Bankruptcy forecasting — Market information with ensemble model

Abstract

We introduce an innovative Ensemble model for predicting firm bankruptcy. This model enhances prediction performance by integrating Boosted Tree, Random Forest, $k$-Nearest Neighbor, and Neural Network models within a stacking structure. Our model incorporates an extensive set of asset-pricing factors, extending beyond traditional financial ratios. The empirical results highlight that market information measuring the equity return, volatility, dividend, downside co-movement, and liquidity demonstrates the strongest predictive power for firm bankruptcy. Our findings offer strong empirical insights for Merton’s credit risk modelling framework. Further, our model notably outperforms benchmarks in the one-, two-, and three-year-ahead testing-sample forecasting of firm bankruptcy for U.S. public companies.