Savings Study: Flummoxing Findings

March 05, 2012

For almost three years, I've been citing a single randomized study suggesting that microsavings does good. By Pascaline Dupas and Jonathan Robinson, it found that offering a savings account to market vendors in Kenya increased their investment, income, and spending on average. The account was a commitment savings account, meaning that it was expensive or impossible to quickly get one’s money out once in. This commitment device mainly appeared to help women. This finding of a stimulus to investment lined up with the studies of microcredit found around the same time that also showed higher investment. But in the microsavings study, unlike the microcredit ones, household spending and income also went up. It is one reason I advocate savings in my book.Last October, two new microsavings appeared as working papers. One is marked DO NOT CITE OR CIRCULATE. But it's on the web so how can it not circulate? And does blogging count as citing? Anyway, I haven't digested it yet.The other is by four economists: Lasse Brune and Dean Yang of the University of Michigan, Xavier Giné of the World Bank, and Jessica Goldberg, who is at the University of Maryland as well as CGD, as a post-doc. The experiment was run through Opportunity International’s bank in Malawi. The study’s results are at once encouraging, in corroborating Dupas and Robinson on the benefits of commitment savings, and confusing, in digging into the why of these benefits and producing unexpected answers. (This is the second study I’ve read in the last week involving an Opportunity International affiliate. Kudos to them for collaborating on rigorous research.)Quantitative studies are better at telling us what happened than why. Done right, they measure whether a new drug reduces heart disease; but that does not tell us through what biochemical pathways the effect traces. Case in point: Dupas and Robinson do not know why women offered the commitment savings account earned more. One theory is that it helped them muster the self-discipline they needed to save for important purchases for their businesses. Another, which I often cite, is that it made it easier for them to say “No” to friends and relatives asking for money. But the black box of human behavior is not impentrable. With cleverness, quantitative researchers can peer inside and measure variables thought to transmit cause to ultimate effect. That way they can rule out some theories. This study does that. More generally, it reflects the growing sophistication of randomized trials in development economics. Like the Mongolia microcredit experiment, which tested group and individual loans side-by-side, this one compares two variants of the core treatment: commitment savings and ordinary liquid savings. And it overlays one experiment with another: months after the two kinds of savings accounts were randomly offered to some farmers, the researchers distributed raffle tickets (for a bicycle) to those who were offered accounts. For every 1,000 kwacha a farmer had saved (about $7), he got one ticket. Some people were handed their tickets in private, others in a very public way, so that their savings balances became public information. Again, the split was random. The thought was that farmers whose balances were made public would come under more pressure to withdraw and share their money unless they had the shield of a commitment savings account, as hypothesized in Kenya.The subjects of the study were male tobacco farmers who have been organized by the tobacco companies into clubs of 10--15 members each. Even before the study, the groups were taking loans from Opportunity International Bank of Malawi, for which they were jointly liable. The loans came not as cash, but as in-kind inputs for tobacco farming, mainly fertilizer. Importantly, the standard package of inputs did not provide most farmers with as much fertilizer as they needed to maximize their incomes. "60.4% of farmers were applying less than the recommended amount of nitrogen on their tobacco plots....83.2% and 84.7% of farmers used less than the recommended amount of phosphorus and potassium, respectively."So here you can see the stark facts of the farmers' economic lives. They are poor, living on an average of $1.50/day in purchasing power parity terms…but in fact, their income arrives once a year, not once a day, when they sell their crops the tobacco companies. They could earn more in the long run if they bought and used more fertilizer. But that would require the discipline and foresight to set aside nontrivial fractions of their income at harvest time for use many months later. You can see why a commitment savings account offers such hope as way to help them summon that discipline.For the experiment, some clubs were offered an ordinary liquid savings account, which allowed deposits and withdrawals at any time. After the next harvest, the payments from the tobacco buyers went, if the farmers requested, straight into these accounts. Farmers in some other clubs were offered a commitment savings account along with an ordinary account; on opening, the account owners could specify how much of the tobacco payments would go into these accounts (the rest going into the ordinary account) and when the bank should allow withdrawals. Some clubs were offered neither kind of account.The headline results are straightforward. As with vendors in Kenya, farmers in Malawi appeared to benefit from the commitment accounts. Although only 21% of those offered a commitment account used it (against 16% for just the ordinary account), and although farmers in both treatment groups withdrew most of their tobacco earnings pretty quickly, those offered commitment accounts cultivated 0.42 more acres than those in the control group in the next growing season, spent 16,500 kwacha more on inputs, sold 34,000 more kwacha of tobacco, and thus earned 19,200 more in profit ($132). Monthly household spending was 410 kwacha ($2.85) higher. Those offered only the ordinary, non-commitment accounts generally saw impacts of the same sign---but much smaller, and not of statistical significance. The authors put the benefit-to-cost ratio of the commitment accounts for society at nearly 4-to-1.As I said, the study was designed to shed light also on why commitment savings accounts help people, by measuring impacts on intermediate variables. And here is where things get strange. Those offered the commitment savings account didn’t actually save more than those offered only an ordinary account. In fact the difference between the 21% and 16% take-up rates I just mentioned is within the margin of error. And while those offered commitment accounts did use them, putting in 2,000 kwacha ($14) on average, this represented only 11% of their total deposits at the bank. The other 89% went into ordinary accounts, from which money could be (and generally was) withdrawn quickly. Somehow the commitment accounts made a big difference even though the people obtaining them made little use of them. By and large, the accounts did not tie their hands, since not much money was in them. They did not use the accounts to discipline their future selves or gird themselves against friends and extended family asking for money.The authors theorize about what is going on, but at the end of the day, they don’t understand it. Perhaps having a commitment account allowed a farmer to tell a supplicant that all his money was locked away, even as he quietly kept it in a liquid account. But that theory is undercut by other evidence. People with access to commitment accounts transferred slightly more money to others outside their family. (Whether they transferred more or less to people within their immediate family is not known.) The results of the raffle experiment too are upside-down from the point of view of this theory. People only offered the ordinary savings account---people whose hands were completely untied---did not withdraw funds once their balances were revealed through the public distribution of lottery tickets. But if such people received their tickets in private they did withdraw more. (Depositors with commitment accounts did not save more or less in response to either type of raffle, which is one what one would expect from people who could at least claim their hands were tied.)Why were the raffle results for people with ordinary accounts upside-down? The paper points out that most people actually kept little money in the accounts. So perhaps only those who had significant balances bothered to show up for the lottery. Their attendance, then, publicly signaled that they had holdings even when the raffle tickets were handed to them in private. But you would think this effect would occur equally for the public and private raffles. Perhaps, the authors conjecture, it was offset for public raffles by social competition: people boosted their deposits in the run-up to the raffle, knowing that the savings prowess would be publicly exhibited. The trouble with this theory is that two distinct effects would be cancelling each other out almost perfectly---rather a coincidence.In talking to author Jessica Goldberg about it, she argued against taking those raffle results too literally. Sometimes flukes happen; statistically, one expects them from time to time. If the same group that saw raffle-reduced savings---farmers with access only to ordinary accounts, who got their raffle tickets in private---had also exhibited differences in key variables such as profit and land cultivated, she would take the raffle results more literally. But they don’t, so she doesn’t.I asked Jessica whether qualitative research---you know, asking people with commitment savings accounts why they didn’t use them much and yet invested more in their farms---might resolve the mystery. She said she has done qualitative work in other projects, but is pessimistic that it would help here. If I asked you how your opening of a bank account six months ago affected how much time you spent at work today, would you know? It turns out that qualitative research too is better at eliciting what happened than why.At any rate, we have the good news that commitment savings accounts make a difference for really poor people. I hope with time we’ll better understand why.


CGD blog posts reflect the views of the authors, drawing on prior research and experience in their areas of expertise. CGD is a nonpartisan, independent organization and does not take institutional positions.