Abstract/Summary

When the assumption of constant parameters fails, the in-sample fit of a model may be a poor guide to ex-ante forecast performance. We expound a number of models, methods, and procedures that illustrate the impacts of structural breaks on forecast accuracy, and evaluate ways of improving forecast performance. We argue that a theory of economic forecasting which allows for model mis-specification and structural breaks is feasible, and may provide a useful basis for interpreting and circumventing systematic forecast failure in macroeconomics. The empirical time series of consumers' expenditure, and Monte Carlo simulations, illustrate the analysis.