A Study on Stacking-Ensemble-Based Corporate Bankruptcy Prediction Combining Stock Liquidity and Financial Economic Indicators

Purpose This study aims to predict corporate defaults by implementing several state-or-the-art machine-learning techniques and assess whether a hybrid feature selection strategy applied to features consisting of both accounting-based and non-financial features enhances the predictive ability.Methods We propose a stacking ensemble model with stock liquidity and financial conomic indicators augmented to accounting-based features to predict default risk. A hybrid approach to feature selection for eliminating less relevant features is also employed to further increase predictive accuracy.Results Experimental results show that our proposed method yields the highest performance regardless of misclassificaion cost considerations, and the results are consisntent with a few recent findings that utilize stacking ensemble models.Conclusion This study shows that predictive accuracy of corporate defualt predition models can be enhanced by both augmenting non-financial fuetures such as stock liquidity and financial conomic indicators to accounting-based features and using a hybrid approach to feature selecion.