Who let the pit bulls out?
November 20, 2020
Extracting Service Quality from Online User Reviews: A Comparison of NLP Techniques
April 14, 2021
JUST is a non-profit micro-finance organization based in Austin, Texas that provides loans to underprivileged communities and minority female entrepreneurs. The organization was founded in January 2016 currently operates in Central and North Texas and is expanding into Houston and El Paso in 2021. JUST lends to borrowers without credit scores or even bank accounts, using a trust-based tech-enabled model using its network of community leaders, i.e. JETAs. Not only do these loans provide relief for people in need, but also integrates borrowers to into the financial system, and help them grow their businesses.1
To accomplish this successfully, understanding the payment behavior of borrowers is crucial. As an analytics and modeling company with expertise in the field of finance and credit scoring, IAMECON analyzed JUST’s historical micro loan data and built a statistical model to evaluate loan re-payment behavior.
Within this scope, we constructed a statistical model to predict the total number of days each repayment to be delayed (i.e. late payments), using borrower characteristics and historical payment patterns. One important mechanism JUST has is the innovative system called JETAs whereby each borrower can have a person called a JETA who is responsible for helping and guiding them through their JUST borrowing. JETAs come from the same communities and are close to the borrowers, which create a social environment that makes the process less foreign and more familiar.
Because JETAs form a grouping which generates our data in a nested fashion, we built a two-level model that incorporates both fixed effects at the individual level and random effects at the JETA level. Specifically, we built a Multilevel Mixed-effects model (MLM), an industry standard model used for estimating such nested effects and, in turn, predicting borrower payment behavior.
According to our MLM results, we find five individual borrower characteristics as significant predictors of days delay: loan amount, income level of a borrower, number of past loans, number of years in business and borrower’s performance on their last loan. Variations in JETA effects on average explain ~ 20% of the unexplained loan-to-loan variation in days delay. In the Shapley diagram above, we display the relative importance of each factor on the borrower’s predicted loan repayment delay.
Our MLM analysis proves to be successful when we examine the accuracy of predictions since ~60% of our predictions are within 20 days of error, without excluding any outliers from our analysis.
