Why is it that microfinance practitioners consistently argue that microfinance has large impacts on borrowers, while impact results from randomized controlled trials reveal only very small effects? In a paper that is newly forthcoming in the journal World Development, we try to answer this question. My co-authors Ram Rajbanshi (a former student and microfinance practitioner in Nepal) and Meng Huang (an econometrician and recently graduated doctoral student at UC San Diego who now works at Freddie Mac) develop a simple model that explains the discrepancy and then we test it with microfinance data from Nepal. What our model illustrates is that much of what practitioners observe about microfinance is something akin to an optical illusion. Practitioners typically base their assessments of microfinance on how much better borrowers seem to be doing after they take microfinance loans. The illusion occurs because the timing at which microfinance borrowers take loans is not random; they take loans when they have opportunities, opportunities that would have paid off to some extent even if they hadn’t taken a microfinance loan, but have a bigger payoff when they do.
But we demonstrate that the bias not only resides on the side of the practitioners. The model in the paper reveals that many of the recent studies carried out using randomized controlled trials are also likely to be afflicted by their own problems, which cause them to estimate microfinance impacts that are likely to be smaller than the overall impacts of microlending. The reason is that, with a few exceptions that we note in the paper, most of the recent rigorous impact studies on microfinance are carried out in areas where microfinance lending already exists. As a result, the studies are forced to study impacts on marginal borrowers who newly opt to take microfinance loans for the first time. But there is good reason to think that impacts on these later-takers of microfinance are not a good measure for overall impact of microfinance across the general population (what economists call the “average treatment effect”). The reason is that the productivity of the small loans comes from two different sources: 1) the general productivity of the borrower, which doesn’t change much over time; and 2) economic opportunities faced by the borrower, which may vary considerably over time. These really productive borrowers (#1) are more likely to be the first borrowers to take microfinance loans. This is because the economic opportunities that existed when loans first became available didn’t need to be extraordinarily high for a microfinance loan to make economic sense for them; even with a normal level of opportunity, these really productive borrowers would benefit after paying off a loan. Later borrowers, however, are likely to be those who only take loans when they face greater than normal levels of opportunity (#2), which perhaps nudge them over the borrowing-decision threshold. As a result, we would expect impacts to be bigger on earliest borrowers, and studies that rely on impacts to later borrowers to underestimate the average impacts of microfinance.
We test this idea with some data Ram collected in Nepal, which was novel data for a couple of reasons. The first was that when the lending organization which we studied began operating, none of the entrepreneurs in the six villages we studied had previously had access to microfinance. Moreover, the borrowers were microfinance-starved when microfinance became available; the take-up rate when it was introduced during the mid-2000’s was extraordinarily high, 51% of our random sample. This made the the region ideal for testing the idea in our model, because we could actually measure the impact of microfinance across all of the people who took up loans from this first microfinance lender in the region. What we find is that about 75% of these “before-and-after” observations by microfinance practitioners is an illusion, although the 25% of impact that remains appears to be larger and more significant than what is reported in some of the recent randomized control trials. Furthermore, our findings seem to parallel quite closely with a new study by Esther Duflo and her co-authors in Morocco, arguably the best study to date, a study that uses a randomized controlled trial, but in an area previously unserved by microfinance. Both of our studies find considerable causal impacts on business expansion, but little impact on consumption.
There seem to be a couple of lessons to be learned by this exercise. First, when we are dealing with an issue as serious as poverty alleviation, we need to avoid the hype and focus instead on what has been rigorously demonstrated to be true. And if nothing has been demonstrated yet, the bandwagon we need to get on is the one that insists on serious research for investigating the true impacts of the program. Second, even seemingly “bomb-proof” impact evaluation studies may not be as bomb-proof as they first appear. Underneath most (pretty darn good) scientific studies there are subtle flaws, flaws that are only discovered by other researchers later on down the road. A study may offer real insights even in the presence of these flaws, but what once seemed so convincing, often isn’t just a few years later. But that doesn’t mean that we don’t apply the most rigorous methods at our disposal to understand to difficult questions. Through doing this we will (hopefully) converge at a reasonable picture over time of where the truth lies.
- 10 Reasons Fair Trade Coffee Doesn’t Work
- New Ways to Think about Christmas Giving