Financial Risk Prediction Based On Time-dependent Covariate Cox Model

In recent years,experts and scholars from various countries have paid more and more attention to the topic of financial distress prediction.Financial distress not only damages the rights and interests of stakeholders,but also costs the entire country and society a huge cost.Therefore,it is very important to establish a high-precision financial distress prediction model.This paper uses the Wind database of 455 companies listed on the Shanghai and Shenzhen stock exchanges from 2000 to 2015 as a data set,and uses the time-dependent Cox model to explore the default risk of Chinese listed companies.Compared with the time-invariant Cox model,the covariate used by the time-dependent Cox model is the panel data that changes with time,which can better reflect the impact of changes in the company’s financial structure on the company’s financial distress.This paper not only considers financial factors,but also considers market factors,macroeconomic factors,and their intersections,adding time-varying intersection factors to reflect the synergistic effect of listed companies’ micro factors,market factors,and macroeconomic factors.In addition,the Lasso method is used to screen out the time-varying factors that have a significant impact on the financial risk of listed companies,and the ADMM algorithm is used to solve such non-smooth problems,which improves the predictive ability of the model.The empirical results show that:(1)The predictive ability of the time-dependent Cox model is better than that of the time-invariant Cox model.For the Cox model that only introduces financial indicators,the prediction accuracy of the time-dependent Cox model is improved by 5.8%compared with the time-invariant Cox model;for the Cox model that adds market indicators and macro indicators,the time-dependent Cox model is compared with the time-dependent Cox model.The prediction accuracy of the invariant Cox model is improved by 5.1%;for the Cox model with the intersection of significant financial indicators,significant market indicators and macro indicators,the time-dependent Cox model has improved prediction accuracy compared to the time-invariant Cox model.9.5%.(2)Regardless of whether it is a time-invariant Cox model or a time-dependent Cox model,for the Cox model that only introduces financial indicators,the Cox model that adds market indicators and macro indicators,and the addition of significant financial indicators and significant market indicators,as well as the intersection of macro indicators For the Cox model,the prediction ability of the model is gradually improved.Among the six models,the most prominent prediction accuracy is the time-dependent Cox model that adds significant financial indicators,significant market indicators and macro indicators.The prediction is accurate.The rate reached89.61%.Compared with the time-dependent Cox model that incorporates market indicators and macro indicators,the accuracy is improved by 8.6 percentage points.There are three main contributions of this paper: First,the time-dependent Cox model is used to build a prediction model for the financial distress of listed companies,which is different from the existing research that uses the time-dependent Cox model to predict the company’s financial distress,which only considers the company’s financial factors.Considering the influence of three different types of explanatory variables: financial factors,market factors and macroeconomic factors;second,different from the existing financial distress studies that only consider the synergy formed by the intersection of micro and macro factors,this paper adds time-varying The cross factors to reflect the synergistic effect of the listed company’s micro factors,market factors and macroeconomic factors on financial distress;And the ADMM algorithm is used to solve this kind of non-smooth problem,which improves the prediction ability of the model.