Model Selection and Combining for forecasting
主 题: Model Selection and Combining for forecasting
报告人: Prof. Yuhong Yang (University of Minnesota)
时 间: 2006-06-20 下午 2:00 - 4:00
地 点: 理科一号楼 1114
Statistical models are commonly used for time series forecasting. Traditional analysis involves transformation, detrending, order-selection, etc. to reach a final model. Model selection criteria have become an essential ingredient there. This general modeling process, however, can be highly unstable, which causes the final forecast more uncertain than necessary. Alternatively one can consider combining the different models. With an information-theoretic approach, the combined forecast converges optimally. Numerical examples show
its significant advantage over traditional model selection methods.