All Change! The Implications of Non-Stationarity for Empirical Modelling, Forecasting and Policy

Economies, societies, and many natural systems evolve and change, sometimes dramatically, so good models and accurate forecasts are vital for policymakers to prepare for and navigate these changes successfully. Yet history is littered with forecasts that went badly wrong, sharply illustrated during the recent recession. A glance at most economic and related time series, such as greenhouse gases, reveals the invalidity of an assumption of stationarity, whereby the mean and variance are constant over time. Nevertheless, many models used in empirical research, forecasting or for guiding policy have been predicated on treating observed data as stationary, when in fact such analysis must take non-stationarity into account if it is to deliver useful outcomes. The problem for policymakers is not a plethora of excellent models from which to choose, but to find stable relationships that survive long enough to be useful. This paper offers guidance for policymakers and researchers on identifying what forms of non-stationarity are prevalent, what hazards each form implies for empirical modelling and forecasting, and for any resulting policy decisions, and what tools are available to overcome such hazards