The Research On Dynamic Fuzzy Integral And SVM Ensemble For Financial Crisis Prediction

The constant development of economic globalization makes the market become more uncertain, the possibility of enterprises facing financial crisis is increasing. Financial crisis will not only hinder the healthy development of enterprises, bring losses to investors, the risk of unemployment will also increase and bank's credit may not be recovered. If a lot of enterprises in the market have suffered financial crisis, which will undoubtedly affect the economic development, financial security and social stability. Therefore, it is very necessary and urgent to build financial crisis prediction model to predict financial crisis of enterprises.The early financial distress prediction method use statistical techniques such as discriminant analysis. After that, with the development of computer technology, machine learning methods have been applied as single classifier to financial distress prediction. In these traditional financial distress prediction techniques, Support Vector Machine (SVM) has been widely used. The method is based on statistical learning theory, which is based on structural risk minimization principle. Support vector machine has strong learning ability and generalization ability. It also has excellent advantages in solving small sample, high dimension and other problems. However, relying on the traditional financial theory knowledge and single classifier to make judgments on the financial situation of enterprises, cannot adapt to the increasingly complex and ever-changing external environment. In the complex and uncertain environment, the fuzzy integral is a very strong reasoning tool. Using single classifier ensemble based on fuzzy integral, not only can take the advantages of each single classifier, but also consider the interaction between the classifier. In recent years, fuzzy integral has become an integrated algorithm which is used successfully in the field of financial distress prediction, and has achieved a lot of research results. Financial distress prediction research generally collects the company's financialdata as an experimental sample. However, the authenticity of financial data is likely to be distorted by the outside world. Some research shows that earnings management is a common behavior. In order to gloss over financial statements, companies are likely to manipulate earnings. 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. A SVM classifiers ensemble framework based on earnings management and fuzzy integral is proposed in this paper.We use financial data in the previous three years to predict the current financial situation of companies and divide the companies in each year into different categories according to whether the companies manipulate the earnings. Then SVM classifiers are trained on the data of different categories. Fuzzy integral based on a new fuzzy measure determination and adjustment method is used to combine the outputs of the SVM classifiers. In this method, a feature of companies' recent financial data that have more valuable information to evaluate the current financial situation is taken into consideration. Also, as the external environment may considerably be changed when using the model trained by historical data to predict the current financial situation of companies, the proposed fuzzy measure is dynamically adjusted by considering the confidence of each single classifier's output, the consistency between each single classifier's output and the diversity among classifiers. In order to verify the performance of SEMFI, an empirical study on real financial data of Chinese listed companies is conducted. The experiment results indicate that the introduction of earnings management and the new fuzzy measure determination method in FDP can significantly enhance the prediction performance.