| Summary: | 碩士 === 東吳大學 === 財務工程與精算數學系 === 101 === VaR (Value-at-Risk) from 1993 put forward by the Group of 30 tools for measuring market risk, has been used by various financial institutions. Due to the financial return on assets volatility clustering and heavy-tailed distribution, we consider asymmetric volatility GARCH, IGARCH and EGARCH model, with the error distribution is normally distributed, and the fat tail distribution of standard T distribution and normal-inverse Gaussian (referred to NIG) distribution. We calculate the value at risk of the financial return on assets and according backtesting assessment of the suitability of the VaR models. The comparison criteria are Kupiec (1995) proposed unconditional coverage test, Christoffersen (1998) proposed conditional coverage test and Christoffersen and Pelletier (2004) proposed rate of return on financial assets is not greater than the duration of the test method of the value at risk. The data during January 4, 2010 to December 31, 2012 of the Asian developing countries, Japan NIKKEI 225 Index, Hong Kong HANG SENG Index, Taiwan Weighted Stock Index and the Korea Composite Stock Price Index daily rate of return for empirical research. We compare various VaR models and historical simulation the performance of the method. We found that the four stock markets under the same GARCH model, it is assumed that the error of the normal distribution VaR models will underestimate the risk, and assuming error NIG distribution VaR models is less underestimates the risk, under the assumption that the same error distribution, EGARCH model is easy to overestimate the risk. Overall, the IGARCH-T VaR model except the Taiwan stock market VaR estimates easier to overestimate the risk, in the other three stock market is better performance. Error distribution for the NIG distribution of the VaR models is mostly performed well in predicting the Hong Kong stock market risk value, it less likely to underestimate the risk.
|
|---|
