The Research On Financial Crisis Prediction Based On DS Evidence Theory And SVM Ensemble

With the rapid development of the market economy, Chinese enterprises have gained more development opportunities. But at the same time, the external environment of the enterprises is becoming more severe and complex, which makes the enterprises face more risks and chanllenges. Because of these reasons, the probability of financial crisis or even bankruptcy of enterprises increased significantly. Enterprises are the basic elements of the market economy and they play an important role in promoting the development of the national economy and the improvement of the living standards of the people. Financial crisis will endanger the banking system and the creditors. Even more frightening is that the financial crisis may lead to economic depression and social unrest. Therefore, it is very necessary and urgent to build financial crisis prediction model to predict financial crisis of enterprises.Research on financial crisic prediction of enterprises receives more and more attention in rescent years. Many prediction models have been developed by the academia and enterprises, such as multiple discriminant analysis, Logistic regression, neural network and so on. These models have their own characteristics, but they also have some shortcomings, and they can not meet the requirements of the modern market economy. So the academia and enterprises are searching for new theories and new methods for financial crisis prediction.Support vector machine is a relatively new financial crisis prediction method. It has strong learning ability and generalization ability and it also has unique advantages in solving the problems of high dimension, nonlinear and small sample. Due to these advantages, Support vector machine is widely used in the field of financial crisis prediction in recent years. Previous prediction models mainly used single classifiers for crisis prediction. But, in recent years a number of studies have been announced concerning classifiers ensemble. Because classifiers ensemble can take advantages of different single classifiers and improve the prediction accuracy of prediction model. Due to its simple structure and flexibility in the measurement of uncertainty, DS evidence theory is widely studied and used for financial crisis prediction. Previous studies ignore an important factor for financial crisis prediction: earnings management. Earnings management is the act of manipulating the financial statement and may influence the authenticity and reliabilityfinancial of financial data in the financial statement. Therefore, financial statement of a company which manipulates the earnings may have different characteristics compared with that of a company which does not manipulate the earnings. So it is reasonable to diviae the companies into two categories according to whether they manipulate the earnings or not. A SVM classifiers ensemble model based on DS evidence theory is proposed in this paper, and earnings management is considered in this model.Financial data in the previous three years are used to predict the current financial situation of companies. According to whether the companies manipulate the earnings or not, all the companies of each year are divided into two categories. Financial data of the two categories of each year are trained on three SVM classifiers based on three different kernel functions.An improved DS combination method is proposed to combine the outputs of classifiers in this paper. The improved DS combination method considers the support degree of the classifier on the test set, the average support degree of the classifier on the training set and the prediction accuracy of the classifier on the training set. The outputs of the three SVM classifiers based on different kernel functions of each year are combined by the improved DS combination method in the first place, and then the combination results of the the three years are combined by the improved DS combination method in the second place to obtain the final results.An empirical study based on real financial data of Chinese listed manufacturing companies is conducted in this paper. The experiment results indicate that SVM classifiers ensemble model based on DS evidence theory and earnings management can significantly enhance the prediction performance for financial crisis prediction.