Used Fractal’s Self-Similarity for N-steps-ahead Forecasting and Discovered the Trading Signal

• Main Authors: Chi-Liang Yang, 楊啟良
• Other Authors: Chao-Fu Hong
• Format: Others
• Language: zh-TW
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/04097148456271573868

碩士 === 真理大學 === 管理科學研究所 === 93 === In the past decade many researchers had used the artificial intelligence (AI) for forecasting the future. These AI methods had two researching directions: one was like the black box’s pattern recognition according to the relationship between input and output value to discovery its predictive function; the other was according to the chaotic theory to discover its fractal and construct the predictive function. From the experimental results of Kumar and Tan [7] we found that the chaotic base had better predictive performance. But, when we used the chaotic base method to run the n-step ahead forecasting, it easily cumulated the residuary error and resulted in hardly controlling the predictive error by the chaotic base method. In this paper we decompose the fractal to some sub-fractals, and let every sub-fractal only run its one-step ahead predicting to avoid the error accumulation. At last, the IDF model was integrating all sub-fractal to build the n-step ahead predicting time series, and the experimental results also imply that the predictive error is as good as the one-step ahead predicting error.