Financial Distress Prediction Of Listed Companies Based On Logit Model, SVM, Random Forests And AdaBoost Algorithm

The purpose of this article is to design a more accurate and effective early-warning model for more effective management of listed companies based on listed companies’ financial situation. Firstly, this article researches the relative documents of listed companies. Secondly, this article uses Logit model, SVM algorithm, Random Forests algorithm, AdaBoost algorithm for constructing models of financial distress to anticipate the financial situation of the listed companies. Thirdly, the risk of listed company will decrease because of the early-warning model. Finally, the loss of listed companies and their stakeholders will be minimized.The data used in this article is the (t-3) annual financial statements published by the financial crisis and normal part of listed companies from2008to2012. The financial index used in this article comes from CSMAR database, which uses150financial indexes of short-term liquidity, operating capability, long-term liquidity, profitability, risk level, shareholder profitability, cash flow ability and development capability describing the listed company’s financial situation.According to this atricle, you know,1)random forest and adaboost algorithm is superior to SVM algorithm and Logit model,2)statistical machine learning algorithm is superior to Logit model (the traditional model),3) as a whole, the forecast accuracy in turn is:adaboost (logistic), adaboost (exponential), the random forest and the SVM and the Logit.The innovative features of this article is mainly shown in follows:1)The selection of indicators. All of financial indicators in this article are selected from CSMAR database, then deleting the financial indicators which seriously missing the data, and make the principal component analysis base on the109financial indicators directly in order to preserve the financial metrics information, with each component rotates at the same time to more effective explain. 2) Sample selection. The article random sample the paired sample base on the appropriate expansion ratio, rather than on the size pair.3) The article use much algorithm to construct model and comparative analysis in order to choose the better algorithm of constructing the model of financial distress prediction of listed companies.