The following is a guest post we’re pleased to share by Billy Jack from Georgetown University’s Economics Department and Tavneet Suri from MIT’s Sloan School of Management
The diffusion of M-PESA in the Kenyan economy has been rapid and deep. While M-PESA was seen early on as a no-frills, if somewhat hi-tech, product aimed at the unbanked, it was initially adopted more widely by the better off. However, over a short period of time the service has spread to households who are poorer, less well connected to the financial system, and located in more remote areas. This pattern of technology adoption mirrors that of other product and service innovations, which are often first used by the better off. But the speed at which the service has reached less well-off Kenyans, and their apparently high valuation of it, is unprecedented, especially in the developing world.
Data we collected in two rounds of a survey of households across most of Kenya (but excluding the sparsely-populated far north and north-east) confirm these findings. In the first survey, conducted in August 2008, we found 43 percent of households had at least one member who had used M-PESA (go here for a summary of the results of the first survey, and here for an update). This share rose to 69% by the time of the second survey in December 2009. Over the same period, on the supply side, in a country with about 850 bank branches (Central Bank of Kenya), the number of M-PESA agents, at which cash can be deposited or withdrawn, increased from 4,000 to 12,000. It currently stands at about 19,000, and counting.
M-PESA’s reach down the socio-economic spectrum is reflected along a number of dimensions. While the representation of all segments of the income distribution in the profile of users has grown, the proportional growth has been highest among those at the bottom. For example, the bottom quartile of the income distribution accounted for just 10 percent of users in 2008, but 14 percent in 2009, while the share of users from the richest 25 percent of household accounted for 34 percent of users in 2009, down from 37 percent in 2008.
Similarly, households that have taken up M-PESA recently differ from those that adopted the service earlier, and from those that had not yet done so by 2009. For example, the level of consumption of non-users was about 40 percent, and that of recent adopters was about 72 percent, of that of early adopters. Despite this balancing of the pattern of adoption, M-PESA remains popular among the well-off: fully 93 percent of households in the top quarter of the income distribution have at least one user.
This same pattern is observed with respect to other indicators of well-being and financial inclusion. In the latest survey, close to all (86%) of households with a bank account also have an M-PESA user – M-PESA is not a service “targeted” to the unbanked. But it is a service used by them – of unbanked households (whose number fell only slightly from 52 to 50 percent of the population), the share who use M-PESA doubled from 25 to 50 percent from 2008 to 2009.
And the representation of rural households, typically poorer and less integrated into the broader real and financial economy, has also increased. While a predictably high share, three-quarters, of urban households used M-PESA in 2009, the share of rural households using the service again doubled from just 29% in 2008 to nearly 60% in 2009. Similar patterns are observed with regard to educational attainment: better educated people are more likely to use M-PESA, but growth is higher among the less-well educated.
Finally, M-PESA is reaching women. While only 38 percent of users were female in 2008, this share had increased to 44 percent by 2009. Amongst adults over 18 years of age, the share of men using M-PESA saw a healthy jump from 25 to 54 percent between 2008 and 2009. But the share of women using the product leapt in comparison, from 15 percent to a level approaching gender parity of 41 percent.
In sum, M-PESA is that rare product that appears to be valued both by those with better opportunities and higher incomes, as well as by the less advantaged and the vulnerable. It is by no means used universally, and barriers remain to its adoption (see below), but it appears to be drawing individuals into a financial network in promising ways.
The fact that people are willing to pay the not-insignificant fees to use M-PESA is evidence that it is valuable. But we have also tried to examine the ways in which M-PESA adds value to people’s lives. Two particularly features of M-PESA are primary candidates for the source of value: first, the technology drastically reduces the costs of sending money over long distances, so could have a positive impact on the ability of households to send and receive remittances – a potentially important source of regular or incidental income in a country with a large number of (internal) migrant workers. And second, in a country with high levels of both violent robbery and police extortion and corruption, the safety, confidentiality, and convenience that M-PESA offers might allow individuals to better manage their day-to-day finances, including their ability and incentive to save.
Our results indicate that M-PESA adoption is associated with a higher frequency of remittances, both as senders and recipients. For example, comparing 2008 and 2009, we found that of those who were M-PESA users in both years, the share who sent any remittance (by M-PESA or otherwise) was little changed between the two surveys, rising from about 66 percent to 69 percent. Similarly, of those who were M-PESA users neither in 2008 nor in 2009, the frequency of remittances sent was relatively unchanged, and indeed fell somewhat (from 26 to 18 percent). However, among those who adopted M-PESA between the two surveys, the frequency of remittances sent jumped from 37 to 55 percent. A similar pattern was observed for remittances received – early adopters and never adopters saw little change, while the share of recent adopters who received a remittance doubled.
We cannot infer from this correlation that M-PESA causes remittances to increase – it is quite likely that people adopt M-PESA, at least partly, when the need to make or receive a remittance arises. However this in itself is evidence of the value of M-PESA. (We argue that it is less likely, though not impossible, that some third factor is affecting remittance behavior and M-PESA adoption coincidentally.)
If M-PESA does in fact facilitate remittances, do we see evidence that this leads to improved household financial management? Our tentative answer is yes: it appears that households with access to M-PESA are better able to protect themselves against downside risks associated with job loss, harvest or business failure, poor health, etc. In particular, we find that for M-PESA users, consumption in general, and food consumption in particular, does not tend to fall in the event of such losses. We have employed a number of statistical techniques to identify a causal connection from M-PESA use to improved risk sharing, but we continue to refine these estimates.
Also, it is difficult to tell from the data exactly how such improved risk management is effected. The obvious channel is, we anticipate, the improved ability of individuals to (literally) call on others to help in times of need. Such calls might go out sooner than they otherwise would have, because the fixed costs of sending funds is small (compared to the cost of catching a bus half way across the country); the calls might go out to people further away, and less likely to have suffered the same negative event; or they might go out to out to a wider pool of potential friends (who are more likely to stay friends for longer, if they are not called on too often!).
Alternatively, M-PESA might improve an individual’s ability to protect herself against risk if it gives her better access to emergency loans. This is closely related to the remittance logic though – better access to loans might arise because they are taken out earlier, with lenders who have not suffered similar downturns, or with a variety of lenders, each of whom has enough funds and is willing to extend credit. (It is planned that M-KESHO, a new financial product offered jointly by Safaricom and Equity Bank earlier this year, could be a formal source of both insurance and credit, extended via the M-PESA network.)
The final way in which M-PESA might provide financial protection is by facilitating self-insurance – that is, precautionary savings. Although we have yet to devise a strategy to distinguish between these potential mechanisms, we can report some evidence on the use of M-PESA as a savings vehicle. M-PESA was reported as being used to save money, both in 2008 (by 75 percent of users) and in 2009 (by 81 percent of users). The definition of “saving” however was quite limited, reflecting whether an individual or household kept money on their phone for more than 24 hours. However, in a country in which individuals often have little or no money on their person, either due to safety or liquidity constraints, this share seems significant.
Perhaps the strongest indication that M-PESA might provide a savings vehicle that facilitates insurance against downside risk is our finding that, when asked if and why they used M-PESA to save, the share of M-PESA users who responded that they used it to save for “emergencies” increased from 12 to 22 percent from 2008 to 2009. Whether savings do in fact lie behind our preliminary risk-sharing results, or if these results are driven more by remittance and/or credit market stories, remains to be seen.
In any event, users appear to be getting used to M-PESA and seem to be more at ease with the service. While in 2008 65 percent reported trusting the M-PESA agent with whom they transacted, by 2009 this share had increased to fully 95 percent. This in turn likely reflected continued improvements in the behavior and performance of agents, reflected in our findings that agents performed their jobs according to specified procedures more diligently (95 percent asked for an ID in 2009, compared with 77 percent in 2008), and that they were less likely to run out of either cash (5 percent in 2009) or e- float (4 percent) than in 2008 (when the rates were 16 percent and 7 percent respectively). In an environment of a rapidly expanding agent network, in which the average quality of the pool of potential agents could well be falling, this improvement in agent performance is particularly striking.
Finally, one reason that users might be reporting fewer problems with agents is that they might need them less. Safaricom has introduced an increasing number of bill-pay and other transaction options that allow e-float to be used as a proxy for money. Those with higher incomes, who tend to have electricity, water, and other utility accounts, as well as a higher opportunity cost of time, appear to value these options significantly.