AN EXAMINATION OF THE PREDICTIVE POWERS OF FILTERED HISTORICAL SIMULATION

Abstract

In this study, we examine how complexity in asset price forecasting models does not necessarily translate into accuracy. We present a comparison of prediction accuracy of two asset pricing models in this thesis – an Exponential Generalized Autoregressive Conditional Heteroskedastic Filtered Historical Simulation model and a simple median returns model – where we analyse daily return forecast errors of up to 2 years for both models. We also conduct tests of prediction accuracy using regressions of forecast errors on market volatility and other controls. The results show that the simple median returns method outperforms the EGARCH FHS, even in volatile periods, an environment that usually favours GARCH models.