Application Of Dynamic Neural Network In Quantitative Investment Forecast

With the development of finance theory and computer technology, quantitative investment strategies with based on data and rules has been gradually on the rise in China. Quantitative model has become a powerful tool for prediction markets and guide investment. However, the stock market is a complex nonlinear dynamical system; there are a lot of limitations with traditional time-series forecasting techniques, neural network predictive theory developed in the last decade in a non-systematic forecasting and modeling show a unique advantage.This thesis involves time series analysis based NARX dynamic neural network model and its application of Chinese stock prices time series forecasts, and compare it with traditional time series models. Through empirical studies on the data of China’s Shanghai and Shenzhen stock index prices, the results show that the accuracy of forecast from dynamic neural network model is superior to which from ARIMA-GARCH or from static BP neural network model.Based on the involved NARX dynamic neural network, the model is creatively applied to pattern classification to judge if the index price is at the period’s "top" or "bottom" and makes it as quantitative timing selection model. The empirical studies results show that the accuracy of dynamic neural network model based pattern classification is significantly superior to which from discriminant analysis or from regular artificial network model based pattern recognition algorithm.In addition, an analogous active portfolio is constructed based on the quantitative timing selection model. Through calculation varies of widely-used performance evaluation measurement the results show that the performance of analogous active portfolio based on the module is superior to which of market index. So it can be successfully used to quantitative investment in China’s securities market.Finally the thesis researches the co-relationship between behavioral finance based market anomalies and efficiency of dynamic neural network and conduct statistical testing, and finds out the source and financial explanation of the forecast efficiency by dynamic neural network model.