| Summary: | 碩士 === 東海大學 === 工業工程學系 === 90 === The study refers to both domestic and overseas literatures of Value at Risk(VaR) evaluation. In order to find suitable VaR model for each Industry and then reach real risk- avoiding effect, various models are used to compute VaR. The purposes of research use Equally Weighted Moving Average Approaches (SMA), Exponentially Weighted Moving Average Approach(EWMA), Historical Simulation (HS), Classical Bootstrap (CB) and Stationary Bootstrap (SB) to forecast VaR of Taiwan stock market and access the model performance among VaR models by the two dimension of Conservation and Accuracy including eight examination method. The conclusions of the study are as the following:
1. Local Valuation method: Considered variance is the viewpoint of the method. Relatively low VaR and high variance are obtained. Conservative valuation makes the result that conform to expect. Users need to consider its applicability for real cases.
2. Fully Valuation method: The method without considered matrix. Relatively accurate VaR and low variance are obtained. Accurate valuation comes out more accurate risk-taking degree. But users still need to consider its applicability for real cases.
3. The results are different under various criteria and through different examination method. When Under different α value, same industry has different result. The study is limit to a almost circulate and can describe the industry. Starting from today, users can choose the appropriate sample to portfolio by this study.
Between these five models, SB final valuation perfrom stable on Conservation index and perfrom the best on Accuracy index, and therefore is the number one. HS perform worse on the Conservation index but only behind SB on Accuracy index, and so is the number two. SMA has average performance both on Conservation index and Accuracy index and follows. Before EWMA is CB, which perform relative bad on Accuracy index and the worst on puncture degree. EWMA has most conservative VaR valuation. But the overly right bias makes it the worst model.
|
|---|
