I’ve been meaning to read for the last month this new paper by Orazio Attanasio and co-authors, which is the latest in the still small number of studies to carry out a randomized experiment to measure the impact of microfinance. David Roodman was quick to give his thoughts on it in this post, but I thought I’d also summarize it briefly for you and offer my thoughts.
Sample: 1148 women in 40 villages in rural Mongolia
Loans: intended for business use, but about a half are used for household uses. Provided by XacBank, the second largest microfinance lender in the country. Grace period is 1-2 months depending on loan length, and interest rate is 1.5-2 percent per month. Average loan length is around 200 days, and average loan size was $411 for individual loans and $279 for an individual getting a loan through a group.
There are several innovative features of their intervention. They started in 2008 by holding information sessions in the 40 villages and telling them that there was a 2/3 chance that XacBank would start lending in their village during the experiment, and that lending may take the form of either group or individual loans. Women who wished to participate could sign up and were asked to form potential groups of about 7 to 15 persons each. They had to have assets of less than US869 and monthly profits of less than US174 to be eligible to participate. The paper doesn’t explain how this was verified or enforced.
Once about 30 people had signed up in each village, a baseline survey was taken before people knew whether they were getting microfinance or not.
Villages were then randomized so that 15 got group loans, 15 got individual loans, and in 10 XacBank didn’t provide loans to the participating women (they continued providing their regular individual loans to wealthier clients in these villages).
For the group villages, once they were told they would get group loans, women had to form groups following certain rules – these groups in many cases were different from the initial sign-up groups.
All women who had signed up and expressed an initial interest in borrowing were visited by a loan officer and received a first loan after a successful screening. After 1.5 years, 54 per cent of all treatment respondents had borrowed from XacBank: 57 per cent in the group-lending villages and 50 per cent in the individual-lending villages. Other MFIs were also operating, so access to credit was actually quite high – 50% of the control group and 74% of the treatment groups had received microcredit during the time of the experiment.
Follow-up Survey: 18 months after baseline, with 14% attrition rate. Attrition was higher in the individual loan villages. The survey is only on average 5.2 months after the first respondents in a village received a loan (6 months in group villages, 4 months in individual villages).
· Individual and group loans are used in similar ways. In both treatments women report using just over half of the loans for business uses. There is no difference in default probabilities between the two types of loans.
· OVERALL TREATMENT EFFECTS: They find a marginally significant but quantitatively large impact of group loans on the probability of individuals owning an enterprise, and no significant average impacts of individual or group loans on business profits. No significant impacts on household labor supply or overall household income. Food consumption increased in the group villages and there is reduction in amounts spent on cigarettes (a temptation good). In both individual and group loans there is an increase in purchases of large household appliances.
· HETEROGENEITY: they examine heterogeneity of impacts by education level, and by the number of months microfinance had operated in the village for. The latter wasn’t randomized, and they don’t provide any tables to show it was close to random. The heterogeneity analysis finds high impacts on business opening for less-educated women getting group loans. Profitability is higher for enterprises in group villages after more months of microfinance.
· The authors interpret the difference between group and individual loans as perhaps due to “Group discipline may not only prevent the selection of overly risky investment projects, it may also ensure that a substantial part of the loans is actually invested in the first place (instead of used for consumption or transfers to others)”.
· One of the big bugbears facing impact evaluations on microfinance is low power, typically driven by low take-up. The strategy here of having people sign up before hand if they want the loan is useful in increasing power here – the take-up rates of over 50 percent are very high compared to other microfinance studies. The downside is a loss in generalizability – we only obtain effects for the types of people that would sign up for a chance of getting microfinance and when they don’t know the type of microfinance they might get. These might be the more risk-seeking and more entrepreneurial individuals in the population – since more cautious and less go-getter types may wait and see.
· Power is still low – in particular, the authors don’t explicitly test for whether the differences observed between individual and group loans are statistically significant. The standard errors are pretty large and looking at the tables it would seem that in most cases where they find a significant coefficient on group and not on individual, we would not be able to reject that they are the same.
· I worry about studies where the effects are mostly in the heterogeneity - it is encouraging to get some headline results on log food consumption and owning a business here for the average effect, and overall the results mostly tell a coherent story – but given they test the impact on 36 different outcomes across various tables, we should worry a bit about multiple hypothesis testing (although clearly business opening and business profits are perhaps the most obvious ex ante outcomes to look at). It would be good to know whether they specified in advance that education was the key dimension of heterogeneity to look at, or whether they tried several and are just reporting the one that gave more differences.
· The authors estimate ITT effects using a difference-in-difference approach with 2 rounds of data (pre- and post-). As my recent paper shows, this can really reduce power when outcomes are not very highly autocorrelated – an ANCOVA specification or just using the post-treatment data will have more power for less correlated outcomes.
· More time is needed – these impacts are really short-run – only 4-6 months after getting loans on average according to the paper – but 13 months according to David Roodman’s blog post. It is surprising they find as much as they do – and it would be useful to return and measure impacts over 2-3 years.