Microfinance and poverty alleviation:
randomized controlled trial design
Ona Kotryna Kajokaite
April 1, 2014
Abstract
This paper proposes a new randomized controlled trial design of evaluation of the effect of introducing the standard microcredit group-lending product in a new market in Allahabad, Uttar Pradesh in Northern India, which has one of the lowest concentrations of mi-crofinance institution ratios in the country. We depart from the ex-isting literature by placing more emphasis on eliminating the issues associated with self-selection bias, duration of the experiment, pseudo-random methods, ethical problems and other threats to internal va-lidity. The resulting theoretical model can be used to evaluate the impact of microfinance on income and literacy levels.
1
Introduction
Microfinance has existed in various forms for centuries, and even longer in Asia where informal borrowing and lending stretch back for several thou-sand years. However, the modern concept of microfinance emerged in the mid 1970s in rural Bangladesh, when Dr. Muhammad Yunus, professor of economics at the University of Chittagong, discovered that very small loans could make a disproportionate difference to poor people [34]. Dr. Yunus saw that without a well-functioning financial system it is impossible to cre-ate sustainable economic growth and independence, which is an imperative goal as many developing countries rely on financial aid from the established economies. His work on provision to effectively serve the poor segment of the population, which was largely ignored by the formal financial systems, gave birth to the first modern microfinance institution (MFI) Grameen Bank, which literally translates as “village bank”. Since Dr. Yunus handed out the first 27 dollars to a group of female villagers in 1976, the Grameen Bank has
expanded to serve over 7 million people in 75,950 villages, distributing a to-tal amount of $6.07 billion. Across the world, nearly 140 million households benefit from the services of 10,000 MFIs [5]. Such expansion demands precise evaluation, which is challenged by both magnitudal and geographical issues of many developing regions. In this paper we design a randomized experi-ment in order to evaluate the effect of introducing the standard microcredit group- lending product in a new market in India.
In India, as well as in other developing economies that lack a well-established financial structure, traditional banks are usually reluctant to pro-vide financial services such as credit to individuals with little to no cash in-come or assets. Microfinance approaches these problems through innovations in lending contracts as the MFIs grant small loans to the impoverished with-out requiring collateral. Some borrowers are required to keep small amounts of savings with the MFIs that are inaccessible during the loan period. These savings act as a collateral. Also, close monitoring of the borrower’s behavior and frequent repayment schedules are vital. The central activity of MFIs is visiting borrowers and collecting the installments at their homes. Borrow-ers are never required to travel anywhere since for most of them this would mean missing a day’s work and thus a day’s salary, or arranging an expensive travel. The costs of these practical inconveniences are borne by the borrowers as they pay interest rates much higher than those of traditional banks but always cheaper than those of village lenders [13]. Also, the initial emphasis on credit has now grown to include savings, insurance, pensions and other financial services.
We choose to examine India as it has the most developed microfinance system in the world [15]. There are two prominent models of microfinance in India. The first one is the Self Help Groups (SHGs), which usually consist of up to 20 people, mainly women. Each member saves a small sum of money regularly and the groups use pooled savings to provide loans to individual members [13]. Another more dominant model of microfinance in India is intermediation by the MFIs. The early institutions were implemented by non-governmental organizations (NGOs) and based on international donors and government subsidies; however, in recent years involvement of private and commercial banks has significantly increased as microfinance turned out to be a demonstrably profitable private sector approach to poverty alleviation [29]. The possibility of a ‘win-win’ poverty reduction with profits draws the attention of policy makers, investors and academics [1]. MFIs now borrow at commercial rates from public and private banks and lend on to SHGs or joint liability groups (JLGs) [6].
In addition to fighting poverty, another prominent goal of microfinance is the empowerment of women. For a long time Indian females were excluded
from the socioeconomic development of the country, especially those living in households with little to no assets [20]. Microcredit programs empower women by giving them tools to establish or strengthen their financial inde-pendence, which leads to more productive communities [7]. The majority of MFI borrowers are women and most of them have excellent repayment records [9]. Empirical evidence also shows that women are less likely to default on their loans than men [3].
One of the most distinguished issues for MFIs is the provision of loans at an affordable cost. The global average interest rate on microloans is estimated at 37 per cent with rates reaching as high as 70 per cent in some markets [28]. High interest rates are caused by high transaction costs of microfinance operations relative to the loan size. The rates on microloans are much higher than those of traditional banks because it inevitably costs more to lend and collect a given amount through thousands of tiny loans than to lend and collect the same amount in a few large loans. Higher administrative costs have to be covered by higher interest rates.
However, there are concerns that poor people are being exploited by exces-sive interest rates, given that the impoverished have little bargaining power and that an even larger proportion of microfinance is moving into for-profit organizations where higher interest rates mean higher returns for the share-holders [28]. The microcredit industry is now being accused of pushing its clients into debt traps [5] as it charges interest rates almost as high as those of the village lenders it meant to replace [26].
What is interesting about this discussion is the scarcity of reliable infor-mation. To this day, quantitative estimates from randomized experiments on microfinance and poverty reduction are limited. The many stories about prosperous entrepreneurs or deeply indebted borrowers do not provide signif-icant evidence on the actual impact of microfinance on the average borrower. The paucity of competent effect evaluations can partly be associated with the high cost of such studies relative to the working budgets of MFIs [14]. Since virtually all studies that conclude a positive effect of microfinance have been conducted by MFI supporters or MFIs themselves, most impact evaluations feature a selection bias [14]. Also, most MFIs purposely choose certain vil-lages over others. This overrules comparisons over time between clients and clients. Even representative data about microfinance clients and non-clients cannot distinguish the causal effect of microfinance access from the selection effect because clients are self-selected and therefore not comparable to non-clients [5].
In the mid 2000s a few independent economists made cases for random-ized controlled trial (RCT) methodologies. While selection bias plagues most of social science research, the key distinguishing feature of RCTs is that they
randomly assign individuals to control and treatment groups. The random assignment ensures that potential outcomes are not contaminated by selec-tion bias where the clients of MFIs might be systematically different from non-clients [14]. In an ideal experiment, randomization ensures that the only difference between the treatment and control groups is the access to micro-credit. A control group is created to isolate the effect of access to credit through MFIs. Obviously, RCTs are hard to implement in places where mi-crofinance is accessible to virtually everyone and not useful for assessing the programs already on the ground.
These problems make evaluation of MFIs particularly difficult, and so far there is no unanimity among economists on their impact. For example, Bateman [6] expresses negative views on microfinance and criticizes the MFI studies for ignoring the issues of displacement and client failure. In 2011, the systematic review conducted by the UK government-funded Duvendack et al. found no identifiable impact in the last 30 years of microfinance movement [6]. Duvendack argues that support for microfinance has been mostly based on politics and not economics.
A few randomized experiments relevant to our research were launched in 2010s and have been delivering their results for a few years already, so we can compare different methodology designs and the results they deliver.
Banerjee and Duflo’s [5] evidence from the randomized experiment con-ducted in Hyderabad, India, shows no significant impact of access to micro-finance on poverty alleviation. However, the particular choice of the deter-minants of access to microfinance could have led to a bias in the estimation of the effect on poverty primarily due to endogeneity issues, errors in mea-surement and other threats to the internal validity of their estimation. First of all, only one MFI was restricted from entering certain villages over the time span of 15 to 18 months. The experiment was conducted in the area of Hyderabad in South India where the distribution of MFIs is the most dense in the country and thus microfinance is accessible to virtually everyone. For this reason, it would have been easier to limit the access to microcredit if the experiment had been conducted in North India, as the Northern and North-eastern regions have the lowest share of MFIs in the country. In 2008, 51.33 per cent of MFIs were located in the South compared to only 3.81 per cent in the North [30]. However, there is a question whether it would be right to extrapolate local impact results in a remote area of India to the national context [32].
Cr´epon et al. [11] investigate the effect of access to microfinance in rural Morocco over the course of 24 months, in areas that had no previous access to microcredit. They find that access to microcredit expands the scale of self-employment activities; however, they do not find any effect on consumption,
health or education.
Angelucci et al.’s [2] evidence from the randomized experiment conducted in north-central Sonora, Mexico, finds that access to microcredit has little to no impact on most of the variables examined. However, just as with Banerjee and Duflo’s experiment, only one microlender Compartamos was restricted from entering “clusters” (neighborhoods in urban areas, towns in rural areas) in the control areas, so people could still borrow from other sources (MFIs, banks or microlenders).
One key result of the model of the studies above is that poor people often choose to sacrifice short-term or even medium-term consumption when they gain access to microfinance and can invest in a durable good. Non-durable consumption may thus fall initially, which accentuates the importance to fol-low the households for a longer period of time. But there are some ethical issues to consider, as we would be restricting individual’s access to microfi-nance for a longer period of time so the time horizons naturally are short [32]. However, in our experiment, we ignore the effect on consumption as it is not a reliable indicator of poverty in the short-run.
In this model microfinance plays a role in helping households make dif-ferent intertemporal choices in consumption but its overall effect on con-sumption, income, education and health remains ambiguous. Banerjee [5] and Cr´epon [11] acknowledge the limitations that these issues impose. The experimental designs of the three studies above are comparable to our pro-posed experiment’s design. All of the above examine group lending, all in developing countries, and all adopt similar time spans.
The following section discusses the proposed research design, describes the sample and provides some summary measures; Section 3 investigates the effect of access to microcredit on income and literacy levels of households. The last section begins with a demonstration of how randomization elim-inates the selection bias, proceeds with the presentation of the employed methodology and the discussion of external validity of our model.
2
Experimental design
The research question that we intend to test is the level of impact that the access to microloans has on poverty alleviation, measured by income, levels of consumption, literacy and women empowerment.
2.1
The Product
SKS Microfinance is the most important microfinance entity active in India in terms of size and access to capital. As of January 2014, it has 7.3 million borrowers, 1,081 branches and an outstanding loan portfolio of $939 million. Ever since SKS transformed into a for-profit nonbank finance company in 2005 in order to tap equity, it has established a reputation for raising capital and debt to keep pace with its ambitious growth plans [8].
SKS uses the Joint Liability group lending model whereby borrowers guar-antee each other’s loans. A group is comprised of five women, and 3 to 10 groups or 15 to 50 women compose a ‘centre’. Weekly centre meetings serve as time to conduct financial transactions. Group loans begin at Rs. 2,000 and after the first loan repayment range to Rs. 12,000 (about $44-$260). The term of the loan is 50 weeks with principal and interest payments due on weekly basis. The flat interest rate is 12.5 per cent or a 24.55 annual percentage rate (APR).
2.2
Demographics
The district of Allahabad has a population of around 6 million people. It is the 13th largest district in India and the most populous district in the state of Uttar Pradesh, which contributes to over 16% of India’s population and has a high population growth rate at about 20% (Uttar Pradesh Census 2011). The overall literacy rate of the district stands at 74.41%, close to the national level of 74%. The literacy rate among men is 79%, while among women it is only 59%. The Planning Commission in 2005 revealed that Uttar Pradesh had 59 million people living below the median poverty line of $2 a day 2005 PPP (Purchasing Power Parity), more than in any other state in India. Agriculture remains the most prevalent industry in the region.
2.3
Experimental design
When choosing the target district, we have to recognize that the geographical distribution of microfinance in India continues to be skewed. Increasing the access to microcredit in the North and North-eastern parts of India is not a challenge that can be met overnight [22]. We choose to examine Allahabad in the state of Uttar Pradesh as the region has less than 5 per cent concentration of MFIs compared to more than 25 per cent concentration in the southern states, e.g. Andhra Pradesh [15], which makes it easier to isolate the treat-ment effect as households in the treattreat-ment group do not have an option to borrow from alternative lenders. A proposal to increase the number of MFIs
in the lesser-served states of India was made by the Tenth Five-Year Plan Working Group on Poverty Alleviation Programmes (Planning Commission 2002). It recommended that the well-established MFIs be asked to set up branches in selected poor districts in the northern states of India. SKS itself has also expressed its plans to expand in the northern parts of the country, especially Uttar Pradesh [33]. SKS claims that there is no need to adopt a different strategy in the northern region.
It is proposed that SKS selects 100 identifiable neighborhoods in Alla-habad, Uttar Pradesh as places in which they are interested opening branches. The population in the neighborhoods proposed for the study ranges from about 50 to 500 households [18]. These neighborhoods are selected based on having no preexisting microfinance institutions in the area. Neighborhoods with high fractions of construction workers ought to be ignored as they have a tendency to move regularly, which makes them undesirable as borrowers [5]. We also exclude the few villages where other MFIs are present. Only families that have been living in the area for over a year are chosen to participate in the trial as MFIs believe that dynamic stimulus, or the promise of credit in the future, are more influential to the permanent dwellers in motivating the repayment.
In each area, we conduct a brief baseline survey in order to attain the most important information on the baseline conditions of the households in the area. We choose to randomize at a household level as it would be very difficult and much more costly to obtain data on each individual. The survey is conducted by a local independent survey firm and includes information on household composition, literacy, employment, asset ownership, expenditure, borrowing, saving, and any business currently operated by the household. After the baseline survey but prior to randomization we eliminate the areas which contain large numbers of migrant-workers [5].
In order to avoid the self-selection bias, we conduct a lottery where sim-ilar households of each village are invited to participate. We choose the selected households by the similarity of neighborhoods, based on average per capita consumption and per-household debt [5]. In the lottery there are equal numbers unlucky and lucky tickets that determine whether the participants obtain a loan. As participation is voluntary, only the households that are in-terested in obtaining a loan are take part in the lottery. As there is an equal number of lucky and unlucky tickets, we obtain equally sized treatment and control groups. The participating households are not allowed to transfer or sell their lucky tickets, which is prevented by registering the outcome of the lottery of each household. The lottery approach allows us to measure the average treatment effect on those households that are interested in obtaining a microloan and not the public as a whole.
SKS then begins to operate in the selected neighborhoods for 24 months. In the intervening period, other MFIs are free to start their operations in all areas. However, it is not very likely as the MFI penetration in the district is very low. After 24 months we conduct an end-line survey in order to attain the information on the endline conditions of the area and to evaluate the changes of the variables that were examined during the baseline survey.
2.4
Pilot
Prior to the full-scale experiment, we conduct a small scale preliminary ex-periment in order to evaluate feasibility, time, cost and adverse effects in an attempt to improve upon the study design. We conduct the pilot study in 10 neighborhoods (approximately 500-1000 households) of the relevant popula-tion in Allahabad district, however, not in the villages that will form part of the final sample. The time span of the pilot experiment is 6 months between the baseline and endline surveys, during which SKS operates as described in the section above.
2.5
Threats to internal validity
Two main concerns about the use of data should be discussed at this point. First, since the analysis uses numerous spans of the data to compute averages, attrition problems may be of significant importance. However, based on previous experiments attrition in our proposed study is expected to be rather small, so the outcomes remain unbiased.
The second concern is spillover effects and Stable- Unit- Treatment- Value-Assumption (SUTVA). SUTVA assumes that the treatment of one individual does not have effect on the others. Imbens and Wooldridge [17] present ex-amples from agricultural and epidemiological interventions that explain that interaction between individuals and spillover effects cannot be prevented al-together. Treated individuals can and most often do have an impact on the outcomes of the non-treated individuals [17]. We are unable to conclusively estimate the extent of externalities of borrowing on non-borrowers and ac-count for spillovers in our proposed study. However, Grameen Bank claims that the poverty level of non-borrowers decreases by 0.3 percentage point every year as a result of positive spillovers from the borrowers.
3
Methodology
We use the econometrics of Stock and Watson [31] to show that randomiza-tion eliminates the selecrandomiza-tion bias. We denote the outcome of interest (income level and level of literacy) Yi∗, while we choose our treatment indicator to be access to microcredit loani = 0, 1. The goal of our research is to determine
whether Yi∗ is affected by access to microcredit loani. For each household i
there are two potential income variables: Y1i∗ if loani = 1 and Y0i∗ if loani = 0.
Y0i∗ is the income level of a household which draws an unlucky ticket and does not obtain a microloan, while Y1i∗ is the income level of a household which draws a lucky ticket and gains access to microcredit. Hence, the causal effect of access to microfinance is ∆i = Y1i∗− Y
∗
0i. As the household is either treated
or not, we can observe only one of the two potential outcomes. Therefore, the observed outcome Yi∗ can be written in terms of potential income variables as:
Yi∗ = Y0i∗ + (Y1i∗ − Y0i∗) ∗ loani
As some households receive the treatment and some do not, the expected difference in observed outcomes between the two groups is E(Yi∗|loani =
1) − E(Yi∗|loani = 0) = E[Y1i∗|loani = 1] − E(Y0i∗|loani = 0]. This
equa-tion is true regardless of how treatment is determined and states that the expected difference between the treatment and control groups is the mean treatment outcome for the treated, minus the mean no-treatment outcome for the untreated:
AT E = E(∆i) = E(Y1∗− Y ∗ 0) = E(Y ∗ 1) − E(Y ∗ 0)
As loani is the treatment indicator, we have the average treatment effect
treated (ATET):
AT ET = E(∆i|loan = 1) = E(Y1∗− Y ∗ 0) = E(Y ∗ 1) − E(Y ∗ 0)
= E(Y1∗|loan = 1) − E(Y∗|loan = 1)
Households with a positive ∆i are more likely to choose treatment. If there
is self-selection,
E(Y1∗) 6= E(Y1∗|loan = 1) E(Y0∗) 6= E(Y0∗|loan = 0)
The households that are interested in receiving a loan draw lottery tickets and in this way are randomly assigned to the treatment and control groups, so loani is distributed independently of personal characteristics and is
the treatment and control groups is
E(Yi∗|loani = 1) − E(Yi∗|loani = 0) = E[Y1i∗|loani = 1] − E[Y0i∗|loani = 0]
= E[Y1i∗ − Y0i∗] where E(Yi∗|loani = 1) − E(Yi∗|loani = 0) is the observed difference in
average income between the treatment and control groups and E(Y1i∗|loani =
1) − E(Y0i∗|loani = 1) denotes the average treatment effect on the treated.
The second equality uses the fact that [Y1i∗, Y0i∗] are independent of loani, so
the mean difference in the experimental outcomes between the two groups is the average treatment effect (ATE) in the population from which the sample was drawn. Thus,
AT E = AT ET
Now we can use ordinary least squares (OLS) to estimate the above regression with some additional control variables W :
Yi = β0+ β1loani+ β2W1i+ . . . + β1+ Wri+ ui
3.1
Access to microcredit and income
We examine the effect of access to microcredit on income and investigate it by estimating the model below by OLS:
ln(income)i = β0+ β1loani+ βmXi+ ζi,
where ln(income)idenotes the natural logarithm of average yearly household
earnings during the period of 24 months. loani is the dummy variable defined
above, β1 is the intent to treat effect and Xi is a vector of control variables,
calculated as area- level baseline values. Since we would like our results to be comparable to the existing literature, the choice of variables is principally based on the specifications of Banerjee et al. [5] and Cr´epon et al. [11]. More specifically, we measure income levels by looking at a number of different variables including area population, total businesses, average expenditure per capita, fraction of household heads who are literate, and fraction of all adults who are literate and average weekly hours worked.
3.2
Access to microcredit and literacy
We examine the effect of access to microcredit on literacy and investigate it by estimating the model below by OLS:
where liti represents the fraction of literate households observed in our
sam-ple and loani is a dummy variable approximating the household’s access to
microfinance. The vector Xi is as defined in section 3.1, vector Yi includes
variables that control for household specific characteristics such as education and health.
4
External validity and multiple inference
The issues of external validity and multiple inference naturally arise in such experimental evaluations, all of which are conducted in very different loca-tions and have the potential to affect many different outcomes. The external validity issue questions whether the results from the experiment in one of the poorest parts of India represent the reality of the whole country. Can they be generalized to the national level? Another issue that naturally arises is the problem of multiple inference. As we are testing many different outcomes, there is always the possibility that the results obtained are the statistical artifacts of multiple hypothesis testing [5].
Cr´epon et al. [11] evaluate Al Amana’s group-lending product program in rural Morocco, a context where there was no preexisting access to micro-credit, even the village lenders, before the introduction of the program and no other MFIs or borrowers have entered the area. Our experiment design is comparable to this study as our proposed area is in the north-eastern part of India, where there is very little to no formal borrowing (there are, though, village loan sharks). Thirteen per cent of the households in treatment villages took a loan, and none in the control villages. In the borrowing households, access to microcredit led to an increase in investment in assets used for self-employment activities, and an increase in profit. However, this increase was offset by a decrease in income from casual labor, so in general there was no rise measured in income or consumption [11].
Angelucci et al. [2] evaluate the largest micro lender Compartamos’ pro-gram in Mexico, which was implemented in both urban and rural areas in the state of Sonora and measured after 18-34 months post-expansion. They find some positive impact and little harm.
The papers’ designs and main results are surprisingly consistent for the most important economic outcomes. Our experiment design is similar to those above, however, our study addresses the self-selection problem, so it is feasible that the outcomes are going to have more statistical significance than those of the above papers.
5
Conclusion
This study, if conducted, would be the longest running randomized control trial of the standard group-lending loan product that made microfinance known in the 1970s. If the outcomes are consistent with the previous experi-mental studies, it is expected to yield results that would prompt to reevaluate the efficiency and importance of microfinance [5].
Several recent randomized evaluations have found that while access to mi-crocredit does lead to a rise in investment in assets used for self-employment activities, it has no significant impact on poverty reduction, measured in overall income and consumption ( [5]; [11]; [2]). One of the possible expla-nations is that the businesses that microborrowers usually choose to invest in yield a low marginal product of capital. Recent randomized evaluations have also found no significant impact on business profits (except Angelucci et al., who found a small positive effect).
As the official goal of microfinance is to help alleviate poverty by encour-aging poor people to invest in microbusinesses, evaluations of MFI programs usually focus on business investments and overall consumption per capita as key measures of success. However, as households borrow from MFIs, we may see them consume less of non-durable goods in the short-term. Hence, if we are mainly interested in how microcredit affects consumption, we need to pay attention to its composition [11]. Future extensions on our design may wish to account for such decompositional factors of consumption. Finally, a relatively long time horizon may be needed to estimate the full effects on poverty and development.
References
[1] Aghion, D., Armendariz, B., Morduch, J.: The economics of microfi-nance, MIT press (2007)
[2] Angelucci, M., Karlan, D., Zinman, J.: Win some lose some? Ev-idence from a randomized microcredit program placement experiment by Compartamos Banco, No. w19119. National Bureau of Economic Research (2013)
[3] . Anthony, D, Horne, C.: Gender and cooperation: explaining loan repay-ment in micro-credit groups., Social Psychology Quarterly (2003): 293-302.
[4] Banerjee, A., Duflo, E.: Poor economics: A radical rethinking of the way to fight global poverty, Public Affairs Store (2011)
[5] Banerjee, A., Duflo, E.: The miracle of microfinance? Evidence from a randomized evaluation., (2013)
[6] Bateman, M.: Why doesn’t microfinance work?, London: Zed Books (2010)
[7] Chattopadhyay, R., Duflo, E.: Women as policy makers: Evidence from a randomized policy experiment in India., Econometrica 72.5 (2004): 1409-1443.
[8] Chen, G. et al: Indian Microfinance Goes Public: The SKS Initial Public Offering., Focus Note 65 (2010)
[9] Cheston, S., Kuhn, L.: Empowering women through microfinance, Draft, Opportunity International (2002)
[10] Collins, D., Morduch, J., Ruthven, O.: Portfolios of the Poor: How the World’s Poor Live on 2 a Day Princeton University Press (2009)
[11] Cr´epon, B. et al.: Impact of microcredit in rural areas of Morocco: Ev-idence from a Randomized Evaluation., Massachusetts Institute of Tech-nology (2011)
[12] Dehejia, R., Montgomery, H., Morduch, J.: Do interest rates matter? Credit demand in the Dhaka slums., Journal of Development Economics 97.2 (2012): 437-449
[13] Duflo, E., Tripathi, R., Walton, M.: Credit Market Failures and Micro-finance: From Promise to Practice-A Case Study of the Andhra Pradesh crisis., (2007)
[14] Duvendack, M. et al.: What is the evidence of the impact of microfi-nance on the well-being of poor people?, (2011)
[15] Ghate, P.: Microfinance in India: A State of the Sector Report,2007, SAGE Publications Ltd (2008)
[16] Holvoet, N.: The Differential Impact on Gender Relations of ’Transfor-matory’ and ’Instrumentalist’ Women’s Group Intermediation in Microfi-nance Schemes: A Case Study for Rural South India., Journal of Interna-tional Women’s Studies 7.4 (2013): 36-50
[17] Imbens, G., Wooldridge, J.: Recent Developments in the Econometrics of Program Evaluation., The Institute for Fiscal Studies, Department of Economics, University College London, Cemmap Working Paper No.CWP 24/08
[18] James, K.S.: Indias demographic change: opportunities and challenges., Science 333.6042 (2011): 576-580
[19] Kaboski, J., Townsend, R.: The Impact of Credit on Village Economies, working paper (2009)
[20] Karim, L.: Microfinance and its discontents: Women in debt in Bangladesh., U of Minnesota Press (2011)
[21] Karlan, D., Zinman, J.: Expanding microenterprise credit access: Using randomized supply decisions to estimate the impacts in Manila., No. 976. Center Discussion Paper, Economic Growth Center (2009)
[22] . Kendall, J.: Designing a research project: randomized controlled trials and their principles., Emergency medicine journal: EMJ 20.2 (2003): 164
[23] Kuma, D.: The effects of the microfinance programme of initiative development-Ghana on women clients in Nima, a suburb of Accra., (2012)
[24] Morduch, J.: Does microfinance really help the poor? Evidence from flagship programs in Bangladesh, Hoover Institution, Stanford University working paper (1998)
[25] Morduch, J.: The role of subsidies in microfinance: evidence from the Grameen Bank., Journal of development economics 60.1 (1999): 229-248
[26] Polgreen, L., Bajaj, V.: India microcredit faces collapse from defaults., New York Times 18 (2010): A5.
[27] Roodman, D., Morduch, J.: The impact of microcredit on the poor in Bangladesh: Revisiting the evidence., NYU Wagner research paper no. 2010-09, (2010)
[28] Rosenberg, R., Gonzalez, A., Narain, S.: The new moneylenders: are the poor being exploited by high microcredit interest rates?, Vol. 92. Emerald Group Publishing Limited, (2010)
[29] Sinclair, H.: Confessions of a Microcredit Heretic: How Microlending Lost Its Way and Betrayed the Poor., Berrett- Koehler Store, (2012)
[30] Singh, N. T. : Micro Finance practices in India: an overview., Interna-tional Review of Business Research Papers 5.5 (2009): 131-146
[31] Stock, J.H., Watson, M.W.: Introduction to econometrics., Vol. 104. Boston: Addison Wesley (2003)
[32] Odell, K.: Measuring the impact of microfinance., Grameen Foundation, Washington (2010): 1-38
[33] Taylor, M.: Freedom from poverty is not for free?: rural development and the microfinance crisis in Andhra Pradesh, India., Journal of Agrarian Change11.4 (2011): 484-504
[34] Yunus, M.: Creating a World Without Poverty: Social Business and the Future of Capitalism., PublicAffairs, New York, (2007)
