| Summary: | 碩士 === 逢甲大學 === 應用數學所 === 97 === Value-at-Risk (VaR) had been a criterion for the investment of market return rates. The wavelet transform does not need assumptions for estimating return rates risk model, and make multiresolution analysis in counter of non-linear return rates. This paper utilizes the data of nonferrous metal, original oil, the exchange rate, and the fund material, using the discrete wavelet transformation to obtain the numerical risk model, using the high frequency data obtain by the wavelet analysis of time and frequency domain countering with different processing period, obtain the model parameter to estimate the return rates by discrete wavelet transformation, then compare it with the real return rates to get the penetrating times to observe the accuracy rate of the estimation. If the forecasting number of exceptions does not fall into the confidence interval in backing test method, then we change the relative energy weight in each level to be between 0.5 to 2, different data categories have different relative energy weight stability to improve penetrating times. As to the relative energy value''s selection at each level, set an algorithm and input the data and numbers, relative energy weight interval, and corresponding value of confidence level, then we get the suitable weight. Moreover, if we take the nonferrous metal for data, the estimated model of multiple-steps ahead forecasts is better than the WDVaR model of one-step ahead forecasts.
|
|---|
