The Comparison Of GARCH And ANN Methods On Stock Price Predictions

It is very interesting and challenging to predict the price pattern of financial assets. As all we know, financial time series could be time-variant, nonlinear, as well as non-stationary. Thus, it is very complex and difficult to make such predictions. Financial econometricians have developed many models, such as linear regression, time series models to accomplish this task. Meanwhile, AI models, such as ANN and genetic algorithms have been introduced to do this as well. In this thesis, we make use both of the econometrics models and AI models to predict stock prices, and compare their performances. This would be important both in the sense of theoretical and practical research.In this thesis, we first survey these two methodologies in the context of stock price prediction, such as ARMA, VAR, GARCH, and ANN, GA. Then, we apply them to Chinese stock market. Thus, we use GARCH, which is implemented in EViews, and BP-ANN, which is implemented in MATLAB, to predict the prices of a single stock and stock index. Finally, we compare the performances of these two methods. It enhances our understandings of both methods.We show that, BP-ANN outperforms GARCH on both the single stock prediction and the stock index prediction contexts. It would be interesting to further investigate these two methods.