Great Expectations: Microfinance and Poverty Reduction in Asia and Latin America

Microfinance institutions (MFIs) are often seen by aid practitioners as a manifestly effective means of improving the position of the poor. Despite this widely held view, detailed research studies have been much more guarded about the impact of MFIs. In particular, several studies have raised doubts about the effectiveness of MFIs in reaching the “core poor”. This paper surveys the evidence from Asia and Latin America and contrasts experiences in the two regions. Studies on the former have been carried out more “rigorously”, but in both regions the evidence that microfinance is reaching the core poor is very limited.


I. Introduction
The microfinance revolution has changed attitudes towards helping the poor in both Asia and Latin America and in some countries has provided substantial flows of credit, often to very low-income groups or households, who would normally be excluded by conventional financial institutions. Much has been written on the range of institutional arrangements pursued in different organizations and countries and in turn a vast number studies have attempted to assess the outreach and poverty impact of such schemes. However, amongst the academic development community there is a recognition that perhaps we know much less about the impact of these programs than might be expected given the enthusiasm for these activities in donor and policy-making circles. To quote a recent authoritative volume on microfinance "MFI field operations have far surpassed the research capacity to analyze them, so excitement about the use of microfinance for poverty alleviation is not backed up with sound facts derived from rigorous research. Given the current state of knowledge, it is difficult to allocate confidently public resources to microfinance development." (Zeller and Meyer 2002). This is a very strong statement of doubt and in part reflects lack of accurate data, but also in part methodological difficulties associated with assessing exactly what proportion of income and other effects on the beneficiaries of micro credit can actually be attributed to the programs themselves. Here we compare poverty impact studies from Asia and Latin America. In particular we ask what is the evidence on three specific issues the success of microfinance programs in reaching the core poor the effectiveness of microfinance initiatives in pulling households out of poverty the cost effectiveness of microfinance as a poverty targeting tool These are very basic questions and the fact that they can still be posed reflects the extent of uncertainty in the literature. Since a number of other surveys are also available we give most attention to evidence produced in the last three or four years 1 and highlight similarities and differences in microfinance as it has developed in Asia and Latin America.
The paper is organized as follows. We first provide a brief overview of some of the distinguishing characteristics of the microfinance industry in Asia and Latin America. Section three discusses the potential for microfinance to combat poverty and methodological issues relating to assessing its success in doing so, and section four goes on to survey the evidence from selected research studies on this point. Section five addresses the question of cost-effectiveness. Finally we draw some brief conclusions.

II. Characteristics of Microfinance in Asia and Latin America
Microfinance developed in Asia and Latin America under very different ideological, political and economic conditions. Hence, there are distinctive differences in the microfinance industry in the two regions. A brief look at the history of two of the most famous MFIs: the Grameen Bank in Bangladesh and Banco Sol in Bolivia, gives an informative picture of how the industry in the two regions can be characterized.
Modern microfinance was born in Bangladesh in the 1970s, in the aftermath of the country's war of independence, when Muhammad Yunus, an economics professor at the University of Chittagong, began an experimental research project providing credit to the rural poor of Bangladesh. That experiment driven by a strong sense of developmental idealism developed into what is now the world's most famous microfinance institution, the Grameen Bank, and institutions that replicate its pioneering methodology worldwide.
Microfinance in Latin America developed under quite different conditions. In Bolivia, a collapsing populist regime led to widespread unemployment. Banco Sol, a pioneering microfinance institution in the region, developed to address the problem of urban unemployment and provide credit to the cash-strapped informal sector. The notion of commercial profitability was embraced relatively early in this approach.
As a result of the different conditions under which the very first microfinance institutions were founded, the industry in the two regions developed distinctive characteristics. In the beginning, " by comparison with Bangladesh, the Bolivian intervention was typically urban rather than rural, less concerned with poverty and more focused on microenterprise. It targeted the 'economically active poor' -people with established businesses that needed capital to grow. … from the start, Bolivian microcredit was itself seen as a business, potentially as a branch of commercial banking " (Rutherford (2003) p.5). Many of these differences still characterize the industry in the two regions today.
For example, data from various sources suggest that Asian MFIs lead the world in terms of both breadth (number of clients) and depth (relative poverty of clients) of outreach. In their analysis of over 1,500 microfinance institutions from 85 developing countries, Lapeneu and Zeller (2001) find Asia accounted for the majority of MFIs, retained the highest volume of savings and credit, and served more members than any other continent. The most recent data from the Microbanking Bulletin 2 , reinforces these findings. Average size of loans and deposits are often taken as a simple proxy of depth of outreach. By this criteria Asian MFIs have among the lowest Loan and Savings Balance per Borrower, even after adjusting for GNP per capita, suggesting that they are effectively reaching the poor. The same data indicates that Latin American MFIs are ahead of Asia in terms of financial viability. On average, Latin American MFIs registered with the Microbanking Bulletin show a higher return in Asia. Latin America MFIs are also further advanced in the process of drawing in external funding through savings deposits with registered MFIs on average in the region have a deposit-loan ratio of 29%, which is roughly double the comparable figure for Asia (Ramirez 2004).
Regional data of course covers up some wide disparities within each region. Microfinance is highly concentrated industry and the giants of the industry -BRI, BRAC and ASA -account for more than 50% of the total number of borrowers from the more than 300 MFIs worldwide, who report to the MIX Market. BRI alone accounts for nearly 40% of their gross loans. Within Asia, Bangladesh, Indonesia, Thailand and Viet Nam have the largest number of members served and the largest distribution of loans and mobilization of savings in terms of GNP in the world. In contrast, the two most populated countries in Asia, India and the PRC, have very low outreach, despite a high concentration of the region's poor. In Latin America, there is very strong skew with MFIs playing a major role as financial providers to micro-enterprise in Bolivia and Central America, but being largely insignificant in the larger countries of Brazil, Mexico and Argentina. There is wide disparity in terms of financial viability as well. Within Latin America there is a wide range, with the larger MFIs showing a return on assets in 2001-02 well above the average for the commercial banking sector in their countries and that of the smaller MFIs in the region, which on average operate at a substantial financial loss when capital costs are calculated at commercial rates.
The strong financial performance of larger MFIs in Latin America is linked with a trend toward commercialization of microfinance in the region. In 1992 Banco Sol became the first example of an NGO transformation to a commercial bank and thus became the first regulated microfinance bank. Banco Sol surpassed other Bolivian banks in profitability and became the first MFI to access international capital markets. Following this successful example, at least 39 other important NGOs worldwide transformed into commercial banks over the period 1992(Fernando (2003). Given that the failure of commercial financial institutions to reach the poor provided the initial impetus for MFIs, this new trend is paradoxical and raises the question of whether the initial poverty reduction objectives of the transformed NGOs will be subjugated to commercial criteria (so-called 'mission drift'). This potential disadvantage is still unexplored empirically, but the advantages of transformation are clear: increased access to funding and regulatory authority freeing the institutions from dependence on donor-funds and capital constraints on growth and allowing them to offer a wider range of financial services.
There is also a recent trend in the opposite direction -traditional banks getting involved in microfinance in a variety of ways. In both regions therefore we see similar trends towards a provision of a wider range of financial services, a move away from traditional group lending to individual loans, and in summary a greater shift towards commercialization of the sector, with Latin America more advanced in this process. However in both regions NGOs remain important providers and in Asia they are still the dominant mode of delivery. The NGO sector is still, with exceptions, not financially sustainable and continues to rely on subsidies of various sorts. In these circumstances, of what seems a fragmenting MFI sector in many countries with a division between NGO-based lending and a commercially-driven banking operation, there is a strong need for studies that shed light on the poverty consequences of different modalities. If NGOs are to continue to draw on subsidized finance there is a need to demonstrate that they can reach the poor and do so in a cost-effective manner, as compared with other forms of poverty targeting interventions. If public policy is to encourage the transformation of NGOs to regulated financial institutions or if the delivery of small loans is to be left to the commercial banking sector, the concern that the client base will change so that poor clients are excluded by application of tighter commercial criteria must be addressed. In such instances there is a need to learn more about the poverty consequences of the ongoing changes in the MFI sector in many countries.

III. Poverty and Microfinance
Here we define poverty as an income (or more broadly welfare) level below a socially acceptable minimum and microfinance as one of a range of innovative financial arrangements designed attract the poor as either borrowers or savers. In terms of understanding poverty a simple distinction can be drawn within the group 'the poor' between the long-term or 'chronic poor' and those who temporarily fall into poverty as a result of adverse shocks, the 'transitory poor'. Within the chronic poor one can further distinguish between those who are either so physically or socially disadvantaged that without welfare support they will always remain in poverty (the 'destitute') and the larger group who are poor because of their lack of assets and opportunities. Furthermore within the non-destitute category one may distinguish by the depth of poverty (that is how far households are below the poverty line) with those significantly below it representing the 'core poor', sometimes categorized by the irregularity of their income. In some Latin American cases for example the core poor or destitute are taken to be those below 50% of the poverty line (although Latin American poverty lines are generally higher than in Asia) In principle, microfinance can relate to the chronic (non-destitute) poor and to the transitory poor in different ways. The condition of poverty has been interpreted conventionally as one of lack of access by poor households to the assets necessary for a higher standard of income or welfare, whether assets are thought of as human (access to education), natural (access to land), physical (access to infrastructure), social (access to networks of obligations) or financial (access to credit) (World Bank 2000:34). Lack of access to credit is readily understandable in terms of the absence of collateral that the poor can offer conventional financial institutions, in addition to the various complexities and high costs involved in dealing with large numbers of small, often illiterate, borrowers. The poor have thus to rely on loans from either moneylenders at high interest rates or friends and family, whose supply of funds will be limited. Microfinance institutions attempt to overcome these barriers through innovative measures such as group lending and regular savings schemes, as well as the establishment of close links between poor clients and staff of the institutions concerned. The range of possible relationships and the mechanisms employed are very wide.
The case for microfinance as a mechanism for poverty reduction is simple. If access to credit can be improved, it is argued, the poor can finance productive activities that will allow income growth, provided there are no other binding constraints. This is a route out of poverty for the non-destitute chronic poor. For the transitory poor, who are vulnerable to fluctuations in income that bring them close to or below the poverty line, microfinance provides the possibility of credit at times of need and in some schemes the opportunity of regular savings by a household itself that can be drawn on. The avoidance of sharp declines in family expenditures by drawing on such credit or savings allows 'consumption smoothing.' In practice this distinction between the needs of the chronic and transitory poor for credit for 'promotional' (that is income creating) and 'protectional' (consumption smoothing) purposes, respectively, is over-simplified since the chronic poor will also have short term needs that have to be met, whether it is due to income shortfalls or unexpected expenditures like medical bills or social events like weddings or funerals. It is one of the most interesting generalizations to emerge from the micro finance and poverty literature that the poorest of the chronic poor (the core poor) will borrow essentially for protectional purposes given both the low and irregular nature of their income. This group, it is suggested, will be too risk averse to borrow for promotional measures (that is for investment in the future) and will therefore be only a very limited beneficiary of microfinance schemes (Hulme and Mosley 1996: 132).
The view that it is the less badly-off poor who benefit principally from microfinance has become highly influential and, for example, was repeated in the World Development Report on poverty (World Bank 2000:75). Apart from the risk aversion argument noted above a number of other explanations for this outcome have been put forward. A related issue refers to the interest rates charged to poor borrowers. Most microfinance schemes charge close to market-clearing interest rates (although these will often not be enough to ensure full cost-recovery given the high cost per loan of small-scale lending). It may be that, even setting aside the risk-aversion argument, such high rates are unaffordable to the core poor given their lack of complementary inputs; in other words, despite having a smaller amount of capital marginal returns to the core poor may be lower than for the better-off poor. If the core poor cannot afford high interest rates they will either not take up the service or take it up and get into financial difficulties. Also where group lending is used, the very poor may be excluded by other members of the group, because they are seen as a bad credit risk, jeopardizing the position of the group as a whole. Alternatively, where professional staff operate as loan officers, they may exclude the very poor from borrowing, again on grounds of repayment risk. In combination these factors, it is felt by many, explain the weakness of microfinance in reaching the core poor. 5 The sector has responded in a number of cases by establishing special programs for the core or 'ultra poor'. The best known of these are in Bangladesh and involve the well-established institutions of BRAC and ASA. The programs essentially aim to provide a range of services, covering training, health provision and more general social development for the disadvantaged, as well as grants of assets or credits. The ultra poor are encouraged to build up a savings fund and to graduate to conventional microfinance programs. Other variants of this approach involve greater flexibility in repayment terms for the poorest (Fernando 2004).
Given the new trends in the sector and their possible effect in diluting the original poverty focus of MFIs, the question of their impact on the poor (and particularly the core poor) is clearly of great policy interest. It might be thought that if such institutions are designed to serve only poor clients and if repayment rates are high, no further detailed analysis is needed. Such a view is misleading for a number of reasons. First, there is no guarantee that only the poor will be served unless strong eligibility criteria (like land ownership) are enforced. Often the aim is to dissuade the non-poor by the inconvenience of frequent 5 An important attempt to address this problem has been the Income Generation for Vulnerable Group Development (IGVGD) program run by BRAC in Bangladesh, which combines measures of livelihood protection (food aid) with measures of livelihood promotion (skills training and micro credit). Hence micro credit is provided as part of a package approach. Matin and Hulme (2003) survey the evidence on how far the benefits of this program actually reach the core poor and conclude that although the program was more successful than more conventional micro credit schemes none the less many target households were still missed.
meetings or the stigma of being a member of a credit group of the poor. Such disincentives need not work and eligibility criteria, where they exist, may not be enforced. Second, high repayment rates may be due to social pressure within a group or family and may not reflect the capacity to repay (if for example loans from moneylenders have to be taken out to repay the microcredit). Third, even if the poor are genuinely served by MFIs as long a public funds are required to finance the MFI there is the issue of how cost-effective this means of reaching the poor is compared with alternatives. This requires a comparison of the cost of transferring the funds through a micro credit institution per unit of benefit received by the target group, as compared with the benefitcost ratio for alternative schemes for reaching the core poor, such as food subsidies, workfare, and integrated regional development initiatives. Such comparisons must take account of not just the administrative costs involved, but also the leakage rate (that is the benefits to the non-poor).
Hence for these reasons there is a strong case for attempting to assess both the depth of outreach of microfinance programs, the impact of access to microfinance services on the welfare of clients and the costs of achieving this impact.
On the first point, assessing the depth of outreach or access of the poor to microfinance programs, it is important to note from the outset that most MFIs probably do not consider their institutional mission to be serving the poorest of the poor. Particularly in Latin America, most MFIs report a broader agenda to provide financial services to poor communities or specific groups such as female entrepreneurs who would not otherwise have access. Among MFIs that report to the Mix Market, slightly less than half of those in Asia identified "specifically targeting very poor clients" as their institutional mission. In Latin America, the share is even smaller: only around 10%. Of the Latin American MFIs that claim to target very poor clients, only two use some sort of targeting tool to identify clients. In Asia, most of the MFIs that specifically target the very poor use some sort of targeting tool, such as a means test, participatory wealth ranking or a housing index to identify the target group.
For those MFIs that do explicitly aim to serve the poorest within their community, recent work on poverty outreach of MFIs has focused on constructing a poverty index that can be used to establish whether the target group is being reached. The Consultative Group to Assist the Poorest (CGAP) has developed a poverty assessment tool (PAT) that can be used to compare clients and non-clients of MFIs in the same community. This is based on the construction of a weighted index of poverty based on a range of indicators covering the human resources of households, characteristics of their dwellings, measures of food security and their assets. The different indicators are weighted by principal components analysis, which allows weights to differ between cases . The approach here is to sort a non-client sample into three equal groups (high, intermediate and low) on the basis of their poverty score. The poverty index scores at the cut-off points between the three groups then become a reference point for the client or participant sample and their distribution between the three categories can be compared with that of the non-clients. As the non-client groups are divided equally, any deviation from equal proportions amongst the clients signals a skew either for or against greater poverty outreach. 6 The PAT is an outreach, as compared with an impact, assessment and therefore does not directly address the question of what impact the programs have on their clients. Conducting a rigorous impact assessment is challenging. It is not simply a case of looking at a group of borrowers, observing their income change after they took out micro loans and establishing who has risen above the poverty line. Accurate assessment requires a rigorous test of the counterfactual -that is how income (or whatever measure is used) with a microcredit compares with what it would be without it, with the only difference in both cases being the availability of credit. This requires empirically a control group identical in characteristics to the recipients of credit and engaged in the same productive activities, who have not received credit, and whose income (or other measure) can be traced through time to compare with that of the credit recipients.
A practitioner-friendly impact assessment toolkit is also available: the result of the Assessing the Impact of Microenterprise Services (AIMS) Project. This assessment tool has been used in longitudinal studies of the impact of programs in Peru (Mibanco), India (SEWA) and Zimbabwe (Zambuko Trust). This procedure looks at change over time and matches pairs of observations between borrowers and members of a control group, where each pair have similar starting values for the impact variable (like income or sales revenue) and other characteristics, like age, gender or sector of activity. Simplifying, this approach identifies impact as: Where Y t and Y t+1 are an impact variable (like income) in period t and t +1 respectively, p refers to matched pairs of borrowers and non-borrowers, where there are n pairs. Thus impact can be rationalized as the average difference between matched pairs of program participants and control group. 7 Where impact is greater than zero (and statistically significant) microfinance will have made a difference and once again initially poor and non-poor borrowers can be distinguished in the analysis. The weakness in the applications of approach to date is that researchers have only been able to control for observable characteristics.
Failure to account for unobservable characteristics may lead to biased measures of impact. Two key sources of bias can arise in empirical work that attempts to assess the impact of microcredit on poor households -selection bias and placement bias. The former arises where there are key differences between borrowers and non-borrowers that cannot be observed, measured and allowed for, with self-selection bias (that is where those with particular characteristics choose to participate in a program) a key problem. Hence whilst differences in education, age or gender can be controlled for statistically there can also be differences in attitude to risk or 'entrepreneurship', which will be basically unobservable. A bias will arise if there is an association between a decision to take a micro loan and these unobserved characteristics. Hence if the more entrepreneurial individuals are those who take out loans, growth in their income relative to income of those who have not taken out a loan may be due in part to the effect of the tended to serve a clientele that is more representative of the communities in which they operate, which may or may not be poorer than the national average. 7 The analysis of covariance (ANCOVA) essentially allows separate parallel regression lines to be fitted through the data for the treatment (borrower) and control groups. The regression lines measure the outcome variable for a given year (t + n) relative to an earlier year (t). Insofar as a program like microcredit has a tangible effect this will be picked up by the distance between the two lines, that is by the difference in intercept terms. The statistical significance of this distance gives a test for the impact of the program. loan itself, but in part to their entrepreneurial ability. Attribution of all of the change to the loan will overstate its impact. Placement bias arises where loans go to locations or activities that are in some way favored, such as villages with better infrastructure or sectors with strong demand growth. Comparing income change for households in a superior location (or sector) who have a loan, with income change for similar households in another location (or sector), who have not taken out a loan, and attributing of all this to the loan will create an upward bias.
Best-practice approaches to resolving these problems employ a form of "difference-indifference" (two-stage least squares instrumental variables) analysis that compares participants and a similar control group and between locations or sectors with and without access to the program. 8 One approach (as used for example by Pitt and Khandker (1998) on Bangladesh) is to use exogenous eligibility criteria for participation in a microfinance program (for example lack of land ownership) as a means of avoiding a self-selection bias. Placement bias is allowed for by comparing those who are eligible with those who are ineligible, both in villages that are covered by programs and those that are not. Hence the analysis based on a double difference can be simplified as follows Where Y is change in an outcome measure (such as income) over the study period, e and i stand for eligible and ineligible households, respectively, and p and n stand for program and non-program villages, respectively. For microfinance to produce positive results Impact must be greater than zero. If poor and non-poor borrowers can be identified, there will be a quantification of poverty impact.
The chief problem with this approach is that many microfinance schemes do not use formal eligibility criteria and those that do may not always enforce them, creating a further source of error. An alternative where no formal criteria are set out but approvals for borrowing are known is to use as a control group those approved for loans who have not yet taken them up (for example as used by Mosley and Hulme (1996) in their country studies). This address the self-selection issue unless not taking up a loan reveals an aversion to risk and is correlated with subsequent outcomes.
A variant of this approach (as applied by Coleman (1999Coleman ( , 2004 for Thailand) draws on the fact that most microfinance activities start in a narrowly defined area and then expand their coverage to similar villages elsewhere or within urban centers. In the rural case, if the villages are similar and if the borrowers can choose to participate, then selfselecting participants in villages that have been identified for later inclusion in a program should provide an accurate control group for current borrowers in villages with a program. Here, again simplifying, this is equivalent to estimating impact as Where Y is as before, P and N stand for (self-selecting) participants and non-participants respectively, t stands for time a program has been operative in a particular village, so t + 1 covers the early and t the late entrant villages.
Here we examine some of the recent 'rigorous' studies on the impact of MFIs based on survey data that employ versions of these methodologies. We do not report the results of work based on more qualititative or participatory approaches. 9 Table 2 summarizes the results of the studies surveyed here for Asia and table 3 does the same for Latin America. In general it is perhaps not surprising that studies based on a rigorous counterfactual find much smaller gains from microfinance than simple unadjusted before and after type comparisons, which erroneously attribute all gains to micro credit. Also although the results are far from consistent, studies on Asia tend to report a stronger poverty impact from microfinance than do comparable work from Latin America.

III.1. Poverty Impact Studies -Asia
One of the early and most widely cited of the poverty impact studies is Hulme and Mosley (1996). This employs a control group approach looking at the changes in income for households in villages with microfinance programs and changes for similar households (for example, in terms of initial income, gender, education, and location) in non-program areas. As far as possible the control groups are drawn from households eligible for loans and who had been approved for loans by the institutions concerned, who had not yet received a loan. Programs in a number of countries are considered including the Grameen Bank in Bangladesh and the Bank Rakyat Indonesia (BRI). In general a positive impact is found on borrower incomes of the poor (1988-92) with on average an increase over the control groups ranging from 10-12% in Indonesia, to around 30% in Bangladesh and India (Hulme and Mosley 1996, table 8.1). Gains are larger for non-poor borrowers, however, and within the group the poor gains are negatively correlated with income. However, despite the breadth of the study and its use of control group techniques, it has been criticized for possible placement bias, whereby microfinance programs may be drawn to better placed villages, so that part of the advantage relative to the control group may be due to this more favorable location. The quality and accuracy of some of the data, particularly in relation to the representative nature of the control groups, has been questioned (Morduch 1999(Morduch :1600. There also appears to be a basic problem with the data side of the case studies, since these are not based on a comparison between baseline data and that for a later survey year. Rather there is at least partial recourse to a recall approach for the earlier years of the period covered, as respondents are asked to estimate their income retrospectively. Finally the major conclusion of the study that there is a positive correlation of gains from microfinance with income, so that poorer borrowers gain proportionately less, has also been challenged on the grounds that their comparison of income changes for different categories of borrowers biases their results in favor of the conclusion. This follows since gains for different income groups are compared with the average for a control group, not with the change for comparable income categories within the control group; in other words gains to very poor borrowers are compared with average gains in the control group not to the gains to the very poor controls (Morduch 2003).
Another major early initiative that has provided some of the firmest empirical work were the surveys conducted in the 1990's by the Bangladesh Institute of Development Studies (BIDS) and the World Bank; these provided the data for several major analyses, such as Pitt and Khandker (1998). Khandker (1998) summarizes a number of different studies conducted in Bangladesh using the 1991/92 survey and focusing on three major microfinance programs, including the Grameen Bank and the Bangladesh Rural Advancement Committee (BRAC). As discussed above impact is assessed using a double-difference approach between eligible and ineligible households (with holdings of land of more than half an acre making households ineligible) and between program and non-program villages. After controlling for other factors, such as various household characteristics, any remaining difference is attributed to the microfinance programs. The study draws a number of conclusions, but the main one is that the program had a positive effect on household consumption, which was significantly greater for female borrowers. On average a loan of 100 taka to a female borrower, after it is repaid, allows a net consumption increases of 18 taka. In terms of poverty impact it is estimated that 5% of participant households are pulled above the poverty line annually. Khandker (2003) follows up this earlier work by employing panel data. He uses the BIDS -World Bank survey conducted in 1998-99 that traced the same households from the 1991-92 survey. He finds apparently strong and positive results. Whilst borrowing by males appears to have no significant impact on consumption, that by females, who are the dominant client group, does have a positive impact. From this analysis a 100 taka loan to a female client leads to a 10.5 taka increase in consumption (compared with 18 taka in the earlier analysis). Allowing for the impact of higher consumption on poverty gives estimates of poverty impact. It is estimated that due to participation in micro finance programs moderate poverty among program participants decreased 8.5 percentage points over the period of seven years and extreme poverty dropped about 18 points over the same period. 10 He also finds evidence of positive spillovers on nonprogram participants in the villages, with the impact greater for those in extreme poverty. Over the study period of seven years poverty for non-participants is found to decline by 1 percentage point due to the programs, whilst extreme poverty declines by nearly 5 percentage points. This impact is due solely to female borrowing.
The same data set has also been used to identify health impacts as opposed to income changes. Pitt et al (2003) find that credit going to females has a large and significant impact in two out of three health measures for children. Male borrowing has no such effect. For example, a 10% increase in credit to females increases the arm circumference of daughters by 6.3%. A 10% increase in female credit on average increases the height of girls by 0.36 cm annually and of boys by 0.50 cms. The relations are stronger for daughters than sons. Hence in Bangladesh micro credit and improved family health appear to be related.
These are strong and positive results and probably are the clearest evidence there is that microfinance is working in the way intended to bring sustained relief from poverty. However a couple of caveats are in order. First, the accuracy of the original results as presented in Pitt and Khandker (1998) has been disputed on the grounds that the eligibility criteria of low land holdings was not enforced strictly in practice. In a reworking of the results focusing on what are claimed to be more directly comparable households no impact on consumption from participation in a program is found (Morduch 1999(Morduch :1605. 11 Second, in the BIDs-World Bank survey data the 'ultra poor' (defined as those with less than 0.2 acres of land) form nearly 60% of participants and the likelihood of participation is strongly and negatively associated with level of land holding. Nonetheless, how much is borrowed depends principally on the entrepreneurship of households, so that the charge that the risk-averse very poor will benefit proportionately less has not been totally dispelled. Furthermore, the panel data reveals a relatively high dropout rate of around 30%, indicating that there may have been problems of repayment for many households.
For Asia, there are examples of other studies that are either inconclusive or provide less convincing results. Coleman (1999) and MkNelly et al (1996) both focus on experiences with village banking in Thailand. Coleman (1999) utilizes data on villages that had participated in village bank micro finance schemes and those control villages that were designated as participants, but had not yet participated. As noted above this allows a double difference approach that compares the difference between income for participants and non-participants in program villages with the same difference in the control villages, where the programs were introduced later. From the results here the poverty impact of the schemes appears highly dubious. Months of village bank membership have no impact on any asset or income variables and there is no evidence that village bank loans were directed to productive purposes. The small size of loans means that they were largely used for consumption, but one of the reasons there is a weak poverty impact is that there was a tendency for wealthier households to self-select into village banks. Coleman (2004) uses the same survey data but reconsiders the estimation strategy to control for self-selection. He argues that the village bank methodology, which relies on self-selection by loan size and monitoring by frequent meetings, may not reach the poorest. As many better-off households tend to be on village bank committees, the failure to control for this leads to systematic biases. The regression results of Coleman (2004) indicate that there is substantial difference between ordinary members and committee members of village banks. The impact of microcredit on ordinary members' wellbeing is either insignificantly different from zero or negative. On the contrary, the impact of microfinance programs on committee members' measures of wealth, such as income, savings, productive expenses and labor time is positive, implying a form of program capture by the better-off in the village, even though this group may not be welloff by national standards. A similar result in terms of rationing micro credit in favor of better-off groups or members is found by Doung and Izumida (2002) in a study of six villages in Viet Nam. There whilst credit availability is linked with production and income household economic position and prestige in a village plus the amount of credit applied for are the main determinants of how credit is allocated. MkNelly et al (1996) evaluated the Freedom from Hunger credit with education program in Thailand operated through village banks. The results show positive benefits, however although non-participants in non-program villages are used as controls, there are problems in accepting the results. No statistical tests are reported, so one cannot judge whether differences between participants and non-participants are significant. There is also a potential measurement bias since the staff responsible for the program also did the interviewing. Chen and Snodgrass (2001) examine the operations of the Self Employed Women's Association (SEWA) bank in India providing low income female clients in the informal sector with both saving and loan services. The study tests for the impact of these services by comparing the bank's clients against a randomly selected control group in a similar geographic area. Two surveys were conducted two years apart. Average incomes rose over time for all groups -borrowers, savers and the control, although the increase was less for the latter. In terms of poverty incidence there was little overall change, although there was substantial 'churning', in that amongst the clients of SEWA there was quite a lot of movement above or below the poverty line. In interpreting these results Meyer (2002) argues that the evidence on the counterfactual -that is what would have happened to the clients in the absence of the services of SEWA -is not sufficiently strongly established to draw any firm conclusions on poverty impact.
The smoothing of consumption over time to protect the poor against adverse shocks is one of the principle objectives of micro credit. Using data again for Bangladesh, Amin et al (2003) compute several measures of vulnerability. 12 They find that the micro credit participants in the two villages covered are more likely to be below the poverty line than if they had been selected at random, so that the programs have reached the poor. However, the vulnerable are more likely to join a micro credit program in only one of the two villages. Further, for the vulnerable below the poverty line in one village there is no evidence that there are more likely to be members of a program and in the other village there is evidence that they have either chosen not to join or are actively excluded, presumably on the grounds that they are a poor credit risk. Hence the very poor and vulnerable do not appear to be reached.
More positive conclusions in terms of the ability of micro finance to reduce vulnerability are found for Indonesia by Gertler et al (2003), who find that access to micro finance helps households smooth consumption in the face of declines in health of adult family members. Having established an empirical relationship between health condition and consumption, the authors test for a relation between access to a financial institution and consumption shortfalls associated with ill health. Using geographic distance as a measure of access they find that for households in an area with a BRI branch health shocks have no effect on consumption. This study does not differentiate within the group of the poor.

III.2. Poverty Impact Studies -Latin America
In Latin America in general the impact of microfinance on poverty has been less well documented both in a methodological sense and in terms of coverage in individual studies, which tend to be concentrated in a small number of countries, principally Bolivia and Peru. The overall impression, however, is that compared with Asia microfinance has reached less far down the income scale and that a significant proportion of borrowers are not in fact below the poverty line, although they may well have below average incomes. This is likely to be due at least in part to a greater commercial orientation with a focus on credit for urban micro-enterprises, with lower rural outreach in Latin America as compared with other regions. A typical requirement for access to credit from an MFI has been that the borrower should be the owner of a micro-enterprise, holding a national identification card and having at least six to twelve months experience in the economic activity for which the loan is to be used (Gulli and Berger 1999:26). It is perhaps not surprising that many of the poor do not meet these criteria.
For example, detailed evidence on the outreach of MFIs in Bolivia is provided by the survey reported in Navajas et al (2000), who use an index of basic needs fulfillment to classify borrowers into poor and non-poor groups. For the urban area of La Paz they find that of three MFIs, two tend to lend disproportionately to those above of the poverty line. For two of the three, the share of 'moderately poor' borrowers (at 29%) was lower than their share of the population (at 38%), although this was not the case for the third MFI, BancoSol (at 47%). However of the very poorest group the share of borrowers in all three institutions (at 2-5%) was well below their share in the population, reinforcing the view that MFIs have difficulty in reaching the very poor. When rural lending activities are also included there is a tendency for a skew in lending towards the 'threshold' group, defined as those just above the poverty line and the 'moderately' poor. Table 4 gives the ratio of the share of groups of borrowers by poverty class in the portfolio of the different MFIs to their share in total population. A figure above unity thus indicates a positive skew towards a particular poverty class and a figure below unity indicates the opposite.
In terms of institutional mix FIE, PRODEM and Sartawi are NGOs, whilst BancoSol and Caja Los Andes are regulated financial institutions. Table 4 shows that being an NGO (like FIE) is no guarantee of strong allocation of loans to the poor and that both regulated institutions had a superior distribution to FIE. However in turn the rural-based NGOs, PRODEM and Sartawi outperform BancoSol by this criteria. This type of evidence on poverty outreach does not address the issue of how far incomes of poor borrowers have been affected. In the limited number of detailed poverty impact studies on Latin America, BancoSol of Bolivia remains by far the most studied institution. Hulme and Mosley (1996, table 4.1) look at a small sample of BancoSol borrowers. Using those approved borrowers who had not yet taken out a loan as a control they find an average annual increase in income of 28% for borrowers compared with an average of 14.5% for the control group. An estimated 8% of borrowers crossed the poverty line in 1992 alone. However in comparison with the MFIs from other countries in their study BancoSol has only a relatively small proportion of borrowers in the sample below the poverty line (29%) and average borrower household income from the sample was nearly five times the national poverty line, which is far higher than for any institution studied in other countries. BancoSol also showed the largest average absolute income increase for borrowers, and the proportionate increases were greater for the poor. Although the Hulme and Mosley study has a reasonable control group criteria (those approved borrowers who had not yet taken out a loan, but who might be expected to share the self-selection characteristics of current borrowers) it suffers from several problems; there is only a small sample of 36 borrowers; it is not clear that the control group matches borrowers exactly in terms of characteristics such as education, gender or sector of activity; and the sample is surveyed at a point in time so that retrospective income estimates are required to derive rates of change.
The last of these problems is addressed for BancoSol, but not the other Bolivian MFIs covered, in Mosley (2001), which resurveys the households covered earlier to obtain income data at two points in time. Mosley (2001) finds that for BancoSol borrowers resurveyed on average income growth was a little more than twice (214%) that of the control group; for the other three institutions the excess income growth for borrowers over the control group was between 132% and 158%. For poor borrowers (who were a minority of those surveyed) gains relative to the average for the control group were lower than for all borrowers, for example 151% in the case of BancoSol. Regression analysis relating income increase per household relative to the control group average to initial income shows a positive relationship, so that proportionate gains from borrowing rise with household income, although at a declining rate. There is a positive poverty impact, although given the fact that only a minority of borrowers (around one third) were poor at the starting point of the analysis in 1993, this is modest. Between 10%-20% of poor borrowers, varying between institutions, crossed the poverty line over the period studied as a result of microfinance. 13 However when the core poor (those in 'extreme poverty' defined in Bolivia as those living on half the poverty line) are considered, it is clear that none of the MFIs studied are reaching them. From a sample of 200 borrowers over six years for four institutions, there is only one case of the removal of extreme poverty and hence this segment of the poor was not reached. Arbuckle (2001a, 2001b) use an analysis of covariance to examine loans to micro-enterprises for 305 households in Lima, Peru by Mibanco. The study draws on data at two points in time 1997 and 1999 and looks at changes in the borrowers relative to a control group of households who had not received a micro-enterprise loan. On average the borrower group appears to be around or slightly above the national poverty line, with approximately 30% below the national poverty line. As noted above, the procedure uses matched observations in the borrower and control groups that have the same starting values for performance variables, like net revenue, assets or employment, as well as the same values for 'moderating' variables, like gender of entrepreneur, sector of activity and location. Change in the performance variables for the matched observations over 1997-1999 are compared to establish if there are significant differences between the borrowers and the control group. The results suggest on average a significant difference in terms of enterprise revenue (roughly $1000 annually), fixed assets and employment creation (as much as nine extra days per month). These results are very substantial. The study however recognizes that it may be difficult to attribute all of these changes to the microcredit program of Mibanco, as the matching system used does not address adequately self-selection bias and the moderating variables used seem crude (for example, sector variables reported are 'commercial, service and industrial' rather than anything more precise such as industrial subsectors).
The poverty dimension of the study as reported in Dunn and Arbuckle (2001b) shows a positive poverty reduction effect. For households starting with the same poverty level, number of income sources and economically active members in 1997, on balance after net effects are allowed for by 1999 borrowers were 6% more likely to be above the poverty line than non-borrowers. There is the contrary result, however, that in the smaller group of new borrowers who took out a loan during 1997-1999, but not initially in 1997, new borrowers were 15% less likely to have moved out of poverty than the control group. 14 The poor and non-poor appear to benefit almost equally in absolute terms, although there is evidence that the poorer borrowers were 20% more likely to liquidate assets in response to a financial shock. Banegas et al (2002) look at the operations of two MFIs in Ecuador (Banco Solidario) and Bolivia (Caja los Andes) utilizing the CGAP poverty index noted above to establish outreach and a logit regression model (where being a client and taking a loan gives a dependent variable of 1.0 and being a non-client a dependent variable of zero) that links participation in a program with income changes and poverty scores. It is found that for both institutions taking a loan is associated with increases in income. However income change is measured not by the size of monetary values but by a simple scoring system (1 for income decrease, 2 for unchanged income and 3 for income increase). The relation with poverty varies since in the case of Banco Solidario lower poverty is associated with a greater probability of taking a loan and in the case of Caja los Andes with a higher probability. On the other hand Banco Solidario has a greater depth of outreach as 75% of its clients belonged to the lower and intermediate groups as defined by the CGAP poverty score, as compared with 48% for Caja los Andes. Again it seems therefore it is the better-off amongst the poor who are benefiting. Limitations of this analysis are the crudity of some of the indicators, for example for income change, and the way in which a control group of non-clients are selected; that is from households in the same locality that have micro-enterprises in the same sector as the borrowers and which have not had a loan from a formal sector institution. This simply ignores the issue of self-selection bias and does not control for factors like education and skills.
From a nutritional perspective MkNelly and Dunford (1999) look at the impact of Credit with Education loans to women in rural Bolivia. A relatively rigorous approach is applied by collecting data two years apart from a participant group and a control group, who would be offered the credit at the end of the study period. In addition amongst the participants a sub-group of those who joined during the course of the study, rather than immediately, is examined separately. Small loans were available in combination for training in health and nutrition, as well as micro-enterprise management topics. Roughly two-thirds of participants reported an increase in income over the study period and their net incomes in 1997 appeared far higher than the control group (perhaps casting some doubt on the representativeness of the latter). However on the key concern of the study, nutritional status (for example child height-for-age or weight-for-height measures), there is little evidence of any impact due to the program. The most positive result is that for households suffering 'food stress', participants are less likely to sell off animals and are more likely to take out loans as a coping strategy, than are non-participants.
In general, for Latin America the available studies suggest that MFIs, whilst they may be flourishing in commercial terms, and providing a valuable service to micro-enterprises often run by poor entrepreneurs, have relatively weak impact on those at the very bottom of the income distribution.

IV. Forms of Microcredit Interventions and Cost-Effectiveness
It is clear that experimentation and local variation are likely to be important aspects of successful MFIs. A few studies (more in Asia than in Latin America) have looked in detail at the impact and cost effectiveness of different forms of intervention. The Hulme and Mosley (1996) cross-country study of 13 institutions in seven countries (Bolivia in Latin America and Bangladesh, India, Sri Lanka and Indonesia in Asia) found that loan impact, in terms of change in borrower income, (which is not necessary the same as poverty impact) was greater in the more financially viable institutions (such as BRI and BancoSol). They explain this in terms of the screening efficiency of higher interest rates and tighter repayment conditions, which deter less financially sound borrowers. The institutions involved used a range of delivery mechanisms and the analysis does not allow firm judgements between these. Within-country comparisons by ownership are made explicitly in Park and Ren (2001), who look at the Chinese experience drawing on household survey data for 1997. They are able to compare three types of program based on ownership characteristics -NGO-based, mixed programs and government ownership. Whether in terms of conventional financial criteria like repayment rates, or measures of initial impact like targeting effectiveness, the NGO programs appear to function best, with the government-run programs the least successful.
Detailed mechanisms for micro lending are examined for Thailand by Kaboski and Townsend (2003) who look at different institutional variants such as production credit groups, women's groups, rice banks and buffalo banks, as well as a variety of services included training and various savings facilities. Of the forms of institution, allowing for a range of other factors, women's groups appear to have the largest positive impact on their members. Of the services offered, training in conjunction with credit appears to work well and the availability of savings facilities appears to be associated with asset growth amongst households. Of the savings services regular 'pledged savings' have the largest positive impact. Explanations offered for this include the use of savings as collateral for further loans either from the institution itself or from other sources, and a reduction in the cost and risk of infrequent deposits and withdrawals. However since the poorest may not be in a position of offer regular savings, this also provides an explanation for why they may benefit relatively less from MFIs. 15 Most studies of the impact of different forms of micro finance do not conduct a full cost effectiveness analysis in order to judge both the effectiveness of different alternatives and how micro finance interventions compare in efficiency terms with other ways of reaching the poor. However there is often a general expectation that MFIs are an effective and efficient means of reaching the poor. For example, Wright (2000) argues that "...microfinance has a particular advantage over almost (and probably) all other interventions" in providing cost-effective and sustainable services to the poor. In fact the evidence to support such a strong claim is not yet available. Bangladesh and Bolivia, the most widely studied countries for microfinance, provide most of the evidence on its cost effectiveness.
The early work by Khandker (1998) attempts to assess the cost-effectiveness of micro credit in Bangladesh (that is costs per taka of consumption for the poor) as compared with more formal financial institutions and other poverty-targeted interventions. His data are summarized in table 5. They appear to be based on the assumption of a zero leakage rate to the non-poor. The interesting result that emerges is that the Grameen Bank is considerably more cost-effective than BRAC and that as expected loans to female borrowers are considerably more cost-effective than loans to males. Further, subsidies to Grameen (but not to BRAC) appear to be a more cost effective means of reaching the poor than various food for work programs. However a food for education scheme appears very cost-effective relative to the food-for-work programs and to BRAC. 16 Formal financial institutions are less cost-effective than Grameen for both female and male borrowers and less cost effective than BRAC in some, both not all, cases examined (Khandker 1998:134-139). The high figure for BRAC is in part due to the range of services, such as training, offered in addition it micro credit, but nonetheless if such services are essential to the success of microcredit, including their cost in a costbenefit assessment of microcredit is legitimate.  Wodon (1998); figure in brackets is the cost effectiveness ratio for the very poor.
The above data provide ambigous support for the idea that micro-finance is a costeffective means of generating income for the poor. The figures for Grameen support this view, whilst those for BRAC do not. More recently a couple of other estimates are available. Burgess and Pande (2003) examine whether the pattern of commercial bank expansion in India into rural areas, previously not served by banks (so-called 'social banking'), has impacted on rural poverty and their work allows a simple comparison with microfinance. Their estimates suggest that it costs 2.72 rupees to generate an additional rupee of income for the poor via social banking program. Compared with the data in table 5 this ratio is higher than the cost-effectiveness ratio for Grameen, but lower than that for BRAC. 17 A further look at the effectiveness of Grameen is provided by Schreiner (2003), who calculates the subsidy-lending ratio at 0.22 over the period 1983-97. This is not directly equivalent to the ratios in table 5, but assuming the same return to borrowing as in Khandker (1998) these figures can be converted into a broadly equivalent ratio of cost to gains to the poor of 1.15. This is consistent with the figures in table 5 which would need to be averaged to give an overall return to male and female borrowing combined. The result confirms Grameen as a relatively cost-effective form of poverty intervention, although it says nothing about how the benefits from its activities are distributed between the poor, the very poor and those above the poverty line.
For Latin America, Mosley (2001) provides a rare, if approximate, estimate of costeffectiveness of MFIs relative to other poverty interventions in Bolivia. He compares the estimated numbers in a particular area brought over the poverty line by four different MFIs, as a result of microcredit, with the organizations' expenditure that can be allocated to activities in that area. This gives a cost per person brought out of poverty for four MFIs that use different approaches. BancoSol and Fundacion para la Promocion y Desarollo de la Microempresa (PRODEM) are more commercial with greater use of individual loans, whilst ProMujer lends largely to women in urban co-operative groups and Sartawi offers both group and individual loans, but also provides a range of training and education services in addition to credit. Cost-effectiveness in the MFIs, defined as the cost per person brought out of poverty, are $603 for BancoSol, $467 for ProMujer, $373 for PRODEM and $589 for Sartawi. These figures are not directly comparable with those for Bangladesh reported in table 5, as the latter are the ratio of MFI costs to benefit in income (or consumption) received by the poor. Although the range is relatively wide, perhaps due to the approximate nature of the calculations, the author himself suggests that they show that there is little difference between the institutions and that no one model dominates microcredit delivery in Bolivia (or indeed elsewhere). There are also some approximate comparisons with the cost of poverty reduction from Social Fund investment in health, education and rural roads, which show microfinance from all of the institutions to be lower cost than the Social Fund programs. 18 However, the cost effectiveness figures found for MFIs Bolivia in dollars per person brought out of poverty are much higher than some of the anecdotal figures used for Bangladesh. The fact these estimates, approximate as they are, provide one of the few indications of the costeffectiveness of MFIs in Latin America is an indication of the undeveloped nature of research on this issue in the region.
In general in terms of cost-effectiveness there is limited support for the view that MFIs can be cost-effective ways of reaching the poor, although the range of figures within both Bangladesh and Bolivia suggest that this is far from inevitable for all types of MFI. BRAC in particular appears relatively high cost. However even if it could be shown that microfinance uniformly outperformed other targeting measures in cost effectiveness terms one could still not conclude that other measures should be abandoned and their funds diverted to microfinance. As Khandker (1998) points out, participants to microfinance borrowing self-select (that is they judge that micro credit suits their particular needs, often for self employed work), whilst microfinance may not be suitable for others amongst the poor. For this latter group, perhaps more risk adverse or more disadvantaged, other forms of targeting will still be required.

V. Conclusions
Despite the current enthusiasm among the donor community for microfinance programs, rigorous research on the outreach, impact and cost-effectiveness of such programs is rare. Design of aid programs would ideally incorporate evidence on all three points, but the research that does exist generally focuses on only one of these criteria: either outreach, impact or cost-effectiveness. In part this reflects the difficulty of establishing an appropriate statistical methodology and implementing those standards in practice, and in part no doubt reflects the variation found in practice in the way in which microfinance operates. The evidence surveyed here suggests that the conclusion from the early literature, that whilst microfinance clearly may have had positive impacts on poverty it is unlikely to be a simple panacea for reaching the core poor, remains broadly valid. Reaching the core poor is difficult and some of the reasons that made them difficult to reach with conventional financial instruments mean that they may also be high risk and therefore unattractive microfinance clients.
Asia has much to learn from Latin America in terms of developing a vibrant commercially oriented MFI sector. However MFIs in Latin America have often been seen as a vehicle for the development of the micro-enterprise sector rather than as a tool for the removal of core poverty, which was its initial focus in much of Asia. Work on Bolivia has demonstrated this at least for that country. There has been an extensive debate that we do not touch on here, on the financial sustainability of MFIs. We would simply make the point that just because an institution needs a subsidy to cover its costs in itself is not a reason for not supporting the institution. The issue would be what benefits, in terms of income gains for the poor, can be achieved with the subsidy and how the ratio of subsidy to benefits compares with that for other interventions. Detailed cost effectiveness studies are rare and those that are available show both high and low scores for MFIs in the same country. Hence there is a need to continually improve design and outreach and to see MFIs as part of the package for targeting the poor, rather than the whole solution.
Our view is that despite the difficulties, there is a need for more careful research on the outreach, impact and cost-effectiveness of microfinance programs -studies that rigorously address the critical issues of selection and placement bias. Such studies can inform the debate on the way forward for microfinance by sharpening the donor community's understanding of the role of microfinance in reaching the poor, its impact in different environments, and its cost-effectiveness as a poverty intervention.