The Application Of Improved Combination Forecasting Method In Company Financial Pre-warning

The stock market is an important foundation for the modern economic system and the "barometer" of its fund-raising function of the real economy and the return on investment for investors features require that the stock market has a healthy and stable operating state, the listed company as the stock marketimportant components, their development and stability or not, to a large extent determines the operational status of the securities market.While listed companies in their industry, size, capital, and development prospects are in the leading position, but the fierce market competition, there is no "undefeated generals, " the discovery of effective and timely listing of business crisis, both the capital markets, strict supervision and guarantee of stable development, but also protecting the interests of investors.The financial situation of crisis as a business crisis, warning signs, open and transparent disclosure regime for financial data to establish quantitative analysis of the financial early warning system to help provide a conducive, so the theory of how an effective way to create an early warning system to predict the financial crisis and into a businesstheoretical studies of the long hot topics at home and abroad.Domestic and foreign scholars of pre-warning of financial crisis is also quite a lot of achievements, has produced a large number of effective financial early warning models.Domestic research is mainly based on the learning methods used to analyze the foreign country’s real problems.The early Chinese scholars also used statistical analysis techniques, Chen used the single-variable analysis and discriminant analysis, Yu Liu, and Kang Su-hua used the Logistic approach and Shinong Wu, and LU Xian-yi compare the single-variable analysis, discriminant analysis, Logistic methods of various methodsprediction.As for the data mining technology, security, etc. Young neural networks, has enriched the study of China’s early-warning analysis.Hu Yuan Cheng, Tian full text using improved BP neural network model, in order to improve accuracy, and comparison with a variety of models to choose a better prediction methods. Jiang Yanhui using decision tree methods, trying to increase the usefulness of early warning models.Min of listed companies using SVM method to predict the financial plight of the issue, and came to SVM method is superior to neural networks, multiple discriminant analysis and Logistic regression results.Zhong-Sheng and Wang Yu, has developed an integrated two discriminant rules, support vector machine model, and for listed companies, financial distress prediction.Yang Haijun, too reeti out the fuzzy support vector machine in the listed company’s financial early-warning applications.In conclusion, the financial early-warning for the area is now a mainstream is the application of statistical methods and data mining algorithms two broad categories.Discriminant analysis, logistics regression analysis is the traditional application of statistical methods in establishing early warning mechanisms has been frequently achievements, predicted better results, but the reality of strict statistical assumptions of data is almost impossible to satisfy, the assumption can not meet the forecast results areWhat is compellingIn recent years, the rise of data mining techniques such as neural networks, decision trees and so on, assuming that there is no constraint as stringent a traditional statistical methods, but also have better self-learning and self-improvement of the advantages of fault-tolerant, in the areas of early warning is graduallyapplications become increasingly wide, gradually have an alternative to discriminant analysis, logistics regression analysis, statistical methods for applications such as momentum.In this paper, the theory of combination forecasting the idea of improving the combination of forecast modeling to compare neural networks, decision trees are two common algorithms for data mining method, as well as an integrated combination of the two methods mixed model prediction results from the effects of view, a listed company property the crisis of a variety of models to predict the accuracy is generally higher than two years before the crisis; decision tree algorithm to predict the financial crisis as effective as neural networks, but costly time to consider, the decision tree algorithm is more of thesimple and clear, while the combination of the two combined forecasting method to improve the accuracy of the projections are higher than for each individual method.We combine the advantages of the use of their own to construct a reasonable combination of predictive models to achieve the purpose of maximizing predictive power, with a view to early warning for listed corporate finance theory to make a certain contribution to the study.