Using Big Data to Extend Loans to the Unbanked

In economics, the process of managing income shortfalls is known as “consumption smoothing,” and loans, along with savings, are the two options used to manage such shortfalls. In the absence of sufficient savings, the availability of loans is crucial to meet expenses.

The Financial Inclusion Insights (FII) survey for Pakistan in 2015 has some interesting indicators on borrowing behavior. Fifteen percent of respondents had borrowed in the last 12 months. Of those who had borrowed, 85 percent had borrowed from their friends, neighbors or relatives, while 7 percent from a bank or microfinance bank.

In my opinion, this preference for informal loans could be because of no associated interest payments. However, “no interest” should not be interpreted as “no cost.” As Karaivanov et al[1] observed in Thailand, while formal loans are based on physically divisible collateral, loans from friends and family are based on indivisible social-capital pledges. This implies that a partial payment could result in partial loss of the physical collateral on formal loans. However, in the case of social capital pledges, the borrower’s whole social capital might be lost in case of a partial repayment. Therefore, borrowers who have access to both sources, formal and informal, are likely to prefer borrowing from friends and family when the amounts are smaller (less likelihood of a default) and prefer formal loans for situations where the amounts are larger. The FII data for 2013 captures a similar difference in loan sizes in Pakistan as the average loan borrowed from friends and family was $140 (PKR 14,881[2]); the average loan from banks or microfinance banks was much higher at $420 (PKR 44,617). This might imply that those without access to formal loans are likely to be more affected by unexpected falls in income than those who have such access. In other words, those with no access to formal loans might find it difficult to smooth out their consumption when compared with those who do have access to formal loans.

The scale of this vulnerability becomes apparent when one considers that, in Pakistan, the credit worthiness of potential borrowers is usually measured by the mainstream, large banks by analyzing their bank account activity and salary slips, thus, effectively barring those without a bank account or a job in the formal sector. According to the Pakistan Labor Force Survey[3] for 2014-15, 73.6 percent of the non-agricultural labor force was employed in the informal sector. As per Pakistan’s FII 2015 data; 91 percent of respondents old enough for a bank account did not have a bank account.

The idea behind checking bank account statements and salary slips is to ensure that the borrowers don’t default. However, as we can see from the self-reported assessments of unbanked borrowers, 76 percent of respondents[4] returned the loan either “before the date” or “on the date” agreed upon. Based on the above, I would conclude that it is very likely that a substantial proportion of Pakistanis who have the ability to repay bank loans are being denied that facility.

The problem might lie in the way creditworthiness is being measured at the moment, and one alternative to bank history and salary slips could be to use “big data.” The term “big data” refers to data sets that are so big and complex that they cannot be analyzed through traditional data processing applications. These include utility bill payments for thousands of households, and the mobile bill payment data for millions of customers.

In the context of Pakistan’s unbanked, telecom data and utility bill payments could be used to rate potential borrowers on the timely payments of their utility and telephone bills. A recent paper[5], based on data from the United States, concluded that non-financial data such as utility and telecom payments could predict financial defaults. The authors argue that this could be because an individual under financial stress is likely to de-prioritize non-financial payments such as utility bills.

Another reliable source for measuring creditworthiness could be data from over-the-counter (OTC) transactions, which currently dominate mobile money transactions in Pakistan. As per the State Bank of Pakistan[6], 66.1 million OTC transactions were carried out amounting to $2.15 bn (PKR 225.9 bn[7]) at the end of September 2015. FII 2015 data predicts that 50 percent of such OTC transactions are initiated by users sending regular remittance payments to friends and family. It is likely that these individuals earn in cash and then the money is spent in cash by their friends and family as well. In many cases, these earnings are from the informal economy and these OTC transactions provide a crucial formal link that could be analyzed more carefully to sift out unbanked individuals with regular and stable remittance histories.  

The unbanked in Pakistan are devoid of an integral financial security service provided by modern banking through loans. On the other hand, banks are unable to tap a market segment that could potentially payback loans. A probable disconnect seems to be the way in which credit worthiness is being measured at the moment. The availability of big data[8] coupled with the rising sophistication in data sciences implies that this disconnect can be addressed through analysis of non-financial data in a manner that is beneficial for both the banks and borrowers.