The impact of group lending in Northeast Thailand

Abstract

Much of the literature on group lending focuses on its high repayment rates rather than its goal of promoting borrower welfare. Most studies that attempt to measure the impact of group lending neglect the issues of self-selection and endogenous program placement, thus leading to biased estimates of impact. One reason for this neglect is the lack of data that would allow for identification of impact. This paper surmounts these problems by using data from a quasi-experiment conducted in Northeast Thailand in 1995–1996. Program participants were identified in six control villages 1 year prior to receiving loans. Surveys were then conducted of these “control” members, “treatment” members in eight older program villages, and nonmembers in both types of village. This survey design allows for straightforward estimation of impact. The results indicate that program loans are having little impact although “naive” estimates of impact that fail to account for self-selection and endogenous program placement significantly overestimate impact.

Introduction

One cause of poverty observed in less industrialized countries (LIC) may be the poor's lack of access to productive capital. The poor often find themselves in a vicious circle: producing at a subsistence level makes it difficult to accumulate savings or other assets, thus making it difficult either to invest in productive resources or to gain access to credit in formal capital markets, which leads to low productivity and continued poverty.

In most LICs, local moneylenders are the principal source of credit to peasant households. One advantage of the village moneylender is that he knows the reputations of his clients and can monitor their activities much more easily and cheaply than potential competitors. However, moneylenders often charge annual interest rates of more than 100%. Given that world interest rates in formal credit markets are in the range of 10 to 20%, many potentially profitable projects are not undertaken in rural areas of LICs. This inefficiency may have a greater impact on poor women (and the children under their care) than on men because women generally have even less access than men to formal credit markets. Hence, delivering sustainable, low-cost credit to the poor, especially to poor women, should lead to increases in efficiency and equity by increasing their income and expenditures on children.

Commercial banks generally do not cater to the needs of the rural poor. The projects that most peasant borrowers would undertake are small scale, requiring small loans; therefore, the costs of obtaining the information necessary to select borrowers, evaluate their creditworthiness, monitor the use of the loans, and enforce repayment outweigh the potential profits to most lending institutions. Hence, previous government-led efforts to deliver formal credit to rural areas have included setting up special agricultural banks or ordering commercial banks to loan a certain minimum percentage of their loan portfolio to rural borrowers. However, such efforts have generally failed because it is often politically difficult for governments to enforce repayment of loans, and because the below-market interest rates that have been charged induce non-price rationing of loans (e.g., loaning to those who can put up the most physical collateral), allowing the rural elites rather than the poor to receive the lion's share of the loans (Adams and Vogel, 1986; World Bank, 1989).

The failure of formal lending institutions and the apparent success of Bangladesh's Grameen Bank in reaching the rural poor have recently inspired numerous non-governmental organizations (NGOs) and LIC governments to establish group-lending schemes to deliver credit at low cost and reasonable interest rates to small-scale rural entrepreneurs. NGOs usually target women¹ in their “village bank” programs, which generally operate on a much smaller scale than government-led programs. In these programs, borrowers form their own peer groups of 20 to 60 members (sometimes further broken down into five-person “solidarity groups”). The NGO lender grants a loan to each member, but group members co-guarantee each other's loans. Loans are generally due in about 6 months. If the group does not meet its collective responsibility to repay all of its members' loans, then all group members are denied future credit. The first loan is the same amount for all members (1500 baht in Thailand²). For each subsequent loan cycle, the member is entitled to borrow an amount equal to her previous loan plus her accumulated savings in the village bank, up to a fixed maximum (in Thailand, the maximum is 7500 baht). Moreover, the group also makes loans to its members (and sometimes to nonmembers) from its members' savings. Loans from the NGO lender are “external account” loans, and loans from members' savings are “internal account” loans.³

Hence, given the tightly knit communities that exist in many villages in LICs, members are well placed to judge the creditworthiness and to observe the actions of their peers, thus mitigating the problems of adverse selection and moral hazard (Stiglitz, 1990; Varian, 1990; Ghatak, 1999). Group lending also provides incentives for a member to avoid excessively risky projects (Stiglitz, 1990), to repay her loan in order to avoid the social sanction of her peer group (Besley and Coate, 1995), to seek assistance from other members if her project is performing poorly or provide similar assistance to other members (Varian, 1990), and to provide insurance to other members in the event that their projects fail (Coleman, 1998b). Given the trust that exists between group members who often have known each other all their lives, there may also be a strong inducement to self-monitor so that monitoring costs to other borrowers are close to zero. Hence, group lending provides a mechanism to overcome some of the informational disadvantages of commercial lenders. Indeed, Wydick (1995b) demonstrates that the social cohesion of groups in Guatemala, by mitigating adverse selection and moral hazard and by encouraging mutual insurance, is the primary determinant of group lending's high repayment rates. The repeated game theory, which states that borrowers repay simply to ensure the continuation of access to loans at favorable loan terms, is not supported.

Group lending is often viewed as a success, primarily because of its high repayment rates, usually over 90%, and low-cost delivery system. Indeed, most research to date focuses on reasons for the high repayment rates. The primary goal of village banks, however, is to improve borrower welfare. It is usually assumed, based on the high repayment rates as well as numerous anecdotes of how individual members pulled themselves out of poverty, that village banks accomplish this goal (Remenyi, 1991). However, the hypothesis that village banks accomplish this goal has not been adequately tested. Indeed, leading members of the “Ohio School” of development finance have recently stated bluntly, in response to the mushrooming of group lending, that “debt is not an effective tool for helping most poor people enhance their economic condition — be they operators of small farms or micro entrepreneurs, or poor women” (Adams and von Pischke, 1992). They argue that access to credit is not a significant problem faced by small agricultural households and that factor and product prices, land tenure, technology, and risk are the factors limiting small farmer development. Thus, given that village banks aim to alleviate rural poverty, their success needs to be judged by other standards in addition to the lenders' costs, profits, and loan recovery rates. Moreover, village banks will not be sustainable unless the benefits to their members are sustainable.⁴ The main purpose of this paper will be to evaluate the village banks against the primary goals for which they have been implemented.

The main problem plaguing attempts to evaluate the success of village banks in promoting borrower welfare is that village bank members self-select and are selected by their fellow members: a potential member must decide that she wants to participate in the program, and she must be accepted as a member by other villagers who have self-selected. Hence, it is likely that there are significant differences between village bank members and nonmembers. To the extent that such differences can be observed and measured (e.g., age, education), they can be controlled for when estimating village bank impact. To the extent that such differences cannot be observed (e.g., entrepreneurship, risk preferences, trustworthiness, attitudes regarding the role of women in the household, attitudes toward belonging to a poverty lending group), direct comparison of village bank members and nonmembers will yield biased estimates of village bank impact. This bias results because the same unobservable characteristics that lead some women to become village bank members will also affect impact measures such as income, accumulation of assets, and spending on education and health care.⁵

Most existing impact studies are nonacademic project evaluations that are of a descriptive nature or suffer from the selection bias problem. Chen (1992) reviews 11 studies of the Grameen Bank in Bangladesh, none of which make any correction for selection bias. Sebstad and Chen (1996) review 32 research and evaluation reports on the impact of micro-credit; none of these 32 studies account for section bias, with the exception of the study by Hulme and Mosley (discussed separately below). Hossain (1988) also presents descriptive and anecdotal evidence of the positive impact of the Grameen Bank. MkNelly and Watetip (1993) evaluate the impact of village banks in Thailand, but use as their control group women from villages that do not have a village bank and who, therefore, have not had the opportunity to self-select.

Among academic studies, Wydick, 1995a, Wydick, 1995c evaluates the impact of group lending in Guatemala on child labor and class mobility, but like MkNelly and Watetip (1993), he uses as his control group entrepreneurs who have never been given the opportunity to self-select, though he does attempt to match his control and treatment groups on observables.

Two exceptions to this lack of attention to selection bias exist. The study by Hulme and Mosley (1996) of micro-lending institutions in several countries included eight institutions that practice group lending, and in two of these, the authors identified a control group that had been accepted for a loan, but who had not yet received a loan. However, the authors present only the means of various outcome variables for both treatment and control groups. No statistical analyses of the differences are conducted. Moreover, their data would not allow them to control for the possibility of endogenous program placement, discussed in greater detail below.

The most thorough attempt to correct for selection bias and nonrandom program placement in group lending is the study by Pitt and Khandker (1998), who use data from a World Bank survey of the Grameen Bank and two other group lending programs in Bangladesh. Specifically, they use a quasi-experimental design in which they sample members and nonmembers from villages with a program, as well as randomly selected households from villages without a program. In this design, availability of a credit program is used as an identifying variable. However, they recognize that there will likely be systematic differences between the two types of villages because program placement may be endogenous. Therefore, they use village fixed effects estimation to control for unobserved differences between villages. But, because the village-specific dummy variables that identify fixed effects would be collinear with the “program availability” variable, they also sample households in program villages which are, in principle, exogenously excluded from the programs — in the programs studied, households with more than a fixed amount of assets are theoretically excluded from membership consideration.⁶

Most group lending programs, however, do not impose such eligibility criteria. Rather, they attempt to attract the relatively poor and dissuade the relatively rich from participating by the small size of loans, the high frequency of meetings, and the stigma of belonging to a poor person's credit program. Hence, the method of Pitt and Khandker could not be implemented in most group lending programs. Moreover, even in the context of the three Bangladeshi programs they studied, their survey found that some 18–34% of program participants in fact had wealth that should have excluded them from participating. Hence, the use of this eligibility criterion as a key exclusion restriction may not be appropriate.⁷

The methods used in this paper do not require the existence and enforcement of exogenously imposed membership criteria to identify program impact. Instead, they rely on data collected using a unique survey that allows for the use of relatively straightforward estimation techniques. Member and nonmember households in 14 villages in Northeast Thailand were surveyed four times over the course of a year. At the time of the first survey, seven of the villages had a village bank for 2 to 4 years, and one village began its village bank immediately after the first survey. Six “control” villages were identified to receive NGO support for a village bank, such support to begin 1 year after they were identified. In these control villages, villagers were allowed to self-select to be village bank members or nonmembers. Hence, the “old” village bank members in the eight “treatment” villages can be compared with the “new” village bank members in the six control villages. Moreover, differences in the length of time that the program has been available to members in the treatment villages is taken into account to obtain more precise impact estimates. Inclusion of nonmembers in all villages allows for the use of village fixed effects estimation to control for the possibility that the order in which these 14 villages received program support is endogenous.

The potential impact of any credit program will depend largely on the context in which it is implemented. The main premise of group lending is that the rural poor in developing countries are credit constrained, having limited access to formal sector credit. In Thailand, however, the semi-statal Bank for Agriculture and Agricultural Cooperatives (BAAC) practices both individual and group lending and counts 84.5% of rural households as its clients. Hence, it is arguable that rural households in Thailand are considerably less credit constrained than their counterparts in other developing countries. The BAAC's outreach in the impoverished Northeast, however, is smaller than the rest of the country. In the 14 villages surveyed for this study, 63% of village households were BAAC members. Moreover, as is often the case in government-led credit programs, the BAAC's clientele is largely male. Although Thai women have traditionally been active participants in the market place and have enjoyed a certain amount of economic autonomy and power, only 29.5% of surveyed BAAC members were women. Hence, only 29.5% of 63%, or 18.6%, of surveyed households included women who had direct access to BAAC credit. However, 25.8% of surveyed households included women who were in debt to moneylenders. Hence, there is indirect evidence that women are credit constrained and may benefit from greater access to institutional credit. On the other hand, the village bank program studied here makes small loans, with a loan ceiling of 7500 baht (US$300). Average size of a BAAC loan in the survey region is 15,134 baht, with some loans as high as 100,000 baht. Average household wealth in the sample is 529,586 baht, which dwarfs village bank loan size.

The results from this research design indicate that village bank loans are having little impact on most household outcomes in Northeast Thailand, although “naive” estimates of impact that fail to account for selection bias and endogenous program placement significantly overestimate impact.

The remainder of the paper is organized as follows. Section 2presents the empirical model and estimation strategy; Section 3discusses the survey design in greater detail; Section 4presents a brief description of the survey area; Section 5presents the empirical results; and Section 6summarizes the results and draws policy implications.