We propose a new method of leading index construction that combines the need for data compression with the objective of forecasting. This so-called principal covariate index is constructed to forecast growth rates of the Composite Coincident Index. The forecast performance is compared with an alternative index based on principal components and with the Composite Leading Index of the Conference Board. The results show that the new index, which takes the forecast objective explicitly into account, provides significant gains over other single-index methods, both in terms of forecast accuracy and in terms of predicting recession probabilities.

business cycles, index construction, principal covariate, principal component, time series forecasting, turning points
Time-Series Models; Dynamic Quantile Regressions (jel C32), Forecasting and Other Model Applications (jel C53), Forecasting and Simulation (jel E17)
hdl.handle.net/1765/10348
Econometric Institute Research Papers
Report / Econometric Institute, Erasmus University Rotterdam
Erasmus School of Economics

Heij, C. (2007). Improved forecasting with leading indicators: the principal covariate index (No. EI 2007-23). Report / Econometric Institute, Erasmus University Rotterdam. Retrieved from http://hdl.handle.net/1765/10348