Forecasting with leading economic indicators
Forecasting with leading economic indicators entails drawing upon a large number of time series variables that, by various means, have been ascertained to lead the variable of interest, typically taken to be aggregate output (the business cycle). The first set of leading economic indicators was developed as part of the business cycle research program at the National Bureau of Economic Research, and was published by Mitchell and Burns (1938). More recent works using this general approach include Stock and Watson (1989) and the papers in Moore and Lahiri (1991).
The use of many variables and little theory has the exciting potential to exploit relations not captured in small multivariate time series models. It is, however, particularly susceptible to overfitting within sample. For example, Diebold and Rudebusch (1991) found that although historical values of the Index of Leading Economic Indicators (then maintained by the US Department of Commerce) fits the growth in economic activity well, the real time, unrevised index has limited predictive content for economic activity. This seeming contradiction arises primarily from periodic redefinitions of the index. Their sobering finding underscores the importance of properly understanding the statistical properties of each stage of a model selection exercise. The development of methods for exploiting large sets of leading indicators without overfitting is an exciting area of ongoing research.