RDP 8604: Leading Indexes – Do They? 1. Introduction
May 1986
Two indexes of leading indicators have recently been developed by the Melbourne Institute of Applied Economic and Social Research and the National Institute of Economic and Industry Research. Each is primarily intended as an aid to forecasting future movements in economic activity. Both have received much attention in the Australian press. This is partly due, no doubt, to the fact that both indexes appear to be signalling a downturn in the economy. Despite the attention paid to these indexes, relatively little appears to be known about their reliability or usefulness in forecasting the future path of economic activity.
A study conducted by the Office of the Economic Planning Advisory Council (EPAC) (1985) provides a brief evaluation of the index published by the Melbourne Institute, and of the usefulness of leading indicators in general. The major criticism of the Melbourne Institute's index made by EPAC was that the lead time between turning points in the leading index and turning points in the coincident index (an index that is designed to track the business, or reference, cycle) was highly variable. More over, the leading index failed on a number of occasions to pick turning points in the coincident index and occasionally signalled turning points that did not eventuate. These criticisms have also been made with respect to leading indexes in the US (see Ratti (1985) and Vaccara and Zarnowitz (1977)).
The EPAC paper also argued that there were three basic areas in which leading indexes were generally defficient. First, although the leading index may be of some use in forecasting the reference cycly, such a forecast may be of limited use to policy makers who are more concerned with the individual series which make up the reference cycle (e.g., employment and production).[1] Secondly, the leading index provides little or no information on the reasons behind movements. These are of considerable importance for policy makers. Thirdly, the reference cycle, which the leading index is devised to predict, does not contain a number of variables with which policy makers are concerned.
Much of the evaluation of the reliability of leading indexes in the literature has been based on a rather ad hoc decision rule whereby two or three consecutive falls (rises) in the leading index signal a downturn (upturn) in the reference cycle. This approach takes little account of the size of the falls in the index and one could argue that the variable lead times observed may be a result of the ad hoc decision rule rather than the properties of the leading indexes themselves. Neftci (1982) has attempted to formalise the decision rule into a probability framework by assuming that the leading index switches probability distributions prior to a turning point in the reference cycle (relying on the assumption of asymmetric business cycles). By setting a subjective probability threshold, a turning point is signalled when the probability that the leading index has switched distributions reaches this threshold. This technique has, however, received little attention in the literature.[2]
In general, these approaches to evaluating leading indexes assume that the model underlying the economy undergoes a discrete change when turning points occur, and that leading indexes are primarily of use in picking turning points. A more desirable method of evaluating the forecasting usefulness of leading indexes may be to look at the relationship between them and variables representing the business cycle at all points. Such an evaluation would show whether the indexes were valuable for forecasting per se rather than for forecasting turning points only. Of course, if standard forecasting techniques perform badly at turning points, and if leading indexes are useful at picking turning points, then such an analysis should show that leading indexes add to the forecasting power of more traditional techniques. A number of studies along these lines have been conducted for the US leading indicators (see Auerbach (1983), Vaccara and Zarnowitz (1977) and Weller (1979)). The general conclusion is that leading indexes may be useful for forecasting economic activity when these indexes are incorporated into distributed lag regressions.
The recently popularised vector autoregression (VAR) methodology (which may be thought of as a multivariate generalisation of the single equation distributed lag methodology) appears well suited to the evaluation of the forecasting ability of leading indexes.[3] In particular, the innovation accounting techniques generally applied to VARs (in order to describe them succinctly) provide detailed information on the patterns and degrees of influence among variables in the VAR.
The aim of this paper is to use the VAR methodology to evaluate the forecasting usefulness of the two leading indexes published by the Melbourne and National Institutes. In particular, the ability of the indexes to forecast individual activity variables (e.g., employment and production) is examined to evaluate the EPAC criticism that these indexes are of limited use in forecasting individual activity variables. A number of small unrestricted VARs consisting of a leading index and a variable representing the business cycle, are estimated. Tests of “Granger-causality” (to determine the intertemporal timing relationships between variables) and the innovation accounting techniques are employed to examine the “usefulness” of leading indexes in forecasting future activity and to determine the lead times between movements in the leading index and movements in activity variables.
The paper begins with a brief discussion of the VAR methodology and, in particular, the innovation accounting techniques employed. The usefulness of each leading index in forecasting its own (or related) coincident index is evaluated in Section 3. The following section evaluates the ability of each index to help forecast individual activity series and, where appropriate, compares the two indexes. Finally, in Section 5, a summary of the main results and some concluding remarks are presented.
Footnotes
The leading index may provide, however, a consistency check on forecasts of such series, since a set of forecasts for individual series implies a forecast for the reference cycle which can be compared to that provided by the leading index. [1]
Palashi and Radecki (1985) is the only application of which we are aware. Attempts by a colleague to apply it to Australian data have been unsuccessful. [2]
There has been a certain amount of controversy over the usefulness of VARs. In particular, the recent contribution of Cooley and LeRoy (1985) argues that they are of limited usefulness for policy analysis. However, all participants of this debate seem to agree that VARs are useful in the realm of forecasting. [3]