A new random subspace method incorporating sentiment and textual information for financial distress prediction

Abstract

Abstract Financial distress prediction aims to provide the early warning signals for corporate governance, which has been widely recognized as a promising way to reduce financial losses. However, non-financial predictive information, such as sentiment and textual information, and the class-imbalance problem were often neglected in previous research. Therefore, incorporating sentiment and textual information into a random subspace method (IST-RS), is proposed for financial distress prediction. Sentiment and textual features are extracted as non-financial features and further integrated with the conventional financial features. To deal with the high-dimension and class-imbalance problems, the ensemble random subspace method is adopted and improved by fusing the lasso regularized sparse method. Experiments on the dataset derived from the China Security Market Accounting Research Database (CSMAR) were conducted to verify the effectiveness and feasibility of IST-RS. The results indicate that the proposed approach enables the performance of financial distress prediction to be significantly improved. Moreover, the proposed approach has outperformed the benchmark methods on high-dimensional datasets, which demonstrates that is suitable for simultaneously solving the high-dimensionality and class-imbalance problems in financial distress prediction.