forecastingprinciples.com Reviews of Important Papers on Forecasting,
1985-1995 Reviews
Review of:

Francis X. Diebold, and Glenn D. Rudebusch (1991), "Forecasting output with the Composite Leading Index: A real time analysis," Journal of the American Statistical Association, 86 , 603 – 610.


This is what William J. Bennett, former US Secretary for Education, had to say about the Census Bureau’s Index of Leading Economic Indicators in the Wall Street Journal on 15 March 1993: "These 11 measurements, taken together, represent the best means we now have of . . . predicting future economic trends." This appears to be the consensus viewpoint on leading economic indicators. Research on leading economic indicators began in the late 1930s. In 1950, an index of eight leading indicators was developed using data from as far back as 1870. The method has spread such that 22 countries now use leading indicators. A summary of this history is provided in the collection of papers edited by Lahiri and Moore (1991).

Interestingly, despite the massive research effort over the past half century, and despite widespread use of this technique, its ability to provide forecasts has never been adequately tested . . . until the study by Diebold and Rudebusch. Once again I am reminded of Martin Arrowsmith, the medical researcher in Sinclair Lewis’ 1924 novel, Arrowsmith. The major theme was that medicines were being used without proper testing. How is it that people adopt and use these economic prescriptions without adequate testing? Perhaps because it is intuitively obvious that such a procedure should be useful. Careful measurements of many series should contain valuable information.

Diebold and Rudebusch tested the obvious. They examined whether the addition of information from the Composite Leading Index (LCI) can improve upon extrapolations of industrial production. The extrapolations were based on regressions against prior observations of industrial production, and four models were developed. One model used simply the preceding observation, the next estimated a regression using the most recent four observations, then a model used the eight preceding observations, and the final model was estimated using the last 12 observations. They examined monthly data from 1950 through 1988 and prepared ex ante forecasts for 1, 4, 8, and 12 periods ahead. Successive updating was used (that is, they made forecasts, advanced 1 month, and added this month into the models, then created a new set of forecasts, and so on until the data were exhausted). This yielded a total of 231 forecasts for each model for each forecast horizon. They evaluated the forecasts using Mean Absolute Percentage Errors and also Mean Square Percentage Errors.

What did they find? First, they confirmed extensive prior research showing that ex post forecasts are improved by use of the LCI. However, when they made ex ante forecasts, the results were surprising and depressing. Inclusion of CLI information tended to make the forecasts less accurate, especially for short-term forecasts (1 to 4 months ahead). Diebold and Rudebusch also examined forecasts of changes instead of levels, and the results were the same.

This well-done study argues against the use of the leading economic indicators to make economic forecasts. While the index may be useful for showing where things are now, we have no evidence to support its use as a forecasting tool. Now will we have to wait another half century for another study on this important topic? Surely this information must have some value to forecasters.

Reference

Lahiri, K and G.H. Moore, 1991, Leading Economic Indicators: New Approaches and Forecasting Records. (Cambridge University Press, Cambridge).