Comparative study on credit risk assessment models for the new energy vehicle industry—predictive analytics using Logit models and BP neural networks based on composite versus single-factor indicator systems

Abstract

The new energy vehicle (NEV) industry urgently requires tailored credit assessment frameworks to address its nonlinear risk characteristics, driven by rapid technological iterations and policy dependency. This study selects 46 listed companies in the NEV supply chain (2019–2023) as samples, innovatively integrating multi-source data including financial metrics, textual tone analysis of annual reports, and ESG ratings. A three-dimensional composite indicator system (“financial robustness–strategic credibility–environmental resilience”) is developed to compare the predictive performance of Logit models and backpropagation (BP) neural networks in estimating corporate default probabilities. Empirical findings reveal: (1) Under the composite indicator system, the BP neural network achieves 81.7% default prediction accuracy, significantly outperforming the Logit model (72.4%), with a 32-percentage-point improvement in identifying defaulted entities; (2) Using single financial indicators, the BP network maintains superiority (58.3% overall accuracy vs 48.3% for Logit), validating its capacity to capture complex risk features; (3) The composite system enhances prediction accuracy by 23.4% (BP) and 24.1% (Logit) compared to single indicators, demonstrating the early-warning value of non-financial metrics. These results suggest that the synergistic application of multi-source composite indicators and BP neural networks substantially improves the precision of dynamic credit risk assessment in the NEV sector, offering methodological support for differentiated financial services and regulatory oversight.