3 Reasons the Impact of Microcredit May be Bigger Than We Thought
Beginning about four years ago, microcredit as an effective poverty intervention seemed dead, or at least on life support. Along with a couple of earlier studies, the six famous randomized controlled trials (RCTs) that appeared in the January 2015 issue of the American Economic Journal: Applied Economics appeared to show that the hype surrounding microfinance in the 1990s and early 2000s was exactly that–or at least this is how many readers of these studies interpreted them. Effects on household income and consumption fell short of statistical significance in any of the studies. There was little evidence of female empowerment. Microloans did foster entrepreneurialism, but at the cost of other sources of income.
However, subsequent research has hinted that rumors of microcredit’s demise as a poverty intervention do appear to be exaggerated, perhaps substantially, and I want to present three reasons why the impact of microcredit around the world, while modest on average, is almost certainly bigger than we thought it was a few years ago. I will mention the first two only briefly, because they have been the subject of a past blog post, and spend a little more time on the third, which presents newer evidence.
Both the first and third reasons relate to critiques of the microcredit RCTs, and the former of these suggests there is good reason to believe that the average impact of microcredit across all borrowers is bigger than the marginal effect of expanding microcredit to new borrowers, which is what most of the well-known studies actually did. In the six famous randomized trials published jointly, Mexico, Mongolia, and Bosnia-Herzegovina were saturated with microcredit loans when the studies tried to estimate the impact of credit with even more microloans.
Why is impact likely to be higher among earlier takers of microloans than it is among the later-takers in these studies? A math-free summary of the argument I lay out in a response to these studies is this: Suppose the return on borrowed capital among a vast pool of entrepreneurs is a product of two separate things: 1) innate productivity; and 2) luck. And suppose productivity differs between borrowers, yet is stable over time, but luck bounces around from good to bad over the years for everyone. At the time when microcredit first becomes available, it is more likely that a productive borrower (for whom microloans will have a bigger impact) takes a microloan right away—because her productivity/luck combination is more likely to exceed the cost of the newly available microcredit loan. Less productive borrowers are likely to come into the pool only later because the cost of borrowing is only worth it when they happen at some point to get very lucky with economic opportunity. And sure enough, the “point estimates” of impact (while still statistically insignificant) are much higher on average in the three non-saturated countries among the six collective studies (Ethiopia, India, and Morocco) than the saturated ones. That’s reason #1.
Reason #2 is related to Reason #1 but with even more convincing data. Sometimes you don’t find out about the impact of something like microcredit until it goes away. This is just what occurred in 2010 when the government of the Indian state of Andhra Pradesh shut down the state’s microcredit sector due to an economic crisis, eliminating $1 billion of microcredit lending in one fell swoop. Cynthia Kinnan and Emily Breza study what happened from the wholesale removal of microcredit from the economy as a whole, and this probably gives us a better picture of what its average impacts are throughout the economy than small experimental interventions. They find that the loss of microcredit caused wages to fall by 6% (ostensibly from lost jobs in microenterprises and entrepreneurs returning to the labor market). Household income also dropped, resulting in large reductions in consumption. Is the opposite of what happened in Andra Pradesh what occurred over a longer period of time when microcredit was introduced? Hard to say, but perhaps it is a better measure of microcredit impact than tinkering with it at the margin.
Reason #3 is the subject of a 2019 working paper by Mahesh Dahal and Nathan Fiala at the University of Connecticut. In this paper they combine the data from the six famous RCTs in the AEJ Applied Economics issue with another previous 2011 RCT from the Philippines and Fiala’s own recent 2018 RCT in Uganda. Their meta-study compliments a recently published paper by Rachel Meager and the conclusions, if vetted over time by the profession, cast the results of these studies in a substantially different light.
What Dahal and Fiala argue is that all of these RCT evaluations of microfinance are vastly underpowered, in the statistical sense. This implies that the sample size used in every one of the studies is far too small to be able to detect the reasonably sized effects on profits that microloans could conceivably generate. Indeed, they argue that the minimum detectable effect (MDE) of these studies, the effect size that would result in a statistically significant finding of microcredit impact 80% of the time, is multiple times higher than any reasonable impact microcredit could have. They contend this is mainly a result of a much lower take-up of the loans than the researchers expected in the experimental studies. What this means in plain English is that even if microcredit had a substantial impact, these studies weren’t equipped to find it. Perhaps a close equivalent would be a biologist concluding that there is no evidence for the existence of a certain species after a search over a geographic area that is simply too small.
Their next step is to combine the data in a single estimation, creating a small meta-study of the data, in which they weight each of the eight countries equally in their analysis. Although they note that even with the combined data the study remains underpowered, they find that the randomized introduction of microcredit resulted in a 28% to 40% increase in profits, statistically significant at the 1% level simply from “assignment to treatment,” meaning that this impact is averaged over many who actually took the offered loans and many who chose not to.
As in many of the individual studies, they also find that non-entrepreneurial income falls substantially among those randomly offered the loans, although they don’t find this effect to be statistically significant. However, the point estimates of these reductions are large, and so we need to balance the quite substantial impact on microenterprises they find with a loss in wage income, as households taking the microloans divert their time and energy away from working for others and toward working for themselves.
As Dahal and Fiala’s synthesis of the microcredit experimental data is peer reviewed, it will be praised and criticized. But either way it will likely create another important piece to the puzzle about whether microcredit is actually helping its 200 million+ borrowers worldwide, clearly an important puzzle to solve.
Follow Bruce Wydick and AcrossTwoWorlds.net on Twitter @BruceWydick.