Improving financial distress prediction using textual sentiment of annual reports

Bo Huang,Xiao Yao,Yinqing Luo,Jing Li

DOI: https://doi.org/10.1007/s10479-022-04633-3

IF: 4.82

2022-03-15

Annals of Operations Research

Abstract:An accurate prediction of financial distress is beneficial to investors and allows banks and other financial institutions to build an early warning system to avoid risk contagion. This study investigated financial distress prediction using textual sentiment extracted from listed firms’ annual reports in the Chinese market. The sentiments reflected by the firms’ management discussions and analysis (MD&A) sections and audit reports were extracted separately through the application of deep learning algorithms. We found that the sentiment score extracted from MD&A sections was more optimistic compared with that extracted from audit reports. Moreover, the experimental results demonstrated that the modeling performance was significantly improved with the incorporation of textual sentiment scores, and the inclusion of sentiment from audit reports lead to a more significant incremental improvement than that from the MD&A sections. However, when both sentiment scores were included in the modeling input, the improvement in predictive accuracy was insignificant compared to the model using audit report scores only. Our study highlights the predictive power of textual information in annual reports, and shows that the textual sentiment of annual reports should be applied in distress modeling. The results provide implications for the utilization of soft information in credit risk modeling in the context of Chinese market, and such application can be further explored in other areas of operational research studies.

operations research & management science