A hybrid approach for predicting corporate financial distress: Integrating Complex network features and machine learning model

Abstract

As interconnection across sectors and institutions within the financial system increases, the insolvency of corporations generally leads to negative repercussions for the financial health of related firms. This research aims to construct a novel hybrid model incorporating a network-characterized multidimensional financial distress indicator system to forecast corporate financial distress dynamically. Firstly, a volatility spillover network is constructed, and network-based features are extracted using the DCC-GARCH model. Subsequently, experiments are conducted using the hybrid AdaPSO-RF model, which integrates Random Forest, AdaBoost, and Particle Swarm Optimization (PSO) for parameter optimization. The empirical findings indicate that the AdaPSO-RF model exhibits enhanced predictive performance when applied to the hybrid feature dataset, surpassing both baseline and comparable models. The incorporation of network-based attributes results in consistent performance improvements, validating the significance of systemic risk data in forecasting financial instability. The integration of volatility spillover network characteristics into ensemble learning markedly enhances predictive accuracy and resilience, offering substantial assistance for financial risk assessment and decision-making.