I worked this week on chapter 6, in which I survey academic studies on the impact of microcredit. In part, I am explaining the grand debate between Jonathan Morduch and Mark Pitt years ago over the the validity of the Pitt and Khandker conclusion that "annual household consumption expenditure increases 18 taka for every 100 additional taka borrowed by women…compared with 11 taka for men.” (That is, microcredit helps families economically, especially when it goes to women.) Muhammad Yunus has indirectly cited this conclusion, via Khandker's book, as showing that the Grameen Bank lifts 5% of its borrowers out of poverty each year.
A bone of contention between Morduch and Pitt was about how well microlenders were enforcing their eligibility requirements, in particular that households owning more than half an acre of land couldn't borrow.
Pitt and Khandker's analysis was founded on the assumption that enforcement was reasonably good. So while households just under the line and households just over probably had similar economic prospects, since officially one kind could get credit while the other could not, the comparison between them could reveal the effects of microcredit. (They studied data from 1991-92.)
Morduch discovered that many households above the half-acre mark actually borrowed despite the rule. To make his case, Morduch displayed graphs showing that the probability that a household in the data set borrowed fell gradually the more land it owned, without any clear break at half an acre. Crunching the data myself, I found that 203 of the 905 households in the data set that borrowed owned more than 0.5 acres before they started with microcredit--1.5 acres on average. (Don't try this at home: in time, I will post the database I built to prep Pitt and Khandker's raw data for analysis.) Back in 1999, Pitt countered that his and Khandker's conclusions did not require perfect enforcement.
I set out to recreate Morduch's graphs and ended up with something different. It turns out that Morduch's graphs include all the households in villages without microcredit programs, whose chance of borrowing was zero. For some reason, the households in these villages owned substantially more land on average. Maybe they were more purely agricultural, farther from the cities, farther from paved roads, so that the microlenders hadn't reached them yet. When Morduch mixed in the zeros from these more-landed non-borrowers, it particularly reduced the apparent probability of borrowing among large landholders.
In contrast, when I made the graphs, I restricted to households that actually could borrow thanks to the presence of a credit group in the village. It turns out that among households in villages where microlenders such as BRAC and the Grameen Bank operated, the more land owned, the more microcredit borrowed--despite an eligibility rule that was supposed to achieve just the opposite outcome.
My first graph is a scatter plot of the raw data with one dot for each household in villages with going credit programs. As in Pitt and Khandker's study, "borrowings" are cumulative from late 1986 to late 1991, adjusted for inflation; if someone borrowed and repaid 1,000 taka, then borrowed 2,000, that counts as 3,000. "Landholdings" are what households owned before they got microcredit, as reconstructed from answers to survey questions about how much they owned in 1991 and how much they had bought and sold in the years before. A vertical line shows the official half-acre limit. If that were strictly followed, all the dots on the right would be squashed down to the bottom:
No obvious patterns here. The next graph erases the dots and shows the moving average of the amount of borrowing as one scans from left to right. I've segregated by gender. One curve shows average borrowings by men in households in villages where male credit groups operated (where the average is taken over all households in such villages that had men, including ones that did not borrow). The other is for women:
The next is the same except that the vertical axis is the probability of borrowing rather than the average amount borrowed:
These graphs surprised me. Richer households borrowed more than poorer ones. This is not necessarily bad: serving richer clients with larger loans may make it more economical to reach poorer people in the same villages--an example of cross-subsidization. And someone whose main asset is an acre or two or riceland in rural Bangladesh is hardly rich by western standards--and might even make better use of the credit. Still, the "mistargeting" of landed families contradicts the public image of microfinance in Bangladesh as targeting the poorest, and probably needs to be better understood.
Update: In response to Asif Dowla's comment (below), here are some more graphs. (The blog system won't allow me to insert pictures in a comment.)
The land variable used above includes cultivable land. I think I tried last summer to break out the land variable by type and got stuck. Below are the graphs done another way; I think they convincingly show the same pattern. They differ in showing the value of land in taka rather the area in acres. This one is for the amount of borrowing as a function of the value of landholdings. The vertical essentially line shows the average value of landholdings for those owning 0.5 acres--so it's meant to serve as a roughly equivalent poverty line:
And this one shows the percent who borrowed as a function of landholdings value:
It's seems hard to make the case that the microcredit lenders were targeting effectively. As I wrote above, this isn't automatically all bad.