微額信貸在減少貧窮與鼓勵教育之顯著性 - 政大學術集成

全文

(1)國立政治大學應用經濟與社會發展英語碩士學位學程 International Master’s Program of Applied Economics and Social Development College of Social Sciences National Chengchi University. 治政碩士論文大. 立Master’s Thesis. ‧. ‧ 國. 學 y. sit. io. er. Nat. 微額信貸在減少貧窮與鼓勵教育之顯著性 n. Microfinance: aThe in v l C Significance ofn iMicrocredit Alleviating Povertyh eand Education n gSupporting chi U. Student: 貝君傑 Carlos Benavides Advisor: 林左裕 Tsoyu Calvin Lin. 中華民國 106 年 6 月 June 2017.

(2) Abstract The following research uses a panel linear model regression to detect possible effects Microfinance and its iterations (namely microcredit, microsavings, microinsurance, and remittance services offered by microfinancial institutions) have on distinct social outcomes such as poverty and secondary school enrollment rates. The study sample consists of 20 countries in the Latin American and Caribbean region observed during a period of 4 years (2011-2014). After subdividing the sample into comparable groups the results yielded statistically significant negative effects on poverty headcount, and statistically significant positive effects on secondary school enrollment rates across the subgroups. The results from other social. 治政 capita were insignificant. Due to the increasing presence of大 Microfinancial institutions in Latin 立is sufficient reason to encourage further research to be done America and the Caribbean, there. welfare dimensions such as health, business creation, household consumption, and income per. ‧ 國. 學. in which more experienced researchers can use more statistically complex models (such as IV, RCTs, quasi-experimental surveys) to try to determine whether or not there is a causal. ‧. relationship between microfinance and the effects herein described. For policymakers and funders of NMPs, the takeaway is that they should encourage both advocates and critics to. Nat. sit. y. present transparent and replicable studies to back their claims.. n. al. er. io. 關鍵詞 : 微型金融， NMP ，微型信貸，微型貸款，微型儲蓄，微型保. i n U. v. Keywords: Microfinance, NMP, Microcredit, Microloans, Microsavings, Microinsurance. Ch. engchi.

(3) Table of Contents Abstract ..................................................................................................................................................... 1 Introduction ......................................................................................................................................... 1 1.1 Microfinance ................................................................................................................................. 3 1.2 Problem ......................................................................................................................................... 5 1.3 Purpose ......................................................................................................................................... 6 2 Literature Review ................................................................................................................................. 7 2.1 Theoretical and Empirical Review ................................................................................................. 7 2.1.1 Evidence in Favor ................................................................................................................... 7. 政治大 3 Empirical Process ............................................................................................................................... 14 立 3.1 Panel Data ................................................................................................................................... 15 2.1.2 Evidence Against .................................................................................................................. 11. ‧ 國. 學. 3.2 Fixed Effects vs. Random Effects................................................................................................. 16 3.3 Random Effects vs. Pooled OLS................................................................................................... 17. ‧. 3.4 Data Collection ............................................................................................................................ 17 4 Empirical Model ................................................................................................................................. 19. sit. y. Nat. 4.1 Model .......................................................................................................................................... 19 4.1.1 Poverty Model......................................................................................................................19. io. er. 4.1.2 Education Models ................................................................................................................ 20. al. n. v i n Ch 4.3 Microfinancial Variable Selection ...............................................................................................22 engchi U 4.4 Correlation .................................................................................................................................. 25 4.2 Handling of Missing Data ............................................................................................................ 21. 5 Empirical Results ................................................................................................................................ 26 5.1 Validity Testing ............................................................................................................................ 28 5.1.1 Controlling for Heteroscedasticity ....................................................................................... 29 6 Conclusion .......................................................................................................................................... 30 6.1 Implications ................................................................................................................................. 30 6.2 Limitations................................................................................................................................... 31 References ............................................................................................................................................ 32. I.

(4) 1 Introduction Inequality across regions has been a main concern of global institutions such as the United Nations for a long time. The combined efforts of the world’s leading development institutions and the majority of the world’s countries have resulted in the establishment of the millennium development goals. These range from the halving of extreme poverty to halting the spread of diseases such as HIV/AIDS and providing universal primary education. Since the establishment of these goals, there have been a number of both national and international efforts working towards the fulfillment of them – but critics still maintain that so far there has been little significant progress towards accomplishing the more urgent matters. The Grameen bank serves as prime example of a national effort that has since its inception become a strong. 治政 bank founded in 1983 by Muhammad Yunus set eliminating 大poverty in rural Bangladesh as its 立 main goal (Grameen Bank, 2015).. movement motivating several other individuals and groups to follow in its steps. The Grameen. ‧ 國. 學. The Grameen bank was the modern incarnation of a microcredit lending institution, removing any kind of required collateral or credit history in order to take out a loan. Instead,. ‧. the system is based on mutual trust, accountability, participation and creativity (Grameen Bank,. y. Nat. 2015). This new type of banking system defied conventional banks and their current selection. sit. policies for handing out loans – and specifically targeted people living in poverty (later also. al. n. Microfinance.. er. io. including unemployed or low-income individuals). He called this revolutionary new system:. Ch. engchi. i n U. v. Yunus reasoned that if financial resources can be made available to the poor people on terms and conditions that are appropriate and reasonable, the millions of small people with their millions of small pursuits can add up to create the biggest development wonder (Grameen Bank, 2015). In this way, Yunus was the recipient of a Nobel peace prize for his strides against poverty. Naturally, his efforts have seen international recognition because of repeated reports of Microfinance having significant effects in reducing poverty. Furthermore, there have been numerous studies reporting that not only did Microfinance have significant impacts on the alleviation of poverty, but also the empowerment of women, the improvement of education, health and job creation (namely boosting entrepreneurship efforts) (Pitt, 1998). These effects were especially true in developing countries with communities with high indexes of poverty and inequality (Pitt, 1998).. 1.

(5) Today, Microfinance includes more services than just small loans targeted at people living below the national poverty line. The array of financial products includes: microsavings, microloans, microinsurance, and the traditional sending and receiving of remittances (Investopedia, 2017). These programs are specially designed to cater to those who are not financially literate and at an individual level do not consist of transactions with high volume – but due to its quick expansion and reception Microfinance has become a force to be reckoned with in the road towards the accomplishment of the MDG.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. 2.

(6) 1.1 Microfinance Though the original meaning of microfinance referred to lending; nowadays microfinance refers to the array of financial services offered by microfinancial institutions. Ultimately, the goal of microfinance is to give low-income people and opportunity to become self-sufficient by providing a way to save money, borrow money and get insurance (Grameen Bank, 2015). Microloans are given out with the intention of permitting clients to participate in productive activities or support a small business. Like conventional loans, lenders will charge interest on loans – and also require specific repayment plans with payments due at regular intervals. Microloans are typically not more than several hundred dollars. Ultimately what differentiates a microloan from a regular loan is the amount, inherent risk, and interest rate.. 治政 While the intended use of the loans is explicitly to 大 provide a road to alleviate poverty, 立used to cover needs such as food, shelter, and short term debt. some loans have reportedly been ‧ 國. 學. As expected, loans that are taken out to satisfy these needs create over-indebtedness and further exacerbate the poverty condition. In a further section, key words such as over-indebtedness. ‧. will be defined.. y. Nat. Microsavings refer to a small deposit account offered to clients at some microfinance. sit. institutions which allow lower income families or individuals to store funds. Similar to a. al. er. io. normal savings account, clients can withdraw and deposit money at any time. These small. v i n minimum balance requirements orCset at an extremelyU h e n g c h i low threshold. Institutions further incentivize the usage of this service by allowing users to save small amounts of money and n. deposit accounts are specially designed around smaller amounts of money, often waiving. having no charges for the service. This practice is mainly seen in Microfinance institutions located in developing countries as a way to encourage low income individuals to have some money set aside for any unforeseen expenses such as health bills or disaster coping expenses. Microinsurance typically refers to insurance services offered to clients with low income and limited access to traditional insurance and risk management services. It is a specially targeted means of protecting low income people against specific risks which they are more likely to experience. The risks to which low income people are exposed to are inherently different due to factors such as lifestyle, natural disasters, poor growing seasons, among others.. 3.

(7) Premiums for microinsurance programs are calculated based on the degree of risk that the client experiences. Furthermore, the service assigns high importance to factors such as: affordability, inclusiveness, simplicity and clarity in documentation, accessible processes, and building trust among target clients. Remittances play a very important part for people living in developing countries, especially in Latin America. Because of its geographical proximity to a developed country such as the United States, many Latin Americans individually emigrate to the United States for work and their families are usually left in their home country. For this reason, these people need a way to send financial resources back to their families in an affordable way. Ever since the year 2000, the remittance flows into Latin America and the Caribbean. 政治大. have seen a growing trend, with exception of 2009 in which as a result of the international financial crisis remittance flows dropped by US$ 8.5 billion. However, the year after that the. 立 continued. trend. to. manifest. itself.. 學 ‧. ‧ 國. growing. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 1 – Source: Multilateral Investment Fund. 4.

(8) 1.2 Problem A number of studies have been conducted in order to verify the effectiveness of Microfinance in accomplishing its declared goals and more importantly measuring its impact. Studies with different kinds of approaches such as: Cluster RCT, Controlled trials, RCT, and with and without approach methodologies have been conducted on different scenarios such as Africa, Latin America, South-east Asia at different points in time; the studies have come up with a plethora of mixed results. These studies measured the impact that microfinance has on dependent variables such as poverty (extreme poverty, living below the national poverty line), women’s empowerment (sometimes measured as prevalence of female household leads, women’s participation in the labor force, prevalence of intra household violence, among others), education (literacy rates, enrollment rates, child labor), job creation, household consumption. 政治大 expenditure structured consumption such as balancing the durable goods expenditure and the 立 temptation goods expenditure of a household), household income, and health (risk of (to cover short term commitments such as groceries, basic service bill payments, or more. ‧ 國. 學. contracting certain diseases which people living in poverty are more prone to) . Many studies have since then found positive effects of an increased penetration of. ‧. microfinance, total portfolio size, and client diversity on women’s empowerment, education,. y. Nat. and job creation. Lacalle Calderon et al. (2008) conducted a with and without impact study on. n. al. Ch. er. io. a positive effect on household accumulation of wealth.. sit. a Group-based lending to men and women in sub-Saharan Africa found that this program had. i n U. v. On the other hand, some cases such as Barnes, Gaile, et al. (2001) in a control trial study. engchi. found that Microfinance with group-based lending to men and women in sub-Saharan African actually has a negative impact on education that reduced enrollment rates of school-aged children. Because of these mixed results, many critics have argued that microfinance is not a panacea for improving living conditions for low-income individuals in every scenario – because of evidence proving otherwise (Duvendack, 2012). Detailed explanations of the results obtained in these and other studies will be discussed further in the literature review section. Moreover, critics have since long argued about there being a mission drift from the original microfinance intended purpose (Banerjee, 2015). Because of increasing microfinance interest rates, many critics argue that modern microfinancial institutions are not prioritizing the. 5.

(9) benefit of the clients and instead are commercializing the industry and questionably making money off of poor people.. 1.3 Purpose The purpose of this study is to identify the existence of an influence of microfinance over social dimensions, in the context of microfinancial services becoming increasingly accessible and diversified. The two main social dimensions being observed are poverty and secondary school enrollment (education). Because of the discrepancies in results, there is motivation to find out what the ultimate effects of the ever-increasing presence of microfinance is doing for the region in terms of contributing to the progress of the completion of the MDGs. After identifying whether or not a causal effect exists between microfinance and the. 治政 improvement/setback of women’s empowerment, the study大 expects to serve as a motivation to other researchers/scholars into 立 further researching of microfinance institutions and their role in betterment/amplification of the poverty situation, increase/decrease of school enrollment rates,. ‧ 國. 學. the Latin American region as a tool to reduce poverty, increase education, and improvement of women’s empowerment.. ‧. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. 6.

(10) 2 Literature Review Many different studies in several different scenarios have yielded mixed results with respect to microfinance and its effects on issues relating to the aforementioned millennium development goals (MDGs). The methodologies used to examine these effects range from RCT, IV research designs, Cluster RCT, Tobit regression, and PSM (propensity score matching). It can be argued that the discrepancy in results is due to vast differences between cultures, economic contexts (growth, recession), microfinance institutions’ business models (NGO or not), lending conditions (group lending, individual lending), client base (some microfinancial institutions loan to just women, just to the extremely poor). Another consideration to keep in mind is the design of the research being presented – some of the research is not replicable (due to data not being provided, complex statistical methods) and simply does not control for. 政治大 of Microfinance and its effects on a society sometimes overlook these differences and 立. unobservable biases, selection biases, among others. This being said, both critics and advocates. generalize the results.. ‧ 國. 學. The following section is divided into evidence in favor and evidence against. These. ‧. mean, respectively, evidence providing reason to believe that microfinance has significant positive impacts in the betterment of social outcomes such as poverty, and evidence providing. sit. y. Nat. reason to refute the aforementioned belief. The section overlooks several of these studies and. io. er. summarizes the results while also providing a brief description of the methods used and considerations to keep in mind while evaluating the evidence. The studies have been conducted. n. al. i n U. v. in different periods of time across vastly different communities while also employing different. Ch. engchi. lending methods and carry different microfinancial models.. 2.1 Theoretical and Empirical Review 2.1.1 Evidence in Favor Several research papers concerning positive effects of Microfinance on numerous dimensions of society such as: poverty, education, women’s empowerment, health, household consumption, business and job creation, income, among others have been written after conducting studies and statistical analyses of the results. Some of the papers that provided evidence that microfinance and its distinct products have a positive effect on resolving societal issues include: Imai et al. (2010), Katsushi et al. (2012), and Banerjee et al. (2015). In the case of Imai et al. (2010), a study that was conducted in India with household data from 2001 was looking for the effects that household access to microfinance had on 7.

(11) reducing poverty. The study used a treatment effects model in order to identify a causal effect between the increase in availability of microfinancial services and the reduction in poverty. The study specifically focused on estimating the poverty-reducing effects of MFI loans lent out for “productive purposes such as investment in agriculture or non-farm businesses on household poverty levels”. In order to verify the significance levels of the treatment effect coefficients, the study relied on two models: The Tobit and the PSM (Propensity score matching). The results they were able to report after the study concluded was that despite the unobservable important determinants of access to a microfinancial institution – there was a significant positive effect of microfinancial productive loans on multidimensional welfare indicators. The point of using the Tobit and PSM models were to estimate poverty-reducing effects of access to MFIs and loans used for productive purposes and to check the robustness. 治政 study used the Indexed based Ranking (IBR) indicator大 which is used to reflect multi立 dimensional aspects of poverty. of the results. In order to measure the effects on several different dimensions of welfare, the. ‧ 國. 學. However, the most important contribution that was made by this study is that when the effects on both rural and urban areas are considered separately some important distinctions. ‧. arise. Namely, that in rural areas when loans were used for productive purposes – there was a. y. Nat. significantly higher estimate of poverty-reducing effect (raising the IBR) than just the fact of. sit. having access to a loan from an MFI. These results allow for the following interpretation: the. n. al. er. io. client’s intended use of the loan is important in determining poverty reduction outcomes.. i n U. v. On the other hand, in the urban areas “simple” access (namely, not necessarily for any. Ch. engchi. specific purpose) had a larger average poverty-reducing effect than taking the loan for a productive purpose. Katsushi et al. (2012) in their study test the hypothesis that microfinance reduces poverty at the macro level using cross-country and panel data which are constructed by the Microfinance Information Exchange data on MFIs and the World Bank data. The research used cross-country data for 2007 and a panel for 2003-2007. This study is very similar in design and premise to this research but important distinctions are made down the road. Primarily in sample selection, approach methodology, and robustness checks. The primary findings of this study are that countries that host Microfinancial institutions with higher gross loan portfolio per capita tend to have lower levels of poverty indices. The study also mentions that their results differ from recent evidence at the micro level in that their results suggest that microfinance. 8.

(12) significantly reduces poverty at macro level. The researchers conclude with the policy implication that these results support the idea that “funds should be channeled from development finance institutions and governments of developing countries into MFIs”. From the start, the study is based on a context of large criticism surrounding the overall feasibility and effects of promoting a quasi-omnipresence of MFIs especially in countries belonging to the developing world. Much of the studies claiming that microfinance and its effects are over-glorified are products of studies which use the randomized control trials. Their results (further detailed in the coming section) point towards little, weak or insignificant effects of microfinance on dimensions such as women’s empowerment (in any of its sub dimensions) and poverty reduction.. 政治大. Importantly, the results reported by this study claim that microfinance loans per capita are significantly and negatively associated with poverty. In a more general (macro) sense, this. 立. means that “a country with a higher MFI gross loan portfolio per capita tends to have lower. ‧ 國. 學. poverty after controlling for the effects of other factors influencing it”. When substituting for other dependent variables (switching from poverty headcount ratio to poverty gap and squared. ‧. poverty gap) the relationship between gross loan portfolio and poverty remains unchanged. The substitution of the dependent variables suggests that microfinance not only reduces the. Nat. sit. y. incidence of poverty but also its depth and severity (multidimensional poverty).. al. er. io. The study concludes on a strong note assuring that the findings completely counter the claims that microfinance is “oversold” and the claims that it has little, weak, or insignificant. n. v i n effects on poverty are “widely off the largely mistaken”. Cmark h e ifnnot gchi U. Banerjee et al. (2015) confronts a similar question in a familiar scenario, India. The. question it seeks to answer is whether microfinance and its iterations have significant effects on poverty alleviation, health, education, women’s empowerment, small business investment, consumption, and business profits. Comparatively, this study sets a very wide scope to search for results. It belongs to the same group of studies that Katsushi et al. (2012) referred to in their concluding remarks. The study uses the RCT (randomized control trial) method to follow the results of a group-lending microcredit program in 52 out of 104 slums in Hyderabad, India. The study worked alongside an MFI and followed up on the institution’s and client’s results after a period of 3 years. The study serves as a focal point of critics of Microfinance and claims of the existence of a mission-drift in the original “plan” of microfinance.. 9.

(13) The study focused on availability of taking out a microfinancial loan in a groupmodality (this means that a group of people are held accountable for the loan at the same time). This group-lending modality is known to be the standard as microfinancial loans usually lack the strict pre-condition of owning reliable collateral or a better-than-average credit history. By using this peer system, MFIs are indirectly able to reduce failure to repay. The study’s findings reveal that demand for microfinancial loans is not as present as some argue it to be. The results report that by the end of a 3-year study period, only 33 percent of households borrow from an MFI – and this is from a subgroup of people that were selected for their higher than average propensity to borrow. The implications of these results are that most households do not have a project with an RoR of at least 24% which was the APR on a loan from the microfinancial the study was following. Furthermore, it could also mean that. 政治大 relatives or money lenders (Collins et al. 2009) 立. instead of borrowing from a formal institution, the borrowers preferred to seek out friends,. ‧ 國. 學. Other results reported that microloans have a positive effect on business expansion for loan holders, but it does not appear to fuel an escape from poverty based on those small. ‧. businesses. The study used the monthly consumption values for the households to measure overall welfare, and based on these values the researchers concluded that monthly household. Nat. sit. y. consumption did not increase equaled to a non-discernible effect on poverty. For the average. io. upper tail of profitability.. al. er. business profitability does not increase, although there are some significant increases in the. n. v i n Finally, the study found that C hmicrocredit does Uaffect the structure of household i (such as stoves, vehicles, furniture) e ndurable g c hgoods consumption. Households will invest in home and will spend less on temptation goods and expenditures on festivals and parties. Microcredit seems to have a deepening effect on the sense of responsibility of borrowers (whether by peerpressure or a sense of owing something in return). Microcredit expands households’ abilities to make different intertemporal choices, including business investment. There are no discernible effects on education, health, or women’s empowerment. The results from Banerjee et al. (2015) can also be considered to correspond to both sides of the argument. Mainly because of the methodology used, the experimental design, and the way the results were interpreted. While it did find supporting evidence to increase the confidence that microfinance has positive effects on social outcomes such as household consumption structure and entrepreneurship, it also found that in the context in which their trial 10.

(14) was conducted – there was no discernible effect of alleviation of poverty, increase in education, overall betterment of health, or the empowerment of women (unless women being in charge of businesses doesn’t fit into the definition of empowerment). When interpreting these results, it is important to keep in mind that the study only followed a single microfinance institution and in a context of a high economic growth. While also considering that there may exist some cultural implications for borrowing money. 2.1.2 Evidence Against Nghiem et al. (2012) in their assessment of the welfare effects of microfinance in Vietnam study the consequences of NGO microfinance programs on household welfare in Vietnam 2004. The study uses household data from 470 observations across 25 villages taking care to overcome any self-selection bias (clients can choose whether or not to become clients).. 政治大 approach in order to deal with the threat of the bias. Member and non-member households were 立. The study states that their survey results were collected using a quasi-experimental survey. organized in such a way that they could share similar characteristics and thusly be available to. ‧ 國. 學. be compared.. ‧. The treatment variable used to indicate the influence of microfinance is the duration of participating in NMP microfinance. This treatment variable encompasses additional. Nat. sit. y. development activities provided by NMPs and is also expected to compound cumulative effects. io. er. of microfinance (the higher the coefficient). With regards to checking for validity of survey design, the researchers ran a Hausman test to examine whether or not the order in which eligible. n. al. i n U. v. villages received microfinance might not be entirely random – so the model the study preferred. Ch. engchi. was considering village fixed-effects; this would control for the order in which the villages received the services. The results presented by this study shows no significant effects of participation in NGO microfinance on household welfare, which was measured by looking at the values of income and consumption per adult equivalent. It is important to highlight that the microfinancial institution looked at was of NGO modality and therefore is not a for-profit model, similar to the original microfinance lending program pioneered by the Grameen Bank in Bangladesh. These NGO microfinancial programs (NMP) are known for having a client base composed of extremely poor individuals, but also follow the group-lending modality. The results at first glance present a statistically insignificant effect of NMP on household welfare.. 11.

(15) Considering these points, microfinance may show not to have statistically significant results because NMP clients have access to fewer services in lower amounts. With lower amounts for loans, the impact could be undetectable because it’s not strong enough to effect the proxy variables. Furthermore, the study states that one important limitation to its study was that it did not examine the impacts of NMP on welfare that cannot be calculated monetarily. Duvendack & Jones (2012) under the context of the continuous debate between the validity of iconic studies which have justified the effectiveness of Microfinance in providing a feasible way to improve social outcomes, re-investigated evidence from two popular studies regarding this subject. Several different studies including Roodman and Morduch (2005, 2010) have replicated the results from Pitt and Khandker (1998 – henceforth referred to as the original study) while also employing other methods of estimation. The results are similar, but the. 政治大 Duvendack & Jones explores the standing evidence and seeks out to replicate the study by 立. interpretation differs in that Roodman and Morduch deny the existence of a causal relationship.. Chemin (2008) which in itself was a replication of the results from the original study.. ‧ 國. 學. Respectively, the original study’s findings show that “microfinance enables the poor to access credit, providing them access to remunerative activities and relieving them of debts;. ‧. additionally, these results are accentuated when the services are targeted at women”. The. y. Nat. second, finds comparatively “positive but lower than previously thought” results. Duvendack. sit. & Jones consistently make the point that replicating the results of the original study is difficult. n. al. code that would allow easy reproduction of their results.. Ch. engchi. er. io. mainly due to failure by the original study to provide the data set which it worked with and the. i n U. v. Duvendack & Jones quote several other studies which perfectly illustrate the ongoing debate on whether microfinance is as effective as iconic studies allege; the main studies focusing on refuting the positive impacts of microfinance on any social dimension: Armendariz de Aghion and Morduch (2005, 2010), Goldberg (2005), Roodman and Morduch (2009), Roy (2010), Bateman (2010), Stewart et al. (2010), Duvendack et al. (2011). These criticisms of initial evidence have been refuted by Pitt (2011a, 2011b) from the original study. From the replication of Chemin (2008), Duvendack & Jones find differences in descriptive statistics from those reported but when compared to Roodman and Morduch (2009) the differences are “only minor”. This means that Chemin did not successfully replicate the dataset used in the original study, whereas Roodman and Morduch (2009) were closer to matching. From this perspective, it is not hard to question why Chemin had different results. 12.

(16) albeit still positive. Amongst the results, the research criticizes the original study for having a “weak research design, complex statistical analysis, poor documentation of the data and the absence of code”. In the same tone, Duvendack & Jones make it clear that their paper does not corroborate the original study’s findings and instead raises concerns about relying heavily on the results of one data set. The main takeaway from this research is that future policymakers would be wise to not rely too heavily on the original study, and instead opt for studies which have facilitated replication and better quality data production. Advocates of microfinance should also encourage replication of analyses in order to decide whether or not that investment is truly beneficial. The concluding note is that it would be better for the scientific community to have more transparent reporting methods and data processing in order to facilitate the corroboration. 政治大. or educated criticism of a scientific study.. 立. Augsburg et al. (2012) designed an RCT to directly analyze the impact of microcredit. ‧ 國. 學. on poverty reduction, child and teenage labor supply, and education. The researchers worked closely with a big established MFI in the area – and cooperating with the management to allow. ‧. for an experimental operation. The agents in charge of handing out the loans would need to. sit. Nat. that assuming a “slightly greater risk” would be accepted.. y. look for clients that would generally be rejected for loans as a result of general screening, but. al. er. io. The results included positive effects of microloans on new business creation, negative effects on savings of marginally more educated households (which they justify as “presumably. n. v i n to complement the loan and achieveC the minimum investment amount”). Augsburg et al. (2012) hengchi U also detected a decrease in consumption in less-educated households.. The most significant takeaway from the results was the substantial increase in the labor supply of children aged 16-19, which occurred simultaneously with a reduction in their school attendance. This implication would shock many advocates of microfinance, as it is the generally accepted consensus, within that community that microfinance and its iterations would benefit this social outcome (Pitt, 1998; Khandker, 2005). However, the results presented by this study show that it is quite the opposite and instead provide the explanation that because a poor household’s marginal utility of having an extra hand contributing to the household income is higher than for the average household – the household heads opt for choosing a working child instead of assisting an educational facility.. 13.

(17) 3 Empirical Process As a first step, literature was consulted in order to have a background understanding of the Microfinance. Sources containing information on: microcredit theory, past studies, critiques of Microfinance, and current events surrounding the industry were all consulted. This provided the researcher with a solid background knowledge of the topic which helped the construction of models and interpretation of results – and further on assisting in the understanding of policy implications. For statistical software, R was chosen because of a higher degree of flexibility in the operations. R has progressively been gaining more recognition amongst seasoned statisticians and has been personally recommended by other researchers. R has a highly resourceful library. 政治大. of packages and commands that can be installed in a quick and easy way. For this study, the R packages of ‘plm’, ‘stats’, ‘lmtest’, ‘coeftest’ and ‘stargazer’ were used. The ‘plm’ stands for. 立. “panel linear models” and contains several intuitive commands that make it simpler to run a. ‧ 國. 學. panel model style linear regression. ‘stats’ and ‘lmtest’ both contain important statistical commands necessary for running the validity testing and other statistical analyses. ‘stargazer’. can be viewed.. ‧. is the package used to compile the results in a neatly organized chart in which regression results. y. Nat. sit. After choosing the statistical software in which to run the regression models, further. al. er. io. literature was consulted in order to obtain the data required to build the benchmark models.. n. Benchmark models are the standard way international entities such as the World Bank, Socio-. Ch. i n U. v. economic development for Latin America and the Caribbean, or the Economic commission for. engchi. Latin America and the Caribbean would measure the effects of certain regressors on a dependent variable. After the data for the benchmark models was collected and consolidated, the data for Microfinance variables (further detailed individually in a later section) was researched and taken from the Multilateral Investment Group, a member of the Inter-American development bank (IDB). Once the information for Microfinance variables was consolidated completely, the next step was to import the dataset into R-studio for evaluation and execution of the regression models. The following table displays a summary of all the models proposed:. 14.

(18) Table 1 – Microfinance Models Model Education. Dependent (Y) School enrollment rates (secondary school) Household consumption Household Income per capita Labor force participation (women) / Household lead (female) Percentage of population living under national poverty line Percentage of labor force that is selfemployed. Household consumption Income per capita Women’s empowerment. Poverty. Entrepreneurship. 立. # of regressors (X) 3. Final revision Yes. 3. No. 4. No. 4. No. 3. Yes. 政 3治大. No. ‧ 國. 學. After the results were reviewed, the models with significant results and good degrees. ‧. of robustness were preferred to be presented in the results section of this research. For a more detailed view of the results, with direct print-outs of the regression results look into the results. y. sit. al. er. io. 3.1 Panel Data. Nat. section.. n. v i n C h of 4 years. Because where 20 countries are observed for a period e n g c h i U of the inclusion of both crossThe data for the model is organized as a longitudinal long-form balanced panel data,. sectional and time series data, panel data was deemed the most efficient model to run the regression on. Several estimators with which to run the regression were considered, such as: pooled OLS estimator, between estimator, fixed effects, and random effects. Further details of the chosen estimator will be detailed in the following section. A pooled OLS regression usually runs under the assumption that there is no panel effect and that all observations in the model should be treated with no individual effect (Lambert, 2013). There are diagnostic tests that can be run in order to confirm whether or not this is the case. The between estimator which takes the individual effects model and averages out the time component resulting in a regression which equals the average change in the dependent 15.

(19) variable as a result of an average change in the explanatory variables (Lambert, 2013). It only works under the assumption that the idiosyncratic errors (α) are random effects. If this was the case however, a researcher would most likely run the regression under the random effects estimator.. 3.2 Fixed Effects vs. Random Effects A standard fixed effects model has the following form:. 𝑌𝑖𝑡 = 𝛽1 𝑋𝑖𝑡 + ⋯ 𝛽𝑛 𝑋𝑛𝑡 + 𝛼𝑖 + 𝜇𝑖𝑡 A standard random effects model has the following form:. 𝑌𝑖𝑡 = 𝛽1 𝑋𝑖𝑡 + ⋯ 𝛽𝑛 𝑋𝑛𝑡 + 𝛼 + 𝜇𝑖𝑡 + 𝜀𝑖𝑡. 政治大. The main difference between the fixed effects estimator and random effects estimator. 立. is that inside a fixed effects model there are individual effect dummy variables fixed into the. ‧ 國. 學. regression. (Lambert, 2013).. A random effects estimation is usually preferred over fixed effects because of higher. ‧. efficiency in fitting the regression line. (Lambert, 2013). sit. y. Nat. In order to decide what estimator is more appropriate for any given model, the researcher can run a diagnostic test known as the Hausman test (Reyna, 2010)(which is also. al. er. io. included as a command in the ‘plm’ package for R). To perform this test, the researcher must. n. v i n C h is the more consistent of the Hausman test is that random effects e n g c h i U and efficient, thus a high pfirst run both regressions and then use the results to run the Hausman test. The null hypothesis. value (>0.05) will confirm that random effects is the more efficient estimator (Reyna, 2010). The Hausman test statistic is denoted by the following: †. 𝐻 = (𝑏1 − 𝑏0 )′ (𝑉𝑎𝑟(𝑏0 ) − 𝑉𝑎𝑟(𝑏1 )) (𝑏1 − 𝑏0 ) . Where † is the Moore-penrose pseudoinverse. . b1 is the random effects estimation. . b0 is the fixed effects estimation. While considering the fixed effects model, there was the possibility that there was a need for inclusion of time fixed effects. The plm package in R also offers several tools in order to identify any need for such inclusion such as the Lagrange Multiplier Test, where the null 16.

(20) hypothesis is that no time-fixed effects are needed (Reyna, 2010). Upon running the test and observing the p-value, the researcher can then decide whether or not there is sufficient evidence to reject or fail to reject this hypothesis.. 3.3 Random Effects vs. Pooled OLS A Breusch-Pagan lagrange multiplier test (LM) which tests for random effects can yield the necessary results to decide between a random effects regression and a simple OLS regression (Reyna, 2010). The prerequisite for using this test is to run a pooled OLS regression of the dataset and then the software will automatically determine whether or not random effects are appropriate for the model. The null hypothesis of this test is that random effects is not the appropriate estimator, implying that there are not any significant evidences pointing towards differences across countries (Reyna, 2010). A high p-value (>0.05) will determine that the. 政治大. researcher fails to reject the null – which means that the model should use the pooled OLS. 立. estimator to run the regression.. ‧ 國. 學. 3.4 Data Collection. Several databases were consulted for the purpose of data collection. Careful attention. ‧. was placed in order to avoid taking data for the same variable from different sources; as this. Nat. sit. different entities at the time of putting together their databases.. y. would cause inconsistency across the data set due to possible different estimates taken by. al. er. io. For all variables concerning microfinance (namely: total clients, total portfolio,. n. v i n C h InvestmentUFund were consulted. These reports reports from 2011-2014 from the Multilateral engchi contain country-specific information about the development and performance of microfinance microfinance lend rates, average credits, and microfinance penetration) a series of annual. institutions. This was used as the primary data source for microfinance related variables because of the thoroughness with which the data was collected, and because the reports are published by an internationally respected institution. These yearly reports are available for public download in their website. For macroeconomic variables such as poverty rate, population, rate of employment. Also due to its reputation in the international scene this source was preferred above all others where data was available. Unfortunately, all the data could not have been taken from the world bank due to access limitations. Important data such as household consumption, income per capita, labor force, literacy rates, school enrollment rates, among others were not available in the world bank database, or incomplete. 17.

(21) In the cases that data was missing in the world bank database, the study proceeded to collect data from equal in thoroughness but lesser known sources such as the United Nations Economic commission for Latin America and the Caribbean and the Socio-Economic Database for Latin America and the Caribbean. Both of these sources filled in the missing data from World Bank, but in cases where the data was available in all three cases the data differed – which indicates a different in their estimations of things such as unemployment rate, poverty, economically active population (also known as active labor force), among others. It is important to keep these estimation differences when replicating the results. In some cases, ideal variables were withdrawn from the final dataset because of missing data in all sources, such as new business creation (volume) which would have helped in the construction of additional models and further strengthen the thoroughness of the study.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. 18.

(22) 4 Empirical Model In order to insure that the estimation of the models with the microfinance variables included in the regressors was consistent with official measures of the dependent variables chosen: reputable sources such as the World Bank and the Socio-economic database for Latin America and the Caribbean were consulted. The next step was to insert the microfinance related variables into the estimation (namely: total microfinance clients, total investment portfolio, and average credit).. 4.1 Model 4.1.1 Poverty Model For the poverty model, the World Bank (Coudouel) was consulted. The report that was. 治政 efficiently. Important details distinguishing the use of household 大 consumption and income per capita were made – where the 立 recommendation was to use household consumption instead of retrieved contained recommendations on what variables to use in order to measure poverty. ‧ 國. 學. income per capita, as it can capture more dimensions of poverty (Coudouel). For the rest of the regressors in the poverty model, variables such as the percentage of. ‧. the labor force occupied in the informal sector, the rate of employment and gini coefficient of. y. Nat. income were chosen. According to the report published by the World Bank (Coudouel), labor. sit. informality plays a large role in measuring poverty because an increase in household poverty. al. er. io. increases the likelihood of employment in the informal sector. The gini coefficient on the other. v i n Ctheh highest income earners into how deep the difference between e n g c h i U and the lowest income earners. n. hand, measures the severity of the disparity in the income per capita – thus provides an insight. The estimator used in the estimation of the coefficients of the poverty benchmark model was fixed effects. This is due to the correlation of individual effects with other regressors and was confirmed by running a Hausman test after running one regression considering fixed effects and another considering random effects. For the sake of consistency and former comparison between benchmark and microfinance models, the preferred estimator for both models was fixed effects. Thusly, the poverty benchmark model is:. 𝑌𝑖𝑡 = 𝛽1 𝑋𝑖𝑡 + 𝛽2 𝑋𝑖𝑡 + 𝛽3 𝑋𝑖𝑡 + 𝛽4 𝑋𝑖𝑡 +∝𝑖 + 𝜇𝑖𝑡 Where:. 19.

(23)      . Yit is the dependent variable (poverty headcount) where i = entity and t = time β1 is household consumption β2 is labor informality β3 is rate of employment β4 is the Gini coefficient for income per capita µ it is the error term. . αi is the unknown intercept for each entity.. Introducing the microfinance variable, the resulting model is:. 𝑌𝑖𝑡 =∝𝑖 + 𝛽1 𝑋𝑖𝑡 + 𝛽2 𝑋𝑖𝑡 + 𝛽3 𝑋𝑖𝑡 + 𝛽4 𝑋𝑖𝑡 + 𝛽5 𝑋𝑖𝑡 + 𝜇𝑖𝑡 . Where β5 is the average credit.. 4.1.2 Education Models. 立. 政治大. In the construction of the education model, the suggested model was consulted from the. ‧ 國. 學. institute for Fiscal Studies (Augsburg, 2012). The report is a study that uses an RCT to analyze the impact of microcredit on poverty reduction, child and teenage labor supply, and education.. ‧. While the sample composition for the RCT in question and the present research is far from similar, the concept that Augsburg et al. propose is sound. If the child labor supply is. Nat. sit. y. indeed being affected by school attendance (and it should), then the inverse should also be true.. er. io. In simple terms, if school attendance is increasing then the labor supply for children of this age group should decrease. At this point, it is important to make the clarification of what exactly is. n. al. Ch. i n U. v. meant by “child”. Similar to the concept used by Augsburg et al., the children in their teenage. engchi. years (namely 15+) are more prone to be used as an alternative source of income for a family. Thusly, observing children of primary school age would yield different results (also there exists a limitation of availability of this type data). Simply, collecting and analyzing data for secondary school aged children in this context makes much more sense. The explanatory variables were selected based on the conclusion and remarks of this RCT. The explanatory variables included in the model were: household consumption, labor supply of children of age 15-18 (% of total), and poverty headcount. Household consumption is used to measure welfare; labor supply of children is used to provide a clearer picture of what is happening to the children in this age group as a result of their parents or guardians receiving a microfinance loan. Poverty is included in order to pinpoint the suggested effect that microfinancial loans might have on the status of a household with respect to poverty. 20.

(24) The preferred estimator used for these model was the random effects in both cases (total client estimation, and average credit estimation). Similar to the poverty model, the Hausman test provides a clear way to choose between what estimator is preferred. In a similar way to the poverty model, the inclusion of the microfinance variable means the inclusion of an additional term. This is why the benchmark for education is omitted. Thusly the education models are:. 𝑌𝑖𝑡 = 𝛽1 𝑋𝑖𝑡 + 𝛽2 𝑋𝑖𝑡 + 𝛽3 𝑋𝑖𝑡 + 𝛽4 𝑋𝑖𝑡 +∝𝑖 + 𝜇𝑖𝑡 + 𝜀𝑖𝑡      . Yit is the dependent variable (secondary education enrollment) where i = entity and t = time β1 is household consumption β2 is labor supply for children aged 15-18 (% of total) β3 is poverty (poverty headcount) β4 is the microfinance variable (total clients) µ it is the error term. . αi is the unknown intercept for each entity.. ‧. ‧ 國. 學. And. 立. 政治大. Nat. 4.2 Handling of Missing Data a. sit. io. Where the only difference is that the microfinance explanatory variable is average credit.. er. . y. 𝑌𝑖𝑡 = 𝛽1 𝑋𝑖𝑡 + 𝛽2 𝑋𝑖𝑡 + 𝛽3 𝑋𝑖𝑡 + 𝛽4 𝑋𝑖𝑡 +∝𝑖 + 𝜇𝑖𝑡 + 𝜀𝑖𝑡. n. v i l n The selection of variables C with which to put together the models was challenging hengchi U because of two main reasons: the countries in the data set have a lot of inconsistent/missing data, which means that finding a variable for which all observations had data available in all years was rarely possible and the values used different estimates for some countries (where data wasn’t available in one source). Instead, all readily available data was collected and at a later time the countries were further sub-divided into groups. The criteria for dividing observations into groups included but was not limited to:  . . Dividing countries into a large subgroup labeled “countries with available data” Dividing countries into a smaller subgroup with highest “x”. X being the dependent variable being studied. (I.E. X is Poverty Headcount in the case of the poverty model) Dividing countries into a smaller subgroup based on geographical location (making a distinction between Central America and South America) 21.

(25) . Dividing countries based on a constructed variable labeled “MF Penetration” which was calculated by (total MF clients / population). The subdivision of groups also took into consideration those countries which presented outliers as data points. Such was the case of Venezuela which presented values far greater than its counterparts in many points and has had an unusually unstable political situation in recent years causing major reductions in both perceived and real welfare of its citizens. The aforementioned models were run for all subdivided groups and statistically significant results were only found for countries with the highest poverty (in the case of the poverty model) and countries which have available data (in the case of the education model). After the division was complete, the resulting observed groups are the following:. 治政大 Education Poverty Guatemala Argentina 立Honduras Honduras. Table 2 - Model Obs. Breakdown. n. Ch. engchi. y. sit. ‧ 國. io. al. El Salvador Dominican Republic Colombia Ecuador Paraguay Bolivia Peru Brazil Panama Costa Rica Uruguay. ‧. Nat. El Salvador Dominican Republic Colombia Ecuador Paraguay Bolivia Peru Brazil N/a N/a N/a. 學. Obs. 1 (2011,2012,2013,2014) Obs. 2 (2011,2012,2013,2014) Obs. 3 (2011,2012,2013,2014) Obs. 4 (2011,2012,2013,2014) Obs. 5 (2011,2012,2013,2014) Obs. 6 (2011,2012,2013,2014) Obs. 7 (2011,2012,2013,2014) Obs. 8 (2011,2012,2013,2014) Obs. 9 (2011,2012,2013,2014) Obs. 10 (2011,2012,2013,2014) Obs. 11 (2011,2012,2013,2014) Obs. 12 (2011,2012,2013,2014) Obs. 13 (2011,2012,2013,2014). er. Model. i n U. v. 4.3 Microfinancial Variable Selection The variables which represent the presence of microfinance were directly extracted from a series of reports known as “Microfinance in Latin America and the Caribbean 20112014” translated from Spanish. These reports were published by the Multilateral Investment Fund as a subgroup within the Inter-American bank for development. The reports contain detailed data on variables ranging from portfolio, number of total clients, microfinance penetration index, the business environment for microfinance institutions, and several measures of performance and depth of the services offered. The reports are available for download from the Multilateral Investment Fund website.. 22.

(26) Total client (tot_client) refers to the total amount of clients in the country, including client bases from both formal and informal microfinancial institutions. For later iterations of the report, the distinction is made between how many clients belong to formal institutions and informal institutions. Amongst the observations it can be noted that some clients have comparatively higher numbers – in order to make a fairer comparison the microfinance penetration index was also calculated which allowed for a straightforward comparison between total clients and total population.. Total Clients 2014. 政治大. AR. ‧ 國. 立 BO. BR. CO. CR. EC. SV. 學. 4500000 4000000 3500000 3000000 2500000 2000000 1500000 1000000 500000 0. GT. HN. PA. PY. PE. DO. UY. ‧. Figure 2- Total Clients Vizualized. sit. y. Nat. Total portfolio (tot_portf) refers to the total amount in current US$ dollars that the. io. n. al. 7E+09. i n U. Total Portfolio 2014. Ch. 6E+09. engchi. er. collective microfinance sector has in current loans.. v. 5E+09 4E+09 3E+09 2E+09 1E+09 0 AR. BO. BR. CO. CR. EC. SV. GT. HN. PA. PY. PE. DO. UY. Figure 3 – Total Portfolio Visualized. 23.

(27) Average credit (avg_credit) refers to the result of. 𝑡𝑜𝑡𝑎𝑙 𝑝𝑜𝑟𝑡𝑓𝑜𝑙𝑖𝑜 𝑡𝑜𝑡𝑎𝑙 𝑐𝑙𝑖𝑒𝑛𝑡𝑠. . The result gives a better. idea of how much the average loan is and serves as a closer estimate to the marginal impact that a microfinancial loan might have on the household level (Multilateral Investment Fund, 2014). This was important because in order to directly compare microfinance with household data (such as household income and household consumption) a more manageable variable was needed.. Average Credit 2014 16000 14000 12000. 政治大. 10000 8000. 立. 6000. 0 BO. BR. CO. CR. EC. SV. GT. ‧. AR. ‧ 國. 2000. 學. 4000. HN. PA. Nat. PE. DO. UY. n. al. er. io. sit. y. Figure 4 – Average Credit Visualized. PY. Ch. engchi. i n U. v. 24.

(28) 4.4 Correlation. 立. 政治大. ‧. ‧ 國. 學. io. al. er. sit. y. Nat. Figure 5- Correlation Chart. n. v i n C h a negative correlation The red coefficients (negative) represent e n g c h i U between the sets of variables, The chart above shows the correlation between all the variables used in both the models.. while the blue coefficients (positive) represent a positive correlation between the variables. Additionally, the chart also includes the estimated correlation coefficient. In a general sense,. the models don’t include a very high inter-correlation between dependent and independent variables.. 25.

(29) 5 Empirical Results The microfinance model for poverty suggests that there is a negative relation between average credit (total clients/total portfolio) and poverty headcount below the national poverty line. The regression table below shows the results for the regression under both estimators. Poverty Model Results Dependent variable: Poverty Rate Fixed Effects (1). Ch. engchi. Observations R2 Adjusted R2 F Statistic. 36 0.79036500 0.66648970 16.58886000*** (df = 5; 22). Note:. *. y. sit. ‧ 國. n. al. -0.00329202** (0.00151460) 0.52757820*** (0.11753140) -0.81455020** (0.30545970) 0.43984180* (0.25035710) 48.33889000 (28.61314000). ‧. io. Constant. Nat. Gini coefficient. -0.01001087 (0.00300352) 0.45602140*** (0.11053930) -0.61868940* (0.34168360) 0.30367800 (0.27433540). (0.00042991). 學. Rate of employment. ***. ***. er. 立. Household consumption Labor informality. 政治大-0.00120083. -0.00081639* (0.00043816). Average credit. Random Effects (2). i n U. v. 36 0.71376510 0.66605930 14.96180000*** (df = 5; 30). p<0.1; **p<0.05; ***p<0.01. The microfinance models for secondary school enrollment suggest that there is a statistically significant positive relation between total clients, and average credit.. 26.

(30) Education Model Results Dependent variable: Secondary School Enrollment Models Total Clients (1) Total Clients Household consumption. Average Credit (2). 0.00000113** (0.00000055) 0.00066976. 0.00068531 (0.00098564) 政治大 -0.61470210. (0.00096841) Labor Student. -0.67089260***. Poverty. (0.12268240) -0.30095110*** (0.09287946). ***. 立. Average Credit. ‧. ‧ 國. 學. (0.12375600) -0.29645130*** (0.08891306). 0.00039846*. y. Nat. (0.00022455). Observations R2 Adjusted R2 F Statistic. 52 0.58334280 0.54788260 16.4506*** (df = 4; 47). Note:. *. er. n. al. 107.50650000*** (9.69347000). sit. 110.59970000*** (9.55171900). io. Constant. Ch. engchi. i n U. v. 52 0.59723120 0.56295310 17.4231*** (df = 4; 47). p<0.1; **p<0.05; ***p<0.01. Under the assumption that these low-resource people would use loans in a productive way (such as creating new businesses, investing, etc.); a realistic way to approach the mitigation of population living under the national poverty line would be to increase portfolio of existing microfinancial institutions or to encourage the amount of existing institutions which would also increase the portfolio.. 27.

(31) Because the children attending secondary school are of “suitable” age to help with the household income (in the form of holding an independent job or helping with the family business). By looking at secondary school enrollment, it is possible to draw the conclusion that if comparatively, school enrollment rates are going up while microfinance clients and average credits go up – then it could be concluded that household heads are preferring children’s education over a job (which means the loan has provided them with enough resources to account for what a child might bring in). Conversely, if the employment rates of children of this age are going down while school enrollment rates are going up – it can be interpreted as less children are working and are opting instead for working on their education. These results support the theory that microfinance is a viable and accessible tool with which to approach the achievement of the MDG. However, further studies are required in order. 政治大 presence and the improvement of poverty and school enrollment rates for secondary students. 立. to identify whether or not there is a causal relationship between the increase in microfinancial. In accordance with other studies conducted in India, Bangladesh, and Sub-saharan Africa: it. ‧ 國. 學. appears that Latin America can also benefit from exploring microfinance and looking forward to the long term effects that can come from microfinancial operations.. ‧. 5.1 Validity Testing. Nat. sit. y. After the initial results were collected, some further tests were conducted in order to. io. er. verify that the selected estimators, and the resulting coefficients were reliable in the context on the data set. Several tests mentioned in the methodology section were run in order to test for. n. al. Ch. i n U. v. robustness and significance of the results. The follow tables present the results of the tests for each model:. engchi. Table 3- Poverty Model Validity Tests Model. F-test Indv. Lagrange M. Random Effect Test (Time Effects effects) (BreuschPagan) Microf. (Avg. 0.01267 0.3598 4.422e-10 Cred) FE Micrf. (Avg. 0.02689 0.1436 2.968e-10 Cred) RE. Heteroskedasticity Hausman test test. 0.7076. 0.06298. 0.4225. 0.06298. Even though the Breusch-Pagan test detected the existence of random effects, this means that when comparing the results of a simple OLS regression with that of the random. 28.

(32) effects regression – the random effects was preferred. However, from the results of the Hausman test it can be argued that since the p-value is so close to the threshold, the estimation can also be run using both estimators. Table 4- Education Models Validity Tests Model. F-test Indv. Lagrange M. Random Effects Test (Time effects effects) (BreuschPagan) Micrf. (Tot. 0.3087 0.3382 2.074e-13 Client) RE Micrf. (Avg 0.3337 0.2068 4.296e-14 cred) RE. Heteroscedasticity Hausman test test. 0.4172. 0.377. 0.002769. 0.31. 5.1.1 Controlling for Heteroscedasticity. 治政 included into the regression, the Breusch-Pagan test for homoscedasticity rejected the null – 大立 of heteroscedastic errors (Reyna, 2010). In order to control which means that there is a presence In the case of the education model in which the microfinance variable of average was. ‧ 國. The results of conducting this test yielded the following results:. 學. for such errors, the ‘coeftest’ command group introduced in Reyna’s presentation of panel data.. ‧. Table 5- Evidence of Heteroscedasticity. Lbr_Scnd. HC0. 10.29937. 0.0001781321 0.0007940220 0.1423507. 0.09407347. HC1. 10.83336. 0.0001873678 0.0008351900 0.1497312. 0.09895093. HC2. 11.07398. HC3. 11.93099. HC4. 11.70319. er. io. sit. Hconsump. y. Avg.Credit. Nat. (Intercept). n. a0.0001938740 0.0008498943 i v0.1529606 l C n 0.1646441 U h e n g0.0009111055 0.0002115795 chi 0.0002136632 0.0008875745 0.1610446. Poverty. 0.10068592 0.10783313 0.10380970. The values correspond to the standard errors given different types of heteroscedasticity HC0-HC4 (Reyna, 2010): . HC0: heteroscedasticity consistent.. . HC1, HC2, HC3 – Recommended for small samples, HC3 gives less weight to influential observations.. . HC4 – Small samples with influential observations.. The important estimates are HC1-HC3 which are the recommended estimates for small samples such as the present one. As is easily observable, the coefficient for labor supply for children of secondary school age has a relatively high standard error. 29.

(33) 6 Conclusion In conclusion, the ongoing debate surrounding Microfinance and its causal effects on social dimensions such as poverty, education, health, women’s empowerment, business creation, income, and household consumption has been widely investigated on a global level. Many studies have been conducted concerning the debate while using different methods of statistical estimation. The results are astonishingly mixed, giving both sides of the debate sufficient expert opinions with which to support their arguments. However, most of the sources cited in this study to be against the idea that Microfinance can have a significant impact on reducing poverty and having positive effects on other social dimensions, simply claim that Microfinance should not be over glorified – as in some way or another these researchers have also found – through replication and reproduction of iconic studies – that Microfinance has. 政治大. some effect on distinct social outcome variables.. 立. This study corroborates the findings of the literature in that it was able to identify. ‧ 國. 學. statistically significant negative effects on the poverty rate, and positive effects on the secondary school enrollment rate for children of secondary school age for the Latin American. ‧. and Caribbean region. However, these results do not indicate a causal relationship between Microfinance and the mentioned social outcome variables. Instead it provides insight into the. sit. y. Nat. general effect of the presence of Microfinance in the region on these variables. The research. io. er. seems to serve as a means to justify the funding or pursuing of further research of the topic in this region. Further research can apply methods of data collection such as household surveys,. n. al. i n U. v. which will allow for more sophisticated estimations which will be able to estimate the causal effect (if any).. Ch. engchi. 6.1 Implications Additionally, the research also hopes to serve as a channel with which to communicate to the pertinent authorities about the importance of publishing consistent data estimates – especially on the macro level. Developing economies have a tendency to lack useful household data from official institutions which can be used to study a wide range of different topics. In most cases if the data is available, it is inconsistent with regional estimates or difficult to put into an international context. The implications for policy makers and governmental entities from the cited studies and this one are several: they provide reasonable arguments as to why Microfinance might be worth looking into to resolve issues of social welfare disparity, Microfinance seems to at least be 30.

(34) worth exploring with regards to advancing the completion of the aforementioned MDGs proposed by the United Nations, and if not directly applying Microfinancial models to a potential solution for high levels of poverty the concept of offering equal opportunities to lowresource households is a good first step. It is possible to identify such efforts already being derived from Microfinance, such as the recently popularized Universal Basic Income which seeks to offer a baseline “salary” for all citizens in a given area or region. Further, it is important to note that the repayment rate for these loans is very high, some sources say that it is around 95% (Banerjee, 2015). Although critics of Microfinance claim that this high repayment rate is inflated by clients who take out additional loans in order to repay existing loans – creating a debt cycle in which they cannot get out of. Nevertheless, many advocates of Microfinance appeal to this fact in order to sustain their claims.. 6.2 Limitations. 立. 政治大. For a study looking to identify a causal relationship between Microfinance and social. ‧ 國. 學. welfare, an ideal scenario would rely on surveys which would allow for the implementation of statistical techniques such as a randomized control trial. These surveys would allow researchers. ‧. to design a study around treatment variables and would thusly provide a clearer relationship between changes in variables such as the various measures of poverty or education. The present. Nat. sit. y. research had difficulty in this aspect because of time constraints, funding and physical distance. io. er. from the observations. To the knowledge of the researcher there is also no readily available database which would allow for such a research design to be performed.. al. n. v i n Calso The results from this research towards a specific type of microfinance U h edonnotgpoint i h c. business model (I.E NMP or otherwise) – which means that the marginal impact that the microfinance variables would have do not point towards a certain business model.. 31.

(35) References Augsburg, H. H. (2012). Microfinance, Poverty and Education. Institute for Fiscal Studies. Banerjee, D. G. (2015). The miracle of microfinance? Evidence from a randomized evaluation. American Economic Journal, pp. 22-53. Brooks, C. (2013, April 4). What is Microfinance? Retrieved from Business News Daily: http://www.businessnewsdaily.com/4286-microfinance.html Chemin. (2008). The benefits and costs of microfinance evidence from Bangladesh. Journal of Development Studies, Vol 44, No. 4, pp. 463-484. Coudouel, H. W. (n.d.). Poverty Measurement and Analysis. World Bank Group. Duvendack, J. (2012, April 27). High Noon for Microfinance Impact Evaluations: Reinvestigating the Evidence from Bangladesh. The Journal of Development Studies, pp. 1864-1880.. 政治大 Economic Commission for Latin America and the Caribbean. (2017). CEPALSTAT | 立 Databases and Statistical Publications. Retrieved from ECLAC - Economic. ‧ 國. 學. Commission for Latin America and the Caribbean: http://estadisticas.cepal.org/cepalstat/WEB_CEPALSTAT/Portada.asp?idioma=i. ‧. Grameen Bank. (2015, December). Introduction to Grameen Bank. Retrieved from Grameen Bank: http://www.grameen.com/introduction/. sit. y. Nat. Imai, A. A. (2010). Microfinance and Household Poverty Reduction: New Evidence. World Development Vol. 38, No. 12, pp. 1760-1774.. n. al. er. io. Imai, G. T. (2012). Microfinance and Poverty—A Macro Perspective. World Development, Vol. 40, No. 8, pp. 1675-1689.. Ch. i n U. v. Investopedia. (2017, April). Microfinance. Retrieved from Investopedia: http://www.investopedia.com/terms/m/microfinance.asp. engchi. Khandker. (2005). Microfinance and Poverty: Evidence using panel data from Bangladesh. The World Bank Economic Review, Vol. 19, No. 2, pp. 263-286. Lambert, B. (2013). Retrieved from Ben Lambert Private Tutor: http://oxbridge-tutor.co.uk/ Microinsurance Network. (2017, April 20). Key Concepts. Retrieved from Micro Insurance Network: http://www.microinsurancenetwork.org/microinsurance/key-concepts Multilateral Investment Fund. (2014). Remittances to Latin America and the Caribbean set a new record hgh in 2014. Nghiem, C. R. (2012, April 18). Assessing the Welfare Effects of Microfinance in Vietnam: Empirical results from an quasi-experimental survey. The Journal of Development Studies, pp. 619-632.. 32.

(36) Pitt, K. (1998). The impact of group-based credit programs on poor households in Bangladesh: does the gender of participants matter? Journal of Political Economy, Vol. 106, No. 5, pp. 958-996. Reyna, O. T. (2007). Panel Data Analysis - Fixed and Random Effects using Stata. Princeton University. Reyna, O. T. (2010). Getting Started in Fixed/Random Effects Model using R. Princeton University. Roodman, M. (2009, June). The impact of microcredit on the poor in Bangladesh: revisiting the evidence. Center for Global Development, Working paper No. 174. Socio-economic Database for Latin America and the Caribbean. (2017). Statistics. Retrieved from SEDLAC - Socio-economic Database for Latin America and the Caribbean: http://sedlac.econo.unlp.edu.ar/eng/statistics.php World Bank Group. (2017). World Bank Open Data. Retrieved from World Bank: http://data.worldbank.org/. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. 33.

(37)