Dynamic effects of microcredit in Bangladesh

By Khandker and Samad, there is now a new study of microcredit and it has a much longer time horizon — twenty years — than the previous “gold standard” studies.  It also finds more positive effects than many of the other treatments:

This paper uses long panel survey data spanning over 20 years to study the effects of microcredit programs in Bangladesh. It uses a dynamic panel model to address a number of issues, such as whether credit effects are declining over time, whether market saturation and village diseconomies are taking place, and whether multiple program membership, which is rising as a consequence of microcredit expansion, is harming or benefiting the borrowers. The paper makes the following observations:

  • Group-based credit programs have significant positive effects in raising household welfare including per capita consumption, household non-land assets and net worth;
  • Microfinance increases income and expenditure, the labor supply of males and females, non-land asset and net worth as well as boys’ and girls’ schooling;
  • Microfinance, especially female credit, reduces poverty;
  • Past credit has a higher impact on income and expenditure than current credit;
  • With higher village-level aggregate current male borrowing, the marginal effect of male borrowing on per capita income gets lower.

The paper concludes that the current microfinance policy of credit expansion alone may not be enough to boost income and productivity, and, hence, sustained poverty reduction.

There is a useful write-up of the paper from The Economist.  In sum, we should up our estimate of the efficacy of microcredit.


Unfortunately the identification strategy used in that paper is questionable. Here is commentary by someone who has looked into the results of this series of studies in detail over the last few years: http://davidroodman.com/blog/2014/04/17/shoddy-microcredit-impact-reporting-in-economist/

I found the second half of the paper, regarding village-level rather than household-level outcomes, to be more interesting than the first. The results there are much more mixed - I was interested to see that more lending in a village tends to *decrease* school attendance (especially of boys) rather than increasing it. I suppose the mechanism there is via job creation, creating a reason to drop out of school?

On a related theme, there's an interesting paper summarized here https://blogs.worldbank.org/impactevaluations/some-unexpected-effects-microcredit-children-guest-post-pedro-aratanha that links expanded availability of microcredit in Brazil to lower school achievement.

What about Latin America? http://www.uclaeconjournal.com/issues/s11/2.pdf Results show that microfinance does not solve poverty by itself. Microfinance provides some improvements for the poor. These improvements are smaller than acknowledged poverty solutions like children education, good governance, strong property rights and free trade. But, while these solutions are worked out, is it too bad to take a painkiller even if you're still sick? Also, microfinance helps to close the gap between the formal and informal economies in Latin America: reduce transaction costs and taxation.

A nice surprise "The study took a three year mean from 2002 to 2004 and found that the average return on equity (ROE) of the top twelve perfor ming MFIs in Latin America was over 33% and the highest was 52%. To put this into perspective, the average ROE for formal banks over the same period was only 11%. Seventeen MFIs boasted a ROE higher than those of Citigroup."

So, it alleviates the suffering of the poor, it can help to reduce transaction costs, it might be useful to tax people in Latin America and produces a nice profit for investors. Can anyone tell why not?

Tyler, I don't think this paper should cause us to update our priors much on the impact of microcredit.

It is great that the authors followed up after such a long period. However, the paper hardly states and never defends its identifying assumptions, the things we things we need to believe about the conduct of microcredit 20 years ago in Bangladeshi villages in order to interpret these results as causal. One of those assumptions is whether the geographical pattern of access to microcredit by gender--whether men, women, both, or neither could get it--was exogenous, i.e., tantamount to random for the purposes of evaluation. That assumption supports the disaggregation of impacts by gender. The other assumption is that those with more than half an acre of land had systematically and exogenously less access to credit, as required under the law creating the Grameen Bank (which states that the Bank shall serve the landless poor, essentially so defined). Jonathan Morduch showed years ago that the data from the study don't support this assumption (http://www.nyu.edu/projects/morduch/documents/microfinance/Does_Microfinance_Really_Help.pdf), that there was tons of borrowing above the half-acre line, creating a strong possibility that treatment was endogenous.

These identifying assumptions could of course yet be true. But in my view, the researchers have a responsibility to fully state and discuss them, so that the reader can form an educated opinion about the possible explanations for the results.

At any rate, the correlation found is very small: an elasticity of per-capita household spending with respect to cumulative borrowing of 0.004. A reasonable, cautious interpretation is that this study corroborates the randomized studies in finding little economically significant effect. The main impact it should have on our priors is to somewhat reduce the worry that the shorter-term randomized trials are missing a big, positive long-term story of impact on average household per-capita consumption.

I also don't find the Economist write-up that useful. In appears to commit McCloskey and Ziliak's "standard error of regressions" in describing that tiny elasticity as significant.

Sorry to be such a grouch.

In a more general sense, I've become skeptical about any such big, broad analyses. Aren't there just way too many knobs to tune the results to get to almost whatever conclusion you want?

Identification, external validity, selection biases, cherry picking metrics, significance shopping, there's literally an endless list of ways to twist conclusions in studies of this kind. I like to think of it as a hypersensitivity of conclusions to precise protocol.

Is my worry / cynicism justified?

Rahul, I think you should be skeptical but not to the point of completely discarding social science, which has taught us many things over the years.

In fact this analysis wasn't that "broad." It followed 1800 real families over 20 years. That's a lot smaller than global statistical analyses of correlates of economic growth.

One reason I rely more on randomized studies (which this one is not) is that there are fewer dials to manipulate. You randomly do something for one group of people and not another. You wait a while. You survey them. You look at whether the averages for one group differ from the averages of the other. The math is pretty simple.

The study at hand is a lot more complex, and does contain a larger number of debatable and potentially arbitrary choices.

I think of randomization as necessary but not sufficient. There's still tons of dials to tweak (I think).

Sometimes, I feel that the randomized studies are the most insidious because the mere fact that they were "randomized" sometimes leads people to trust them far more than they really ought to.

Comments for this post are closed