Related Lending and Financial Development

Transcription

1 Related Lending and Financial Development Robert Cull (World Bank) Stephen Haber (Stanford University) and Masami Imai (Wesleyan University) Abstract: This paper examines how related lending affects financial development and whether institutional environments work to alter the effects of related lending by using two unique data sets, (1) crosscountry data on the permissiveness of related lending and (2) bank-level data on the amount of related lending in Uganda. Our cross-country analysis indicates that related lending, on average, does not have any effects on financial development. However, it finds that related lending tends to retard financial development when rule of law is weak while it tends to promote financial development when rule of law is strong. The paper also finds that related lending adversely affects bank performance in Uganda only in domestic banks, but not foreign banks, suggesting that the institutions of corporate governance might partly determine whether related lending is being used in ways that hurt financial development. Taken as a group, our results indicate that whether related lending is positive or pernicious depends critically on the institutional context in which it takes place, and that there is no single best policy regarding related lending. Corresponding author: Robert Cull, Development Research Group, the World Bank, MC3-300, 1818 H St. NW, Washington, DC, rcull@worldbank.org. Fax ; Tel: Contact information for other authors: Stephen Haber, Department of Political Science, Stanford University, Stanford, CA haber@stanford.edu. Fax ; Tel: Masami Imai, Department of Economics, Wesleyan University, Middletown, CT mimai@wesleyan.edu. Tel: The views are those of the authors and not necessarily those of the World Bank or its affiliate institutions. Imai gratefully acknowledge financial support from Wesleyan University for Mellon Faculty Carreer Development Mini Grant. We thank Varun Kshirsagar for his excellent research assistance and Thorsten Beck, Richard Grossman, and Aldo Musacchio for valuable comments. An earlier version of this paper was presented at the Harvard Business School International Research conference. All errors are our own. 1

2 1. Introduction There is a broad consensus that bankers in developing countries engage in related lending. They commonly extend credit to firms owned by their close business associates, members of their own families or clans, or businesses that they own themselves. However, there is not yet a consensus on how related lending affects financial development. Broadly speaking, there are competing views. The first, the looting view, which is informed by recent LDC financial crises and which has come to be the conventional wisdom at multilateral aid organizations, holds that related lending is pernicious. It allows insiders (bank directors) to expropriate outsiders (minority shareholders, depositors, and, when there is under-funded deposit insurance, taxpayers). The incentives for insiders to expropriate the outsiders are particularly strong during an economic crisis, when the insiders have incentives to use the resources of the bank to rescue their other enterprises (Akerlof and Romer, 1993; La Porta, López-de-Silanes, Shleifer, and Vishny, 1997, 1998; Rajan and Zingales, 1998; Johnson, Boone, Breach, and Friedman, 2000; Johnson, La Porta, López-de-Silanes, and Shleifer, 2000; Laeven, 2001; Bae, Kang, and Kim, 2002; Mitton, 2002; Habyarimana, 2003; La Porta, López-de-Silanes, and Zamarripa, 2003). 1 Outsiders, of course, know that they may be expropriated, and therefore behave accordingly: they refrain from investing their wealth in banks, either as shareholders or depositors. The combined effect of tunneling by directors, the resulting instability of the banking system, and the reluctance of outsiders to deploy their wealth in banks is a small financial system. 1 In a crisis, loan repayment by unrelated parties worsens, and thus it becomes more difficult to reimburse depositors and continue operating as a bank. The insiders therefore find it in their interest to make loans to themselves, and then default on those loans in order to save their non-bank enterprises. 2

3 The second view, the information view, which is informed by the economic histories of the United States, Germany, and Japan, is that related lending has a positive effect on financial system development: it allows banks to overcome information asymmetries, and creates mechanisms for bankers to monitor borrowers (Gerschenkron, 1940; Aoki, Patrick, and Sheard, 1994; Lamoreaux, 1994; Calomiris, 1995; Fohlin, 1998). Scholars who stress this view of related lending would wonder why, if related lending is pernicious, it characterized the banking systems of advanced industrial countries during their periods of rapid growth. They would also make the point that related lending is still widespread in those same countries (Kroszner and Strahan, 2001). Not only do the looting and information views differ from each other in terms of bankers economic motivation behind related lending and its potential effects on financial development, but also these two competing views may have entirely different policy implications. For instance, the looting view can be used to justify stringent regulatory restrictions on cross-ownership between banks and non-financial firms and/or quotas on the amount of loans to related parties so as to reduce the opportunities and incentives of bank insiders to expropriate outsiders. On the contrary, the information view of related lending suggests that such regulatory restrictions might be counterproductive as they are likely to diminish the ability of banks to overcome asymmetric information problems. The primary goal of this paper is to bring more empirical evidence to this unsettled issue. It does so by examining empirical relations between related lending and financial outcomes using two unique data sets: (1) a cross sectional data set for 76 countries on regulatory restrictions on related lending and private credit growth, and (2) a microeconomic bank-level data set on related lending, bank ownership, and bank 3

4 performance in Uganda. We move away from the typical good or bad approach, and investigate whether the institutional environment alters the manner in which related lending retards (or promotes) financial development and banking performance. We advance and test the hypotheses that rule of law, depositor monitoring, and the quality of corporate governance affect the relation between related lending and financial development. We find three notable empirical results. First, cross-country analysis reveals no clear relationship between regulatory restrictions on related lending and financial development; i.e., the average effects of related lending on financial development are neither positive nor negative. However, when we allow the effects of related lending to vary with rule of law, we find that related lending has negative effects only in countries where rule of law is weak and that it has positive effects where rule of law is strong. These results suggest that when judicial systems are corrupt and public officials can enforce laws and regulations selectively, bank insiders are more likely to be able to expropriate bank outsiders with impunity, resulting in a small financial system. Second, unlike the rule of law, depositor monitoring does not seem to affect the statistical relation between related lending and financial development, suggesting that this monitoring might not be an effective mechanism to detect and curtail potential looting by bankers. Third, our case study of Uganda, a country with weak rule of law and weak depositor monitoring compared to developed countries, shows that related lending has pernicious effects on bank performance only for domestic banks, not foreign banks, suggesting that good corporate governance and perhaps reputational capital associated with foreign ownership might work to alter the nature of related lending. These findings 4

5 indicate that a negative or positive view of related lending might be too simplistic. Instead, the success of related lending might depend critically on the institutional environment. The rest of this paper is organized as follows. In section 2 we briefly discuss some theoretical considerations and present our hypotheses. In section 3 we describe our crosscountry data and empirical methodology. Section 4 contains the main cross-country results, while section 5 contains those for our case study of related lending in Uganda. In section 6 we present a series of robustness checks for our cross-country results. Finally, in section 7 we offer concluding remarks. 2. Hypotheses How do institutional environments alter the nature of related lending? Simply put, bank insiders might use related lending as a device to expropriate outsiders when the costs of looting (e.g., the possibility and severity of legal punishment) are less than the benefit, when bank insiders are not closely monitored (i.e. looting does not get detected), and when bank insiders incentives are not appropriately aligned with bank outsiders and other stakeholders. In this section, we argue that strong rule of law, intense depositor monitoring, and good corporate governance can reduce the incentive and opportunity for bank insiders to use related lending as a mechanism to expropriate. That is, we hypothesize that these institutional characteristics determine whether related lending is positive or negative for financial development. When judicial systems are corrupt and public officials can enforce laws and regulations selectively, bank insiders are more likely to be able to expropriate bank outsiders with impunity. When there is strong rule of law, however, insiders face 5

6 sanctions if they expropriate the outsiders. As a result, they may engage in related lending, but they do not use it as a mechanism to tunnel. Instead, they use it as a mechanism to overcome information asymmetries and monitor borrowers. Therefore, related lending might have negative effects on financial development when rule of law is weak, but positive effects when the rule of law is strong. This hypothesis is motivated by two findings of the empirical literature on economic growth: indices of institutional quality (rule of law, property rights, and the like) are strongly correlated with growth (Knack and Keefer, 1995; Kaufman, Kraay, and Zaido-Lobaton, 1999); and there is a causal connection between financial development and growth (see, e.g., Levine, 2005). Our second hypothesis is that related lending has negative effects on financial development when there is a high level of deposit insurance, but has positive effects when the level of deposit insurance is low. That is, when depositors have strong incentives to police bank insiders, and therefore sanction them by removing their funds, the insiders may engage in related lending, but they do not use it as a mechanism to tunnel. Rather, they use it to overcome information asymmetries and monitor borrowers. This hypothesis is motivated by a growing literature that suggests that generous deposit insurance lessens depositors incentives to monitor the activities of their banks, resulting in greater banking sector instability and slower financial sector development (Demirgüç- Kunt and Kane, 2002; Demirgüç-Kunt and Detragiache, 2002; Demirgüç-Kunt and Huizinga, 2004). Our third hypothesis is that related lending has negative effects on financial development when the institutions that enforce good corporate governance are weak, but has positive effects when those institutions are strong. When minority shareholders 6

7 cannot police bank insiders, or when bank insiders have weak incentives to police each other, bank insiders have strong incentives to divert the bank s resources to their own firms particularly during a crisis that threatens the existence of those firms. When the institutions of corporate governance are strong, however, insiders may engage in related lending, but they cannot use it as a mechanism to tunnel. Rather, they use it to overcome information asymmetries and monitor borrowers. 2 It follows that related lending will have particularly positive effects when the rule of law is strong, when there are binding limits on deposit insurance, and when the institutions of corporate governance create strong incentives for bank insiders to monitor each other. In point of fact, all three features were present in the case that is often cited as strong evidence for the positive view of related lending: New England during the nineteenth century. There was no government-instituted deposit insurance scheme in New England except for Vermont ( ). 3 Banks, instead, created private payment and mutual coinsurance systems (e.g., the Suffolk system) in which member banks closely monitored one another to prevent moral hazard problems (Calomiris and Kahn 1996; Kroszner 2000). It has also been shown that related lending was successful in New England mainly because bank directors did not want to sacrifice their reputation and future access to capital (Lamoreaux 1994) and that bank insiders closely monitored one another to prevent opportunistic behavior (Meissner 2005). 2 This hypothesis is motivated in part by research by Maurer and Haber (forthcoming) on related lending in early twentieth century Mexico, in which they found that related lending did not cause tunneling or credit misallocation, even though rule of law was weak and Mexico went through the kind of externally-generated financial crisis that is usually thought to cause bank insiders in LDCs to loot their own banks. What prevented directors from doing so were institutions that provided strong incentives for directors to monitor one another. 3 Five other states, New York ( ), Indiana ( ), Michigan ( ), Ohio ( ), and Iowa ( ), adopted deposit insurance in the 19th century (Calomiris 1990). 7

8 It also follows that related lending might have particularly pernicious effects when the rule of law is weak, when there is unlimited deposit insurance, and when the institutions of corporate governance create weak incentives for bank insiders to monitor each other. In fact, all three of these features were present in the case that is often cited as strong evidence for the view that related lending is a manifestation of looting: Mexico from 1995 to 1998 (La Porta, López-de-Silanes, and Zamarripa, 2003). The Mexican government enforced the law selectively, and the judicial system was, by most accounts, corrupt. Depositors did not monitor directors because they were protected by unlimited deposit insurance (Martinez Peria and Schmukler, 2001). Bank directors did not monitor one another because they had very little of their own capital at risk: during the 1991 privatization, bank insiders borrowed much of the capital to purchase the banks, some of it from the very same banks they were buying, with the collateral being shares in the banks (Mackey 1999). In addition, as of 1995 bank depositors, bank debtors, and bank stockholders were protected by a taxpayer-financed bailout of the banking system (Haber 2005). In short, given the institutional environment in Mexico, it might have been expected that related lending would become a vehicle for insiders to loot their own banks. 3. Data and Methodology (Cross-Country Regression) Our basic empirical strategy is to relate financial development to regulatory restriction on related lending while controlling for macroeconomic factors and institutional features that also might affect financial development. In order to carry this out, we assemble a data set on financial system development, the extent of related lending, the strength of the rule of law, and the intensity of depositor monitoring for 76 countries. We then employ this data set to estimate regressions designed to determine 8

9 whether related lending is, on average, positively or negatively associated with financial development, and whether the relationship between related lending and financial development depends on the strength of rule of law and depositor monitoring. In this section, we first describe how we measure our key variables (financial development, related lending, rule of law, and depositor monitoring) and then present our regression model Measurement and Data A. Financial System Development We measure the development of financial systems using a standard metric in the literature: the ratio of private credit to GDP (see Beck, Demirgüç-Kunt, and Levine, 2000). Because data for countries that were previously members of the Soviet Union are not available until 1990, our data set runs from 1990 to More formally, for country i we compute the average annual rate of growth of the indicator as: Credit Growth i T (Pr ivate Credit / GDP ) /(Pr ivate T Credit 1 / GDPit ) it it it 1 t= = 1 (1) T in our case is fourteen since 1991 is the first year for which a growth rate can be computed. Credit Growth is used as a dependent variable that captures financial development. The data on private credit and GDP are obtained from IMF s International Financial Statistics. B. Related Lending Because regulatory authorities define related lending in different ways, a key challenge is measuring its extent across countries. It seems highly likely, however, that related lending is more prevalent in countries where regulators are more tolerant of cross- 9

10 ownership between banks and non-financial firms and where restrictions on the ownership of bank capital by related parties or a single owner are less binding. We therefore construct a proxy for the prevalence of related lending drawing on the database of bank regulation and supervision created by Barth, Caprio, and Levine (hereafter BCL, 2000, 2003). 4 Box 1 reports the questions that underlie our proxy. The first two questions refer to the ownership structure of banks (the maximum share that a single entity or individual may own, and the maximum that a related party may own). The second two refer to the degree to which banks are restricted from owning non-financial firms, and vice versa. Survey responses come from the regulatory and supervisory authorities in each country. Box 1. Questions on Relatedness (from Barth, Caprio, and Levine database of Bank Regulation and Supervision 2000, 2003) 1. What is the maximum percentage of bank capital that can be owned by a single owner [0-100]? 2. What is the maximum percentage of a bank s capital that can be owned by a related party [0-100]? 3. What is the level of regulatory restrictiveness for non-financial firms ownership of a bank [0 = Prohibited; 33.3 = Restricted; 66.7 =Permitted, with prior authorization or approval; 100=Unrestricted]? 4. What is the level of regulatory restrictiveness for bank ownership of nonfinancial firms [0 = Prohibited; 33.3 = Restricted to some figure below 100 percent; 66.7 = Permitted to own 100 percent of equity, but ownership is limited based on the bank s equity capital; 100 = Unrestricted, a bank may own 100 percent of the equity of any nonfinancial firm]? We rescale the qualitative responses to questions 3 and 4 on ownership interlinkages with non-financial firms (prohibited, restricted, permitted subject to certain 4 See Barth, Caprio, and Levine (2001) for a description of the database. The survey is available at: pagepk: ~pipk: ~thesitepk:469382,00.html 10

11 stipulations, and unrestricted) so that they conform to the responses from the first two questions (scaled from 0 to 100). We assign prohibited a value of 0, restricted a value of 33.3, permitted a value of 66.7, and unrestricted a value of 100. Our index of related lending, Related Lending, is the simple average of the responses to the four questions. 5 As Box 2 indicates, some of the responses to the questions are significantly correlated, especially questions 1 and 2 (the percentage ownership limits of banks for individual entities and related parties). To a lesser extent, responses to those two questions are also correlated with the responses to question 3 (limits on non-financial firms ownership of banks). By contrast, the responses to question 4 (limits on bank ownership of non-financial firms) are not significantly correlated with the responses to any of the other questions. Despite some significant correlations, the overall pattern suggests that each question provides a separate source of information. Box 2. Pairwise Correlations for Components of Connectedness Index Single Owner Related Party Non-Financial Firm Bank Ownership of Limit Limit Ownership of Banks Non-financial Firms Single Owner 1 Limit (n=76) Related Party *** 1 Limit (n=76) (n=76) Non-Financial Firm *** *** 1 Ownership of Banks (n=76) (n=76) (n=76) Bank Ownership of Non-financial Firms (n=76) (n=76) (n=76) (n=76) 5 The BCL questions in Box 1 were asked in and again in 2002, and thus we have two observations for our index of related lending for each country. We compute the index for each country using the responses and assign that value to all observations prior to For observations after 2002, we use the responses to the 2002 survey. We then take the average of those two observations for each country (weighted by the number of pre- and post-2002 observations), which we use in the regressions. As a practical matter, there was not much variance in responses from the survey to the 2002 survey. Thus, our results are not sensitive to the weighting scheme. For example, the simple average of the responses from the two datasets generates an index that is correlated at the.98 level with the one we use. 11

12 C. Rule of Law In order to measure the strength of the rule of law we use the index developed by Kaufman, Kraay, and Zaido-Lobaton (1999, hereafter KKZ). 6 The KKZ data are now available for 1998, 2000, 2002, and We use the index values from 2004 (and denote it by Rule of Law) because that year offers the widest country coverage, although we note that our results are not sensitive to the index year chosen, as the KKZ rule of law index does not vary dramatically over time. 8 D. Depositor Monitoring We use the level of deposit insurance to measure the intensity of depositor monitoring. More specifically, we compute, for each country i, the inverse of the ratio of deposit insurance coverage to per capita GDP: 1 Depositor Monitoring i = (2) ( Coverage Limit / GDP Per Capita ) The coverage ratios are for 2000, and are taken from a database developed at the World Bank (Demirgüç-Kunt and Sobaci, 2001). We use the inverse of the deposit coverage ratios in our index so as to allow this variable to capture potential positive effects of depositor monitoring, and to be consistent with our treatment of the other sources of monitoring and enforcement, such as rule of law. One potential issue with the construction of Depositor Monitoring is the treatment of countries without explicit deposit insurance. If we interpret the lack of an explicit deposit insurance program literally, we might assume that such countries have zero 6 In the regressions that follow, we add 2.5 to the KKZ index in order to create an index that runs from zero to five. 7 KKZ data set and its description are available at 8 The correlation between the 2004 data and those from other years runs from.91 to.98. i i 12

13 coverage. Governments, however, often end up covering depositors via nationalization and/or forced mergers and acquisitions, even when the country has no explicit deposit insurance program, which suggests that the anticipated coverage is not zero. We assume that the anticipated coverage is not zero and yet small for these countries, and set the value of Depositor Monitoring for those non-deposit insurance countries to the highest value from the countries with explicit insurance (i.e., the lowest coverage limit relative to GDP per capita). One might also argue that the level of anticipated coverage in implicit insurance can be quite large if governments face a credibility problem without coverage limits that are legally binding. 9 In a robustness check, we abandon our assumption that no deposit insurance implies low levels of insurance. Instead, we assume that the absence of an explicit insurance program implies a very generous level of deposit insurance, and construct Depositor Monitoring for no-deposit insurance countries as if they provide the same level of coverage as those countries that have the most generous explicit deposit insurance programs in the sample. To foreshadow our results a bit, no matter the method with which we treat countries without explicit deposit insurance, the interaction between depositor monitoring and our proxy for related lending is never associated with greater financial development. However, when we treat implicit coverage as being most generous, we do find that our index of depositor monitoring is positively associated with credit growth, in line with the aforementioned findings in the literature. 9 In this view, the absence of explicit insurance actually increases moral hazard (e.g., Gropp and Vesala 2004). 13

14 E. Other Control Variables We include the initial level of private credit to GDP (i.e, in 1990) in our regressions to control for the possibility that countries with better developed banking sectors tend to experience slower credit growth. These conditional convergence effects were first made known in the context of cross-country growth regressions, but have also proved important in financial development regressions (e.g., Detragiache, Gupta, and Tressel 2005). We also include the growth rate of GDP per capita and the inflation rate as macroeconomic controls. Both are averaged over the full sample period. We expect private credit growth to be faster in countries with low inflation and high growth. The data on macroeconomic controls are from the World Development Indicators database maintained by the World Bank. Correlation coefficients between these variables and those measuring financial development, depositor monitoring incentives, and the rule of law index are reported in Table 2. Neither the correlation between credit growth and related lending nor the correlation between our index of depositor monitoring and related lending is significant. Similarly, the correlations between the related lending index and our control variables are also insignificant, except for the negative correlation with GDP growth. We will deal with the potential endogeneity of related lending in multiple ways in the section on robustness checks. Before we present our base models, however, we note that the index is not related to measures of financial or institutional development in any straightforward way Regression Model Our baseline regression is: 14

15 Credit Growth i = β 0 + β 1 Related Lending i + β 2 Rule of Law i + β 3 Depositor Monitoring i + β 4 Initital Credit i + β 5 Macro Controls i + ε i (3) where Credit Growth i is the rate of growth in the ratio of credit to the private sector to GDP in country i from 1990 to Related Lending is the related lending index, Rule of Law is the KKZ rule of law index, Depositor Monitoring is the inverse of insurance coverage-to-gdp ratio, Initial Credit is the ratio of private credit to GDP in 1990, and Macro Controls are the two macroeconomic variables (inflation and GDP growth) described above. 10 In order to examine whether the effects of related lending depend on the institutional and regulatory environment, we add the interactions between Related Lending and Rule of Law and Depositor Monitoring as follows: Credit Growth i = β 0 + β 1 Related Lending i + β 2 Rule of Law i + β 3 Depositor Monitoring i + β 4 (Related Lending i )*(Rule of Law i ) + β 2 (Related Lending i )*(Depositor Monitoring i ) + β 4 Initital Credit i + β 5 Macro Controls i + ε i (4) In this specification, positive coefficients for (Related Lending)*(Rule of Law) and (Related Lending)*(Depositor Monitoring) would suggest that rule of law and limitations 10 The regressions that we present are estimated via OLS with White s Heteroskedasticity-consistent standard errors. We also estimated those models using least median squares (or quantile) regression. Though White s standard errors address heteroskedasticity, estimated coefficients could still be heavily influenced by outliers. Such outliers have less influence on the coefficients in quantile regressions. We also ran simple OLS models and STATA s robust regressions and found similar results. The results for the simple OLS, least median squares, and robust regressions are available from the authors upon request. 15

16 on deposit insurance coverage enable related lending to have a positive effect on financial development. 4. Regression Results The base results appear in Table 1. In the first column, we include only the related lending index and the control variables (initial private credit/gdp, inflation and real GDP growth). The coefficient for Related Lending is insignificant indicating that on average related lending is not strongly associated with private credit growth. Even when we add Rule of Law to the regressions in column 2, thus providing some institutional context, the coefficient on Related Lending remains insignificant. These results are not consistent with either the looting view or information view of related lending. When we introduce the interaction between Rule of Law and Related Lending in specification 3, however, we do find a relationship between related lending and private credit growth. The coefficient for Related Lending becomes negative and highly significant; that for the interaction becomes positive and significant. This pattern suggests that the pernicious effects of related lending are evident only when the rule of law is weak. The regressions indicate that the countries with the fastest rates of growth of private credit to GDP are those which have both strong rule of law and a high level of related lending. Specification 3 implies that for countries that score above 3.29 on the Rule of Law index, related lending is positively associated with private credit growth. For reference, the countries with scores closest to 3.29 are Italy (3.24), Netherlands Antilles (3.25), Israel (3.27), Monaco (3.27), Qatar (3.29), Taiwan (3.33), and Cyprus (3.35). The mean for the full KKZ sample is

17 On the right-hand side of Table 1 are specifications that include Depositor Monitoring (models 4-8). Whether Depositor Monitoring appears alone (models 4 and 6) or is interacted with our index of related lending (models 5 and 7), it is not significantly associated with private credit growth. The results (or non-results) are not due to the absence of control for Rule of Law as indicated in columns 6, 7, and 8. Nor is the monitoring variable significant when it is interacted with the rule of law (column 8). The lack of significance on the monitoring variable and its interactions with other variables indicates that depositors might not be effective monitors of looting by bank insiders. However, related lending tends to promote credit growth in countries with strong rule of law even when we control for Depositor Monitoring (Columns 6 and 7). That is, the net effect of the related lending coefficient and its interaction with rule of law is positive and significant for countries with high values for rule of law. When rule of law is weak but related lending is permitted, however, private credit grows more slowly than in countries that do not permit related lending. When we construct Depositor Monitoring based on the assumption that countries without explicit insurance provide the same level of coverage as those countries that have the most generous explicit deposit insurance programs, the results mirror those we obtained when we assumed that the absence of deposit insurance implied low levels of coverage with one exception. 11 There is a significant positive association between depositor monitoring and private credit growth in line with previous findings in the 11 Because countries that switched from implicit to explicit insurance during the sample period complicate the analysis, we restrict ourselves to those countries that had an explicit scheme prior to 1990 and those that never adopted explicit insurance. When we restrict the sample in the same way and re-run the base regressions in Table 1, the results for Related Lending and its interaction with Rule of Law are qualitatively unchanged. 17

18 literature. Our findings about related lending still hold, and thus we do not present these results in the table. 5. Governance of Related Lending in Uganda In this section, instead of relying on cross-country analysis, we make use of unique microeconomic bank-level data on related lending, bank ownership and performance to examine the importance of governance. 12 Our data set, which is drawn from Uganda from 1996 to 2005, provides information about the total amount of loans to related parties. Such information is usually not available at the bank level, especially in developing countries, and thus our data set provides a unique opportunity to examine the relationships between related lending and bank performance. 13 Our data set also contains information about the ownership of banks. Since a growing body of research from developing countries indicates that foreign-owned banks are better governed than 12 We also tried estimating whether a measure of corporate governance (used in La Porta, López-de- Silanes, Shleifer, and Vishny 1998; Djankov, La Porta, López-de-Silanes, and Shleifer 2005) affected the relation between financial development and related lending just as rule of law seems to. The results, however, turn out to be fragile. This might be because the indices developed by La Porta, López-de- Silanes, Shleifer, and Vishny (1998) and Djankov, La Porta, López-de-Silanes, and Shleifer (2005) are imperfect measures of corporate governance. See Spamann (2006) for a critique of these indices. Alternatively, these indices might not accurately capture the quality of governance in banking sector as well as that in non-financial firms since complex banking regulations and laws often affect the governance of banks in each country. Another possibility is that there might be important variation in corporate governance within each country which is not well-captured by the indices. 13 Thailand is another important exception. One problem with Thailand as a candidate for case study is that the Thai government started disclosing information about related lending only after the banking crisis of which led to the nationalization of distressed banks. Since the Thai government has been quite reluctant to divest these shares, many of the Thai domestic banks remain under government-control, which makes it difficult to distinguish traditional related lending from lending arising from political motivations and soft budget constraints (see, for example, Sapienza 2004). In a companion paper, however, we find that bank performance is poorer in government-owned banks than private banks, and more importantly, that bank performance is negatively correlated with the share of loans to related parties among state-owned banks, but not among privately owned banks in Thailand. These results indicate that related lending at government-owned banks, but not at private banks, might be a form of looting in Thailand. 18

19 domestic banks, we use foreign ownership as a proxy for the quality of corporate governance Background on Uganda BCL do not provide regulatory information for Uganda, and thus we lack a value for our index of related lending. We do, however, know that related lending was a serious enough concern to compel the Ugandan authorities to monitor it on a quarterly basis for each of the country s banks. At the same time, Uganda scores on the low end of the scale of the rule of law (1.7 relative to a mean of 2.5). 15 Additionally, Uganda s explicit deposit insurance was among the most generous in the world in the mid-1990s. Its coverage level (Shs 3,000,000) was 12.7 times larger than nominal GDP per capita in Of the 990 country-year observations for which Demirgüç-Kunt et al. (2005) provide an estimate, that figure ranks in the top seven percent. Only ten other countries with explicit deposit insurance produced a figure as large in any year, and more than half of those high figures were for years prior to The Ugandan coverage level was times the per capita level of bank deposits, the seventh highest country-year observation for that variable in the Demirgüç-Kunt et al. data set, which ranks well within the top 1 percent. Uganda was also in the midst of a banking crisis during the period under analysis. Caprio and Klingebiel (2003) classify Uganda from 1994 to 2003 as experiencing a systemic banking crisis, which they define as much or all of banking capital being exhausted. In fact, between 1994 and 1998, half of the banking system sector faced 14 This may be because foreign bank directors insist on better loan screening, because foreign directors are less likely than domestic directors to have kinship ties with the directors of non-financial enterprises, or because the mix of foreign and domestic directors makes it more difficult for directors to collude against minority shareholders and depositors (Berger, Klapper, and Udell, 2001; Clarke, Cull, D Amato, Molinari, 1999; Clarke, Cull, Martinez Peria, and Sanchez, 2003, 2004; Haber and Musacchio, 2005). 15 By African standards, however, Uganda s score is not low. 19

20 problems of insolvency. The years that followed saw continued contraction of the banking system. Two banks were closed in 1998, and another two in 1999 (Brownbridge, 2002; Kasekende, 2001). Even as late as 2002, according to Caprio and Klingebiel, the banking system was not fully solvent: they report that one small bank was intervened and two others were experiencing difficulties. In Uganda, we have data from foreign-owned banks, domestically-owned banks, and one state-owned bank. The state-owned bank, Uganda Commercial Bank (UCB), was also facing difficulties. It was recapitalized and privatized in 1998, only to require intervention and re-privatization to more competent private owners, the Stanbic Bank of South Africa, in 2000 (Brownbridge, 2002; Kasekende, 2004). For most of the period under study, therefore, UCB was under state control. The foreign banks operating today are not newcomers to the Ugandan markets: one can view their operations in recent years as steady state. Five of the foreign banks operating today (the so-called ex-colonial banks) trace their origin to the 1960s. Although their activities were restricted and many of their assets were nationalized as part of Idi Amin s indigenization policies in the 1970s, they resumed operations in the 1980s. Beginning in 1991, bank entry restrictions were relaxed as part of a broad financial sector liberalization program (Kasekende and Sebudde, 2002). There were eleven foreign banks and nine domestic banks operating in Uganda during our period of study. 5.2 Regression Results Our basic empirical strategy in this case study is simple. We use the balance sheet data on Ugandan banks to regress profitability (return of assets, hereafter referred to 20

21 as ROA) and portfolio quality (as proxied by the ratio of non-performing loans to total loans, hereafter referred to as NPL) on bank size, ownership type, and the share of lending to parties related to bank owners: 16 Bank Performance it = β 0 + β 1 Size it + β 2 Ownership it + β 3 Related Lending it + β 4 (Related Lending it )(Ownership it ) + ε it (3) where performance for bank i in period t is either ROA or NPL; Size is total assets; Ownership is a vector of dummies for foreign and private domestic ownership; 17 and Related Lending is the share of lending to bank insiders. We expect that for ownership types with weaker internal governance (domestic banks), related lending will be negatively associated with ROA and positively associated with NPL. Appendix 1 shows the summary statistics of relevant variables used in our empirical analyses of Uganda: the share of related loans to total loans, NPL, and ROA. Compared to the widely cited Mexican case, the amount of loans to related borrowers is small in Uganda (Appendix 1, Table A1); according to La Porta, Lopez-de-Silanes, and Zamarripa (2003), the Mexican banks commonly allocated as much as 10-30% of total loans to related borrowers. Note, however, that the standard deviation of related lending is around 10 percent for both domestic and foreign banks, suggesting that some of these banks directed a sizable share of their loans to related parties. More importantly, such variation in related lending under both types of ownership makes this data well-suited for analyzing the differential effects of related lending on bank performance under different 16 The data are taken from the Central Bank of Uganda. 17 The base group is a lone state-owned bank. 21

22 ownership. 18 Some of the variables we employ in our analysis are strongly correlated with one another in ways that we would expect (Appendix 1, Table A2). NPL is negatively correlated with ROA, whereas it is positively correlated with the share of related lending. That is, the banks with more NPL are less profitable and provide more loans to related borrowers. Table 4 presents the regression results. We present results estimated via quantile regression (least median squares) and OLS with White s robust standard errors. 19 We also clustered standard errors by bank, and these specifications produced similar results. (See Appendix 2, models 3, 4, 7, and 8). Finally, we included bank fixed effects in the ROA and NPL regressions, and obtained qualitatively similar results (see Appendix 2, models 1, 2, 5, and 6). 20 Specifications 1, 2, and 6 of Table 4 indicate that banks with higher shares of lending to insiders had lower profitability and higher shares of non-performing loans than other banks. In specifications 3 and 4 for ROA and 7 and 8 for NPL share, we interact the share of lending to bank insiders with ownership type to test whether the effects of related lending are similar for foreign and domestically owned banks. Because the lone state bank, UCB, had essentially no lending to insiders (see Appendix 1, Table A1), we need only interact insider lending with the domestic and foreign ownership dummies. The coefficient on the domestic ownership interaction with insider lending is negative and significant in the ROA regressions, positive and significant in the NPL regressions. The 18 One might wonder whether foreign banks in Uganda are engaged in related lending at all. It turns out that foreign banks, on average, provide the most related lending (Appendix 1, Table A1). 19 We include quarterly dummies to capture unobserved economy wide shocks although the coefficients on these dummy variables are not reported to keep the table uncluttered. 20 We also estimated the regression equation for NPL using tobit because the dependent variable is censored at zero. The results are qualitatively similar and thus not reported to conserve space. 22

23 foreign ownership interaction term is not significant, except in specification 7 where it indicates that foreign banks with higher shares of insider lending tended to have fewer non-performing loans. We can also draw inferences about the sectoral composition of related lending in foreign and domestically-owned banks. In specification 10 there is a positive association between related lending and agricultural credit, controlling for bank ownership type. When we add the interactions between bank ownership type and related lending in specifications 11 and 12, the coefficient for lending to insiders for domestic banks is positive and highly significant. By contrast, there is no significant relationship between insider lending and agricultural lending for foreign-owned banks (i.e., insider share*foreign ownership is not significant). These results indicate that domestic banks that lent heavily to insiders also tended to allocate a relatively high share of their total credit to agriculture. Taken together, these results suggest that the pernicious effects of insider lending are attributable to domestic rather foreign owners, even though some foreign banks had substantial shares of their loan portfolios (10 to 50 percent) allocated to insiders. The results also suggest that within the group of domestically-owned banks, those that made the most related loans had the poorest performance, and many of those loans were (ostensibly) to support agricultural activities. In short, our results indicate that corporate governance matters: internal control mechanisms within foreign banks meant that related loans were not associated with poorer bank performance. Indeed, foreign banks that had 23

24 high proportions of related loans also tended to have lower rates of non-performing loans Robustness Checks Results from the previous section indicate that foreign ownership of banks can help curtail abuses associated with related lending in developing countries. However, curtailing abuses is not the same as promoting private credit growth, and indeed the results from Uganda indicate that foreign banks have not used related lending to boost substantially the level of private credit relative to GDP. To enjoy the benefits of related lending as in the New England experience likely requires that domestic banks lend to insiders in the context of formal institutions and/or reputational mechanisms that can capably monitor those activities. Our cross-country results provide a rough indication that, when these institutions are in place, related lending can contribute to growth in private credit. To this point, our proxy for institutional development has been adherence to the rule of law, which we admit is a bit of a broad, amorphous construct. In this section therefore we push our main result further to provide more specifics about the policies, institutions, and contexts that increase the likelihood that related lending makes a positive contribution to financial development. In so doing, we also demonstrate the robustness of our main result and offer additional evidence that the relationship that we have found is a causal one. 21 One might be tempted to argue that the poorer performance of related loans by domestically-owned banks is offset by a faster rate of credit expansion. We therefore estimated regressions, similar to those reported in Table 4, in which the rate of growth of credit was the dependent variable. Those regressions, not reported here, lend no support to the argument. To the degree that they pick up anything, they indicate that domestic banks that made more related loans actually expanded credit more slowly than all other bank types. 24

25 6a. Replacing Rule of Law with More Exogenous Proxies Although the correlations in Table 2 indicate that there is no simple relationship between the rule of law and related lending, a skeptical reader might still maintain that there is an endogenous relationship among these variables: rule of law and related lending are both legal institutions, which are jointly determined by some more fundamental, underlying factor. That same fundamental factor might also be causing the rate of growth of financial development. We therefore replace our measure of institutional quality the rule of law with variables that capture fundamental exogenous factors that drive institutional outcomes. Work by Easterly, Ritzen, and Woolcock (2006) shows that ethnic fractionalization undermines social cohesion, which undermines the rule of law, which in turn affects economic growth. 22 Obviously, any institutional measure is going to contain some component that is endogenous to politics and political institutions. One would be hard pressed, however, to argue that ethnic fractionalization is caused by political factors. We therefore substitute ethnic fractionalization for the rule of law variable in Table 3, models 1 and 2. Related lending is positive and significant in both specifications, while related lending interacted with ethnic fractionalization is negative and significant. The regressions indicate that at low levels of social cohesion, the impact of related lending on credit growth is negative, but that at high levels of social cohesion related lending has a positive impact on credit growth. 23 These findings are therefore consistent with those we obtained from the rule of law regressions. They also provide additional 22 They also show that these measures of social cohesion pass over-identification tests for excludability in growth regressions. 23 The measure of ethnic fractionalization used by Easterly et al. ranges from 0 to 1 with higher values indicating more fractionalization. In our sample, the mean value is 0.44 and the highest value is 0.93 for Uganda. The lowest values are for Japan (.01) and Korea (.002). 25

26 information about the type of society in which related lending is less likely to offer benefits. 6b. An Instrument for Related Lending Though ethnic fractionalization poses fewer endogeneity problems than rule of law, the robustness checks in the previous sub-section do not address the potential endogeneity of the related lending policies themselves. The difficulty, of course, is finding an appropriate instrument for those policies. We propose as a candidate the index of official supervisory powers constructed by BCL (2001). That index, which is described in detail in Appendix 2, is based on sixteen questions about the powers granted to supervisors in monitoring and disciplining banks. We view this as an indication of a society s general propensity to use official mechanisms to monitor market activities. To the extent that a society views its banking sector as being tightly regulated and supervised, fears about the negative effects of related lending might be diminished. We would therefore expect a positive relationship between our related lending index and the BCL index of official supervisory powers, and indeed the correlation between the two is 0.42, which is significant at the one percent level. When we use the official supervisory powers index as an instrumental variable in a two-stage least squares regression, the index of related lending is not significant (Table 3, column 3). However, when we restrict the sample to countries that began our period of study with relatively low levels of financial development (private credit/gdp < 0.7 in 1990), the index of related lending is negative and significant (column 4). This filter excludes eight industrialized countries and China from the active observation set Those industrialized countries are Switzerland, France, Japan, Malta, Netherlands, New Zealand, Singapore, and the United Kingdom. 26