| Summary: | 碩士 === 國立高雄應用科技大學 === 金融資訊研究所 === 99 === The numerous kinds of financial assets and the convenience of information circulation increase not only the liquidity of assets, but also the risk. The question of how to measure the risk using a common standard has become an important one. With the advantages of being simple and easily understood, Value-at-Risk can quantify potential losses into numbers; however, it might not evaluate risk correctly if the return of assets is not of a normal distribution. In this study, Conditional Value at Risk (CVaR) is used as a portfolio optimization tool, which can perform risk assessment even in the case of an unknown distribution of return; at the same time, using the optimal CVaR, the optimal VaR and investment portfolio weights can be obtained. In addition, a genetic algorithm (GA) is applied in this study to fit the optimal CVaR. This algorithm is designed on the basis of Darwin’s “evolution theory”, and can fit the best result under a given environment. However, no matter the risk model, the accuracy of risk assessment is very important. BIS specifies that financial institutions must review the accuracy of risk assessment periodically. In this study, three evaluation criteria – conservancy, accuracy and efficiency – are used as performance indicators to measure risk estimates. From the empirical results, it can be seen that GA performed VaR and CVaR optimization more conservative than traditional optimization mothod. Because CVaR is more conservative than VaR, CVaR-GA is the most conservative risk model, but its failure rate is lowest and loss coverage rate is highest. Mean-VaR-NLP is the most in line with the requirements of theory but less conservative and higher failure rate. By applying the GA to CVaR, it would be more conservative than those produced by the traditional optimization method.
|
|---|
