MASTER THESIS FINANCIAL MANAGEMENT TILBURG UNIVERSITY RAOUL GEURTS S948067

MASTER THESIS FINANCIAL MANAGEMENT TILBURG UNIVERSITY RAOUL GEURTS S948067

Micro Finance Institutions: Does Investing in Latin America s MFI Assets Yield a Positive Value for Western Bond Investors? Tilburg University Faculty of Economics and Business Administration Department of Finance August 2010 Name: ANR Number: Name Supervisor: Raoul Geurts S948067 Dr. M.R.R. van Bremen 2

Table of Contents Section 1...4 1.1. Introduction...4 1.2. Problem Definition...5 1.3. Continuation of the Thesis...5 Section 2: MFI Performance...6 2.1. Measurement of MFI Performance and MFI Benchmark...6 2.2. Discussion and Results of MFI Performance...7 Section 3: Correlation of MFI Performance with Benchmark Performance...8 3.1. Relevance of the Correlation between MFI Performance and Benchmark...8 3.2. Results Correlation between MFI Performance and Benchmark Performance...8 3.3. Conclusion...9 Section 4: Differences among MFI Performance...10 4.1. Assessment of Comparison Groups...10 4.2. Differences in Performance...10 4.3. Results...10 Section 5: Hypotheses...11 Section 6: Methodology...12 6.1. Introduction...12 6.2. Statistical Tests used...13 6.2.1. Hypothesis 1 Testing: Differences between MFI Return and Bond Return?.13 6.2.2. Hypothesis 2 Testing: Relation between MFI Performance and Bond Performance...14 6.2.2.1. Correlation...14 6.2.2.2. Coefficient of Determination...14 6.2.3. Hypothesis 3 Testing: Difference in Return among MFIs?...15 6.3. Robustness Tests...16 6.3.1. Heteroskedasticity...16 6.3.2. Independency of Error Random Variables...17 Section 7: Data...18 Section 8: Empirical Results: Hypotheses Testing and Discussion...19 8.1. Hypothesis 1...19 8.2. Hypothesis 2...20 8.3. Hypothesis 3...22 Section 9: Summary, Conclusion and Discussion...25 Section 10: References...27 Appendix A: Mean and Variance Outputs for Hypothesis 1...29 Appendix B: Correlation Outputs for Hypothesis 2...30 Appendix C: Outputs for Hypothesis 3...32 Appendix D: Outputs for Heteroskedasticity Tests...35 Appendix E: Outputs for Durbin-Watson Test...37 Appendix F: Results Robustness Tests...40 3

INTRODUCTION Section 1 1.1. Introduction Rahman (1998) defines Micro Credit as the extension of small collateral-free institutional loans to jointly liable poor group members for their self employment and income generation. The basic idea of Micro Finance is rather simple. A financial institution provides loans (and financial services) to poor people who under the conventional banking system would not be able to request a loan. The loan is aimed to be invested in a small business projects which can generate income. With the income they create, borrowers are able to repay the loan and the accrued interest. By means of microfinance, borrowers should be able to fulfil their own and family s financial requirements and the community should be able to develop itself. Most MFIs started as Mission Driven NGOs. The main disadvantage of an NGO structure is that funds are limited, which can impede growth. For this reason many NGOs transformed themselves into Micro Finance Institutions (MFI) partly or entirely funded by private capital (Krauss and Walter 2009). A number of micro credit funds have been created for investors. Some of these funds main objective is socially responsible investing, whereby performance of the fund is not the only objective. In this thesis it is believed that attracting private capital to a large extent is only sustainable if the return for the investors is attractive. That is, in order to be sufficiently financed in the long run, MFIs need to compete with the mainstream capital markets. By this point of view it is necessary to investigate investors relevance, both for the sake of investors and the sustainability of MFIs. Most academic literature about Micro Finance Institutions has been published relatively recently. The vast majority of the academic literature was published from the end of the 1990s. Most literature focuses on the Micro Finance Institutions themselves and address subjects like borrowers characteristics 1, the role of social capital 2 or macro economic 1 Edgcomb et al. (2002); among others 2 Woolcock (2001); among others 4

INTRODUCTION consequences 3, among many other research fields. Among the vast academic literature which has been published about MFIs in the past 15 years, there is very few finance-related literature. The providers of capital have not been investigated for a long time. 1.2. Problem Definition A lot of literature has been published about the social role of microfinancing, but still there is not much literature about the relevance of investors to invest in Micro Finance Institutions, while this is an important issue for the sustainability of MFIs in the long run. Is it lucrative for investors to invest in MFIs? In order to reduce the scope, this research will focus on the Latin American market. The research question will be: Does investing in Latin America s MFI assets yield a positive value for western bond investors? 1.3. Continuation of the Thesis In order to assess whether investing in MFI assets creates value for western bond investors, the value creation should be divided into three elements. Firstly, the most straightforward element is performance. Performance of MFIs compared to the performance of mainstream capital market alternatives provides the first element of value creation for western investors. The second element is the correlation of MFI Performance to the investors existing portfolio. As will be explained in chapter 3, a low correlation provides, ceteris paribus, a higher value creation. The last element is the creation of subgroups. By this last element it is possible to assess not only if western bond investors should invest in MFI assets, but also in which MFI assets they should invest. In chapter 2 to 4 a literature review is provided. Each chapter of the literature review will elaborate one of the three elements of the value creation for western bond investors. Each element with the corresponding literature results in a hypothesis. These hypotheses are presented in chapter 5. Section 6 and 7 respectively describe the methodology and data to test these hypotheses. The results and the discussion of the results are in section 8. Finally, section 9 summarizes, concludes and discusses. 3 Ahlin and Lin (2006), Ahlin et al. (2009); among others 5

LITERATURE REVIEW Section 2: MFI Performance The first element of investors relevance to invest in MFIs is MFI Performance. Different measures for MFI performance and the benchmark are used in the literature; therefore the first paragraph describes how MFI performance and the benchmark are measured. The second paragraph provides the discussion and the results. 2.1. Measurement of MFI Performance and MFI Benchmark The literature proposed many different measurements of performance. Measuring performance can be achieved by measuring return in the broadest sense of the word and by measuring risk. As a matter of fact a high return and a low risk imply a high performance. The most common and useful measure to assess investors performance is the return on equity. This measure, however, is not that obvious to assess MFI performance since return on equity of MFIs can be highly biased (see next paragraph). The literature so far uses rather different performance measures. So far, there is no consensus in the literature which performance measures reflect MFI performance best. Krauss and Walter (2009) use the broadest set of performance measures, which is composed of both return and risk measures. The set of performance measures they use is composed of (1) the percentage change in Net Operating Income (NOI), (2) the Equity, (3) the Profit Margin, (4) the percentage change in Total Assets, (5) Loan Portfolio change and (6) Portfolio at Risk. Krauss and Walter (2009) do not look at the values of these performance measures, but correlate these measures to the benchmark portfolios and the domestic GDP. So, in spite of assessing the performance of MFIs for investors, Krauss and Walter (2009) just assess performance measures which are tested to have a correlation to their benchmark or not. This procedure will be elaborated in the next chapter. Gonzales (2007) roughly applied the same methodology to assess MFI performance. Also Gonzales (2007) assesses performance measures which he correlates to a benchmark. He only uses risk measures, while Krauss and Walter (2009) use both return and risk measures. The risk measures Gonzales (2007) uses are (1) Portfolio at Risk 90 days (2) Portfolio at Risk 30 days (3) Loan Loss Rate Ratio (4) Write off Ratio. Galema et al. (2008) use a different methodology to assess MFI performance. They assess the return on equity and the return on assets of MFIs and use the variance of the return on equity 6

LITERATURE REVIEW and the variance of the return of assets as risk measure. By using a mean-variance spanning test, Galema et al. (2008) determine whether MFIs can improve investors portfolios. About the benchmark of performance is more consensus. Krauss and Walter (2009) and Galema et al. (2008) both use the return of developed stock market indices and the return of emerging stock market indices. Moreover, Krauss and Walter (2009) use the return of Equities of Listed Emerging Market Indices and domestic GDP and Galema et al. (2008) use global and emerging bond returns. Gonzales (2007) only uses Gross Net Income as a benchmark. 2.2. Discussion and Results of MFI Performance Both Krauss and Walter (2009) and Gonzales (2007) do not assess performance directly. They determine some MFI performance measures and regress those on the returns of portfolios available to investors and GNI respectively. This procedure will be elaborated in the next chapter. Krauss and Walter s (2009) benchmark is incomplete. They use stock markets as benchmark portfolios, but the majority of the MFIs is not traded so that the structure of MFIs is more similar to bonds than to equity. Also Galema et al. s (2008) benchmark is incomplete since their benchmark (Gross National Income) does not reveal directly the investors relevance. Galema et al s. (2008) paper assesses MFI return on assets and MFI return on equity directly. The drawback of this paper, however, is that MFI return on equity could be highly biased since no market data of MFIs is available. The equity value used is the book value of equity. This book value tends to be different from the real or market value of equity. The difference could bias the results substantially. 7

LITERATURE REVIEW Section 3: Correlation of MFI Performance with Benchmark Performance The second element of investors relevance of investing in MFIs is the correlation between MFI performance and its benchmark. The reason why this correlation is relevant for investors is described in the first paragraph. So far, two papers have researched the correlation between MFI Performance and a benchmark. The results of these two papers are elaborated in the second paragraph. Finally, the third paragraph provides the conclusion. 3.1. Relevance of the Correlation between MFI Performance and Benchmark The correlation between MFI performance and the benchmark indicates the potential diversification benefits. A low correlation between MFI performance and its benchmark implies diversification benefits since the MFI performance and its benchmark s performance tend to move independently. A negative correlation implies that the performance of MFIs and its benchmark tend to move in the opposite direction. The next paragraph describes two papers which investigated the correlation between MFI performance and the benchmark. Both papers suggest that a low correlation between MFI performance and its benchmark implies diversification benefits for investors. Based on Krauss and Walter (2009), also Dieckmann (2008), senior economist at the Deutsche Bank, mentions the potential low correlation between MFI performance and its benchmark as an advantage for investors. As an explanation for the low correlation he mentions some characteristics of MFIs which are (1) low integration in the formal economy (2) low affection by currency fluctuations due to the reliance on domestic products and services (3) short maturity of the loans (4) frequent repayment of the loan instalments. He notes that the low correlation might rise over time, when the MFI sector becomes more integrated into the mainstream financial sector. 3.2. Results Correlation between MFI Performance and Benchmark Performance Krauss and Walter (2009) regress several MFI performance measures on the returns of U.S. Stock Indices, Equities of Listed Emerging Market Institutions and Equities of Listed Emerging Market Banks. They find no statistically significant relationship between the MFI performance measures and global market movements. However, they find a weak relationship between MFI performance and Emerging Market Institutions performance. Finally, most MFI 8

LITERATURE REVIEW performance measures show a statistically significant correlation with domestic GDP. Krauss and Walter (2009) conclude that MFIs may have positive diversification value to international portfolio investors, a less strong positive diversification effect to emerging market domestic investors and a no positive diversification effect on domestic investors. Gonzales (2007) main objective is to see whether MFI risk is related to changes in GNI per capita. He argues that no relation between MFI risk and changes in GNI per capita implies that microfinance portfolio have resilience to macroeconomic shocks. Gozales (2007) uses four MFI risk measures, but only finds a statistical relationship between growth in Gross Net Income and Portfolio at Risk 30 days. Since his research suggests no correlation between most MFI risk measures and Gross Net Income, Gonzales (2007) concludes that microfinance has a high resilience to economic shocks. In other words, if MFI risk increases, the Gross Net Income tends to stays the same. The other way around, if Gross Net Income decreases (which usually goes along with an economic crisis), MFI risk measures do not change. He finds, however, a correlation between Portfolio at Risk 30 days and Gross Net Income. Gonzales (2007) concludes that economic crises increase the portfolio at risk on the short run, but do eventually not lead to default. 3.3. Conclusion So far, two academic papers investigated the correlation of MFI performance with benchmark portfolios. Krauss and Walter (2009) conclude that investing in MFIs yields diversification value for international investors, but no diversification value to domestic investors. Gonzales (2007) concludes that MFI risk and economic crises are not related, thereby suggesting that MFIs are to a certain extent detached from macro economic shocks. Since usually firms performance is not detached from macro economic shocks, MFIs could be an addition to existing investors portfolios. 9

LITERATURE REVIEW Section 4: Differences among MFI Performance The third element of investors relevance to invest in MFIs is the creation and comparison of subgroups. This section elaborates the cross-sectional differences in MFI Performance. The first paragraph describes the assessment of comparison groups, the second paragraph describes the cross-sectional differences in MFI Performance. The last paragraph provides the conclusion. 4.1. Assessment of Comparison Groups So far, only Galema et al. s (2008) paper has distinguished and compared different types of MFIs for the determination of performance for investors. Galema et al. (2008) distinguish six different regions (i.e. Africa, East Asia & the Pacific, Eastern Europe & Central Asia, Latin America, Middle East & North Africa and South Asia) and five different industries (i.e. Bank, Cooperative/Credit Union, Non-bank Financial Institution, Non-Profit and rural bank). 4.2. Differences in Performance Galema et al. (2008) find that MFIs in Latin America can improve investors portfolios in terms of return and variance. They do, however, not find this result for MFIs in Africa, the Middle-East and North Africa. Eastern Europe and Central Asia are found to improve bond portfolios and East Asia and the Pacific are found to improve equity portfolios. No clear result is found for South Asia. Galema et al. (2008) find that MFIs in the bank industry and the rural bank industry can improve the performance of both stock portfolios and bond portfolios for investors in terms of return and variance. For the other industries no clear patterns are found. 4.3. Results So far only one academic article deals with the comparison of MFI performance for investors. This article suggests that the performance vary substantially among regions and industries. As noted in chapter 2, the quality of the data of this paper should be questioned. The data could be highly biased. 10

HYPOTHESES Section 5: Hypotheses As noted in the problem definition in paragraph 1.2. The research question of this thesis will be: Does investing in Latin America s MFI assets yield a positive value for western bond investors? In paragraph 1.3. the potential value creation was divided into three elements. Each element was elaborated by a literature review. The first element of value creation for western bond investors is the performance of MFIs. In order to test whether MFIs add value in terms of performance, the first hypothesis will test whether there is a statistical difference between MFI Performance and the benchmark performance. (Hypothesis 1) 1. Micro Finance Institutions in Latin America have the same performance as bond portfolios The second element of value creation for western bond investors is the (non-positive) relation between the MFI performance and its benchmark. The second hypothesis will be as follows. (Hypothesis 2) 2. The return on Assets of Micro Finance Institutions is not related to the return of benchmark bond portfolios Finally, the third element of value creation is the creation of subgroups. As will be described in the methodology, the creation of subgroups is not possible due to a relatively small number of observations. Therefore, some countries that are expected to bias the results of hypothesis 1 and 2 are omitted from the sample. The remaining countries are notated as the less developed countries in Latin America. (Hypothesis 3) 3. MFIs in less developed countries in Latin America have the same return as their benchmark 11

METHODOLOGY Section 6: Methodology 6.1. Introduction Like the papers described in section 2 to 4, this paper investigates whether MFIs can improve investors portfolios. The use of correlations between MFI performance and benchmark portfolios, as Krauss and Walter (2009) do and the correlation between MFI risk measures and GNI, as Gonzales (2007) does, gives an idea whether MFIs can improve existing portfolios, but it does not confirm this hypothesis well. The methodology of Galema et al. (2008) seems to reveal better whether MFIs can improve existing portfolios, but the main problem is that the authors use accounting return on equity and the results could therefore be severely biased (see paragraph 2.2). As benchmark portfolio they take international developing markets stock index. Using international developing markets stock index in this thesis would not be precise, since the Latin America MFI market is investigated. Therefore, each MFI return on asset observation is compared with a bond portfolio from the same country and the same year. Although the accounting rate of return is not a perfect measure for firms performance, the literature does not provide a good alternative. Kapler (2000) is able to slightly improve the accounting rate of return for economic purposes. However, for his calculations detailed financial information is required, which cannot be obtained for MFIs. Many papers use the accounting rate of return on assets, due to an absence of alternatives. Although it is recognized that accounting ROA is not the perfect measure to assess firms performance, the literature accepts to use it. Stark (2004) argues that the firms Internal Rate of Return is the best measure for economic performance. This does, however, only hold under certain conditions. He concludes if the firm is not in a steady-state growth situation, it is difficult to apply cash flow pattern fitting methods in the absence of a full enough history of investment outflow data. If such is not available, it is possible that the Accounting Rate of Return is the only measure that it is possible to use. MFIs do not have a full history of investment outflow data and are certainly not in a steady-state growth situation, which suggests that, according to Stark (2004), the Accounting Rate of Return is the best measure for economic performance to use. 12

METHODOLOGY 6.2. Statistical Tests used This paragraph describes the statistical tests used to investigate the three hypotheses. Unless stated otherwise, all tests are based on a 95% confidence interval, which is equal to α = 0.05. 6.2.1. Hypothesis 1 Testing: Differences between MFI Return and Bond Return? Since the variances of MFI asset return and bond portfolio return are assumed to be different, an unequal variance T-test will be used to test whether the average asset return of MFIs and the return of bond portfolios is the same. The average returns of MFI assets and returns of bond portfolios will be compared with this statistical test. If the averages of MFI asset return and bond portfolio return are different, the difference between these averages is not equal to 0. If the averages are the same, the subtraction should be equal to 0. With the following procedure will be tested whether the means are statistically different. Testing Problem: H0: µ1 µ2 = 0 H1: µ1 µ2 0 Test Statistic: Rejection Region: G = (X1 X2) / ((S1^2)/n1) + (S2^2/n2)) G -t0.025;m OR G t0.025;m Where µ stands for the sample mean G stands for unequal variance t-test X1 stands for the mean of MFI asset return X2 stands for the mean of the benchmark bond portfolio S1 stands for the variance of the MFI asset return S2 stands for the variance of the bond portfolio return n1 stands for the sample size of MFI asset return observations n2 stands for the sample size of the benchmark bond portfolio return t stands for t-distribution m stands for the degrees of freedom, which is equal to min(n1,n2) 1 13

METHODOLOGY Note: n1 is always equal to n2 in this research, since each MFI asset return observation is linked to a benchmark return. 6.2.2. Hypothesis 2 Testing: Relation between MFI Performance and Bond Performance In order to test for the relation between MFI asset return and bond portfolio return (hypothesis 2), both the correlation and the coefficient of determination are taken into account. This subparagraph is divided into two parts. The first subparagraph describes how the correlation is tested, the second describes how the coefficient of determination is tested. 6.2.2.1. Correlation The first test will test whether the correlation coefficient is statistically smaller than or equal to 0. H0: The correlation coefficient between MFI return on assets and return on bond portfolios is equal to or smaller than 0 H1: The correlation coefficient between MFI return on assets and return on bond portfolios is bigger than 0 It should be noted that this hypothesis is not so strict in case it is rejected. That is, if the null hypothesis is rejected, but the correlation-value is low, it could still be concluded that the MFIs improve existing bond portfolios. 6.2.2.2. Coefficient of Determination The coefficient of determination (R-square) indicates which percentage of the variance in the MFI return can be explained by the return on the bond portfolio. As noted in paragraph 3.1, a weak relation between MFI return and bond portfolio return indicates diversification benefits and is therefore, ceteris paribus, desirable for most investors. Since the coefficient of determination indicates which percentage of the variation in the MFI return is explained by the bond return, Krauss and Walter (2009) argue that a low coefficient of determination would indicate a low exposure to the benchmark portfolio and vice versa. 14

METHODOLOGY Krauss and Walter (2009) find that Emerging Market Institutions (EMIs) and Emerging Market Commercial Banks (EMCBs) have an R-square up to 6% and 11.5% respectively with the MSCI World Index. (Morgan Stanley Capital International). These coefficients of determination will be taken as a benchmark for MFI asset return relation to bond portfolio return. The hypotheses for the coefficient of determination will therefore be: H0: The coefficient of determination is smaller than 0.10 H1: The coefficient of determination is equal to or larger than 0.10 If the null hypothesis is confirmed, that would imply that MFI returns have a lower exposure to the benchmark portfolio than the exposure of EMIs and EMCBs performance to their benchmark. The coefficient of determination of a simple regression like this is straightforward to calculate, since it is equal to the square of the correlation coefficient. Still it is useful to use the coefficient of determination next to the correlation coefficient, since it can be compared with Krauss and Walter s (2009) findings. As a matter of fact, a higher correlation implies a higher coefficient of determination. 6.2.3. Hypothesis 3 Testing: Difference in Return among MFIs? A one way ANOVA F-test is useful to test for differences in subgroups of a sample. This test is not used to test for the third hypothesis since the sample size of the subgroups would be too small to obtain reliable results. In order to test for differences in subgroups, without losing too many observations, the same test as for hypothesis 1 is used. By having a close look at the data, it turns out that the more developed economies of Latin America (Brazil, Argentina, Chile and Mexico) often have a strongly negative return on assets. It could be the case that these countries bias the results obtained for hypothesis 1 and 2. By repeating the tests for hypothesis 1 and 2, but without the observations of the more developed economies, it will be tested whether the returns of less developed economies of Latin America show a different relation to their benchmark. 15

METHODOLOGY 6.3. Robustness Tests This paragraph describes the robustness tests which are conducted to test for the quality of the statistical tests of hypothesis 1 and 3. The results of the robustness tests can be found in appendix F. 6.3.1. Heteroskedasticity In order to test for heteroskedasticity, the Breusch-Pagan test is conducted. This test is used to assess whether the variance of the error terms is dependent on the independent variable. In this research it is tested whether the variance of the error terms of return on MFI assets is dependent on the bond portfolio returns. An f-test will be conducted to test for this relation. The following formula is applied to test for heteroskedasticity. (Ei)^2 = β0 + β1 * Bond Returns + ε Testing problem: H0: Bond return does not explain the error terms H1: Bond return explains the error terms Test statistic: SPSS Output Rejection Region: Where: f F0.05,n1-1,n2-1 (Ei)^2 stands for the variance of error term of return on MFI assets f stands for an f-distribution n stands for the number of observations In case heteroskedasticity takes place, the results are less reliable and should therefore be considered with more caution and less confidence. 16

METHODOLOGY 6.3.2. Independency of Error Random Variables In order to test for the independency of the error random variables, the Durbin-Watson Test is conducted. This test reveals whether there is correlation within the sequence of residuals (first-order autocorrelation). Since it is not clear beforehand whether the potential autocorrelation is positive or negative, a two-sided Durbin Watson test will be used. The following procedure is applied: Testing problem: H0: There is no first-order autocorrelation H1: There is first-order autocorrelation Test statistic: Value obtained from SPSS Rejection Region: d dα/2,l OR d 4 - d α/2,l Conclude H1 d α/2,l < d < d α/2,u OR 4 - d α/2,u < d < 4 - d α/2,l Conclude Inconclusive d α/2,u d 4 - d α/2,u Conclude H0 Where: L stands for lower bound U stands for upper bound By filling out the α, the lower and upper bound value can be found in a Durbin-Watson Table. In case autocorrelation takes pace, there is no clear way to improve the research. In order to avoid first order correlation, not the bond portfolio prices but the bond portfolio returns are used in this research. If the Durbin-Watson test would indicate first order autocorrelation, that would imply that the results in section 8 are less trustable. The results of the tests should therefore be considered with more caution and less confidence. 17

DATA Section 7: Data For this research, only data from Latin America is used. Data for the MFIs is extracted from the MIX Database. Countries are selected for which both data on MFI asset return and data on benchmark bond portfolios is available. Data for the benchmark portfolios is extracted from the Datastream database. This database provides returns on bond portfolios. Only national bond portfolios are used. That is, only bond portfolios that contain bonds from one particular country are used, in order to assess a more precise benchmark. In case more benchmark bond portfolios are available for a country a weighted average of these bond portfolios is taken. Only bond portfolios quoted in US dollars are selected, since the MFI asset return is also quoted in dollars. Data are selected from 2005 to 2009. Since MFIs arose relatively recently, there are not enough observations of MFI return on assets before 2005. Finally, only bond portfolios that are easily assessable for western investors are selected. That is, bond portfolios that can be easily traded from Europe and the United States. Benchmark bond portfolios that can only be traded in Latin America are not taken into consideration. For each MFI asset return a corresponding bond portfolio return from the same year is selected. All MFIs that both come from the same year and the same country are attributed the same benchmark return. It should be noted that participation to the MIX Database is voluntary. This voluntary participation could bias the results, since bad performing MFIs can choose not to participate. Voluntary participation, however, does not necessarily bias the results, since investors can choose to invest only in MFIs that participate in the MIX database. Gonzales (2007) states that the MFIs in the MIX Database are a random sample of the best MFIs in the world ( ), but definitely not a random sample of all MFIs. This research could therefore be seen as a research in MFIs that participated in the MIX Database. The countries that fulfil all requirements for the data are Argentina, Brazil, Chile, Colombia, Costa Rica, Ecuador, Guatemala, Mexico, Panama, Peru and Venezuela. For hypothesis 3 the more developed economies are: Argentina, Brazil, Chile and Mexico. Consequently, the less developed economies are Colombia, Costa Rica, Ecuador, Guatemala, Panama, Peru and Venezuela. This selection of countries does not include all Latin America countries. Since it is not possible to investigate all Latin America countries, this selection could imply a bias which is unavoidable. This research does not investigate literally Latin America, but investigates the countries mentioned above. 18

EMPIRICAL RESULTS: HYPOTHESES TESTING AND DISCUSSION Section 8: Empirical Results: Hypotheses Testing and Discussion 8.1. Hypothesis 1 Table 1 provides the inputs for the hypothesis 1 testing statistic (see appendix A for SPSS outputs). The second column provides the average MFI asset return for the years 2005 to 2009. The third column provides the average benchmark portfolio return for the years 2005 to 2009. The fourth and fifth column provide the variance of the MFI asset return and the variance of the benchmark portfolio return respectively. The last two columns provide the number of observations. At first glance, the average MFI return on assets and the average benchmark portfolio return seem to be quite different. In order to statistically test this, an unequal variance t-test is conducted. The outcomes of this test, which is called the G-value, can be found in table 2. The rejection regions are obtained by a t-distribution table. Year Average MFI Asset Return (µ1) Average Benchmark Return (µ2) Variance MFI Return (S1) Variance Bechmark Return (S2) N1 N2 2005 0.00019 0.06908 0.03141 0.00224 119 119 2006 0.00065 0.06845 0.02651 0.01557 171 171 2007 0.00089 0.14513 0.01681 0.02945 183 183 2008 0.00074-0.1887 0.01514 0.08028 229 229 2009 0.02230 0.12591 0.00509 0.01988 139 139 Table 1: Mean and Variance Return Data Year G-value Rejection Region 2005-4.0964 G -2.2701 OR G 2.2701 2006-4.3224 G -2.2613 OR G 2.2613 2007-9.0732 G -2.2600 OR G 2.2600 2008 9.2814 G -2.2562 OR G 2.2562 2009-7.7304 G -2.2659 OR G 2.2659 Table 2: Unequal Variance T-test Results and Rejection Regions 19

EMPIRICAL RESULTS: HYPOTHESES TESTING AND DISCUSSION Table 2 shows that the G-value for all years is rejected. That means that the mean return on assets of MFIs and mean return of the benchmark bond portfolios are statistically different. For the years 2005, 2006, 2007 and 2009, the mean return of the benchmark bond portfolio is statistically higher than the mean return of the MFI return on assets. Only for 2008 the opposite holds. This implies that for the years 2005, 2006, 2007 and 2009 it can be concluded with a rather high confidence that investing in the bond portfolios of the separate Latin American countries on average yields a higher return than investing in the assets of MFIs of the same countries. For 2008 the opposite holds, which means that investing in the assets of MFIs yields a higher return than investing in the bond portfolios of the same countries. The economic crisis took place in 2008, so that this finding is in line with Gonzales (2007) finding that MFIs are to a certain extent detached from macro economic shocks. 8.2. Hypothesis 2 In this paragraph two hypotheses will be tested. The first hypothesis is as follows: H0: The correlation coefficient between return on assets and return on bond portfolios is equal to or smaller than 0 H1: The correlation coefficient between return on assets and return on bond portfolios is bigger than 0 Table 3 provides a summary of correlation data. The SPSS outputs can be found in appendix B. Year Correlation Coefficient Significance 2005 0.334 0.000 2006-0.372 0.000 2007 0.108 0.073 2008 0.033 0.311 2009 0.019 0.414 Table 3: Correlation between MFI Asset Return and Benchmark Bond Portfolios 20

EMPIRICAL RESULTS: HYPOTHESES TESTING AND DISCUSSION Table 3 shows that the correlation between MFI asset return and the benchmark bond portfolios is close to zero and not significant for 2007, 2008 and 2009. The correlation coefficients for 2005 and 2006 are highly significant and is positive for 2005 and negative for 2006. The null hypothesis is confirmed for the years 2006, 2007, 2008 and 2009, since for these years the correlation between MFI asset return and the benchmark bond portfolios is statistically equal to 0 or negative. The null hypothesis is only rejected for the year 2005, since this year shows a positive and significant correlation. It can be concluded that investing in MFI assets creates a positive diversification value between 2006 and 2009. For 2005, this value creation is less strong. The second hypothesis tested in this paragraph is: H0: The coefficient of determination is smaller than 0.10 H1: The coefficient of determination is equal to or larger than 0.10 YEAR R-SQUARE 2005 0.1116 2006 0.1384 2007 0.0117 2008 0.0011 2009 0.0004 Table 4: R-square of the relation between MFI Asset Return and the Benchmark Bond Portfolios Table 4 shows similar results as table 3.The null hypothesis is rejected for 2005 and 2006 and confirmed for 2007, 2008 and 2009. This implies that for 2005 and 2006 the benchmark bond portfolio return explains the variance of MFI return on assets to the same extent as Krauss and Walter (2009) found. For 2007, 2008 and 2009 there is no statistical evidence that the benchmark bond portfolio return explains the variation in the MFI return on assets. In total it is hard to draw conclusions from these results. 21

EMPIRICAL RESULTS: HYPOTHESES TESTING AND DISCUSSION 8.3. Hypothesis 3 In order to see whether investing in MFI assets of less developed economies of Latin America has a different return than investing in assets of MFIs of the whole dataset, the tests described in paragraph 6.2.1 and paragraph 6.2.2 will be repeated with the dataset in which the observations of the more developed economies of Latin America are omitted. Table 5 provides the inputs for the testing statistic (See appendix C for the SPSS outputs). The second column provides the average MFI asset return for the years 2005 to 2009. The third column provides the average benchmark portfolio return for the years 2005 to 2009. The fourth and fifth columns provide the variance of the MFI asset return and the variance of the benchmark portfolio return respectively. The last two columns provide the number of observations. Year Average MFI Asset Return Average Benchmark Return Variance MFI Return Variance Bechmark Return N1 N2 2005 0.03467 0.08246 0.004 0.001 93 93 2006 0.03035 0.03915 0.003 0.010 126 126 2007 0.02754 0.18037 0.003 0.035 134 134 2008 0.02462-0.2558 0.004 0.091 153 153 2009 0.02030 0.11070 0.002 0.010 105 105 Table 5: Mean and Variance Return Data of the Less Developed Latin America Economies In order to test for the difference in mean between return on MFI assets of less developed countries in Latin America and return on bond portfolios of less developed economies in Latin America, the unequal variance t-test used to test for hypothesis 1 is repeated. The results of this test can be found in the G-value column of table 6. The rejection regions are obtained by a t-distribution table. 22

EMPIRICAL RESULTS: HYPOTHESES TESTING AND DISCUSSION Year G-value Rejection Region 2005-6.5183 G -2.2783 OR G 2.2783 2006-0.8661 G -2.2685 OR G 2.2685 2007-9.0756 G -2.2669 OR G 2.2669 2008 11.2545 G -2.2637 OR G 2.2637 2009-8.4558 G -2.2740 OR G 2.2740 Table 6: Unequal Variance T-test Results and Rejection Regions Table 6 shows that the MFI return on assets of less developed countries in Latin America is lower than the return on benchmark bond portfolios for the years 2005, 2006, 2007 and 2009. In 2008 the opposite holds, which means that investing in the assets of MFIs yields a higher return than investing in the bond portfolios of the same countries. The results of the unequal variance t-test in table 6 show that the return on MFI assets in countries of less developed economies in Latin America show roughly the same difference to the return on benchmark bond portfolios as the whole Latin America sample. Comparing these results with the results of hypothesis 1, it is noted that the return on MFI assets stays lower than the return on the benchmark bond portfolios for the years 2005, 2006, 2007 and 2009. The opposite stays the same for the year 2008. The only difference is that the difference between MFI asset return and the benchmark bond portfolio return is not statistically different anymore for the year 2006. Year Correlation Coefficient Significance 2005 0.033 0.378 2006 0.079 0.189 2007-0.107 0.109 2008 0.060 0.230 2009 0.082 0.204 Table 7: Correlation between MFI Asset Return and Benchmark Bond Portfolios 23

EMPIRICAL RESULTS: HYPOTHESES TESTING AND DISCUSSION Table 7 shows that the correlation coefficients of the return on MFI assets of the less developed Latin America countries to their benchmark portfolios are not statistically significant for any of the years investigated. This implies a diversification benefit for western investors who invest in bond portfolios. 24

SUMMARY, CONCLUSION AND DISCUSSION Section 9: Summary, Conclusion and Discussion This thesis investigates whether investing in Latin America s MFI assets yields a positive value for western bond investors. In order to answer this question the value creation was divided into three elements: (1) Performance of MFIs (2) Correlation of MFI performance to the benchmark (3) Cross sectional differences in MFI performance. Since the literature does not provide a satisfying measure for MFI performance, the MFI return on assets is selected as a measure for MFI performance. As a benchmark return on bond portfolios from the same countries is chosen. Data is collected for the years 2005, 2006, 2007, 2008 and 2009. (Hypothesis 1) The average return on assets of MFIs is statistically lower than the average return on benchmark bond portfolios for all years, except the crisis year 2008. This implies that, from a return-perspective, an investor better invests in bond portfolios than in the assets of MFIs. This finding does not hold for the year 2008, which could indicate that investing in MFIs is an attractive alternative during economic crises, like Gonzales (2007) suggests. (Hypothesis 2) The correlation between MFI performance and the benchmark performance is a second element of potential value creation. A low or negative correlation could imply diversification benefits for investors. In this research, there is a negative or statistically insignificant relation for the year 2006, 2007, 2008 and 2009. Only in 2005 there is a positive and significant correlation. This implies that there was a diversification benefit for bond portfolio investors to invest in MFIs (except for 2005). This finding is in line with Gonzales (2007) results, but contradicts with Krauss and Walter s (2009) results, since Krauss and Walter (2009) found a positive correlation between MFI performance and the local market performance. (Hypothesis 3) The last element of potential value creation is the selection of the most profitable MFIs in Latin America. Galema et al. (2008) suggest that MFI performance varies substantially among regions. By looking at the individual MFI return on assets observations it is noted that the more developed economies of Latin America (Brazil, Mexico, Argentina and 25

SUMMARY, CONCLUSION AND DISCUSSION Chile) often have a strongly negative return on assets. In order to test whether the observations of these four countries have a big impact on the average MFI performance and the correlation of the MFI performance to their benchmark, the tests of hypotheses 1 and 2 are repeated, but without the observations from Brazil, Mexico, Argentina and Chile. It turns out that omitting these countries, does not have a big impact on the results obtained for hypotheses 1 and 2. It could be concluded that only investing in the less developed economies of Latin America yields roughly the same results as investing in all Latin America s countries investigated. Taking the three elements of value creation for western bond investors together it could be concluded that investing in Latin America s MFI assets is not a good alternative for investing in bond indices in terms of return. It could, however, be a good diversification strategy to invest in MFI assets, but the cost of a low return is too high. Investing in the less developed economies of Latin America does not provide a good alternative for investing in all Latin American countries. The drawback for this research and the papers written so far, is that very few MFIs issued bonds or stocks. For that reason it is not possible to directly assess whether MFIs add value to existing investors bond or stock portfolios. The comparison of MFI return on assets with the return on bond portfolios, is believed to be more precise than the methodology of the other papers, but still it is not perfect. The test for heteroskedasticity reveals that the results of 2005, 2006 and 2007 should be treated with more caution, since heteroskedasticity might have taken place in these years. For the less developed countries in Latin America heteroskedasticity might have taken place in 2005 and 2007. The Durbin-Watson test, which tests for the independency of error variables, either reveals that the error random variables are indeed independent, or the test cannot provide a conclusive result. It might be that in the next years more MFIs issue debt. It would be worthwhile to repeat this research when data on MFI bonds is available. Moreover, it would be worthwhile to repeat this research when the MFI industry is more mature. Finally, when time passes, it will be possible to investigate whether MFI performance is indeed resistant to economic crises. 26

REFERENCES Section 10: References Ahlin, C., and J. Lin (2006), Luck or Skill? MFI performance in Macroeconomic Context, BREAD Working Paper 132 Ahlin, C., J. Lin and M. Maio (2009), Where does Microfinance Flourish? Microfinance Institution Performance in Macroeconomic Context, Working Paper Brau, J.C., and G. Woller (2004), Microfinance Institutions: A Comprehensive Review of the Existing Literature and an Outline for Future Financial Research, Journal of Entrepreneurial Finance and Business Ventures 9(1), pp. 1-26 Cull, R., A. Demirguc-Kunt and J. Morduch (2009), Microfinance Meets the Market, Contemporary Studies in Economic and Financial Analysis 92, pp. 1-30 Dieckmann, R. (2007), Microfinance: An Emerging Investment Opportunity, Deutsche Bank Research Dieckmann, R. (2008), Microfinance: An Attractive Dual Return Investment Opportunity, Capco Journal of Financial Transformation, pp. 108-112 Edgcomb, E. L. (2002), What Makes for Effective Micro Enterprise Training? Journal of Microfinance 4 (1) Evans, J. (2010), Microfinance s Midlife Crisis, The Wall Street Journal (March 1) Galema, R., R. Lensink, and L. Spierdijk (2008), International Diversification and Microfinance, University of Groningen Gonzales, A. (2007), Resilience of Microfinance Institutions to National Macroeconomic Events: An Econometric Analysis of MFI Asset Quality, MIX Discussion paper (1) Hull, J.C. (2006), Options, Futures and Other Derivatives, University of Toronto 27

REFERENCES Kapler, J.K. (2000), Measuring the Economic Rate of Return in Assets, Review of industrial organisation 17, pp. 457-463 Krauss, N., and I. Walter (2009), Can Microfinance Reduce Portfolio Volatility?, New York University Nieuwenhuis, G. (2008), Statistic Thinking and Methodology For Business and Economics, University of Tilburg Rahman, A. (1999), Micro-Credit Initiatives for Equitable and Sustainable Development: Who Pays?, World Development 27(1), pp. 67-82 Stark, A.W. (2004), Estimating Economic Performance from Accounting Data a Review and Synthesis, The British Accounting Review 36, pp. 321-343 Woolcock, M. (2001), Micro Enterprise and Social Capital: A framework for Theory, Research, and Policy, Journal of Socio-Economics 30 (2) 28

APPENDICES Appendix A: Mean and Variance Outputs for Hypothesis 1 2005 Descriptive Statistics N Mean Variance assets 119,109160% 0,031412 Benchmark 119,06908022,002242 Valid N (listwise) 119 2006 Descriptive Statistics N Mean Variance assets 171,646374% 0.0265112 Benchmark 171,06845448,016 Valid N (listwise) 171 2007 Descriptive Statistics N Mean Variance assets 183,889016% 0.0168057 Benchmark 183,14513393,029 Valid N (listwise) 183 2008 Descriptive Statistics N Mean Variance assets 229,741223% 0.0151444 Benchmark 229 -,188724,081 Valid N (listwise) 229 29

APPENDICES 2009 Descriptive Statistics N Mean Variance assets 139 2,230288% 0.0050948 Benchmark 139,12591614,020 Valid N (listwise) 139 Appendix B: Correlation Outputs for Hypothesis 2 2005 Correlations assets V13 assets Pearson Correlation 1,334(**) Sig. (1-tailed),000 N 119 119 Pearson Correlation,334(**) 1 Benchmark Sig. (1-tailed),000 N 119 119 ** Correlation is significant at the 0.01 level (1-tailed). 2006 Correlations assets V13 assets Pearson Correlation 1 -,372(**) Sig. (1-tailed),000 N 171 171 Pearson Correlation -,372(**) 1 Benchmark Sig. (1-tailed),000 N 171 171 ** Correlation is significant at the 0.01 level (1-tailed). 30

APPENDICES 2007 Correlations assets Benchmark assets V13 Pearson Correlation 1,108 Sig. (1-tailed),073 N 183 183 Pearson Correlation,108 1 Sig. (1-tailed),073 N 183 183 2008 Correlations assets Benchmark assets V13 Pearson Correlation 1,033 Sig. (1-tailed),311 N 229 229 Pearson Correlation,033 1 Sig. (1-tailed),311 N 229 229 2009 Correlations assets Benchmark assets V13 Pearson Correlation 1,019 Sig. (1-tailed),414 N 139 139 Pearson Correlation,019 1 Sig. (1-tailed),414 N 139 139 31

APPENDICES Appendix C: Outputs for Hypothesis 3 Mean and Variance Outputs 2005 Descriptive Statistics N Mean Variance assets 93 3,466774% 0,0036441 Benchmark 93,08246229,001 Valid N (listwise) 93 2006 Descriptive Statistics N Mean Variance assets 126 3,035317% 0,0028893 Benchmark 126,03915029,010 Valid N (listwise) 126 2007 Descriptive Statistics N Mean Variance assets 134 2,753582% 0,0029323 Benchmark 134,18036727,035 Valid N (listwise) 134 2008 Descriptive Statistics N Mean Variance assets 153 2,461765% 0,0039988 Benchmark 153 -,255825,091 Valid N (listwise) 153 32

APPENDICES 2009 Descriptive Statistics N Mean Variance assets 105 2,030286% 0,0021186 Benchmark 105,11069895,010 Valid N (listwise) 105 Correlations Outputs 2005 Correlations assets Benchmark assets V13 Pearson Correlation 1,033 Sig. (1-tailed),378 N 93 93 Pearson Correlation,033 1 Sig. (1-tailed),378 N 93 93 2006 Correlations assets Benchmark assets V13 Pearson Correlation 1,079 Sig. (1-tailed),189 N 126 126 Pearson Correlation,079 1 Sig. (1-tailed),189 N 126 126 33

APPENDICES 2007 Correlations assets Benchmark assets V13 Pearson Correlation 1 -,107 Sig. (1-tailed),109 N 134 134 Pearson Correlation -,107 1 Sig. (1-tailed),109 N 134 134 2008 Correlations assets Retun on Benchmark assets V13 Pearson Correlation 1,060 Sig. (1-tailed),230 N 153 153 Pearson Correlation,060 1 Sig. (1-tailed),230 N 153 153 2009 Correlations assets Benchmark assets V13 Pearson Correlation 1,082 Sig. (1-tailed),204 N 105 105 Pearson Correlation,082 1 Sig. (1-tailed),204 N 105 105 34