Credit default prediction for micro-enterprise financing in India using ensemble models

Assessment of default risk for micro-enterprise financing is altogether distinct from the financing of large corporations. Credit assessment officers bear dual pressure from a policy perspective to grant more credit to micro-enterprises and also an internal pressure of minimising default to them. The conventional approach of evaluating borrower-centric default risk to micro-enterprises, which uses the ability to pay criterion, has proven to be irrelevant in the absence of a first-generation potential borrower's financial data implying the need to model a set of variables capable of predicting credit default. We ensemble the findings of the multinomial logistic regression, neural network, and CHAID algorithms using the most significant variables developed from lender's package of credit granting framework for credit default prediction to improve prediction ability. We use a database of 3,013 micro-enterprises obtained from a cluster of micro-enterprises who are first-time borrowers of a financial institution based in Delhi, covering a period from 2007-2010. We find that our model is robust as predictive accuracy results confirm its validity and it can be used by policy-makers and the central bank (RBI), which can change the entire philosophy of financing for micro-enterprises in India.