Font Size: a A A

Study Of The Stock-Forecast Using ARMA And Grnn Technology

Posted on:2005-11-02Degree:MasterType:Thesis
Country:ChinaCandidate:W M LiFull Text:PDF
GTID:2168360125466764Subject:Software engineering
Abstract/Summary:PDF Full Text Request
Stock market is a nonlinear system. Mostly the relative stock price forecasting is not good in using traditional time series model in this dissertation and many other relative researches, because it is difficulty in mapping nonlinear system and defining the suitable model by using it. As a modern means in dealing with intelligent information, neural network can adaptively learn by datas and map nonlinear system strongly, so it suits to deal with complicated nonlinear system such as stock price.This dissertation constructs share index series model and analyzes it using the traditional model firstly. From the result of the share index forecast, we know the trend of the share index in the future largely. Then, the author selects General Regression Neural Network(GRNN) in neural network technology. Compared with BP network, the GRNN technology has many merits, such as computing quickly and being stead in result,being few parameters selected by people. All the tests of this article are on the GARCH and ANN tools in MATLAB6.5 .We construct the ARMA and GRNN model of the share index, and forecast the share index in the future. The result shows that the GRNN is better than traditional time series model in mapping the nonlinear system. Finally, the author puts forward a new way ARMA-General Regression Neural Network that combines ARMA model with GRNN in order to improve the precise of stock-forecast. The results indicate that the scheme composed by ARMA model and GRNN is better than ARMA model and GRNN that takes the share index a s t he o nly i nputting p arameters. T his i ndicates t hat ARMA m odel b rings s tatistical information into GRNN. The precisely result in short time reflects the ability of the combined model ARMA-General in mapping nonlinear system.
Keywords/Search Tags:ARMA, neural network, GRNN, time series, innovation series, forecasting
PDF Full Text Request
Related items