G7 Countries Unemployment Rate Predictions Using Seasonal Arima Garch Coupled Models

Abstract

Despite the unemployment data have been recently released as seasonally adjusted, seasonality may still exist in moving average (MA) or auto-regressive (AR) terms. This can be detected by searching for a regular pattern in auto-correlation function (ACF) and partial ACF (PACF) diagrams. Therefore, models that aim to forecast unemployment rates should consider their seasonal properties so as to obtain better mean equation estimations. Univariate models mostly employ integrated ARMA (ARIMA) or generalized auto regressive conditional heteroscedastic (GARCH) models or any combination of them. Once the mean equations are structured better, GARCH estimations of variance equation is expected to perform better accuracy in forecasts. This study first examines the ACF's and PACF's of seasonally adjusted unemployment rate data in G-7 countries for 1995-2019 period. Then it compares the 4-quarter and 8-quarter ahead forecast performance of the seasonal ARIMA (SARIMA) coupled volatility models of GARCH in mean, absolute value GARCH, GJR-GARCH, exponential GARCH and asymmetric GARCH models. The performance of these models is also compared to SARIMA and MA filtered volatility models. The results show that seasonality should be re-examined even in seasonally adjusted unemployment data, since SARIMA models outperform ARIMA models in terms of out of sample forecast errors. Besides SARIMA-GARCH models provide better out of sample prediction accuracy.