Overdue debts and financial exclusion

Berlinger, Edina; Dobránszky-Bartus, Katalin; Molnár, György Article Overdue debts and financial exclusion Risks Provided in Cooperation with: MDPI – Multidisciplinary Digital Publishing Institute, Basel Suggested Citation: Berlinger, Edina; Dobránszky-Bartus, Katalin; Molnár, György (2021) : Overdue debts and financial exclusion, Risks, ISSN 2227-9091, MDPI, Basel, Vol. 9, Iss. 9, pp. 1-21, https://doi.org/10.3390/risks9090158 This Version is available at: https://hdl.handle.net/10419/258242 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ Risks 2021, 9, 158. https://doi.org/10.3390/risks9090158 www.mdpi.com/journal/risks Article Overdue Debts and Financial Exclusion Edina Berlinger 1,*, Katalin Dobránszky-Bartus 1 and György Molnár 2 1 Department of Finance, Corvinus University of Budapest, Fővám tér 8, 1093 Budapest, Hungary; [email protected] 2 Centre for Economic and Regional Studies, Institute of Economics, Tóth Kálmán u. 4, 1097 Budapest, Hungary; molna[email protected].hu * Correspondence: [email protected]; Tel.: +36-(30)5541075; Fax: +36-(1)4825212 Abstract: We examine the impact of overdue debts in small villages in one of Hungary’s most disadvantaged regions. We find that a significant number of debtors with overdue debts permanently escape from debt collectors. Accordingly, in our sample, overdue debts reduce the likelihood of declared work by 14 percentage points on average. The lack of declared work alone reduces the probability of opening a bank account by 21 percentage points, and overdue debts further reduce it by 9 percentage points. The negative effect of overdue debts on health is almost as large as the positive effect of a high school diploma. In addition, the health-destroying effect extends not only to the debtor but to all members of the household. Therefore, overdue debts create a poverty trap mechanism exacerbating financial exclusion, hence resulting in significant losses for both the individual and society. We recommend paying more attention to smoothing credit cycles and resolving nonperforming debt obligations. Keywords: financial inclusion; poverty trap; overdue debts; debt relief 1. Introduction Non-performing, overdue debts are regularly incurred in credit cycles, usually to an accelerated extent after a crisis. It is well-documented in the literature that employment is an important factor in repaying household loans. If someone loses their job, the probability of non-payment increases significantly (Ben-Galim and Lanning 2010; Balás et al. 2015; Campbell and Cocco 2015; Dimitrios et al. 2016). Some recent papers recognized, however, that there is reverse causality, too, as overdue debts have a negative effect on employment as well. According to Mian and Sufi (2014), through the housing net worth channel, a decline in the housing net worth reduces consumer demand, hence labor demand. Similarly, Verner and Gyöngyösi (2020) showed that overdue foreign currency mortgage loans reduced aggregate demand, leading to job losses, and, thus lowering employment levels and economic growth. However, overdue debts can reduce not only labor demand but also labor supply. For example, through the financial distress channel, deteriorating creditworthiness reduces an employee’s chances of finding a job, keeping a job, or choosing working conditions (Herkenhoff 2019; Dobbie et al. 2020). In addition, through the housing lock channel, the deterioration in the collateral value of mortgages leads to a decrease in or a complete lack of labor mobility (Bernstein and Struyven 2017). Bernstein (2017) introduced the household debt overhang channel, which means that due to the renegotiation of overdue debts, repayments become income-contingent, hence debtors become motivated to hide their incomes. In the present study, we investigate the effects of overdue debts on employments, as well, but in a specific context. Filling a gap in the literature, we do not analyze the debtors in the average situation, but people living in small villages in a disadvantaged region of Hungary. A novelty of our research is that we investigate the relationship between overdue debts and employment in the context of financial exclusion. Financial exclusion is a Citation: Berlinger, Edina, Katalin Dobránszky-Bartus, and György Molnár. 2021. Overdue Debts and Financial Exclusion. Risks 9: 158. https://doi.org/10.3390/risks9090158 Academic Editor: Stelios Markoulis Received: 10 July 2021 Accepted: 25 August 2021 Published: 31 August 2021 Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. Copyright: © 2021 by the authors. Submitted for possible open access publication under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/). Risks 2021, 9, 158 2 of 21 well-researched area, especially in the US and UK. Still, some vulnerable groups such as rural inhabitants are completely neglected (Fernández-Olit et al. 2019), overdue debts do not receive enough attention (Krumer-Nevo et al. 2017), and researchers do not ask unbanked people directly (Koku 2015). Our study fills a gap in all three respects. We define debts in a broader sense, including all types that can trigger a debt collection process in the case of non-payment (utility bills, bank loans, and tax liabilities). We introduce a new channel that has so far been unexplored in the literature, such as escape from debt collection and deductions. At the first sight, it is similar to the household debt overhang channel described by Bernstein (2017), but in our case, debts are not renegotiated, so they remain unresolved for decades. The escape from debt collection can curb economic growth in several ways: it can reduce legal employment and the willingness to use a bank account. Moreover, it can damage the mental and physical health of the debtors and their families in the long run due to the constant stress they live with (Fitch et al. 2011). Relative to the financial literature focusing on the relationship between overdue debts and employment, in our sample, overdue debts are more common while the institutional system dealing with financial difficulties is less developed and accessible. On the one hand, the Hungarian personal bankruptcy system is much stricter, hence less attractive than in the US and most EU countries. On the other hand, in this special segment, debt renegotiations with banks or debt collectors are much less effective. Debts are relatively small, communication with debtors is more difficult and costly, and the collateral is less valuable compared to the average indebted household. These conditions may explain why small overdue debts remain unsettled en masse and for decades, while large nonperforming loans are renegotiated much more efficiently and quickly. Furthermore, larger borrowers can obtain significantly larger discounts (Tirole 2006) contrary to moral intuition (Kornai 2016). In our research, the negative effects of overdue debts last typically longer than in the financial literature and undermine economic growth and social cohesion across several business cycles. Escape from debt collection creates a special poverty trap, as it creates a positive feedback mechanism through which poverty is reproduced and even exacerbated (Azariadis 1996). Due to overdue debts, the lender avoids declared work and electronic payment, trying instead to make a living from casual work in the black economy and paying for everything in cash. Thus, overdue debtors benefit less from the services of the welfare systems (unemployment benefits, health care, pensions, etc.) and formal financial services (payment services, savings opportunities, loans, etc.), becoming more vulnerable and being forced to make worse compromises (Allen et al. 2016). Escape from debt collection as a life strategy severely limits the debtors’ and their families’ capabilities as defined by Sen (2014). Several kinds of poverty traps, although based on different mechanisms, were presented by Banerjee and Duflo (2011), mainly in relation to education, health, and financial systems. Mullainathan and Shafir (2013) examined the mechanism of the debt trap, its psychological, behavioral effects, the ‘scarcity mindset’, and the role of unexpected expenditures. The thought experiment underlying our empirical analysis is what would happen if long-standing non-performing loans were renegotiated and restructured (combined with partial debt relief), thus the threat of recovery over the debtors’ heads would be averted. Our research aims to find out what impact such a program is expected to have on employment, bank accounts, and the health of the population. Krugman (1988) introduced the concept of the so-called debt relief Laffer-curve. According to this concept, debt relief can increase lenders’ income (just like tax cuts can increase tax revenue of the state through improved tax incentives) and, at the same time, a borrower’s well-being by removing barriers to employment and financial inclusion (World Bank 2012). Kanz (2016) examined the impact of the most extensive debt relief program in economic history targeting households (agricultural entrepreneurs) in India and found that the expected positive effects had not materialized. On the contrary, the Risks 2021, 9, 158 3 of 21 saved debtors had piled up their informal debts and decreased their investments, and their productivity had fallen compared to debtors who had not been saved. Mukherjee et al. (2018), analyzing the same debt relief program, showed that the situation of those who found themselves in difficult situations due to exogenous shocks (weather) and not their own faults have significantly improved and became financially more included. Similar conclusions were drawn by Dobbie and Song (2020) on a different sample, performing a randomized controlled experiment on overdue credit card debts in the US. Ong et al. (2019) found that debt relief programs positively affected the mental state of the debtors; particularly, anxiety and present-biasedness decreased. Therefore, they supported debt relief programs in addressing poverty. Academic views regarding the effectiveness of debt relief programs are mixed. Moreover, there are only a few studies in the field, not least because quality data are not available for research. Creditors have the interest to hide information if they agree with the debtor on partial or complete debt relief of non-performing debts, as debt reliefs increase moral hazard (Fudenberg and Tirole 1990) and lead to the soft budget constraint syndrome (Kornai 1998). In addition, it is the payment discipline of not only the borrowers who have received the debt relief that deteriorates in the future (if they expect to be rescued again and again) but also of other previously performing debtors if discounts become known. However, examining the decision of US mortgage debtors on strategic default, Guiso et al. (2013) found that the willingness-to-pay depends also on non-pecuniary factors, such as fairness and morality. Bhutta et al. (2017) also concluded that US mortgage borrowers are reluctant to walk away even if it were beneficial for them, thus, moral hazard can be lower than suspected. It is clear, therefore, that debt relief programs can have significant individual and social costs and benefits in the long run. In this light, instruments that can efficiently prevent the excessive build-up of households’ credit risk and are also able to address complex tradeoffs in case of crisis can create huge value. The present study is aimed at contributing to the development of more efficient debt relief instruments. Section 2 discusses the hypotheses of the study. In Section 3, the considered database is presented. Section 4 provides a comparative description of households living with and without overdue debts. Section 5 analyzes the impacts of overdue debts in a multivariable setting. Finally, the conclusions are summarized in Section 6. 2. Development of Hypotheses At the start of this research, we conducted 14 in-depth interviews with local residents, mostly women, in a chosen settlement of Borsod-Abaúj-Zemplén (BAZ) County. We collected information on households’ financial management, their savings, and borrowing habits. During the interviews, the issue of utility, bank, and other debts and problems arising from default was raised several times. In many cases, the pattern emerged that the household obtained general-purpose, consumer, or mortgage credit(s) from banks either in Hungarian forint (HUF) or in foreign currency and then failed to repay them due to some exogenous shock (job loss, health deterioration, exchange rate and interest rate changes, etc.). Banks handed over non-per- forming loans to debt collectors, and, since then, overdue debts have just further accumulated. The formal process for dealing with overdue debts can be summarized as follows. We distinguish three stages if the debtor does not pay his debt. The recovery phase is the first 90 days after arrears, during which the creditor actively contacts the debtor and seeks alternative solutions for payment. If the debtor is overdue for more than 90 days (with the repayment of a bank loan, utility bill, etc.), the unpaid financial obligation enters the claim phase. The lender will usually continue to look for alternative solutions, but the claim will be handed over to a claim manager. After 180 days from the date of non-payment, in the enforcement phase, the contract is terminated, the claim is sold, and debt is collected based on out-of-court or in-court proceedings. In addition to penalty interest, the collection and Risks 2021, 9, 158 4 of 21 enforcement costs borne by the debtor can significantly increase the debt’s value. The debtor’s assets, income, and movable and immovable property are under the scope of enforcement. Hungarian legislation does not recognize the institution of datio in solutum. If the collateral of the loan is not sufficient to repay the outstanding debt, the debtor remains liable for the outstanding part of the debt. During the enforcement, movable and immovable property may be sold to cover the outstanding debt, and a certain amount may be automatically deducted from the debtor’s registered tax-paying income before the debtor receives it (MNB 2019). Act LIII of 1994 on Judicial Enforcement states that the claim must be recovered primarily from the debtor’s wages; the amount recovered may not exceed 33%, or, exceptionally, 50% (e.g., in case of child support or multiple foreclosures). Accordingly, the debt collector is required to examine whether the debtor has a declared job. Our interviewees, who reported overdue debts, were already in the enforcement phase without exception, so all their legal income was subject to a 33% or 50% deduction. As the market value of the movable and immovable property is typically very low in this segment, foreclosures and evictions were not worthwhile for the debt collectors; consequently, unresolved debts have persisted for many years, and the penalty interest has accumulated. The initial loans of a few hundred thousand forints have since grown to debts of several million. The creditors renounced ever being able to repay these huge sums, so they gave up trying. Letters sent by debt collectors are not even opened, the exact amount of the debt is not known. The debtors equipped themselves to hide their incomes and potential savings from debt collectors throughout their lives. Interviewees reported that, in many cases, they do not apply for registered jobs or open a bank account specifically because of overdue debts. Overdue debts have a negative effect also on debtors’ mental and physical health. They are angry at banks and debt collectors, feel misled, and do not want to have any business with the banks, thus accepting their long-term financial exclusion. Based on the existing literature and the findings of the in-depth interviews, we formulate the following three hypotheses in relation to the negative impacts of overdue debts: Hypothesis 1 (H1). Debtors with overdue debts are less likely to apply for a registered job. Hypothesis 2 (H2). Debtors with overdue debts are less likely to open a bank account. Hypothesis 3 (H3). Debtors with overdue debts suffer from worse mental and physical conditions than debtors without overdue debts. In the following sections, we examine these hypotheses in detail based on the questionnaire survey. 3. Data Collection To collect targeted data, Soreco Research Limited conducted a questionnaire-based survey in March and April 2019, on behalf of the Corvinus University of Budapest and under the framework of the “Financial and Public Services” research project of the Higher Education Institutional Excellence Program. All the procedures were performed in compliance with relevant laws and institutional research ethical guidelines. The research focused on the financial management of households in the small settlements of BAZ County. The sample is representative of non-urban households in the county. Data were collected by personal interviews, via the so-called multistage stratified random sampling procedure. In the first stage of the sampling procedure, the settlements to be sampled were selected and the number of households to be interviewed in each settlement was determined to reflect the proportion of the non-urban population of the districts. In the second stage Risks 2021, 9, 158 5 of 21 of sampling, the interviewers selected the households to be interviewed using a predetermined selection algorithm, the so-called random walk method. In other words, households were not selected based on a preliminary address list but randomly. In the given settlement, there was a pre-recorded starting point. From this starting point, based on a fixed-route algorithm, every fifth household was selected for an interview. In each household, the household member most competent in financial matters was asked to complete the questionnaire. If the randomly chosen household refused to respond, the nearest household was contacted. Sampling was carried out anonymously, and no personal data were collected, so households cannot be identified. In total, we have information on 504 households and 1794 individuals; 1196 were of active age (18–65 years), and 179 had overdue debts. Households with no active-age members were excluded from the analysis. Of the remaining 496 households, 136 were inhabited by individuals who had some form of overdue debts (177 individuals in total). The questionnaire included questions for each adult, for the household, and for the respondent only. Economic activity, the possession of a bank account, and overdue debts play a key role in our analysis, and information on these is known for all adult members of the interviewed households. The majority of the respondents (72%) were women as they were more familiar with household financial management. See Appendix A for details on the variables used in the analysis. 4. Descriptive Statistics In this section, we use data obtained from the questionnaire to characterize the individuals and households with and without overdue debts. First, we present the results of a direct inquiry on what the respondents think about the causal relationships formulated in H1, H2, and H3; second, we perform a one-dimensional comparative analysis of their characteristics. 4.1. Direct Inquiry In the questionnaire, we asked directly whether the interviewee knows someone who, because of his or her overdue debts and the fear of debt collection does not take up a declared job (H1), does not open a bank account (H2), or has experienced a deterioration in his or her health (H3). Respondents also had to indicate whether this kind of causality exists for him- or herself, for a close family member living in the household, for somebody in the wider family, or for someone living in the settlement or in a wider circle of acquaintances. Nearly half of all respondents (49%) know someone who does not have a declared job because of overdue debts. Similarly, nearly half of the respondents reported the negative effects of overdue debt on bank accounts and health (45% and 58%, respectively). Considering only the households with overdue debts, almost a quarter of the respondents reported that there is at least one person in their household who, specifically because of the outstanding overdue debts, does not take up a declared job (18%), does not open a bank account (21%), or has experienced a health deterioration (34%). It is noteworthy that the deterioration in health, both in a narrower and wider context, was given greater emphasis by the respondents than the other two consequences. We examined those households where no one had a bank account (30% of all households) in more detail. In previous research, representative of the whole country, the proportion of such households was lower: 24% (Illyés and Varga 2015) and 17% (Horn and Kiss 2019). In our case, the main reason for the lack of a bank account was that it is not needed (60%) and/or it is too expensive (51%), but 21% of the respondents referred to the fear of debt collection. At the national level, 90% of those who did not have a bank account said they did not need it, 25% said it was too expensive, 10–11% did not trust credit institutions, and 3–4% feared security risks (Illyés and Varga 2015). We note that the questionnaire of Illyés and Varga (2015) did not include the possibility of fear of debt collection, Risks 2021, 9, 158 6 of 21 hence such fears probably appeared in the answers “do not trust credit institutions” and “fear of security risks.” Clearly, in our sample (villages of BAZ County), costs play a significantly larger role than in the national-level sample, which may be due to lower incomes. Although not listed in Table 1, we also directly asked the interviewees about the extent to which overdue debts are a problem in general. According to respondents, this is a serious problem in their immediate environment (66%), in the village (81%), and nationally (89%). Table 1. Results of the direct inquiry. Do You Know Someone, Who—Due to Their Overdue Debts and the Fear from Debt Collection... % of All Respondents (N = 496) % of Respondents with Overdue Debts (N = 117) Within the Household In a Wider Circle of Acquaintances * Within the Household … does not take up a declared job, because their wage would be decreased by deductions. 6% 49% 18% … does not open a bank account, as their debited money would be decreased by deductions. 8% 45% 21% … experienced a deterioration in their health. 11% 58% 34% * Including the household. Source: Questionnaire-based survey, small settlements of BAZ County, Hungary, 2019. Note: The table shows how respondents answered our direct inquiry, i.e., whether they knew someone in their environment who did not take up a declared job, did not open a bank account, or had experienced a deterioration in their health specifically because of overdue debts. The answers to the above questions are consistent, and we have no reason to suspect that the interviewees did not understand the questions or that the answers were significantly distorted for other reasons, although smaller biases are possible in both directions. At the level of the closest family members, due to the personal involvement and the need for self-discharge, the impact of overdue debts may be somewhat exaggerated. On the other hand, respondents are likely to admit neither overdue debts nor hiding from enforcement. Therefore, the two opposite biases may extinguish each other to some extent. At the same time, at the level of a wider circle of acquaintances, the lack of information may have skewed the responses downwards. Nevertheless, if there is some bias, the respondents are more likely to present the problem as being less severe than it is. The above results, therefore, support our hypotheses based on in-depth interviews. 4.2. Comparative Analysis Next, analyzing the answers to the questionnaire, we examine whether the statistical characteristics of the sample are consistent with our hypotheses. Our primary goal is to characterize individuals and households with and without overdue debts. In some cases when data were missing, observations were removed from the sample. Table 2 shows the extent to which the subsamples of individuals with (179 individuals) and without (1017) overdue debts differ from each other. Variables in bold are included in the multivariate regression analysis to examine the combined effects of the variables (in Section 5). Risks 2021, 9, 158 7 of 21 Table 2. Differences between individuals and households with and without overdue debts. Number of Observations Variable With Overdue Debts Without Overdue Debts Significance of the Difference, p-Value Variables related to individuals 1196 Full-time job 11% 42% *** 0.000 1196 Declared work 13% 45% *** 0.000 1183 Gender (women) 50% 50% 0.970 1196 Age (year) 42.9 39.2 *** 0.000 1196 Education (less than elementary school) 21% 4% *** 0.000 1196 Education (high school diploma) 3% 27% *** 0.000 1196 Net income (thousand HUF) 60.4 95.8 *** 0.000 1157 Bank account 32% 60% *** 0.000 286 Loan instalment (thousand HUF) 21.0 28.9 *** 0.000 Variables related to households 1196 Household members 3.3 3.6 ** 0.029 1196 Children 1.7 1.3 *** 0.004 1151 Income per capita (thousand HUF) 52.3 90.6 *** 0.000 1160 Forex loan 21% 20% 0.576 1143 Ability-to-pay 1.0 1.2 *** 0.000 1196 Social aversion to overdue debts 0.2 0.3 ** 0.034 1196 Social aversion to non-declared work 0.2 0.2 0.964 1165 Settlement development 40.9 42.4 0.112 1165 Chronic illness 31% 21% *** 0.010 Variables related to the respondents 496 Smoking 56% 38% *** 0.001 496 Alcohol 1% 3% * 0.093 493 Medication 38% 38% 0.893 496 Stressed 60% 36% *** 0.000 495 Hopeless 48% 22% *** 0.000 495 Tired 65% 46% *** 0.000 496 Unhappy 51% 24% *** 0.000 496 Not socializing 37% 55% *** 0.000 496 Satisfied with health 48% 56% 0.126 Source: Questionnaire-based survey, small settlements of BAZ County, Hungary 2019. Note: The table shows the average values of the variables for the subsamples of individuals or households with and without overdue debts. The description of the variables can be found in Appendix A. Differences were tested with the independent t-test, the Mann–Whitney test, or the Kruskal–Wallis test depending on the distribution of variables in the subsamples. We reject the null hypothesis of independence if the p−value is smaller than the benchmark significance levels of 1%, 5%, and 10% (marked with ***, **, or *, respectively). Looking at the full-time jobs or all declared jobs in Table 2, the difference is notable for the two subsamples (31 and 32 percentage points, respectively) and statistically significant. Hence, there is a close negative association between employment and overdue debts. However, it is not clear whether overdue debts cause lower employment (overdue debts → employment) or vice versa, the lack of employment causes non-payment (employment → overdue debts). It is likely that both effects occur simultaneously and this positive feedback loop creates a vicious circle leading to a poverty trap. Considering that the year 2019 (when the survey was taken) was characterized by a strong economic boom and general labor shortages, we can assume that the causality of interest (overdue debts → employment) was strongly present as someone who wanted a full-time and declared job during this period had plenty of opportunities. Anecdotes also support the fact that, Risks 2021, 9, 158 8 of 21 in many cases, employers offer special employment contracts specifically designed to avoid deductions (for example, by paying the wages in cash with daily settlements); moreover, some foreign job opportunities have been advertised explicitly as a tool to evade debt collections. There is no difference in gender between those with overdue debts and those without. Those with overdue debts are a few years older, but the difference, while statistically significant, is not remarkable. Figure 1 shows that the relationship is nonlinear (middle-aged people are more likely to take out a loan and more likely to become insolvent than young or older people). Figure 1. Distributions of age in the subsamples of individuals with and without overdue debts. Source: Questionnairebased survey, small settlements of BAZ County, Hungary, 2019. In terms of education, we found a significant difference (Mann–Whitney test p-value = 0.000) between the two subsamples. Those with overdue debts have a lower education level in general. Among them, the proportion of those who have not completed the eight classes of elementary school is significantly higher, and the proportion of those who do not have a high school diploma is significantly lower. As education can have an effect on both overdue debts and employment, we will control for this variable in the multivariate regression model. The net income from work in the previous month is strongly correlated with declared and full-time jobs. The difference is significant in this respect, too, as those without overdue debts have an income advantage of more than HUF 35,000. Figure 2 shows the distribution of net incomes across the two subsamples. 0 2 4 6 8 10 12 14 16 15 20 25 30 35 40 45 50 55 60 65 % With overdue debt Without overdue debt Risks 2021, 9, 158 15 of 21 Figures 4 and 5 show the distribution of perceived social aversion of each behavior in the studied population. To recap, interviewees were asked to select from a pre-defined list and rank the five most convicted behavioral patterns, based on which we calculated a composite indicator for overdue debts (utility bills, bank loans, tax). We then extended this indicator to all members of the household. The composite indicator—the perceived social aversion to overdue debts—is presumably not strongly related to the control variables as the question refers to the opinion of the village. At the same time, it is possible that the individual view on the opinion of the village is also determined by individual and household characteristics. In any case, in our sample, the perceived social aversion to overdue debts hardly correlates with any other control variables; therefore, the hypothesis of independence cannot be rejected. It is possible that there is a link between various deviant behaviors, such as undeclared work and refusal to pay loans. However, we did not find any indication of this in a statistical sense. It is also possible that societal expectations may change, for example, because of an exogenous shock. The foreign currency credit crisis in Hungary (Bethlendi 2011) could be considered as an exogenous shock, as a result of which, the society may become less likely to convict those who do not pay their debts if the proportion of non-paying households increases (Becker and Murphy 2000). In our database, however, there is no significant relationship between foreign currency loans and the perceived social aversion (Mann–Whitney test p value is 0.391). The relationship is examined by breaking down by settlements, but the t-tests and Mann–Whitney tests do not indicate a relationship between the two variables. Further, the development of the settlement can, in principle, be related to social aversions, but we do not see a close correlation here either (+0.05). Logically, we can assume that the perceived social aversion to overdue debts affects the repayment of loans but does not directly affect employment, except through the channel of overdue debts, so it can be considered as a suitable instrumental variable in this context. In a linear probability model (LPM), we first regress the variable X (overdue debts) and then the variable Y (declared work) on the IV (perceived social aversion to overdue debts), and we get the values of −0.033 and +0.133 for the coefficients (the standard error is 0.018 and 0.024, respectively), which implies 0.133/−0.033 = −4.03 coefficient for the relationship between X and Y. Although the signs are as expected, the coefficient of 4.03 is incomprehensible in the LPM. By performing the same analysis but in a logit regression model, we conclude that overdue debts reduce the chances of the debtor having a declared job by 14%. Despite the theoretical and practical limitations of our analysis based on the selected instrumental variable, the estimated coefficient of 14% seems realistic, considering the results of the in-depth interviews, direct interviews, and multivariate regression analyses accounting for the effects of the potential omitted variables. Bernstein (2017) found on a sample of US mortgage borrowers representative for the whole population that overdue debts decrease employability by 2–6%. Our estimate (14%) is much higher, not least because our sample is representative of people living in small villages in one of the most disadvantaged counties of Hungary. 5.2. Overdue Debts and Bank Account As a next step, we examine our second hypothesis more closely, namely that overdue debts have a negative effect on bank account ownership. Bank account ownership is the best proxy for financial inclusion as this is the prerequisite for all other financial services (Allen et al. 2016). Several studies showed a close relationship between declared work and a bank account (Illyés and Varga 2015; Koku 2015; Allen et al. 2016; Krumer-Nevo et al. 2017; Horn and Kiss 2019; Fernández-Olit et al. 2019). Thus, if overdue debts negatively affect declared work (see Table 3), these can indirectly affect bank accounts too. However, based on the in-depth interviews, we assume that there is a direct channel between overdue debts and the bank account, as well, through the escape from debt collection channel. Risks 2021, 9, 158 16 of 21 Table 4 shows the results of the multivariate regression analysis. To make our results comparable with the results of similar, albeit nationally representative research (Illyés and Varga 2015; Horn and Kiss 2019), we extended our regression model with net income and declared work. Table 4. Overdue debts and bank accounts. Y = Bank Account N = 1131 Modified R2 = 0.323 Beta t p C Intercept −0.14 −0.97 0.332 X Overdue debts −0.09 −2.31 ** 0.021 Z1 Gender (female) 0.08 3.14 *** 0.002 Z2 Age 0.00 0.50 0.620 Z2 Age^2 0.00 −0.61 0.541 Z3 Education: Elementary school 0.07 1.27 0.204 Z3 Education: Vocational exam 0.21 3.78 *** 0.000 Z3 Education: High school diploma 0.32 5.38 *** 0.000 Z3 Education: University degree 0.40 5.00 *** 0.000 Z4 Net income 0.00 5.78 *** 0.000 Z5 Ability-to-pay ratio 0.03 1.01 0.312 Z6 Chronic illness −0.04 −1.14 0.256 Z7 Declared work 0.21 6.52 *** 0.000 Source: Questionnaire-based survey, small settlements of BAZ County, Hungary, 2019. Note: The table summarizes the results of the OLS-based regression model. C is the intercept, X is the main explanatory variable, and the Zs are the control variables. In the case of education, the completion of fewer than 8 classes of elementary school is the reference. Instead of settlement development, we applied district dummy variables. The results are similar when calculating robust standard errors. If the coefficients are significant at the levels of 1% or 5%, the p-values are marked with *** or **, respectively. Opening a bank account is impacted by the banking services’ accessibility which is not necessarily reflected in the settlement development indicator. Therefore, we used district dummy variables instead of the settlement development indicator. District dummies proved to be significant, indicating remarkable differences between different regions in terms of the available banking services. According to Table 4, overdue debts are negatively related to bank accounts: if an individual has overdue debts, the likelihood of using a bank account decreases by 9 percentage points on average ceteris paribus. A declared job increases the chances of using a bank account by 21 percentage points, and, in line with our expectations, the income has a positive impact on bank account ownership. If net income and employment is left out from the model, the coefficient of overdue debts increases to 15 percentage points. Women are more likely to open a bank account, but this finding is not robust across different model specifications. The use of a bank account initially increases with age, which is consistent with the results of (Illyés and Varga 2015; Horn and Kiss 2019). However, in our sample, there is no subsequent negative effect of age, probably because we only examined those of active age (18–65 years). Education also has a strong effect on our dataset: the higher the degree, the higher the probability of having a bank account, and the magnitude of this effect is similar to the findings of (Illyés and Varga 2015; Horn and Kiss 2019). The coefficients of the variables “ability-to-pay” and “chronic illness” have the expected sign, but they are only significant in model specifications without the net income and employment variables. Risks 2021, 9, 158 17 of 21 5.3. Overdue Debts and Health In this section, we investigate our third hypothesis i.e., overdue debts have a negative impact on health. Health is measured by a factor derived from 10 variables related to mental and physical health (smoking, alcohol, medication, stressed, hopeless, tired, unhappy, no socializing, dissatisfied with health, and chronic illness) using a principal component analysis. Based on the KMO value (=0.605) and the Bartlett test (p-value is 0.000), we determined one common factor, the so-called unhealthy index. The higher the value of the unhealthy index, the worse the health condition of the individual (the value of the index varies between −1.54 and +2.37). In Table 5, the outcome variable is the unhealthy index and the main explanatory variable is overdue debts again, and we control for important individual-, household-, and settlement-level variables. Variables determining the unhealthy index are known only for the respondents, therefore, in this model, the number of cases is only 480 (some did not provide the information needed to produce the unhealthy index). While in the case of employment and the bank account, the negative effects occur only in the case of a person with overdue debts; the stress caused by overdue debts can destroy the mental and physical health of the whole family. So, in this model, the main explanatory variable—overdue debts—indicates whether there are overdue debts in a given household or not. Table 5. Overdue debts and health. Y = Unhealthy Index Y = Unhealthy Index N = 481 Modified R2 = 0.322 N = 480 Modified R2 = 0.328 Beta t p Beta t p C Intercept 0.17 0.71 0.479 0.12 0.47 0.636 X Overdue debts in the household 0.23 4.88 *** 0.000 0.21 4.36 *** 0.000 Z1 Gender (female) 0.07 1.73 ** 0.084 0.07 1.56 0.120 Z2 Age 0.01 0.69 0.491 0.01 0.93 0.353 Z2 Age^2 0.00 0.58 0.561 0.00 0.29 0.772 Z3 Education: Elementary school −0.09 −1.14 0.256 −0.09 −1.09 0.276 Z3 Education: Vocational exam −0.29 −3.46 *** 0.001 −0.28 −3.30 *** 0.001 Z3 Education: High school diploma −0.30 −3.33 *** 0.001 −0.28 −3.12 *** 0.002 Z3 Education: University degree −0.51 −4.53 *** 0.000 −0.51 −4.28 *** 0.000 Z4 Net income 0.00 0.58 0.560 Z5 Forex loan 0.05 1.08 0.283 Z6 Ability-to-pay ratio −0.09 −2.51 *** 0.013 −0.09 −2.52 *** 0.012 Z7 Settlement development 0.00 −1.35 0.177 0.00 −1.20 0.231 8 Declared work −0.10 −1.83 ** 0.069 Source: Questionnaire-based survey, small settlements of BAZ County, Hungary, 2019. Note: The table summarizes the results of OLS-based regression models. C is the intercept, X is the main explanatory variable, and Zs are the control variables. In the case of education, the completion of fewer than 8 classes of elementary school is the reference. The results are similar when calculating robust standard errors. If the coefficients are significant at the levels of 1% or 5%, the p−values are marked with *** or **, respectively. As expected, the sign of the coefficient of overdue debts is positive and significant in both specifications, which means that overdue debts destroy health. Table 5 also suggests that, in absolute terms, the size effect of overdue debts on health is comparable to that of a vocational exam or a high school diploma after completing elementary school. Age is not significant, but gender is, women have slightly worse health in this sample. Education matters again, the more someone studies, the healthier they live. Similarly, the ability-to- pay ratio, which is a more balanced family budget, has a positive impact on health too. Net income is irrelevant, and surprisingly, declared work has a weak negative impact on health. Having had a foreign currency loan in the past has no significant impact. Similarly, the settlement development is not significant, and controlling for this effect at the district Risks 2021, 9, 158 18 of 21 level instead of the settlement level does not significantly change the results. The above models, thus, support the negative relationship between overdue debts and health revealed during the in-depth interviews and through the direct inquiries. 6. Conclusions We examine the negative impacts of overdue debts on employment, bank accounts, and health using several methods (in-depth interviews, direct inquiries, and statistical analyses). According to our estimations, overdue debts reduce the likelihood of having a declared job by nearly 14 percentage points. Not having a declared job reduces the probability of owning a bank account by 21 percentage points; in addition, overdue debts further decrease the probability of having a bank account by 9 percentage points. Furthermore, overdue debts have a negative effect on the health of the family members living in the household of the debtor, and this negative effect is roughly of the same magnitude as the positive impact of obtaining a vocational certificate or a high school diploma after completing primary school. Overdue debts slow down economic growth not only through the decreased labor demand, as (Mian and Sufi 2014; Verner and Gyöngyösi 2020) showed, but also through the decreased labor supply as we presented. Overdue debts can have negative effects not only in a crisis but also in a boom for many decades thereafter. Our research methodology has limitations. Most of all, one can never be sure that endogeneity is fully excluded from the models. Endogeneity can emerge from simultaneous causality, omitted variables, and measurement errors (Bascle 2008). In our multivariate regression models with control variables, all three are relevant issues. Therefore, we performed an additional analysis using an instrumental variable (perceived social aversion to overdue debts) as well. Literature review, in-depth interviews, direct inquiries, multivariate analysis, instrumental variable method, and robustness checks strengthen each other and indicate that our results are reliable. The intuitive new channel (escape from debt collection) and the comprehensive and representative survey data also add to the quality of our research. We can conclude that overdue bills, tax, and bank loans have similar effects on employment, bank services, and health through the escape from debt collection channel. The estimated size effects refer to a disadvantaged population in Hungary, and these proved to be highly significant in economic terms. Results can be generalized for other populations of overdue debtors, too, as debt collection can create perverse incentives and poverty trap mechanisms everywhere if overdue debts are not settled effectively. Of course, side effects can vary widely depending on institutions, personal bankruptcy systems, labor market tendencies, cultural factors, etc. To reduce the negative social and economic impacts of overdue debts, policymakers should pay more attention to attenuating credit cycles (debt control rules, consumer protection, and other anticyclical policies) and settling non-performing debts, especially in this fragile segment of the society. Thus, specific debt relief programs are needed, and state intervention can be justified by the positive external effects in terms of employment, growth, tax income, subsidies, health care costs, children’s perspective, black economy, etc. Most of all, we argue in favor of more lenient personal bankruptcy regulations to promote the fresh start of overdue borrowers and their families. A bad financial decision should not ruin entire families. Market-based debt renegotiations should also be more effective for example by using FinTech solutions (for example, online platforms for renegotiations and bargaining, income-contingent repayments, smart contracts and decentralized clearing, etc.). According to (Kshetri 2017; Fernández-Olit et al. 2019) this is an under-researched but promising direction. Our results indicate that well-designed debt relief programs could be attractive for borrowers, too, as hiding from credit collectors for a lifetime has high personal costs. In Risks 2021, 9, 158 19 of 21 these programs, moral hazard issues should be carefully addressed. International evidence suggests, however, that moral hazard can be much less serious as it is widely believed (Guiso et al. 2013; Bhutta et al. 2017). As far as development policies are concerned, we believe that financial inclusion is not possible without the settlement of existing overdue debts. This must be the first step and well before promoting saving accounts and regular savings plans. Our database is not suitable for estimating the extent of the problem at the national level. This would require a sufficiently detailed representative sample of the entire population with a large number of observations. A critical study of existing debt relief programs and the development of possible solutions would also require a separate study. Author Contributions: Conceptualization, E.B. and G.M.; methodology, K.D.-B. and G.M.; software, K.D.-B.; validation, E.B. and G.M.; formal analysis, E.B.; investigation, K.D.-B.; resources, E.B.; data curation, K.D.-B.; writing—original draft preparation, E.B.; writing—review and editing G.M.; visualization, K.D.-B.; supervision, E.B. and G.M.; project administration, E.B.; funding acquisition, E.B. All authors have read and agreed to the published version of the manuscript. Funding: This research was supported by the Higher Education Institutional Excellence Program 2020 of the Ministry of Innovation and Technology in the framework of the ‘Financial and Public Services’ research project (TKP2020-IKA-02) at Corvinus University of Budapest. Institutional Review Board Statement: Not applicable. Informed Consent Statement: Not applicable. Data Availability Statement: Not applicable. Conflicts of Interest: The authors declare no conflict of interest. Appendix A Table A1. Description of variables. Variable Description 0 Overdue debts The value is 1 if the individual has any formal overdue debts (loan, utility bills, tax), otherwise 0, for each active age member of the household. 1 Full-time job The value is 1 if the individual has a full-time registered job (public work excluded), otherwise 0, for each active age member of the household. 2 Declared work The value is 1 if the individual has full or part time declared job, otherwise 0, for each active age member of the household. 3 Gender The value is 0 if male, 1 if female, for each member of the household. 4 Age Calculated based on the year of birth, in years, for each member of the household. 5 Education Category variable indicating five categories: less than 8 cl asses of elementary school, elementary school, vocational exam, high school diploma, university degree. For each member of the household. 6 Net income Net amount in thousand HUF received last month from employment (including pension, caretaking support, unemployment benefit, child-support, maternity, etc.). For each active-age member of the household. 7 Bank account If the individual has a bank account, the value is 1, otherwise 0. 8 Loan instalment Instalment (in thousand HUF) of the individual’s bank loan(s), for each active-age member of the household. 9 Household members Number of household members living under the same postal address (weight is 1 if above 18 years, 0.5 if below 18 years of age). 10 Children Number of children (below the age of 18) living in the same household (i.e., under the same postal address) (weight is 1 for all children). 11 Income per capita Sum of net incomes of all household members divided by the weighted number of household members. Risks 2021, 9, 158 20 of 21 12 Forex loan If somebody in the household had (ever) taken out a foreign exchange denominated loan, the value is 1, otherwise 0. 13 Ability-to-pay ratio The sum of net monthly incomes of the household divided by the monthly expenses. 14 Perceived social aversion to overdue debts Perceived social aversion to the non-payment of utility bills, loans, tax according to the respondent (higher value, greater aversion). 15 Perceived social aversion to black work Perceived social aversion to undeclared work according to the respondent (higher value, greater aversion). 16 Settlement development Settlement development indicator based on societal, demographic, living conditions, local economic, employment-related, infrastructural, and environmental factors) 17 Chronic illness The value is 1 if there is someone in the household, whose health condition prevents employment, otherwise 0. 18 Smoking If the individual is smoking regularly, the value is 1, otherwise 0. 19 Alcohol If the respondent consumes alcohol more than once a week, the value is 1, otherwise 0. 20 Medication If there is anyone in the family who permanently needs medication, the value is 1, otherwise 0. 21 Stressed If the respondent was stressed for more than half of the days in the last two weeks, the value is 1, otherwise 0. 22 Hopeless If the respondent was hopeless for more than half of the days in the last two weeks, the value is 1, otherwise 0. 23 Tired If the respondent was tired for more than half of the days in the last two weeks, the value is 1, otherwise 0. 24 Unhappy If the respondent was unhappy for more than half of the days in the last two weeks, the value is 1, otherwise 0. 25 Not socializing The value is 1 if the individual goes out (religious, cultural, sport, recreational purposes) less than once a year, otherwise 0. 26 Satisfied with health The value is 1, if the individual is totally or rather satisfied with her/his health, otherwise 0. References (Allen et al. 2016) Allen, Franklin, Asli Demirgüç-Kunt, Leora F. Klapper, and Maria Soledad Martinez Peria. 2016. The foundations of financial inclusion: Understanding ownership and use of formal accounts. Journal of Financial Intermediation 27: 1–30. (Azariadis 1996) Azariadis, Costas. 1996. The economics of poverty traps. Journal of Economic Growth 1: 449–86. (Balás et al. 2015) Balás, Tamás, Adam Banai, and Zsuzsanna Hosszú. 2015. Modelling probability of default and optimal PTI level by using a household survey. Acta Oeconomica 65: 183–209. (Banerjee and Duflo 2011) Banerjee, Abhijit V., and Esther Duflo. 2011. Poor Economics: A Radical Rethinking of the Way to Fight Global Poverty. New York: Public Affairs, Perseus Books Group. (Bascle 2008) Bascle, Guilhem. 2008. Controlling for endogeneity with instrumental variables in strategic management research. Strategic Organization 6: 285–327. https://doi.org/10.1177/1476127008094339. (Becker and Murphy 2000) Becker, Gary S., and Kevin M. Murphy. 2000. Social Economics, Market Behavior in a Social Environment. Cambridge: Harvard University Press. (Ben-Galim and Lanning 2010) Ben-Galim, Dalia, and Tess Lanning. 2010. Strength against Shocks. Low-Income Families and Debt. London: The Institute for Public Policy Research (IPPR), pp. 1–22. Available online: http://library.bsl.org.au/jspui/bitstream/1/1494/1/strength_against_shocks%5B1%5D.pdf (accessed on 4 February 2021). (Bernstein 2017) Bernstein, Asaf. 2017. Negative equity, household debt overhang, and labor supply. The Journal of Finance 1–79. Available online: https://leedsfaculty.colorado.edu/AsafBernstein/Asaf_Bernstein_NegativeHomeEquityHouseholdDebtOverhangLaborSupply _20180614.pdf (accessed on 6 February 2021). (Bernstein and Struyven 2017) Bernstein, Asaf, and Daan Struyven. 2017. Housing Lock: Dutch Evidence on the Impact of Negative Home Equity on Household Mobility, SSRN Scholarly article. Available online: https://ssrn.com/abstract=3090675 (accessed on 25 April 2021). (Bethlendi 2011) Bethlendi, András. 2011. Policy measures and failures on foreign currency household lending in Central and Eastern Europe. Acta Oeconomica 61: 193–223. Risks 2021, 9, 158 21 of 21 (Bhutta et al. 2017) Bhutta, Neil, Jane Dokko, and Hui Shan. 2017. Consumer Ruthlessness and Mortgage Default during the 2007 to 2009 Housing Bust. The Journal of Finance 72: 2433–66. (Campbell and Cocco 2015) Campbell, John Y., and Joao F. Cocco. 2015. A model of mortgage default. The Journal of Finance 70: 1495– 554. (Dimitrios et al. 2016) Dimitrios, Anastasiou, Louri Helen, and Tsionas Mike. 2016. Determinants of non-performing loans: Evidence from Euro-area countries. Finance Research Letters 18: 116–19. (Dobbie et al. 2020) Dobbie, Will, Paul Goldsmith-Pinkham, Neale Mahoney, and Jae Song. 2020. Bad credit, no problem? Credit and labour market consequences of bad credit reports. The Journal of Finance 75: 2377–419. (Dobbie and Song 2020) Dobbie, Will, and Jae Song. 2020. Targeted debt relief and the origins of financial distress: Experimental evidence from distressed credit card borrowers. American Economic Review 110: 984–1018. (Fernández-Olit et al. 2019) Fernández-Olit, Beatriz, José María Martín Martín, and Eva Porras González. 2019. Systematized literature review on financial inclusion and exclusion in developed countries. International Journal of Bank Marketing 38: 600–26. https://doi.org/10.1108/IJBM-06-2019-0203. (Fitch et al. 2011) Fitch, Chris, Sarah Hamilton, Paul Bassett, and Ryan Davey. 2011. The relationship between personal debt and mental health: A systematic review. Mental Health Review Journal 16: 153–66. https://doi.org/10.1108/13619321111202313. (Fudenberg and Tirole 1990) Fudenberg, Drew, and Jean Tirole. 1990. Moral hazard and renegotiation in agency contracts. Econometrica 58: 1279–319. (Guiso et al. 2013) Guiso, Luigi, Paola Sapienza, and Luigi Zingales. 2013. The determinants of attitudes toward strategic default on mortgages. The Journal of Finance 68: 1473–515. (Herkenhoff 2019) Herkenhoff, Kyle F. 2019. The impact of consumer credit access on unemployment. The Review of Economic Studies 86: 2605–42. (Horn and Kiss 2019) Horn, Daniel, and Hubert Janos Kiss. 2019. Who does not have a bank account in Hungary today? Financial and Economic Review 18: 35–54. (Illyés and Varga 2015) Illyés, Tamás, and Lóránt Varga. 2015. Show me how you pay, and I will tell you who you are—Socioeconomic determinants of paying habits. Financial and Economic Review 14: 26–61. (Kanz 2016) Kanz, Martin. 2016. What does debt relief do for development? Evidence from India’s bailout for rural households. American Economic Journal: Applied Economics 8: 66–99. (Koku 2015) Koku, Paul Sergius. 2015. Financial exclusion of the poor: A literature review. International Journal of Bank Marketing 33: 654–67. (Kornai 1998) Kornai, Janos. 1998. The place of the soft budget constraint syndrome in economic theory. Journal of Comparative Economics 26: 11–17. (Kornai 2016) Kornai, János. 2016. Breaking Promises. The Hungarian Experience. Working Paper. Budapest: Corvinus University of Budapest Faculty of Economics. (Krugman 1988) Krugman, Paul R. 1988. Market-Based Debt-Reduction Schemes. Working paper 2587. NBER Working Papers. Cambridge: National Bureau of Economic Research. (Krumer-Nevo et al. 2017) Krumer-Nevo, Michal, Anastasia Gorodzeisky, and Yuval Saar-Heiman. 2017. Debt, poverty, and financial exclusion. Journal of Social Work 17: 511–30. (Kshetri 2017) Kshetri, Nir. 2017. Potential roles of blockchain in fighting poverty and reducing financial exclusion in the global south. Journal of Global Information Technology Management 20: 201–4. (Mian and Sufi 2014) Mian, Atif, and Amir Sufi. 2014. What Explains the 2007–2009 Drop in Employment? Econometrica 82: 2197–223. https://doi.org/10.3982/ECTA10451. (MNB 2019) MNB. 2019. Magyar Nemzeti Bank 2/2019. (II.13.) on Consumer Debt Management Activities. Budapest: Hungarian National Bank. Available online: https://www.mnb.hu/letoltes/2-2019-koveteleskezeles.pdf (accessed on 24 April 2021). (Mukherjee et al. 2018) Mukherjee, Saptarshi, Krishnamurthy Subramanian, and Prasanna Tantri. 2018. Borrowers’ Distress and Debt Relief: Evidence from a Natural Experiment. The Journal of Law and Economics 61: 607–35. (Mullainathan and Shafir 2013) Mullainathan, Sendhil, and Eldar Shafir. 2013. Scarcity: Why Having Too Little Means so Much. Times Books. New York: Henry Holt and Company. (Ong et al. 2019) Ong, Qiyan, Walter Theseira, and Irene Y. H. Ng. 2019. Reducing debt improves psychological functioning and changes decision-making in the poor. Proceedings of the National Academy of Sciences 116: 7244–49. (Sen 2014) Sen, Amartya. 2014. Development as freedom 1999. The Globalization and Development Reader: Perspectives on Development and Global Change. New York: Alfred Knopf, p. 525. (Tirole 2006) Tirole, Jean. 2006. The Theory of Corporate Finance. Princeton: Princeton University Press. (Verner and Gyöngyösi 2020) Verner, Emil, and Győző Gyöngyösi. 2020. Household debt revaluation and the real economy: Evidence from a foreign currency debt crisis. American Economic Review 110: 2667–702. (World Bank 2012) World Bank. 2012. Reducing Vulnerability and Promoting the Self-Employment of Roma in Eastern Europe through Financial Inclusion. Washington, DC: World Bank.