The recent poor performance of economic forecasting in the UK has emphasized its inability to predict the turning-points in the economic cycle and the need to exploit all available information. In Discussion Paper No. 833, Research Fellow Michael Artis, Robin Bladen-Hovell, Denise Osborn, Graham Smith and Wenda Zhang consider the possible contribution of the `longer' and `shorter' leading indicators published by the Central Statistical Office (CSO). The `shorter' index combines the change in volume of consumer credit, the real value of corporate (non-oil) trading profits, new car registrations and responses to CBI surveys on changes in new orders and expected changes in stocks and materials. The `longer' index combines the real value of companies' financial surplus, the interest rate on three-month prime bills, dwelling starts, the FT Actuaries Index and responses to a CBI survey about `optimism'. The CSO's `coincident index' measures activity as a function of GDP, output of production, volume of retail sales, and responses to CBI surveys on capacity utilization and raw material stocks.

Artis et al establish that the longer index leads the coincident series by ten months at the peak and thirteen at the trough on average, and the shorter index by five months at the peak and nine at the trough; but there is substantial variation. The simple direction of their monthly movement is a poor predictor of the economic cycle, and evidence of persistence is needed to draw reliable inferences. They use the CSO data now available to consider both a simple `three consecutive declines/upturns' rule and a more sophisticated `sequential probability' approach, in which the probability of an upturn (downturn) increases as confirmatory observations accumulate. The shorter index has very little information content, but the longer index performs much better; the sequential probability approach yields good predictions at leads of four-to-six months for peaks and six months for troughs. While the use of `real time', unrevised data would no doubt impair its performance, a determined search might also produce a better set of component series; their combination into an index may also be wasting potentially useful information, which is left as an exercise for further research.

Turning Point Prediction for the UK Using CSO Leading Indicators
Michael J Artis, Robin C Bladen-Hovell, Denise R Osborn, Graham W Smith and Wenda Zhang

Discussion Paper No. 833, September 1993 (IM)