Applying the Fractal Based Parallel Mechanism Artificial ARCH Model in Stock Price Multi-Steps Ahead Forecasting

• Main Authors: San-Long Yu, 余三郎
• Other Authors: Chao-Fu Hong
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/55372053953055507994

碩士 === 真理大學 === 管理科學研究所 === 94 === Title of Thesis： Applying the Fractal Based Parallel Mechanism Artificial Intelligence ARCH Model in Stock Price Multi-Steps Ahead Forecasting Key words：Fractal Theory, ARCH, Chaos Theory, Genetic Programming Name of Institute：Graduate School of Management Science, Aletheia University Graduate Date： September, 2006 Degree Conferred：Master Name of Student：San-Lang Yu Advisor： Chao-Fu Hong, Ph.D. 余三郎 洪朝富 Abstract： Many research results supported that the chaotic forecasting models have high predicting precision in one-step ahead predicting. In this study, we tried to use the dimension of the chaos to explain why these models can have high precision forecasting. From Our experimental results, if the chaotic dimension of stock price time series is smaller than 2-5, the one-step ahead stock price will be predictable. Furthermore, in this study we also try to implement the parallel computing into the chaos base one-step ahead predicting model and expand it to n-step ahead prediction. Here, the delay time (τ) is used to separate the stock price time series to parallel τ’s sub-fractal time series, and then the GP is used to look for the hidden rule in every sub-fractal time. Finally, this algorithm is not only can avoid the error propagation, but also can do the n-step ahead predicting. In addition, we assume the ARCH effect can influence the predicting precision. Therefore, we define an AIARCH model (Artificial Intelligence Autoregressive Conditional Heteroskedastic), and expect it can further decrease the n-step ahead predicting error. The experimental results also evidence that AIARCH model can decrease the error propagation, and at the same time the results of pair- t test verify that it can improve the predicting precision in n-step ahead. At last we analyze the result of the error distribution (MAPE and error distribution figure). The error distribution also quite similar to the one-step predict model. These results can evidence that the methods of the artificial intelligence prediction also need to add the ARCH effect into their method in order to improve predict accuracy(Heteroskedastic).