A Novel Hierarchical Framework for Financial Distress prediction Based on Online Q&A Text

Financial distress prediction (FDP) can effectively identify a company's potential financial crisis, providing critical insights that support informed and strategic decision-making. As markets develop and competition among enterprises intensifies, some companies resort to more concealed methods of financial fraud to maintain market share and attract investors, aiming to enhance short-term profit performance. As for the frequently used FDP approaches, most of them try to discover anomalous patterns based on financial features, which cannot meet the ever-changing fraudulent means. In this paper, we characterize each company more precisely with the combination of financial indicators, governance indicators, and strategic indicators. Also, we incorporate topic features from online Q&A text for the purpose of the public opinion about each company. Based on those abundant features, a novel hierarchical FDP framework has been proposed, constructed by Financial, Governance, Strategy and embedded Q&A layers in total. Each layer employs convolutional neural networks for feature extraction. Experimental results demonstrate that, comparing to the classic machine learning approaches, the proposed framework has high predictive precision, indicating the topic features contain incremental information related to financial distress. In addition, this study possesses substantial practical implications for the optimization of risk warning mechanisms and the facilitation of decision-making processes among stakeholders.