Analyses of forecasting that assume a constant, time-invariant data generating process (DGP), and so implicitly rule out structural change or regime shifts in the economy, ignore an aspect of the real world responsible for some of the more dramatic historical episodes of predictive failure. Some models may offer greater protection against unforeseen structural breaks than others, and various tricks may be employed to robustify forecasts to change. We show that in certain states of nature, vector autoregressions in the differences of the variables (in the spirit of Box-Jenkins time-series modelling), can outperform vector equilibrium-correction mechanisms. However, appropriate intercept corrections can enhance the performance of the latter, albeit that reductions in forecast bias may only be achieved at the cost of inflated forecast error variances.
