This paper aims to enhance the econometric models mainly used by the financial services firms to predict prepayments of fixed-rate institutional loans. Upon deploying several model types for prepayment prediction on Euro-currency, fixed-rate institutional loans between 2012 and 2020, we found that tree-based machine learning methods significantly outperform logistic regressions in predictive powers. We recovered the expected inverse relationships between changes in interest rates and subsequent prepayments. We also identified other driving features that correlate with subsequent prepayments, like loan costs and changes in business conditions. We ascertained the directional covariance of significant features of non-linear inference models by means of a Shapley plot. Further, we also draw inferences on the prepayment volumes in Euro amounts and timing of prepayments.
This paper aims to enhance the econometric models mainly used by the financial services firms to predict prepayments of fixed-rate institutional loans. Upon deploying several model types for prepayment prediction on Euro-currency, fixed-rate institutional loans between 2012 and 2020, we found that tree-based machine learning methods significantly outperform logistic regressions in predictive powers. We recovered the expected inverse relationships between changes in interest rates and subsequent prepayments. We also identified other driving features that correlate with subsequent prepayments, like loan costs and changes in business conditions. We ascertained the directional covariance of significant features of non-linear inference models by means of a Shapley plot. Further, we also draw inferences on the prepayment volumes in Euro amounts and timing of prepayments.