Application Of Stock Returns Forecasting Based On Wavelet Denoising And Clustering Pattern Mining

A large number of data and the uncertainty factors is a remarkable feature of financial management.Because of the massive financial data,based on the traditional statistical models which are too much assumptions to gather in practical application on.Data mining is a new technology of statistical decision-making which rose in the mid 1990s.Its process is to find useful model in mass data,and its purpose is to use the pattern that help explain the behavior or predict the future.It provides useful information to people with easy understand format.Stock market is a very complicated system which affected by policy,economy and psychological factors.It has typical complex uncertainty characteristics.At the same time,stock time series is essentially nonlinear, nonstationary and low signal-to-noise ratio.Data mining is to find some of the sequence of essence rule.But high noise on the existence would weaken the significant laws, on the other hand, it may provide some false information, thus seriously affect the effect of data mining.In view of the above two key problems, this paper studied wavelet de-noising clustering model mining in stock returns predict application.This paper has completed the following jobs:Firstly,this paper points out the deficiency of modern financial econometrics, and discusses the importance and necessity on application of data mining technology in the financial field,then analyses the possibility of digging the implied mode from the mass financial data.Secondly, considering high noise of financial time series data, this paper discusses that using wavelet de-noising methods on data pretreatment, and puts forward the improvement of the wavelet threshold denoising method. It selectes suitable wavelet function, threshold value criterion, threshold processing function and decomposition layers of data to deal with the noise.Thirdly,this paper introduces temporal sequence pattern mining, then combines clustering analysis method with temporal sequence pattern mining to construct clustering temporal sequence pattern model.It improves the traditional model,optimizes the deficiency in the algorithm.We use random data simulation to verify the model, and the results is very good and it proves its validity.At last,we select 20 stocks in Shanghai and Shenzhen Stock Mrackets using data from Jan.31,2005 to Dec.31,2007 for empirical research. One of the first half of data as the digging object,the other half of data as model validation.The results shows that the average returns of symptom patterns are higher indicating that the constructed data mining techniques can be used for future decision-making and forecasting.