Estimates and Forecasts of GARCH Model under Misspecified Probability Distributions: A Monte Carlo Simulation Approach

Abstract

The effect of misspecification of correct sampling probability distribution of Generalized Autoregressive Conditionally Heteroscedastic (GARCH) processes is considered. The three assumed distributions are the normal, Student t, and generalized error distributions. The GARCH process is sampled using one of the distributions and the model is estimated based on the three distributions in each sample. Parameter estimates and forecast performance are used to judge the estimated model for performance. The AR-GARCH-GED performed better on the three assumed distributions; even, when Student t distribution is assumed, AR-GARCH-Student t does not perform as the best model.