November 2018 Georgia Indebtedness_Pov ertyNote_Nov 5 Analysis Based on Integrated Household Survey Natsuko Kiso Nozaki, Alan Fuchs Tarlovsky, and Cesar A. Cancho POVERTY AND EQUITY GP, ECA GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Contents Overview ....................................................................................................................................................... 2 Macroeconomic Evidence ............................................................................................................................. 4 Poverty and Prevalence of Borrowing ........................................................................................................ 10 Indebtedness and Its Impact on Household Wellbeing – Regression Analysis .......................................... 16 Conclusion .................................................................................................................................................. 20 References ................................................................................................................................................... 23 Appendix. .................................................................................................................................................... 26 Appendix 1. Loan-related variables in IHS ............................................................................................ 28 Appendix 2. Interest Rates ...................................................................................................................... 28 Appendix 3. Literature Review, Model Specification, Estimation Strategy and Data............................ 29 Summary Statistics – Selected Variables .................................................................................................... 38 1 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Overview There is considerable public concern about the level of household indebtedness in Georgia. The new regulation expected to come into force on November 1, 2018 addresses this concern by enforcing the responsible credit framework targeting the commercial banks 1 . A recent study by the Finance, Competitiveness & Innovation (FCI) Group named Borrowing by Individuals: Capacity, Risks and Policy Implications, Summary Note also emphasizes the over indebtedness of individual borrowers which -if the issue is generalized and representative at the national level- can be a potential source of vulnerabilities that could trigger macroeconomic financial distress. Without the institutional mechanisms in the event of financial distress, the adverse consequences of over-indebtedness on household welfare as well as the overall macroeconomic implications may be severe for Georgia, compared to more advanced countries. The objective of this note is twofold. First, the note presents micro-level evidence using the nationally representative household survey to understand households’ borrowing pattern s with supporting evidence from perceptions surveys. The high level of indebtedness of households to bank loans, especially among the poor and vulnerable, may harm economically and socially their drive for escaping poverty. Household profiling is based on quantitative measures complemented by analysis using a set of subjective measures represented at the national level. Second, the note examines plausible causal effects of over-indebtedness on household’s welfare. Much of the solid empirical evidence illustrating the causal relationship between financial development and poverty reduction is at the macro-level given the limitations of nonexperimental data. Doubts have been raised about the welfare impact of bank loans at the micro-level. With excessive debt, there is a risk for poor and vulnerable households to be caught in a spiral of debt and high interest rates that could lead them to poverty traps. Taking advantage of the survey instrument that enables to address the issue of selection bias, the note provides preliminary results on the impact of bank credit on well-being at the household level. Findings are indicative of the financial distress illustrated in the report prepared by the FCI Group. The main messages are: 1. Georgia has seen significant increase in households’ bank borrowing, causing public concern about its economic and social impact. Focusing on the formal banking sector, share of borrowing households has almost doubled from 2011 to 2016, with largest increase in the share of poor households. Estimates from the national representative survey show that over 40 percent of all households uses some type of financial services in 2016, with majority borrowing only from formal commercial banks. Moreover, between 2011 and 2016, share of poor households in the bottom quintile increased the most (by 3.2 percentage points) followed by those in the second lowest quintile (1.2 percentage points) among the borrowing households in contrast to a drop in its share among the richest quintile (negative 4.4 percentage points). Macroeconomic indicator also shows that the credit developments in recent years between 2014 and 2017 have been driven by households as opposed to corporate sector. 2. Banks have become increasingly the main source of loan provision for the households as opposed to informal lenders. However, public trust in banks has fallen drastically to its lowest in 2017 since the financial meltdown in 2008 – only 26 percent of the population had trust in 2017, 1 Caucasus Business Week, October 18, 2018. http://cbw.ge/banking/commercial-banks-against-national-bank-new- regulations-to-take-effect-in-november/ . 2 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 corresponding to less than half the share in 2008. 2 This level of trust is also low from international perspective. The declining trend of public trust in banking after the global financial crisis in 2008 is commonly found in many other countries, but they tend to rise or stabilize at around year 2011/2012. Interestingly, in Georgia, that is not the case – the perception toward banks has continued to deteriorate since 2010 with larger share of individuals expressing distrust towards banks. The trend is accentuated when asked about the National Bank in particular – percentage of individuals rating National Bank as “favorable� has dropped drastically from 67 percent in 2011 to 21 percent in 2017, only slightly increasing to 22 percent in 2018.3 International comparison based on Life in Transition Survey (LiTS) also shows that the level of trust towards banks in Georgia is among the countries with relatively high rates of “distrusts� at 48 percent of all population. 3. Indebtedness, as measured by the ratio of unpaid debt to household total income, has no significant impact, and if any, will increase the household’s likelihood of being in poverty. By isolating causality from mere correlation based on more sophisticated econometric methodology compared to the naïve OLS estimates, we show that: (1) increasing bank loans do not increase the household welfare in terms of per capita consumption, (2) higher indebtedness, measured either by the ratio of borrowing amount or unpaid debt over total household income, have negative (but insignificant) impact on household’s per capita consumption, and that (3) we cannot reject the hypothesis that the higher indebtedness increases the household’s likelihood of being in poverty or in vulnerable sta tus. Results confirm that descriptive statistics and naïve OLS estimates seem to be biased and overestimate the impact of borrowing from banks. 4. Given the dramatic increase in household debt and indication of increasing debt stress, there is an urgent need to gather basic facts from the demand and supply side of the financial market. Only with systematic observations of credit market and dynamics we would be able to reach concrete policy implications tailored to Georgian context. Efforts are needed to validate the magnitude of over- indebtedness and irresponsible lending at the national level. International comparison and variation of financial development within Georgia would also be essential in designing regulatory and policy interventions without being overly restrictive. This paper is structured as follows. Section 2 provides macroeconomic indicators and findings from perception survey as the background evidence. Section 3 illustrates the prevalence of borrowing among the households and identifies type of households that borrow from different sources. Section 4 shows results from the causal impact analysis of bank loans on household welfare. Section 5 concludes with directions for future research. 2 2012 Edelman Trust Barometer: U.S. Financial Services and Banking Industries. https://www.slideshare.net/EdelmanInsights/2012-edelman-trust-barometer-us-financial-services-and-banking- industries . 3 Opinion poll conducted by Baltic Surveys/The Gallup Organization (2018). 3 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Macroeconomic Evidence There is considerable public concern about household indebtedness in Georgia. In Georgia, it is estimated that between 3 – 5 percent of households could have moved below poverty line due to financial conditions.4 Taking on debt can increase consumption beyond what one’s income can support, it can smooth consumption in face of shocks and it can represent an investment in the future. However, over indebtedness may result in significant financial distress, ultimately capturing households in poverty traps. Indebtedness may thus signal irresponsible spending, a lack of self-control, or low level of financial literacy. To address increasing household indebtedness, the National Bank of Georgia (NBG) has established a cap on loans to households without verifiable income (25 percent of banks’ regulatory capital), awaiting upcoming legislation to promote responsible lending.5 The level of household debt in Georgia has been rising steadily over the years until 2016 and declined slightly in 2017. According to the World Development Indicators, credit to households and other sectors reached 62.05 percent of GDP in 2016 compared to 35.52 percent of GDP in 2011. More specifically, IMF reports that household debt had reached 34 percent of GDP at end-2017, which had doubled in the last five years.6 The rate for Georgia is still significantly low compared to Euro Area estimates 7 and close to the average of all ECA countries excluding high income, but higher than countries such as Albania, Armenia and Azerbaijan (Figure 1). Figure 1: Trend in Loans by Households and Other Sectors* – Cross Country Comparison 4 June 5, 2018 REZONANSI: “MORE 5% OF POPULATION BECAME IMPOVERISHED!� 5 Programs are underway to enhance financial education and sensitive households on risks associated with financial imprudence, over-indebtedness, and FX borrowing (IMF Georgia Article IV, June 2018). 6 Ibid. 7 Ibid. 4 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Credit to Households and Other Sectors* (as % of GDP) 80 Georgia, 58.84 70 Turkey Europe & Central Asia 62.05 Rus sian Federation (excl uding high income) 60 54.77 49.17 Armenia 50 43.56 % of GDP 40 38.17 Al ba nia 35.52 Azerba ijan 30 20 10 0 2011 2012 2013 2014 2015 2016 2017 Source: World Development Indicators (as of September 14, 2018). Note: * Includes gross credit from the financial system to households, nonprofit institutions serving households, nonfinancial corporations, state and local governments, and social security funds. Credit growth is driven by households (Figure 2), and its magnitude is proven empirically to be an important factor for economic growth and poverty reduction. A study shows that the relation between financial depth (as defined as private credit as a share of GDP) and poverty is not only causal and statistically significant but also sizeable. Even after controlling for other variables, almost 30 percent of the cross-country variation in changing poverty rates can be attributed to cross-country variation in financial development.8 Although the level of household debt and the size of non-performing loans (NPL) in Georgia are still at the reasonable level compared to its peers and developed countries (Figure 3), the size and stock of household debt may trigger concerns over financial distress in the medium to long terms. Figure 2: Share of Household Debt Figure 3: Household Debt – Cross Country Comparison 8 Interesting example The World Bank, 2008. 5 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Households Outstanding Loans with Commercial Banks and Bank Non-Performing Loans to Total Gross Loans (2016) 45 35 40 30 35 25 30 25.96 % Total Gross Loans 25 20 % of GDP 20 15 15 10 10 3.45 5 5 0 0 BLR BIH IRL UVK ISL TUR AZE MDA UKR LVA ALB LTU SRB FRA JPN CZE PER MNE HRV MEX BGR HUN POL KAZ USA RUS DEU SVN SVK EST GEO ROU MKD ARM Outstanding Loans with Commercial Banks (HHs) as % of GDP Bank nonperforming loans to total gross loans (%) Source: IMF Article IV, June 2018. Source: Non- performing loans from World Development Indicators (as of September 14, 2018) and households outstanding loans from Financial Access Survey, IMF (as of September 16, 2018). http://data.imf.org/?sk=E5DCAB7E-A5CA-4892-A6EA-598B5463A34C . Banks had become increasingly the main source of loan provision as against informal lenders. However, trust in banks has fallen drastically to its lowest in 2017 since the financial meltdown in 2008 – only 26 percent of the population had trust in 2017, corresponding to less than half the share in 2008 (Error! Reference source not found., left). The declining trend of public trust in banking after the global financial crisis in 2008 is commonly found in many other countries, but they tend to rise or stabilize at around year 2011/2012.9 Interestingly, in Georgia, that is not the case – the perception toward banks has continued to deteriorate since 2010 with larger share of individuals expressing distrust towards banks. The trend is accentuated when asked about the National Bank in particular. Opinion poll conducted by Baltic Surveys/The Gallup Organization in 201810 shows that the National Bank was the least trusted institutions with highest share of individuals rating “unfavorable� (67 percent), second only to Political Parties (68 percent “unfavorable�). Figure 4: Poor Public Perception of Banks in Georgia 9 2012 Edelman Trust Barometer: U.S. Financial Services and Banking Industries. https://www.slideshare.net/EdelmanInsights/2012-edelman-trust-barometer-us-financial-services-and-banking- industries . 10 Public Opinion Survey: Residents of Georgia (April 10-22, 2018). The sample consists of 1500 residents of Georgia, representative of the general population by age, gender, region and size/type of settlement. For details, see http://www.iri.org/sites/default/files/2018-5-29_georgia_poll_presentation.pdf . 6 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Source: The Caucasus Research Resource Centers. Caucasus Barometer, 2008 – 2017 Georgia. Retrieved through ODA - http://caucasusbarometer.org on October 24, 2018 (left) and Baltic Surveys/The Gallup Organization, 2018 (right). Trust in banks and financial system in Georgia is also low by international standard. The Life in Transition Survey (LiTS III) allows international comparison on the level of trust towards banks and the financial system. Trust varies significantly across regions, and Georgia is among the countries with relatively high rates of “distrusts� (48 percent). Figure 5: International Comparison of Perception Towards Bank – Selected Countries Trust - banks and the financial system 100% 90% 80% 70% 21.11 % of Population 60% 28.7 50% 5.03 23.76 35.54 40% 22.45 12.98 33.6 17.11 28.87 20.55 30% 16.86 20.2 11.68 48.13 30.45 12.72 11.14 17.55 42.77 20% 21.79 25.01 20.91 27.32 11.01 16.8 14.74 22.89 6.48 8.65 16.96 19.38 16.54 14.37 23.93 10% 16.54 5.62 10.66 4.89 7.3 0% Not applicable Don't know Complete distrust Some distrust Neither trust nor distrust Some trust Complete trust Source: Author’s estimation using LiTS III (2014). One of the causes for poor public perception of commercial banks may be the lack of debt relief policy measures such as debt counselling, restructuring and personal insolvency framework as addressed by FCI. FCI’s Individual Indebtedness Survey (IIS) reveals severe debt pressures among households with over-indebtedness. Given the choice-based sampling frame adopted by the IIS, the sample does not provide national representation of borrowing households in Georgia. Yet, IIS is a valuable source of information 7 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 that can help assess the type and degree of financial distress, households’ tendency for over-indebtedness and its implication to debt traps.11 Excess indebtedness is a legitimate concern given its potential economic and social impact. However, it is also important to assess the magnitude of the problem by assessing households’ borrowing behavior and prevalence of indebtedness at the national level. If over-indebtedness associated with severe debt pressure is truly widespread across nation, then establishing debt resolution processes may be one of the urgent policy measures to maintain stable financial system. This note addresses this concern by using nationally representative household survey to examine the prevalence of borrowing and how it varies with observed characteristics at the national level. The note also tries to examine the causal impact of indebtedness on household welfare by addressing issues of endogeneity. Box 1: Survey Overview and Potential Bias of the Estimates This note reveals that the IHS-based estimates differ substantially from the ones from the Individual Indebtedness Survey. Among others, the difference comes from sample design, sample size, unit of collection, and the objectives in conducting the surveys which is described briefly below. Data Description of Georgia IHS The data used for the analysis is the series of Georgia Integrated Household Survey (IHS) from 2011 to 2016 collected by the National Statistics Office of Georgia (Geostat), unless otherwise noted. The IHS is a nationally representative household survey, whose stratification is based on 2002 census. It collects information on household and individual’s socio demographic characteristics, as well as consumption using a 7 -day diary, expenditures in the last three months, and income from labor, social assistance, private transfers, and agricultural activities. It’s major focus is to allow for distributional analysis on multiple topics based on income, consumption and wealth. The survey estimates are made representative not only at the national level but also at the regional level, as well as for urban and rural areas. Because regions with small number of population (Racha-Lechkhumi and Kvemo Svaneti) were joined to an adjacent region, and two regions not under the control of the central government of Georgia were omitted (Tskhinvali and Abkhazia AR), households are divided into 10 regions as specified in the main report. The sample is composed of roughly 11,000 observations per year comprising around 3000 households interviewed four times throughout the year (one per quarter) to correct for seasonal bias. Households are replaced by another randomly selected households from the same cluster after one cycle (household rotation). The survey is structured as a rotating panel where households are visited in four consecutive quarters. Attrition rates are available from the Geostat and in general, they range in levels acceptable for this type of surveys. Drawbacks in the IHS sample design are the ones common to most household surveys. Most importantly, although sample households are representative geographically based on stratified sampling, they are not necessarily representative of households’ financial characteristics, which is the focus of this study. Ideally, if sufficient information were available, the sample would use a design that minimized the expected sampling error for a weighted combination of financial variables, where weights may reflect the relative importance of the variables of interest.12 It is unclear whether the sample households overrepresents or underrepresents the households’ borrowing behavior and financial position, as its comparison with national account shows mixed results ( Table 1). Data Description of Georgia Individual Indebtedness Survey 11 A. Prigozhina, et al., 2018. 12 Kennickell and McManus, 1993. 8 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Individual Indebtedness Survey (IIS) used for the note, “Borrowing by Individuals: Capacity, Risks and Policy Implication, Summary Note13� used choice-based sampling frame and are collected at the much smaller scale. It is focused on individual’s borrowing behavior and has advantage in allowing in-depth analysis on capacity of individual borrowers to manage their debt repayments and the characteristics of households with and without debt by type of loans. About 4000 residents throughout Georgia were interviewed during October 2017 – January 2018 by Caucasus Research Resource Center (CRRC) under the World Bank Financial Deepening and Inclusion Project. Out of 4000 residents, 3500 had current outstanding loans and about 500 had no current borrowing. Micro-level data on financial access and usage is limited and only few surveys focus on this topic. This survey is thus an important effort in improving our understanding of households’ indebtedness. These are the only way to get detailed information on who uses which financial services from which types of institutions, including informal ones. However, major concern of using the survey is the possible bias introduced through choice-based sampling and limited sample size. Samples were formed conditional on four choices: (1) currently have at least one loan from a commercial bank but have no current loans from other financial sources; (2) individuals who currently have at least one loan from any non-bank source in addition to commercial banks; (3) currently have at least one loan from a non-bank but have no current loans from banks; and (4) individuals who currently have no loans. This entails over- sampling of households with loans and the distribution of these three types of borrowers in the population is unknown. Without correcting for weights that is validated against administrative data, the sample is likely to over- represent certain types of borrowers. 14 It is important to note that both studies suffer from their own limitations – over-representatives of certain types of borrowers in case of Indebtedness Survey, and potential under-representativeness of borrowers in case of IHS since typical surveys fail to capture the subtle distributional properties at the very top of the distribution. However, given that the sampling frame of IHS is based on census to assure representativeness at three levels (national, regional and urban/rural) and designed to correct for seasonal bias with equal number of observations for each quarter throughout a year, estimates based on IHS is expected to be more reliable with bias smaller in magnitude. Consistency between National Account and IHS Estimates The comparison of survey data with data derived from administrative sources is a familiar approach in the scientific literature. Table 1 shows the ratio of households’ consumption, income, and amount borrowed from banks reported in IHS against the data reported in national accounts (available from National Bank of Georgia and Geostat, as of September 13, 2018). Table 1: Comparison of Estimates by Data Source (2016) Data Source Intergrated Household Ratio (in millions, Lari) Survey (IHS) External Source (=Survey/External) Consumption* (in millions, Lari) 8710.0 21272.3 0.409 Income* (in millions, Lari) 11418.7 32340.8 0.353 Loan from commercial banks* (in millions, Lari) 555.7 2385.0 0.233 Number of Borrowing Households/Individuals from Commercial Banks** (in millions) 0.2 1.4 0.116 13 Finance, Competitiveness & Innovation, The World Bank, 2018. 14 For details, see Methodological Report by Caucasus Research Resource Center, 2017. 9 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 *External Sources for consumption, income and loan from commercial banks are reported figures from National Bank of Georgia and Geostat, as of September 13, 2018. **External Source for number of borrowers is Financial Access Survey (FAS), IMF (as of September 16, 2018). http://data.imf.org/?sk=E5DCAB7E-A5CA-4892-A6EA-598B5463A34C. IHS reports number of households while FAS refers to number of individuals. The level of discrepancies between figures from survey and external source in consumption and income are not surprising and common in other countries. For example, in Armenia, household expenditures in the survey accounted for 37 percent of that in the national accounts and income around 40 percent 15. Table also shows that IHS captures 23.3 of the households’ loan from commercial banks against the loan amount reported in external source. This ratio (i.e., total from household survey against total from external source) is lower than the ratio for consumption and income. Larger downward bias in loan amount, compared to those in consumption and income, is somewhat expected as household survey often fails to capture households at the top end of the wealth distribution and positive correlation of loan amount and household wealth is anticipated in the population.16 Number of borrowers captured in the survey is also low compared to that reported in external source (0.12). The downward bias may be due to the difference in unit of analysis (being household in the survey and individual in the external source), or non-observation bias (due to omission of wealthy households as mentioned above), or mis- or under-reporting of borrowing behavior. However, IHS estimates on prevalence of borrowing do resonate with those in the Caucasus Barometer, which is representative of all population of ages 18 and over. The survey is collected annually about socio-economic issues and political attitudes in the three South Caucasus countries: Armenia, Azerbaijan and Georgia. The project started in 2004 and data is available since 2008 by the Caucasus Research Resource Centers (CRRC) (Annex). Source: Inchauste and Lustig, eds., 2017. Note: “NA� refers to national account. Final consumption expenditures of households include expenditures for purchasing consumer goods and services and also other consumption of goods and services in kind, produced for own use (available by quarter and annual). National account on “commercial bank loans (excluding interbank loans) to households by loan purpose� includes other items such as “business loans for large enterprises,� “business loans for SME,� “lombard loans,� and “other loans.� Poverty and Prevalence of Borrowing Financial development has a pronounced impact on changes in relative and absolute poverty with disproportionate impact on the poor.17 But much remains to be learned about the channels through which financial development affects income inequality and poverty reduction. Cross country studies show that greater financial development induces the incomes of the poor to grow faster than average per capita GDP growth, which lowers income inequality.18 This impact may come from direct access of the poor to credit or indirectly through better financial access for nonpoor entrepreneurial households. 19 Relative importance of these channels on growth and poverty reduction may differ by country and needs more in-depth research at the household level to derive effective policy implications. 15 Inchauste and Lustig, eds., 2017. 16 A study on savings behavior in Austria reports that underestimation of deposits due to undercoverage of most affluent tail of the distribution can be relatively minor (Andreasch and Lindner, 2014). 17 Beck, et al., 2007. 18 Ibid. 19 The World Bank, 2009. 10 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Analysis of financial access and indebtedness at the household level has been scarce. Most of the empirical evidence has been at the country level. Having established the importance of financial development at the macro-level, the next task is to go beyond the national level and focus on the level of households and firms. One of the focuses of this note is to explore whether there is a risk for household well-being to be worsened through increased debt burden from bank loans. This question is legitimate as debt may be viewed as a welfare enhancing mechanism as well as potential channel to poverty trap when used imprudently and excessively without institutional mechanisms for households to deal with debt distress. This section illustrates the pattern and levels of financial exposure of poor and non-poor households in Georgia to formal and informal credits. This will contribute to the literature by providing evidence of household indebtedness in Georgia at the micro level. By using the IHS, a nationally representative survey, the section highlights the trend and extent to which households rely on different financial sources. Over 40 percent of all households uses some type of financial services and the share has been increasing over time for the poor and the non-poor (Figure 6), figure consistent with the estimates from Caucasus Barometer (Appendix). These are households that either borrowed and/or repaid back to the financial organizations within the past 3 months of the interview.20 Survey identifies two sources of loans – (1) banks or other financial organizations, and (2) private persons. Without further details, (2) private persons can include any informal sources, such as professional moneylenders, pawnbrokers, tradespeople, and associations of acquaintances. Following analysis would thus interpret (1) as formal banking sector and (2) as informal credit institutions. Figure 6: Prevalence of Borrowing – 2011 – 2016 Trend 20 Households that had borrowed and/or repaid back only to banks are categorized as “bank only� borrowers. Similarly for “private only� borrowers. 11 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Share of Households Borrowing/Repaying with Share of Credit Source Any Type of Credit Source 50% 50% 5.34% 5.55% 45.85% 40% 5.50% 6.19% 5.90% 6.03% 8.37% 5.59% 5.64% 7.29% 5.77% 41.85% 6.90% % of Households 40% 4.43% 8.74% 30% 3.60% 9.69% 11.51% 36.43% 12.28% 11.22% 16.00% % of Househol ds 12.65% 18.89% 15.73% 30% 33.07% 20% 18.50% 32.25% 33.01% 26.50% 29.09% 20% 22.56% 23.77% 24.57% 10% 19.35% 19.64% 15.02% 17.12% 10.97% 10% 0% Non Poor Non Poor Non Poor Non Poor Non Poor Non Poor Poor Poor Poor Poor Poor Poor 0% 2011 2012 2013 2014 2015 2016 2011 2012 2013 2014 2015 2016 Non Poor Poor Bank Only Private Only Both Source: Author’s calculation using Georgia IHS. Note: Poor households are defined using per adult equivalent consumption aggregates and national poverty lines (125.9 and 137.13 GEL for years 2011 and 2016 respectively). “Bank Only� refers to households that borrowed/repaid only to formal banks, and “Private Only� refers to those that only borrowed/repaid to private source. Commercial banks had become the major source of credit over time for both poor and non-poor (Figure 7). The share of poor households borrowing from banks had increased significantly over time, reversing the relative importance of formal vs. informal source since 2011. This is true for the households in the richest quintile as well as for those in the poorest quintile. Figure 7: Prevalence of Borrowing, by Quintile - Trend Share of Households that Borrowed, by Quintile 50% 40% % of Households 30% 20% 10% 0% 2011 2016 2011 2016 2011 2016 Any Source Bank Only Private Only Poorest quintile Richest quintile Source: Author’s calculation using Georgia IHS. 12 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 While formal and informal finance coexists, they are used as substitutes and not as compliments by households. Interestingly, share of households that borrow from both sources is small (Figure 6).21 This is understandable if credit contracts differ substantially between these two sectors and thus there is only very limited inter-sector competition. Greater importance of informal private sources among the poor households and vice versa among the non-poor reflects typical market failure stemming from imperfect information, moral hazard as well as lack of collateral to prevent moral hazard.22 Focusing on the formal banking sector, the share of poor and vulnerable households in the bottom two quintiles increased the most, with increase driven by the growing share of borrowers in the bottom quintile. Poor are not over-represented among the bank borrowers (Appendix), but the increase in the share had been the highest among the households in the bottom quintile (3.18 percent points) followed by those in the second lowest quintile (1.18 pp) in contrast to the drop in its share among the richest quintile (negative 4.37 pp). Figure 8: Distribution of Households among Borrowers by Source Poverty Quintile 100% 100% 19.03 14.08 19.14 16.98 25.8 33.38 29.01 37.76 80% 80% % of Households % of Households 60% 60% 40% 20.22 19.98 40% 80.97 85.92 74.2 20% 16.33 62.24 15.15 23.2 22.93 9.41 12.59 20% 0% 2011 2016 2011 2016 0% Bank Borrowers Private Borrowers 2011 2016 2011 2016 Quintile Lowest 1 Quintile 2 Bank Borrowers Private Borrowers Quintile 3 Quintile 4 Non Poor Poor Quintile Highest 5 Source: Author’s calculation using Georgia IHS. Note: Poor households are defined using per adult equivalent consumption aggregates and national poverty lines (125.9 and 137.13 GEL for years 2011 and 2016 respectively). “Bank Borrowers� refers to households that borrowed only from banks, and “Private Borrowers� refers to those that only borrowed from private source. Borrowers are unevenly distributed throughout the regions, with largest share of borrowers in Tbilisi in the formal credit market. The role of informal finance is diminishing and continues to serve rural households. Figure 9 shows that borrowers are unevenly distributed throughout the regions. Most concentrated region is Tbilisi followed by Imeriti for the formal banking, where the share had remained stable over time. Although smaller in magnitude, the share of Samegreb has increased. These two regions were identified as location with large growth potential in tourism, industry, and trade, and the World Bank also supports multiple regional development projects.23 21 There is a huge variation in the pattern of borrowing across countries. Share of borrowers getting credit from both formal and informal sources varies from 70 percent in India (Das-Gupta et al., 1989) to 13 percent in rural Thailand (Giné, 2011) 22 Literature suggests that formal and informal finance coexist in markets with weak legal institutions and low levels of income (Madestam, 2014). 23 The World Bank, 2015. 13 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Figure 9: Regional Distribution Regional Distribution of Borrowers 100% Urban/Rural 100 90% 18.09 18.31 19.08 Imereti, 26.84 80% 10.18 13.51 80 40.8 43.56 70% 11.3 Samegrelo, 5.65 62.31 % of Households 64.29 60% 2.75 2.49 60 5.89 50% Samtskhe-Javakheti, 14.07 31.87 29.24 40% 40 30% 16.56 21.13 23.11 20% 27.33 27.2 Tbilisi, 12.6 20 27.33 27.2 10% 16.56 12.6 0 0% 2011 2016 2011 2016 2011 2016 2011 2016 Bank Borrowers Private Borrowers Bank Borrowers Private Kakheti Tbilisi Shida Kartli Kvemo Kartli Borrowers Samtskhe-Javakheti Ajara Guria Samegrelo Tbilisi Rest Urban Rural Imereti Mtskheta-Mtianeti Source: Author’s calculation using Georgia IHS. Note: “Bank Borrowers� refers to households that borrowed only from banks, and “Private Borrowers� refers to those that only borrowed from private source. Lower regional borrowing rate is uncorrelated with regional poverty rate. Instead, for the formal sector lending, there is a positive correlation between drop in poverty rate and increase in borrowing rate between 2011 and 2016 (Figure 10). Negative correlation is observed for the informal banking. From the supply side of credit, this indicates the strategic placement decision of formal banks based on market potential and profitability. From the demand side, households seem to increasingly switch borrowing channels from informal to formal source. Figure 10: Correlation of Decline in Poverty Rates in Increase in Borrowing Rates 14 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Changes in Poverty and Borrowing Rates, Bank Only Changes in Poverty and Borrowing Rates, Private Only 30% (2011 - 2016) (2011 - 2016) 20% 25% Samtskhe-Javakheti Increase in Borrowing Rate (% of HHs) Sa megrelo Increase in Borrowing Rate (% of HHs) 10% Imereti y = -0.7555x + 0.0127 20% Shi da Kartli y = 0.4802x + 0.0789 Tbi lisi R² = 0.3069 Ajara R² = 0.201 0% 15% Imereti Kvemo Ka rtli Guri a -10% Tbilisi 10% Aja ra Ka kheti Kvemo Kartli Shida Kartli Sa mts khe-Javakheti Mtskheta-Mtianeti -20% Samegrelo 5% Guria Mts kheta-Mtianeti Kakheti -30% 0% 0% 5% 10% 15% 20% 25% 0% 5% 10% 15% 20% 25% Reduction in Poverty Rates (% of pop) Reduction in Poverty Rates (% of pop) d_borrowed_bankonly Linear (d_borrowed_bankonly) d_borrowed_privonly Linear (d_borrowed_privonly) Source: Author’s calculation using Georgia IHS. Note: Size of the bubbles reflect the relative size of the population across regions. Borrowing also varies by household type and its pattern remains constant over time with huge parallel shift – upward shift for formal banking and downward shift for informal banking. In addition to geographic variation, Figure 11 illustrates the borrowing rates by household type. Interestingly, the patterns have shifted parallelly between 2011 and 2016 – households of all type increased borrowing from formal banks and ceased from informal credits. For formal credit, higher rates are visible among the following groups: larger households; households with educated heads; households whose heads are married/living together; multiple member households; and families with children. Young households – characterized either by having young head or with lower share of elderlies within a household – are groups associated with higher borrowing rates. These are mostly not surprising and can be explained by need for consumption smoothing in face of shocks or need for investment in human capital. Households with low educated head may be one of the groups excluded from the formal credit market, while elderlies may be associated with lower demand for credit (due to universal coverage and reasonable generosity of old-age pension). Figure 11: Share of Borrowing Households by Demographic Type (2011 and 2016) 15 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Prevalence of Borrowing - Bank Only Prevalence of Borrowing - Private Only 50% 30% 40% % of Households % of Households 30% 20% 20% 10% 10% 0% 0% HH with Children & Older Head Young Head (15<=age<=29) None/=30) Self Employed Living together Employee Married Single Member HH HH without Children (0-15) Widow/er Special Sec Multiple Member HH Unemployed Retired HH with Children (0-15) Female Head BELOW national avg Large HH Size (>=6) OVER national avg HHsize <=5 Young Head (15<=age<=29) None/=30) Employee HH with Children & Young Head Married Single Member HH HH without Children (0-15) Widow/er Special Sec Retired Multiple Member HH Unemployed Female Head BELOW national avg Large HH Size (>=6) HH with Children (0-15) OVER national avg HHsize <=5 HH size Gender Education of LF status of Head Marital Status of Age of Share of Single Children HH size Gender Education of Head LF status of Head Marital Status of Age of Share of Single Children Head Head Head 66+ and Age of Head Head 66+ and Age of within Head within Head HH HH Bank Only 2011 Bank Only 2016 Private Only 2011 Private Only 2016 Source: Author’s calculation using Georgia IHS. Note: Differences are significant at 10 % significance level or lower for all the categories among bank only borrowers. Differences are significant for selected classification for informal borrowers (share of 66+, age of head, single/multiple). Indebtedness and Its Impact on Household Wellbeing – Regression Analysis The objective of this section is to estimate the causal relationship between bank borrowing and household’s welfare. There is a growing concern over households’ indebtedness and its effect on household welfare in Georgia. Taking on debt can increase consumption beyond what one’s income can support, it can smooth consumption in face of shocks and it can represent an investment in the future. However, over indebtedness may result in significant financial distress, forcing households to be caught in poverty trap.24 By drawing on lessons from the empirical literature on microcredit, the note tries to estimate the causal impact of bank loans on household welfare. Policy implications – whether and how Government should promote or repress financial intermediation – will be discussed at the end. Credible evidence on whether bank loans can reduce poverty remains limited. The main reason for this is the nonrandom nature of the borrowing practice. From the demand side, there is a concern for self-selection bias which comes from unobserved household attributes (such as endowments of entrepreneurial ability, innate health, and productivity). If household’s decision to borrowing is based on unobservable attributes that simultaneously affect outcome, then estimates of the effect of bank loans will be biased. Market imperfections – such as moral hazard and adverse selection that arise from serious information asymmetries and enforcement problems – may lead to an unequal distribution of credit in favor of the wealthy households. There is also an endogeneity with respect to bank’s spatial distribution, or, placement bias, from the supply side. Banks are expected to make strategic placement decisions based on specific features of markets depending on their motivation – either areas with vibrant market potential for profitability or relatively poor areas because of social concerns. Selection bias can go in either direction. Drawing from the literature on microcredit and project evaluation, this paper uses interest rate averaged over sample households in each location-year-season group as instrumental variables (IV) 24 Implications of indebtedness are described, for example, in Mannah-Blankson 16 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 to address the classis issues of endogeneity when using nonexperimental data to evaluate the effect of bank loans on outcomes such as household welfare. To measure the effect of bank loan on household welfare, we estimate a restricted welfare equation that conditions household’s per c apita consumption welfare on the household’s decision to take loans from the bank. Taking up the loan cannot be treated as exogenous because households that apply and succeeded in obtaining loans may systematically differ from those that do not apply for or applied but denied bank loans. Thus, the model comprises two stages in which IV is used to estimate the first stage in modelling the decision to take the bank loan. The price of bank loans – the average interest rate of the area in which household reside in specific quarter in a given year – is used as an identifying instrument. By taking the average of the reported interest rates by the borrowing households within each group, we can partial out the portion of interest rate that may be correlated with household’s attributes known to lenders but unknown to researchers and treat it as exogenous to the wellbeing of households. Details of the model and estimation strategy as well as literature review on the methodologies are described in the Appendix. Focusing on the formal banks, share of borrowing households has almost doubled from 2011 to 2016. Table 1 presents the percentage of households that had borrowed from formal banks and the average per capita consumption aggregate and its logarithm. As shown earlier, percentage of households taking bank loans has increased steadily over the years by 2 – 3 percentage points from 2011 to 2015 slowing down to less than 1 percentage points from 2015 to 2016. Table 2: Weighted Mean and Standard Error of Per Capita Consumption Aggregates Mean Log Per % HH Mean Per Cap Standard Cap Standard Year Household Type # Obs Borrowing Consumption Error Consumption Error Non Borrower 9395 2321.97 23.38 7.46 0.008 2011 16.9 Borrower 1811 3027.56 227.30 7.69 0.016 Non Borrower 9070 2513.48 25.26 7.55 0.008 2012 20.64 Borrower 2195 2832.21 49.74 7.70 0.015 Non Borrower 8589 2707.78 26.15 7.65 0.008 2013 24.38 Borrower 2502 3422.97 65.09 7.87 0.014 Non Borrower 8315 2854.79 28.53 7.71 0.008 2014 27.24 Borrower 2843 3403.80 62.40 7.89 0.013 Non Borrower 7844 2898.92 29.75 7.73 0.008 2015 30.72 Borrower 3155 3103.23 42.02 7.82 0.012 Non Borrower 7639 2908.77 28.03 7.73 0.008 2016 31.49 Borrower 3219 3166.14 47.34 7.84 0.012 Source: Author’s calculation using Georgia IHS. Note: All values are weighted except for number of observations in the third column. Descriptive statistics show that borrowing households consistently have higher per capita consumption than non-borrowing households (Table 2). Table 2 provides some indication for the relationship between household wealth and bank borrowing - average per capita consumption and its logarithm are higher for borrowers across all years considered. The gaps are all statistically significant, with null hypothesis that these mean differences are equal is rejected at the 0.00 significance level. However, there are many possible factors generating these gaps. For example, borrowers are better off than non- borrowers because banks strategically select wealthy households; households that chose to borrow may also be different from those that chose not to borrow in their attributes including their entrepreneurial abilities and prospects for the future. Combinations of demand and supply side factors are at play. To disentangle causation from correlation, we turn to regression analysis addressing selection biases from both demand and supply sides. 17 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Empirical Results Estimates show that there is no impact of bank loans on household’s well-being. Furthermore, size of debt has negative impact on household’s wellbeing, if any. First, we estimate the impact of bank borrowing on logarithm of per capita consumption.25 The first two columns in Table 3 report coefficients from the OLS regressions controlling for household attributes as well as area-, seasonal- and year-specific unobservables. Specification [2] also includes proximate of household’s cognitive ability expected to capture attributes such as entrepreneurial ability, self-confidence, and aspirations for the future as an attempt to minimize the selection bias. The naïve OLS estimates show that households with bank loans consume 12.5 percent more than households without the loan (specification [1]) and 9.3 percent more when controlling for the household’s cognitive skills (specification [2]). However, once we take into account the selection bias by using IV methodology, the impact disappears – columns [3] and [4] show that the coefficient becomes highly insignificant.26 Specification [4] includes debt level as regressors as well, which indicates that the magnitude of debt relative to household income may have negative effect on household consumption, although they are statistically insignificant. Table 3: Model Results, Estimates of the Effect of Bank Loans on Log (Per Capita Consumption Aggregate) [1] [2] [3] [4] OLS OLS IV IV Added HH Perception on Financial Same as [3] but added state during the next 12 months and relative size of debt to Perception on Income Needed to Meet Same as [2] and used total HH income as HH's Need (GEL) as regressors IV regressors Dummy=1 if borrowed/repaid to Bank ONLY in the past 3 months 0.125*** 0.0930*** 13.20 12.51 (0.00640) (0.00731) (27.72) (23.98) Degree of Indebtedness (Reference Level = 0) Debt ( <50% of HH Total Income) -5.688 (11.45) Debt ( 50 - 100 % of HH Total Income) -5.390 (10.97) Debt (>= 100% of HH Total Income) -5.085 (10.49) Number of observations 49,252 31,855 31,855 31,854 Source: Author’s calculation using Georgia IHS. Note: In specifications 2,3,4,5, bank loan dummy is treated as endogenous. Sample are restricted to households in years 2013- 2016 in specifications [2]-[4] due to availability of perception questionnaire. Perception variables are jointly significant at 0.00 significance level. Moreover, estimates suggest that we cannot reject the hypothesis that higher indebtedness would worsen the household welfare. Instead of using dummy for borrowing from the bank, this model uses amount of unpaid debt to the banks (measured as borrowed amount minus repaid amount over household’s total income) as the variable of interest. Here, log of per capita household consumption is regressed on the ratio of unpaid debt to household’s total income in the past 3 months. 27 From the results reported in Table 25 A description of the independent variables with their mean and standard errors are reported in Table A1 in the appendix. 26 Endogeneity test is rejected with p-value of 0.00, suggesting that significant positive impact estimated in the OLS regression is biased as expected. However, test results also indicate that the weak identification test cannot be rejected (for example, with F- statistic equal to 0.38 in specification [4] and 0.378 in the last specification [5]). Weak instrument may be the result of measurement error and how missings were treated in the dataset. Specifically, for households that did not borrow from the bank in the past 3 months, value for this variable is missing. The average interest rates reported by the households in the same year – region – location classification was assigned to the non-borrowers and borrowers with missing interest rates. 27Because actual amount borrowed from banks and spent on repayment are available only for households that borrowed within the past 3 months, households defined as “borrowers� are restricted compared to the first model where we defined households that either borrowed or repaid in the past 3 months as “borrowers.� 18 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 4, we find that higher unpaid debt ratio has positive correlation with per capita consumption when the debt ratio is treated as exogenous (specification [1]) or when it is treated as endogenous but without controlling for the set of household’s attributes that are assumed to be correlated with household’s entrepreneurship and cognitive skills ([2]). However, once these household attributes are taken into account (specifications [3], [4], [5]), the impact turns negative although statistically insignificant. Results again indicate that there is tendency for better off households to borrow more, which lead to overestimate the impact of borrowing. Table 4: Model Results, Estimates of the Effect of Unpaid Debt (in GEL) on Log (Per Capita Consumption Aggregate) [1] [2] [3] [4] [5] OLS IV IV IV IV Added HH perception on financial state during Added GEL per month the next 12 months as needed to meet HH Added two HH perceptions regressors need as regressors as regressors Ratio of Unpaid Debt to HH Income (= (Borrowed - Repaid to Banks in the past 3 months) / Total Income in the past 3 months) 0.00904*** 4.228 -14.55 -8.814 -11.35 (0.00179) (14.54) (67.69) (26.43) (43.58) Number of observations 49,251 49,251 31,854 31,854 31,854 Source: Author’s calculation using Georgia IHS. Indebtedness, as measured by the ratio of unpaid debt to household total income, has no impact, and if any, will increase the household’s likelihood of being in poverty. Table 5 and Table 6 show estimates from additional analyses by regressing household’s poverty status on the size of unpaid debt to banks. Unpaid debt, or indebtedness, is measured by borrowed amount minus repaid amount over household’s total income, all in the past 3 months. Estimates are all insignificant, but specifications controlling for additional household characteristics ([3], [4], [5]) consistently show positive sign – that percent increase in the ratio of unpaid debt over total income would increase the likelihood for households to be in poor/vulnerable status. Table 5: Model Results, Estimates of the Impact of Unpaid Debt on Poverty Status (1 if per adult equivalent is less than national poverty line, 0 otherwise) [1] [2] [3] [4] [5] OLS IV IV IV IV Added HH perception on financial state during Added GEL per month the next 12 months as needed to meet HH Added two HH perceptions regressors need as regressors as regressors Ratio of Unpaid Debt to HH Income (= (Borrowed - Repaid to Banks in the past 3 months) / Total Income in the past 3 months) -4.88e-05 -1.105 5.573 3.425 4.350 (0.00116) (3.893) (25.93) (10.29) (16.72) Number of observations 49,251 49,251 31,854 31,854 31,854 Source: Author’s calculation using Georgia IHS. Table 6: Estimates of the Impact of Unpaid Debt on Likelihood of being in Bottom 40% (1 if HHs belong to the bottom 40%, 0 otherwise) 19 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 [1] [2] [3] [4] [5] OLS IV IV IV IV Added HH perception on financial state during Added GEL per month the next 12 months as needed to meet HH Added two HH perceptions regressors need as regressors as regressors Ratio of Unpaid Debt to HH Income (= (Borrowed - Repaid to Banks in the past 3 months) / Total Income in the past 3 months) -0.00369*** -2.022 7.524 4.583 5.844 (0.00129) (6.997) (35.02) (13.76) (22.46) Number of observations 49,251 49,251 31,854 31,854 31,854 Source: Author’s calculation using Georgia IHS. Conclusion The note contributes to the literature by revealing the pattern of households’ borrowing behavior and estimating causal impact of bank loan on household welfare at the micro level. However, there are important limitations to the study that need further analyses. First, Integrated Household Survey – nationally representative survey used in the note – does not capture all debt from all possible sources. Moreover, impacts are restricted to marginal borrowers and not inframarginal borrowers who borrowed before the reference period defined in the questionnaire (past 3 months). This is a strength in the sense that marginal borrowers are the focus of much theory, practice and policy. But it is a weakness in the sense that impacts on inframarginal borrowers are key to understanding the overall impact of bank loans, and especially if credit market is potentially saturated. Thus, more innovation is needed in combining data from different sources – from credit bureaus and focus surveys with nationally representative household data - to disentangle the relationship between poverty and indebtedness and to assess longer-term impact. Second, it does not answer to the question on when and why households get into debt or too much debt. By questionnaire design, the analysis falls short of capturing the magnitude of indebtedness beyond past 3 months or the use of existing loans, which prevents us from pinning down the causes of possible negative impact on household welfare. Only by using better data, we can test various hypothesis and identify sources of struggle and distress that may possibly worsen the household welfare. How to define “too much debt�, or over-indebtedness, is also a topic that may be revisited. 28 Third, the analysis is capable of providing additional possible policy measures that may influence households’ borrowing behavior without providing concreate policy recommendations until further data and assessments become available. Assistance to the financial sector and support for household debt management have already been proposed by FCI and policies have been put in place or underway. However, to have the overall picture, accurate assessments of market penetration, irresponsible lending practices, over-indebtedness, households’ borrowing sensitivities to credit contracts, are among the few that needs to be identified from supply side and demand side data. Loan pricing is one of the measures that can be effective if done right based on extensive empirical research. Given the dramatic increase in household-level credit and indication of increasing debt stress, there is an urgent need to gather basic facts from the demand side, such as estimates of households’ loan demand curve with respect to interest rate (in other words, households’ price elasticity of demand for bank credit). Only with information on households’ sensitivity to interest rates, 28 For example, D’Alessio and Iezzi (2013) and Banerjee (2013). 20 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 policy makers can effectively design optimal rates in targeted markets. If there is heterogeneity in the price elasticities, loan pricing can be used for targeting certain group of households. Loan maturity is also considered as effective policy parameter that affects demand for credit.29 Market-based regulations, such as compulsory affordability assessments, establishment of credible credit bureaus, and mechanisms to address adverse incentive are priorities to assess lending environment from the supply side. From the supply side, there is a need to validate the magnitude of reckless lending practices and if there is a sign of vicious cycle where increasingly irresponsible lending leading to over-indebtedness of households at the national level. South Caucasus Barometer showed deteriorating public perception towards banks, which may be an early indication of potential debt stress. As increased debt stress can result in social unrest and political repercussions30, actions should be taken to accurately assess the supply side risks and implement monitoring mechanisms at an early stage. In addition to recognizing Georgia’s level of financial development by international standard, assessing variation of financial development within Georgia – with higher market saturation in particular geographic areas or with particular population groups – would also be a key in implementing any policy measures (Box 2). Obtaining a systematic view of credit market and dynamics within would be essential in designing regulatory and policy interventions tailored to Georgian context. Overly restrictive or uniformly prescriptive regulatory environment should also be avoided if Georgia’s financial development is identified as the expansion stage of credit market cycle. Box 2: Variation of Accessibility within Georgia As stated in the Access to Finance and Development: Theory and Measurement31, improving access and building inclusive financial system is a goal that is relevant to economies at all levels of development. There is also empirical evidence that show positive correlation between financial depth and poverty reduction - that better developed financial systems experience faster drops in income inequality and faster reduction in poverty. 32 In many developing countries, however, less than half of the population has access to formal financial services and in most of Africa less than one in five households has access 33. Figure 12 provides a crude indication of geographic access or lack of physical barriers to access for selected countries. First, it is clear that geographic access varies greatly across countries. Focusing on the branches of commercial banks (Figure 12, left), density of branches relative to the population shows that among the peers, Georgia has high rate of 32.7 per 100,000 population, which is equivalent to US at 32.6. On the contrary, Georgia fairs low in terms of geographic distance (branches per 1,000 km2). Combined, indicators illustrate that branches are not distributed equally across country but are clustered in cities and some large towns. 34 As indicators may also reflect the inclusiveness of the financial system, there are high variability in borrowing pattern across regions and urban/rural as will be presented in the main text. Financial services are provided also by the informal sector, such as credit unions and financial cooperatives (Figure 12, right). This channel appears to be extremely uncommon in Georgia as shown earlier in the main text. High penetration of credit union in Poland can be explained by the fact that the country is a member of European Network of Credit Unions, and Germany as the country to first establish the credit union in the 1850s. 35 These figures suggest that in Georgia, other, perhaps more informal intermediaries may be the important financial source. 29 Karlan, D. S. and J. Zinman (2008). 30 Davel, G. (2013). 31 Ibid. 32 Back, T., et al., 2009, 2007 and the World Bank, 2008. 33 Ibid. 34 Better measure would be the average distance from the household to the branch, but these data are available for very few countries. 35 http://www.creditunionnetwork.eu/ 21 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Figure 12: Accessibility to Financial Services – Cross Country Comparison Accessibility of Commercial Banks Accessibility of Credit Unions and GEO Financial Cooperatives POL per 100,000 adults ROU POL per 100,000 adults ARM DEU Branches of credit unions and financial CZE Branches of commercial banks HUN ALB TUR TUR HUN ROU DEU GEO cooperatives CZE POL DEU DEU per 1,000 km2 per 1,000 km2 CZE POL ROU HUN ARM TUR ALB GEO ROU TUR CZE HUN GEO 0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35 Source: Financial Access Survey, IMF (as of September 16, 2018). http://data.imf.org/?sk=E5DCAB7E-A5CA-4892-A6EA- 598B5463A34C 22 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 References Andreasch, M. and P. Lindner, 2014, “Micro and Macro Data: A Comparison of the Household Finance and Consumption Survey with Financial Accounts in Austria,� Working Paper Series, No. 1673, May 2014, European Central Bank. Banerjee, A., D. Karlan, and J. Zinman, 2015, “Six Randomized Evaluations of Microcredit: Introduction and Further Steps,� American Economic Journal: Applied Economics 2015, 7(1): 1-21. Banerjee, S. S., 2013, Credit Market Saturation: Anatomy of a Recent Debate, CGAP Blog 11 July 2013. Beck, T., A. Demirgüç-Kunt, and P. Honohan, 2009, “Access to Financial Services: Measurement, Impact, and Policies,� The World Bank Research Observer, 24 (1): 119-145, April 2009. Beck, T., A. Demirgüç-Kunt, and R. Levine, 2007, “Finance, Inequality and the Poor,� Journal of Economic Growth, March 2007, Vol. 12, Iss. 1, pp. 27-49. Burgess, R. and R. Pande, 2005, “Do Rural Banks Matter? Evidence from the Indian Social Banking Experiment,� The American Economic Review, Vol. 95, No. 3 (Jun, 2005), pp. 780-795. Cancho, C. and E. Bondarenko, 2017, “The Distributional Impact of Fiscal Policy in Georgia,� The Distributional Impact of Taxes and Transfers, Inchauste and Lustig eds. Caucasus Research Resource Center, 2017, Individual indebtedness survey Georgia: Methodological Report, January 16, 2017. Conning, J., and C. Udry, 2005, “Rural Financial Markets in Developing Countries,� The Handbook of Agricultural Economics, Vol. 3, Agricultural Development: Farmers, Farm Production and Farm Markets, edited by Evenson, R.E., P. Pingali, and T. P. Schultz. D’Alessio, G. and S. Iezzi, 2013, Household Over-Indebtedness: Definition and Measurement with Italian Data, mimeo. Davel, G., 2013, “Regulatory Options to Curb Debt Stress�, CGAP Focus Note, No. 83, March 2013. Donou-Adonsou, F., and K. Sylwester, 2016, “Financial development and poverty reduction in developing countries: New evidence from banks and microfinance institutions,� Review of Development Finance 6 (2016) 82-90. Eckerstorfer, P. et al., 2015, “Correcting for the missing rich: An application to wealth survey data,� Review of Income and Wealth, DOI: 10.111/roiw.12188. European Commission, 2015, EU Youth Report – Communication from the Commission to the European Parliament, The Council, The European Economic and Social Committee and the Committee of the Regions, Brussels, 2015. Giese, J., H. Andersen, O. Bush, C. Castro, M. Farag and S. Kapadia, 2014, “The credit-to-GDP gap and complementary indicators for macroprudential policy: Evidence from the UK,� International Journal of Finance & Economics, Vol 19, Iss 1, 2014. https://doi.org/10.1002/ijfe.1489 23 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Giné, X., 2011, “Access to capital in rural Thailand: An estimated model of formal vs. informal credit,� Journal of Development Economics, Vol 96, Iss 1, September 2011, pp. 16-29. Heckman, J. J., 1981, “The Incidental Parameters Problem and the Problem of Initial Conditions in Estimating a Discrete Time-Discreate Data Stochastic Process,� in C.F. Manski and D. McFadden (eds.) Structural Analysis of Discrete Data with Econometric Applications, Cambridge, MA: The MIT Press. Inchauste, Gabriela, and Nora Lustig, eds., 2017, The Distributional Impact of Taxes and Transfers: Evidence from Eight Low- and Middle-Income Countries, Directions in Development, Washington, ED: World Bank. https://openknowledge.worldbank.org/handle/10986/27980 License: CC BY 3.0 IGO. International Monetary Fund, 2018, Georgia 2018 Article IV Consultation, IMF Country Report No. 18/198, June 2018, Washington, DC., IMF. Karlan, D. S., and J. Zinman, 2008, “Credit Elasticities in Less-Developed Economies: Implications for Microfinance,� American Economic Review 2008, 98:3, pp. 1040-1068. Kennickell, A. and D. McManus, 1993, “Sampling for Household Financial Characteristics Using Frame Information on Past Income,� Proceedings of the Survey Research Methods Section, American Statistical Association, Vol. 1, pp.88-97, 1993. King, R.G., and R. Levine, 1993, “Finance and Growth: Schumpeter Might be Right,� The Quarterly Journal of Economics, Vol. 108, No. 3 (Aug. 1993), pp.717-737. Krauss, A., L. Lontzek, and J. Meyer, 2013, Does Market Saturation Increase the Risk of Over- indebtedness?, CGAP Blog 06 May 2013. Lang, J. H. and P. Welz, 2017, “Measuring credit gaps for macroprudential policy,� Financial Stability Review May 2017 – Special features. Levine, R., 1997, “Financial Development and Economic Growth: Views and Agenda,� Journal of Economic Literature, Vol. XXXV (June 1997), pp. 688-726. Lombardi, M., M. Mohanty, and I. Shim, 2017, “The real effects of household debt in the short and long run,� BIS Working Papers, No. 607, January 2017, Bank for International Settlements. Madestam, A., 2014, “Informal finance: A theory of moneylenders,� Journal of Development Economics, Vol. 107, pp. 157-174. Mannah-Blankson, T., “Implication of Microfinance Debt Burden for household Welfare: Lessons from Ghana,� mimeo. McKinnon, R. I., 1973, Money and Capital in Economic Development. Washington, DC: Brookings Institution. Prigozhina, A., N. Tsivadze, and R. Pratt, 2018, Georgia: Indebtedness of Individuals, Household Indebtedness Survey (HIS) Review, May 2018, mimeo. Schularick, M. and Taylor, A. M., “Credit Booms Gone Bust: Monetary Policy, Leverage Cycles, and Financial Crises, 1870-2008�, American Economic Review, Vol. 102(2), 2012, pp. 1029-1061. 24 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 The World Bank, 2018, The Pending Mobility Challenge: Spatial Disparities in the South Caucasus, September 2018, Washington, DC., The World Bank. The World Bank, 2015, The World Bank Group – Georgia Partnership Program Snapshot, April 2015, Washington, DC., The World Bank. The World Bank, 2009, Finance for All? Policies and Pitfalls in Expanding Access, Washington, DC., The World Bank. 25 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Appendix. Figure 13: Poverty Rates – National and Regional National Poverty Rates 32.50% 35% 29.31% % of Population/Households 30% 25% 21.28% 18.05% 20% 15% 10% 5% 0% Year Year Year Year 2011 2016 2011 2016 Individual Level Household Level Source: Author’s calculation using Georgia IHS (left) and from World Bank 2018. Note: Poverty rates are calculated using national consumption aggregates and national poverty lines of 125.9 and 137.13 GEL (per adult equivalent per month) for years 2011 and 2016 respectively. Figure 14: Non-Borrowers and All Borrowers (either from Bank or Private) Share of Poor Households 35% 32.39% 30% % of Households 25% 28.49% 20.60% 20% 15% 17.59% 10% 5% 0% 2011 2012 2013 2014 2015 2016 Non-Borrowing HHs Borrowing HHs Source: Author’s calculation using Georgia IHS. Note: Poverty rates are calculated using national consumption aggregates and national poverty lines of 125.9 and 137.13 GEL for years 2011 and 2016 respectively. Analysis at the household level. Appendix Figure 1: Households and Individuals with Bank Account or a Bank Card 26 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Source: The Caucasus Research Resource Centers. Caucasus Barometer, Armenia and Georgia 2015 (left) and 2011 – 2017 Georgia (right). Retrieved through ODA - http://caucasusbarometer.org on October 24, 2018. Appendix Figure 2: Households and Individuals with Debts DEBTPERS: Do you have any personal debts? (%) 70 66 60 58 57 53 53 50 46 46 43 41 40 34 30 20 10 0 2011 2012 2013 2015 2017 Yes No DK/RA Source: The Caucasus Research Resource Centers. Caucasus Barometer, 2008 – 2010 Georgia (left) and 2011 – 2017 Georgia (right). Retrieved through ODA - http://caucasusbarometer.org on October 25, 2018. 27 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Appendix 1. Loan-related variables in IHS The survey questionnaires focused in the study are as follows from module labelled, “Shinda05�: Q1_3: Please specify, how much did you borrow from a private person during the past three months (GEL). (Write 0 if you did not borrow). Q1_5: Please specify, how much credit did you obtain from the banks or other financial organizations during the past three months (GEL). (Write 0 if you did not obtain any credit.) Q1_9: Please specify, how much GEL did you spend on repayment of the debt to a private person during the past three months. (Write 0 if you did not spend anything.) Q1_10: Please specify, how much GEL did you spend on repayment of the bank credit during the past three months. (Write 0 if you did not spend anything.) Q1_5a: If you obtain a bank credit, please specify the interest rate. (Annual) Appendix 2. Interest Rates Figures are provided here to show the sample variation of interest rates used for the IV. Figure 15 shows the distribution of interest rates reported by households on bank loans by year (response to question Q1_5a). Figure 15: Annual Interest Rates Reported by Households Source: Author’s calculation using Georgia IHS. For households with missing interest rates and for those who did not borrow, average interest rate within the given year, region and urban/rural was assigned. The distribution of the average rates is shown below. 28 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 Source: Author’s calculation using Georgia IHS. Appendix 3. Literature Review, Model Specification, Estimation Strategy and Data Literature Review At the macro-level, there is a large empirical body of literature that identifies positive correlation between financial development, economic growth and poverty reduction using cross-country data.36 While theory provides conflicting views about the impacts of financial development on the economic growth, inequality and poverty reduction37, there are ample empirical evidence demonstrating a strong, positive link between financial development and economic growth at the macro level.38 Some studies even show that the level of financial development is a good predictor of future economic development.39 More recent studies examine the causal relationship between financial development and poverty reduction at the macro level, addressing the endogeneity associated with financial development. Causality at the macro level using panels are reported in Boukhatem (2016), Uddin et al. (2014), cross- sectional datasets in Rewilak (2017). Macro-level evidence also exists in identifying causal impact of microfinance on poverty reduction. The positive impact of microfinance has been reported in studies such as Burgess and Pande (2005), Lopatta and Tchikov (2017), Miled and Rejeb (2015). 36 See for example, Levine (1997), The World Bank (2009), Beck, T. et al. (2009), Donou-Adonsou and Sylwester (2016), Lombardi et al. (2017). 37 See Levine (1997) for a comprehensive review on theoretical and empirical analyses. 38 Seminal book by McKinnon (1973) studies the relationship between the financial system and economic development in Argentina, Brazil, Chile, Germany, Korea, Indonesia, and Taiwan in the post-World War II period. 39 King and Levine (1993), for example, studies 80 countries over the period 1960-1989. 29 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 At the micro-level, empirical analyses have struggled to identify causal impact of microfinance on poverty reduction. This is due to the well-known selection biases that can come from both the demand- side and supply-side as described in Banerjee et al. (2015). Attempts were made for example by Pitt, Rosenzweig, and Gibbons (1993), Pitt and Khandker (1998), McKernan (2002), Townsend (2011), Breza and Kinnon (2018), and critical assessment of the methodology by Morduch (1998), Chemin (2008), Chowdhury (2009), Duvendack and Palmer-Jones (2010), Roodman and Morduch (2013) among others. Given the limits in using nonexperimental data to evaluate causal impact of microfinance at the micro-level, researchers increasingly turned to randomized controlled trials. Recent studies include six studies published in American Economic Journal: Applied Economics (2015)40, and Banerjee et al. (2017), Coleman (1999). Quasi-experimental analyses based on treatment and control groups are also conducted in countries such as India, Thailand and Malaysia.41 Acknowledging the virtue of randomization, this paper relies on instrumental variable (IV) approach to address the causality. Due to the non-randomized nature of bank loans in our setting and cross-sectional nature of the dataset, we use the interest rate each household reports in the survey as the IV for taking up the bank loan. The analysis draws on the methodology adopted in the microfinance literature. The objective is to estimate the causal relationship between bank borrowing and poverty. There is a growing concern over households’ indebtedness and its effect on household welfare in Georgia. Taking on debt can increase consumption beyond what one’s income can support, it can smooth consumption in face of shocks and it can represent an investment in the future. However, over indebtedness may result in significant financial distress, forcing households to be caught in poverty trap.42 By drawing on lessons from the empirical literature on microcredit, the note tries to address the causal impact of bank loans on household welfare. Policy implications – whether and how Government should promote or repress financial intermediation – will be discussed at the end. Model For the implementation of IV method, the first-stage equation for taking bank loans 𝐿𝑖𝑗 is, 𝐿 𝐿 𝐿𝑖𝑗 = 𝑋𝑖𝑗 𝛽𝐿 + �𝑖𝑗 𝜋 + 𝜇𝑗 + 𝜖𝑖𝑗 , (1) where 𝐿𝑖𝑗 is a dummy for taking bank loans such that 𝐿𝑖𝑗 = 1 if a household i in area j either has borrowed or repaid to banks (or both) in the past 3 months of the survey and 𝐿𝑖𝑗 = 0 otherwise, 43 𝑋𝑖𝑗 is a vector of household characteristics (e.g., age, education, marital status, and labor market status of the household head and household’s demographic features), �𝑖𝑗 is a set of household or village characteristics distinct from the X’s in that they affect 𝐿𝑖𝑗 but not on other household behaviors conditional on 𝐿𝑖𝑗 , 𝛽𝐿 and 𝜋 are unknown 𝐿 𝐿 parameters, 𝜇𝑗 is an unmeasured determinant of 𝐿𝑖𝑗 that is fixed within an area j, and 𝜖𝑖𝑗 is a nonsystematic 40 Banerjee et al. (2015). 41 Coleman (1999) for Thailand, Samer et al. (2015) for Malaysia, 42 Implications of indebtedness are described, for example, in Mannah-Blankson 43 Ideally, 𝐿𝑖𝑗 would measure the stock amount of bank credit. However, dataset only has the amount of “flows�, i.e., the amount of bank loans that occurred in the past 3 months. which would highly underestimate the effect of borrowing. In the later analysis, the model will be restricted to households that had borrowed within the past 3 months allowing the estimation of the impact of indebtedness (current bank loan less amounts repaid) on household welfare status. 30 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 𝐿 𝐿 error term that reflects unmeasured determinants that vary over households such that 𝐸(𝜖𝑖𝑗 |𝑋𝑖𝑗 , �𝑖𝑗 , 𝜇𝑗 )= 0. The conditional demand for outcome 𝑦𝑖𝑗 (such as household’s welfare) conditional on 𝐿𝑖𝑗 -whether the household has a bank loan (or had borrowed from the bank in the past) is, 𝑦 𝑦 𝑦𝑖𝑗 = 𝑋𝑖𝑗 𝛽𝑦 + 𝐿𝑖𝑗 𝛿 + 𝜇𝑗 + 𝜖𝑖𝑗 , (2) where 𝑦𝑖𝑗 is the continuous variable measuring household’s per capita consumption aggregates,44 𝛽𝑦 and 𝛿 𝑦 𝑦 are unknown parameters, 𝜇𝑗 is an unmeasured determinant of 𝑦𝑖𝑗 that is fixed within an area, and 𝜖𝑖𝑗 is a nonsystematic error reflecting unmeasured determinants of 𝑦𝑖𝑗 that vary over households such that 𝑦 𝑦 𝐿 𝑦 𝐸(𝜖𝑖𝑗 |𝑋𝑖𝑗 , 𝐿𝑖𝑗 , 𝜇𝑗 ) = 0. The estimation issue arises as a result of the possible correlation of 𝜇𝑗 with 𝜇𝑗 and 𝐿 𝑦 of 𝜖𝑖𝑗 with 𝜖𝑖𝑗 . Econometric estimation that does not take these correlations into account may yield biased estimates of the parameters in equation (2) due to endogeneity of taking bank loans, 𝐿𝑖𝑗 . In the model set above, the exogenous regressors �𝑖𝑗 in equation (1) are the identifying instruments. We apply the approach motivated by demand theory – that is, to use the price of the endogenous variable as an identifying instrument. In our case, the most obvious measure of the price of bank loan is the interest rate charged, which is available in the dataset. There is sufficient level of variation across the sample as each household reports interest rate charged from the banks. Admittingly, using reported interest rate does not entirely address the issue of endogeneity as it is likely that some of the variation in interest rates may reflect unmeasured household attributes unknown to researchers but known to the lender and likely to be 𝐿 part of the error term, 𝜖𝑖𝑗 , which violates the exclusion restriction of IV. Unfortunately, other variables that gives exogenous variation to obtaining bank loans that can justifiably be used as an IV are difficult to find. Without panel data on households before and after the availability of bank loans, interest rate is the best candidate for our instrument to estimate 𝛿 . However, we have tried to address this issue by taking the average of reported interest rates within the area in which the household resides (urban/rural within a specific region) in a specific quarter within a given year. By using the average of interest rates within the specific location – season – year group, we can treat the interest rate as exogenous to household’s wellbeing but correlated with the take-up of loans. To control for other household specific attributes that might affect both outcome (household welfare) and decision to borrow, included in 𝑋𝑖𝑗 ’s are household’s subjective views on well-being 45 : household’s perception on the changes in financial status in the past 12 months, and income needed in order to meet the basic needs (in GEL). With the exception of question on income needed for basic needs, these are categorical variables classified into 5 to 6 categories. To control for area-specific unobservables, we use area-specific fixed-effects (FE) technique that 𝐿 𝑦 treats the area-specific errors 𝜇𝑗 and 𝜇𝑗 as parameters to be estimated and thus control for area-specific 44 When both the behavioral outcome (in our case, borrowing from banks) and the outcome variable are measured as binary indicators, identification of the behavioral impact is generally not possible (Pitt and Khandker, 1991). Thus, in our model specification, the outcome variable 𝑦𝑖𝑗 is continuous. For robustness check when restricting the sample to borrowing households within the past 3 months to focus on recent flow of borrowing, we may relax this restriction and use dichotomous indicator as the outcome variable in the second stage. 45 Other perception variables were dropped due to weak statistical power (e.g., household’s perception on own economic status given their income levels, expectations about the financial conditions in the next 12 months, and perception of the housing condition). 31 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 unobservables.46 In our case, areas are defined by urban/rural divide within each of the nine regions which is expected to control for nonrandom placement of bank branches. Due to the lack of village-level survey or any variable at the village-level, no other village-level attributes are controlled for in the model specification. Year and quarter dummies are included to control for year trend and seasonal bias. Data Data used for the analysis is the Georgia Integrated Household Survey (IHS) from 2011 to 2016 as described in Box 2. For some specifications, years are restricted due to the availability of perception questions or use restricted sample for robustness check. Dependent variable is the logarithm of per capita consumption aggregate temporally and spatially adjusted. As described earlier, our variable of interest 𝐿𝑖𝑗 is a dummy variable equal to one if the household either has borrowed from the bank in the past 3 months and/or has repaid back to the bank in the past 3 months, and zero otherwise. The reason for adopting this definition is that, there are quite a number of households that had not borrowed but has reported to have been repaying back during the specified period (past 3 months from the survey date). Because actual amount borrowed and spent on repaying are available for households that borrowed/repaid in the past 3 months, subsample of households is used in the later analysis when estimating the effect of indebtedness. Weighted Means and Standard Errors of Variables used for Regression Analysis – Household Attributes Table A1: Weighted Means and Standard Errors of Variables Household Attributes 46 Fixed-effect estimation with limited dependent variables raises the issue of consistency, which may become an issue in the later section when we use household’s poverty status as the dependent variable. However, Heckman (1981) provides Monte Carlo evidence that with eight or more observations per fixed-effects unit, the inconsistency issue becomes relatively inconsequential. In our sample, as we use region-urban/rural as the fixed-effect unit with more than eight observations in each, this would not become an issue. 32 GEORGIA INDEBTEDNESS_POVERTYNOTE_NOV 5 2011 2012 2013 2014 2015 2016 N mean se(mean) N mean se(mean) N mean se(mean) N mean se(mean) N mean se(mean) N mean se(mean) Age of Head 11206 59.67508 0.143129 11265 59.4418 0.142925 11091 59.42879 0.141119 11158 60.10503 0.138953 10999 60.08219 0.142577 10858 60.11582 0.139497 Age of Head Squared 11206 3790.661 16.72592 11265 3763.422 16.75217 11091 3752.634 16.49088 11158 3828.032 16.34006 10999 3833.438 16.79729 10858 3825.183 16.44114 HH Size 11206 3.617313 0.018542 11265 3.611596 0.017828 11091 3.598659 0.017946 11158 3.602532 0.018042 10999 3.593326 0.018037 10858 3.542118 0.01785 Dummy Female HH head 11206 0.352638 0.004514 11265 0.346275 0.004483 11091 0.331822 0.004471 11158 0.334463 0.004467 10999 0.330326 0.004485 10858 0.337364 0.004538 Share of 0-6 y.o. within HH 11206 0.057532 0.001107 11265 0.063975 0.001157 11091 0.062187 0.001166 11158 0.061142 0.001161 10999 0.06336 0.001201 10858 0.064534 0.001228 HH Demographics Share of 7-15 y.o. within HH 11206 0.077349 0.001364 11265 0.076189 0.001332 11091 0.07407 0.001327 11158 0.078196 0.001364 10999 0.07677 0.001375 10858 0.074652 0.001351 Share of 66+ y.o. within HH 11206 0.231759 0.003264 11265 0.220503 0.003175 11091 0.21574 0.003147 11158 0.225208 0.003168 10999 0.225853 0.003206 10858 0.219755 0.003171 Dummy Own House 11200 0.916231 0.002618 11256 0.9213 0.002538 11091 0.933507 0.002366 11158 0.936886 0.002302 10999 0.931385 0.002411 10858 0.944437 0.002198 Living Space (in m2) 10382 73.76712 0.423923 10111 77.46922 0.442137 9876 78.44082 0.456775 9731 80.77607 0.453114 9685 83.37673 0.482721 9610 84.49064 0.51069 Whole Space (in m2) 10378 113.4254 0.672571 10105 117.9554 0.693822 9839 122.4599 0.745511 9766 126.1546 0.778555 9705 126.2781 0.761612 9607 125.2065 0.770879 Cultivation Area (in hectar) 11206 0.300721 0.008994 11265 0.353874 0.016531 11091 0.361385 0.010581 11158 0.383121 0.009664 10999 0.388766 0.013973 10858 0.392542 0.012074 Dummy HH has car/motorcycle/truck/tractor 10937 0.265596 0.004223 11123 0.300055 0.004346 11049 0.30873 0.004395 11115 0.333664 0.004473 10952 0.375643 0.004628 10829 0.388045 0.004683 Dummy HH has TV/PC 10937 0.965597 0.001743 11123 0.970854 0.001595 11049 0.975742 0.001464 11115 0.984606 0.001168 10952 0.985249 0.001152 10829 0.985458 0.00115 Dummy HH has heating device Assets (heater/ac/gas elec stove) 10937 0.674946 0.004479 11123 0.748284 0.004115 11049 0.826935 0.003599 11115 0.864174 0.00325 10952 0.893973 0.002942 10829 0.916905 0.002653 Dummy HH has basic electronics (fridge/washing mach/vacuum) 10937 0.711976 0.00433 11123 0.773727 0.003968 11049 0.821247 0.003645 11115 0.859799 0.003293 10952 0.89991 0.002868 10829 0.920974 0.002593 Number of mobile phone per HH 10937 1.35813 0.011688 11123 1.547518 0.011361 11049 1.754752 0.012046 11115 1.952687 0.012579 10952 2.131029 0.01308 10829 2.138687 0.012562 Quantity of cow/bull/baffalo/cattle 11206 0.746494 0.01437 11265 0.797531 0.015652 11091 0.842863 0.016903 11158 0.000877 0.000828 10999 1.06575 0.028653 10858 1.042387 0.029578 Quantity of donkey/horse 11206 0.034067 0.001962 11265 0.035238 0.002126 11091 0.02967 0.001805 11158 0 0 10999 0.034175 0.00233 10858 0.044863 0.003889 Quantitiy of poultry 11206 5.0461 0.088788 11265 4.589345 0.080846 11091 5.049869 0.090971 11158 0.005818 0.002533 10999 5.776831 0.109149 10858 5.311808 0.097032 Quantitiy of other (pig/sheep/goat/rabit/bee) 11206 0.629364 0.036368 11265 0.568698 0.04376 11091 0.784241 0.077307 11158 0.001162 0.000756 10999 2.813378 1.05981 10858 1.976237 0.727561 Dummy =1 if HH received retirement pension in past 3 mon 9873 0.55524 0.005002 10181 0.535786 0.004943 9774 0.550475 0.005032 9736 0.588787 0.004987 9116 0.623623 0.005075 8849 0.630358 0.005132 Social Benefit Dummy =1 if HH received assistance for socially vul fam in past 3 mon 9873 0.147972 0.003574 10181 0.141347 0.003453 9774 0.153992 0.003651 9736 0.135621 0.00347 9116 0.126108 0.003477 8849 0.117833 0.003428 Incomplete 5-12 686 522 705 436 332 259 General Sec 5,050 5,082 4,424 5,158 5,308 5,208 Education of HH Special Sec 2,655 2,783 2,782 2,837 2,709 2,635 head Tertiary 2,487 2,574 2,498 2,535 2,550 2,660 None/66) within HH (>avg (1) 66) within HH (>avg (1) 66) within HH (>avg (1) 66) within HH (>avg (1) 66) within HH (>avg (1) 66) within HH (>avg (1)