Automated Financial Risk Assessment in Automotive Financing Using Explainable Machine Learning

Abstract

Research in automotive financing has explored a variety of data, models, and methods to either improve decision-making capabilities or reduce operational costs associated with risk assessment applications. Despite this trend, the most well-known and widely used product in the area—credit risk scoring—continues to be approached using classic statistical techniques and limited data sources. In traditional scoring implementations, risk signals are generated without thorough evaluation or consideration of auxiliary data that could provide additional insights. These issues call for the development of a credit-risk score generator based on automated machine-learning techniques that can leverage a wider range of macroeconomic and alternative data sources. Although automated financial risk scoring systems represent a breakthrough for the industry, one fundamental characteristic must be added to fully exploit replacements: explanation. Financial scoreboards play a regulatory role in fostering bank stability; it is hence fundamental that users can understand how decisions are made. Making a clear, understandable, and replicable financial risk-scoring tool through machine-learning automation could therefore represent a decisive turning point for the sector. The first step in closing the gap is thus the development of an automatic default-score generator capable of exploiting macroeconomic, industry-specific, and alternative data. Different algorithmic approaches are tested, distinguished on performance and interpretability grounds. The constructed generator sheds light on the most important variables affecting credit-risk prediction in the automotive sector. The described module represents an initial contribution to achieving an automated financial-risk-scoring solution characterized by explanatory capabilities. By applying recent developments in explainable artificial intelligence and integrating them into the validation framework, forthcoming research can provide insurance companies with a multilayered understanding of credit-risk dynamics and patterns.