microcredit impact assessment: the brazilian and chilean …the brazilian data show a high positive...

100

PANORAMA SOCIOECONÓMICO AÑO 27, Nº 39, p. 100 - 112 (December 2009)

Microcredit Impact Assessment: The Brazilian and ChileanCases

Evaluación del Impacto del Microcrédito: Los Casos de Brasil yChile

Patricio Aroca 1, Geoffrey J.D. Hewings 2*

1Ph.D. IDEAR, Universidad Católica del Norte, Chile, e-mail: [email protected] 2Ph.D. REAL, University of Illinois,Urbana, IL, USA, e-mail: [email protected]*The authors highly appreciated the comments provided by Bill Maloney and Pablo Fajnzylber, the valuable coordinationof the data collection used in this work by Gianni Romani (Chile) and Tales Andreassi (Brazil). We would also like toacknowledge the participation of Leonardo Basso, Andre Ng and Marcelo Lufin in different stages of the project andthe efficient research assistance of Orlando Martinez. The financial support of the Inter-American Development Bankis gratefully acknowledged.

Abstract . Two different sources of data, Brazilian and Chilean banks and NGOs, were accessed to evaluate microcreditprograms. Using propensity score and matching techniques, we compare the average income of individuals whoreceived microcredit with the income of control groups, formed by people with similar characteristics. The results forthe Brazilian data show a high positive impact of microcredit programs, especially for those administered by banks.In the Chilean case the evidence is weaker for the microcredit administered by banks. As for NGO-based programs,the evidence suggests that their impact on the average income of their clients is actually negative.

Keywords : Methods, microfinance, credit, impact assessment, evaluation.

Resumen . Se tuvo acceso a dos fuentes de información, bancos brasileños y chilenos, así como varias ONG, paraevaluar programas de microcrédito. Usando técnicas de Propensity Score Matching (PSM), comparamos el ingresopromedio de los individuos que recibieron microcrédito con el ingreso de grupos de control formado por gente concaracterísticas similares. Los resultados con los datos brasileños muestran un impacto positivo importante de losprogramas de microcrédito, especialmente para aquellos administrados por bancos. En el caso chileno, la evidenciaes más débil para los microcréditos administrados por los bancos. En cuanto a los programas basados en las ONG,la evidencia sugiere que su impacto en el ingreso promedio de sus clientes es en realidad negativo.

Palabras clave : Métodos, microfinanzas, crédito, evaluación de impacto, evaluación.

(Recibido: 7 de octubre de 2009. Aceptado: 9 de diciembre de 2009)

INVESTIGACIÓN / RESEARCH

101

Evaluación del Impacto del Microcrédito: Los Casos de Brasil y Chile

Patricio Aroca, Geoffrey J. D. Hewings

INTRODUCTION

The award of the Nobel Peace Prize toBangladeshi economist Muhammad Yunus and theGrameen Bank that he created focused attention onthe role of micro-credit in the economic developmentof many second and third-world countries. As Moore(2006) noted: “The Grameen Bank’s pioneering useof micro-credit has been duplicated across the globesince Yunus started the project in his home villagethree decades ago. Loans as low as $9 have helpedbeggars start small businesses and poor women buycellular phones and basket-weaving materials”.

The award highlights the need to consider moreextensively the fact that entrepreneurship is not theexclusive purview of either large companies, or busi-ness people in western economies but can beconsidered to be a more widespread phenomenon,often targeting thousands of people whose economicimpact might be limited when considered in isolation– but aggregated across individuals, the impacts canoften be outstanding (It should be noted that a Chicagocommunity bank, ShoreBank, that had pioneeredloans to low-income communities, was recruited byYunus in 1983 to help him set up Grameen Bank.Grameen Bank now has over 6 million borrowers whohave accessed loans totaling almost $6 billion with a98% repayment rate). There is increasing interest inmeasuring the impact and viability of microcreditprograms. The evidence to date is not uniformlyencouraging, notwithstanding the Bangladeshexperience, as Sebstad and Chen (1996) describe intheir survey of the impact of microcredit programs:“Income increased for at least half of the enterprisesin most of the studies, while it remained the same oreven declined for a significant proportion”. Results areambiguous and many programs are kept alive onlyby the injections of government or sponsor subsidies(Schreiner and Yaron, 2001).

Amin et al. (2003) report the results from researchin northern Bangladesh. They distinguish among thepoor between vulnerable and non vulnerable,concluding that in the richer villages the microcreditprograms do not reach the vulnerable while in thepoorer ones they are excluded from the programs. Inboth case, measuring the impact of the microcreditprogram could be biased for the reasons articulatedby Coleman (2006), namely, self selection of theborrowers and endogenous program placement. Thus,microcredit does not reach the vulnerable poor in therelatively richer village B, and appears to exclude thevulnerable poor in the poorer village A. In other words,poor households that do join tend to have betteraccess to insurance and smoothing devices than those

who do not. The vulnerable poor may either choosenot to join or they may be excluded by the microcreditprogram.

What has been the experience in Latin America?Berger (2006) makes the following observation:

“In a region of great inequality and economicinstability, microfinance is a capitalist paradox. Inthe past 20 years, microfinance has gone from anobscure development to a multibillion dollarenterprise bringing banking to millions of people.Although the industry has grown globally and thereare star performers in every region, institutions inLatin America stand out for their integration intothe formal financial system and their impressivegrowth, outreach and profitability indicators.”Drawing on Miller and Martínez (2005), she notes

that by the end of 2004, 80 of the top microfinanceinstitutions (both NGO and formal financial institutions)in this continent were serving 4 million clients with aloan portfolio valued as $4 billion. These areimpressive numbers but also generate some importantquestions. The most important obviously focuses onthe success of the program; but here, there are severaloptions that could be considered as appropriatemetrics – the volumes of loans, the profitability of thesystem, the percentage of loans repaid, jobs andbusinesses created by the program and so forth. Oneimportant criterion that seems to not to have receivedmuch attention is the degree to which recipients ofthe program are better off than non recipients.Obviously, in this case, there are a variety ofappropriate welfare measures that could be employed,for example, did the microloan recipient successfullylaunch a new venture or extend or enhance an existingone? In this paper, we follow the suggestion of Aghionand Morduch (2005) who focused on the causal impactof microfinance on borrower income; however, as theynote, there are many other measures that may beequally important (extending the viability of theenterprise, increasing the productivity of the business,or generating positive spillover effects on otherenterprises).

One of the major challenges of any programevaluation that involves a subset of a population is tomeasure success of the program participants incomparison to a control group of non-participants. Inmost cases, data are only available about the programparticipants, generating a need to build the controlgroup from nonsurvey information. The objective ofthis paper is to evaluate the impact on micro-entrepreneurs income of two Brazilian and Chileanmicrocredit programs. Drawing on two unique sourcesof data, control groups are built using the propensityscore to match beneficiaries of micro-credit programs

102


with non-beneficiaries with similar characteristics.Then, the average incomes are compared. The resultsfor the Brazilian data show a high positive impact ofthe microcredit programs, especially for thoseprograms administered by banks. In the Chilean case,the evidence is weaker for bank-based programs andNGO (non-governmental organization) basedprograms appear to have no positive impact at all.

In the next section, the theoretical underpinning ofthe microfinance programs will be explored, togetherwith a focus on the challenges facing small-scaleentrepreneurs in Latin America. The need to developa control group is presented along with themethodology of propensity score and matchingestimator is explained in the following section. Thesecond and third section presents the results from theChilean and Brazilian cases. The final sectionsummarizes our main conclusions and offers somepolicy evaluation.

The Microfinance Program: Economic Foundationsand Challenges

An emerging entrepreneur in the United States canapproach a financial institution with an idea for a newbusiness; she will be encouraged to develop a busi-ness plan and asked to consider various forms ofcollateral so that a loan can be underwritten. In manycases, the collateral offered may be in the form of anadditional mortgage on her house or other property;in very exceptional circumstances, a loan may begranted without significant collateral if the businessconcept has been through a successful experimental“proof of concept” phase. Without collateral, theentrepreneur may be forced to approach venturecapitalists that, for a share in the profits, might bewilling to underwrite the new business. Theseexplorations assume an efficient market whereinformation is easily accessed, the process is quiterational and the outcome can usually be predictedrather well. Now consider an entrepreneur in LatinAmerica with few tangible assets (i.e. little or nocollateral). As Goldmark (2006) reminds us, earliermicrofinance programs often offered a “bundle” ofservices that extended beyond the loan itself, suchas training programs to help clients managebusinesses and money. While there would be areasonable expectation that the US entrepreneurmight possess some of these skills (such as familiaritywith a prior loan), there was little expectation of thisbeing the case with the majority of microfinance clientsin Latin America. However, the more recent experiencesuggests that microlenders have moved away frommore extensive training programs to focus on the credittransaction itself.

In essence, Goldmark’s (2006) concerns parallelthose of Aghion and Morduch (2005) who aim tobroaden the conversation about the role and impactsof microfinance and to address some of the prevailingmyths. For example, they argue that microfinanceneeds to be complemented by providing better waysfor low-income households to save and insure. Thechallenge the received wisdom that microfinance hasa clear record of social impacts (e.g. povertyreduction); they note:

“Relatively few rigorous studies of impacts havebeen completed, and the evidence on statisticalimpacts has been mixed so far. There is not yet awidely acclaimed study that robustly shows strongimpacts, but many studies suggest the possibility.Better impact studies can help resolve debates…”(Aghion and Morduch, 2005, p.4).In addition, Aghion and Morduch (2005) address

the question of why capital does not naturally flow tothe poor. To begin with a simple binary example, thinkof an enterprise with little capital and one withextensive capital. Given a concave productionfunction, the rates of return from investment in thesmall capitalized firm should be higher than the largerone; however, empirical evidence does not revealcapital flowing disproportionately to poorerentrepreneurs. To the contrary, issue of risk,government rile (such as usury laws) proven a marketoperating since, ceteris paribus, given the expectedlarger marginal returns for the poorer entrepreneur,she should be capable of paying a higher interest rate.There are additional problems – such as adverseselection that may result in banks charger a riskpremium than may drive customers with the ability topay out of the market or the issue of moral hazardthat posits the problem of clients not making therequired effort to make their projects successful. A fi-nal problem may stem from poor enforcement of le-gal contracts. Collecting the requisite informationabout clients in poorer, rural areas presents a dauntingproblem; microfinance is seen as an option that mayreduce risk and costs (such as information collection)for lenders, encourage broad participation andultimately offer benefits to both lender and client.

Aghion and Morduch (2005) also advance the ideaof a shift in nomenclature, from microcredit tomicrofinance; while the two are often regarded assynonyms, they argue that the former refers to loanswhile the latter embraces part of what Goldmark (2006)would term “business services” – collecting savings,providing insurance or even assisting in the marke-ting of products.

The issues raised here are extensive; our accessto the available data precludes our addressing many

103



of them but we feel, following Aghion and Morduch’s(2005) comments about the dearth of impact studies,that it is important to at least begin the analysis of theeconomic impacts at the individual level. Hence, thepurpose of this paper is very simple: is it possible todemonstrate that recipients of microfinance werebetter off (in terms of income) than non recipients?As we noted in the introduction, formal survey dataare only available for recipients; our task was to createa control group using nonsurvey methods. Themethod adopted is described in the next section.Essentially, the goal was to define two groups –recipients and non-recipients of microcredit – whowere “identical” in terms of a set of non-income relatedcharacteristics thus affording the opportunity to claimthat differences in income could be attributed to themicrocredit program.

METHODOLOGY

Impact assessment requires a group affected bythe program intervention (microcredit recipients), anda control group (non recipients) to compare theoutcomes. Then, the differences between the twogroups will provide an important component of thetotal impact of the program. Hulme (2000) notes that:

“Impact Assessments assess the difference in thevalues of key variables between the outcomes on«agents» (individuals, enterprises, households,populations, policy-makers, etc) which haveexperienced an intervention against the values ofthose variables that would have occurred had therebeen no intervention.The fact that no agent canboth experience an intervention and at the sametime not experience an intervention generatesmany methodological problems.”In our study, the intervention will be the microcredit

program. However, one of the main obstacles toassessing the impact is finding or building theadequate control group. Mosley (1997) suggestsseveral different alternatives, although all of thempresent serious limitations of the kind discussed byHulme (2000).

Coleman (2006) described two kinds of bias in themeasurement of a microcredit program impact whenthe control group is not properly chosen. The first isassociated with the comparison of people with initiallydifferent characteristics such as education, skill,wealth, etc., that might lead to the identification ofdifferences that are not necessarily caused by themicrocredit program. He also noted a tendency formicrocredit to go to people with some particular setsof skills. A second source is associated with themicrocredit institutions that chose the village in which

to launch their microcredit program, ones that mayhave more entrepreneurship capacity installed thatincrease the success of the programs (endogenousprogram placement). Coleman (2006) addressedthose bias issues in his study through the surveydesign and sample process.

In this paper, we will use a methodology that wasinitially applied in the health literature and has recentlybeen adapted by a research group led by Heckman(Heckman et al., 1997 and 1998) for evaluatingprograms in economics. It addresses the problemsdescribed by Coleman (2006) and it is especiallysuitable when there is a data base available(household survey) from which to build the controlgroup in a way that significantly reduces the researchcost. This methodology is known as a matchingestimator and it is based on Rosenbaum and Rubin(1983). The basic idea is that using a set of similarattributes (X: characteristic, variables or regressors)for two groups of people, one subject to the treatmentand the other not; a propensity score for each indivi-dual in each group can be calculated. Then, thebalancing property of the propensity score makes itpossible to obtain the same probability distribution ofX for treated and non-treated individuals in matchedsamples. Following Sianesi (2001), we will evaluatethe causal effect of microcredit programs on householdincomes of the participants, relative to a constructedcontrol group, which has not received credit. Note thatthe evaluation is limited to income – other metrics couldnot be compared (such as job creation, success inmarketing, participating in supplier networks etc.).

Let Y1 be the outcome that would result if the indi-vidual receives microcredit and Y0 the outcome thatwould result if the same individual does not receivemicrocredit. Let D = {0, 1} denote the binary indicatorof microcredit (D = 1 if microcredit, 0 otherwise). For agiven individual, i, the observed household income isthen Y

i = Y

0i + D

i (Y

li –

Y

0i). In addition, assume that X,

the set of attributes, is not affected by the microcreditprogram.

As Hulme (2000) noted, no individual can bothreceive and not receive microcredit at the same time,so that either Y

li or Y

0i is missing for each i. Since it is

impossible to observe the individual microcredit effectand thus to make causal inference without makinggenerally un-testable assumptions, we attempt toidentify the average treatment effect in the population,or in a sub-population, which requires generally lessstringent assumptions.

Thus, following Heckman (1997 and 1998) andSianesi (2001), we could attempt to identify thefollowing parameters:• The average treatment effect: E(Y

1 –

Y

0)

104


is the average income difference between the twogroups: the micro-entrepreneurs that receivemicrocredit and the rest that does not.• The average treatment effect on the non-treated: E(Y1 – Y0ID=0) is the average incomedifference between the potential or expected incomethat the entrepreneurs who did not receive microcredit(D=0) would get if they had [E(Y1)] and the real incomethat they earned (Y

0).

• The average treatment effect on the treatedE(Y1 – Y0ID=1). This parameter is the one receivingthe most attention in the evaluation literature andmeasures the average income difference between theincome that the entrepreneurs earned who receivedmicrocredit and the income that they would obtainedif they had not received credit.

Two unknown values: E(Y1ID=0) and E(Y0ID=1)prevent direct inference. Therefore, we need to makeestimates based on some usually un-testableidentifying assumptions that justify the use of theobserved E(Y

1ID=1) and E(Y

0ID=0) .

Sianesi (2001) notes that treated individuals maynot be a random sample of the population, but theymay receive treatment on the basis of characteristics,which also influence their outcomes. For example,microcredit institutions may try to pick out the bestcandidates such that micro-entrepreneurs who receivemicrocredit are of better quality than the rest; hence,these individuals could be expected to perform well,generating selection bias. This would result in an over-estimate of the impact of the microcredit program:E(Y

1ID=1)–E(Y

0ID=0) would in general be an upward-

biased estimate of the effect of treatment on thetreated.

Statistical matching offers a way to construct acontrol group to partially address the selection biasissue (see Sianesi 2000). Rosenbaum and Rubin(1983) show that treatment and non-treatmentobservations with the same value of the propensityscore have the same distribution as the full vector ofregressors. It is thus sufficient to match exactly onthe propensity score to obtain the same probabilitydistribution of the explanatory variables for treated andnon-treated individuals in matched samples.In the next two sections, we will review, in turn, theChilean and Brazilian cases.

THE CHILEAN CASE

In order to obtain the information about the Chileanmicro-entrepreneurs, we developed a questionnairefollowing the methodologies developed by Barnes(1996), Chen (1997), Hulme and Mosley (1996),Hulme (2000), Mosley (1997), Sultana and Nigam

(1999) and Tsilikounas (2000). The survey was run inFebruary and March of 2002. The sample wasobtained randomly from the databases of the bankBandesarrollo and the NGO Propesa: respectively 56observations for Antofagasta (II Region) and 30observations for Melipilla (Metropolitan Region).

The CASEN (the Chilean survey of national so-cio-economic characterization) for the year 2000 wasused to build the control group. This survey was runin November 2000. A set of variables from bothsurveys (location, age and employment status) wasused to identify the control group from CASEN. Weselected people who were living in Melipilla orAntofagasta, who were employees and who wereolder than 17 and younger than 66. The final controlgroup sample contained 715 observations. Aftercleaning the sample extracted from the Bandessarollodatabase, we were left with 81 cases of “treated”individuals leaving us with a total of 796 observations.The variables dictionary for both groups is shown inTable 1.

The set of variables was chosen according to thematching techniques’ requirements. First, it wasimportant that these variables should not be affectedby the microcredit program. With the exception ofincome, all the other variables comply with thisrequirement. To build the control group, we estimatedthe propensity score as a function of those variables.We used a probit specification and the results areshown in Table 2 for the total sample, the bank clients(Bandesarrollo) and the NGO clients (Propesa). Tomeasure the impact of the microcredit program wecompare the average income of the people whoreceived microcredit with the average income of the“similar” people who did not receive microcredit in theconstructed control group: E(Y

1–Y

0ID=1).

The results shown in Table 2 are as expected andwith a high value for the explained variance for thesekinds of models. In general, older women who do nothave a husband, with some education and householdheads are the ones with the highest probability ofreceiving microcredit. However, there are somesignificant differences between the bank and NGOclients. While the bank clients who belong to theproductive or service sectors have a higher propensityto receive microcredit, the NGO clients that belong tothe same sectors have a lower than averagepropensity to receive microcredit. This fact is reflectedin the sign change of the coefficients on the variablesproductive sector (spro) and service sector (sserv) forthe bank clients (positive) and the NGO clients(negative).

Evidence for the quality of the matching for thetotal sample, the bank clients and the NGO clients is

105

presented in Appendix A, where the mean for eachvariable is calculated for the treated and the controlgroups, for the whole sample and for the matchedsub sample. The difference between the treated andcontrol group is lower for the matched sub samplethan for the whole sample. The results further showthat the matching procedure worked better for thewhole sample and for the NGO clients, while for bankclients, the improvement from matching were not assignificant. The main explanation is that even thoughthere was some improvement in matching the indivi-dual according to the chosen variables, the matchingof household head and household head spouse wasnot good.

The impacts of the microcredit programs are shown

at the end of Table 2 . The first two results are positivealthough not statistically significant. The microcreditprogram as a whole has a positive impact on the ave-rage income of the micro-entrepreneurs. This positiveimpact means that those receiving microcredit earnon average 25% more than those that did not. Thisimpact is about 38% (Ch$119,432 or about US$ 220)for those who received credit from the Bandesarrollo.Those receiving credit from the NGO program appearto have a negative and significant impact on incomesof about 50%.

Alternatives explanation can be given for thisfinding. First, it could result from bias in the quality ofthe matching. Appendix A shows that we only obtainpositive results from the worst matching.



106


THE BRAZILIAN CASE

The Brazilian data were collected in February andMarch 2002. Five institutions provided informationabout their micro-entrepreneur clients: Microcred(bank from São Paulo), Socialcred (bank from Rio deJaneiro), CEAPE (NGO from Goias), Bancri (NGOfrom Santa Catarina) and Bco Povo Sto Andre (NGOfrom São Paulo). In addition, the information collectedby PNAD (Brazilian National Survey on Households)1999 was used to build the control group.

Table 3 shows the variables that were used in theanalysis. These variables are not exactly the same

as the ones used in the Chilean case because theChilean CASEN uses a different questionnaire thanthe Brazilian PNAD. However, we tried to provideproxies for the same concepts, even if measured bydifferent variables. The only variable that was notavailable in the PNAD is marital status. In addition,for hours worked by week and number of employeesin the firm, PNAD provides ranges and not number ofhours and workers respectively. It is worth noting,however, that PNAD does provide information on va-riables that are not available in the Chilean case, suchas the location of the micro-firm (home or outside).

In total, we were able to use 198 observations from

107

micro-entrepreneurs of five different states that hadreceived microcredit, either from NGOs (152) or frombanks (46). For the control group, we selected 34.887observations from PNAD, for the same five states.

In order to make income comparisons acrosstreatment and control groups, we need to take intoaccount the different reference periods for the incomeinformation available for microcredit clients (January2000) and the control group from PNAD (September1999). The price variation in Brazil during these 4months is shown in Table 4: income differences ofthat order of magnitude across treated and controlgroups can be explained purely by price changes.

The large sample collected in Brazil allowed forbetter comparisons than in Chile. We estimated theeffect of microcredit programs using the total sampleand comparing the entrepreneurs that receivedmicrocredit, first, with the whole set of employers andself-employed in the PNAD sample, and secondly withthe PNAD sample of salaried workers.

We are aware of the possibility that some of theemployer or self-employed individuals in the controlgroup may also be clients of a microcredit program.However, we do not have the information required torecognize them. Therefore, when the PNAD sampleof individuals in the employer and self-employed sta-tus (we will hereafter use the term employer to referto both employers and self-employed individuals) isused as a control group, we can expect the differencebetween treated and non-treated to be a downwardlybiased estimate of the impact of the microcreditprogram hereby considered. On the other hand,estimates based on the control group built using onlysalaried workers from the PNAD sample couldoverestimate the effect of the microcredit programbecause those workers may lack the entrepreneurialskills that account for a share of the income ofindividuals in the treatment group.

Table 5 shows the results for the three alternativecontrol groups (total PNAD sample, employers only



108


and salaried workers only), while Appendix B presentsevidence on the respective quality of the matching.From Appendix B, we see that in all three cases thematching procedure generally leads to significantreductions in the average differences among treatedand control groups. This is especially the case whenthe matching is performed using the PNAD sample ofemployers as a control group. It is worth noting,however, that although gains are considerable for mostvariables, there are some variables for which avera-ge differences across the two groups actually increaseafter the matching.

In contrast to our results on Chile, the microcreditprograms examined in Brazil appear highly effective,with high and statistically significant increases in theaverage income of their clients. If we adjust the meanincome in the PNAD sample for the inflation betweenSeptember 1999 and January 2000 (see Table 4 ),the difference between entrepreneurs who receivedmicrocredit and other entrepreneurs or workers withsimilar characteristics is still above 100%. When wecompare the estimates obtained with control groupsof, respectively, employers and salaried workers, wefind, as expected, that the impact of microcredit islower in the first case. However, this impact is stillhigh (R$ 1,351 or about US$ 350 of the averagemonthly income) and highly significant (t - statistic =5.68).

As in the Chilean case, we compare the differencesbetween those who received microcredit from a bank-based program to those who were clients of an NGO-based program. Results are shown in table 6 for bothtypes of control groups (employers and salariedworkers). Before further commenting on these results,note that the average income of bank clients is about20% higher than that of NGO clients. As for thematched control groups, in the case of bank-basedprograms, the average income of the employers con-trol group is 80% higher than that of the salariedworkers control group (R$ 934 versus R$ 516). This

is different from the NGO case, where the averageincome of the employers matched control group islower than that of the salaried workers control group(R$860 and R$1,053, respectively). Moreover, theaverage income of the employers matched controlgroup is higher for bank- than for NGO-basedprograms, while the opposite is true whencomparisons are made across salaried workers con-trol groups. The main conclusion arising from theseresults is that banks and NGOs have very differentclients in terms of their incomes, a finding that wasalso observed in Chile.

Although in the case of the control groups usedfor bank clients, salaried workers earn much less thanemployers with similar characteristics, the behaviorof average incomes in the matched control groupsconstructed for the sample of NGO clients is verydifferent. While the average income of employersmatched to NGO clients is 9% lower than the avera-ge income of employers matched to bank clients, theaverage income of salaried workers matched to NGOclients is larger than the average income of employersmatched to NGO clients by 22%. However, this lastresult is not surprising if one takes into considerationthe educational level and number of workers in thefirm, both of which are positively and significantlyrelated to the propensity to receive microcredit in thesub-sample of salaried workers.

One characteristic of the Brazilian labor market isthat people tend to have more than one job to increasetheir income. On the other hand, micro-entrepreneurstend to over-estimate the number of hours that theyreally work, especially in the case when they use theirhome as the business location. These elements havetwo implications for our study. First, we decided tocompare average monthly income rather than avera-ge hourly income. Secondly, we re-estimated avera-ge income of the control group using all the incomesources of salaried workers and employers.

Table 7 shows the results of comparing the income

109

of the micro-entrepreneurs that received micro-creditwith the total income, from all sources, of matchedsalaried workers and employers. Although the

microcredit program effect is lower than the one shownin Table 6 , it is still high and significant both for clientsof bank- and NGO-based programs.



110


111

CONCLUSIONS

We estimate the income impacts of two Chilean andfive Brazilian microcredit programs, attempting toprovide a small impacts study to contribute to thedearth of such studies noted by Aghion and Morduch(2005). We evaluated the bank and the NGO programsseparately because the previous literature suggeststhat they address different shares of the market. Weuse a relatively new method to build control groups,in order to deal with some of the more seriousproblems that have plagued previous assessmentsof microcredit programs. In many cases, the programwas evaluated in terms of metrics that focused on thevolume of loans, number of clients, repayment ratesand so forth but did not address a fundamental issue– did the program make a difference in the incomesof recipients in comparison to the incomes of non-recipients?

We find weak evidence of positive impacts for theChilean bank-based program. As for the Chilean NGOclients, it seems that the impacts of microcredit onincome are not positive but negative. On the otherhand, the Brazilian evidence shows a highly positiveand significant impact of microcredit programs onclients’ income, especially in the case of thoseadministered by banks. The next logical steps wouldbe to assess the impacts of these programs ondevelopment more broadly defined – beyond thecalculation of just the impacts on clients’ incomes. Didthey create sustainable enterprises that in turnemployed innovative methods/techniques to developproducts or services that extended to the communityas a whole? Did the successes, such as those in theBrazilian case, generate a milieu that encouragedothers to assume the risk of obtaining a loan? Werethese entrepreneurs able to participate in formalsupply chains in marketing their goods or services?Entrepreneurial development at this level is likely tobe modest in scope when viewed from the perspectiveof the individual micro-loan, but as noted in theBangladeshi case, summed over thousands or millionsof clients, the impacts begin to accumulate to reachrather impressive levels.

Finally, a one-time evaluation of a program needsto be complemented by an on-going monitoringprocess: Were the income gains noted in Brazil forprogram participants sustained over time? To whatdegree does location play a role – Are there criticalcommunities whose participation and success in theseprograms lead to a greater and more successfulevolution (in terms of spread)? There are other issuesthat need to be considered – especially the role of thesuppliers of credit – before a full evaluation of the

impacts can be assessed.

REFERENCES

AGHION B.A., MORDUCH J., 2005. The economicsof microfinance. Cambridge, MA., MIT Press.

AMIN S., ASHOK S.R., TOPA G., 2003. Doesmicrocredit reach the poor and vulnerable?Evidence from northern bangladesh. Journal ofDevelopment Economics 70, 59-82.

BARNES C., 1996. Assets and the impact ofmicroenterprise finance programs. Washington,DC: Management Systems International.

BERGER M., 2006. The Latin American model ofmicrofinance. In BERGER M., GOLDMARK L.,MILLER-SANABRIA T., (ed.) An inside view of LatinAmerican microfinance. Inter-AmericanDevelopment Bank, Washington, DC. pp.1-36.

CHEN J., 1997. A guide for assessing the impact ofmicroenterprise services at the individual level.Washington, DC: Management SystemsInternational.

COLEMAN B.E., 2006. Microfinance in northeastthailand: who benefit and how much? WorldDevelopment 34, 1612-1638.

GOLDMARK L., 2006. Beyond finance: microfinanceand business development services. In BERGERM., GOLDMARK L., MILLER-SANABRIA T., (ed.)An inside view of Latin American microfinance.Inter-American Development Bank, Washington,DC. pp. 193-234.

HECKMAN J.J., ICHIMURA H., TODD P.E., 1997.Matching as an econometric evaluation estimator:evidence from evaluating a job training programme.Review of Economic Studies 64, 605-654.

HECKMAN J.J., ICHIMURA H., TODD P.E., 1998.Matching as an econometric evaluation estimator.Review of Economic Studies 65, 261-294.

HULME D., 2000. Impact assessment methodologiesfor microfinance: theory, experience and betterpractice. World Development 28, 79-98.

HULME D., MOSLEY P., 1996. Finance againstpoverty. London: Routledge, 1-2.

MILLER J., MARTÍNEZ R., 2005. Banking:championship league. MicroEnterprise Americas.pp. 5-11.

MOSLEY P., 1997. The use of control groups in impactassessment for microfinance mimeo. Washington,DC: Consultative Group to Assist the Poorest(CGAP) and also (2001) Working Paper 19, ILO,Geneva.

MOORE M., 2006. Micro-credit pioneer wins peaceprize economist, bank brought new opportunity topoor. Washington Post Foreign Service Saturday,



112


October 14, p. A01.ROSENBAUM P.R., RUBIN D.B., 1983. The central

role of the propensity score in observational studiesfor causal effects. Biometrika 70, 41-55.

SCHREINER M., YARON J., 2001. Developmentfinance institutions: measuring their subsidy. TheWorld Bank, Washington D.C., USA.

SEBSTAD J., CHEN G., 1996. Overview of studieson the impact of microenterprise credit. Papersubmitted to Assessing the Impact ofMicroenterprise Services (AIMS), WashingtonD.C., USA.

SIANESI B., 2001a. Implementing propensity scorematching estimators with STATA. UK Stata UsersGroup, VII Meeting London, May 2001.

SIANESI B., 2001b. An evaluation of the active labourmarket programmes in Sweden. Working Paper2001:5, Office of Labour Market Policy Evaluation,Sweden.

SULTANA M., NIGAM A., 1999. Impact assessmentstudy of the family development fund, Egypt.UNICEF, New York, USA.

TSILIKOUNAS C., 2000. Impact survey. ICMC andProject Enterprise. Bosnia and Herzegovina.

microcredit impact assessment: the brazilian and chilean …the brazilian data show a high positive...

Documents