A Machine Learning Approach for Estimating Probability of Default in Lending Portfolios

This study introduces an innovative framework based on machine learning techniques to assess the influence of climate change on credit risk within lending portfolios. The proposed framework utilizes a hybrid methodology that integrates bottom-up and top-down approaches to determine borrowers' likelihood of default (PD) across different climate change scenarios. The bottom-up method employs financial and emissions data specific to individual companies. In contrast, the top-down methodology utilizes industry-level data for companies with limited information. The framework utilizes climate scenarios, specifically those formulated by the Network for Greening the Financial System (NGFS), to generate pathways for risk factors and make necessary adjustments to financial data at the company level. Subsequently, the modified data calculates PD utilizing sophisticated credit scoring models. The evaluation of the framework's performance is conducted through the utilization of scenario analysis, benchmarking, and sensitivity analysis. The findings illustrate the capacity of the framework to assist financial institutions in evaluating and controlling credit risks associated with climate change, thereby facilitating the creation of lending portfolios that are more resilient in the context of climate change.