by Dean Karlan
But here was the big surprise: It turned out that Green Bank of Caraga had no reason to be afraid! Individual-liability and group liability groups had the same repayment rates across the board. In fact, thanks to the randomized study design, we can say something even more encouraging. Even if our observation was a fluke and there truly is a difference in repayment rates, that difference is almost certainly very small.
The view from the bottom line is pretty good too. Even in the worst case, if clients really are more likely to default on individual-liability loans, the ability of individual liability to attract (and retain) more clients still means a net gain for the bank overall. Money lost to any realistic increase in delinquency would be easily offset by the revenue generated by additional borrowers.
So what can microlenders learn from the Green Bank’s experience? What can anyone learn who wants to do something about world poverty?
First, microcredit has outgrown the stock image that we’ve become used to—women in brightly colored saris, sitting in a circle and talking about their vegetable stands. An accurate portrait of microcredit would include a cast of characters from Philip to Mercy, and everyone in between. To serve such a vast spectrum of ambitions and needs, we need variations on the theme. We in the United States have a smorgasbord of credit options to choose from—mortgages, car loans, student loans, lines of credit for business, credit cards, cash advance on credit cards, and payday loans, just to name a few. Why should we expect a single type to meet the needs of the billions of poor people around the world?
Which brings us to the second point: We need to be curious and relentless. We need to develop new programs, tweak existing ones, and find out what makes them tick.
For microlenders, that might mean making changes to classic group lending, or offering different products to some clients, or trying something completely new. While they experiment, they need to monitor and respond to the results: Does a new approach serve them better than the one it replaces? More important, does it serve the poor better? If we want to make real progress against poverty, we must get used to improving things in quantifiable, proven ways—and then going through the same steps to improve on our improvements. It is, and will always be, an ongoing process.
The good news is that we don’t need to be fatalistic. We can improve microcredit—dramatically!—without demolishing it and rebuilding from scratch. Some parts of the time-tested group lending model work well; let’s find out what they are and hold on to them. As for the parts that don’t, let’s fix them or cast them aside.
What Makes Group Lending Tick? A Game of Trust
So let’s try to get to the heart of it: What really fuels the group lending model? In our project with Green Bank of Caraga, we saw people continuing to borrow and repay responsibly, even after the legal tie that bound the group together was severed. That suggested there was more to the group dynamic than the lender’s contract terms. And, as we saw in the last chapter, when outside rules come up against people’s personal priorities, the rules often take a backseat. So maybe success really has less to do with the rules and more to do with who we are as individuals, or how we interact socially.
It struck me that, if personal integrity and natural social dynamics could induce borrowers to do a good job of screening out risky types up front, there would be no need for group liability. Trustworthy clients would make every effort to repay on time, even without the aid of peer pressure. Right?
I would have liked to ask borrowers point-blank: Are some of you defaulting because you’re just not trustworthy? But you can guess how valuable that would have been.
So instead of asking a loaded question, I set up an experiment with some of FINCA Peru’s loan clients. In each of the group meetings, I asked clients to play a game. First, every person in the room was given three soles (about a dollar). Then the borrowing group was divided randomly into pairs (some people were paired with members of other borrowing groups too), and each person got a letter (A or B). As soon as they saw the identity of their partner, the Bs were sent into another room.
Then I explained the game to the As: “You can keep the three soles, or you can pass one, two, or all three of them to your partner in the other room. I will double anything you give, so if you pass two soles, for instance, your partner will receive four. Then your partner will choose to pass back however much she wants out of what she received.”
Since partners had no opportunity to talk, the game required some implicit trust and trustworthiness. Would A trust B, and pass all three coins? Would A expect B to send back at least as much as was given, and maybe even more? And would B exploit A’s generosity by sending back nothing? Or would she prove trustworthy by playing nice?
Traditional economics can tell us exactly how Econs will play the Trust Game: Regardless of how much she receives, B will not pass anything back, because that is the “profit-maximizing” move. And knowing this, A will not pass anything to B. In spite of this tidy, rational explanation, Humans don’t always play that way. Some people pass because of social norms; others may pass because they fear retribution after the game is over. About three-quarters of the As in the experiment passed at least one sole to their partners, and more than three-quarters of the Bs who received any soles sent at least one sole back.
I wanted to know whether acting trustworthy in the game translated to being trustworthy in real life. Would Bs who sent back more money to their partners be more likely to repay their loan to FINCA Peru one year later?
In a word, yes. After a year had passed, those Bs who had chosen to return more of As’ generosity also repaid more of their loans to FINCA Peru. In fact, I saw something even stronger. Trustworthiness, as far as it was measured by B’s action in the game, appeared to transcend the boundaries of the borrowing group. B’s actions in the game predicted real-life loan default equally well whether A was a member of the same borrowing group or not. That suggests that the game wasn’t just capturing a dynamic between the members of particular borrowing groups, but getting at real and substantive personal characteristics.
But default hinges on many things, not just borrowers’ innate trustworthiness. Though clients’ screening of one another goes a long way in keeping groups free of deadbeats, even the best of us have bad moments. Everyone is susceptible to unexpected shocks. The strength of the group matters when individuals are stretched to their limits.
So even if intrinsic virtues like trustworthiness keep a borrower constantly on her best behavior, there may still be weeks when her business struggles, when her child falls ill, or when she just makes a mistake. To stay on course through those weeks, the group needs to know how to help. Sometimes people need a break; other times they should be called to task.
Like trustworthiness, the group’s unique social dynamic doesn’t stem directly from the liability clause in the loan contract. It arises naturally from the way people interact. If we can understand what social factors make some groups stronger and others weaker, then maybe we can build strong groups without sticking them with group liability. With that in mind, let’s look at a couple studies that explore what makes groups tick.
The Importance of Hats
In the spring of 1999, immediately after finishing my exams at MIT, I got on a plane to Peru to go see the Lanao family. Iris Lanao was the executive director of FINCA Peru (the organization I would eventually work with on the Trust Game study we just saw, and Business Training projects discussed in chapter 5). Her parents ran the Ayacucho branch, which serviced most of FINCA Peru’s clients. The Lanao family cared deeply about their clients and their community, and they were deep thinkers themselves, constantly questioning the way they did things, looking for ways to improve. They were curious, in the best sense of the word. So when I arrived, they gave me the best possible marching orders: Explore.
I had visited the Ayacucho branch before, during my stint as a consultant to FINCA before graduate school, but back then I’d spent my time hunched over an ancient computer, progr
amming a software system. I had noticed group repayment meetings taking place, but spent very little time in them. This time, I dove into the group process, observing, attending meetings to learn how they worked, and talking to clients about what they thought.
I could tell: There was something about the hats.
I played over the meetings in my head and thought about the hatted and unhatted women. They sat and talked in separate circles. Then I imagined a meeting of only hatted women, all sitting in one big circle and carrying on. Would a completely hatted (or completely unhatted) group get along better than a half-hatted group? Would members be more supportive of, or more attentive to, one another? And if so, would they be better borrowers?
In fact, my question was about more than millinery. Hats were just felt and ribbon, but they stood for the very blood of the people who wore them. The hatted women were indígenas, native Andeans. They wore long, thick skirts and plaited their black hair in long, thick braids; and though they could get by in Spanish, they spoke Quechua in their circle. The unhatted were mestizas, women of mixed or European ancestry. They spoke only Spanish, wore blue jeans and makeup, and had modern hairstyles.
Though they sat in separate circles, the women got along just fine. They were always civil to one another. But I wanted more than civility—I wanted to see how social connections made the group tick. It seemed to me that groups whose members were “close” (in some hat-related sense of the word) might have an advantage. If they sat in one big circle together, maybe they would know more about one another, and be able to push each other more effectively toward making payments.
FINCA Peru had an unusual way of combining borrowers that made it the ideal partner for answering this question. It had a kind of randomization built right in. Instead of forming groups on their own, like most microcredit clients, people who wanted to apply for loans simply came to a branch and added their names to a list. When the list got long enough, FINCA Peru broke off the first thirty names into a new group. That meant people were thrown together based solely on when they happened to register—not on their relationships with one another. Consequently, the level of social connectedness among groups varied more or less randomly. This is what economists call a “natural experiment,” or an RCT by dumb luck.
The remaining challenge involved quantifying and capturing information on both social connections and borrowing behavior. The latter was easy: FINCA Peru already kept track of clients’ repayment records. But grasping social connections was not so straightforward. What does it mean in practical, empirical terms to say people are socially connected, anyway?
I settled on two types of connectedness. The first, a culture index, was really inspired by the hats. It was a number between 1 and 8 that captured the “Westernness” or “indigenousness” of each person, based on a few simple observations—language, clothes, and, of course, headwear. A client’s culture score was the portion of group members who shared her culture index. The second, a geographic measure, was the portion of group members who lived within ten minutes’ walk of the client’s house.
The question was whether these kinds of connections between group members led to better performance as clients. After tracking some six hundred clients for almost two years, the answer was clear. Social connections did matter. Clients with higher culture and geographic scores were more likely to make their payments on time and significantly less likely to drop out (or be forced out) of their groups, even if they had missed some payments.
Apparently the improvement didn’t come from tough love alone. Well-connected groups were more likely to forgive delinquents than poorly connected ones. Surveying a year into the project, I found clients were more likely to know the circumstances surrounding a default in the group if they were culturally similar to the defaulter in question. That suggested that clients in a well-connected group were monitoring one another more effectively—specifically, they could tell when a delinquent had a good excuse for missing a payment, and they cut the offender some slack. Of course, too much leniency could have been an invitation for bad behavior; but given that well-connected groups had higher repayment rates, it looks like the clients had it right.
Now, the simple fact that socially connected groups tend to behave better may not be earth-shattering, but knowing how it happens is potentially very valuable. If a lender knows, for instance, that culturally similar clients monitor one another particularly well, then it could actively promote monitoring by constructing culturally similar groups.
Meetings Matter
Even if intrinsic and unchanging characteristics like trustworthiness or hat preference play a significant role in the success of borrowing groups, one of the big reasons for the world’s excitement about microcredit is its promise of personal and group transformation. Clients, whether they arrive hatted or un-hatted, are supposed to learn to thrive together through the process of borrowing—through talking, sharing knowledge, covering for and monitoring one another. Indeed, fostering close relationships within the group is one of the purported benefits of—and is frequently a justification for—the time-consuming practice of holding group repayment meetings as often as once a week.
Do frequent and mandatory interactions between members really lead to success for borrowing groups, no matter who their members are? Weekly repayment schedules dominate the microcredit landscape—but what is so special about weekly? Why not every two weeks? Why not once a month? The answers have big implications for the design of group lending programs. If a different repayment regime could still generate acceptable repayment rates, spacing out the schedule might be an attractive option: It would save valuable time for both clients and loan officers, and would give clients more financial flexibility by reducing the number of occasions for which they need cash on hand to make payments. The critical question is: How does the frequency of group meetings actually impact the group’s success?
This is exactly the kind of relationship that can and should be tested in the field. But, as with the choice between group and individual liability, in practice most microlenders make a decision about repayment schedules (often with little or no information guiding them) and stick with it. The irony is that if lenders choose unwisely, they’re unlikely to realize it: Since most opt for a conservative, labor-intensive weekly schedule, they’re most likely erring on the side of too much caution. Such errors sometimes go unnoticed.
This was the scenario when two development economists, Erica Field and Rohini Pande of Harvard University, approached the management of Village Welfare Society (VWS), a major microlender in Kolkata, India, in 2006. Based on its year-end report, VWS was doing quite well. In eleven years of lending it had grown to serve some forty thousand borrowers—all women—with loans of up to three hundred dollars. Its 22 percent interest rate was competitive in India’s microcredit market. But the most remarkable figure you could find in the annual report was the clients’ repayment rate of 99.1 percent. That’s impressive by any standard; by comparison, the repayment rate for American small business loans in the same year was about 94 percent. (If anything, one could argue that VWS’s repayment rate was too high, a sign that the bank was not funding risky businesses that, on average, could have been profitable and good for growth.)
Still, Field and Pande suspected there might be room for improvement. They also saw an opportunity to understand how—and not just whether—frequent group meetings led to success. With everything running smoothly, VWS could easily have told the economists to keep walking, but happily it heard them out.
VWS’s credit product at the time was based closely on Yunus’s Grameen model: Women borrowed in groups, were liable for one another’s loans, and repaid in equal installments at forty-four weekly meetings. Field and Pande, with Benjamin Feigenberg (now a graduate student at MIT and J-PAL), designed an RCT to explore the relationship between meeting frequency, group dynamics, and client default. They set aside a hundred new borrowing groups and randomly assigned each to a schedule: Thirty groups got t
he standard weekly package, and the remainder got monthly meetings. Over the following two years, during which time most clients repaid their initial loans and took at least one more, they tracked individuals’ repayment records and surveyed to capture group dynamics as well.
The differences didn’t pop out immediately, but they were there. Over the course of the initial loan, it looked like monthly meetings were a free lunch. There was no difference in default or dropout rates between weekly and monthly repayment clients. But over time it became clear that the more frequent meetings had slowly but surely built stronger groups. Five months in, members of weekly repayment groups were 90 percent more likely than their monthly repayment counterparts to know other members’ family members by name, and to have visited them in their homes. After more than a year, members of weekly repayment groups were more likely to socialize together and more likely to say they’d help one another in the case of a health emergency.
The social closeness of the weekly repayment groups was also reflected in their economic choices. About a year after the initial loans had been repaid, the economists set up a game with real money, a lottery experiment, to see how the group members felt about one another. It was a carefully designed lottery, with a nice twist to try to see whether clients simply felt more altruistic toward their fellow group members, or were actually more able to trust them, to share risk.
Clients were presented each with a ticket to a two-hundred-rupee (about five-dollar) raffle. They were told that the raffle had eleven tickets in all, ten of which had been given to people in other groups. Clients could keep their tickets and participate in the raffle as described (with a one-in-eleven chance of winning), or they could give up to nine additional tickets to other members of their own group (which would reduce their individual chance of winning all the way down to one in twenty, but would increase the odds that someone in their group would win).
