RDP 9102: Indicators of Economic Activity: A Review 2. Data and Methods

The variables included in the study are indicators of real expenditure and activity which are judged to be in frequent use in published analyses of the economic cycle. The variables are listed below in Table 1.

Table 1: Summary of Indicators Included in the Study
1. Expenditure Aggregates 2. Partial Indicators
Dwelling investment Building approvals
Housing finance
Housing sales
Plant and equipment investment Capital expenditure survey
CAI-Westpac survey
Non-dwelling construction Construction approvals
Capital expenditure survey
CAI-Westpac survey
Consumption Retail trade
Car registrations
Consumer sentiment index
Change in non-farm stocks  
Exports  
Imports  
GDP Employment Job vacancies

Note: For further details on sources and definitions, see Appendix.

As noted in the introduction, the analysis proceeds in two stages, the first stage looking at relationships between variables from group one in the above table, and the second stage studying the usefulness of the partial indicators in forecasting individual expenditure aggregates. Conceptually, the most appropriate method for dealing with these issues is to use vector autoregressions (VARs) or forecasting equations. That is, we estimate equations of the form

and conclude that xt leads yt if the b coefficients are jointly significantly different from zero. It may be noted that these methods are subject to a certain amount of controversy, particularly when the aim is to make inferences about causality. For example, it is well known that such systems of equations are misspecified unless all relevant variables in a causal system are included. Also, results can be very sensitive to design features such as the choice of lag lengths and the length of the sample period.[2]

These problems are less important when the aim is only to draw conclusions about forecasting, since the equations are interpreted only in the more limited sense of showing whether particular variables add information to a given forecasting system. It is nonetheless our experience that the results in this paper are quite sensitive to the design features mentioned above, and we have therefore chosen to supplement the VAR results with simpler techniques based on bivariate VARs, correlation coefficients and visual inspection of the data. This follows a similar approach to that of Bullock, Morris and Stevens (1989) and Stevens and Thorp (1989) in their analyses of financial indicators.[3]

The VARs are estimated with lag lengths of up to four quarters. Generally speaking, the data used are in quarterly log-differenced form, although in some cases it is possible to estimate relationships using monthly data (for example, in estimating the relationship between employment and job vacancies). Further details on data sources are provided in the Appendix.

Footnotes

For example, Thornton and Batten (1985, p166) state that “individuals could arrive at different, but equally legitimate, conclusions concerning the Granger-causal relationship between time series due solely to differences in their lag-length selection criteria”. [2]

Early studies by Beck, Bush and Hayes (1973) and by Bush and Cohen (1968) looked at an exhaustive list of indicators using various statistical techniques. [3]