Examining the Beliefs of Professional Forecasters

where ${\epsilon }_t$ is the error term and α is a constant term. Our coefficient of interest β represents the change in the current quarter-ahead forecast for variable y in response to current news about variable x. Table 1 reports the OLS estimates of β.

Table 1: Relationship between Forecast Revisions and News

Note: The table reports the OLS estimates of β in equation (1).

We find that professionals’ quarter-ahead forecasts are particularly sensitive to news about the unemployment rate and the short-term TBill rate. When the unemployment rate is higher than expected, forecasters revise downward their quarter-ahead expectations for real GDP growth, inflation, and interest rates, but they revise upward their quarter-ahead expectations for the unemployment rate. This result is consistent with professional forecasters’ having a demand-side view of the economy, as documented in Candia, Coibion, and Gorodnichenko (2023).⁵ If the three-month TBill rate is higher than expected, forecasters revise upward their expectations for real GDP growth, inflation, and future interest rates, and revise downward their expectations for the unemployment rate.⁶ On the other hand, we find that news about inflation or real GDP growth has a relatively small effect on professionals’ quarter-ahead forecast updates.

We then consider a richer setting: we regress the average forecast revisions of variable y on news about any of the four variables, that is, real GDP growth, the unemployment rate, inflation, and three-month TBill rate:

Our coefficients of interest ${\beta }_{GDP}^y$, ${\beta }_U^y$, ${\beta }_{\pi }^y$, and ${\beta }_{TB}^y$ capture the associations between forecast revisions of variable y and news about real GDP growth, unemployment rate, inflation, and the three-month TBill rate, respectively; α is a constant term, and ${\epsilon }_t$ is an error term. Because news about different variables is likely correlated with one another, we estimate equation (2) using the least absolute shrinkage and selection operator (LASSO) technique.⁷ This prevents the model from being too sensitive to the data (overfitting), and it sets the weight of irrelevant news to 0.

We summarize our findings in Figures 1 and 2. Each subplot reports the optimal β estimates resulting from the LASSO technique applied to equation (2). Figure 1 reports the β estimates pertaining to forecast revisions in real GDP growth and unemployment, whereas Figure 2 plots the estimates related to forecast revisions in the inflation and three-month TBill rates.⁸ The figures highlight two key findings. First, news about the unemployment rate or the short-term TBill rate is the relatively more important news in terms of professionals’ updating their forecasts, while news about inflation or the real GDP growth has little to no effect on professionals’ forecast updates.

Second, news about the unemployment rate carries a particularly important weight when professionals update their annual forecasts about real GDP growth or unemployment itself, whereas news about the three-month TBill rate is important for how professionals update their inflation or TBill rate forecasts. A 10-basis-point increase in the unemployment rate is associated with 4.6-basis-point decrease in the forecast for year-ahead annual real GDP growth and with a 27-basis-point increase in the average forecast for the unemployment rate in the next year. The latter effect is relatively big, and it could reflect the fact that, historically, once unemployment has started to increase, the sustained rise has been generally steep. On the other hand, news about the short-term TBill rate carries a relatively higher importance for updates of inflation or TBill rate forecasts. Specifically, a 1 percentage point unanticipated increase in the TBill rate is associated with a 14-basis-point increase in the forecast for the year-ahead annual inflation rate and with a 73-basis-point increase in the average forecast for the TBill rate during the next year. As we argued earlier, the effects of news about the short-term TBill rate are likely a reflection of the information effect discussed in Nakamura and Steinsson (2018). Namely, if professionals associate a positive surprise in the TBill rate with an unexpected increase in the federal funds rate, they might interpret that to mean that policymakers anticipate a stronger economy featuring stronger real GDP growth, a lower unemployment rate, or higher inflation. This interpretation can prompt professionals to revise their own forecasts and report higher real GDP forecasts and lower unemployment rate forecasts. Consistent with this view, Figure 1 shows that an unexpected rise in the TBill rate is associated with a slight downward revision of the unemployment forecast but an upward revision of the real GDP growth outlook.⁹

Finally, our results indicate that inflation news has little effect on professionals' outlook for real activity, and unemployment news has little bearing on their inflation outlook. This contrasts with some existing findings in the literature that look at the correlation between inflation and real GDP growth forecasts for consumers and firms in the United States. For instance, Kamdar (2019) finds that US consumers associate higher future inflation with lower real GDP growth, whereas Candia et al. (2026) document a negative correlation between firms’ inflation and real GDP growth outlooks in the United States.¹⁰ Consistent with the results of this Economic Commentary, Candia et al. (2023) document a weak link between inflation and the real GDP growth outlook of professional forecasters in the Livingston survey.

Conclusion

Our analysis reveals patterns in how professional forecasters update their forecasts in response to macroeconomic news. The evidence demonstrates that forecasters are particularly responsive to surprises in unemployment and the three-month TBill rate, with these variables serving as powerful signals that trigger substantial revisions across multiple forecast dimensions. When the rate of unemployment is higher than expected, forecasters revise downward their outlook for real GDP growth and upward their unemployment outlook. On the other hand, and what is likely a reflection of the information effect, we find that unexpected increases in the three-month TBill rate lead professionals to anticipate higher future inflation and TBill rates but lower future unemployment.  While the literature has documented a strong, negative link between firms’ or consumers’ outlooks for inflation and real GDP growth, we find a weak relationship between professionals’ outlook for inflation and real GDP growth.

This Economic Commentary sheds new light on modeling expectations formation processes that, as economic theory shows, affect macroeconomic fluctuations and the transmission of monetary policy. Professionals tend to compartmentalize their forecast updates by economic domain: they use news about the real side of the economy to update forecasts about real variables, while they rely on news about nominal variables to update forecasts about nominal variables.