Multi-step Forecasting in Unstable Economies: Robustness Issues in the Presence of Location Shifts

To forecast at several, say h, periods into the future, a modeller faces two techniques: iterating one-step ahead forecasts (the IMS technique) or directly modelling the relation between observations separated by an h-period interval and using it for forecasting (DMS forecasting). It is known that unit-root non-stationarity and residual autocorrelation benefit DMS accuracy in finite samples. We analyze here the effect of structural breaks as observed in unstable economies, and show that the benefits of DMS stem from its better appraisal of the dynamic relationships of interest for forecasting. It thus acts in between congruent modelling and intercept correction. We apply our results to forecasting the South African GDP over the last thirty years as this economy exhibits significant unstability. We analyze the forecasting properties of 31 competing models. We find that the GDP of South Africa is best forecast, 4 quarters ahead, using direct multi-step techniques, as with our theoretical results.