Ensemble Classifiers Financial Distress Dynamic Modeling Based On Time Series Orient

Since the reform and opening, the economy of China has experienced 30 years of continuous development and has got remarkable achievements. However, since the global financial crisis at the year of 2008, China’s economic growth began to slow down. The pressure of economic downturn increased continuously. In the unstable economic environment, many enterprises have to bear the same pressure as well, and many of them fall into financial distress due to unscientific management, some even have to petition for bankruptcy. Meanwhile, the pressure enterprises are facing will be passed to the banking industry to some extent. Enterprises unable to repay the debt because of the financial distress, and the banks’ bad debts increase.and this would bring the whole national economy system normal, huge hidden troubles. In such a situation, the enterprises require a financial distress predict system, which should be complete, efficient, and accurate. And for the bank industry, it also requires such kind of system to help make decision and scale the lenders who may get into financial distress to protect themselves against losses.Financial distress prediction (FDP) has always been a hot spot of concern to scholars. However, most researches are built on static data sample, and accuracy of models built on static data samples may get lower when time pass by and new samples keep added into the test data-set.Against this background, through a large number of literature read, this article summarizes and concludes previous research results. We built 3 models combined time weight methods and multiply classifiers ensemble to deal with dynamic data flow, they are Integration of Batch Weighted Method with Classifiers Combination, IBW-CC; Double Exports Vote Ensemble Method Based on Timeboost and Adaboost, DEVE-AT; Adaboost SVM Based on Time Weighted, ADASVM-TW. This article to listed 466 companies in Shanghai and Shenzhen stock market as a research object, collection time interval in 2003-2012 for two consecutive fiscal years of losses or net assets per share lower than the face value of the shares was the special treatment (ST) sample as a financial crisis, and in principle of same industry and similar size, we collected 466 normal companies as the samples of normal. In the aspect of indexes selection, we chose 42 indexes including solvency, operating capability, profitability, risk level, cash flow ratios, and development ability. Through theoretical analysis and simulation experiments, we proved that all these three models perform better than traditional classify algorithm when dealing with dynamic financial data flow.