

Financial surveys are very fashionable lately and for a great motive. They inform us how the parents being surveyed—skilled forecasters, households, agency managers—really feel in regards to the economic system. So, as an example, the New York Fed’s Survey of Shopper Expectations (SCE) web site shows an inflation uncertainty measure that tells us households are extra unsure about inflation than they have been pre-COVID, however a bit lower than they have been a number of months in the past. The Philadelphia Fed’s Survey of Skilled Forecasters (SPF) tells us that forecasters believed final Could that there was a decrease threat of unfavorable 2024 actual GDP progress than there was final February. The query addressed on this submit is: Does this data even have any predictive content material? Particularly, I’ll concentrate on the SPF and ask: When skilled forecasters point out that their uncertainty about future output or inflation is greater, does that imply that output or inflation is definitely turning into extra unsure, within the sense that the SPF could have a tougher time predicting these variables?
Measuring Uncertainty in Probabilistic Surveys—a Fast Recap
In my companion submit I lined some materials that’s going to be useful for understanding this submit. Let me summarize it for you:
- Probabilistic surveys ask respondents not just for level predictions (that’s, one quantity, akin to the reply to the query “What’s output progress going to be in 2024?”), but in addition attempt to elicit the whole likelihood distribution of potential outcomes. The SPF asks for each level forecasts and possibilities. Particularly, SPF forecasters present probabilistic forecasts for actual GDP progress and GDP deflator inflation for the present and the next yr. That is the survey used on this submit.
- There are a selection of approaches for translating the reply to probabilistic surveys into measures of uncertainty. On this submit I’m going to make use of the method developed in a current Workers Report with my coauthors Federico Bassetti and Roberto Casarin. For some questions the method used issues as mentioned within the first submit on this sequence.
- Skilled forecasts massively disagree on how a lot uncertainty there’s on the market, and so they additionally change their evaluation of uncertainty over time (as is the case for the SCE inflation uncertainty measure talked about above).
- These variations can in precept be defined by a idea referred to as the noisy rational expectations (RE) speculation: forecasters obtain each private and non-private alerts in regards to the state of the economic system, which they don’t observe. Variations within the variance of the personal alerts clarify variations in uncertainty throughout forecasters (mainly, some people are higher forecasters than others), and modifications within the variance of both the general public (for instance, as a consequence of one thing like COVID) or personal alerts can clarify why they modify their thoughts about uncertainty.
Does Subjective Uncertainty Map into Goal Uncertainty, That Is, Forecast Errors?
Nonetheless, beneath rational expectations it higher be that forecasters which can be extra unsure are really worse forecasters, within the sense that they make on common worse forecast errors. And that in the event that they really feel financial uncertainty has declined, they need to be capable of really predict the economic system higher if RE holds. By regressing ex-post forecast errors (particularly, the logarithm of their absolute worth) on subjective uncertainty (particularly, the logarithm of its normal deviation) throughout forecasters and time, we take a look at whether or not subjective uncertainty maps into goal uncertainty. That’s, we take a look at whether or not it’s the case that when ex-ante uncertainty will increase by, say, 50 p.c, both over time or throughout forecasters, ex-post uncertainty additionally will increase by the identical quantity. As a byproduct, we additionally take a look at whether or not the noisy rational expectations speculation holds water.
Within the Workers Report I discussed we run this regression for the panel of particular person SPF forecasters from 1982 to 2022 for each output progress and inflation, and for eight totally different forecast horizons—from eight to at least one quarter-ahead (recall that SPF forecasters present probabilistic forecasts for progress and inflation for the present and the next yr; therefore the eight quarter-ahead makes use of the surveys performed within the first quarter of the yr earlier than the belief; the one quarter-ahead makes use of the surveys performed within the fourth quarter of the identical yr). The thick dots in charts beneath present the OLS estimates and the whiskers point out 90 p.c posterior protection intervals based mostly on Driscoll-Kraay normal errors.
Do Variations in Subjective Uncertainty Map into Variations in Forecast Accuracy?

Supply: Creator’s calculations.
For output progress, for horizons longer than 5 quarters we can’t reject the speculation that the coefficient is zero and the purpose estimates are all small: In sum, we discover no relationship between subjective uncertainty and the scale of the ex-post forecast error for horizons past one yr. Because the forecast horizon shortens, the connection turns into tighter, and for horizons of three quarters or much less we can’t reject the speculation that the coefficient is one, that’s, we can’t reject the noisy RE speculation. (Recall that the coefficient measures the extent to which ex-post uncertainty modifications when ex-ante uncertainty modifications, both over time or throughout forecasters. A worth of zero signifies that modifications in subjective or ex-ante uncertainty haven’t any bearing by any means on ex-post forecast errors; a worth of 1 signifies that ex-ante and ex-post uncertainty transfer in unison, as RE would indicate.)
For inflation, the OLS estimates hover between 0.2 and 0.5 for longer horizons, however improve towards one because the horizon shortens, with estimates which can be additionally not considerably totally different from one for horizons of three quarters or much less. The appendix of the Workers Report reveals that these outcomes are broadly sturdy to totally different samples and particularly don’t depend upon together with the COVID interval.
The chart beneath reveals the estimates controlling for time, forecaster, and each time and forecaster-fixed results. The outcomes with time-fixed results (prime panels) take a look at whether or not cross-sectional variations in subjective uncertainty throughout forecasters map into related variations in forecast errors, controlling for elements that have an effect on all forecasters (akin to recessions and the COVID interval). For output, the reply is totally not at lengthy horizons, though for brief horizons the correspondence between the 2 turns into tighter, if not fairly one-to-one. For inflation the connection can also be removed from one-to-one at longer horizons, however for horizons lower than two quarters the coefficient is indistinguishable from one.
The underside panels present the outcomes controlling for forecaster-fixed results. They present that when subjective uncertainty modifications over time—both due to combination uncertainty shocks or as a result of the standard of their personal sign has modified—on common it has no bearing for forecast accuracy for longer horizons, however maps one-to-one into corresponding modifications within the absolute forecast errors for brief horizons. The outcomes with forecaster-fixed results make clear whether or not forecasters accurately anticipate intervals of macroeconomic uncertainty. They clearly don’t for any horizon longer than one yr. However when they’re already in a excessive uncertainty interval (that’s, for brief horizons), that is mirrored of their subjective uncertainty.
Controlling for Fastened Results

What will we conclude from the proof mentioned on this sequence of posts? Skilled forecasters clearly don’t behave on common just like the noisy rational expectations mannequin suggests. However the proof additionally signifies that any proposed idea of deviations from rational expectations higher account for the truth that such deviations are horizon dependent. No matter biases forecasters have, whether or not as a consequence of their previous experiences or different elements, appear to go away because the horizon will get nearer. In different phrases, skilled forecasters can’t actually predict what sort of uncertainty regime will materialize sooner or later, however they appear to have a great grasp of what regime we’re in in the intervening time.

Marco Del Negro is an financial analysis advisor in Macroeconomic and Financial Research within the Federal Reserve Financial institution of New York’s Analysis and Statistics Group.
The right way to cite this submit:
Marco Del Negro, “Can Skilled Forecasters Predict Unsure Instances?,” Federal Reserve Financial institution of New York Liberty Road Economics, September 4, 2024, https://libertystreeteconomics.newyorkfed.org/2024/09/can-professional-forecasters-predict-uncertain-times/.
Disclaimer
The views expressed on this submit are these of the creator(s) and don’t essentially mirror the place of the Federal Reserve Financial institution of New York or the Federal Reserve System. Any errors or omissions are the duty of the creator(s).