Does Saving Increase Micro-Finance Customers’ Ability to Repay Their Loans?

By Marie-Eve G. Augier

Micro-Finance Institutions (MFIs) provide financial services to the poor, typically persons who are undesirable candidates within the formal banking sector. The industry is marked by three major challenges, increasing its scale, depth (to poorer, more remote populations), and lowering costs all whilst meeting its characteristic “double bottom line” of generating profit and improving social welfare within communities (CGAP, 2006). The research question explored here is in line with these challenges. Investigating whether a MFI model including deposit services reduces loan defaults, simultaneously considers increasing MFI liquidity and thus its ability to scale up, reach more clients and lower costs. If true, it also addresses social welfare, arguing for MFIs to provide more services to its clients in a way that does not increase their risk of default, which can lead to further impoverishment, community exile and suicide.

Deposit services could either put a strain on clients’ ability to repay loans (many MFI’s require deposits with lending) or it could provide them with a much needed buffer to aid repayment. To investigate which effect is more salient, I use data from MixMarket.org, the largest, openly available collection of MFI data. Data was available from 1999 – 2014 although not every MFI reported for every year. This left me with 2584 observations over 899 MFIs.

In addition to data on loan loss rates (dependent variable, DV) and average deposit balances per depositor as a share of host country Gross National Income (GNI) per capita (independent variable, IV) of each MFI, I included data on percentages of female borrowers, average loan balances per borrower as a share of host country GNI and MFI age as controls. I expect each control to have an effect on both the DV and IV hence the need to control for them. Female borrowers are less likely to default on loans and more likely to save than men; the higher the loan balance the riskier it is and the lower the ability of the client to save due to higher repayment installations. Lastly, older MFIs are more trusted in their communities and a certain level of trust is needed before clients feel safe depositing their money. This also allows the MFI more time to perfect its loan product and recovery methods.

In my model, I include a lag of the IV to reduce concerns of reverse causality i.e. that the DV may cause the IV rather than the converse, allowing me to say confidently that the IV changed first and thus caused the DV to change. Secondly, I graphed both DV and IV to gauge the relationships across MFIs. I noted that most observations were grouped in the lower left corner, which indicated the need to take the log of IV. After this, there is still significant skew but less than before.

Thirdly, because there are factors specific to each MFI and the country in which they are based that could affect the variables (e.g. MFI business model or a countrywide financial shock); I used a regression technique called Fixed Effects, which controls for such unobservable and incomparable MFI effects and also did one regression with dummy variables for each country. Lastly, because there was very little variation across the DV, I also did a Between Effects estimate that considers time-invariance; however, this estimation typically requires many more controls than available in this model so its results should be taken sparingly.

Table 2 shows the results. The results of the country effects model are shown in the first column. Although the sign of the estimate for the IV is negative as expected, it is not statistically significant. Only the control Young was significant at the 5% level, meaning that there is only 5% chance that this number would have been estimated if there were no true relationship. Column 2 displays the MFI fixed effects model. Unfortunately, this model returned neither the expected sign nor any significance for the IV. In the final column is the between effects model. In this model, the IV is not negative but also significant at the 5% level. The estimate suggests that if MFI1 has an average deposit balances that is 1% higher than MFI2, then its loan loss rate is 0.000064 percentage points lower than MFI2’s loan loss rate. This may seem small, but depending on how large the loan portfolio of the MFI is, this may be a sizable loss. If the loan portfolio is $1 billion dollars, the loss could mean a loss of $64,000. However, in this model, none of the controls are significant.

While I would love to seize on the last model and claim that maintaining a deposit account has a negative effect on default rates, as mentioned previously, between effects models require many more controls which is difficult to achieve due to lack of data thus the estimates might very well be biased. Thus, in conclusion, considering the results of models 1 and 2 both showed no statistically significant effect, I would say that maintaining deposits do not have a statistically significant effect on default rates. Perhaps, the effects theorized in an earlier paragraph cancel each other out across such a broad population or are more salient in say one MFI, community or country given the specific context of the sample population.