The effect of seasonal adjustment on forecasting accuracy

This paper investigates the effect of seasonal adjustment on the forecasting power of structural time series models. The empirical work is based on 18 quarterly and monthly Australian time series (both macro and sectoral). Models are estimated for the seasonally unadjusted and seasonally adjusted time series, and out-of-sample forecasts are generated from both models. The seasonally adjusted forecasts are subsequently deseasonalised to facilitate comparison. Having done that, the AGS test is performed to find out if there is a statistically significant difference between the root mean square errors of the two models. Since the model that is estimated from the seasonally adjusted data tends to be a more powerful forecasting tool than the other model, we conclude that seasonal adjustment is not generally detrimental to the forecasting accuracy of time series models.