| Summary: | 碩士 === 國立臺灣大學 === 商學研究所 === 88 === In line with the growing sophistication of risk management models, supervisors permit financial institutions to use their own VaR models in calculating the required minimum regulatory-capital to be held against those market risks. While the relative performance of different VaR models may change through time, it is expensive for banks to change their VaR model. For regulatory purposes, banks are not permitted to switch between different models frequently. For these reasons, banks need a model that is going to perform well over a prolonged of time.
The purpose of the research is comparing the three dimensions of model performance (conservatism, accuracy, and efficiency) among four VaR models by the statistical methods. Then trying to set a measuring standard for risk management.
The results of the research are as followings:
1.Among four models in my research, GARCH model gets the worst performance in all dimensions.
2.There may exist some interactive performance between historical simulation and Monte Carlo simulation. Historical simulation is better than Monte Carlo simulation in all dimensions.. When the confidence level is decreased, these two models get better performance.
3.Under 99% confidence level, extreme value estimation shows the best performance.
4.Changing confidence level affects the models’ performance. Using portfolio data of this thesis, under 99% confidence level we should choose the extreme value estimation and under 95% or 90% confidence level we should choose the historical simulation.
5.Comparing the required regulatory-capital calculated by VaR models and standard regulatory model, we conclude that BIS model tend to get more conservatism.
|
|---|
