Time varying coefficient models and their forecasting performance

The last decade has seen the spasmodic development of causal models with coefficients that vary over time and this paper aims to examine the effectiveness of the approach, with particular reference to ex-post forecasting performance. It discusses the empirical and theoretical reasons for models of this type, briefly identifies estimation methods, surveys published work and presents new empirical studies on whisky, tobacco, football and inflation. The paper proves conclusively that the approach significantly improves forecasting performance and concludes that it should be automatically considered by any management scientist undertaking the modelling of causal relationships over time.