While there appears to be some reason to think microsavings and microinsurance are more effective at relieving poverty than microcredit, this post will focus on microcredit due to its greater popularity and the appearance of a recent World Bank working paper that promises to shed light on its long-term benefits.
Microcredit has received considerable public attention, much more so than many other development interventions. Despite this, there has been relatively little high-quality research on the intervention. In particular there is a lack of randomised control trials (RCTs) compared to many medical interventions; research has relied on methods which give a less accurate idea of how much good microcredit does.
From the current research, we believe that the effects of microcredit are small at best. Until now, however, available studies have been done over a period of a couple of years. This has left open the possibility that, while microcredit has an unclear effect over a couple of years, there may be longer-term benefits that make microcredit a more effective intervention.
A new paper, released in March this year by Shahidur Khandker and Hussain Samad, begins to try to answer this question. Conducted over a period of 20 years in Bangladesh, the research investigates the effect of both past and current microcredit borrowing on outcomes like poverty and income. On top of being done over a considerably longer time period, the current study is also quite large, investigating over 3,000 households. The Economist has greeted its results with considerable enthusiasm, writing: “Bangladesh’s government should look at the study—and stop interfering with Grameen’s efforts to cut poverty.”
The data was gathered in 3 rounds – one in 1991/2, one in 1998/9, one in 2010/11 – by the World Bank and partners BIDS (Bangladesh Institute of Development Studies) and InM (Institute of Microfinance). Households were surveyed to see which microfinance programs they were a part of, for their background (how many people in the household, gender balance, etc.) and for outcomes (household income, labour supply, school enrolment, poverty, etc.).
Then a regression analysis was performed – examining the correlation between microfinance program participation and outcomes like extreme poverty. To isolate causation in a study of this kind, one tries to account for all potential confounding factors and determine the strength of the remaining correlation.
In this study, past credit was considered separately to current credit, so as to isolate the correlation of e.g. past credit and net worth, as opposed to just credit and net worth. This allows us to then see how much of the effect on net worth is down to participants’ having loaned in the past, and how much is down to them having loans out now. If we then observe that the effect of past credit is much higher than current credit, we have evidence that microfinance’s effects take a long time, so the other studies in the area – which are done over a year or two – might be too pessimistic due to their inability to account for the true benefits of microfinance.
One potential confounding factor is attrition: 7.4% of households surveyed in 1991/2 couldn’t be traced in 1998/9, and 10% of households surveyed in 1998/9 couldn’t be traced in 2010/11. This is an issue if attrition is not random: for example, if the households that can’t be traced are ones whose homes were repossessed since they failed to pay off a loan. The authors accounted for this by using ‘inverse probability weights’: roughly, if two households are similar and are surveyed in the first round, and one drops out, then the one that’s left in is weighed twice as much in the second round.The study’s authors also tried to account for household split-off, where some members of a household split off to make their own (e.g. when household members grow up and marry). They only included the households which were surveyed in all 3 rounds, so that the new households that arise from a split-off are ignored.
This seems somewhat problematic, and indeed is different to the method done when only the first two rounds were considered (Khandker 2005), where the two households were aggregated (having checked this wouldn’t bias the outcome). For example, household income per capita will appear to go up in these households simply because some people have left the household, rather than because income has actually gone up. Equally, household net worth may go down because the household members who split off may take some of the worth to start the new household. If either of these happen more commonly to those who take part in microfinance programs then it will look like microfinance increases household income per capita or decreases net worth, even though that is not down to the microfinance program (or at least should not be attributed to it). The reverse could also be observed if these happen less commonly for those who take part in microfinance programs.
Another confounding factor, raised by David Roodman on his blog, is that those who take part in microfinance programs may be more entrepreneurial or better able to use resources, and so would do better than any control group even without microfinance. One way to get around this is to randomise the trial, so that there is no way for people who are disproportionately entrepreneurial to be more or less likely to receive microfinance. However, that is not done here and means the results of the paper must be interpreted with some caution.
Firstly, the authors found that increasing loans to women reduced extreme poverty, and increased assets and school enrolment of both boys and girls, though it did not increase income or expenditure. This is not uncommon; past research of microcredit has given a mix of papers showing a positive effect and papers showing no effect. This research, while showing a positive effect, is relatively in line with the consensus, as the statistically significant effects are small: for example, an increase of loans to women by 10% increased non-land assets by a mere 0.25% (though still large enough to be statistically significant).
What is more interesting in this paper is how much of the change in outcomes is down to current borrowing, and how much is down to past borrowing. In particular, several outcomes were (statistically significantly) bettered by past loans. An increase of 10% in past loans to women increased per capita income by 0.16%, whereas a 10% increase in current loans to women increased per capita income by only 0.04% - too small to be statistically significant. Similarly, extreme poverty was reduced by both past and current loans, suggesting that microcredit brings some people out of extreme poverty in the short-run, and brings further people out of poverty in the long-run.
However, the research suggests microcredit does not have a positive effect on all outcomes over the long-run. A 10% increase in past loans to women decreases non-land assets by 0.12%, and school enrolment by 4-5%. In these outcomes, current loans have a bigger effect than past loans. A 10% increase in both past and current loans will still increase non-land assets and school enrolment. However, borrowing may not increase non-land assets considerably over the long-term and holdings could revert to pre-loan levels if one doesn’t continue to take out further loans.
The new paper gives us a first insight into the long-term effects of microcredit. It suggests that there is some reason to be optimistic about the long-term effects of microcredit on poverty and household income, but equally some reason to think that short-run increases in school enrolment or assets may be short-lived. This assumes that the paper is methodologically sound. Given the small effects, and concerns over the non-randomized methodology noted earlier, perhaps the best thing to conclude from the paper is that there is weak evidence that microcredit has some benefits and some costs over the long-term, but certainly not enough to be convinced of the effectiveness of microfinance to the extent The Economist is. David Roodman goes so far as to say that he is “surprised by how weak [the article] is”.
Still, this is just one paper, and there is plenty of scope for more research to change the picture, and it is encouraging to see more research in this area. Research would at this point likely be better done on microsavings and microinsurance. There is greater scope to expanding savings and insurance coverage and higher potential effectiveness. It remains unlikely, however, that microcredit will be recommended by us in the near future, due to research showing either small or no effects, unlike our current recommended charities. David Roodman in Due Diligence agrees, writing: “less money for microcredit and more for bed nets would be a double win.”