A comparison of volatility prediction between ARIMA-GARCH and VAR models

Abstract

In this thesis the authors use ARIMA-GARCH and VAR to predict future volatility of 6 macroeconomic variables from the US. The data is monthly and spans the period 1964-2014, where the last 20 years are used as the out-of-sample period. The univariate GARCH models are widely used in volatility estimations in the elds of macroeconomics and nance, and the authors feel that there are better ways of predicting volatility. We nd that VAR outperform both GARCH and EGARCH when it comes to predicting future volatility out-of-sample, which is consistent with previous research. The naive model is found to outperform both GARCH and EGARCH, while there is no conclusive answer for which GARCH model is superior.