Long Memory Analysis And Forecasting Of Financial Time Series

Financial system is the core of modern economy. The complexity and volatility of the financial system widely exist in countries and each phase of society. China joined in World Trade Organization(WTO). From then on, China is increasingly opening its economy to foreign countries. Chinese financial market joined in WTO 5 years later. It brings us both the development opportunity and challenges that never faced before. While it is influencing the world, Chinese finance has to confront the volatility from the international financial market. How to take the advantageous position during the proceeding of finance internationalization and take the positive measures to reply to the volatility of the capital market have become the urgent affairs in our development of financial market.To show the characters of financial time series accurately, we must establish the proper models which accord with its characters. The memory character of time series is one of the key factors. Studying the memory of the prices of financial assets not only provides dependable foundation for authorities making policies and macro-economy regulations but also gives practical suggestions to the institutions and private investors. Only investing according to the characters of the financial assets considering both the long and short relativities can catch the essence of the market and adjust the portfolio in time to avoid the risk.Based on the above, the main works of the thesis are as follows.1,Use R/S, modified R/S and V/S analysis to research the long memory of financial time series. Study the impact factors of the long memory of financial time series from time and event point of view. The result indicates that varying time segments and special events can make the conclusions totally different. We also study the short sensitivity of V/S analysis.2,Based on the traditional GM model and ARMA model, use IGM model to estimate the error between the real value and the estimate value from the FIGARCH model. Propose the IGM-ARFIMA model to estimate the expectation of the long-term financial time series. Financial time series are forecasted with these models. The results indicate that modified model outperforms the original model.3,Based on the GM-GARCH model, according to the volatility forecast of long-term financial time series establish IGM-FIGARCH model using IGM model to correct error in FIGARCH model. That is to forecast the random error in FIGARCH model using IGM model and add the forecast value to the FIGARCH model to correct the influence of the uncertainty. The demonstrations indicate that IGM-FIGARCH model outperforms the GM-GARCH model.4,A modified Elman network is proposed. It has 2 feedback cells. Combine the modified Elman network and phase space reconstruction technique to establish a feedback chaos neural network. Stock price time series are forecasted with these methods. The experiments on the prediction of the specific financial series are carried out. The results indicate that multivariate chaos neural network outperforms the univariate one.