Improving accuracy of financial distress prediction by considering volatility: an interval-data-based discriminant model

Financial distress prediction models are much challenged in identifying a distressed company two or more years prior to the occurrence of its actual distress, on the grounds that the distress signal is too weak to be captured at an early stage. The paper innovatively proposes to predict the distressed companies by a factorial discriminant model based on interval data. The main idea is that we use a new data representation, i.e., interval data, to summarize four-quarter financial data, and then build a interval-data-based discriminant model, namely i-score model. Interval data makes both average and volatility information comprehensively included in the proposed prediction model, which is expected to improve prediction performance on the distressed companies. A comparison based on a real data case from China’s stock market is conducted. The i-score model is compared with five commonly used models that are based on numerical data. The empirical study shows that i-score model is more accurate and more reliable in identification of companies in high risk of financial distress in advance of 2 years.