RDP 8604: Leading Indexes – Do They? 5. Conclusion

In this paper we have attempted to evaluate the usefulness, for forecasting the business cycle and related variables, of the leading indexes published by the Melbourne and National Institutes. The methodology involved the estimation of a number of small unrestricted vector autoregression models and the use of the related innovation accounting techniques.

The results indicate that each leading index contains information that is useful for forecasting its own or related coincident index, in the sense of significantly reducing the forecast variance. Large differences in the lead times were observed, with the Melbourne Institute's leading index leading its coincident index by ten months and the National Institute's leading index leading its coincident index by just four months. Lags in the publication of these two leading indexes reduce the informational or effective lead times by three or five months. In the case of the National Institute's leading index this results in an informational lag of one month – i.e., the leading index they publish this month tells us about the business cycle last month. It was argued on the basis of further results, however, that this was due more to differences in the two coincident indexes rather than in the leading indexes themselves.

Partly because of this, and partly because of the nebulous nature of any single measure of the business cycle, we also considered whether the leading indexes were useful for forecasting individual activity variables that move procyclically. The results indicate that the leading indexes may be of some use in forecasting these individual activity series. Both leading indexes help forecast the ANZ index of industrial production, employment and (the inverse of) unemployment; although neither helps forecast motor vehicle registrations or retail sales.

However, there did not appear to be a consistent lead time between either leading index and each activity variable. This, along with the different timings of the coincident indexes, highlight the difficulties one faces in trying to obtain a single measure of the business cycle and, therefore, a single measure of future movements in this cycle.

The results of a comparison of the two leading indexes offered no firm support for the dominance of one index over the other. The Melbourne Institute's index greatly outperformed the National Institute's index with regard to forecasting the production index^[25] while the reverse was true (in a weaker form) for the employment related variables. The one constant factor, however, was that a given movement in these activity variables is associated with a much larger movement in today's National Institute leading index, than in today's Melbourne Institute leading index.^[26]

Finally, a word of caution. All of these results have been in terms of the extent to which the information in the lags of a leading index is a useful supplement to the information incorporated in the lags of variable being forecast. However, whether the leading indexes provide additional information over and above that which is normally used to provide forecasts is an open question. Further work involving ex-ante forecasting with leading indexes compared to traditional forecasts and actual outcomes may be able to shed some light on this issue.

Footnotes

The results for quarterly non-farm GDP, presented in Appendix A, are similar to those for the production index. [25]

These results probably reflect the different methodologies the two institutes use in constructing their leading and coincident indexes. Unfortunately not enough details of these methodologies are available to enable a comparlson to be made. [26]