| Summary: | This dissertation aims to find a better-performed model in estimating risk measures for certain countries. The risk measures are estimated under five distributional assumptions (normal, Student-t, skewed Student-t, historical distribution, and generalized pareto) for five financial markets (Nasdaq, FTSE100, SSEC, BVSP, and Nifty50), three estimation windows (250, 500, 1000), and two significance levels (0.05 and 0.01). A two-stage ES backtest and a direct ES test are conducted. The violation ratio comparisons and the backtesting results indicate that student’s t distribution and normal distribution have overall the best and the worst performance. Unconditional models also have its deficiencies in modeling financial markets.
|
|---|