Microfinance is the development panacea of the 21st century. Or so we might believe by the hype and eye-popping expansion of microlenders across the globe. But does microlending actually deliver on development and poverty alleviation?
Amazingly, we have almost no idea. There’s been little hard evidence to go on, even after 20 years of a global microfinance explosion.
Until now. A number of projects, many coming out of IPA or J-PAL, are starting to show experimental results. Dean Karlan presents a new research paper here in Shanghai that (by my reading) damns microfinance with faint praise.
Dean and Jonathan Zinman run a randomized trial of a micro-lending program to small entrepreneurs in Manila–loans equal to about a tenth of their income.
What do they find? Not much in the way of business success; there’s little effect on incomes, savings, and assets (and no impact on more subjective happiness measures). Rather, the number of household economic activities actually decreases, as does the number of household members involved.
They suggest an explanation: after the loan, junior household members switch from the family business into schooling. Loans promote education, not business. So microfinance is good for the household, just not the way we think.
This seems plausible, but (as Dean and Jonathan note) we have to take the schooling result with caution. It is small and somewhat weak, and only appears in the families of a subset of loan recipients –males.
If we run 100 regressions, 10 are going to show up as significant at the 10 percent level, especially as we delve into subgroups. Dean and Jonathan account for this, and do much better than 1 in 10 in their results. This uncertainty, however, is the big argument for more replication. Theirs is just one data point (and fortunately we can look forward to several more in the next few years).
With a single data point, it’s cavalier to suggest all of microfinance is damned with faint praise. But it’s better than pushing microfinance with virtually no data–the state of the world for the last twenty years. I wonder if microfinance blogger David Roodman would agree?
Listen to Dean discussing the results with BBC World.
7 Responses
“If we run 100 regressions, 10 are going to show up as significant at the 10 percent level, especially as we delve into subgroups. Dean and Jonathan account for this, and do much better than 1 in 10 in their results. This uncertainty, however, is the big argument for more replication. Theirs is just one data point (and fortunately we can look forward to several more in the next few years).”
This is not really true. One regression is not just one data point. The “one regression” may include thousands of pieces of data, even millions. Suppose you run one regression that includes a million individuals and find the results will only occur just by pure random chance only one time in one trillion — and the results are also very economically significant — and the underlying assumptions of the regression are very reasonable. It’s not true that that’s weak evidence because it’s just one regression.
Other papers providing experimental evidence on microlending’s effects:
Experimental Evidence on Returns to Capital and Access to Finance in Mexico
David McKenzie, and Christopher Woodruff#
April 4, 2007
Expanding Microenterprise Credit Access:
Using Randomized Supply Decisions to Estimate the Impacts in Manila*
Dean Karlan Jonathan Zinman
Hi Chris. I agree generally with the tenor but would add a few important caveats. First, as I note in my own post about the study, the sample has an average income of about $15,000/year and is more educated than the United States. There is not much poverty to be eliminated. To an extent, high incomes are inherent in Karlan and Zinman’s approach of introducing randomness into credit scoring, a filtering technique that is uneconomical for the truly poor who must be lent to through groups (to be economical). Average income in K&Z’s related South Africa study was ~$8,000/year, I think.
I did not notice any adjustment for “multiple hypothesis testing,” so it seemed to me that some of their significant results were attributable to chance. But of course statistically insignificant results are still scientifically significant, which is the point of your post; lack of significance is not a criticism.
The JPAL study of microcredit in Hyderabad gets much more to people living on $1/day (my review here). ~15 months on, it finds effects on business formation and profits but none on bottom-line indicators such as income, consumption, and health.
Also worth noting is the new Dupas and Robinson impact study of microsavings, which finds benefits for women in Kenya. (My review here.)
Bigger caveat: microcredit, microfinance more generally, can help the poor even if it does not increase average incomes. It can help them smooth their typically volatile incomes and deal with crises against which they lack insurance. This can give them more control over their circumstances, which fits Amartya Sen’s definition of “development as freedom.” This is why in my book and blog I am looking at microfinance through multiple perspectives. Microfinance probably does not live up to the mythology about microenterprise as a ladder out of poverty. But perhaps some forms of it are doing respectably by the standards of foreign aid (public and private)?