| Summary: | 碩士 === 銘傳大學 === 財務金融學系碩士在職專班 === 100 === The risks and rewards of the two major issues are often the primary attention points in the investment market, the investors hope to predict future values in order to make better investment decisions, this paper is by constructing a prediction interval can be more appropriate predictive capability for the future value we want to observations. In the past, use of the bootstrap method to construct the prediction model are assuming that the specific order model, but the choice of the specific order without going through an appropriate model of the process may generate an error inference, therefore, according to Thombs and Schucany (1990) proposed to construct a single model AR (p) prediction interval algorithm, given the candidate model and the weight (the Akaike, 1978,1979), using non-parametric bootstrap method proposed the concept of model averaging to construct this paper discusses the prediction interval model for future reward. Specific order of the AR model and model averaging in the bootstrap method to construct the prediction interval, whether the model averaging can be obtained relatively stable results; and analysis average coverage probability and the average interval length of the single model of the AR (p) and model averaging under different sample size and prediction phases in the numerical simulation and seven different file stocks of the Taiwan stock market. The results showed that: when different number of samples and prediction phases, the model averaging method to construct the prediction interval does not have a significant impact, and the results are still similar to the real model, showing the stability of the model averaging, and model averaging compared to the specific order of the AR (p) model method with a better simulation results.
|
|---|
