Big-time kudos to Calculated Risk, who uncovers this paper by Dean Croushore that tests the value of consumer sentiment measures when forecasting with data that is available in real time. (That is, data before it is revised using information that becomes available after a forecast has been made.) Dean, now at the University of Richmond, was one of the driving forces behind the Philadelphia Fed's indispensable real-time data collection. His work may set the new standard on this issue.
Could a researcher or policy analyst use data reported from surveys of consumer confidence to improve forecasts of consumer spending? This issue has been examined in the literature previously, which reached the conclusion that consumer confidence data helped improve the forecasts slightly, but not statistically significantly. But that research was based on latest available data and thus did not use the data that would have been available to forecasters in real time. This paper remedies that shortcoming, using the Real-Time Data Set for Macroeconomists to analyze the quality of forecasts made with indexes of consumer confidence. We conjecture that using real-time data might show greater marginal significance for consumer confidence because the surveys might capture effects that will not appear in the data until they are revised. The main finding is that the indexes of consumer confidence are not of significant value in forecasting consumer spending. In fact, in some cases, they make the forecasts significantly worse, suggesting that consumer confidence surveys are no better than government data agencies in capturing information about consumer spending.
UPDATE: CR has the response from the University of Michigan and Conference Board folks in its post on the Croushore paper (via this AP post.) They beg to differ with Dean. I say fine -- let's see the evidence, and let the debate begin (er. continue).