ABSTRACT

The primary purpose of this research is to construct a 99% one-day VaR of the CSX index based on the volatility model that does not use the traditional statistical technique to measure the return volatility. The volatility of the index return is predicted by using the conditional variance equation of AR(1) GARCH(1,1) model to estimate VaR. To obtain the sample parameters of this model, the Quasi-Maximum Likelihood Estimation (QMLE) method is applied under three different assumptions of the disturbance term of the mean equation, namely the Normal, Student-t and GED distributions. The total sample size for this study is 1000 observations, but the estimated model with respect to each distribution assumption is performed with three sub-samples, which are 250, 500 and 750 daily returns. The best estimate of the 99% one[1]day VaR is derived from the model under the Student-t distribution with 750 observations since the exception, failure, or penalty rate of the model is only 0.93% as referring to the Backtesting, which is lower than the 5% by the Basel Committee on Bank Supervision.

Keywords: Value at Risk, GARCH Model, Backtesting.