RDP 2001-09: What do Sentiment Surveys Measure? 5. Discussion

The results presented in the previous two sections indicate that we can explain a substantial proportion of variation in the sentiment indices. Furthermore, the unexplained variation does not seem particularly useful for predicting the future. The indices themselves have some predictive power, especially with respect to employment growth, but this simply corroborates a point well made in the existing literature. Indeed it would be surprising if a linear combination of lagged activity and financial variables did not predict some variation in economic activity, since GDP itself is likely to be included in that combination.

As far as the business surveys are concerned, only the residuals from the average NAB Actual Conditions index and the two Expected Employment indices appear to have much predictive power (vis-à-vis employment growth), suggesting that this index might be more useful than others as a leading indicator. The consumer confidence residuals (especially those gauging personal financial conditions) perform somewhat better overall, but again usually only as predictors of employment growth.[22] The success of specific filtered indices as predictors of employment growth is not necessarily surprising. Indeed, it is intuitively reasonable that people might factor a variety of informal information about their own employment prospects, and those of their friends, peers and employees into their assessments of economic conditions.

Indices corresponding to different forecast horizons (in both their original and filtered forms) exhibit limited variation in their predictive power. The regressions show that longer horizon variables (Expected Business Conditions (12-month Outlook), and Index 4 from both consumer confidence surveys) tend to perform more poorly than shorter horizon variables. Moreover, the residuals from regressions of forward-looking indices generally predict the future no better than those from backward-looking indices. A plausible interpretation could be that people are better at making assessments of current conditions and conditions over the next quarter than they are at developing more extended forecasts. After all, much the same can be said of professional forecasters.


We view the relatively strong results for Expected Employment – 12-Month Outlook and Index 1 of both consumer confidence surveys with caution, however, as these residuals were taken from regressions that showed signs of parameter instability. [22]