A0150
Title: Discriminating direct from induced equilibrium-mean shifts
Authors: David Hendry - University of Oxford (United Kingdom) [presenting]
Jennifer L Castle - Oxford University (United Kingdom)
Jurgen Doornik - University of Oxford (United Kingdom)
Abstract: Equilibrium-mean, or location, shifts can result directly from changes in intercepts with constant dynamics, or be induced by shifts in dynamics (or other parameters) when data means are non-zero. The impacts of in-sample induced shifts substantively modify previous taxonomies of forecast errors. Step-indicator saturation helps detect any resulting location shifts. However, even when all relevant variables in the data generation process (DGP) and all indicators matching DGP shifts are selected in the forecasting model, mis-forecasting can occur. To discriminate direct from induced shifts, we add to the model multiplicative indicators formed by interacting all selected step indicators with the lagged regressand. When equilibrium-mean or location shifts are induced by changes in dynamics, forecasts can be markedly improved when these interactive indicators are included.
