Quantifying & Predicting Prepayments in the Microfinance Environment

 

By Nandan Sudarsanam and Dibu Philip, Indian Institute of Technology – Madras

Abstract:

Financial institutions that lend to customers are interested in understanding and predicting repayment patterns. Two critical subjects of interest in this context are the advances and delays in payment with respect to the stipulated repayment schedule. In this research we concern ourselves with the advances in payment, also referred to as prepayment, of loans on a time-quantum scale. While this translates to interest relief for the customer, it could potentially result in a loss for the institution, albeit small. More importantly, from the institution’s perspective, it could indicate a form of preterm attrition, a sign of the unsuitability of the product or the customer’s preference of a competitor. This could in turn lead to pre-closures, and even an overall attrition of customers.

The contributions of this research are two-fold. First, we present a framework for quantifying prepayments on a fixed scale over the duration of the loan. Second, we recommend and demonstrate the use of a machine learning technique called Temporal Difference (TD) Learners to improve the performance of predicting the customer’s prepayment state in the future. TD Learners work with traditional predictive modeling techniques to improve their performance in environments of sparse data. The recommended approach shows an overall improvement in predictive capacity over the conventional approach of using only the predictive model. Specifically, with the sample data set, we find that at best the proposed method is 57% better than the traditional approach, and at worst is indistinguishable in performance. We discuss the specific suitability of such an approach in the microfinance context, where institutions could be looking at unexplored products, demographics or locations and thereby operate in environments which are not data rich.

Download: Full Paper | Information Brief

Back to NSE-IFMR Finance Foundation Financial Deepening & Household Finance Research Initiative