Turning Point Prediction for the UK using CSO Leading Indicators

This paper examines the performance of alternative models for predicting turning points in the UK growth cycle. The models are based upon an interpretation of movements in the CSO's composite longer and shorter leading indicators. The difference between the models lies in the choice of method adopted for separating and classifying observations into a pattern corresponding to an upturn and downturn regime, together with the decision rule applied in recognizing when a regime shift has occurred. The models involved include a simple mechanical rule based upon an interpretation of consecutive movements in the leading indicator and two probabilistic methods, namely a standard Bayesian model and the sequential probability model developed by Neftci (1982). The results of the exercise suggest that usefulness of the shorter leading index is extremely limited; prediction based upon this series is typically outperformed by naive, non-indicator methods. The information content of the longer leading index appears somewhat greater. The signal extracted by the sequential probability model is particularly well-defined in this respect giving rise to a lead time of between four and six months at peaks and six months for troughs. At horizons beyond six months, however, the sequential probability model is outperformed by a more conventional Bayesian method.