Using evaluation data to predict loan performance among poor borrowers: The case of BRAC’s asset transfer and microcredit programmes

Motivation Anti‐poverty programmes can work as a comprehensive data source of poor households’ economic behaviour and performance, a resource that is particularly scarce in environments without formal credit scores or households that have minimal involvement in economic activities. Purpose This study examines the potential role of information generated by an anti‐poverty programme on self‐selection by borrowers (i.e., applying for a loan), screening applicants by lenders (i.e., loan approval), and borrower performance in the microcredit market. Methods and approach We apply the logistic regression, Least Absolute Shrinkage and Selection Operator (LASSO), and Random Forest (RF) methods to predict self‐selection, screening, and post‐loan outcomes. Findings We show that the rate of accurate prediction is about 70% for self‐selection and screening. We find that the prediction accuracy rate is 68% for productive use and 91% for repayment difficulty. Objective indicators (e.g., income, assets, age of the household head, savings) stand as the most influential predictors of self‐selection, screening, and post‐loan outcomes. Policy implications Development programmes can improve availability of data needed to predict creditworthiness, suggesting that there could be potential to expand credit access among poor borrowers through partnerships between development agencies and financial institutions.