Share

Working Paper No. 174, June David Roodman and Jonathan Morduch (2009)

This paper attempts to resolve conflicting findings about the impact of microcredit programs.  The authors replicate three leading microfinance studies and find that the evidence for microcredit reducing poverty and smoothing consumption, especially when targeted towards women and the extremely poor, is weaker than generally assumed.

Many of the conclusions about microfinance effectiveness were drawn from three studies conducted in Bangladesh in the 1990s.  Pitt and Khandker (1998) find that microcredit, especially as loaned to women, boosts household consumption.  Morduch (1999) finds that microcredit decreases consumption volatility but does not affect overall consumption levels.  Also using data from the 1990s, Khandker (2005) finds that microcredit is especially beneficial for the extremely poor. 

In this paper, Roodman and Morduch replicate these studies to examine the credibility of the findings.  They focus on the impact of microfinance on household income, explaining that “by a definition often used by program evaluators and academic researchers, the test is whether the interventions measurably improved the lives of the poor, such as through higher or more stable household consumption” (1). 

The early literature on microfinance did not use experimental methods and therefore could not establish causality.  Using a quasi-experimental approach, Pitt and Khandker based their work on three World Bank surveys conducted by the Bangladesh Institute of Development Studies in 1991 and 1992.  In a work judged by the Grameen Foundation to be perhaps the “most reliable impact evaluation of a microfinance program to date,” Khandker examined the fourth round of this survey conducted in 1999 and concluded that microfinance is associated with an increase in annual consumption spending of 18 per 100 taka borrowed by women and 11 per 100 taka borrowed by men. 

However, these findings have drawn some criticism.  Pitt and Khandker used a complicated econometric estimator of the effect of microcredit on consumption that may obscure the actual effect.  In addition, neither of these studies were randomized.  The authors report that “After going through a replication exercise—applying the same methods to the same data as in PK, Morduch, and Khandker—and performing related Two-Stage Least-Squares (2SLS) regressions, we come to doubt the positive results in all three” (3).  They found the opposite result as in Pitt and Khandker, suggesting that there were significant omitted variable concerns in the original work.  In particular, “the endogenous credit-consumption relationship varies substantially by sub-sample, as well as borrower sex, which can explain the seeming gender differential in impact” (3).  The authors further argue that in Khandker, “exploiting the panel dimension does not compensate for the lack of clearly exogenous variation in the treatment variable” (3).  The striking conclusion of this replication exercise is that “30 years into the microfinance movement we have little solid evidence that it improves the lives of clients in measurable ways” (3-4).

In terms of future research, the authors call for more demanding standards in identifying causality and cast doubt upon the corrective powers of sophisticated econometric techniques as used in Pitt and Khandker.  Though the replication does not directly contradict the findings of the three works, the authors “assert, however, that decisive statistical evidence in favor of them is absent from these studies and extraordinarily scarce in the literature as a whole” (40).  Therefore, this paper casts some doubt on the widespread praise of microfinance.