Microfinance Effect on Income Inequality in Latin America: A cross-country panel data study on the effects of microfinance on the income inequality in Latin America

(1)

Microfinance Effect on Income

Inequality in Latin America

A cross-country panel data study on the effects of microfinance

on the income inequality in Latin America

Bachelor Thesis

Authors: Gabriel Antoine & William Möllestam Supervisor: Tobias König

Examiner: Mats Hammarstedt Term: VT20

Subject: Economics Level: Bachelor level Course code: 2NA11E

(2)

Abstract

This paper examines if increased microfinance intensity reduces the income inequality in 11 Latin American countries from 2005 to 2015. Gini coefficient was used as a measure of income inequality, while microfinance intensity was derived by dividing the number of active borrowers by the country's population. A panel data was constructed with 384 microfinance institutes present in the countries studied. To examine the relationship, a pooled OLS and a country clustered fixed-effects model was conducted using the specific-to-general method. Both methods showed a significant negative relationship between the Gini coefficient and microfinance intensity. However, it was a relatively small impact at -0.004% for every percent increase in microfinance, which confirms our hypothesis that a higher MFI participation leads to a decrease in income inequality. These results are in line with previous studies conducted, although, to our knowledge, this is the first macroeconomic framework study conducted on multiple Latin American countries at once.

Key words

Microfinance, Income inequality, Cross-Country analysis

Acknowledgments

We want to give big thanks to our supervisor, Tobias König for his invaluable advice, support and expertise during our time writing this paper.

We would also like to thank Adam Lidvall, who provided feedback in his discussion. Lastly we want to thank the faculty of Linnaeus School of Business and Economics.

(3)

Appendices

Appendix 1: Variance inflation factors

Appendix 2: Breusch-Pagan/Cook-Weisberg test for heteroscedasticity

(4)

Appendix 4: Hausman test

(5)

1

1 Introduction

“The poor are very creative. They know how to earn a living and how to change their lives. All they need is an opportunity;” (Yunus, 1999) is an assumption that pioneered the way microfinancing is in use today.”

Microfinance institutes (MFIs) are banks for the unbanked, providing loans and other financial services, such as money transfers and insurances to the lower-income people of developing countries. With an estimated 1.7 billion people considered unbanked 2019 (Worldbank, 2019), microfinance is a solution that has been tried and tested, with varying degrees of success.

While critics cite that it does not reach the poorest of the poor (Datta, Dipanka, 2004) and that microfinance institutes charge exorbitant interest rates, which leads to debt traps (Mitra, 2009). Proponents of microfinance (MF) see it as a road out of relative poverty and a possibility to not only get employment but also create employment for their community while increasing earnings in rural areas of developing countries.

Microfinance (MF), as known today, has its roots in 1970s Brazil, Accion International began providing MF in Recife, Brazil to informal businesses. In four years, they managed to provide 885 loans that helped create or stabilize 1300 jobs ("Our History | Accion" 2020). On a similar note, Yunus inspired the poor following the famine in Bangladesh 1974 by giving small loans of 27 dollars to 42 struggling families. In 1976, by applying his research and experience expanded his experiment with the help of Grameen Bank in the rural village Jobra. Since then, much interest has been given to microfinance as a means to help the poor. (Yunus, 1999)

This study will examine if increased microfinance intensity (MI) reduces the income inequality in the 111 Latin American countries, from the macro perspective. GINI will be used as a measure of income inequality, where a higher coefficient indicates a higher inequality and vice versa. The expectation is that increased participation in MI will have an inverse relationship to GINI, leading to lower income inequality in the countries studied.

1_{The countries included in the sample is Bolivia, Brazil, Colombia, Costa Rica, Dominican Republic, Ecuador, El}

(6)

2 Latin America has been chosen because it is a perfect place for MFIs to grow because of their favorable conditions for MF; this includes institutional factors. Further on, MFIs and banks have shown to be profitable, which gives an incentive for more actors to enter the market, or for existing to expand. Further on, to our knowledge, there has been no study that has examined the effects of MF on income inequality in Latin America, and the data availability for the countries studied allowed for studying a reasonable time period for examining the effects that MF has on GINI.

The disposition will be as follow; the second section will look at past studies surrounding microfinance in developing countries with a focus on their overall impact. The third section looks at the theoretical framework, with a brief history of microfinance and its purpose. In the fourth and fifth sections, the methodology and data are presented. The sixth section presents the results, followed by the last two sections with a discussion and conclusion.

2 Literature review

There have been differing results about the effectiveness of microfinance, with varying results depending on region, timeline, and the financial development of countries studied. Many new studies have used randomized controlled trials (RCT) to study the impact of the introduction of microfinance in a community. A randomized controlled trial was conducted in Ethiopia 2003-2006 in rural areas where agriculture and animal husbandry comprised the most significant share of the economic activity in the communities. To track the result, they conducted surveys pre- and post-implementation with the treatment and control groups and the results showed that borrowing increased, however, none of the socioeconomic factors, such as education, health, and empowerment of women, showed significant results (Tarozzi et al., 2015). Several studies have conducted similar randomized controlled trials, such as Crépon et al. (2015) in Morocco, where there was no increase in consumption or income but an increase in investment used for self-employment and profit. The results were consistent with Banerjee et al. (2015), where both showed increases in borrowing, profits for existing businesses, and no change in the socioeconomic factors or consumption in Hyderabad, India.

The studies, as mentioned above, have all been RCTs that are reasonable for assessing the results of a region. However, they will not paint a fair picture of the overall impact of microfinancing when looking at a broader spectrum. As this paper focuses on cross-country

(7)

3 comparison over a more extended period, with 11 countries from the same continent, it will help assess the impact on the impact of Latin America, rather than a single town.

Similarily, Angelucci et al. (2013) found that households that obtained microcredit were better off in terms of liquidity and risks. When looking at a short period, the authors did not find any substantial evidence of income or consumption benefits, which is in line with the studies previously mentioned. However, they theorized that those effects could have appeared if they looked at a more extended period. They found no support of the hypothesis that MF hurt poor people and that microcredit on average was beneficial in Mexico. However, it did not create as much progress as proponents advertise. By looking at a more extended period, the goal of this paper is to answer if there are any income benefits long term from MF.

Most of the studies that have been conducted within the subject have focused on the microeconomic view, where most of them have been case studies such as household surveys or population monitoring, such as above. Some studies have tried to analyze the microfinance effect on inequality with a macroeconomic framework. Mahjabeen (2008) did a case study on Bangladesh, where he presented two models with and without the impact of MFIs. The author assumed that the MFIs used a group lending method to provide micro-loans and aimed for rural households. The study showed evidence of an increase in income and consumption when microfinancing was taken into account.

Further on, Mahjabeen also concluded that microfinancing led to a decrease in income inequality and was beneficial for poor people in the country. He concluded that MF is an effective policy for poverty reduction. Beck et al. (2007) arrived at a similar conclusion when looking at how financial development is related to income distribution and poverty levels. Microfinancing increased the income of the poorest quintile while at the same time reduced income inequality. By looking at the distribution of income and the income of the poor by examining the Gini coefficient, the result showed that microfinance is particularly beneficial to the poor.

Clarke et al. (2006) studied 83 developing countries in a long period and saw that inequality is less when financial development is greater. The authors found that when the financial sector development increases at a low level, inequality tends to increase rather than decrease. They rejected the hypothesis that financial development only benefited the wealthier population and concluded that financial development, such as MFIs, reduces inequality. Bangoura et al. (2016) found that in countries with high microfinance intensity, income

(8)

4 inequality was lower, and if the individual loans are small, the inequalities are more significant. To give higher loans to the poor provides income-generating activities and reduces income inequality.

On the other hand, Bangoura et al. could conclude that the link between inequality and microfinance was country-specific, which means that the outcome depends on the MFIs targeting strategy. Buchenrieder et al. (2019) found a positive impact on income in the short-run, but a negative long-term effect on microfinance when looking at income-generating activities in parts of Sub-Saharan Africa. The authors concluded that the MFIs strategy is essential for evaluating the long-term impact of microcredit to avoid a poverty trap.

3 Theoretical Framework

3.1 Microfinance

Microfinance is a financial service offered to individuals or businesses that usually sits at the lower-income quintile. Although the microfinance known to the world today has its roots in 1970s Bangladesh and Brazil, earlier initiatives had been available since the beginning of time with informal savings and credit groups. More similar practices can be traced back to 18th century Ireland, where the Irish Loan Fund System operated, which at its peak was lending to around 20% of the Irish households and 19th century Germany during the winter famine. (Helms, 2006)

The 20th century saw the most rapid development in terms of microfinance institutes. Today, the line between MFI and a traditional bank is starting to blur as MFIs have also started to offer services such as insurance, money transfers, and savings accounts. On the opposite end, traditional banks have also entered the market of offering microfinance. (Helms, 2006)

Services are targeted to individuals with no other alternative, with loans ranging from as low as 100 to 25,000 dollars with the ultimate goal of giving the impoverished a road to self-sufficiency. MFIs services range from providing basics such as savings account to startup capital and capital for expanding business. (Chowdhury, 2009)

According to Weiss & Montgomery (2005), MF is a mechanism for poverty reduction since if access to credit is expanded, the poor would be able to finance productive activates that will lead to income growth. Further on, they state that the households that are on the brink of poverty, microcredit can help stabilize their businesses or consumption in uncertain times.

(9)

5 A potential hindrance to microfinance can be attributed to the interest rate charged to borrowers. Many MFIs charge market interest rates. These are often not high enough for recovery on account of the cost of providing loans in combination with the average principal amount (Weiss & Montgomery, 2005). However, there are also cases with predatory interest rates as a survey conducted in 2008, where 350 MFIs reported interest rates ranging between 20-40% per year (Morduch, 2008), which has been defended because it is needed to sustain the MFIs long term.

(10)

6

4 Methodology

This section will describe our econometrical methods and description of the variables that has been included in the study.

4.1 Measure of income inequality

Income inequality can be measured in several different ways, most established and cited figures being 50-10 wage gap, 90-10 wage gap, and Gini coefficient. The two former are measured just how they sound by giving the percent difference between the workers in the 90th_{or 50}th percentile to the worker in the 10th_{percentile (Borjas, 2015). 50-10 wage gap would then} measure the wage gap between the middle class and the low-income workers and 90-10 the range of the income distribution or the gap between the highest earners and the lowest earner. This method of measuring wage gaps can also be applied to other deciles, such as the 95/50 wage gap.

Gini coefficient, on the other hand, is based on wage distribution for a country that allows

comparing differences in time and also between countries. Gini summarizes the income distribution for a whole country with one single number between 0 (perfect equality) and 1 (perfect inequality), which means that the Gini coefficient decreases as a country become equal in terms of incomes and vice versa.

The calculations are based on the Lorenz curve developed by the American economist Max Lorenz in the early 20th century, and later, based on the Lorenz curve, the Italian statistician Corrado Gini brought forward the Gini measurement. Gini can be illustrated as follows;

(11)

7 Table 1 represents the income levels with the cumulative level of the population and income levels, which in turn creates the blue line, which is what is known as the Lorenz curve. Through this graph, a formula for calculating the Gini coefficient can be formulated as follows;

𝑒𝑞. 1 𝐺𝑖𝑛𝑖 = 𝐴𝑟𝑒𝑎 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 𝑝𝑒𝑟𝑓𝑒𝑐𝑡 𝑒𝑞𝑢𝑎𝑙𝑖𝑡𝑦 𝑎𝑛𝑑 𝑎𝑐𝑡𝑢𝑎𝑙 𝐿𝑜𝑟𝑒𝑛𝑧 𝑐𝑢𝑟𝑣𝑒 𝐴𝑟𝑒𝑎 𝑢𝑛𝑑𝑒𝑟 𝐿𝑜𝑟𝑒𝑛𝑧 𝐶𝑢𝑟𝑣𝑒

The perfect equality area is the orange line. If there is perfect equality in the country, table 1 would be inputted as a percentage of population = percentage of income, meaning everyone in the country receive the same income. Therefore, there would be no Lorenz curve. The numerator in eq.1 is the area between the orange line and the blue line, and the denominator is the area under the blue line. As the Lorenz curve diverges from the perfect equality line, more income inequality is present.

There is one main weakness in only using Gini as a measure of income inequality, as it does not take into account the subtleties that can happen behind the curtains. For instance, a shift from the bottom quintile to the top would lead to an increase in Gini; the same increase can happen by shifting someone in the second or third quintile to the top. Although the change in the Gini coefficient is the same, the redistribution of income is not, which is where other measures can complement the Gini coefficient.

Table and figure 1, sample Lorenz curve %of population % of income 0 0 10 0 20 5 30 10 40 15 50 25 60 35 70 45 80 55 90 75 100 100

(12)

8

4.2 Independent variables

The control variables have been picked based on empirical research that links them with income inequality and microfinance, such as Milanovic (2002), stating that inflation negatively affects the income shares of the poor and the middle class. While microfinance intensity is a measure of the share of the population with a microloan, the gross loan portfolio determines its size. This is measured in US dollars across all countries. FDI, according to Choi (2006), showed that FDI increases income inequality, especially in developing countries with unequal income distribution, in the same article, Choi also argues that growth in GDP per capita decreases income inequality.

The number of active borrowers is the number of individuals with an outstanding loan

balance in one of the country's MFI or is responsible for repaying at least one dollar of the gross loan portfolio. An individual counts as a single borrower even if that individual has multiple loans in different MFI. The numbers are based on the number of individual borrowers and not the number of groups. The variable gives an indication of how developed the MFI is in that country and how many individuals it reaches out to. The number of active borrowers is collected from the databank MIX market on The World Bank (2020). The number of active borrowers is gathered to use for calculating Microfinance intensity (MI) by dividing the number of active borrowers with the population of the country; the number is presented in percentage. MI is a measure of the proportion of the population that is participating in a microfinance program. MI is the main explanatory variable, as it is expected that with a higher MI, there will be an inverse relationship with GINI, leading to more income equality in Latin America. The negative relationship can be confirmed by looking at this scatter plot below, which shows a small negative relationship; this will be examined further to determine if there is any statistical significance.

(13)

9 A drawback to using the number of active borrowers can be attributed to that an individual can be counted twice if said individual has two loans with different MFIs. However, this is not expected to distort the result since individuals with multiple loans with different MFIs are not expected to combine for a large number.

Gross loan portfolio (GLP) is a similar measure to the one above, but with one key difference being, it measures the total value of the loans distributed by the MFIs. This is added to control for the size of the loans distributed. The direction of GLP is theorized to head the same way as MI; as the gross loan portfolio increase, the Gini coefficient will decrease.

Foreign Direct Investments (FDI) is a cross-border investment and is the sum of equity capital, long-term capital, reinvestment of earnings and other capital. In developing countries, MFIs have received foreign funding, especially in Latin America. Foreign investments can lead to more intense and improved MFIs. The reason why foreign investments tend to invest in Latin America is that the MFIs are more developed and regulated in this region. FDI can be a crucial factor when it comes to technology transfer, business models, competitiveness and productivity. Therefore, FDI can help to reduce income gaps and improve productivity in the region. Foreign Direct Investments are collected from the World Bank (2020). The impact FDI will have, or if it even has an impact on inequality is hard to predict. Studies such Chen (2016) found that FDI increased urban-rural income inequality in China, while Choi (2006) found that FDI increases as a percentage of GDP inequality increases.

.4 4 .4 6 .4 8 .5 .5 2 .5 4 0 2 4 6 8 10 Microfinance Intensity GINI Fitted values

(14)

10 Interest expense on borrowing is a tool for the MFIs to cover operating costs such as administration costs, loan losses and cost of capital (including inflation). However, it also works as a provision to increase their equity. The interest rate of microcredit tends to be higher than the banks’ interest rate because the costs of making small loans are usually higher in percentage compared to a large loan. Even if the interest expense on borrowing is high, the poor consider credit access to be greater than the actual interest cost. High-interest rates can create a problem for the poor in the countries when they cannot afford to borrow due to the high interest rate. The loans will not reach the poorest and the income inequality will increase (Mitra, 2009). Interest expense on borrowings is collected from the databank MIX market on The World Bank (2020)

GDP per capita (GDP) is calculated by dividing the gross domestic product (GDP) over a country's population. It is a measure of a country's economic growth. GDP is the primary measure when looking at productivity, and by dividing it with each citizen of the country, GDP per capita also works as a measure of prosperity. Economic growth is associated with higher investment and a higher employment rate. However, economic growth can both reduce and increase income inequality within a country depending on which deciles benefit from the investments. Therefore, GDP per capita will work as a control variable to see how the country's economic growth has changed over time. The data is collected from The World Bank (2020)

Inflation affects the income inequality in a country, especially for the lower deciles since inflation can reduce the spending power so that the poorest are forced to spend their entire income on essential products. This is true for people lacking the ability to save and invest with banks as their savings vanish at the rate of inflation. This affects generations since the poor will have no money to spend on education and other crucial human capital to improve the individual income level. In this sense, inflation will increase income inequality and poverty (Cardoso, 1992). Zou and Li (2002) also found that inflations worsen the income distribution and that higher inflation has a negative effect on the income shares of the poor. The data for inflation is collected from The World Bank (2020).

All the variables collected from Mix market have one key weakness, which will be discussed more in detail in the data section. However, a summary of the weakness is that missing data from years Mix Market had no available data for individual MFIs can potentially underestimate the size of the metrics; this applies to MI, GLP and interest rate expense.

(15)

11

4.3 Econometrical Specification

The data is a strongly balanced panel data which contains measurements on the same 11 Latin American countries from 2005 to 2015 and the specification as follows.

𝑒𝑞. 2 𝐺𝐼𝑁𝐼𝑖𝑡= 𝛽0+ 𝛽1𝑀𝐼𝑖𝑡+ 𝛽2𝐼𝑛𝑓𝑙𝑎𝑡𝑖𝑜𝑛𝑖𝑡+ 𝛽3𝐿𝑜𝑔𝐼𝑛𝑡𝑒𝑟𝑒𝑠𝑡𝑅𝑎𝑡𝑒𝑖𝑡+ 𝛽4𝐿𝑜𝑔𝐺𝐿𝑃𝑖𝑡+ 𝛽5𝐿𝑜𝑔𝐹𝐷𝐼𝑖𝑡+ 𝛽6𝐿𝑜𝑔𝐺𝐷𝑃𝑖𝑡+ 𝜀𝑖𝑡

Where GINI is the dependent variable and is our measure of income inequality, MI is the main explanatory variable and represents the microfinance intensity. The relationship between GINI and MI is expected to show a negative relationship, as more people take a microloan, income inequality will decrease. Inflation is the annual change in consumer price index in percentage. LogInterestRate is the logarithm of the interest expense accrued on borrowings. LogGLP is the logarithm of gross loan portfolio. LogFDI is the logarithm of foreign direct investments, which is a cross-border investment, LogGDP, which is the logarithm of the gross domestic product per capita, and lastly e is the random disturbance with i for country and t (2005, 20006……2015) for the year.

4.3.1 Pooled Ordinary Least Square

The first regressions we did was a pooled OLS regression by using our panel data. A pooled OLS is in simple terms an OLS technique that runs on a panel data, but it ignores all country-specific factors that make them unique. Therefore, if the unobserved effect 𝑎_𝑖 is correlated with the outcome variable, it will result in biased and inconsistent results. (Wooldridge, 2016) The pooled OLS will serve as a benchmark, but it is important to highlight the fact that it is a flawed measure in terms of panel data.

4.3.2 Fixed Effects Model

To reduce the problem of unobserved factors, we conducted a FE regression, which removes time-constant factors. Since the unobserved variables are allowed to have any correlation with the observed variables, the FE model allows for such associations and can treat the unobserved variables as fixed parameters. This is as if these unobserved variables had been included and measured in the regression. In turn, it helps in dealing with the omitted variable bias.

The FE regression also solves the problem with the correlated error by using each specific variable as its control. If the coefficients of the main variable of interest do not change much compared to the pooled OLS regression, this indicates that the unobserved factors do not play

(16)

12 any large role in the regression, and that the pooled OLS already gives a good estimate (Allison, 2009).

Since the intercept in the fixed effects model is not constant, 𝑎_𝑖 is added to the regression to account for the unknown intercept for each country (Wooldridge, 2016). This gives us the following specification for a multiple regression.

𝑒𝑞. 3 (𝐺𝑖𝑛𝑖𝑖𝑡− 𝐺𝑖𝑛𝑖̅̅̅̅̅̅̅̅ = 𝛽𝑖) 1(𝑀𝐼𝑖𝑡− 𝑀𝐼̅̅̅̅̅) + 𝛽𝑖 2(𝑋𝑖𝑡− 𝑋̅ ) + 𝑎𝑖 𝑖 + (𝜀𝑖𝑡− 𝜀̅) 𝑖 𝑡 = 2005, 20006 … .2015

Where 𝛽2 includes all the control variables, 𝛽1 is the main explanatory variable, Gini is the dependent variable and e is the random disturbance.

4.3.3 Specific-to-general method

We have used the specific-to-general method to build our models where the approach is to start with simple regression and then, in succession, start adding to it so that for every modification becomes a better description of the reality. This also allows for an overview of how the added control variables affect the main explanatory variably. (Brooks, 2019)

5 Data

The data is collected from the Socio-Economic Database for Latin America and the Caribbean (CEDLAS) and mix market on The World Bank for the period 2005 to 2015.

When collecting the data, we used the latest years available for GINI—the availability in the data varied among the Latin American countries after 2015, while only missing three single data points for the whole period studied. To deal with this, the average of neighboring values was calculated to account for the single year with missing data, i.e., GINI coefficient 2010 for Brazil and in the case where two consecutive years was missing, the closest observation was carried forward and backward such as for Peru between 2007 to 2008. Although limitations apply to both these methods, we decided that the change in GINI between the years where missing data was present was minimal and thus would not distort the overall result. (Lydersen, 2019).

Table 1 below shows the mean and standard errors for all 11 countries and the average at the bottom for all at the bottom. The average GINI for all countries is 0.5 and average MI is 4.47%.

(17)

13

Table 1: Descriptive Statistics

Country GINI MI % Inflation Log GLP Log FDI Log INT Log GDP Bolivia 0.502 (0.044) 9.348 (1.958) 6.200 (3.388) 21.542 (0.766) 20.087 (0.721) 17.152 (0.512) 7.584 (0.382) Brazil 0.534 (0.016) 0.857 (0.439) 5.810 (1.469) 20.690 (0.922) 24.670 (0.647) 16.936 (0.737) 9.123 (0.333) Colombia 0.536 (0.013) 4.654 (1.159) 4.078 (1.506) 22.136 (0.485) 23.100 (0.340) 17.737 (0.554) 8.669 (0.307) Costa Rica 0.489 (0.008) 0.426 (0.130) 7.407 (4.149) 17.158 (0.537) 21.565 (0.271) 14.748 (0.408) 8.959 (0.305) Dominican Republic 0.478 (0.022) 3.270 (1.654) 1.753 (2.025) 19.784 (0.853) 21.417 (0.307) 16.222 (0.304) 8.594 (0.185) Ecuador 0.486 (0.031) 6.076 (2.459) 4.086 (1.734) 21.356 (0.726) 20.007 (0.681) 17.060 (0.515) 8.400 (0.262) El Salvador 0.441 (0.025) 2.513 (0.441) 2.752 (2.348) 19.646 (0.279) 19.811 (0.705) 15.919 (0.481) 8.031 (0.138) Honduras 0.540 (0.031) 2.195 (0.195) 6.303 (2.216) 19.217 (0.611) 20.642 (0.381) 16.054 (0.612) 7.517 (0.190) Panama 0.520 (0.012) 0.712 (0.346) 3.838 (2.284) 18.096 (1.296) 21.736 0(.537) 13.838 (1.486) 9.037 (0.350) Paraguay 0.504 (0.020) 8.370 (3.724) 5.881 (2.816) 20.305 (0.733) 19.113 (1.452) 16.306 (1.603) 8.300 (0.400) Peru 0.464 (0.028) 10.758 (2.814) 2.934 (1.235) 22.428 (0.682) 22.564 (0.490) 18.199 (0.858) 8.473 (0.323) Average .500 (0.039) 4.471 (0.393) 4.642 (2.891) 20.214 (1.739 21.337 (1.700) 16.379 (1.453) 8.430 (0.593)

(18)

14 As the case with the data collected from the Mix market present at World Bank, the data for active borrowers, GLP, FDI, and Interest rate on borrowings was added for each country and year. Limitations in regards to Mix Market data can be attributed to the intervals at which data is collected. Some MFI has data for only a portion of the years, while some have for all years; this can create an underestimation for certain years for active borrowers, GLP, FDI, and interest rate on borrowings. However, we chose these years as this interval had the least amount of missing data, and further on the Mix Market data is the most comprehensive when it comes to country-level macroeconomic data with no better alternative for the purpose of this kind of study.

A total of 384 MFIs with an average of 35 per country, but the range was large with 7 operating MFIs in Paraguay and 77 in Ecuador.

6 Results

In this section, we will present two different regressions based on equations 2 and 3, which are the specifications for the pooled OLS and the fixed-effects model. Table 2 presents the results of estimating the relationship between microfinance intensity and GINI using pooled OLS. Secondly, in table 3, we see the results of the same estimation but using country clusters in a fixed-effects model. Both these tables apply the specific-to-general method by adding control variables one by one to see whether the relationship between Microfinance intensity and our dependent variable remains statistically significant and the impact each control variable has on the coefficients. Our first regression, as seen in column 1 in table 2 and 3, presents the result when only looking at the correlation between microfinance intensity and GINI. In both the OLS and the fixed-effect model, we see a negative relationship between MI and GINI when looking at the isolated effect. This can be interpreted as if microfinance increased by one percent; it will lead to a 0.004 percentage point decrease in GINI when it comes to our OLS estimate and a 0.009 percentage point decrease with the fixed-effects model, both of which are significant at the 1% level. In other words, a higher MI tends to lower the income inequality, when only looking at the isolated effect of MI on GINI.

When it comes to the pooled OLS, the MI coefficient remains negative and statistically significant at all levels by adding control variables. The impact of MF on Gini stays almost constant when adding new variables meaning that the impact of MF on income inequality stays

(19)

15 the same as suggested earlier. Inflation, FDI, and GDP are also significant at 1%, while GLP and interest expense on borrowing are significant at 5% level.

The R2_{, which shows the total variance in Gini, explained by the outcome variables, is 37.6%.} While R2 always increases as more variables are added, the adjusted R2 increases only if the new variables improve the model more than expected by chance, the adjusted R2 is 34.3%

Table 2: Pooled OLS

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

As is the case with most models, the problem of multicollinearity, which can affect the coefficient and make the p-value unreliable, a test for variance inflation factors (VIF) was

(1) (2) (3) (4) (5) (6)

VARIABLES GINI GINI GINI GINI GINI GINI

Microfinance intensity -0.004*** -0.003*** -0.004*** -0.005*** -0.004*** -0.004*** (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Inflation 0.004*** 0.004*** 0.005*** 0.005*** 0.003***

(0.001) (0.001) (0.001) (0.001) (0.001) Log of interest expense 0.002 -0.008 -0.009* -0.011**

(0.003) (0.005) (0.005) (0.005) Log of GLP 0.012*** 0.011** 0.009** (0.005) (0.005) (0.004) Log of FDI 0.003 0.010*** (0.002) (0.003) Log of GDP -0.030*** (0.007) Constant 0.516*** 0.497*** 0.460*** 0.389*** 0.372*** 0.534*** (0.005) (0.007) (0.046) (0.052) (0.054) (0.063) Observations 121 121 121 121 121 121 R-squared 0.136 0.215 0.219 0.263 0.271 0.376 Adjusted R-squared 0.129 0.202 0.199 0.238 0.240 0.343

(20)

16 conducted to see which of our independent variables was correlated and to which degree. When interpreting the VIF, 1 indicates that there is no correlation, 1-5 shows a moderate correlation, VIF>5 a high correlation and VIF>10 indicates serious multicollinearity and is a cause for concern. Both LogGLP and LogINT showed a high level of collinearity (See Appendix 1), which was expected, seeing as both are expected to increase as MI increase. However, because none of them exceeded 10, which indicates serious multicollinearity, they were kept in the model. Lastly, a test for heteroscedasticity (See Appendix 2) was run to see if any was present in the pooled OLS model. The Breuch-Pagen/Cook Weisberg test failed to detect any presence of heteroscedasticity.

The fixed-effects model was then conducted similarly, as seen below in table 3. As opposed to the pooled OLS, more variation across entities was observed as more control variables were added. Microfinance intensity decreased from -0.009 to -0.004, which means the impact of MI is lower than initially suggested by the simple regression. Only MI and log of GDP were significant at 5%, while the log of FDI was significant at the 10% level. Further on, 56% of the variation in Gini can be explained by the included variables in the regression.

Table 3: Fixed-effect with country cluster

(1) (2) (3) (4) (5) (6)

VARIABLES GINI GINI GINI GINI GINI GINI

Microfinance intensity -0.009*** -0.008*** -0.007** -0.004* -0.005** -0.004** (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)

Inflation 0.001 0.000 -0.000 -0.000 -0.001

(0.001) (0.001) (0.001) (0.001) (0.001) Log of interest expense -0.007* 0.002 0.001 0.003

(0.003) (0.003) (0.003) (0.003) Log of GLP -0.016** -0.017** -0.007 (0.006) (0.005) (0.004) Log of FDI 0.003 0.007* (0.004) (0.004) Log of GDP -0.045** (0.018) Constant 0.538*** 0.535*** 0.642*** 0.822*** 0.789*** 0.853***

(21)

17 Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

A heteroscedasticity test was conducted with a Wald test (See Appendix 3), where we failed to reject the null hypothesis and could conclude heteroscedasticity in our fixed-effects model. The heteroscedasticity is mitigated by clustering the standard errors at the country-level. By clustering, each cross-sectional is defined as an observational cluster for every t. Therefore, arbitrary correlation and changing variance are allowed within each cluster (Wooldridge, 2016). The corrected standard errors shown in parenthesis in table 3 is higher across the board compared to the pooled OLS estimation, which reflects the sampling error from the pooled OLS.

Lastly, a Hausman test was performed, a model misspecification test that helps in choosing between the fixed-effects model and the random-effects model. The null hypothesis is that random effects are preferred model to run by looking to see if there is any correlation within the errors in the regression models.

We failed to reject the Hausman test (see Appendix 4), which points to either a large sampling variation or the models are sufficiently close that it does not matter which one is used. (Wooldridge, 2016). The latter assumption holds when looking at both coefficients, and their robust standard errors, where the only visible difference is with the control variables, where the difference in all, are around 0.001 log points and with the exact same estimates for MI and inflation. Therefore, both of these models are interchangeable. For random effects regression, see appendix 5.

All three models are consistent in terms of microfinance intensity on income inequality (-0,004). However, we see deviations and inconsistencies with theory in terms of the relationship for the interest rate and income inequality in the Pooled OLS model. Further on, as opposed to theory, both FE and RE models show a negative relationship between inflation and income inequality. However, the models failed to reach any significant level in terms of inflation, which makes it harder to estimate the real value.

(0.010) (0.013) (0.055) (0.110) (0.121) (0.115)

Observations 121 121 121 121 121 121

Number of countries 11 11 11 11 11 11

R-squared 0.392 0.395 0.433 0.509 0.512 0.560 Adjusted R-squared 0.387 0.385 0.418 0.492 0.491 0.537

(22)

18

7 Discussion

The results from all three models indicate a negative relationship between MI and Gini, in line with theory. However, the impact across all models is minimal in comparison to the required investment. With a 1% increase in MI, there would be a 0.004 percentage point decrease in Gini. This can be seen as the required increase in MI, to decrease Gini by one unit, from the mean of 0.5 to 0.49, would have to be 250%, holding everything else constant.

The relationship goes in line with past studies that have looked into the effect of microfinance. Various randomized controlled trials ((Tarozzi et al., 2015; Crépon et al., 2015; Banerjee et al., 2015; Angelucci et al., 2013) showed that an increase in microfinance led to increased profits and investments used for self-employment, however, found no evidence of an increase in income. It can be argued that the increase in profits has spillover effects for the more impoverished community’s long term, which leads to increasing incomes, which are in line with Angelucci et al. conclusion.

The reasons for the minimal impact could be for many reasons. Most of the loans are targeted to small businesses or individuals that are in the poorer parts of the country. This, in effect, would mean that their clientele also is weak in purchasing power, which means that to get ideal results, you would have to look at a more extended period. Even though the purpose of this study was to look at an extended period, we are of the opinion that a more extended period could paint a fairer picture of the results. It could also be because of the small reach of MFIs in the countries of our sample, compared with other banking institutes, and generally speaking, MFIs usually are small scale operations which have to charge a higher interest rate to cover their operating expenses because they cannot get the benefit of economies of scale that traditional banks have. We saw a positive but with no significance in terms of interest rates effect on the Gini coefficient. However, it is reasonable to assume a lower interest rate, would give the loan-takers more flexibility to invest their money and expand their income-generating activities.

Further on, we have not been able to pinpoint the exact nature of the shift in Gini. There is a possibility that a microloan has been effective enough to shift a low-income earner up to the top decile, which in turn could increase the Gini coefficient.

The results also indicate that the unobserved factors do not have a big impact on the estimation as both the pooled OLS and fixed effect model end up with the same coefficient in terms of microfinance intensity, which indicates that the pooled OLS already gave a good estimation.

(23)

19 Lastly, the fact that MFIs do not only provide capital but also provide savings, investment opportunities, and insurance, which has not been accounted for. These factors do not directly influence the income of the individuals, but gives them the stability that can influence their net worth rather than their income.

To summarize, the results in our paper are in line with past studies in terms of the positive albeit small impact of microfinance on income inequality. Although, one of our purposes was to try and provide a different angle in comparison to the studies using RCTs and studies mainly looking at one country.

8 Conclusion

This paper set out to examine the effect that microfinance has on income inequality and to what degree by using a cross-country framework with panel data methodology. This has been achieved by using Gini as a measure of income inequality and microfinance intensity as the main explanatory variable. We performed panel data regressions using both pooled OLS and fixed effects estimations. Our results confirm the hypothesis that higher participation in MFI is associated with lower income inequality, albeit modest.

This section will proceed to discuss policy implications, our limitations, and further discuss how future research can deal with the limitations.

8.1 Policy implications

Based on our results, we conclude that microfinance can be an effective tool for reducing income inequalities in a country. However, it is our opinion that it would require more centralization and scaling up the firms operating in this branch. The majority of MFIs during the period studied were a mix of non-governmental organizations, small MFIs, and in some cases, state-owned. There is reason to believe that if traditional banks were to get involved in a bigger scale, and thereby making the sector more effective and affordable by pushing down interest rates, we could see a bigger impact on reducing income inequality. Many of the MFIs are just MFIs, which means their main operating income, is interest rates provided by the relatively small loans. Subsidies could create incentives for banks to enter the market for microloans

(24)

20

8.2 Limitations and Future Research

The study has some limitations that are important to highlight. Firstly, as discussed in the data section, the missing observations could have underestimated our key independent variable. It is difficult to estimate the total impact of a complete data set; however, the missing data could have been for two reasons. Either it is a simple as missing data because of no records, which would mean we have underestimated the size of the Mix Market variables, or the MFIs have stopped operating, which in that case, our data is correctly specified. Nevertheless, since we had data on 384 different MFIs and the MFIs generally only had Spanish websites, it was difficult to source credible information that would lead us to credible information regarding the years with missing data. For future research, this can potentially be looked in to, to fill in the missing data if possible, or otherwise omit the MFIs that have no possibility of tracing the data. Further on, future research could take advantage of using more measurement for income inequality, which allows the researcher to pinpoint the exact nature of how the shift in Gini has occurred. This can be done by examining changes in income deciles and look into 90/10 or 50/10 wage gaps.

For future research, a more in-depth empirical knowledge needs to be done to truly understand the true effect that microfinance has on income inequality. A fiscal approach was not made in this study; future studies may analyze a way to combine fiscal policies with the effect of MFIs to gain better results. Variables such as taxes and transfers may affect the outcome of the regressions and the results as they directly contribute to an individual’s income.

It is also important to note that we have presented correlation, but not causation as experimental variation is hard to come by and thus link with income inequality at the societal level. This can be experimented with, for instance by using an average of MI for years’ prior as a control variable.

(25)

21

9 References

Angelucci, M., Karlan, D. and Zinman, J., 2013. Win Some Lose Some? Evidence from a Randomized

Microcredit Program Placement Experiment by Compartamos Banco. SSRN Electronic Journal,.

Allison, P., 2009. Fixed Effects Regression Models. 1st ed. Los Angeles: SAGE.

Banerjee, A., Duflo, E., Glennerster, R. and Kinnan, C., 2015. The Miracle of Microfinance? Evidence from a

Randomized Evaluation. American Economic Journal: Applied Economics, 7(1), pp.22-53.

Bangoura, L., Khary Mbow, M., Lessoua, A. and Diaw, D., 2016. Impact of Microfinance on Poverty and

Inequality A Heterogeneous Panel Causality Analysis. Revue d'économie politique, 126(5), p.789.

Beck, T., Demirgüç-Kunt, A. and Levine, R., 2007. Finance, inequality and the poor. Journal of Economic

Growth, 12(1), pp.27-49.

Borjas, G., 2015. Labor Economics. 7th ed. McGraw-Hill Education.

Brooks, C., 2019. Introductory Econometrics For Finance. 4th ed. Cambridge: Cambridge University Press.

Buchenrieder, G., Nguefo Gnilachi, J. and Benjamin, E., 2019. Poverty alleviation through microcredit in

Sub-Saharan Africa revisited. Agricultural Finance Review, 79(3), pp.386-407.

Cardoso, E., 1992. INFLATION AND POVERTY. National Bureau of Economic Research, 4006. From

Chen, C., 2016. The impact of foreign direct investment on urban-rural income inequality. China Agricultural

Economic Review, 8(3), pp.480-497.

Choi, C. 2006. Does foreign direct investment affect domestic income inequality?. Applied Eco1nomics Letters,

13(12), 811-814. doi: 10.1080/13504850500400637

Chowdhury, A. (2009). Microfinance as a Poverty Reduction Tool —A Critical Assessment.United Nations,

Department of Economic and Social Affairs (DESA) Working Paper, (89).

Choi, C., 2006. Does foreign direct investment affect domestic income inequality?. Applied Economics Letters,

13(12), pp.811-814.

Crépon, B., Devoto, F., Duflo, E. and Parienté, W., 2015. Estimating the Impact of Microcredit on Those Who

Take It Up: Evidence from a Randomized Experiment in Morocco. American Economic Journal: Applied Economics, 7(1), pp.123-150.

Dipankar, D., 2004. Microcredit in Rural Bangladesh: Is It Reaching the Poorest?,." Journal of Microfinance,

6(1).

Birgit, H., 2006. Access for All : Building Inclusive Financial Systems. World Bank,.

(26)

22

Mahjabeen, R., 2008. Microfinancing in Bangladesh: Impact on households, consumption and welfare. Journal

of Policy Modeling, 30(6), pp.1083-1092.Mitra, S. (2009). Exploitative Microfinance Interest Rates. Asian Social Science, 5(5).

Morduch, J 2008. How can the poor afford microfinance? Financial Access Initiative, Wagner Graduate School,

New York University, New York.

Our History | Accion. Accion.org. 2020. Retrieved 30 April 2020, from https://www.accion.org/about/history.

Rhyne, E., 2001. Commercialization and Crisis in Bolivian Microfinance. Findevgateway,.

Tarozzi, A., Desai, J., & Johnson, K. 2015. The Impacts of Microcredit: Evidence from Ethiopia. American

Economic Journal: Applied Economics, 7(1), 54-89.

Weiss, J., & Montgomery, H. 2005. Great Expectations: Microfinance and Poverty Reduction in Asia and Latin

America. SSRN Electronic Journal.

Wooldridge, J., 2016. Introductory Econometrics. 6th ed. Boston: Cengage Learning.

Yunus, M. 1999. Banker to the Poor (p. 136). Public Affairs.

Zou, H. and Li, H., 2002. Inflation, Growth, and Income Distribution: A Cross-Country Study. Annals of

(27)

23

10 Appendix

Appendix 1: Variance Inflation Factor (VIF)

Appendix 2: Test for heteroscedasticity

Appendix 3: Wald test for groupwise heteroscedasticity

Prob > chi2 = 0.3185 chi2(1) = 0.99

Variables: fitted values of GINI Ho: Constant variance

Breusch-Pagan / Cook-Weisberg test for heteroskedasticity . estat hettest

(28)

24

Appendix 4: Hausman test

Appendix 5: Random-effects GLS regression

(1) (2) (3) (4) (5) (6)

VARIABLES Gini Gini. Gini. Gini. Gini. Gini.

Microfinance intensity -0.008*** -0.008*** -0.006*** -0.004** -0.004** -0.004**

(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)

Inflation 0.001 0.001 0.000 0.000 -0.001

Prob>chi2 = 0.0000 chi2 (11) = 254.42

H0: sigma(i)^2 = sigma^2 for all i in fixed effect regression model

Modified Wald test for groupwise heteroskedasticity . xttest3

(V_b-V_B is not positive definite) Prob>chi2 = 0.1946

= 8.64

chi2(6) = (b-B)'[(V_b-V_B)^(-1)](b-B) Test: Ho: difference in coefficients not systematic

B = inconsistent under Ha, efficient under Ho; obtained from xtreg b = consistent under Ho and Ha; obtained from xtreg LogGDP -.0448254 -.0439882 -.0008372 .0094189 LogFDI .0065152 .0078353 -.0013201 .0013882 LogGLP -.0069601 -.005739 -.0012211 .0027725 LogInterest .0029147 .0018659 .0010487 .0003336 Inflation -.0009395 -.0007967 -.0001429 .0000632 MI -.0037827 -.0040002 .0002174 .0004862 fixed random Difference S.E.

(b) (B) (b-B) sqrt(diag(V_b-V_B)) Coefficients

(29)

25 (0.001) (0.001) (0.001) (0.001) (0.001)

Log of interest expense -0.006** 0.000 0.000 0.002

(0.003) (0.002) (0.003) (0.003) Log of GLP -0.011** -0.012** -0.006** (0.005) (0.005) (0.003) Log of FDI 0.001 0.008*** (0.005) (0.003) Log of GDP -0.044*** (0.009) Constant 0.535*** 0.530*** 0.626*** 0.739*** 0.737*** 0.810*** (0.015) (0.018) (0.049) (0.099) (0.111) (0.095) Observations 121 121 121 121 121 121 Number of Country 11 11 11 11 11 11 R-sq Within 0.392 0.394 0.432 0.504 0.507 0.558 R-sq Between 0.079 0.094 0.066 0.029 0.032 0.088 R-sq Overall 0.136 0.153 0.136 0.094 0.096 0.224

Cluster Country YES YES YES YES YES YES

Robust standard errors in parentheses

Microfinance Effect on Income Inequality in Latin America: A cross-country panel data study on the effects of microfinance on the income inequality in Latin America