Women and repayment performances in Microfinance
Amsterdam Business School
Name Roelien Timmer
Student number 10432914
Program Economics & Business Specialization Economics and Finance Number of ECTS 12
Supervisor Ilko Naaborg
Abstract
Micro Financial Institutions focus more and more on women, because women are viewed as more reliable clients. This paper investigates of this belief is justified on the basis of repayment performances. Three indicators for these repayment performances are selected. These indicators are the portfolio at risk > 30 days, portfolio at risk > 90 days and the write-off rate. A data analysis is constructed on the basis of panel data. Data of 2367 MFIs across 120 different countries worldwide are analyzed. The data is from 2000 till 2010. The results show that a higher percentage of female borrowers is related to lower portfolio at risk and lower write-off rates. From that prospective can be concluded that women have indeed better repayment performances.
Keywords: microfinance, gender, portfolio at risk, write-off rate, repayment
Table of Contents
1. Introduction 2
2. Literature review 4
3. Hypothesis, Methodology and Data 6
3.1. Hypothesis 6
3.2. Methodology 8
3.3. Data and descriptive statistics 12
4. Analysis 14
4.1. Empirical Results 14
4.2. Robustness check 17
5. Conclusion and discussion 18
5.1 Conclusion 18
5.2 Discussion 18
5.3 Limitations 19
5.4 Further research 20
1. Introduction
In the 1970s, Bangladesh was excited about the start of microfinance. Micro financing should be the solution to their everlasting poverty. Later on, the rest of Asia would also adopt this very promising instrument. The world was keeping an eye on Bangladesh and their results. The start in Bangladesh was promising and micro financing was expanding very fast. This new type of lending became very popular in Asia and later on in many more
developing countries.
The principle of microfinance is very simple. Banks provide small loans to entrepreneurs who do not have access to regular banking. Subsequently, the client invest their new money in their own small business. Step by step, the company is growing and becomes more competitive. Profits are rising and more money is invested. An upward spiral which improves the financial situation of the whole family.
Muhummad Yunus, born and raised in Bangladesh, is the pioneer of microfinance. He received a lot of rewards and media attention during the years. He and his Grameen bank are even rewarded by the Nobel Peace Prize in 2006. They are commended "for their efforts through microcredit to create economic and social development from below". From that day, the world paid even more attention to the development of the Grameen Bank and microfinance.
Unless the success of the Grameen Bank and the Nobel Peace Prize, microfinance is controversial. Critics claim that only the success stories are exposed. Microfinance would very often drive people into an endless situation. Clients use the borrowed money for wrong purposes. They increase their consumption instead of investing in long-term projects. Poor family’s often end up in a debt trap and are despaired. Critics claim that microfinance aggravate their financial situation instead of enhancing. Johnson and Kiddler share these concerns (1999). They declare the problems by the idea that not all borrowers are born entrepreneurs.
Due to all the criticism, microfinance is still in process of developing. Researchers are looking for ways to improve the system. Group lending is one of these important
developments. Team wise lending increases the sense of responsibility. Clients would try harder not to fail and the repayment rates are substantially higher.
Another important development is providing more loans to women. Women are viewed to be more capable regarding loans. They use their loans more efficient and show higher repayment rates. Women are more risk averse and focus more on the well-being of their brood. But, most often micro financing is viewed as a way to empower women. By providing small loans to women, they can become financially independent. They obtain more self-esteem and a higher status in the family.
Nowadays, many Micro Financial institutions (MFIs) provide only loans to women. The Grameen bank provides loans almost exclusively to women. In the dataset used by this research 62% of the borrowers are women. This is even more impressive if you realize that plenty of women refuse micro financing. There is still an upward trend going on by providing more loans to women.
In 1997, the Micro-credit Summit in Washington reported:
“The time has come to recognize microcredit as a powerful tool in the struggle to end poverty and economic dependence. We have assembled to launch a global campaign to reach 100 million of the world’s poorest families, especially the women of those families, with credit for self-employment and other financial and business services by the year 2005”.
MFIs belief more and more that women are better lenders. In developing countries, it becomes harder to find affordable loans for men. Therefore, it is very important to make sure that this belief of MFIs is justified. This paper will investigate of women are indeed better microfinance clients.
Several papers have already written about the relationship between gender and micro financing. Nevertheless, this paper is renewing and interesting. The research will be conducted on a very extended worldwide database, which has never done before.
Therefore, the results give a more reliable result than all the other researches until now. Besides, a new indicator of the success of repayment rates will be analysed. This indicator contains to portfolio at risk for more than 90 days. More about this indicator will be discussed later. After this research is conducted, MFIs will be more sure if their policy of targeting female clients is justified.
This research will focus on the success of microfinance for women. The indicator used for this success is the portfolio at risk. Portfolio at risk a measure for the repayment of loans. To be more concrete, the question this paper will answer is: “Do female clients have better repayment scores than men?” A big database of 2367 MFIs of 120 different countries worldwide will be analyzed.
This introduction will be followed by a literature review. Previous literature will be discussed with respect to this research. Thereafter, the hypothesis will be stated with the
corresponding methodology. Details about the dataset will be given, including their source. Consequently, the empirical result will be given along with its interpretation. At the end, a conclusion will be presented followed by a discussion. This discussion will contain the limitations of this research and a proposed future research.
2. Literature review
Previous research showed different results about the repayment performances of women. A lot of research is performed on country level. First, the previous results will be discussed. This will be followed by potential reasons.
A lot of research has been done in Bangladesh, because of the popularity of microfinance this country. Husain reported about the Grameen Bank in Bangladesh (1988). A survey study of 975 lenders of the bank was conducted. The results founded was that 81% of the male clients had repayment problems. Only 71% of the female clients reported to have
repayment problems. Khandker conducted the same type of research years later (1995). This research was also conducted for the Grameen Bank. The repayment differences between men and women have not changed. Sharma did research about the repayment
rates of group lending in Bangladesh as well (1997). The results showed that groups with more female members have better repayment performances.
Godquin did also research about the repayment rates Bangladesh (2004). Godguin used more control variables. His results do not show significant higher repayment rates for women. Although there is a small positive relationship, but this relationship is not significant.
D’espaillier is the only research conducted on a worldwide level about microfinance and gender (2011). He founded that a higher percent of female borrowers is accompanied by a lower portfolio at risk. By the use of an ordinary least square regression, there was a statistical significant relation of 5%. Besides, D’espaillier found that MFIs with a higher percentage of female clients have lower write-off rates. This negative relation also holds for portfolio at risk for 30 days. This would suggest that women have better repayment results. A similar model to the one of D’espaillier will be used for this research. More about this will be discussed in the methodology.
In the discussed papers above is clear that most research find a positive relation between women and repayment performance. Next, the potential reason behind this behavior will be discussed.
First, one of the big differences between female and male entrepreneurs is their motivation. Women are more motivated to become an entrepreneur for family reasons (DeMartino et al., 2003). While men are more motivated to become successful and to obtain a higher status.
One of the important characteristics of women is their risk averse behavior. Jianakoplos et al. conducted a research about the risk taking of U.S. households (1998). They compared the holding of risky assets between gender. The results suggested that single women are more risk averse than single men. Women settle down for lower returns in exchange for lower risk. This preference for financial stability could lead to higher repayment rates of loans.
Besides, most women have very limited experience in doing business. Therefore, they are very cautious in the beginning. They prefer to start with less risky projects (Sharma, 1997). Women use more of their income for their households. A report of Women’s
Entrepreneurship Development Trust Fund (WEDTF) showed that an increase in income benefits their children more than an increase of the income of men(2001). More money is invested in education, healthcare, clothing and healthy food. An increase in income is for 55 percent spend on household goods.
Besides, the WEDTF states that women tend to rely more on other entrepreneurs. Especially at the beginning of their borrowing period, women prefer to operate in groups. They can use the experience of their group members to build up confidence. In addition, women prefer to split the risk between the group members. This works two-sided. On the one hand, if a project fails the whole group is responsible. On the other hand, operating as a group works as an insurance. If the woman is on a certain moment forced to stop working, the group will continue to generate income for her. The tendency of women to cooperate could potentially lead to a more successful use of microloans.
3. Methodology and Data
3.1 Hypothesis
The main goal of this paper is to investigate the relationship between gender and repayment performances in microfinance. As stated above in the literature review, this relationship is not very clear. Due to the big consequences, this relation will be tested on a big scale.
This leads to our hypothesis of this research is:
This hypothesis will be investigated by three different indicators. These three indicators are the portfolio at risk > 30 days, portfolio at risk > 90 days and the write-off rate. This will also leads to three different hypotheses.
Portfolio at risk is an important indicator for the success of a loan. It is very often used as a measure of portfolio performance. Portfolio at risk is a very useful variable to access the performance of a MFI. If an instalment is late the loan is viewed to be risky. There is a substantial chance that the borrower will fail to pay off his debt. The portfolio at risk is computed by dividing the outstanding loans with payment arrears by the total amount of outstanding loans.
Portfolio at risk of 30 days is common used in microfinance. This implies the percentage of loans which is 30 days or more overdue. This leads to the first sub-hypothesis.
H1: MFIs with more female clients have less portfolio at risk > 30 days.
Portfolio at risk is a common measure and a good indicator for the risk. As stated earlier, there is substantial chance that the borrower will never pay off his debt. It could be that borrowers pay off their debt just after this 30 days limit. To check whether this is the case, a portfolio risk of 90 days will also be investigated. A portfolio at risk of 90 days is often denoted as a non-performing loan. Using portfolio at risk > 90 days as an indicator of repayment performances leads to the following sub-hypothesis.
H2: MFIs with more female clients have less portfolio at risk > 90 days
The third indicator of a successful repayment is the write-off rate. The lender decides to report a loan as a loss or an expense. This happens when the lender lost their faith in a repayment of the loan (mixmarket.org). The write-off rate is calculated by the write-off per year divided by the average gross loan portfolio over the year. The last sub hypothesis will be:
3.2 Methodology
All the three hypothesis will be tested on the basis of a similar model. The first model is the one where portfolio at risk > 30 days will be regressed. The following model is therefore constructed:
𝑃𝑃𝑃𝑃𝑃𝑃30𝑖𝑖,𝑡𝑡 = 𝛽𝛽0+ 𝛽𝛽1𝐹𝐹𝐹𝐹𝐹𝐹𝑖𝑖,𝑡𝑡 + 𝛽𝛽′2𝑍𝑍𝑖𝑖,𝑡𝑡+ 𝜇𝜇𝑖𝑖,𝑡𝑡
𝑃𝑃𝑃𝑃𝑃𝑃30𝑖𝑖,𝑡𝑡 stands for portfolio at risk > 30 days for MFI i in year t and is the independent variable of the model.
𝐹𝐹𝐹𝐹𝐹𝐹𝑖𝑖,𝑡𝑡 is the percentage of female borrowers of an MFI. In this paper the focus will be on the relation between the repayment performances and the percentage of female
borrowers.
𝑍𝑍𝑖𝑖,𝑡𝑡 represents seven different MFI-specific control variables. i stands for the specific MFI and t for the year in which this MFI is analysed. It is important to include variables which potentially affect the portfolio risk. The control variables will be shortly discussed below. 𝐴𝐴𝐴𝐴𝑖𝑖,𝑡𝑡 is the total of the net asset accounts of MFI. D’espaillier found a significant relation between the assets of a MFI and their portfolio at risk rate of 30 days. Besides, Sharma et al. found a significant negative relation between the loan size and the repayment
performances (1997). Therefore it is desirable to add 𝐴𝐴𝐴𝐴𝑖𝑖,𝑡𝑡 as a control variable.
𝐴𝐴𝐴𝐴 𝑖𝑖,𝑡𝑡 is computed by dividing the gross loan portfolio by the number of active borrowers. Gordon et al. find that the loan size is related to the repayment of microfinance loans (1965). The repayment decreases when the loan size increases. This variable will control the influence of the loan size.
𝐵𝐵𝑃𝑃𝐴𝐴 𝑖𝑖,𝑡𝑡 stands for active borrowers per staff member. Borrowers per staff is an indicator of the efficiency of a MFI. This terms is very often used to measure the productivity (MIX, 2005). The MIX emphasizes their worries about the efficiency of MFIs. The trade-off between the amount of borrowers per staff and the quality is discussed. More efficiency would lead to lower quality and lower repayment rates.
𝐴𝐴𝐴𝐴𝐹𝐹 𝑖𝑖,𝑡𝑡 is dummy variable which describes the experience of a MFI. 0 stands for a “new” MFI, 1 for a “young” MFI and 2 for a “mature” MFI. D’espaillier found a positive relationship between age and portfolio at risk. Therefore, including experience as a control variable would lead to a better fit of the model.
𝑁𝑁𝐴𝐴𝑁𝑁𝑖𝑖,𝑡𝑡 is a dummy which is 1 if the MFI is a 𝑁𝑁𝐴𝐴𝑁𝑁𝑖𝑖,𝑡𝑡 and 0 otherwise. A 𝑁𝑁𝐴𝐴𝑁𝑁𝑖𝑖,𝑡𝑡 is a non-profit organization registered for tax purposes. D’espaillier et al. researched NGOs and micro financing (2009). They found a significant positive relationship between NGOs and portfolio at risk >30 days. To control for this effect, NGO is added as a variable to the model. 𝑃𝑃𝑅𝑅𝑃𝑃𝐴𝐴𝐴𝐴𝑖𝑖,𝑡𝑡 is also dummy control variable. 𝑃𝑃𝑅𝑅𝑃𝑃𝐴𝐴𝐴𝐴𝑖𝑖,𝑡𝑡 is 1 if the MFI operates in a rural area and 0 otherwise. A rural area is a non-urban region where generally agricultural activities are conducted. D’espaillier found that portfolio at risk for 30 days and operating in a rural area are significant related (2010). Providing loans in rural areas is related to a lower portfolio at risk. The expectation is that loans in rural would bring more risk and lower repayment rates. But, D’espaillier found the opposite relation.
Finally, the error term is given by 𝜇𝜇𝑖𝑖,𝑡𝑡.
Replacing 𝑍𝑍𝑖𝑖,𝑡𝑡 and 𝑋𝑋𝑖𝑖∈𝐶𝐶,𝑡𝑡 by their corresponding control variables gives us the following equation:
𝑃𝑃𝑃𝑃𝑃𝑃30𝑖𝑖,𝑡𝑡 = 𝛽𝛽0+ 𝛽𝛽1𝐹𝐹𝐹𝐹𝐹𝐹𝑖𝑖,𝑡𝑡 + 𝛽𝛽′2𝐴𝐴𝐴𝐴𝑖𝑖,𝑡𝑡+ 𝛽𝛽′3𝐴𝐴𝐴𝐴 𝑖𝑖,𝑡𝑡 + 𝛽𝛽′4𝐵𝐵𝑃𝑃𝐴𝐴 𝑖𝑖,𝑡𝑡+ 𝛽𝛽′5𝐴𝐴𝐴𝐴𝐹𝐹 𝑖𝑖,𝑡𝑡+ 𝛽𝛽′6𝑁𝑁𝐴𝐴𝑁𝑁𝑖𝑖,𝑡𝑡+
𝛽𝛽′7𝑃𝑃𝑅𝑅𝑃𝑃𝐴𝐴𝐴𝐴𝑖𝑖,𝑡𝑡+ 𝜇𝜇𝑖𝑖,𝑡𝑡
For the second hypothesis, the exact same model will be regressed, except that the
𝑃𝑃𝑃𝑃𝑃𝑃30𝑖𝑖,𝑡𝑡 is replaced by 𝑃𝑃𝑃𝑃𝑃𝑃90𝑖𝑖,𝑡𝑡. 𝑃𝑃𝑃𝑃𝑃𝑃90𝑖𝑖,𝑡𝑡 stands for portfolio at risk > 90 days. This leads to the second regression model:
𝑃𝑃𝑃𝑃𝑃𝑃30𝑖𝑖,𝑡𝑡 = 𝛽𝛽0+ 𝛽𝛽1𝐹𝐹𝐹𝐹𝐹𝐹𝑖𝑖,𝑡𝑡 + 𝛽𝛽′2𝐴𝐴𝐴𝐴𝑖𝑖,𝑡𝑡+ 𝛽𝛽′3𝐴𝐴𝐴𝐴 𝑖𝑖,𝑡𝑡 + 𝛽𝛽′4𝐵𝐵𝑃𝑃𝐴𝐴 𝑖𝑖,𝑡𝑡+ 𝛽𝛽′5𝐴𝐴𝐴𝐴𝐹𝐹 𝑖𝑖,𝑡𝑡+ 𝛽𝛽′6𝑁𝑁𝐴𝐴𝑁𝑁𝑖𝑖,𝑡𝑡+
𝛽𝛽′7𝑃𝑃𝑅𝑅𝑃𝑃𝐴𝐴𝐴𝐴𝑖𝑖,𝑡𝑡+ 𝜇𝜇𝑖𝑖,𝑡𝑡
(M2)
The last hypothesis is based on the write-off rate. The write-off rate is given by the abbreviation of 𝑊𝑊𝑁𝑁𝑃𝑃𝑖𝑖,𝑡𝑡. Using the write-off rate as an indicators leads the following equation:
𝑊𝑊𝑁𝑁𝑃𝑃𝑖𝑖,𝑡𝑡 = 𝛽𝛽0+ 𝛽𝛽1𝐹𝐹𝐹𝐹𝐹𝐹𝑖𝑖,𝑡𝑡 + 𝛽𝛽′2𝐴𝐴𝐴𝐴𝑖𝑖,𝑡𝑡+ 𝛽𝛽′3𝐴𝐴𝐴𝐴 𝑖𝑖,𝑡𝑡 + 𝛽𝛽′4𝐵𝐵𝑃𝑃𝐴𝐴 𝑖𝑖,𝑡𝑡+ 𝛽𝛽′5𝐴𝐴𝐴𝐴𝐹𝐹 𝑖𝑖,𝑡𝑡+ 𝛽𝛽′6𝑁𝑁𝐴𝐴𝑁𝑁𝑖𝑖,𝑡𝑡+ 𝛽𝛽′7𝑃𝑃𝑅𝑅𝑃𝑃𝐴𝐴𝐴𝐴𝑖𝑖,𝑡𝑡+ 𝜇𝜇𝑖𝑖,𝑡𝑡 (M3)
These three models will be regressed. The Ordinary Least Squares (OLS) regression technique of Stock and Watson will be used (2007). The averages of the MFIs over the period between 2000 and 2010 is taken. It could be that a MFI has only reported the data of one year. The value this year will immediately be the mean value. By this method, the panel data is converted into cross-sectional data. It is likely that variables are omitted which affects the portfolio at risk and the write-off ratio. If these variables are also correlated with the independent variables, OLS can give a biased result. Still, a lot of data is missing. To get a more reliable result, all the observations with missing values about the percentage of female clients is deleted.
A better test for panel data is the random effect regression. The random effect test gives estimators which are more efficient. Besides, the p-values presented are more precise. The random effects model assumes that the variation between different MFIs is random. This is a strong assumption which appears very occasionally. Therefore, it is important to check whether this assumption holds for the dataset.
Therefore, a Hausman test will be conducted. The Hausman test investigates whether the errors are correlated with the independent variables. The null hypothesis states that errors and variables are independent. If the Hausman test shows that they are dependent, it is better to use a more appropriate test. A more appropriate regression will be the fixed effect model. The fixed effects regression is very suitable for omitted variables which will influence our dependent variable.
The panel data used for the random effect regression of the fixed effect regression contains a lot of gaps. Gaps in panel data are commonly solved by using interpolation. The use of interpolation is only for the two dummy control variables.
For two of the dummy variables, the interpolation is very straightforward. If a MFI is reported as a non-governmental organization, we assume that this status lasts over the years. So the NGO variable is stable, so 0 or 1 and no values in between. The same
assumption holds for the operating area. A MFI is assumed to constantly focus on the same area. So if the dummy of the rural area is 1 at one moment, this value will also be used for the missing years. For the fixed effect model these two dummies are omitted, the rural area and the NGO. Rural and NGO are assumed to be constant over time as discussed previously. For the fixed effect regression are only variables added which vary over time.
The other missing values of the other data is kept empty. Interpolation was a possible solution, but in this case not preferred. It is not clear on what the relationship is between variables. There is not enough reliable theories to made strong assumptions. For that reason, using interpolations would be very tricky. Besides, interpolation on basis of the type of MFI would lead to mean values. Regression on the basis of mean values has already been done in the OLS regression.
3.3 Data and descriptive statistics
The data for this research is obtained from the Microfinance Information eXchange (MIX). The MIX is a non-profit organization founded in 2002. The MIX collects data from more than 1700 MFIs worldwide. MFIs provide their own self-reported data to the MIX. Thereafter, the MIX checks the data on the basis of the internationally accepted accounting standards. Besides, they check on outliers and extraordinary results. After this review, the MIX publics partly of the data on their public website. All data used for this paper is available for free on www.mixmarket.org.
The dataset used for this analyses consists of 1537 MFIs across 109 different countries. The countries are located in North America, South America, Europe, Asia, Africa and Australia. As expected, no MFIs are reported in Western Europe. Besides, most MFIs are located in South Asia, especially India. The MIX possesses data of 100 different Indian MFIs. Different types of MFIS are represented in the data set. 35% are NGOs, 26% non-bank financial institution and 31% regular banks.
The data of the years 2000 till 2010 will be analyzed. Because MFIs have to deliver to own data, the dataset contains a lot of gaps. There are only 10093 observations of borrowers per staff. While there are 11410 observations about the nature of the organization, if the MFI is a non-governmental organization.
Figure 1 reports a summary of the main characteristics of all the variables, including the control variables.
Portfolio at risk >30 days is 8662 times reported. The portfolio at risk >30 days is on average 6.96%. This implies that 6.96% of the loans is at least 30 days are in arrears. The first
quartile is .0089 and the median value is .03485. On basis of these values can be derived that the portfolio at risk is negatively skewed.
Portfolio at risk > 90 days is only reported 6170 times. The mean value of portfolio at risk >90 days is 4.87%. This is 2.09 percent lower than the portfolio at risk >90 days. Which
implies that 2.09% percent of the clients pay of their debt after they are 30 days overdue and before 90 days overdue. The portfolio at risk > 90 days is also negatively skewed.
Figure 1
This figure shows a summary of all the variables of the model. The variables age, non-governmental organization and rural area are dummy variables. This summary contains the number of observations, mean, standard deviation the minimum value and maximum value. Besides, 25th percent percentile, the median and the 75th percentile are given. The regression is based on annual data of 2000 till 2010.
The amounts are given in US dollars.
Variable N Mean Std. Dev. Min Max Q1 Q2 Q3 Portfolio at risk > 30 days 8662 .0695516 .1459826 0 6.8431 .0089 .03485 .0794 Portfolio at risk > 90 days 6170 .048716 .104962 0 4.6115 .0052 .0217 .053 Write-off rate 7497 .0185222 .046454 0 1.2686 0 .0053 .0181 Percentage of female borrowers 8441 .6487203 .2854226 0 1 .4351 .6499 .9266 Total assets 11001 3.25e+07 2.36e+08 0 1.92e+10 847636 2974152 1.20e+07 Average loan per
borrower
10368 2563.414 117588 0 1.19e+07 139 387 1160.5 Borrower per staff 10093 131.2449 131.2449 0 13709 55 100 161 Age 11189 1.360533 .8084075 0 2 1 2 2 NGO 11410 .3512708 .4773883 0 1 0 0 1 Rural area 11410 .0607362 .2388563 0 1 0 0 0
The write-off rate is on average 1.85%. This implies that on a certain moment 1.85% of the total loans is written off. The standard deviation is 0.046. The first quartile is 0 and the second is 0.0053. The write of rate is also negatively skewed.
64.87% of all the clients are female in the dataset. The minimum value is 0% and the maximum value 100% . This implies that the dataset contains MFIs which have only female clients, but also MFIs which have no female clients at all. The standard deviation is 0.29. In figure 2 are the correlations between all the variables given. The correlation between portfolio at risk > 30 days and portfolio at risk > 90 days is 0.8616. This correlation is very high as expected. The correlation values between all the other variables are low. This implies the linear relation between variables are weak.
Between the portfolio at risk > 30 days and the percentage of female borrowers is a
percentage of female borrowers is -0.0768. The correlation between the write of rate and the percentage of female borrowers is -0.0111. This implies that there exists a weak negative linear relation between the percentage of female borrowers and the repayment performances. This in accordance with the hypotheses.
Figure 2
Cross-correlation table
PaR30 PaR90 WOR FEM AS AL BPS AGE NGO RURAL PaR30 1 PaR90 0.8620 1 WOR 0.2546 0.2163 1 FEM -0.1316 -0.0768 -0.0111 1 AS -0.0223 -0.0316 0.0226 -0.0687 1 AL -0.0284 -0.0348 -0.0790 -0.3317 0.0599 1 BPS 0.0355 0.0359 0.2361 0.1631 0.0674 -0.1288 1 AGE 0.0695 0.0947 -0.0267 -0.0170 0.1004 -0.0199 0.0467 1 NGO -0.0459 -0.0117 0.0508 0.3502 -0.0953 -0.1734 0.0559 0.1785 1 RURAL 0.1174 0.0769 -0.0873 -0.1834 -0.0399 -0.0517 -0.0317 0.1883 -0.2003 1 4. Analysis 4.1. Empirical Results
First, a Hausman test was performed to check if the random effect test or the fixed effect test is performed. This test is performed separately for the portfolio at risk >30 days, portfolio at risk >90 days and the write-off rate. The Hausman test is conducted on the basis of a 0.05 significance level. The results are presented in figure 3. The portfolio at risk > 30 days and the portfolio at risk > 90 days will be regressed by a random effect test. The write-off rate will be analyzed by a fixed effect test.
Figure 3
This table present the results of the Hausman test. Panel data from 2000 till 2010 is used. The MFIs which did not report any data in this whole time interval about the percentage of female clients are excluded.
Portfolio at risk >30 days Portfolio at risk > 90 days Write-off rate
Chi2 4.45 8.10 71.59
Prob>chi2 0.2169 0.0879 0.0000
From the OLS regression follows a significant relation between the percent of female clients and the portfolio at risk > 30 days. The corresponding coefficient is -.06 and the significance level is 1%. This implies that an increase of 10 percentage point of female borrowers
decreases portfolio at risk > 30 days by 0.5 percentage point. This significance relation also holds for the portfolio at risk > 90 days and the write-off rate. The coefficients are -0.03 and -0.02 respectively. The negative relation found between women and repayment
performances are in line with the results of the D’Espallier (2011). All the results of the regression are presented in figure 4.
The control variables also give several significant results. The average loan per borrower and the rural area are significant related to the portfolio at risk > 30 days. A higher average loan per borrower corresponds to a lower portfolio at risk. Operating in a rural area as a MFIs corresponds with a higher portfolio at risk > 30 days.
The average loan per borrower also has a significance negative relationship with the portfolio at risk > 90 days. The total assets is negative related with the portfolio at risk >90 days. Age, the experience of a MFI, has a positive impact on the portfolio at risk > 90 days. The repayment indicators are also analyzed on the basis of the panel data. The random effect regression finds a highly significant relation between the percent of female clients and the portfolio at risk > 30 days and the portfolio at risk > 90 days. The corresponding coefficients are -0.05 and -0.04 respectively.
Some noticeable results are found for the control variables. Most results from regression of the control variables matches the expectations. For the active borrowers per staff are varying results. The relation depends on the regression type used. For the write-off rate is a significant positive relation found of .0000355 using the OLS regression technique. While, using a fixed effect regression gives a significant negative relation with a coefficient of -.0000401.
Besides, on basis of Gordon et al. a negative result is expected between the average loan size and the three indicators (1965). The regression o basis of this database found an
opposite results. Using a OLS regression significant negative relations are found. This implies that increase the average loan size per borrower leads to a lower portfolio at risk > 30 days, portfolio at risk > 90 days and the write off rate. So, the repayment performances are better instead of worse.
Figure 4
This table looks at the determinants of the repayment behavior of micro loans. For the repayment success are three indicator selected. These indicators are the portfolio at risk > 30 days, portfolio at risk > 90 days and the write-off rate. Column 1-3 report the results of the ordinary least squares regression. Column 4-6 report the results of the random effect regression and the fixed effect regression. The decision whether to choose the random effect regression or the fixed effect regression is based on the Hausman test. Rural area is a dummy variable which takes 1 if the MFI operates in an rural area. Non-governmental organization is dummy which takes 1 if the MFI is a non-governmental organization. N stands for the number of observations. For the ordinary least squares method are averages used of the MFIs data. For the other two regression methods is panel data used. The constant were included in the regression, but they are not shown here. *,**, and *** indicate a significance level of 10%, 5%, and 1% respectively.
cross-sectional data panel data
(1) (2) (3) (4) (5) (6) Portfolio at risk > 30 days Portfolio at risk >90 days Write-off rate Portfolio at risk >30 days Portfolio at risk >90 days Write-off rate Percent of female clients -.0590883 (0.000)*** -.0293921 (0.001)*** -.0193797 (0.000)*** -.0482214 (0.000)*** -.0354253 (0.000)*** -.010129 (0.069)* Age .004253 (0.267) .0076356 (0.011)** -.000946 (0.472) .0099629 (0.000)*** .0102576 (0.000)*** .0050572 (0.000)*** Total assets -3.32e-11
(0.197) -3.16e-11 (0.088)* 1.90e-12 (0.825) -7.02e-12 (0.724) -1.02e-11 (0.466) 4.44e-12 (0.629) Average loan per
borrower
-1.47e-06
(0.044)** -1.80e-06 (0.021)** -5.37e-07 (0.031)** -8.64e-07 (0.119) -8.63e-07 (0.209) -1.61e-07 (0.557) Borrower per staff .0000113
(0.351) .0000136 (0.190) .0000355 (0.000)*** 2.48e-06 (0.829) 8.27e-06 (0.399) -.0000401 (0.000)*** Rural area .0240966 (0.042)** .0108659 (0.212) -.013223 (0.001)*** .0283639 (0.011) ** .0192787 (0.017)** Non-governmental organization -.0052752 (0.424) -.0006896 (0.892) .0034271 (0.126) .0007174 (0.897) .006437 (0.126) N 1806 1558 1735 7544 5426 6543 Adj. R-sq. 0.0221 0.0145 0.0451 0.0091 F-statistic 6.83*** 4.27*** 12.71*** 8.83 Prob > F 0.0000 0.0001 0.0000 0.0000 LR Chi2 61.22 58.24 Prob>Chi2 0.0000 0.0000
4.2 Robustness check
As shown in figure 2 there are only weak correlations between independent variables. This indicates there exists no multicollinearity.
When plotting the residuals versus the fitted value no noteworthy graph is drawn. To make sure that the assumption of heteroscedasticity is fulfilled a regression is run. The hypothesis for this regression assumes that the error variance is constant as the portfolio at risk > 30 days, portfolio at risk > 90 days and the write-off rate increase. The results of this
regression are given in figure 5. The tests indicate for portfolio at risk > 30 days and the write-off rate significant Chi2 values. This implies that the condition of heteroscedasticity is fulfilled. For the portfolio at risk > days a p-value of 0.1525 is found. Therefore, the null hypothesis is not rejected. It cannot be assumed that heteroscedasticity holds. This is a violation of an important assumption of the OLS-regression.
Figure 5
This table represents the result of a test for heteroscedasticity. *,**, and *** indicate a significance level of 10%, 5%, and 1% respectively.
Chi2 Prob > Chi2 Portfolio at risk > 30 days 38.78*** 0.0000 Portfolio at risk > 90 days 2.05 0.1525
Write-off rate 169.59*** 0.0000
Besides, the dataset does not contain extensive outliers. The Microfinance Information Exchange checked for outliers and outstanding results. Therefore, the most outstanding data is already excluded. When analyzing the summary of the statistics (see figure 1) no outstanding results are detected.
De R-squared value is very low. This implies that the variations in PaR30 is explained by a small part of the model. An important potential reason is the omission of some key
limitations. Besides, low R-squared values are very common in the field of economics. The repayment performances depend on so many factors which are often not clear.
5. Conclusion and discussion
5.1 Conclusion
Nowadays, MFIs focus more and more on women. Mainly , because women should be better in paying off their loans. This paper researched if there indeed exists a relation between gender and repayment performances for MFIs. This investigation has been done on the basis of three indicators for the repayment performance. The three indicators are the portfolio at risk > 30 days, portfolio at risk > 90 days and the write off rate. All these
indicators have their own sub hypothesis. The main hypothesis is: MFIs with more female clients have better repayment performances.
On basis of the three given estimators can be concluded that women have indeed better repayment performances. The data regression showed a significant negative relation between the percentage of female clients and the portfolio at risk >30 days and > 90 days. Besides, the relation between the percentage of female clients and the write-off rate is negative and significant as well. All the three hypothesis are confirmed. These three results regarding the indicators are in line with each other. From this consistent results, it can be concluded that a higher percentage of female clients lead indeed to better repayment performances.
5.2 Discussion
The result of this research is very favorable for current policy makers of MFIs. According to the results, the idea that women are better in paying of their debt is justified. This confirms the idea of MFIs that women are better in paying off their debt. Therefore, the MFIs are on the good way considering their policy. It is very attractive for MFIs to focus on women. Focusing on women would lead to lower portfolio at risk and lower write-off rates.
Why women are better in paying off their loans is not determined yet. This is important to keep in mind for policy makers. It is not clear if the difference comes from the
characteristics of women or their circumstances. To determine the reason further research is needed.
5.3 Limitations
The dataset consist of self-reported data by MFIs. This has two big disadvantages. First, there exist a big chance that only better organized MFIs deliver their data. Therefore, the results can give a too optimistic view. Second, the MFIs do not have to report about all the variables. The consequence is that MFIs deliver just little information about their
performances. This leads to a lot of gaps in the dataset.
For the random effect and the fixed effect regressions is panel data used. To fill up the gaps, three important assumptions are made. First, the status of a MFI is constant over time. This means that if a MFI is reported as an non-governmental organization, this value is used for the other empty years. This assumption has no big consequences, because in general, MFIs are time consistent considering their status.
The second assumption state that the operating area of MFIs are constant over time. This implies as a MFI reported itself operating in a rural area, this dummy value is used for the empty years. This assumption is also very plausible. MFIs rarely switch their operating area, due to the involved organizational costs.
The last assumption concerning unavailable data is about the existence of a MFI. Non available data of a MFI could have two different reasons. First, the MFI did not deliver any data. The second scenario is that the MFI did not exist in this year. This scenario is simply ignored in this paper. Due to all these limitations, an analyses of panel is a rough estimation. Group lending has been left out as a control variable in this regression. Unfortunately, there was no data available about this form of lending. Previous research found a positive relation
between group lending and microloan repayments. Adding group lending as variable could highly increase the fit of the model.
5.4 Further research
This research confirms the idea that women have better repayment performances. But, it is still not clear what the reason behind this behavior is. Therefore, it would be interesting to deep further into the motivation of women. A possible explanation could be that women are more risk averse. Maybe, because they are more motivated due to their limited
opportunities. Or women cooperate more and better when investing in their business. This list of potential reasons could be easily extended.
Group lending has been left out as a control variable in this regression. Unfortunately, there was no data available about this form of lending. Previous research found a positive relation between group lending and microloan repayments. Adding group lending as variable could highly increase the fit of the model. It is highly recommended to add this variable in a further research.
Like every research, there are data limitations. For micro finance, the data availability is even worse. As mentioned earlier, the dataset used contained a lot of missing values. Increasing the data collection would give more precise results. Especially, because the data used is self-reported, the results can be biased. By obligating MFIs to report, the results will better represent the MFI community.
References
- D’espallier, B., Guérin, I., & Mersland, R. (2011). Women and repayment in microfinance: A global analysis. World Development, 39(5), 758-772.
- DeMartino, R., & Barbato, R. (2003). Differences between women and men MBA entrepreneurs: exploring family flexibility and wealth creation as career
motivators. Journal of business venturing, 18(6), 815-832.
- Micro Finance Exchange. Cross-Market Analysis. Retrieved May 20, 2015, from http://mixmarket.org/profiles-reports/crossmarket-analysis-report - Godquin, M. (2004). Microfinance repayment performance in Bangladesh: How
to improve the allocation of loans by MFIs. World Development, 32(11), 1909-1926.
- Hossain, M. (1988). Credit for alleviation of rural poverty: The Grameen Bank in
Bangladesh (Vol. 65). Intl Food Policy Res Inst.
- Jianakoplos, N. A., & Bernasek, A. (1998). Are women more risk averse?.Economic
inquiry, 36(4), 620.
- Johnson, S. and Kidder, T. (1999) ‘Globalization and gender—dilemmas for microfinance organizations’, Small Enterprise Development 10(3):4–15.
- Khandker, S. R., Khalily, M. B., & Khan, Z. H. (1995). Grameen Bank: performance
and sustainability (Vol. 306). World Bank Publications.
- Lafourcade, A. L., Isern, J., Mwangi, P., & Brown, M. (2005). Overview of the outreach and financial performance of microfinance institutions in
Africa.Microfinance Information eXchange, Washington, DC.
- Micro-Credit Summit. (1997). The micro-credit summit report. Washington, DC: Grameen Foundation.
- Schubert, R., Brown, M., Gysler, M., & Brachinger, H. W. (1999). Financial decision-making: are women really more risk-averse?. American Economic
Review, 381-385.
- Sharma, M., & Zeller, M. (1997). Repayment performance in group-based credit programs in Bangladesh: An empirical analysis. World development, 25(10), 1731-1742.
- Stock, J. H., & Watson, M. W. (2007). Introduction to econometrics. Boston, MA: Pearson Education.
- Women’s Entrepreneurship Development Trust Fund (WEDTF). Information on microfinance and empowerment of women. Zanzibar, Tanzania: WEDTF, 2001. - Freimer, M., & Gordon, M. J. (1965). Why bankers ration credit. Quarterly Journal
