exam- ple, if the 90 percent prediction interval of the profit margin for next year is –5 percent to +25 percent, it is estimated that there is 90 percent certainty that next year’s profit margins will be between –5 percent and +25 percent. Premium rate The price of the index insurance product, typically expressed as a percentage of the sum insured. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 282 Glossary Probabilistic model A system whose output is a distribution of possible values. In contrast to a deterministic model, in a probabilistic (stochastic) model randomness is present, and variable states are not described by unique values, but rather by probability distributions. Probability density chart A graph that shows a probability density function. Probability density function A function that describes the relative likelihood that a random variable will take on different values. The probability density func- tion can be integrated to obtain the probability that a continuous random variable takes a value in a given interval. Probability distribution A list of probabilities or probability densities associ- ated with each of a random variable’s possible values, together with those values. Probability mass function Relates the possible value of a discrete variable to its probability of occurrence. Probability of fund ruin The annual probability that the net funds position at the end of the year (after paying out all claims) will be negative. Probability of negative profit The annual probability that the profit margin on the index insurance policy will be negative. Probability of profit below target profit margin The annual probability that the profit margin on the index insurance policy will be less than a certain profit target. Probable maximum loss Level representing the largest economic loss likely to occur for a given policy or set of policies (portfolio) when a disaster occurs. Product design basis risk Basis risk resulting from the failure of the chosen prox- ies to capture inventory damage caused by the named peril. Product design team A group of specialists charged with the responsibility for designing index insurance structures. Projected loss ratio The loss ratio (and its uncertainty range) that may occur dur- ing a growing season or year. The loss ratio is the proportion of claims paid (or payable) to premiums earned, usually expressed as the total gross claim divided by the total gross premium. A loss ratio greater than 1 (or 100 percent) indicates that the amount of the claim paid by the insurer exceeds the amount of the premiums collected from the insured. Projected losses The possible amount of losses (and its uncertainty range) that may occur during the growing season or year. Projected profit margin The possible profit margins (and its uncertainty range) that may occur during the growing season or year. The profit margin is the amount by which revenue from premiums exceeds expenses and claims. Proportional (or pro rata) reinsurance A type of reinsurance in which premiums and losses are shared by the insurer (cedant) and the reinsurer on a propor- tional basis. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Glossary 283 Prototype A product or concept in a reduced but fully functional form for the purpose of testing its functionality in the real world. The product or concept is not yet scaled up to full commercial scale or introduced to the broad market. Proxy A figure that can be used to represent the value of something else, for example, lack of rainfall can represent inventory damage. Qualitative classification of past damages The classification of historical inven- tory damage in different qualitative categories such as mild, mild-to-medium, medium, medium-to-severe, and severe. Random draw See iteration. Reinsurance Purchase of insurance by an insurance (ceding) company from another insurance (reinsurance) company for the purpose of spreading risk and reducing the loss from a catastrophe. Reinsurer The entity from which an insurance company may buy reinsurance; often described as the insurer of insurer. Required capital The amount of capital needed to cover payouts of an insurance product with a certain confidence. For example, if the required capital at a 95 percent confidence level is $1 million, we can be 95 percent confident that $1 million will be enough to cover all losses that may be triggered over the season or year. Required return on capital Return that shareholders require to keep their capital in a certain investment or line of business. See also cost of capital. Residual risk The risk that remains even if all practical and economical risk man- agement measures have been implemented. Typically refers to the amount of risk transferable to the insurer. Responsible finance or responsible insurance The performance of commercial activities in the financial or insurance sector in accordance with guidelines regarding responsible and positive societal behavior. Retrospective analysis An analysis aimed at evaluating how a given product would have performed in the past if it had been in force. Return period The expected time between two occurrences of a specific magni- tude of loss event; defined as the inverse of the annual probability of the event occurring. For example, a return period of 100 years corresponds to an annual probability of 1 percent. Risk-free rate The rate of return on an investment with zero risk. Frequently, the rate of return on U.S. Treasury notes is used as a proxy for the risk-free rate. Risk management committee A team responsible for implementing the risk man- agement guidelines set by a company’s top management. Often, the risk management committee has access to or is part of the board of directors. Risk mitigation The process of making decisions and implementing measures that will minimize the probability or impacts of adverse effects on an indi- vidual or an entity or organization. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 284 Glossary Risk modeling The process of designing, building, and using a probabilistic model to gain an understanding of the risks and uncertainties of a certain situation or system. Risk period The period of time during which an insurable event as defined in the insurance policy is covered. Risk pooling Aggregation of individual risks for the purpose of managing the consequences of independent risks. Pooling large numbers of homogeneous, independent exposure units can produce an average loss that is close to the expected loss. It provides a statistically accurate prediction of future losses and helps determine premium rates. Scenario A postulated sequence of development of events. Within a Monte Carlo simulation, an iteration or trial is often referred to as a scenario, or when the model concerns the prediction of certain outcomes, a scenario is often referred to as a possible future scenario. Secondary uncertainty Both aleatory variability and epistemic uncertainty. Severity The magnitude of a loss to the inventory. Sharpe ratio The expected return earned in excess of the risk-free ratio per unit of risk, calculated as the return above the risk-free rate divided by its standard deviation. In other words, the Sharpe ratio shows how much more return (above the risk-free rate) can be expected from an investment per unit of risk. The Sharpe ratio allows the returns on investments with different levels of risk to be compared. Solvency The ability of an insurer to meet its financial obligations as those obliga- tions become due, including those obligations resulting from insured losses that may be claimed several years in the future, based on existing policies. Spearman’s rank order correlation coefficient A nonparametric statistic for quantifying the correlation relationship between two variables. Starting fund value The amount of funds available to potentially pay out claims and expenses at the start of offering an index insurance product. Stochastic Events or systems that are unpredictable because of the influence of a random variable. Sum insured The monetary value attached to the inventory that is insured, either expressed per insured unit or for an overall region. Tail The extremes of a probability distribution. Tail value at risk (TVaR) A risk measure that describes the expected value of a variable, given that an event above a certain probability level has occurred, over a certain period. For example, the TVaR 95 percent is the expected value above the 95th percentile of a probability distribution. Consider a total claims distribution for which the TVaR 95 percent is $10 million. In this case, we expect that 1 in every 20 years (1–195 percent]) the total claims will be $10 million. Also called conditional value at risk (CVaR), mean excess loss, or mean shortfall. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Glossary 285 Total funds at risk The total amount available to pay claims triggered during the risk period. Total sum insured (also called total insured value) Value of all the assets or pro- duction covered by the insurance contract. Trigger An event that causes a payout because an index crosses an agreed-on point. For example, rainfall of less than a certain amount could trigger payout of an index insurance policy to policyholders. TVaR of projected losses See tail value at risk and projected losses. Uncertainty See epistemic uncertainty. Underwriting The process of selecting risks to insure and determining in what amounts and on what terms the insurance company will accept the risk. Units of exposure Properties or lives at risk from a specific event (for example, floods, earthquake, drought). Variability See aleatory uncertainty. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 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 dis- consumption, chemical use, green- tances decreased, resulting in reduced paper ­ house gas emissions, and waste. We follow the recommended standards for paper use set by the Green Press Initiative. The majority of our books are printed on Forest Stewardship Council (FSC)–certified paper, with nearly all containing 50–100 percent recycled ­ ­ content. The recycled fiber in our book paper is either unbleached ­ rocessed chlorine-free (PCF), or or bleached using totally chlorine-free (TCF), p enhanced elemental chlorine-free (EECF) processes. More information about the Bank’s environmental philosophy can be found at http://www.worldbank.org/corporateresponsibility. Risk Modeling for Appraising Named Peril Index Insurance Products http://dx.doi.org/10.1596/978-1-4648-1048-0 Named peril index insurance has great potential to address unmet risk management needs for agricultural insurance in developing economies, potentially contributing to increased agricultural sustainability and improved food security. However, the development and appraisal of index insurance business lines is not without challenges. Insurers must rigorously evaluate the quality of the products they offer and take care to ensure that distributors and policyholders understand the benefits and limits of the purchased coverage. Without these important steps to ensure responsible insurance practices, insurers can damage the implementation and potential of index insurance in the market. Risk Modeling for Appraising Named Peril Index Insurance Products: A Guide for Practitioners helps stakeholders in the named peril index insurance industry appraise new and existing products. Part 1 of the guide provides a summary of the insights and decisions required for the insurer to make an informed decision to launch and expand an index insurance business line. Insurance managers are the primary audience for part 1. Part 2 provides a step-by-step guide to calculating the decision metrics used by the insurance manager in part 1. These metrics are calculated using probabilistic modeling that provides insights into risks related to the index insurance product. Actuarial analysts are the primary audience for part 2. In an increasingly competitive insurance market, creative product development and imaginative business strategies are becoming the norm. This guide will help emerging market insurers who seek to stay on the cutting edge to successfully and sustainably penetrate new market segments. ISBN 978-1-4648-1048-0 SKU 211048