Empirical Research on Financial Crisis Early-Warning Model of in China A-Stock Market
|Keywords||Listed company Financial Crisis Early-Warning Random Forest Lasso|
Financial crisis is the most prominent and overall crisis in enterprises. Due to the financial crisis, many listed companies in China A-stock have got special treatment in recent years. This not only made the companies suffer from huge losses, but also had significantly bad influence on stakeholders and even the market. Information asymmetry exists in China due to its weak efficiency, therefore it has important implications that we use statistical methods to establish a reliable and stable financial crisis early-warning model.Initiated by background introduction and potential contribution, this paper establishs the research framework and potential innovative ideas based on current study. It then systematically reviews the literature research by both domestic and foreign scholars in the area of enterprise financial crisis on aspects of definition, cause of formation, index systems and model building particularly. Based on these, the paper clarifies the definition of financial crises and also the primary approach to index selection from an empirical point of view.The paper then analyzes relative theoretical findings on the random forest (RF) algorithm. Random forest performs data classification or regression by refining the sample information on the basis of non-parametric decision trees and baggingalgorithm without overfitting the data too much. RF is therefore a great approach in dealing with problems such as insufficient prior information and high dimensional data. The paper then elaborates the variable selection function of RF, and points out the biasness of variable importance calculated in RF. This further brings in the conditional inference framework based forest (Cforest), which can provide unbiased variable importance.Based on previous studies and a sample set of234non-ST companies and78ST companies, this paper uses RF and Cforest to select feature indexes and compares the results of them. From a financial perspective, the variable importance calculated in Cforest is more reasonable. After setting the index system and optimal parameters, this paper establishes the financial crisis early-warning model based on RF and evaluates its performance based on confusion matrix. Meanwhile, the model is used in different market conditions, and it is proved to work well, showing a good adaptability and stability. In addition, the paper incorporates Lasso, another high dimensional index selection method with logistic regression to establish another financial crisis early-warning model. Evaluation results show that the Lasso-logistic model performs well, but it is still slightly inferior compared with RF model in terms of prediction accuracy.Based on these findings, the paper forms some suggestions to the potential users of the model, such as the listed corporations, investors and so on. In the end, this paper discusses the further study direction and framework.