Are the effects of microcredit access for men
larger than for women? Results from a
Randomized Control Trial in Bosnia and
Herzegovina.
Master thesis, International Economics and Business University of Groningen, Faculty of Economics and Business
June 20, 2017
Floor M. hooge Venterink f.m.hooge.venterink@student.rug.nl
S2165244
ABSTRACT
Using data from a randomized control trial performed amongst marginal borrowers in Bosnia and Herzegovina in 2008 to 2010, this paper estimates the different impact access to
microcredit might have on men and women. The results show that as a result of microcredit access, female borrowers consume more nondurables and save more. Furthermore, labor hours increase more per household member of female borrowers in comparison to male borrowers. However, with male borrowers given microcredit access, children aged 6 to 15 years attend school significantly more as opposed to female borrowers. No significant
differences between men and women are found for outstanding loans, microbusiness activities and household income.
PREFACE
This thesis is the last part of my Master International Economics and Business at the
University of Groningen. During my Master, I discovered microfinance as an interesting field of research and therefore I decided to choose it as a topic for this thesis. At first it seemed not possible for me to write about microfinance, as this was a topic belonging to a different department at my faculty. Fortunately, Robert Lensink was so kind as to supervise my thesis. I am very glad that I was thus able to write about such an interesting topic, of which I hope it can improve the lives of many people among the world.
I would like to thank my supervisor Robert Lensink for guiding me through the process and providing me with valuable feedback to improve my thesis. Furthermore, I would like to thank my parents and my sisters for proofreading my thesis, giving new insights and for their constant support and encouragement.
TABLE OF CONTENTS
1. INTRODUCTION 1
1.1 Microcredit effects 1
1.2 Microcredit for men versus women 2
1.3 Randomized Control Trials 2
1.4 Research question 3
1.5 Research results 3
1.6 Outline 3
2. RESEARCH FRAMEWORK 4
2.1 Impacts of microcredit access on women 4
2.2 Impacts of microcredit access on men 6
2.3 Empirical methods revised 7
2.4 Theoretical perspective 8
2.5 Determinants of the impact assessment 9
2.5.1 Outstanding loans 9
2.5.2 Microbusiness activities 10
2.5.3 Household income 11
2.5.4 Consumption and savings 12
2.5.5 Labor 13 2.5.6 Social environment 14 3. RESEARCH DESIGN 15 3.1 Data collection 15 3.1.1 The loan 16 3.2 Sample 16
3.3 Ordinary Least Squares estimation 16
3.4 Balance check 17
3.4.1 Results of men 18
3.4.2 Results of women 20
3.5 Attrition 22
3.6 Estimating causal impact for borrowed amounts 22
4. ANALYSIS 23
4.1 Results 23
4.1.1 Outstanding loans 23
4.1.2 Microbusiness activities 24
4.1.3 Household income 25
4.1.4 Consumption and savings 26
4.1.5 Labor 28 4.1.6 Social environment 30 4.2 Comparison results 31 5. CONCLUSION 32 5.1 Summary of findings 32 5.2 Policy recommendations 33 5.3 Limitations 33
5.4 Suggestions for future research 34
6. REFERENCES 35
7. APPENDIX 38
A. Number of men and women 38
B. Variables definitions 38
C. Attrition 41
1. INTRODUCTION
Since a couple of decades, microcredit is seen as a tool to help eradicating poverty by
providing access to credit for the poor so they can earn an income through their own business (van Rooyen, Stewart, & De Wet, 2012). Microcredit involves the provision of “small loans to very poor people for self-employment projects that generate income, allowing them to care for themselves and their families” (Reed, 2015). The microcredit world has been growing rapidly. According to the Microcredit Summit Campaign, Microfinance Institutions reached 211 million clients in 2013 coming from 13 million in 1997. More than half of those 211 million borrowers were living in extreme poverty, meaning people who live on less than $2 a day (World Bank definition). Therefore, the World Bank regards universal financial access as a key contributor to ending extreme poverty (Reed, 2015). Microcredit is said to not only give an opportunity to the poor to achieve economic development, but it could also improve health, education levels and promote women empowerment (Sharma & Puri, 2013). The focus of this research will be on the possible different effects access to microcredit might have on men as compared to women. These effects will be analyzed, because so far, previous research showed various results. Furthermore, knowledge about these effects is important for Microfinance Institutions (MFIs) and policy makers to give the concept a more meaningful role in developing countries.
Underneath, the various results of previous studies are briefly discussed to provide the context of this research, as well as the different impacts for men and women and the research methods. This is followed by the research question and the results, to conclude with the outline of this thesis.
1.1 Microcredit effects
Despite the outreach of this global movement and the positive expectations of the concept, there are doubts about the impacts of microcredit as the evidence is mixed. A study in India showed that access to microcredit led some borrowers to expand their business, but the business profits of most borrowers did not increase. Monthly consumption levels remained unchanged, though incomes were spent on different goods. Furthermore, education, health and women empowerment were not affected by access to microcredit (Banerjee, Duflo, Glennerster, & Kinnan, 2014). Coleman finds only a positive effect of microfinance1
for the more wealthier villagers in Northeastern Thailand (2006). Hermes and Lensink (2011) point out that access to finance might cause increases in income but it is uncertain whether it substantially reduces poverty. On the other hand, Khandker (2005) concludes, using panel data from Bangladesh, that microcredit access does reduce poverty, especially for female borrowers. Similar results were found in 1998 already. Researchers Pitt and Khandker (1998) found larger effects for female participants in a credit program in Bangladesh, on household consumption expenses, women’s labor supply and schooling of girls. A study into the empowerment of women due to microcredit access in Ghana reports varying results. Some
1 In this thesis, microcredit and microfinance are used interchangeably, as microcredit falls under the heading of
women were indeed empowered when having access to credit, while other women
experienced no empowerment as someone else took control over the use of their loans and yet other women were worse off being subjected to harassment and no longer able to repay their loans, ending up with debts as well (Ganle, Afriyie, & Segbefia, 2015).
1.2 Microcredit for men versus women
As women are often targeted by MFIs, many microfinance studies focus on credit programs for women (Ghosh & Vinod, 2017). Commercial banks which provide formal financial services are more inclined to favor men, therefore MFIs are the solution for financing women. According to Armendariz and Morduch (2010) women constitute a majority of the poorest microfinance clients; thus, this accords well with poverty-reduction goals of NGOs. Besides the increase in income, women could be empowered by microfinance too, having an income of their own and thus being able to make their own economic decisions (Pitt, Khandker, & Cartwright, 2006). Women would also have better repayment rates, tend to be more risk-averse than men and spend more on household expenditures as compared to men when their income grows. Moreover, women are less mobile, which makes monitoring easier and cheaper for MFIs and the related social control causes women to default less (Armendáriz & Morduch, 2010).
Yet, the fact that MFIs focus on women for all the above reasons does not imply that microfinance access has a larger effect on them as compared to men when looking at
individuals. The bias in favor of women is contradicted by studies where the effects of microcredit access are smaller for women or even larger for male borrowers. For instance, Kevane and Wydick (2001) developed a model to compare the results of male and female entrepreneurs in a Guatemalan credit program and concluded that gender differences were small. Though the model does show small differences in changes in enterprise employment, differences in business income are not statistically significant. In addition to this, increased microcredit access caused profits of male entrepreneurs in Manila to increase (Karlan & Zinman, 2009). A field experiment in Sri Lanka, where grants were provided to male and female microenterprise owners, resulted in large income increases only for the men in the sample. Moreover, the women were not able to realize a return on their investments even though they invested the same amounts or more as the men did (de Mel, McKenzie, & Woodruff, 2009).
1.3 Randomized Control Trials
consider the increasing application of randomized control trials (RCT’s) as a key development.
1.4 Research question
As randomized control trials are scarcely used to look into the different effects from
microcredit access on men compared to women, this will be done in this thesis. In this way, a different manner of research will contribute to a small but expanding literature on the impact of microcredit access on men and women. This research is based on data from an experiment that took place in Bosnia and Herzegovina in 2008-2010, performed by Augsburg, De Haas, Harmgart and Meghir (2015). In a special volume on randomized impact evaluations of microfinance published by the American Economic Journal, the authors report various results. They present some evidence of higher self-employment, increases in inventory and profits and a reduction in work wages, consumption and savings for their marginal loan applicants. All effects are average effects for the whole sample, yet there are hints towards different effects for different subgroups. For instance, effects might differ between younger and older borrowers, between higher and lower educated borrowers, or between female and male borrowers (Banerjee, Karlan, & Zinman, 2015). Therefore, this paper broadly follows their method, an Ordinary Least Squares Estimation, in order to answer the question whether microcredit access has larger positive effects for men as compared to women on outstanding loans; self-employment activities; household income; consumption; labor; and social
environment.
1.5 Research results
In short, the results of this research are described here, but will be thoroughly discussed in section 4. They show that male borrowers have increased their outstanding loans at the time of the end line survey but there is no significant difference with female borrowers.
Furthermore, men who have access to microcredit also see a decrease in their possibility to earn income through wages and government benefits but, again, there is no significant difference with women who have microcredit access. Concerning consumption, male
borrowers spend less in total. This might be viewed in a positive way as they spend especially less on cigarettes, alcohol and nondurables, but unfortunately also save less. These last two effects are significantly different from female borrowers who have microcredit access. Similarly, as a consequence of microcredit access the total hours worked by the household members of female borrowers have increased in comparison to the hours of the male household’s members. Finally, a significant result is found for men with microcredit access showing they have less children aged 16 to 19 years attending school. However, when
compared with female borrowers, another significant effect shows male borrowers have more children attending school aged 6 to 15 years.
1.6 Outline
Also in this part, the final sample is described, as well as the estimation method and balance checks. Moreover, attrition is addressed in three ways and for several variables which appear unbalanced, another test will be discussed. The results and discussion of the analysis are presented in section 4, to conclude with a summary and the recommendations in section 5. Limitations and suggestions for further research are given in this last section as well.
2. RESEARCH FRAMEWORK
From earlier research, it becomes clear that access to microcredit influences the economic situation of poor people in various ways and to various extents. Studies are performed in many different countries with different circumstances, which might influence the results. For instance, the provided products vary; loans of different sizes with different repayment rates are offered or grants are given. The main focus of Microfinance Institutions are on women, more than half of the borrowers around the world are women (Reed, 2015). Therefore, effects for different population segments are researched. In some studies, average effects on
borrowers are examined, in other studies women are compared with men or relatively wealthy borrowers versus poorer borrowers. Sometimes business profits, incomes, consumption, labor and savings are of interest, while other researchers are more focused on social impacts in relation to gender such as women empowerment, education, health and nutrition. Moreover, research methods vary widely. Thus, this research will focus specifically on microcredit access and the effects it might have on male participants as compared to female participants, concerning outstanding loans, microbusiness activities, income, consumption and savings, labor and social environment, by using data from household surveys obtained in a randomized control trial. Someone is said to have microcredit access when offered to borrow an amount from MFIs or banks.
Underneath, a review of the literature on microcredit access is given, followed by theoretical reasons for possible differences in the impact of microcredit access on men and women. After this, hypotheses are formulated based on the determinants for the impact assessment.
2.1 Impacts of microcredit access on women
gender, as women who followed a higher education tended to borrow less, but also based on resources. Households with less resources, especially land assets, had a higher demand for microfinance loans. Women who did borrow, experienced a higher increase in consumption as compared to male borrowers. Furthermore, when women borrowed there were more spillover effects to non-borrowers in the village and poverty rates declined more amongst female participants. Yet, this was not the first research done by Khandker where evidence was given for the claim that microcredit access is more beneficial for women.
Already in the 1980’s women were successfully targeted by the Grameen Bank, since clients consisted out of women for almost three-quarters of Grameen’s members and these numbers were rising (Armendáriz & Morduch, 2010). The preference of women as clients of microfinance was confirmed by another research, amongst others, of Khandker in
collaboration with Mark Pitt. Again in Bangladesh, the impact of credit programs was
evaluated with a special focus on gender differences (Pitt & Khandker, 1998). In cooperation with three MFIs, Grameen Bank, Bangladesh Rural Advancement Committee and Bangladesh Rural Development Board, the researchers examined the effect of access to microcredit on household behaviors and on the intrahousehold distribution of resources to determine if targeting women with microcredit really matters. For this, they used a quasi-experimental survey design conducted in households of villages where programs of the three MFIs took place and in households of villages where no programs took place. Household consumption, assets, labor supply and schooling were the effects of interest, because credit was found to be a significant determinant for these impacts. Credit offered to women was more likely to influence these outcomes than credit offered to men. Considering household consumption; an increase in credit provision for women led to a higher increase in household consumption as compared to men. Most likely, this is related to the increased productivity of women’s labor supply when women become microcredit clients. Furthermore, female participation in credit programs caused the value of nonland assets to rise more than with male participation. As for the schooling of boys and girls, boys schooling increased when credit increased for both men and women, while girls schooling only increased when women received more credit.
The earlier work described above was the basis for a study into the underlying mechanisms of these gender differences by Pitt, Khandker and Cartwright (2006). They assumed that: “participating in micro credit programs is an empowering experience for women whose life choices are otherwise restricted through poverty, patriarchy, and societal or religious norms”. This assumption is based on the believe that both empowerment and income and substitution effects can lead to different outcomes of microfinance for men and women. Empowerment was examined for both men and women who participated in a credit program, by making use of survey data obtained in Bangladesh on women autonomy and gender relations within the household. Empowerment was indexed with various indicators on household’s income and consumption patterns, but also impacts on the political and social environment. The findings showed consistency with the above results; that female
valued more. In contrast, the effects on women’s empowerment indicators when men participated in a credit program were negative.
2.2 Impacts of microcredit access on men
Many impact assessments focus on the effects for women and, thus, perform studies in projects where the microcredit products are offered to women. However, with gender differences one has to take into account the region- and culture specific circumstances (Armendáriz & Morduch, 2010). Therefore, questions raised about whether microcredit has indeed more effect on women due to the results of several researches in other countries. In the early 2000’s, a lively discussion on a possible trade-off between increasing household welfare due to women having access to credit at the expense of economic growth was going on. Recall that Pitt and Khandker showed in 1998 already that when women had the possibility to loan, their household consumption increased more than men’s and they also found positive effects on schooling of girls and boys. Besides this, there was also evidence of gender differences concerning investments, while women are more risk-averse wanting to ensure a certain household income, men are said to invest in riskier, but higher-yielding projects (Downing, 1990). Consequently, men showed not only higher rates of return on capital but also contributed more to long-term economic growth by expanding their enterprises more. Kevane and Wydick (2001) attempted to contribute to this debate with their research on gender differences in entrepreneurial behavior. According to them, several gender
asymmetries might have an influence on these differences. First of all, access to credit is poor in many developing countries, but especially so for women. Moreover, traditionally, women take care of the children at home, while men have a job to earn an income for the family. Following this line of thinking, it is more difficult for female entrepreneurs to hire and supervise male workers. With these asymmetries in mind, a model was build and estimations were performed on first-hand survey data of entrepreneurs in Guatemala. Differences did exist in employment generation between male and female entrepreneurs, mainly because of the time constraints for women in childbearing and child raising years, but there was little difference in the generation of increasing sales.
Nevertheless, the microcredit industry kept growing, women were still targeted and research continued as well. Advocates remained firm in their belief that expanding credit access contributed to fighting poverty and promoting growth. However, critics such as Karlan and Zinman argued that there is little evidence for these positive effects on borrowers.
Lastly, the effects seemed to be stronger for male borrowers as well as relatively high-income borrowers, thus questioning whether targeted groups are actually affected by microfinance access. The researchers conclude: “that microcredit works broadly through risk management and investment at the household level, rather than directly through the targeted businesses”.
Following both the ideas and results of Kevane and Wydick (2001) and Karlan and Zinman (2009); de Mel, Mckenzie and Woodruff (2009) looked not only into gender differences concerning business profits, but also in relation to entrepreneurial behavior. Beforehand, they expected women to show larger increases in income as they are assumed to be more credit constraint. Moreover, they are aware of the findings of Kevane and Wydick suggesting female borrowers are more focused on short-term household consumption at the expense of long-term growth of their microenterprise. However, they take into account the competing demands from individual household members. In their view, the bargaining power of individual household members will affect the decisions on consumption and production, among others. Here, they expected efficient outcomes of household asset allocation. To study both these relations, the researchers set up an experiment in Sri Lanka, where microcredit is provided in the form of small or large grants to microenterprise owners by conducting a random prize drawing. In the first two years, firms were interviewed four times per year and in the last year this was reduced to two times per year. The interviews mainly focused on business profits, revenues and expenses, capital stock and inventory. At the beginning of the experiment, halfway and in the end, participants were also asked about household
consumption, education and labor. The study showed that female borrowers who received the smaller grant, invested little of the amount in comparison to men, but this amount increased when women were given the larger grants. Yet, the larger amounts did not result in increased profits for women. On the other hand, for male owners the larger grants did result in a larger income. They try to explain these differences and found that differences in entrepreneurial abilities, credit access, risk aversion and profit reporting played no significant role. In considering the model of efficient household decision-making, they found very limited evidence for women to spend more on schooling for their children and no evidence for an increased investment on health or household durable goods. Expenditures on household durable goods even increased more with male owners. Lastly, de Mel, McKenzie and Woodruff show similar gender differences in Brazil and Mexico, thus giving some evidence that their findings also hold in other contexts. In their view, an explanation as for why these differences exist is an important topic for future research.
2.3 Empirical methods revised
effects. Instead of quasi-experiments, they favor randomized trials in order to eliminate selection bias. The reason for this, is that borrowers may be significantly different from non-borrowers and this might influence the estimates (Odell, 2010). Moreover, they believe more replication studies should be done. This in order to be able to make the findings more robust. In reply of the replication done by Roodman and Morduch, Pitt (2014) wrote a response on the given criticism. According to him, the claims made are based on a flawed understanding of econometrics and he argues that the researchers did not make an effort to understand his statistical model. A heated debate about the best research method for the impact of
microcredit access on male and female borrowers and their communities continues nowadays. RCT’s appear to be a valuable method to provide causal evidence of microcredit impacts (Banerjee, Karlan, & Zinman, 2015). Yet, quasi-experimental and experimental studies have a limitation in common, as they often only assess average impacts. In the case that the effects for half of the borrowers are positive and for the other half negative, almost no effect might be reported (Odell, 2010). Therefore, this study looks into the effects for two groups, men and women, as the impact might be different for different groups.
2.4 Theoretical perspective
Besides the empirical evidence for a different impact of microcredit access on men and women, several theoretical concepts could also explain these differences. First of all is the standard neoclassical assumption on returns to capital. This assumption states that those who have relatively little capital should be able to earn higher returns to their investments
(D'espallier, Guerin, & Mersland, 2013). The assumption works through the principle of diminishing marginal returns to capital. Under this principle, it is assumed that those who have more capital and thus invest more, are expected to earn higher returns. However, each additional invested amount of capital only brings smaller marginal gains, due to curved production functions. In these production functions, the first investment brings the biggest gains and subsequent investments most likely only lead to smaller incremental gains (Armendáriz & Morduch, 2010). Thus, someone who has little capital to invest, should be able to earn the larger gains. In this particular case, this assumption would imply that women should be able to earn higher returns to capital as compared to men, since one of the reasons behind women being targeted by MFIs is that they have less access to finance, and once they do have microcredit access, this should thus have a larger impact on women (Armendáriz & Morduch, 2010). This is especially so because the marginal participants in this research are normally not eligible to borrow at this MFI.
riskier and higher yielding projects results in the opposite effect. They show more risk-taking investment behaviors and tend to expand their business more aggressively, thus increasing the impact of microcredit access (Downing, 1990).
Thirdly, the intra-household decision making process might explain different
microcredit access impacts. Though not directly, this process might give reasons for different changes due to microcredit access between men and women in consumption and savings patterns, labor and education. The unitary approach developed by Gary Becker suggests an efficient labor time allocation of all household members between market production and household activities (Armendáriz & Morduch, 2010). According to Becker, it is most beneficial for men to work in strength-intensive market production activities, while women stay in the household or work in the informal sector. Yet, nowadays, another intra-household decision making process is considered, in which all household members act not as one unit, but men and women might make different choices about consumption, labor and education. More specifically, this collective approach to household decision making is changing the allocation pattern of a household’s resources. Theoretically, women are expected to have stronger preferences for spending income on their children, their education and household needs, while men are expected to invest their income in their business or spend money on cigarettes and alcohol (de Mel et al., 2009).
2.5 Determinants of the impact assessment
As mentioned earlier, the data which is used in this research was collected for the article: ‘The Impacts of Microcredit: Evidence from Bosnia and Herzegovina’ by Augsburg, De Haas, Harmgart and Meghir (2015). This research was published as part of a series of randomized control trials which differ in sampling, data collection, research methods and econometrics, but share the same goal of trying to find evidence for the impact of microcredit access. Furthermore, all articles estimate intention-to-treat effects which implies that participants are offered the possibility to loan, but effects are measured regardless of whether the loan is actually taken up (Odell, 2010). The same determinants of the impact are estimated, which are outstanding loans, microbusiness activities, household income, consumption patterns, labor, and social environment. These outcome groups are measured by several variables which are tested separately with linear regressions. Underneath, hypotheses are formulated for the same determinants based on existing literature. Also, the expected outcomes of the variables which are tested for each hypothesis are provided in Tables 1-6. In all hypotheses for the male and female borrowers the variables move in the same direction, though some increase or decrease more for either the male or female borrowers depending on the hypothesis.
2.5.1 Outstanding loans
To determine whether microcredit access affects the outstanding loans of male and female borrowers, four variables are estimated. Participants are asked if they had any loan
outstanding at the beginning and the end of the experiment, and also which institution provided these loans, either a bank or a MFI. The outcomes are most likely to be positive, as male borrowers might have more access to credit from banks, and female borrowers are able to attract more loans from MFIs. Moreover, Banerjee et al. (2014) find that access to
are experiencing, they tend to invest in projects with lower risks, thus enabling them to repay their loans. In turn, this might cause women to increase their chances on follow-up loans. As women are said to have less access to formal financial services from banks, it is more likely that women will borrow their follow-up loans at MFIs (Armendáriz & Morduch, 2010). Therefore, the following hypothesis is formulated:
H1: Outstanding loans are for women in Bosnia and Herzegovina more positively affected by microcredit access than for men.
Table 1 - Variables of H1 and their expected outcome
Any loan outstanding at end line Increased Number of loans outstanding at baseline Increased Any loan outstanding to any MFI at end line Increased Any loan outstanding to any bank at end line Decreased 2.5.2 Microbusiness activities
Presumably, men and women who are offered the possibility to borrow will see a change in the activities concerning their business. This will be evaluated by the revenue, expenses and profit of the respondent’s main business, but also by the value of their assets, whether they have inventory or not and the chance on any self-employment income. Pitt and Khandker (1998) discovered, though a larger effect for women as for men, that for both, asset values tend to increase due to microcredit access. According to Banerjee et al. (2014) microcredit access also has a positive influence on having inventory, and therefore on business expenses, but also revenues, profits and income tend to increase, for men and women. Increasing revenues for men are confirmed by Kevane and Whydick (2001), yet Coleman (2006) finds revenues only to increase for the more wealthier borrowers. Both Karlan and Zinman (2009) and de Mel et al. (2009) report evidence for increasing profits for male borrowers. Positive effects for male entrepreneurs on their income are also shown by the results of Kevane and Whydick (2001) and de Mel et al. (2009). Under the neoclassical assumption on capital returns, as women have indeed less access to capital, their returns will be higher. This might imply higher investments in their business and thus increased asset values, inventory, revenue, expenses, profit and ultimately, self-employment income. On the other hand,
women’s fear of social sanctions and their tendency towards risk-averse investment behavior, decreases these effects. Moreover, the more aggressive, risk-taking investment behavior of men and their tendency to invest in higher-yielding business can increase their asset values, inventory, revenue, expenses, profit and thus, self-employment income.
Furthermore, if the respondents own a business and if so, in which industry - either services or agriculture - will be taken into account too. The possibility that the respondent closed or started a business in the months during the survey is considered as well. As
business might become less likely due to microcredit access. Underneath the hypothesis for these variables is developed:
H2: Microbusiness activities are for men in Bosnia and Herzegovina more positively affected by microcredit access than for women.
Table 2 - Variables of H2 and their expected outcome
Asset value Increased
Ownership of inventory Increased
Respondent’s main business’ revenue Increased Respondent’s main business’ expenses Increased Respondent’s main business’ profit Increased Any self-employment income Increased
Business ownership Increased
Business in services Increased
Business in agriculture Increased
Whether or not the respondent has started a business in the last 14 months
Increased Whether or not the respondent has closed a
business in the last 14 months
Decreased
2.5.3 Household income
Probably one of the most researched relationships is the effect of access to microcredit on income. In this research, several types will be taken into account. First of all, the possibility that the men and women earn an income through self-employment activities and
consequently, also the amount might change. As described above this relationship is likely to be positive (Kevane and Whydick, 2001; de Mel et al., 2009). Again, the neoclassical
assumption on capital returns might increase women’s income, but indirectly their risk-averseness and fear of social sanctions might decrease their self-employment income. Since men are expected to invest in higher-yielding businesses, their self-employment income might increase. Included in income also, are the wages from salaried work done in several industries such as agriculture, commercial services, finances and government. Again, the possibility and the amount are estimated. Here, several substitution effects might play a role, as borrowers might start working more in their own business at the expense of their salaried work, thus decreasing income by wages. Also, incomes brought in by children, but who are now send to school, might be replaced by an increase in business incomes (Karlan & Zinman, 2009). The unitary household decision making process, in which men should focus on strength-intensive market production activities, might positively influence self-employment income. Though for women, who should focus on the household activities, this process might influence wages in a negative way. Thirdly, the possibility and amount of an income through remittances are considered. Remittances are transfers of funds across borders, often made by friends or family who work and live abroad (The World Bank, 2016). Karlan and Zinman (2009) find no
significant effect for remittances, but as business income grows, they might no longer be necessary and thus decrease. The last type of income taken into account is the possibility and amount of an income from government benefits. Odell (2010) states as a goal from
villages to be less dependent on government aid in the future”. This might infer that access to microcredit will decrease government benefits for male borrowers. The next hypothesis results from these findings:
H3: Household income is for men in Bosnia and Herzegovina more positively affected by microcredit access than for women.
Table 3 - Variables of H3 and their expected outcome
Possibility to earn an income and the amount from self-employment
Increased Possibility to earn an income and the amount
from wages
Increased Possibility to earn an income and the amount
from remittances
Decreased Possibility to earn an income and the amount
from government benefits
Decreased
2.5.4 Consumption and savings
Following the changes in income, a change in consumption patterns due to microcredit access for men and women is a topic of interest too. In the surveys, participants are asked about their total consumption amount, but also the distribution of this amount over the following
categories: durables, nondurables, food, education, cigarettes and alcohol and recreation. Both the studies of Pitt and Khandker (1998) and Khandker (2005) show an increased total
consumption, though higher for women as for men. In addition, Banerjee et al. (2014) report a small positive effect, however statistically insignificant for total consumption and
nondurables. They do find a significant effect for increased spending on durables and decreased spending on goods such as cigarettes, alcohol and food outside the home. Credit access seems to have an insignificant effect on consumption of more qualitative food products (Karlan & Zinman, 2009). However, in a different study in South Africa the same authors do report a positive influence on food quality as well as food quantity (Karlan & Zinman, 2010). Mixed evidence is also found for education expenditures (Odell, 2010). Banerjee et al. (2014) discover no statistically significant results, but several researchers do find a positive effect on education participation (Pitt and Khandker, 1998; Karlan and Zinman, 2009; de Mel et al., 2009), which might imply an increased spending on education. Furthermore, 18 home durable goods are collected in a Home Durable Goods Index (HDGI). These are goods for the house, business or land. As mentioned before, both having inventory and business expenses
increased, which indirectly suggests consumption on durable goods for the business to
RCT’s, savings are positively influenced by microcredit access. Based on these findings, this hypothesis is formulated:
H4: Consumption and savings are for women in Bosnia and Herzegovina more positively affected by microcredit access than for men.
Table 4 - Variables of H4 and their expected outcome
Total consumption per capita Increased
Durables Increased
Nondurables Increased
Food Increased
Education Increased
Cigarettes and alcohol Decreased
Recreation Decreased
HDGI Increased
Savings Increased
2.5.5 Labor
Another determinant for the impact assessment of access to microcredit on men and women regards labor. This is measured by hours worked per week in total, on business activities and on other activities. These hours are calculated for adults and teens in the household together, but also for teens only and per household member. Not many researchers have looked into the effects of microcredit access to labor, but Coleman (2006) did find a significant relationship exists, even for what he defines as “rank-and-file” members of the community, who tend to be poorer, instead of the wealthier “committee” members. The estimated impact of microcredit access appears to be negative for labor time. For instance, when a farmer is able to buy an ox, due to microcredit access, to work on his lands, this might decrease his labor hours necessary to harvest his crops. This decrease could occur, because the farmer is able to make more efficient use of his labor hours. Results on household member employment in other businesses seem to be negative too, especially for male borrowing family members, and though not statistically significant, the possibility that a household member helps in a family business decreases too (Karlan & Zinman, 2009). Finally, Karlan and Zinman also show that more children appear to be in school instead of working (2009). In households of male
borrowers this might be less, since they only tend to send their boys to school, whereas female borrowers also send their girls to school (Pitt & Khandker, 1998). According to the unitary household decision making process, it is most beneficial for men to work on market
production activities, thus influencing their labor hours worked in the business and probably hours worked in total. Recalling the example of the ox, this might decrease these labor hours. On the other hand, women should also devote their time to household work and are expected to care more for their children and their education. This might influence the hours worked in female households in total and on other activities. In this research, a positive effect is a
Table 5 – Variables of H5 and their expected outcome
Total hours worked by all household members and per household member
Decreased Hours worked on business by all household
members and per household member
Decreased Hours worked on other activities by all
household members and per household member
Decreased Total hours worked by teens in household and
per household member
Decreased Hours worked on business by teens in household
and per household member
Decreased Hours worked on other activities by teens in
household and per household member
Decreased
2.5.6 Social environment
The last group of outcomes, which might be affected by offering microcredit access to men, concerns two social factors, the schooling of boys and girls of different age groups and the stress level experienced by the participants. In total, five variables are estimated, namely the share of children who are 6 and 15 years old and who attend school, the share of children aged 16 to 19 years who attend school, which seems to increase as evidence of Pitt and Khandker (1998), Karlan and Zinman (2009) and de Mel et al. (2009) shows. While, Pitt and Khandker find that access to microcredit for both men and women increases schooling for boys, schooling for girls is only influenced by microcredit access for women. This is supported by the collective household decision making approach which allows for different preferences amongst men and women. This theory expects women to want to invest more in education for their children as compared to men. On the contrary, Karlan and Zinman did not find evidence for this gender difference and de Mel, McKenzie and Woodruff found limited evidence that women invest more in the schooling of children. Thirdly, the possibility of the household having children who are 16 to 19 years old is considered and, fourthly, the number of children in the household who are 16 to 19 years old. These are merely to see whether an increasing share of children aged 16 to 19 years going to school might be caused by an increased number in children of these ages. The last variable is a stress index. This stress index is based on several questions asked to participants on their perceived stress level. A small significant and increasing effect is found for male borrowers (Karlan & Zinman, 2009). On the contrary, women’s fear of social pressure and sanctions and their aversion against taking risks might increase their stress levels. From these results, the next hypothesis is formulated:
H6: Social environment is for women in Bosnia and Herzegovina more positively affected by microcredit access than for men.
Table 6 - Variables of H6 and their expected outcome
3. RESEARCH DESIGN 3.1 Data collection
For the data collection, a field experiment was conducted with the help of a large Micro Finance Institution active in Bosnia and Herzegovina. Bosnia and Herzegovina is chosen, because many impact assessments are already conducted in Asian countries such as Bangladesh, the Philippines and Sri Lanka, but this Eastern European country is under assessed in this field. For this experiment, marginal microcredit clients were targeted.
Marginal borrowers are defined here as those clients who are poorer and more disadvantaged as compared to the regular clients. Normally this population segment is not eligible to
participate in loan programs, however now they were considered under the acceptance of slightly more risk. Though this might limit the external validity of the results, it also provides the opportunity to look into the effects of microcredit access for the poorer segment of the population. Moreover, in Bosnia and Herzegovina a proper microcredit system was already in place during this data collection process, thus making it impossible to form an appropriate control group out of the regular borrowers.
The loan officers of the Bosnian MFI made a selection of the marginal applicants based on the regular screening procedures. They started by determining whether the
applicants met the MFIs requirements, such as the available amount of collateral, repayment capacity, overall creditworthiness, the client’s business capacity and credit history. Also, the loan officers had to rate 12 characteristics, on whether they thought the client was competent, reliable, aggressive, trustworthy, clever, stable, experienced, knowledgeable, well-integrated into society, a risk-taker, insecure or a fighter. With these criteria, potential marginal clients were identified and they were informed on the conditions of the loan. As they were not
eligible for a loan before, but the MFI was changing its policies, they would have a 50 percent chance of being provided with a loan under the condition that the participant would agree to participate in a survey now and in the next year. In this way 1,241 marginal applicants were reviewed, of which 1,196 were approved for the final selection and subsequently, interviewed in the baseline survey. Interviews started in 2008, were conducted on the phone by a
professional survey company and lasted around 1 hour.
3.1.1 The loan
The loans offered to participants in the research were similar to the loans the MFI normally offers with regards to the interest rate and maturity. The repayment of the individual-liability loans started immediately and were done every month. The Annual Percentage Rate (APR) was 22% (as compared to the regular rate of 21%), which was paid over the remaining
amount. The loan amounts depended on the business plan, but varied from 300 BAM to 3,000 BAM with a mean of 1,653 BAM. The average maturity was 57 weeks (Augsburg et al., 2015).
3.2 Sample
In the baseline survey, 1,196 applicants participated of which slightly more than 50 percent were allocated to the treatment group, namely 627 applicants, and the control group consists of 569 applicants. However, several participants did not participate in the follow-up survey for various reasons, such as refusals, moves, invalid contact information or personal
circumstances. Thus, in the end line survey 995 participants were left, 550 applicants in the treatment group and 444 in the control group. The attrition rate seems to be higher for the control group, which might influence the results and therefore several checks for attrition are done. Detailed information on the number of men and women in the treatment and control group can be found in Table A1 in the Appendix. To ensure that the men and women in the treatment and control group share similar characteristics, balance tests are performed. Also, other econometric problems are addressed, but first the estimation approach for the main analysis is explained. The definitions of all variables are provided in Appendix Table B1. 3.3 Ordinary Least Squares estimation
As the participants are randomly assigned to either the treatment group or the control group, the effects that might occur are due to microcredit access. To examine this direct relationship between microcredit access and the determinants of the impact for men and women, an Ordinary Least Squares (OLS) estimation is performed. This is done for the hypotheses on outstanding loans, microbusiness activities, household income, consumption patterns, labor, and social environment. The variables which are used to measure the impact of each
hypothesis are estimated separately. All regressions are done for the complete post attrition sample, together with the set of baseline characteristics, in order to be able to compare the effects of microcredit access for men and women. All findings are intention-to-treat effects, as participants in the treatment group are not obliged to take up the offered loans, but either way, effects are measured. The standard errors are robust to heteroscedasticity, thus allowing the variance of the errors to be different for different observations (Carter Hill, Griffiths, & Lim, 2012).
Broadly, this approach is similar to the one used by Augsburg et al. (2015). In this way, the results will have the same meaning but will be different in size, sign and
The model for the linear regression of each dependent variable is as follows:
Y = ß0 + ß1Female + ß2Treatment + ß3(Treatment*Female) + ß4Baseline + µ where Y is the outcome of the regressed dependent variable, ß0 is the intercept parameter, ß1Female is a dummy variable taking the value of 1 when the respondent is female,
ß2Treatment is a dummy variable taking the value of 1 when the respondent was offered to loan, ß3(Treatment*Female) is the interaction variable of ß1Female and ß2Treatment,
ß4Baseline is a set of baseline characteristics, such as age, marital status, highest school grade, household composition (females, employed, school, retired) and number of kids of various ages and µ is the error term. Each variable measures different effects. ß0 indicates the average effect on the dependent variable for the men who do not have the possibility to loan.
Together, ß0 and ß1Female give the average impact on the dependent variable for the women who do not have the possibility to loan. When ß2Treatment is added to ß0, the results show the average effect of microcredit access on the dependent variable for the men who are offered a loan. ß3(Treatment*Female) gives the difference between the average impact of microcredit access for men and the average impact of microcredit access for women. Collectively, ß2Treatment and ß3(Treatment*Female) would give the average effect of microcredit access on the dependent variable for the women who are offered a loan. However, this is not presented in the tables, as the estimate of interest is ß3(Treatment*Female).
3.4 Balance check
The allocation of participants into the treatment and control group was random and according to Augsburg et al. (2015) there are no statistically significant differences between those groups at the time of the baseline survey. However, the sample could still be unbalanced when looking only to men in the treatment versus control group or when looking only to women. Therefore, balance tests are performed with linear regressions for several variables for both the male and female participants using the post attrition sample. Men and women who participated in the baseline and the end line survey are compared on their household composition, access to credit, borrowed amounts, self-employment activities, consumption and living location. For each of these categories, dependent variables were regressed upon treatment status taking into account if they participated in the follow-up survey and their gender. In this way, the regressions tested whether the men in the control and treatment groups are similar and whether the women in the control and treatment groups are similar. After this, attrition is regressed on treatment status, again taking into account the gender of participants. This provides information on the attrition rate in the control group as compared to the treatment group for both men and women.
3.4.1 Results of men
When looking at the p-values in Table 7, it can be concluded that there are almost no
significant differences between the two groups. However, compared to the men in the control group, significantly more men of the treatment group participated in the end line survey, the difference in response rates is around 10 percentage points.
Households of male respondents consist on average of 3 members of which 2 members are often adults. The respondents have an average age of 36 years and a third of them only went to primary school. There are some participants who have had access to loans before joining this program, but this does not count for all male respondents. In the year before the baseline survey male households earned 19,000 BAM, of which 8,459 BAM came from self-employment activities and 464 BAM from agricultural activities. Typically, 1 member in the household is employed and less than one is unemployed or retired. The respondent works around 57 hours a week, of which 36 hours for a business. As for
Table 7 – Men
Control group Control-Treatment Obs. Obs. Mean SD Coeff. p-value Panel A. Post attrition household
sample Household composition Number of members 587 264 3.436 1.519 0.156 0.209 Number of adults 587 264 2.606 1.034 0.168 0.055 Number of children 587 264 0.758 1.040 0.001 0.991 Respondent age 587 264 36.523 12.285 1.254 0.216
Respondent with at most primary education
587 264 0.307 0.462 -0.010 0.801
Access to credit
Any loan outstanding 587 264 0.549 0.499 -0.038 0.354 Number of outstanding loans 587 264 0.693 0.751 -0.034 0.596 - From an MFI 310 145 0.597 0.495 -0.058 0.306 - From a bank 310 145 0.497 0.502 0.067 0.239 Loans used for business investment
(%)*
306 144 49.538 45.882 -3.124 0.547
Amount borrowed (BAM)
Total amount of three main loans 587 264 3,541 7,280 -325.1 0.613 Average amount borrowed from MFI 583 263 1,260 4,057 -428 0.117 Average amount borrowed from bank 584 264 2,286 6,063 122.1 0.836
Self-employment activities
Number of income sources 587 264 2.527 1.020 -0.022 0.797 Total HH income last year (BAM) 587 264 19,000 17,000 -1,500 0.220 Income from self-employment (BAM) 587 264 8,459 14,000 -1,400 0.170 Income from agriculture (BAM) 587 264 464 1,772 -108.13 0.419 Number of HH members unemployed 587 264 0.674 0.854 0.038 0.608 Number of HH members employed 587 264 1.098 0.930 0.059 0.439 Number of HH members retired 587 264 0.318 0.513 -0.018 0.686 Hours worked by respondent, last week 581 261 57.100 25.354 -2.275 0.274 Hours worked on business by
respondent, last week
545 244 35.680 29.005 -1.488 0.553 Consumption (BAM) Food consumption 587 264 105.2 86.7 1.61 0.82 Nondurables consumption 587 264 206.3 886.4 98.08 0.32 Durables consumption 583 264 2857.1 5781.9 -378.16 0.40 Location (km) Distance to Sarajevo 587 261 141.8 367.8 -3.28 0.91 Distance to nearest city 587 261 63.8 364 -3.65 0.90
Panel B. Attrition
Not surveyed at end line 717 345 0.235 0.424 -0.105 0.000
Notes: Households are the unit of observation. The sample in panel A includes only the households of the men who were also surveyed at end line. The sample in panel B includes all households of the men who were surveyed at baseline. * Total average amount of three main loans outstanding. BAM is the Bosnia and
3.4.2 Results of women
The p-values of Table 8 do show some significant differences between the control and
treatment groups. It appears that women in the control group borrow more than women in the treatment group (1,765 BAM). Therefore, it seems logical that women in the treatment group borrow significantly less at a bank as compared to women in the control group (1635 BAM). As well as with the men, it can be concluded that more women in the control group do no longer participate in the follow-up survey, the difference is around 8 percentage points.
The female household composition is rather similar to those of the male respondents as on average there are 3 members of which 2 adults. The average age for women is 38 years, slightly older as compared to the men. When looking at access to credit, women also have been able to borrow before participating in this program. The year income for the female households prior to the baseline survey was 16,000 BAM. Through self-employment activities 5,977 BAM was earned and another 228 BAM was earned with agricultural
Table 8 – Women
Control group Control-Treatment Obs. Obs. Mean SD Coeff. p-value Panel A. Post attrition household
sample Household composition Number of members 408 180 3.467 1.416 0.209 0.161 Number of adults 408 180 2.450 1.058 0.120 0.274 Number of children 408 180 0.967 0.991 0.104 0.317 Respondent age 407 179 37.944 11.465 1.306 0.263
Respondent with at most primary education
408 180 0.328 0.471 0.089 0.065
Access to credit
Any loan outstanding 408 180 0.633 0.483 0.007 0.884 Number of outstanding loans 408 180 0.961 0.988 -0.031 0.748 - From an MFI 260 114 0.649 0.479 -0.087 0.152 - From a bank 260 114 0.535 0.501 -0.021 0.733 Loans used for business investment
(%)*
258 113 43.842 42.424 1.110 0.835
Amount borrowed (BAM)
Total amount of three main loans 408 180 4,983 10,216 -1,765 0.032 Average amount borrowed from MFI 408 180 1,207 1,864 -126.7 0.495 Average amount borrowed from bank 407 180 3,776 10,153 -1635.5 0.044
Self-employment activities
Number of income sources 408 180 2.561 1.084 0.084 0.432 Total HH income last year (BAM) 408 180 16,000 13,000 695.45 0.586 Income from self-employment (BAM) 408 180 5,977 12,000 526.58 0.616 Income from agriculture (BAM) 408 180 228 978.37 -22.70 0.817 Number of HH members unemployed 408 180 0.700 0.927 -0.003 0.977 Number of HH members employed 408 180 1.094 0.895 0.090 0.327 Number of HH members retired 408 180 0.306 0.519 0.023 0.664 Hours worked by respondent, last week 406 179 37.659 28.146 -0.153 0.956 Hours worked on business by
respondent, last week
317 137 27.511 25.739 4.772 0.098 Consumption (BAM) Food consumption 408 180 107.01 78.42 -2.42 0.77 Nondurables consumption 408 180 224.03 1,055 49.42 0.65 Durables consumption 406 179 1,950 3860 -103.09 0.8 Location (km) Distance to Sarajevo 403 179 116.14 62.74 5.96 0.35 Distance to nearest city 403 179 39.05 24.18 -1.00 0.69
Panel B. Attrition
Not surveyed at end line 479 223 0.193 0.395 -0.083 0.012
Notes: Households are the unit of observation. The sample in panel A includes only the households of the women who were also surveyed at end line. The sample in panel B includes all households of the women who were surveyed at baseline. * Total average amount of three main loans outstanding. BAM is the Bosnia and
3.5 Attrition
Since the balance tests indicated that attrition amongst the control group for both men and women is higher than attrition amongst the treatment group, it seems appropriate to look further into the influence of attrition. This is done in three different ways. First, Appendix C1 presents five linear regression models. To control for non-random attrition, the characteristics of attritors in the treatment group are compared with those of the attritors in the control group. Though the results show that attrition amongst men might be non-random, attrition amongst women appears to be random.
This is followed by a estimation of the impact regressions in Appendix C2. The re-estimation is done with weights accounting for the probability that respondents end up in the follow-up survey. Overall, the signs of these reweighted results remain unchanged. Yet, the size and statistical significance of some outcomes does vary slightly. The most remarkable change is that the difference indicating that the labor hours of the household members of female borrowers increase more due to microcredit access is no longer statistically significant. Other changes are described in detail in Appendix C2.
Thirdly, in Appendix C3 the attritors are compared with the stayers on the variables also used in the balance check. These linear regressions are performed to determine whether the attritors are significantly different from the stayers. This analysis shows that men who dropped out of the sample have a significantly higher income from self-employment
activities. The most important difference for the female participants concerns the outstanding loans. Female attritors had rarely any loans outstanding at the end line survey.
3.6 Estimating causal impact for borrowed amounts
The balance check showed that one category for the female participants might not be balanced, namely the borrowed amounts. Three variables are regressed in this category; the total borrowed amount of three main outstanding loans, the amount borrowed from a MFI and the amount borrowed from a bank. Furthermore, attrition might also have an influence on these variables. Therefore, in order to identify whether the differences in these variables amongst the treatment and control group are caused by the offered loans, another test will be performed for these variables besides the Ordinary Least Squares estimation. To do this, there are various options which are addressed in Appendix D. In short, difference-in-differences is the preferred method for this setting, because it tests the assumption that without the loan the changes would be the same for the treatment and control group. Furthermore, with the short time frame and the method being especially useful in determining whether the treatment and control group differ systematically, it is an appropriate method for this setting (Duflo, 2001).
4. ANALYSIS 4.1 Results
The Ordinary Least Squares estimations are provided for each hypothesis in Tables 9 to 14. The data presented is the same for each table as the number of observations for each
dependent variable are given first, followed by the results for ß0, ß1Female, ß2Treatment and ß3(Treatment*Female). Standard errors are in parentheses. The results of ß4Baseline are not presented, to keep an overview of the results. As all measured effects are intention-to-treat effects, effects might be stronger for those who accepted the loans and weaker for those who did not borrow. However, there were only 8 respondents who did not accept the loan
(Augsburg et al., 2015). Thus, when referring to borrowers in the following sections, men or women in the treatment group are considered. Underneath, the results for each hypothesis are discussed, followed by a comparison with the results of the article by Augsburg et al. (2015). 4.1.1 Outstanding loans
As mentioned before, some of the men and women among both the treatment and control group had loans outstanding before the experiment. Logically, one would expect by providing loans to the treatment group, at the end line participants would have an increased number of loans outstanding. This is also confirmed by previous literature. In Table 9, the results can be found for outstanding loans at the follow-up survey. Male participants who were not offered a loan are, nevertheless, almost 72 percentage points more likely to have an outstanding loan at end line and 38 pp more likely to have this loan outstanding at a MFI. Female participants who were not offered a loan, are only 8 pp more likely to have a loan outstanding after the follow-up survey. The men who were offered a loan, are almost 23 pp more likely to have an outstanding loan at end line and 46 pp more likely to have this loan outstanding at a MFI. Yet, they were almost 6 pp less likely to have this loan outstanding at a bank. When compared with women in the treatment group, the findings show that men are more likely to have a loan outstanding (8 pp) and more likely to borrow at a MFI (5 pp), while women are more likely to borrow at a bank (0.5 pp). However, these effects are not statistically significant. Moreover, this is not in line with the expectation that due to social pressure, women being risk-averse and their inclination towards repayments, women would be able to increase their outstanding loans at the end of the experiment more in comparison with men. Apparently, for this
category women’s investment behavior is not enough to explain the changes.
Table 9 – Outstanding loans
Any loan outstanding
Number of loans outstanding
At least one loan outstanding at a MFI
At least one loan outstanding at a bank Observations 994 994 994 994 ß0 0.717*** (0.156) 0.839** (0.372) 0.381** (0.170) 0.0285 (0.0847) ß1Female 0.0823* (0.0447) 0.262** (0.107) 0.0106 (0.0473) 0.0112 (0.0292) ß2Treatment 0.227*** (0.0340) 0.463*** (0.0762) 0.459*** (0.0370) -0.0576*** (0.0208) ß3(Treatment *Female) -0.0812 (0.0520) -0.0838 (0.137) -0.0485 (0.0594) 0.00491 (0.0333)
Notes: Unit of observation is the respondent. Standard errors are robust to heteroscedasticity. BAM is the Bosnia and Herzegovina convertible mark. The exchange rate at baseline was US$1 to BAM 1.634 (Augsburg et al., 2015).
***Significant at the 1 percent level ** Significant at the 5 percent level * Significant at the 10 percent level Source: End line household survey (Augsburg et al., 2015).
4.1.2 Microbusiness activities
The impact of credit access on microbusiness activities can be found in Table 10.
Except for the probability to have to close a business in the last 14 months, all variables are expected to be more positively affected for men as for women. Significant changes are found for the men and women in the control group concerning their asset value. Whereas the asset value of men tends to increase, the asset value of women tends to increase less in comparison. The asset value of men who received a loan appear to increase less, while a larger positive effect on asset value is found for women who received a loan. Yet, this difference of 2,583 BAM is not significant. A slightly larger but again insignificant effect of 2 percentage points is found for the probability of inventory ownership for female borrowers. This could be explained by the neoclassical assumption on returns to capital; that women are indeed able to increase their earnings more, because of their smaller access to capital. Consequently, they would invest this earned capital in their business. For asset values this assumption is
Due to the insignificance for all the differences in effects of microcredit access between men and women and several variables showing higher increases for women, the second hypothesis is not confirmed. Thus, it cannot be said that microbusiness activities are more positively affected for men in Bosnia and Herzegovina when compared to women.
Table 10 – Microbusiness activities
Main business of respondent Asset value (BAM) Inventory ownership Revenue (BAM) Expenses (BAM) Profit (BAM) Any self-employment income (HH) Observations 994 985 976 967 994 994 ß0 55,050* (32,744) -0.0810 (0.118) 2,680 (5,559) 1,571 (2,973) 2,222 (3,131) 0.459** (0.179) ß1Female -28,659*** (8,087) -0.0376 (0.0297) -2,121 (1,428) -761,3 (798.7) -1,662* (859.7) -0.0797* (0.0467) ß2Treatment -1,479 (7,622) 0.0430 (0.0268) 1,859 (1,560) 933.1 (979.2) 771.0 (824.7) 0.0464 (0.0374) ß3(Treatment *Female) 2,583 (10,438) 0.0205 (0.0415) -1,165 (1,933) -812.3 (1,159) -242.6 (1,071) 0.0338 (0.0601) Business ownership Services business Agriculture business Started business Closed business Observations 994 994 994 994 994 ß0 0.101 (0.184) 0.0246 (0.144) 0.173 (0.154) 0.207* (0.123) 0.459*** (0.169) ß1Female -0.191*** (0.0485) -0.151*** (0.0358) -0.0299 (0.0421) 0.0379 (0.0341) 0.0682 (0.0429) ß2Treatment 0.0415 (0.0406) 0.0287 (0.0362) 0.0237 (0.0361) 0.0296 (0.0277) -0.0155 (0.0336) ß3(Treatment *Female) 0.0413 (0.0634) 0.00599 (0.0465) 0.0277 (0.0558) -0.0210 (0.0457) -0.00328 (0.0544)
Notes: Unit of observation is the respondent. Standard errors are robust to heteroscedasticity. BAM is the Bosnia and Herzegovina convertible mark. The exchange rate at baseline was US$1 to BAM 1.634 (Augsburg et al., 2015). Businesses started or closed concerns the last 14 months. The results of the dependent variables business ownership, services business, agriculture business, started business and closed business are provided under the other variables due to space limitations.
***Significant at the 1 percent level ** Significant at the 5 percent level * Significant at the 10 percent level Source: End line household survey (Augsburg et al., 2015).
4.1.3 Household income
through wages of -6.5 pp is statistically significant. This indicates a shift towards wage work instead of increased self-employment activities.
Though male borrowers were expected to earn more income from both
self-employment activities and wages, the results show a larger increase in the self-self-employment amount of around 725 BAM for women as compared to men. Yet, the result is not significant. On the other hand, male borrowers seem to increase their wages more, with a difference of 1,330 BAM, even though this result is insignificant. The possibility to earn an income through remittances and government benefits is lower for male borrowers, respectively, 171 BAM and 176 BAM. This was predicted beforehand.
Since only wages are for men more positively affected than for women, which is an insignificant effect, the third hypothesis is not confirmed by the results. Therefore, it cannot be said that income is more positively affected by microcredit access for men in Bosnia and Herzegovina when compared to women.
Notes: Unit of observation is the respondent. Standard errors are robust to heteroscedasticity. BAM is the Bosnia and Herzegovina convertible mark. The exchange rate at baseline was US$1 to BAM 1.634 (Augsburg et al., 2015).
***Significant at the 1 percent level ** Significant at the 5 percent level * Significant at the 10 percent level Source: End line household survey (Augsburg et al., 2015).
4.1.4 Consumption and savings
Beforehand one would expect an increased consumption because of access to microcredit. Either, this might be caused by an increased income due to borrowers being able to successfully invest the loan or borrowers decide not to invest the loan into the business completely but spend parts of it on durables, education, or start to save, for instance. While consumption on certain products or services might increase, the consumption on others might decrease. This could be explained by the varying preferences of men and women. Table 12 summarizes the impact of microcredit access on consumption and savings. Various significant results are found for women who are not offered a loan. They show decreases in all
categories, except the HDGI, with varying significance levels. These decreases can be explained by the reversed assumption that when not having the opportunity to borrow,
consumption levels might decline. The small resources women do have, they tend to spend on goods in the HDGI, which shows an increase of 11 pp. On the other hand, men in the control
Table 11 – Income
Self-employment Wages Remittances Government benefits
