| Summary: | 碩士 === 東吳大學 === 企業管理學系 === 92 === As global financial markets become more and more liberalized, the interrelationships between international markets consistently increase. While the traditional concept of value-at-risk is incapable of completely controlling the losses caused by critical financial events, stress testing can make up for what value-at-risk is unable to provide. Through the setting of stress test scenarios, the maximal probable losses when rare events occur can be estimated. In this article, several stress-testing methodologies that have been adopted by scholars in the past are considered, including the normal distribution model as well as a mixture normal distribution model and an extreme value distribution model. These models are used to estimate the stress test losses of stock assets in Taiwan, Hong Kong and China. Models comprising both a single risky asset and a portfolio are considered, and the accuracy test and the loss coverage ratio are used to evaluate which model is better. The results show that the extreme value distribution method provides the most accurate stress-VaR estimator, and the mixture normal distribution method provides the highest loss coverage ratio in the case of a single risky asset. Moreover, the Kupiec model and the Kim & Finger model outperform the other models in terms of the accuracy measure, and the extreme value distribution method is found to be superior to the other methods based on the coverage ratio in the case of a portfolio.
|
|---|
