Forecasting Australian Economic Activity Using Leading Indicators

This paper examines the contribution leading indicators can make to forecasting measures of real activity in Australia. In a policy context, we are interested in forecasting the levels or growth of policy relevant variables throughout the cycle. We are less interested in forecasting turning points in the cycle or in forecasting coincident indices, which are subjectively defined overall measures of economic activity. This gives us a different focus to much of the recent work done in this area. We use a simple forecasting framework (bivariate VARs) to compare the Westpac-Melbourne Institute (WM), NATSTAT and ABS leading indices’ predictive performance for real GDP, employment and unemployment in Australia. Within sample we find all three indices help predict all of the activity variables, although with varying leads. Out of sample evidence, however, is weaker. Within our framework, we only find evidence in favour of the WM index when used to forecast GDP. Otherwise, the indices do not make any substantive contribution to forecast quality. To gauge the usefulness of the simple bivariate VAR models, we compare the out of sample forecasts of GDP, using the WM index, to those from a single equation structural model due to Gruen and Shuetrim (1994). Over a forecasting sample of relatively stable growth, the WM index model performs quite well relative to the Gruen and Shuetrim model. Over a longer forecasting sample period, one which includes the downturn in the early 1990s, there is some evidence that the WM index model performs relatively poorly.