Forecasting Of Stock Price Index Based On Wavelet Analysis And ARAM-SVM Model

Nowadays stock market is playing a more and more important role in the national economy. As Stock Prices Index (SPI) can reflect the complex fluctuation of the prices of different stocks, many researchers are focusing on the forecasting of SPI. From macroscopic aspect, the forecasting of SPI may provide valuable information for the government to establish stock market policy, while from the microscopic aspect, it is useful for the market participators to construct investment strategy. Of all the methods that has been applied in SPI forecasting, time series analysis is proved to be an efficient one. There are mainly two methods of time series analysis:traditional time series model method--typically ARMA model, and data mining (DM) method--typically SVM model. And the development of wavelet theory makes it possible to construct a time series forecasting model combining the advantages of traditional time series model method and data mining method.With time series of SPI as the object of this study, we combine wavelet analysis, ARMA model and SVM theory in the forecasting of SPI time series. The main work of this study is as follows:1. Comparison of different mathematic tools already applied in the research of Stock Price Index time series.2. Introduction of the theory foundation of the Wavelet-basis ARMA-SVM model, including wavelet theory, auto regressive moving Average (ARMA) model, and support vector machine (SVM) theory.3. Based on the theories, we conducted Wavelet-basis ARMA-SVM integrated model to increase the forecasting accuracy. After wavelet denoising and decomposition, SVM model is applied in the forecasting of scale variation series and ARMA is used to analyze and forecast the wavelet transformed series. Series gained from the two methods are finally combined to form the forecasted result.4. This model is proved to be of better accuracy against wavelet basis SVM-only model or SVM-only models.