Font Size: a A A

The Application Of Nonparametric Multi-group Genetic Neural Network In Stock Index Prediction

Posted on:2013-10-26Degree:MasterType:Thesis
Country:ChinaCandidate:J S WangFull Text:PDF
Abstract/Summary:PDF Full Text Request
The modern stock market is a huge investment, financing market, every day there has a large capital inflow and outflow and it plays more and more important role in the social stability and development. If who can master the trend of the development of the stock market and make investment decision in a short time, then the one can get huge profit from this great investment and financing market.If we want to grasp the trend of the stock market, and make the correct decision in time, the forecast and analysis of stock market seems necessary. The mainstream view of the existing stock market forecast and analysis method contains fundamental analysis and technical analysis, but these analysis methods have their inherent defects and unbelievable. With the development of the stock market, the stock market analysis based on financial data time series forecast method have begun to emerge, but most of the method is based on linear under assumptions and against the stock market nonlinear characteristics.But the neural network has its own special advantage in solving this kind of nonlinear problems, therefore, this article combine the neural networks with multiple population genetic algorithm, and build a multiple population genetic algorithm neural network then applied it to the prediction of stock price index. The combined model not only overcome the traditional neural network’s defect such as easily trapped in local minimum、difficult in determining initial weights, and also avoid the standard genetic algorithm which exists in the premature convergence problem. At the same time, the model also has the advantage of training fast, having stable results and accurate results, provides a new way for the stock index prediction.This paper first elaborated the neural network method and the learning algorithm of BP neural network model, then analyzed the shortcomings of the model. We introduced the concept of genetic algorithm and multiple population in order to overcome these disadvantages.After expounding the concept of genetic algorithm and multiple population, this paper combined population genetic algorithm with neural network method,and designed the multiple population genetic neural network model. At the same time, this paper also considerate the model’s input layer index variables and influence stock index variables respectively from the macro fundamentals and technical analysis, and use nonparametric kernel estimation methods for selecting indicator variable. The method improve the traditional method such as principal component analysis and the least squares estimation method in variable selection method.Finally, this paper used the eight indicator variable as input layer index for population genetic neural network model and the traditional BP neural network model respectively, and did the empirical research for SSE composite index. The empirical results showed that:the multiple population genetic neural network model is better than the traditional BP neural network model in prediction accuracy, this also showed that the multiple population genetic neural network model used in this paper is scientific and reasonable.
Keywords/Search Tags:stock index forecasting, neural network, genetic algorithm, multiple population genetic neural network, nonparametric kernelestimation
PDF Full Text Request
Related items