1
The effect of formal
financial inclusion on the
outreach of MFIs to the
poor
Evidence from Bolivia
2
The effect of formal financial inclusion on the
outreach of MFIs to the poor
Evidence from Bolivia
Student number: s2201437
Name: J.W.R. Peerlings
Study program: MSc Finance Supervisor: dr. F. Cecchi
Date: 23-06-2015
3
Acknowledgements
4
Abstract
Arguments are made that, with the microfinance sector of Bolivia coming of age and commercial banks also participating, microfinance institutions (MFIs) are abandoning their mission to serve the poor. However, arguments to the contrary are proposed as well. In the light of this debate, this thesis examines, using panel data estimation methods, whether an increase in formal financial inclusion in Bolivia leads to an increase or decrease of the outreach of MFIs to poorer borrowers. Outreach is investigated using number of active borrowers, average loan size, and percentage of female borrowers. Results show that outreach of MFIs to more borrowers is complementary to that of commercial banks. However, it is found as well that for-profit MFIs there is mission drift.
5
Table of contents
List of tables and figures ... 6
1. Introduction ... 7
2. Literature review ... 9
3. Theoretical model ... 12
3.1. Index of formal financial inclusion ... 12
3.2. Outreach ... 13
3.3. Analytical framework ... 14
4. Data ... 17
5. Empirical model and estimation ... 20
6. Results ... 22
6.1. Estimation ... 22
6.2. OLS ... 22
6.3. Panel Data Models ... 26
6.3.1. Attrition bias ... 26
6.3.2. NAB ... 27
6.3.3. ALB and FEM ... 29
6.4. Extensions and suggestions ... 32
7. Discussion and conclusions ... 33
Appendix ... 35
A.1. Correlation table ... 35
A.2. Attrition ... 36
A.3. Fixed effects model FEMALE ... 38
A.4. Diamonds ... 39
6
List of tables and figures
Table 1: Number of MFIs per year... 17 Table 2: Number of years that MFI is in the dataset ... 17 Table 3: Description of the model variables ... 18 Table 4: Description of the MFIs divided on the basis of their target market, legal status,
outreach, profit status and regulation status (number and percentage of MFIs per
characteristic) ... 18 Table 5: OLS with number of active borrowers, average loan balance per borrower ($), or
percentage of female borrowers (%) as dependent variable [t-values between
brackets] ... 23 Table 6: Estimation results for panel data models with number of active borrowers as
dependent variable [t-values between brackets] ... 28 Table 7: Estimation results for panel data models with average loan balance per borrower ($)
as dependent variable [t-value between brackets] ... 30
Table A1: Correlation explanatory and dependent variables ... 35 Table A2: Nijman-Verbeek test for attrition bias [t-values between brackets] ... 36 Table A3: Nijman-Verbeek test for attrition bias omitting smallest MFI [t-values between
brackets] ... 37 Table A4: Estimation results for panel data models with percentage of female borrowers (%) as
dependent variable [t-values between brackets] ... 38 Table A5: Description of the panel (MFIs per year) based on panel of 4 and 5 diamond
observations ... 39 Table A6: Number of years that a MFI is in the dataset based on panel of 4 and 5 diamond
observations ... 398 Table A7: Description of the MFIs divided on the basis of their target market, legal status,
outreach, profit status and regulation status (number and percentage of MFIs per
characteristic) based on panel of 4 and 5 diamond observations ... 40 Table A8: Estimation results for panel data models with number of active borrowers, average
loan balance per borrower ($), or percentage of female borrowers (%) based on
7
1. Introduction
The microfinance sector of Bolivia is considered one of the most mature and successful in the world (Mosley, 2001; Robledo, 2008). It consists of around twenty active microfinance institutions (MFIs) and includes almost half a million borrowers. According to Robledo (2008) high competition between MFIs, leads to borrowers now getting the best financial services at the lowest cost and, moreover, most of the population having access to financial services. Other characteristics that MFIs face in the highly competitive microfinance market in Bolivia are thin margins, currency risks and high leverage. With the microfinance sector of Bolivia coming of age and commercial banks also participating, the question can be asked whether MFIs are abandoning their mission to serve the poor (Dichter and Harper, 2007). The mission of all MFIs is to provide banking services to the poor, in other words, lend small sums to poor borrowers (Mersland and Strøm, 2010). The access of the poor to financial services can be defined as formal financial inclusion.
With commercial banks expanding their financial base, MFIs may reach out to even poorer borrowers. It is argued that a more commercialized microfinance industry leads to more outreach to poorer borrowers, because competition and profit motives lead MFIs to be more efficient and seek out new markets for their products (Mersland and Strøm, 2010). However, Hermes, Lensink and Meesters (2011) provide evidence that outreach and efficiency of the MFI are negatively related. Critics argue that MFIs are becoming too focused on making profits, reducing their outreach to poor borrowers. Commercialization can be defined as ‘’the application of market based principles to the management of MFIs’’ (Armendáriz and Morduch, 2010). Although in principle seeking profit and serving the poor can go hand in hand, they argue that a trade-off between profitability and outreach exists (Cull, Demirgüç-Kunt, and Morduch, 2007) and that MFIs experience mission drift (Mersland and Strøm, 2010).
In the light of this debate, this thesis examines the impact of formal financial inclusion on MFIs. More specifically, it analyses whether increased formal financial inclusion in an economy leads to a change in the client base of MFIs towards a larger share of poor borrowers. The following two hypothesises can be stated:
H1: More formal financial inclusion leads to MFIs having a larger share of poor borrowers
H2: More formal financial inclusion leads to MFIs having a smaller share of poor borrowers
8 complementary or substitute to formal financial inclusion? Moreover, does formal financial inclusion lead to mission drift of MFIs?
Methodology
In order to answer the research objectives first a literature research has been performed in order to select variables that can explain MFI’s outreach to the poor. Using this literature review a theoretical model is formulated. Next after discussing the data different empirical models are estimated to investigate whether formal financial inclusion explains outreach and to what extent. Estimation has been done using both OLS, and fixed effects and random effect estimators. Given the lack of availability of indicators for formal financial inclusion this thesis contributes to the literature by constructing such an index.
Overview
9
2. Literature review
This chapter starts with a definition of formal financial inclusion. From the analysis of the available literature it becomes evident that no universal definition of formal financial inclusion exists. Differences arise from the context in which formal financial inclusion is defined. In many countries, including Bolivia, the number of financially excluded exceeds the number of people living with less than $2 a day. This chapter describes the relationship between formal financial inclusion and economic growth and argues that although a positive relation exits, the relation is imperfect. Moreover, in Latin America, some of the best regulatory environments for microfinance exists. Therefore, the relationship between formal financial inclusion and financial sector stability is described as well. More specifically, in Bolivia, micro deposit taking flourishes while a heavy regulatory framework is present. In recent years, however, microfinance has become more commercialized. This chapter concludes with an examination of whether MFIs experience mission drift in serving the poor and whether, if present, this differs for MFIs of different legal status.
In their paper Amidžić, Massara, and Mialou (2014) define formal financial inclusion in terms of what it is not or, in other words, in terms of financial exclusion. Formal financial exclusion arises from either voluntary-exclusion (self-exclusion) or involuntary-exclusion. Involuntary-exclusion results from the income of individuals being too low and the fact that they propose a too high risk for credit providers. Individuals may also be involuntary-excluded due to market imperfections or government failures. The latter arise due to incomplete/imperfect information leading to moral hazard and adverse selection problems (Amidžić, Massara, and Mialou, 2014). Information problems are common in developing economies. Formal financial inclusion can be defined as the opposite of financial exclusion.
10 (Aduda and Kalunda, 2012). This definition is shared by the World Bank. They define formal financial inclusion as ‘’the broad access to financial services, without price or non-price barriers to their use’’ (Demirgüç-Kunt and Klapper, 2012). It should be noted that MFIs on their own cannot bring about formal financial inclusion as they do not offer the range of financial products and services which are considered the bare minimum to qualify as availability of banking services (Chakrabarty, 2010). Numerous studies have focused on the relationship between economic development and formal financial inclusion. It is concluded that a positive relationship between the two exists. In Bolivia rapid economic growth in the past years has increased the level of formal financial inclusion significantly. However, Hanning and Jansen (2010) find that the relationship between economic development and formal financial inclusion is imperfect. Based on data of the World Bank from 2008 they find that formal financial inclusion is an issue well beyond the population living under the $2-a-day poverty line.
Less studies focus on the impact of formal financial inclusion. Lending to the lowest income groups is associated with idiosyncratic risk as well as high costs. Hanning and Jansen (2010) notice that the exposure of financial institutions to risk of low-income clients depends on the share of the revenues of this group relative to the total revenues of the financial institution. MFIs are most prone to this risk as they have the largest share of low-income groups. However, depending on regulation and supervision these idiosyncratic risk profiles can be managed. In their theoretical research, Aduda and Kalunda (2012) examine the impact of formal financial inclusion on financial sector stability. They conclude that a trade-off exists and, therefore, formal financial inclusion should be taken with care to avoid financial instability. The role of the government should be carefully defined and proper banking models should be carefully designed. Hanning and Jansen (2010) argue that balance between formal financial inclusion and financial stability can be achieved, as is shown in some Latin-American countries such as Bolivia.
11 have left their mission to save the poor. Tchakoute-Tchuigoua (2010) examines in his paper the relationship between the performance of MFIs and their legal status, looking at private companies, non-governmental organisations (NGOs) and cooperatives. It is argued that the performance of private and regulated companies is higher than that of non-profit MFIs. The results found are in line with the findings of Mersland and Strøm (2010), in that profit oriented MFIs (private companies and cooperatives) can achieve better social performance than NGOs. This because private companies and cooperatives collect deposits unlike NGOs and are therefore better able to increase their investment (lending).
Kyereboah-Coleman (2007) examines whether the performance of a MFI is affected by its capital structure. The research is performed by examining 52 MFIs from Ghana between 1995 and 2004. It is concluded that highly leveraged MFIs perform better ‘’by reaching out to more clientele base and reducing default rates.’’ High leverage allows MFIs to reap economies of scale, and, therefore, enables MFIs to be ‘’better able to deal with moral hazard and adverse selection and to accommodate risk’’ (Armendáriz and Morduch, 2010; Kyereboah-Coleman, 2007).
12
3. Theoretical model
This chapter gives a description of the theoretical model used in this thesis. First, measures of both formal financial inclusion and outreach are discussed. It is argued that formal financial inclusion can be measured by a variable resulting from factor analysis using the variables ‘’number of commercial bank branches per 1,000km2’’ and ‘’outstanding deposits with commercial banks (% of GDP).’’ Moreover, three variables are identified that measure outreach. The theoretical model is therefore specified for these three variables.
3.1. Index of formal financial inclusion
In order to test the hypothesis a robust and comprehensive measure of formal financial inclusion in an economy is needed. However, despite being recognized as an important factor for economic development, a formal financial inclusion dataset comprising multiple years does not exists. For example, the Global Financial Inclusion (Global Findex) only looks at the year 2011 (Demirgüç-Kunt and Klapper, 2012). Moreover, the literature lacks an extensive measure for measuring the level of formal financial inclusion in an economy. From a practical point of view, formal financial inclusion should be measured in terms of its dimensions. Hanning and Jansen (2010) define four dimensions to measure formal financial inclusion; access, quality, usage and impact. Sarma (2010) defines formal financial inclusion as ‘’as a process that ensures the ease of access, availability and usage of the formal financial system for all members of an economy.’’ He measures formal financial inclusion in terms of accessibility, availability and usage and provides an index of formal financial inclusion (IFI) that takes a value between 0 and 1. The dimensions he takes into account are banking penetration (number of bank accounts as a proportion of the total adult population), availability of banking services (the number of bank outlets per 1,000 people and/or the number of ATMs per 1,000 people), and usage (the volume of credit and deposit as proportion of the country’s GDP).
13 outreach dimension, because ‘’the physical distance tends to be an important barrier to formal financial inclusion.’’
In this thesis, similar to Amidžić, Massara, and Mialou (2014), a factor analysis is performed in order to calculate the formal financial inclusion index. Factor analysis is a method to create one or more indices out of variables that measure similar things (Princeton, 2015). It has the advantage that it does not assume equal weights, but calculates the optimal weights for the variables based on eigenvalues. Based on Sarma (2010), and Amidžić, Massara, and Mialou (2014), this thesis constructs the index of formal financial inclusion from the variables ‘’number of commercial bank branches per 1,000km2’’ and ‘’outstanding deposits with commercial banks (% of GDP).’’ The ‘’number of commercial bank branches per 1,000km2’’ measures the width and ‘’outstanding deposits with commercial banks’’ the depth of formal financial inclusion. Due to similar time trends in variables and data constraints, the number of ATMs as well as other variables proposed by Sarma (2010), and Amidžić, Massara, and Mialou (2014) are not used in the index. Since the two variables are highly correlated (0.9085) it is appropriate to combine them into a composite variable (Berry and Feldman, 1985). Since there are only two variables, factor analysis results in a single index of formal financial inclusion. Since this index of formal financial inclusion is measured at the commercial bank level, endogeneity between the explanatory and dependent variables is avoided, since the dependent variables (i.e. outreach) are measured at the level of the MFIs.
3.2. Outreach
Schreiner (2002) describes in his paper six dimensions of outreach to the poor: worth, cost, depth, breath, length and scope. However, since these dimensions are difficult to measure due to data constraints, the focus in this research lies on the depth of outreach, where depth of outreach means ‘’the value that society attaches to the net gain of a given client. In welfare theory, depth is the weight of a client in the social-welfare function. If society has a preference for the poor, then poverty is a good proxy for depth. For example, society likely prefers that a street child or a widow get a given net gain than that a richer person get the same net gain’’ (Schreiner, 2002).
14 considered to be a proxy for (depth of) outreach since poor borrowers generally have smaller loan accounts. Higher values of the average loan balance per borrower indicate less depth of outreach, since in this case the MFI is expected to provide fewer loans to poor borrowers. In contrast, a higher percentage of women borrowers indicates more depth of outreach, since lending to women is associated with lending to poorer borrowers. Moreover, a higher number of credit clients also indicates a higher level of outreach.
3.3. Analytical framework
Based on the literature review and Sections 3.1. and 3.2. the theoretical model is now specified. The general model is given by:
𝑂𝑈𝑇𝑖𝑡 = 𝛼 + 𝐹𝐼𝑡𝛽 + 𝑋𝑖𝑡𝛾 + 𝜀𝑖𝑡 (1)
where:
𝛼 the constant, 𝛽 and 𝛾 coefficients for formal financial inclusion and control variables respectively, 𝑂𝑈𝑇𝑖𝑡 outreach of MFI i in year t, 𝐹𝐼𝑡 index of formal financial inclusion in year t, 𝑋𝑖𝑡 vector of control
variables for MFI i in year t, 𝜀𝑖𝑡 error term of MFI i in year t.
From equation 1 and the three indicators of outreach (i.e. the total number of credit clients, average loan balance per borrower and the percentage of female borrowers) the following three equations are formulated:
𝑁𝐴𝐵𝑖𝑡 = 𝛼 + 𝐹𝐼𝑡𝛽 + 𝑋𝑖𝑡𝛾 + 𝑟𝑖𝑡 (2)
𝐴𝐿𝐵𝑖𝑡 = 𝛼 + 𝐹𝐼𝑡𝛽 + 𝑋𝑖𝑡𝛾 + 𝑠𝑖𝑡 (3)
𝐹𝐸𝑀𝑖𝑡 = 𝛼 + 𝐹𝐼𝑡𝛽 + 𝑋𝑖𝑡𝛾 + 𝑢𝑖𝑡 (4)
where:
𝑁𝐴𝐵𝑖𝑡 total number of credit clients of MFI i in year t, 𝐴𝐿𝐵𝑖𝑡 average loan balance per borrower for
MFI i in year t, and 𝐹𝐸𝑀𝑖𝑡 the percentage of female borrowers for MFI i in year t. The vector of
control variables 𝑋𝑖𝑡 consists of: 𝑃𝑀𝑖𝑡 the profit margin of MFI i in year t, 𝐷𝐸𝑖𝑡 debt-equity ratio of
MFI i in year t, 𝐶𝐵𝑖𝑡 cost per borrower of MFI i in year t, 𝐿𝑃𝑖𝑡 gross loan portfolio to total assets of
MFI i in year t, 𝑅𝐼𝑖𝑡 loans outstanding that are at substantial risk of default (30 days) for MFI i in year
t, 𝑁𝑃𝑖𝑡 non-profit status of MFI i in year t, 𝐹𝐼𝑁𝑃𝑖𝑡 interaction term between formal financial inclusion
15 The total number of credit clients of the MFI (𝑁𝐴𝐵𝑖𝑡), the average loan balance per borrower
(𝐴𝐿𝐵𝑖𝑡), and the percentage of female borrowers (𝐹𝐸𝑀𝑖𝑡) all three proxies for the dependent
variable, outreach. As formulated in the hypothesises, formal financial inclusion is the variable of interest, and is given by the index of formal financial inclusion (𝐹𝐼𝑡) in Bolivia calculated for each year
with the factor analysis. The remaining variables are control variables.
The profitability of the MFI is expressed as the profit margin (𝑃𝑀𝑖𝑡). MixMarket (2015) defines the
profit margin of the MFI as the net operating income divided by financial revenue. It is argued by Cull, Demirgüç-Kunt, and Morduch (2007) that a trade-off exists between profitability and serving the poorest, however, their simplest regression is insignificant suggesting that the pursuit of profit and outreach to the poor can go together and mission drift is not inevitable. It is expected that the debt-equity ratio of the MFI (𝐷𝐸𝑖𝑡) has an positive effect on outreach. The more leveraged the MFI the
better it can handle moral hazard and adverse selection and the better it can accommodate risk, therefore, the more it can increase its lending to poor borrowers. Higher costs per borrower (𝐶𝐵𝑖𝑡)
may lower the outreach of the MFI since it becomes relatively more costly to lend to the poor due to their smaller loan sizes. The gross loan portfolio of a MFI to the total assets of the MFI (𝐿𝑃𝑖𝑡 ) may
provide an indication as to whether it is advantageous for the MFI to increase its gross loan portfolio even further by reaching out to more borrowers. The value of the loans outstanding that are at substantial risk of default (𝑅𝐼𝑖𝑡) is measured as: ‘’the value of all loans outstanding that have one or
more instalments of principal overdue for more than 30 days it includes the entire unpaid principal balance, including both past-due and future instalments, but not accrued interest. It also does not include loans that have been restructured or rescheduled’’ (CGAP 2003, 6; MixMarket, 2015). It is expected that this variable has a negative effect on outreach to the most poor borrowers.
The variable of interest in this thesis is the dummy variable 𝑁𝑃𝑖𝑡 which denotes whether the MFI has
a non-profit (𝑁𝑃𝑖𝑡 = 1) or for-profit (𝑁𝑃𝑖𝑡 = 0) status. As mentioned in the introduction of this
thesis, arguments can be made that MFIs are experiencing mission drift since they become too focused on making profits (Cull, Demirgüç-Kunt, and Morduch, 2007). By looking at MFIs that have a non-profit status, compared to those that are for-profit, the effect of possible trade-offs between profitability and outreach, as well as other control variables and outreach can be extensively examined.
Additionally, the interaction effect between formal financial inclusion and the non-profit status of the MFI (𝐹𝐼𝑁𝑃𝑖𝑡 = 𝐹𝐼𝑡∗ 𝑁𝑃𝑖𝑡) is examined. Since formal financial inclusion (𝐹𝐼𝑡) is measured for
17
4. Data
The annual data on formal financial inclusion are collected from the IMF’s Financial Access Survey (FAS) database. The Global Financial Development Database from the World Bank is not used since it does not provide data beyond 2011. The annual data on MFIs are taken from MixMarket (2015). The study focuses on a 10-year period (2004-2013) and provides information on 28 Bolivian MFIs. The study starts in 2004 because the FAS database starts in 2004, and, moreover, this is after the indebtedness crisis of Bolivia (Marconi and Mosley, 2006). It ends in 2013 because this is the most recent year for which annual data on formal financial inclusion variables in Bolivia are available. The dataset is an unbalanced panel and consists of 228 observations. Table 1 describes the dataset in terms of the ten-year period and the number of MFIs per year for which data is available. Table 2 shows the number of year observations per MFI. Table 3 gives the descriptive statistics of the variables and, finally, table 4 divides the MFIs on the basis of several characteristics and gives the number and percentage of MFIs per characteristic.
Table 1: Number of MFIs per year
Year Number of MFIs per year
2004 18 2005 20 2006 25 2007 26 2008 24 2009 25 2010 25 2011 24 2012 21 2013 20 Total 228 Source: MixMarket (2015)
Table 2: Number of years that a MFI is in the dataset
Number of years Number of MFIs
1 1 2 1 3 0 4 2 5 0 6 2 7 1 8 4 9 4 10 13 Total 28
18 Table 3: Description of the model variables
Variable N Mean Std. Dev. Min Max
Number of borrowers 224 9.5726 1.5876 4.0073 12.3795 Loan balance ($) 219 7.3514 1.1075 4.9904 10.5769 Female borrowers (%) 209 0.5520 0.1859 0.1661 1 Financial inclusion 228 0.0000 0.9532 -1.1797 1.9947 Profit margin (%) 216 0.1212 0.1698 -0.6426 0.7934 Debt-to-equity (%) 216 5.6744 7.7347 -13.2200 98.5100
Cost per borrower ($) 208 5.3682 0.8980 2.3979 7.5305
Interest rate on portfolio (%) 200 0.1430 0.0872 -0.0110 0.4980
Loan portfolio/asstes ($) 222 0.8135 0.0987 0.4644 1.0967
Risk (%) 209 -3.5902 1.2058 -7.2644 -0.3979
Non-profit 228 0.6579 0.4755 0 1
FINP 228 -0.0153 0.7619 -1.1797 1.9947
Source: Calculated from MixMarket (2015)
Table 4: Description of the MFIs divided on the basis of their target market, legal status, outreach, profit status and regulation status (number and percentage of MFIs per characteristic)
Characteristic Number of MFIs Percentage
Target market High end 55 24.7%
Low end 11 4.9%
Small 35 15.7%
Broad 122 54.7%
Legal status Cuco 14 6.2%
NBFI 1 0.4%
NGO 136 60.2%
Bank 75 33.2%
Outreach Medium/Small 145 65.0%
Large 78 35.0%
Profit status Non-profit 150 65.8%
For-profit 78 34.2%
Regulation status Not regulated 139 61.0%
Regulated 89 39.0%
Source: Calculated from MixMarket (2015)
20
5. Empirical model and estimation
By graphically describing the variables with the use of histograms, all variables are found to be normally distributed with the exception of number of active borrowers (𝑁𝐴𝐵𝑖𝑡), average loan
balance per borrower (𝐴𝐿𝐵𝑖𝑡), cost per borrower (𝐶𝐵𝑖𝑡) and loans outstanding that are at substantial
risk of default (𝑅𝐼𝑖𝑡). For that reason the natural log of these three variables is taken. First, a simple
OLS regression is performed. However, since this thesis uses panel data, pooling the data and using OLS has some major drawbacks. Most importantly, pooling the data implicitly assumes that the average values of the variables and the relationships between them are constant over time and across all the cross-sectional units in the sample (Brooks, 2014). Therefore, besides the OLS, fixed and random effects models are estimated. The data used in this thesis is panel data comprising both time series and cross-sectional elements (i.e. MFIs). In the fixed effects model a MFI specific intercept is assumed. In the random effects model there is a MFI specific error term. The general specification of the OLS, fixed effects and random effects model is given below.
OLS: 𝑂𝑈𝑇𝑖𝑡 = 𝛼 + 𝐹𝐼𝑡𝛽 + 𝑋𝑖𝑡𝛾 + 𝜀𝑖𝑡 (5) Fixed effects: 𝑂𝑈𝑇𝑖𝑡 = 𝛼𝑖+ 𝐹𝐼𝑡𝛽 + 𝑋𝑖𝑡𝛾 + 𝜀𝑖𝑡 (6) Random effects: 𝑂𝑈𝑇𝑖𝑡 = 𝛼 + 𝐹𝐼𝑡𝛽 + 𝑋𝑖𝑡𝛾 + 𝜀𝑖𝑡 (7) Where 𝜀𝑖𝑡 = 𝜏𝑖+ 𝜐𝑖𝑡
In order to decide on the best panel estimator approach, the Hausman test is conducted (Hausman, 1978). The null hypothesis is that the preferred model is random effects against the alternative of the fixed effects model. The Hausman test tests whether the unique disturbance term, 𝜏𝑖, is correlated
with the explanatory variables. The panel used in this thesis is a micro panel since it does not comprise a long time period (i.e. 10 years). Therefore, testing for cross-sectional dependence and serial correlation is not necessary.
It should be noted that the non-profit status of the MFI (𝑁𝑃𝑖𝑡) in this dataset is time-invariant and,
21 inclusion leads to more outreach to poorer borrowers by MFIs, looking at the difference between non-profit and for-profit MFIs, the rest of the fixed effects panel data estimations focus on the interaction effect 𝐹𝐼𝑁𝑃𝑖𝑡.
For equations 2-4, presented in Section 3.3., first a simple OLS will be performed. Because of multicollinearity not all dummies are included in the estimations. Thereafter, it will be tested whether the models are fixed or random effects models using the Hausman test. Additionally, this will be repeated for equations where the control variables as well as the interaction effect between formal financial inclusion and non-profit MFIs (𝐹𝐼𝑁𝑃𝑖𝑡) are included. Furthermore, a year fixed or
22
6. Results
This chapter starts with the correlation tables of the variables to analyse possible problems of multicollinearity. Thereafter, this chapter presents the regression estimation results from the equations with the non-profit status of the MFI. Moreover, panel estimations (i.e. the fixed and random effects model) are conducted for equations 8-10 focusing on the interaction term between the non-profit status of the MFI and formal financial inclusion. Furthermore, the Nijman Verbeek test (1992) is performed to test for selectivity nonresponse bias (i.e. attrition) in the panel data models. Finally, some possible extensions will be discussed.
6.1. Estimation
Table A1 in the appendix provides the correlation table of the variables based on equations 8-10. From the table it can be noticed that the explanatory variables are not highly correlated and, therefore, problems of multicollinearity and resulting biased estimates are not a problem in the estimation. The only exception could be the correlation between the formal financial inclusion variable (𝐹𝐼𝑡) and its interaction term (𝐹𝐼𝑁𝑃𝑖𝑡). With respect to the correlations between the dependent variables (i.e. outreach) and explanatory variables, the average loan balance per borrower (𝐴𝐿𝐵𝑖𝑡) and the cost per borrower (𝐶𝐵𝑖𝑡) are highly positively correlated. Since a higher
cost per borrower will make relatively small loans more costly, it is expected that, due to cost advantages, relatively larger loans per borrower will be more attractive. From this correlation it is expected that the independent variable has a large explanatory power for the dependent variable 𝐴𝐿𝐵𝑖𝑡. Looking at the formal financial inclusion variable (𝐹𝐼𝑡) and its correlation with the three
dependent variables it can be observed that the correlation between 𝐹𝐼𝑡 and the dependent variable
percentage of female borrowers (𝐹𝐸𝑀𝑖𝑡) is especially low as well as for 𝐹𝐼𝑁𝑃𝑖𝑡 and 𝐹𝐸𝑀𝑖𝑡. This low
correlation may mean that these two independent variables cannot really predict 𝐹𝐸𝑀𝑖𝑡.
6.2. OLS
Table 5 shows the result of the OLS estimation between the dependent variables 𝑁𝐴𝐵𝑖𝑡 (total
number of active borrowers), 𝐴𝐿𝐵𝑖𝑡 (average loan balance per borrower) and 𝐹𝐸𝑀𝑖𝑡 (the percentage
of female borrowers), and the explanatory variable 𝐹𝐼𝑡 (formal financial inclusion) in columns (1), (3)
and (5). Columns (2), (4) and (6) present OLS with all the controls except for the interaction term and with the year dummies. It was found that, especially for 𝐴𝐿𝐵𝑖𝑡, the years in which the MFIs are
23 Table 5: OLS with number of active borrowers, average loan balance per borrower ($), or percentage of female
borrowers (%) as dependent variable [t-values between brackets]
NAB ALB FEM
(1) (2) (3) (4) (5) (6) Financial inclusion 0.4072 0.3154 0.3639 0.0880 0.0017 0.0054 [3.74]*** [2.10]** [4.86]*** [2.02]** [0.13] [0.31] Profit margin (%) -0.4089 0.8968 -0.0436 [-0.73] [5.50]*** [-0.71] Debt-to-equity (%) 0.0019 -0.0035 -0.0004 [0.17] [-1.02] [-0.34]
Cost per borrower ($) -0.6103 1.1106 -0.1212
[-4.19]*** [26.19]*** [-7.50]*** Loan portfolio/assets (%) 0.1335 0.8503 -0.1729 [0.12] [2.73]*** [-1.46] Risk (%) -0.4085 0.0649 -0.0585 [-4.78]*** [2.61]*** [-6.17]*** Non-profit -1.7309 -0.3617 0.0823 [-7.56]*** [-5.43]*** [3.23]*** year2 -0.4101 0.0809 -0.0022 [-0.98] [0.66] [-0.05] year3 -0.3575 0.1962 0.0009 [-0.89] [1.68]* [0.02] year4 -0.3405 0.3814 -0.0546 [-0.90] [3.47]*** [-1.29] year5 -0.1096 0.2723 -0.0406 [-0.30] [2.59]*** [-1.00] year6 -0.1787 0.2355 -0.0187 [-0.52] [2.36]** [-0.49] year7 -0.2830 0.2053 -0.0077 [-0.88] [2.19]** [-0.21] year8 -0.2341 0.1738 -0.0057 [-0.68] [1.47]* [-0.15] year9 -0.1952 0.1549 -0.0090 [-0.56] [1.52] [-0.22] Cons 9.5795 12.7675 7.3612 0.8937 0.5552 1.0942 [92.88]*** [7.89]*** [103.30]** * [1.90]* [42.75] *** [6.13]*** N 224 201 219 201 209 196 Adj. R2 0.0550 0.4382 0.0940 0.8989 -0.0048 0.5314
*=significant at 10%; **=significant at 5%; ***=significant at 1%
NAB – From column (1) of table 5 it can be concluded that formal financial inclusion significantly increases the number of active borrowers of the MFI (𝑁𝐴𝐵𝑖𝑡), providing early evidence that more
24 active borrowers (𝑁𝐴𝐵𝑖𝑡), providing evidence for the first hypothesis that it leads to more outreach
to poorer borrowers by MFIs. Based on column (2) an increase in formal financial inclusion increases the number of active borrowers of all the MFIs by around 32%.
ALB – From column (3) of table 5 it can be concluded that formal financial inclusion significantly increases the average loan balance per borrower (𝐴𝐿𝐵𝑖𝑡), providing early evidence that more formal
financial inclusion leads to lower outreach to poorer borrowers by all the MFIs. This result is opposite to the result found for 𝑁𝐴𝐵𝑖𝑡. The result remains unchanged when the control variables and the year
dummies are added. Looking at column (4) it can be seen that an increase in formal financial inclusion significantly increases the average loan balance per borrower (𝐴𝐿𝐵𝑖𝑡), providing evidence
for the second hypothesis that it leads to less outreach to poorer borrowers by MFIs and thus evidence for mission drift (Mersland and Strøm, 2010).
FEM – From both columns (5) and (6) it can be concluded that the percentage of female borrowers of the MFIs (𝐹𝐸𝑀𝑖𝑡) is not significantly influenced by an increase of formal financial inclusion (𝐹𝐼𝑡).
Table 5 shows that, although not significant, an increase in formal financial inclusion results in a higher percentage of female borrowers of the MFIs providing suggestive supporting the first hypothesis that more formal financial inclusion leads to more outreach to poorer borrowers by the MFIs.
Control – Looking at the control variables in columns (2), (4) and (6) a large number of significant effects are shown. Firstly, evidence of the trade-off between profitability and outreach, as suggested by Cull, Demirgüç-Kunt, and Morduch (2007), is found. Looking at the relationship between the profit margin (𝑃𝑀𝑖𝑡) of the MFI and the average loan balance per borrower (𝐴𝐿𝐵𝑖𝑡), significant evidence is
presented that an increase of the profit margin of the MFI increases the average loan balance of the MFI, i.e. it decreases the MFI’s outreach. However, evidence of this trade-off between profitability and outreach is not found when looking at the other two proxies for outreach (i.e. 𝑁𝐴𝐵𝑖𝑡 and
𝐹𝐸𝑀𝑖𝑡).
The cost per borrower (𝐶𝐵𝑖𝑡) is significant for all three dependent variables. Columns (2), (4) and (6)
provide evidence that an increase in the cost per borrower decreases the outreach of the MFI since (𝐴𝐿𝐵𝑖𝑡) increases and (𝑁𝐴𝐵𝑖𝑡) and (𝐹𝐸𝑀𝑖𝑡) decrease. This results confirms the expectations that
higher costs per borrower make it relatively more expensive to reach out to poorer borrowers.
Based on column (4), only for the average loan balance per borrower (𝐴𝐿𝐵𝑖𝑡) does an increase in the
25 this is also suggested in column (6). The relatively bigger the loan portfolio of the MFI the less profitable it becomes to reach out to poorer borrowers.
Looking at the value of all loans outstanding that are at substantial risk of default in the next 30 days (𝑅𝐼𝑖𝑡), it significantly lowers the total number of active borrowers (𝑁𝐴𝐵𝑖𝑡) as well as the percentage
of female borrowers (𝐹𝐸𝑀𝑖𝑡). Moreover, it increases the average loan balance per borrower
(𝐴𝐿𝐵𝑖𝑡). Therefore, it can be concluded that the higher the loans that are at substantial risk of default in the next 30 days the less the MFIs reach out to poorer borrowers. MFIs reduce the risk of default of the portfolio by providing less and bigger loans.
Looking at the dummy variable non-profit (𝑁𝑃𝑖𝑡), indicating whether or not the MFI has a non-profit
status, it can be concluded that non-profit MFIs have a significantly lower number of active borrowers (𝑁𝐴𝐵𝑖𝑡) and a smaller average loan balance per borrower (𝐴𝐿𝐵𝑖𝑡). Moreover, non-profit
MFIs have a significantly higher percentage of female borrows (𝐹𝐸𝑀𝑖𝑡). So, although they are
significantly smaller than for-profit MFIs, non-profit MFIs have a larger outreach to poorer borrowers indicated by the lower number of active borrowers (𝑁𝐴𝐵𝑖𝑡) and the higher percentage of female
borrows (𝐹𝐸𝑀𝑖𝑡) compared to for-profit MFIs. This result is expected, because the fact that these
MFIs do not target a profit allows them to reach out to poorer borrowers. Nevertheless, non-profit MFIs do reach out to a smaller group of borrowers.
Caveats – In analysing the regression estimations presented in table 5 some important caveats can be signalled. First, and most importantly, this thesis uses panel data comprising both time series and cross-sectional elements (i.e. MFIs), and, therefore, pooling the data and using OLS has some major drawbacks. Most importantly, pooling the data implicitly assumes that the average values of the variables and the relationships between them are constant over time and across all the cross-sectional units in the sample (Brooks, 2014). The results presented above should therefore be interpreted with caution.
26 Finally, the estimations presented in table 5 do not include any interaction terms. The advantage of using interaction terms is that the effect of an explanatory variable on the dependent variable can be specified for a particular subgroup compared to those that do not belong to this subgroup. As mentioned in Section 3.3. this thesis measures formal financial inclusion (𝐹𝐼𝑡) for commercial (for
profit) financial institutions. It is, therefore, interesting to test whether the effect of formal financial inclusion on outreach is significantly different for MFIs that are for-profit or non-profit. The presence of a significant interaction will indicate that the effect of formal financial inclusion on the outreach of MFIs to poorer borrowers is significantly different for MFIs that have a non-profit status or that have a profit status.
6.3. Panel Data Models
This section describes the estimation results of the panel data models for the three different dependent variables as described in equations 8-10. Since the non-profit dummy variable (𝑁𝑃𝑖𝑡) is
time invariant, it will drop out in the fixed effects estimations and, therefore, the variable is not included in the panel data estimations for the average loan balance per borrower (𝐴𝐿𝐵𝑖𝑡) and the
percentage of female borrowers (𝐹𝐸𝑀𝑖𝑡). Based upon the suggestions made in Section 6.2. the panel
data models will include the interaction term between formal financial inclusion and the non-profit dummy variable (𝐹𝐼𝑁𝑃𝑖𝑡). Moreover, the panel data models performed in this section will test for the
presence of heteroskedasticity and include robust standard errors if needed. Furthermore, similar to the OLS, it was found that, especially for 𝐴𝐿𝐵𝑖𝑡, the years in which the MFIs are present in the data
are not homogeny and of significant influence on the results and, therefore, the preferred specification includes year dummies to account for these time effects.
6.3.1. Attrition bias
Since the panel used in this thesis is unbalanced, first it should be tested whether non-random attrition is present. Using a unbalanced panel without correcting for selectivity bias, may result in biased estimates if the nonresponse is endogenously determined (Verbeek and Nijman, 1992). In order to test for selectivity nonresponse bias in the panel data models, the Nijman-Verbeek test (1992) is performed. The results of the tests, for all three measures of outreach, are shown in table A2 in the appendix. 𝐹𝑌𝑃𝑅𝐸𝑉𝑖𝑡 denotes the dummy variable for which 𝐹𝑌𝑃𝑅𝐸𝑉𝑖𝑡 = 1 indicates that
in year t, in the previous year t-1 the MFI was present in the dataset and 𝐹𝑌𝑃𝑅𝐸𝑉𝑖𝑡 = 0 if otherwise.
Since the dataset starts in 2004, and 2003 does not exist, 𝐹𝑌𝑃𝑅𝐸𝑉𝑖𝑡 is always unidentified in 2004.
Similar, 𝐹𝑌𝑁𝐸𝑖𝑡 indicates whether in year t, in the next year t+1 the MFI is present in the dataset.
27 Table A2 in the appendix shows that for the number of active borrowers (𝑁𝐴𝐵𝑖𝑡) and for the
percentage of female borrowers (𝐹𝐸𝑀𝑖𝑡), 𝐹𝑌𝑃𝑅𝐸𝑉𝑖𝑡 and 𝐹𝑌𝑁𝐸𝑖𝑡 are not significant indicating that
attrition bias can be ignored in the estimations for these two dependent variables. Only when looking at the number of active borrowers (𝑁𝐴𝐵𝑖𝑡) as dependent variable, 𝐹𝑌𝑁𝐸𝑖𝑡 is significant at the 1%
level. Therefore, as shown below, an adjustment is made to the estimations for 𝑁𝐴𝐵𝑖𝑡.
6.3.2. NAB
Table 6 starts with a random effects model between the number of active borrowers (𝑁𝐴𝐵𝑖𝑡) and
formal financial inclusion (𝐹𝐼𝑡). Furthermore, the Hausman test conducted in this section was found
to be negative. Since the coefficients are extremely consistent, random effects was chosen as the preferred estimation. Therefore, all estimation results in this section are random effects estimates. Column (1) shows the most simple model. Column (2) presents the random effects model with all the controls but including the interaction effect, column (3) presents the random effects model with the controls, interaction and year dummies and, finally, column (4) presents the random effects model with controls, interaction, year dummies and robust standard errors.
However, when using the likelihood ratio test for the random effects model presented in column (3) the tests failed to find presence of heteroskedasticity. Therefore, column (3) and not column (4) is here the preferred specification. Nevertheless, when testing for attrition, it became evident, from table A2 in the appendix, that 𝐹𝑌𝑁𝐸𝑖𝑡 is significant and, therefore, non-random attrition is present.
In order to solve for this problem, the Nijman-Verbeek test (1992) was redone using the same sample but excluding the smallest MFI, which makes the examined sample smaller. The smallest MFI was selected based on its assets and gross loan portfolio. As shown in table A3, 𝐹𝑌𝑁𝐸𝑖𝑡 is no longer
28 Table 6: Estimation results for panel data models with number of active borrowers as dependent variable
[t-values between brackets]
Re: random effects, Rer, random effects with robust standard errors, Ret: random effects with year dummies, Retr: random effects with year dummies and robust standard errors, Retr*: random effects with year dummies and robust standard errors based on sample excluding smallest MFI in order to account for non-random attrition
NAB (1)Re (2)Rer (3)Ret (4)Retr (5) Retr*
Financial inclusion 0.3347 0.4564 0.4930 0.4930 0.6019 [13.19]*** [4.58]*** [8.20]*** [4.48]*** [7.56]*** Profit margin (%) -0.2662 -0.3059 -0.3059 -0.1968 [-0.95] [-1.36] [-1.19] [-0.95] Debt-to-equity (%) 0.0005 -0.0005 -0.0005 -0.0015 [0.14] [-0.15] [-0.16] [-0.45]
Cost per borrower ($) -0.3270 -0.3711 -0.3711 -0.7210
[-1.94]* [-4.59]*** [-1.81]* [-4.53]*** Loan portfolio/asstes (%) 1.0217 0.9647 0.9647 0.8908 [1.89]* [2.35]** [1.53] [1.67]* Risk (%) -0.0701 -0.0513 -0.0513 -0.0107 [-1.05] [-1.45] [-0.80] [-0.22] Non-profit -1.5423 -1.6072 -1.6072 -1.7129 [-2.78]*** [-3.95]*** [-2.88]*** [-3.35]*** FINP -0.0813 -0.0842 -0.0842 -0.0941 [-0.68] [-1.45] [-0.69] [-0.85] year2 0.1243 0.1243 0.0664 [1.01] [1.83]* [1.01] year3 0.2118 0.2118 0.2068 [1.78]* [2.72]*** [2.40]** year4 0.2483 0.2483 0.2349 [2.20]** [2.39]** [2.26]** year5 0.3654 0.3654 0.3668 [3.39]*** [3.77]*** [3.96]*** year6 0.1897 0.1897 0.1625 [1.84]* [1.97]** [1.83]* year7 0.1844 0.1844 0.2029 [1.92]* [2.34]** [2.78]*** year8 0.3014 0.3014 0.3749 [2.91]*** [3.47]*** [5.28]*** year9 0.3564 0.3564 0.4251 [3.40]*** [4.36]*** [6.41]*** Cons 9.3096 11.2215 11.4164 11.4164 13.7388 [28.77]*** [9.23]*** [14.76]*** [7.75]*** [13.23]*** N 224 201 201 201 198 R2 (within) 0.4702 0.5671 0.6302 0.6302 0.6937 R2 (overall) 0.0593 0.3936 0.3852 0.3852 0.4195
29 From column (5) it can be concluded, looking at the dummy variable 𝑁𝑃𝑖𝑡, that non-profit MFIs have
a significantly lower number of active borrowers than for-profit MFIs. Moreover, column (5) shows that an increase in formal financial inclusion (𝐹𝐼𝑡) significantly increases the number of active
borrowers (𝑁𝐴𝐵𝑖𝑡) for all MFIs. Looking at the interaction term (𝐹𝐼𝑁𝑃𝑖𝑡), when formal financial
inclusion increases, the number of active borrowers increases less for non-profit MFIs than for-profit MFIs. However, this effect is not significantly smaller for non-profit MFIs. The findings indicate that for all MFIs, for-profit and non-profit, when formal financial inclusion increases their outreach to poorer borrowers also increases. So, instead of being a substitute to formal financial inclusion, the results found here indicate that the MFIs are a complimentary to the increased formal financial inclusion of commercial banks.
The estimation results here do not indicate evidence of a trade-off between the profit margin of the MFI and outreach. Although the sign indicates that a higher profit margin lowers the number of active borrowers, this result is not significant, similar to the result found by Cull, Demirgüç-Kunt, and Morduch (2007). The debt-to-equity ratio (𝐷𝐸𝑖𝑡) is not found to significantly increase the level of
outreach to poorer borrowers which does not confirm the suggestions put forward by Armendáriz and Morduch (2010) and Kyereboah-Coleman (2007). The results presented in column (5) suggest a trade-off between a higher debt level compared to the equity level and the outreach of the MFI. This may be because the MFI already faces a too high risk and repayments and therefore cannot accommodate more borrowers. The 𝐶𝐵𝑖𝑡 is highly negatively significant. Higher cost per borrower as
decreases the number of active borrowers for the MFI. This results from the fact that it is relatively more expensive to serve the poor. Finally, column (5) shows a small positive significant effect for the loan portfolio to total assets (𝐿𝑃𝑖𝑡). The relatively higher the loan portfolio the better the MFI can
accommodate an increase in its number of active borrowers. 6.3.3. ALB and FEM
ALB – Table 7 starts with a random effects model between the average loan balance per borrower (𝐴𝐿𝐵𝑖𝑡) and formal financial inclusion (𝐹𝐼𝑡) (column (1)), since the null hypothesis of random effects
30 Table 7: Estimation results for panel data models with average loan balance per borrower ($) as
dependent variable [t-values between brackets]
Re: random effects, Fer: fixed effects with robust standard errors, Fet: fixed effects with year dummies, Fetr: fixed effects with year dummies and robust standard errors
ALB (1)Re (2)Fer (3)Fet (4)Fetr
Financial inclusion 0.3394 0.1988 0.2171 0.2171 [14.70]*** [4.38]*** [6.74]*** [5.37]*** Profit margin (%) 0.1350 0.1011 0.1011 [1.89]* [0.84] [1.02] Debt-to-equity (%) -0.0021 -0.0023 -0.0023 [-1.03] [-1.22] [-1.18]
Cost per borrower ($) 0.6471 0.6472 0.6472
[7.63]*** [14.49]*** [8.07]*** Loan portfolio/asstes (%) 0.5309 0.5119 0.5119 [2.66]** [2.32]** [2.38]** Risk (%) -0.0104 0.0030 0.0030 [-0.53] [0.16] [0.14] FINP -0.0913 -0.0949 -0.0949 [-1.70] [-3.06]*** [-1.77]* year2 0.0554 -0.0554 [0.84] [1.01] year3 0.1089 0.1089 [1.71]* [1.93]* year4 0.1819 0.1819 [3.01]*** [3.27]*** year5 0.1388 0.1388 [2.41]** [2.23]** year6 0.0780 0.0780 [1.41] [1.84]* year7 0.0928 0.0928 [1.81]* [1.84]* year8 0.1587 0.1587 [2.87]*** [3.21]*** year9 0.1548 0.1548 [2.76]*** [3.87]*** Cons 7.4131 3.3841 3.3495 3.3494 [35.73]*** [7.02]*** [9.33]*** [7.27]*** N 219 201 201 201 R2 (within) 0.5278 0.8290 0.8465 0.8465 R2 (overall) 0.0981 0.8273 0.8320 0.8320
*=significant at 10%; **=significant at 5%; ***=significant at 1%
Based on the caveats put in Section 6.2., column (4) provides the most accurate estimation on the effect of formal financial inclusion on the dependent variables. From column (4) it can be concluded that an increase in formal financial inclusion (𝐹𝐼𝑡) significantly increases the average loan balance per
31 the interaction term (𝐹𝐼𝑁𝑃𝑖𝑡), significant evidence at the 10% level is found that for non-profit MFIs
the average loan balance per borrower increases less than for for-profit MFIs when formal financial inclusion increases. From these results it can be concluded that an increase in formal financial inclusion leads to a lower depth of outreach by MFIs, but only significantly for for-profit MFIs. This result confirms partly hypothesis 2 in that an increase in formal financial inclusion leads to MFIs having a lower share of poor borrowers. These findings suggest that a trade-off between profitability and outreach exists and that for-profit MFIs experience mission drift and become too focused on making profits, reducing their outreach to poor borrowers (Cull, Demirgüç-Kunt, and Morduch, 2007). This suggestion is supported by the fact that the relationship between the profitability of the MFI (𝑃𝑀𝑖𝑡) and the average loan balance per borrower (𝐴𝐿𝐵𝑖𝑡) is positive although insignificant.
In contrast to the paper of Mersland and Strøm (2010), evidence is found that MFIs experience mission drift, but only for for-profit MFIs. It is found that profit oriented MFIs drift away from their original purpose of serving the poor and move into new customer segments in order to become more profitable. Moreover, dissimilar to Tchakoute-Tchuigoua (2010), it was not found that profit oriented MFIs achieve better social performance than non-profit MFIs. Even though for-profit MFIs can collect deposits their lending was not increased to poorer borrowers. Based on Hermes, Lensink and Meesters (2011), it can be argued that for-profit MFIs decrease their lending to poorer borrowers to become more efficient and increase their profits.
Looking at the control variables, the debt-to-equity ratio (𝐷𝐸𝑖𝑡) is not found to increase the level of
outreach to poorer borrowers which does not confirm the suggestions put forward by Armendáriz and Morduch (2010) and Kyereboah-Coleman (2007). The cost per borrower (𝐶𝐵𝑖𝑡) is positive
significant related to 𝐴𝐿𝐵𝑖𝑡 indicating that a higher cost per borrower leads to MFIs having a higher
average loan balance per borrower in order to reduce costs. The loan portfolio to total assets (𝐿𝑃𝑖𝑡) is
positive significant indicating that the higher their loan portfolio to total assets the less risk MFIs take and the less they reach out to more riskier poorer borrowers. Finally, the value of the loans outstanding that are at substantial risk of default in the next 30 days (𝑅𝐼𝑖𝑡) does not significantly
decrease the outreach of the MFIs to poorer borrowers.
FEM – The results of the panel data estimations are presented in table A4 which can be found in the appendix. Since no significant results were found on the effect of formal financial inclusion (𝐹𝐼𝑡) on
the percentage of female borrowers of the MFI (𝐹𝐸𝑀𝑖𝑡), this model will not be discussed further.
However, it should be mentioned that, in contrast to the OLS, suggestive evidence for a decrease in outreach by MFIs is found. Similar as for 𝐴𝐿𝐵𝑖𝑡, the results shown in table A4 suggest that evidence
32 6.4. Extensions and suggestions
The dataset used in this thesis is clustered, with regression errors and regressors correlated within the clusters. The usual solution to control for clustering is to use cluster-robust standard errors, but this only works when the sample is large enough. The dataset used in this thesis consist of 28 MFIs which is just below the asymptotic threshold of around 30 clusters (Cameron, Gelbach and Miller, 2008). When the number of clusters is too low, standard asymptotic tests can over-reject, because they are not conservative enough and, thus, they underestimate the true standard errors. The statistical power of the tests calculated is too low and type 2 errors, in which results may be called insignificant even if they are actually significant, may occur (Armendáriz and Morduch, 2010). Cameron, Gelbach and Miller (2008) therefore suggest in their paper to use cluster bootstrap-t procedures that provide asymptotic refinement. More specifically, they recommend to use wild bootstrap-t procedures to obtain more accurate cluster-robust inferences. This suggestion is, however, beyond the scope of this thesis and it is recommended to account for a too small sample size in further research.
Within the scope of this thesis a robustness check for the estimated results is done. MixMarket provides amongst others a diamond rating per observation indicating the MFI’s level of transparency and supporting documentation for all data. Four diamonds indicates that complete data including audited financial statements are disclosed and five diamonds indicates a diamond level four and a rating or due diligence report is published for the year (MixMarket, 2015). Based on Kwon (2010) and Ahlin, Lin, and Maio (2011), this thesis also performs the estimations discussed with only observations from MFIs with four or five diamond ratings. The initial dataset is therefore redefined to a dataset of 181 observations. Moreover, the number of MFIs is reduced to 24. The descriptive statistics of this extension and the results of the panel estimations for the preferred specification can be found in Appendix 41. This section only looks the outputs for the preferred panel effects estimations. Overall the results from the entire panel and from the panel consisting of data from four and five diamond are comparable. The only difference is that now for all three dependent variables random effects is the preferred specification. Moreover, when testing for attrition, 𝐹𝑌𝑁𝐸𝑖𝑡 is
significant at the 10% level with 𝑁𝐴𝐵𝑖𝑡. However, since it is only significant at the 10% level and
omitting the smallest MFI does not change this result, attrition is further ignored. Furthermore, similar to the panel estimations for the entire panel, no presence of heteroskedasticity was found and table A8 presents both the random and robust random estimations for 𝑁𝐴𝐵𝑖𝑡. The comparable
results indicate that the results discussed for the entire panel are robust and are not influenced by faulty or incomplete data.
1
33
7. Discussion and conclusions
This thesis has used OLS and fixed and random effect models to examine whether a relationship exists between increased formal financial inclusion and the outreach of MFIs to poorer borrowers. Using a sample of 228 observations and looking at a ten year time period, conflicting evidence is found on whether increased formal financial inclusion leads to a higher depth of outreach of MFIs. More specifically, in essence two research questions have been answered. Looking at the number of active borrowers of the MFI, it can be asked whether the outreach of MFIs is complementary or substitute to formal financial inclusion. Looking at the average loan balance per borrower and the percentage of female borrowers, it can be asked whether increases formal financial inclusion leads to mission drift of MFIs.
With respect to the first question, significant evidence is found that an increase in formal financial inclusion leads to all MFIs reaching out to more borrowers. Although non-profit MFIs have significantly less borrowers than for-profit MFIs, for both for-profit and non-profit MFIs their outreach to more borrowers increases. This result indicates that there is no substitution between in formal financial inclusion and the outreach of MFIs but that they are in fact complementary and go hand in hand. However, it should be noted that it cannot be said with certainty whether these new borrowers actually are poorer borrowers. Although this result suggests to confirm the first hypothesis, further research is needed to determine more precisely if the MFIs reach out to poorer borrowers and what consequences this may have for the MFIs.
34 female borrowers suggested, similar with the average loan balance per borrower, that for-profit MFIs may experience relatively more mission drift than non-profit MFIs.
Limitations to this research are that the IMF’s Financial Access Survey (FAS) database did not provide data prior to 2004 on Bolivia. By extending the research over a longer time period, including the microfinance crisis in Bolivia, possible developments in outreach to the poor may have been discovered. However, economic growth in Bolivia and the resulting increase in formal financial inclusion, which provide the framework for this research, are more of the last ten years, which justifies the time period used in this research. Moreover, this research only focuses on Bolivia. Bolivia has a highly regulated and competitive microfinance sector and the results provided in this thesis may therefore differ for other countries. Furthermore, when extending the research to more countries, for example Latin America as a whole, the variation in formal financial inclusion will be at the country level which may provide more interesting insights. Additionally, extending the research to more countries will increase the number of MFIs reducing the problem of over-rejection discussed in Section 6.4. Finally, similar to the suggestions put forward by Mersland and Strøm (2010), additional insights may be given in the outreach of MFIs and their mission to serve the poor by looking at MFIs that are transformed from non-profit to for-profit. Unfortunately, this data was not the case in the dataset used in this thesis from MixMarket (2015).
35
Appendix
A.1. Correlation table
Table A1:Correlation explanatory and dependent variables. N=188
NAB ALB FEM FI PM DE CB LP RI
36 A.2. Attrition
Table A2: Nijman-Verbeek test for attrition bias [t-values between brackets]
NAB ALB FEM
Financial inclusion 0.5718 0.2338 -0.0244 [9.05]*** [6.15]*** [-1.62] Profit margin (%) -0.2320 0.1866 0.0385 [-1.16] [1.55] [0.82] Debt-to-equity (%) -0.0007 -0.0017 0.0005 [-0.22] [-0.91] [0.74]
Cost per borrower ($) -0.3484 0.6335 -0.0175
[-4.77]*** [14.40]*** [-1.00] Loan portfolio/assets (%) 1.1447 0.6487 -0.1029 [3.43]*** [3.23]*** [-1.31] Risk (%) -0.0266 -0.0094 -0.0083 [-0.86] [-0.50] [-1.13] FINP -0.1277 -0.1087 0.0244 [-1.79]* [-2.52]** [1.43] FYPREV -0.1136 0.0552 0.0006 [-0.52] [0.42] [0.01] FYNE 0.5769 -0.0422 -0.0318 [2.78]*** [-0.34] [-0.65] Cons 10.0629 3.3656 0.7227 [17.74]*** [9.85]*** [5.38]*** N 172 172 170 R2 (within) 0.5749 0.8309 0.0795 R2 (overall) 0.0577 0.8248 0.2039
37 Table A3: Nijman-Verbeek test for attrition bias omitting smallest MFI [t-values between brackets]
NAB ALB FEM
Financial inclusion 0.6699 0.1981 -0.0280 [10.24]*** [4.85]*** [-1.71]* Profit margin (%) -0.1310 0.1974 0.0512 [-0.54] [1.30] [0.85] Debt-to-equity (%) -0.0014 -0.0014 0.0006 [-0.47] [-0.77] [0.77]
Cost per borrower ($) -0.6106 0.7259 -0.0085
[-6.34]*** [12.07]*** [-0.35] Loan portfolio/assets (%) 1.2032 0.6174 -0.1080 [3.75]*** [3.08]*** [-1.35] Risk (%) -0.0134 -0.0134 -0.0085 [-0.45] [-0.72] [-1.14] FINP -0.1251 -0.1047 0.0258 [-1.79]* [-2.39]** [1.46] FYPREV -0.0730 0.0460 0.0010 [-0.35] [0.35] [0.02] FYNE 0.2300 0.0748 -0.0222 [1.06] [0.55] [-0.41] Cons 11.8434 0.0748 0.6713 [17.14]*** [0.55] [3.85] N 169 169 167 R2 (within) 0.6175 0.8147 0.0642 R2 (overall) 0.0335 0.8644 0.0759
38 A.3. Fixed effects model FEM
Table A4: Estimation results for panel data models with percentage of female borrowers (%) as dependent variable [t-values between brackets]
Re: random effects, Fer: fixed effects with robust standard errors, Fet: fixed effect with year dummies, Fetr: fixed effects with year dummies and robust standard errors
FEM (1)Re (2)Fer (3)Fet (4)Fetr
Financial inclusion -0.0064 -0.0189 -0.0135 -0.0135 [-1.21] [-1.20] [-1.10] [-0.91] Profit margin (%) 0.0574 0.0565 0.0565 [1.17] [1.31] [1.17] Debt-to-equity (%) 0.0006 0.0005 0.0005 [1.95]** [0.78] [1.77]*
Cost per borrower ($) -0.0275 -0.0307 -0.0307
[-1.08] [-1.92]* [-1.28] Loan portfolio/asstes (%) -0.0986 -0.1004 -0.1004 [-1.10] [-1.27] [-1.01] Risk (%) -0.0096 -0.0086 -0.0086 [-1.17] [-1.24] [-0.97] FINP 0.0265 0.0263 0.0263 [1.36] [2.29]** [1.34] year2 0.0050 0.0050 [0.21] [0.43] year3 0.0367 0.0367 [1.59] [1.71]* year4 0.0061 0.0061 [0.28] [0.34] year5 -0.0047 -0.0047 [-0.22] [-0.23] year6 0.0031 0.0031 [0.15] [0.23] year7 0.0102 0.0102 [0.55] [0.76] year8 0.0055 0.0055 [0.28] [0.61] year9 0.0017 0.0017 [0.08] [0.16] Cons 0.5428 0.7370 0.7528 0.7528 [14.91]*** [4.63]*** [5.81]*** [4.76]*** N 209 196 196 196 R2 (within) 0.0082 0.1087 0.1388 0.1388 R2 (overall) 0.0001 0.3775 0.3740 0.5779
39 A.4. Diamonds
Table A5: Description of the panel (MFIs per year) based on panel of 4 and 5 diamond observations
Year Number of MFIs per year
2004 13 2005 15 2006 18 2007 20 2008 21 2009 20 2010 19 2011 20 2012 20 2013 15 Total 181 Source: MixMarket (2015)
Table A6: Number of years that a MFI is in the dataset based on panel of 4 and 5 diamond observations
Number of years Number of MFIs
1 1 2 0 3 2 4 2 5 0 6 1 7 3 8 4 9 3 10 8 Total 24
40 Table A7: Description of the MFIs divided on the basis of their target
market, legal status, outreach, profit status and regulation status (number and percentage of MFIs per characteristic) based on panel of 4 and 5 diamond observations
Number of MFIs Percentage
Target market High end 53 29.8%
Low end 11 6.2%
Small 8 4.5%
Broad 106 59.6%
Legal status Cuco 3 1.7%
NBFI 0 0.0%
NGO 120 66.7%
Bank 57 31.7%
Outreach Medium/Small 104 58.8%
Large 73 41.2%
Profit status Non-profit 123 68.0%
For-profit 58 32.0%
Regulation status Not regulated 121 66.9%
Regulated1 60 33.1%
41 Table A8: Estimation results for panel data models with number of active borrowers, average
loan balance per borrower ($), or percentage of female borrowers (%) as dependent variable based on panel of 4 and 5 diamond observations [t-values between brackets] Re: random effects, Rer: random effects with robust standard errors
NAB ALB FEM
Re Rer Rer Rer
Financial inclusion 0.6146 0.6146 0.1836 -0.0067 [10.52]*** [7.41]*** [4.81]*** [-0.36] Profit margin (%) -0.0846 -0.0846 0.1453 0.05991 [-0.36] [-0.45] [1.26] [1.21] Debt-to-equity (%) -0.0018 -0.0018 -0.0029 0.0002 [-0.60] [-0.67] [-2.15]** [0.72]
Cost per borrower ($) -0.6009 -0.6009 0.7874 -0.0729
[-5.87]*** [-4.45]*** [6.76]*** [-1.84]* Loan portfolio/assets (%) 0.8906 0.8906 0.6256 -0.0784 [2.41]** [1.91]* [2.45]** [-0.75] Risk (%) -0.0062 -0.0062 0.0062 -0.0157 [-0.18] [-0.14] [0.31] [-1.93]* Non profit -1.8238 -1.8238 -0.6214 0.0793 [-4.63]*** [-3.33]*** [-3.76]*** [1.93]* FINP -0.1062 -0.1062 -0.1028 0.0285 [-1.96]** [-3.33]*** [-2.16]** [1.48] year2 0.0811 0.0811 0.0873 -0.0068 [0.76] [1.39] [1.75]* [-0.58] year3 0.1735 0.1735 0.0991 0.0287 [1.65]* [2.23]** [1.94]* [1.48] year4 0.2412 0.2412 0.1823 -0.0001 [2.47]** [2.15]** [3.25]*** [-0.01] year5 0.3588 0.3588 0.1687 -0.0025 [3.91]*** [3.74]*** [2.63]*** [-0.12] year6 0.1421 0.1421 0.1278 0.0110 [1.55] [1.49] [3.11]*** [0.84] year7 0.2096 0.2096 0.0894 0.0060 [2.37]** [2.58]*** [1.60] [0.50] year8 0.3372 0.3372 0.1556 0.0100 [3.67]*** [4.49]*** [3.34]*** [0.89] year9 0.3832 0.3832 0.1508 0.0076 [4.08]*** [5.17]*** [3.99]*** [0.63] Cons 13.2524 13.2524 2.9204 0.8786 [15.68]*** [17.76]*** [4.27]*** [3.04]** N 169 169 169 167 R2 (within) 0.7469 0.7469 0.8462 0.1952 R2 (overall) 0.4341 0.4341 0.8752 0.5319
