Across developing countries, only 63 percent of adults have a bank account, according to our friends over at the Findex. And we’ve seen a couple of papers with targeted populations that suggest savings vehicles could be good for some development outcomes. So is it time for a big push on banking the unbanked?
According to a recent paper by Pascaline Dupas, Dean Karlan, Jonathan Robinson, and Diego Ubfal: not so fast. Dupas and co. set up a very nice experiment to get some cross-country traction. They go to Chile, Malawi and Uganda. Within each country, they target poorer areas with high densities of unbanked individuals. And then they randomly offer the unbanked folks no frills savings account and assistance opening the accounts (this last bit isn’t trivial because in Malawi and Uganda you need three passport photos and certification from the local village council that you are who you say you are). The bank accounts offered here are free – business as usual in Chile, but fees in Malawi and Uganda are significant when compared to mean expenditure for these folks. Finally, keep in mind that real interest rates are negative.
So what happens? By design, prior to the offer, many folks are unbanked – 74 percent in Uganda, 85 percent in Malawi, but only 26 percent in Chile. When the unbanked are offered a free account some people do in fact open bank accounts – 54 percent in Uganda, 69 percent in Malawi and 17 percent in Chile.
Ahh, but do they use it? Not so much it seems. In Uganda 42 percent of folks use the account at least once, with 41 percent in Malawi and 6 percent in Chile. Dupas and co. go on to define active users as those who make at least 5 deposits in the first two years. Here the numbers are smaller still: 17 percent in Uganda, 10 percent in Malawi and 3 percent in Chile. Chile clearly has take up issues, so Dupas and co. leave them aside for further analysis (we’ll return to them below).
These active users may not be a huge group, but they’re interesting. They average 13 deposits during the 2 year study in Uganda, and 12 in Malawi. And they average about $22-24 per month in deposits. But they take this money out pretty regularly too – overall the usage patterns of these active folks seems to suggest they are using the bank as a place to accumulate cash and pull it out when it reaches some target.
What makes an active user? Dupas and co. look at the correlates and in both Malawi and Uganda the following show up as significant: distance to the bank (negative), years of education (positive) and wealth (positive). But they do note that their R-squared is quite low, so other things are going on here as well. Two pertinent non-results show up here too – in contrast to Dupas and Robinson’s earlier work, gender and occupation don’t significantly predict being an active user.
Going back to the overall effects and focusing on the average effect of offering the account, Dupas and co. do find that it increases bank savings in Uganda by $8.8 (22 percent of the control mean) and reduces savings elsewhere (especially at home) by around $4, with a net effect on savings of about $5. In Malawi bank savings go up by $3.9 (28 percent of the control mean) and the crowd-out is $2.5, leaving us with an insignificant aggregate effect.
Dupas and co. then tackle a number of other potential outcomes, including business outcomes, expenditures, transfers and health and education spending. No significant effects.
These results are sobering. So Dupas and co. ask folks why they aren’t using the accounts. Two explanations emerge. First, people are too poor. And indeed, in Malawi individuals are reporting expenditures of $15 per month, with Uganda coming in at $32. The second explanation is the need for liquidity and easy access to cash.
In a nice effort to contextualize these results and to provide some useful policy guidance, Dupas and co. then situate their results in the context of 16 recent studies that are related. It turns out that their take up rate isn’t that far off from a number of other studies. As noted above, the folks in this study are pretty poor – and likely poorer than the folks we see in a number of the other studies. Studies showing more significant downstream impacts tended to have lower transaction costs (e.g. mobile branches), often had a nudge (e.g. lockboxes, logbooks), and/or a commitment device (e.g. withdrawal penalties). So clearly, as they put it, “no one size fits all” and there is still more work to do to match products to different groups of poor people.
This paper also has a couple of other neat and/or admirable methodological points.
According to a recent paper by Pascaline Dupas, Dean Karlan, Jonathan Robinson, and Diego Ubfal: not so fast. Dupas and co. set up a very nice experiment to get some cross-country traction. They go to Chile, Malawi and Uganda. Within each country, they target poorer areas with high densities of unbanked individuals. And then they randomly offer the unbanked folks no frills savings account and assistance opening the accounts (this last bit isn’t trivial because in Malawi and Uganda you need three passport photos and certification from the local village council that you are who you say you are). The bank accounts offered here are free – business as usual in Chile, but fees in Malawi and Uganda are significant when compared to mean expenditure for these folks. Finally, keep in mind that real interest rates are negative.
So what happens? By design, prior to the offer, many folks are unbanked – 74 percent in Uganda, 85 percent in Malawi, but only 26 percent in Chile. When the unbanked are offered a free account some people do in fact open bank accounts – 54 percent in Uganda, 69 percent in Malawi and 17 percent in Chile.
Ahh, but do they use it? Not so much it seems. In Uganda 42 percent of folks use the account at least once, with 41 percent in Malawi and 6 percent in Chile. Dupas and co. go on to define active users as those who make at least 5 deposits in the first two years. Here the numbers are smaller still: 17 percent in Uganda, 10 percent in Malawi and 3 percent in Chile. Chile clearly has take up issues, so Dupas and co. leave them aside for further analysis (we’ll return to them below).
These active users may not be a huge group, but they’re interesting. They average 13 deposits during the 2 year study in Uganda, and 12 in Malawi. And they average about $22-24 per month in deposits. But they take this money out pretty regularly too – overall the usage patterns of these active folks seems to suggest they are using the bank as a place to accumulate cash and pull it out when it reaches some target.
What makes an active user? Dupas and co. look at the correlates and in both Malawi and Uganda the following show up as significant: distance to the bank (negative), years of education (positive) and wealth (positive). But they do note that their R-squared is quite low, so other things are going on here as well. Two pertinent non-results show up here too – in contrast to Dupas and Robinson’s earlier work, gender and occupation don’t significantly predict being an active user.
Going back to the overall effects and focusing on the average effect of offering the account, Dupas and co. do find that it increases bank savings in Uganda by $8.8 (22 percent of the control mean) and reduces savings elsewhere (especially at home) by around $4, with a net effect on savings of about $5. In Malawi bank savings go up by $3.9 (28 percent of the control mean) and the crowd-out is $2.5, leaving us with an insignificant aggregate effect.
Dupas and co. then tackle a number of other potential outcomes, including business outcomes, expenditures, transfers and health and education spending. No significant effects.
These results are sobering. So Dupas and co. ask folks why they aren’t using the accounts. Two explanations emerge. First, people are too poor. And indeed, in Malawi individuals are reporting expenditures of $15 per month, with Uganda coming in at $32. The second explanation is the need for liquidity and easy access to cash.
In a nice effort to contextualize these results and to provide some useful policy guidance, Dupas and co. then situate their results in the context of 16 recent studies that are related. It turns out that their take up rate isn’t that far off from a number of other studies. As noted above, the folks in this study are pretty poor – and likely poorer than the folks we see in a number of the other studies. Studies showing more significant downstream impacts tended to have lower transaction costs (e.g. mobile branches), often had a nudge (e.g. lockboxes, logbooks), and/or a commitment device (e.g. withdrawal penalties). So clearly, as they put it, “no one size fits all” and there is still more work to do to match products to different groups of poor people.
This paper also has a couple of other neat and/or admirable methodological points.
- Savings at home. This is something I am always worried about when I work on surveys looking at assets. I think people keep a fair amount at home (although in one project in Ghana we found quite a few storing cash with friends). Dupas and co. take this seriously and work hard to measure it (you can find their instruments here – one innovation is asking about “a secret place”). And it seems to bear some fruit: 97 percent of people in Uganda report keeping cash at home (or in that secret place).
- Administrative data versus survey responses. A group of us were working on some survey data and comparing it to bank account administrative data for the same respondents recently. The numbers were not the same (shock!). And so it is for Dupas and co. who note that the average response on deposits across the 3 survey rounds is substantially lower than the average from the administrative data – in Uganda survey response are 72 percent of the admin data and in Malawi they are 31 percent. The balances are closer to the truth, and winsorizing (hopefully in the McKenzie definition) helps. More on in this in a future blog.
- When faced with the really low take up in Chile, not only do Dupas and co. not do follow up quant surveys, they roll out a qualitative survey to understand what is going on. And this leads to some interesting discussion in the paper.
- For some of their pertinent non-results, Dupas and co. go the extra mile and tell us what size of effects they can reject. This helps us understand what kind of non-effect we’re looking at (i.e. zero or I don’t know).
- As noted above, they save the heavy literature review for the end and this is really helpful for thinking of the external validity and policy relevance (including a neat table). The literature they cite up front is mostly geared towards motivation rather than the comprehensive “this is where our contribution fits” discussion.
Join the Conversation