Financial Distress Model Research Based On The Lasso Regression

With the rapid development of global economic integration, economic systems become more and more complex. While the listed company as an integral part of the economic system, the results of good or bad in operations and financial condition not only has a direct impacted on their government's macroeconomic policy, but also the orderly operation of capital markets.A number of influential status of listed companies may even affect the global economy's financial crisis.Therefore, it is great significance to understand the financial situation, the financial crisis ahead of forecasts, evaluation of business conditions for national regulatory authorities, corporate managers, investors, creditors and the community. We need to establish early-warning model of listed companies financial crisis to protect investors and the interests of creditors and other stakeholders.In this paper, we choose the specially treated (ST) of the listed company that come fromShanghai and Shenzhen Stock Exchange as the research object. We select the first time ST of the 58 companies as the financial crisis in 2006, corporate and non-ST of the 58 companies as financial normal business in Guo Tai'an database.We select financial ratios to build 120 new index system. We build the financial crisis models. The main process is as follows:(1) To choice financial ratios. The former scholars relied on considered, statistics to selected the financial ratios, for example heteroscedasticity, multi-collinearity, etc. These methods to screen out indicators help to improve the processing accuracy of model building, but the statistics is mainly based on the large sample, when the relatively small number of samples does not have the stability. Meanwhile, in the process of dealing with these indicators is a relatively tedious process. It increases the subjective factors that need to experience, knowledge, increases the difficulty of model building. In this paper, the choice of financial indicators do not have pre-inspection, the variable sellection and model building will be unified, so that the target selection and model building in one step.(2) Sovle the equivalence by constructing least-squares regression model and Fisher discriminant analysis model The result shows they have the same recognition rate, also indicates the two models results not very good when the variable did not do in the strict choice.(3) Establish a warning model of financial crises based on the lasso regression, This is the first time that Lasso regression applied to the financial crisis early-warning model.The experimental results show the lasso regression model recognition rate is higher than the least-squares regression model and Fisher discriminant analysis model. Lasso regression is not strict requirements for multi-collinearity, heteroscedasticity. Previously applied principal component analysis and discriminant analysis of some models such as the establishment of indicators for the last coefficient is not sparse, and not easy to model the interpretation of lasso regression not only can improve the recognition rate, and the model indicator coefficient is sparse, and facilitate the model interpretation.