Banerjee et al. (2015a) review six experiments randomising access to micro loans. The six studies use one of two different designs, and the specific design affects how we should interpret the results. Each of the six experiments plays on one of two margins. Some randomise access to credit at the neighbourhood or village level (Angelucci et al. 2015; Attanasio et al. 2015; Banerjee et al. 2015b; Crépon et al. 2015; Tarozzi et al. 2015), whereas others randomise credit to marginal clients within all neighbourhoods or lender branches (Karlan and Zinman 2011; Augsburg et al. 2015) using credit scores to identify the relevant sample. It is important to appreciate that in both of these designs, we are learning about the effects of expanding microfinance beyond its current penetration rate; we are not learning about the effect of microfinance on previously existing borrowers.
Karlan and Zinman (2010) sort applicants to their partner microfinance institution (MFI) according to their credit score. Applicants with a score of 60 or above are all offered loans; those with a score below 35 are all denied loans; and those with a score between 35 and 59 are randomly sorted into one group that is offered a loan and another that is not. This implies that we can learn about the effect of loans for borrowers with credit scores between 35 and 59, but we can learn nothing about the effect of loans for borrowers with credit scores above 60, where the effect of credit could well be different (and perhaps larger). Similarly, although Banerjee et al. (2015b) randomise at the neighbourhood level, around 18% of households in their control neighbourhoods obtain a micro loan around the time of their experiment. The fact that the percentage in the treatment neighbourhoods (around 27%) is larger allows the researchers to estimate the effect of microcredit on various outcomes. But this effect is estimated on the 9% of households that are marginal borrowers – i.e., those who would not borrow in the control neighbourhoods but do in the treatment neighbourhoods. The effects on the sample affected by the experiment may well be different from the average effect across all borrowers.
With these caveats in mind, the experiments show that credit has a limited effect on the growth of microenterprises. With regard to enterprise outcomes, Banerjee et al. (2015a) summarise the results of the six studies as follows:
the lack of transformative effects is not for lack of trying in the sense of investment in business growth. There is pretty strong evidence that businesses expand, though the extent of expansion may be limited, and there are hints (eyeballing the pattern of positive coefficients across studies) that profits increase.
These findings are underwhelming in light of the much larger returns to capital found in experiments providing capital shocks through grants. Meager (2018) analyses the six studies reviewed by Banerjee et al. (2015a) plus Karlan and Zinman (2010) in a synthetic manner using Bayesian hierarchical analysis. She sumarises in her abstract that “…the impact on household business…is unlikely to be transformative and may even be negligible”.
