系统性风险与企业财务危机预警——基于前沿机器学习的新视角

本文采用Logit回归模型以及随机森林模型、梯度提升模型等前沿机器学习方法,深入考察系统性风险指标对我国企业财务危机的预测能力。结果表明,系统性风险对中下游企业的财务危机具有显著的预测能力,而基于因子分析构建的系统性风险指标,结合随机森林模型可取得更好的预测效果。本文进一步区分财务危机的不同成因并发现,基于随机森林模型和Logit回归模型的预测框架能够对我国大多数财务危机事件进行有效预警。在此基础上,本文对我国上市企业监管提出相关建议,从而为完善金融风险处置机制提供一定参考。

Systemic Risk and Corporate Financial Distress Forecasting from the New Perspective of Machine Learning

The implementation of a financial security strategy was made a high priority in the 14^th Five-Year Plan, which includes the aim to “improve financial risk prevention, early warning, handling and accountability systems.” Frequent black swan incidents have accentuated the shocks of systemic risk on global production activities and enterprise financial stability. Thus, using systemic risk indicators to improve predictions of financial distress is of academic and practical value.
This paper contributes to the research on financial distress forecasting. First, few studies consider the potential impact of systemic risk on production. Thus, we introduce systemic risk indicators into our prediction of financial distress to provide a more comprehensive analysis. Second, research into financial distress mainly focuses on bankruptcy events, although reductions in corporate valuation and creditor losses typically occur before bankruptcies (Gupta and Chaudhry, 2019). Instead of bankruptcy events, we regard special treatment (ST) designation as a signal of corporate financial distress, as it increases foresight and enables the early warning of potential corporate crises, thus preventing huge losses for firms and creditors. Third, current prediction models of financial distress mainly forecast bankruptcy events only one month or one quarter ahead, and thus, they cannot provide early warnings. Instead, we predict the risk event one year in advance, which gives regulators sufficient time to intervene. Finally, conventional parametric models are typically applied for predicting bankruptcies, whereas we utilize newly developed machine learning models to improve the accuracy of our prediction, and we provide references for optimizing the early warning system for corporate financial distress.
Our sample consists of 3,806 Chinese listed companies, with the time period spanning from January 1, 2010 to May 31, 2020. All of the data are obtained from the CSMAR and Wind Databases.
First, we run logit regressions and find that systemic risk indicators serve as powerful predictors of the financial failure of firms, particularly those in mid and downstream industries. Next, a relative importance analysis suggests that the predictive power of systemic risk indicators is independent from that of financial information and market performance. As the partial dependence plots show, the probability that a company faces a financial crisis rises sharply in response to its systemic risk level reaching a certain threshold. We also apply the out-of-sample test and out-of-time test methods proposed by Petropoulos et al. (2020) to compare the predictive abilities of the conventional logit model and our newly developed machine learning model. The results suggest that systemic risk indicators improve the prediction accuracy of these models and that the random forest model outperforms other predictive models, thus highlighting the relative advantages of machine learning methods in financial distress forecasting. In addition, the combination of systemic risk indicators derived from factor analysis and random forest is better suited to the forecasting system in China than other construction methods of systemic risk indicators and predictive models. After we remove the samples in which the distress is due to financial fraud, shrinking net worth, or sudden operating losses, the prediction framework based on random forest and logit regression provides effective early warnings for most financial failure events in China.
Based on this, our study offers three policy implications concerning the listed company regulations in China. First, systemic risk indicators should be included in the early warning system for the financial distress of enterprises to improve the mechanism for the long-term prevention of financial risks. Second, regulators should apply discriminative supervision to different industries. Finally, from a technical perspective, the introduction of newly developed machine learning methods can conduce to a prudent regulation mechanism. This will not only improve the forecasting system for financial distress but also enhance the financial risk disposal mechanism.