A Closer Look at the Behavior of Uncertainty and Disagreement: Micro Evidence from the Euro Area

• Date: 01 February 2021
• Author: ROBERT RICH,JOSEPH TRACY
• DOI: http://doi.org/10.1111/jmcb.12728
• Published date: 01 February 2021

DOI: 10.1111/jmcb.12728

ROBERT RICH

JOSEPH TRACY

A Closer Look at the Behavior of Uncertainty and

Disagreement: Micro Evidence from the Euro Area

ABSTRACT: This paper examines point and density forecasts of real

GDP growth, ination, and unemployment from the European Central

Bank’s Survey of Professional Forecasters. We analyze individual uncer-

tainty measures as well as introduce individual point- and density-based

disagreement measures. The analysis indicates forecasters’ uncertainty and

disagreement display substantial heterogeneity and persistence, with the lat-

ter feature challenging a key prediction of expectations models emphasiz-

ing information frictions. We also nd that uncertainty is characterized by

prominent respondent effects and disagreement by prominent time effects,

suggesting these divergent properties underlie the well-documented weak

uncertainty–disagreement linkage. Takentogether, our results provide a ba-

sis for further development of expectations models.

I’     area of con-

siderable research interest because of its importance for understanding decision-

making, as well as for explaining movements in economic and nancial variables.

The majority of studies have focused on expectations, but other dimensions of fore-

cast behavior such as disagreement and uncertainty can also play important roles

in such analyses. Despite their importance, the measurement of disagreement and

uncertainty—like the measurement of expectations—is challenging because of the

inherent difculty in observing individuals’ subjective assessments. While manysur-

veys offer information about point forecasts and their dispersion across respondents,

The authors thank three anonymous referees and Ken West (the editor) for helpful comments on an

earlier version of this paper. The authors have benetted from the suggestions of Stefania D’Amico as

well as conference participants at the Fall 2017 System Committee meeting and the 5th Workshop on

Empirical Macroeconomics. Michael Stewart, Michael Fosco, Daniel Reuter, Max Sterman, and Victoria

Consolvo provided excellent research assistance. The viewsexpressed in the paper are those of the authors

and do not necessarily reect those of the Federal Reserve Bank of Cleveland, the Federal Reserve Bank

of Dallas, or the Federal Reserve System.

R R is at Federal Reserve Bank of Cleveland (E-mail:Robert.rich@clev.frb.org). J

T is at Federal Reserve Bank of Dallas (E-mail:Joseph.Tracy@dal.frb.org).

Received July 24, 2018; and accepted in revised form February 28, 2020.

Journal of Money, Credit and Banking, Vol. 53, No. 1 (February 2021)

© 2020 The Ohio State University

234 :MONEY,CREDIT AND BANKING

most typically do not provide information about the degree of condence that respon-

dents attach to their point forecasts.

This paper examines point and density forecasts from the European Central Bank’s

Survey of Professional Forecasters (ECB-SPF) that elicits euro area expectations for

real GDP growth, a harmonized index of consumer prices (HICP) ination and the

unemployment rate. In particular, we use the ECB-SPF data to investigate forecast-

ers’ uncertainty and disagreement. While other work has focused on these topics, a

key aspect of our study is that we conduct the analysis at the individual level. This

approach offers a deeper exploration into the behavior of uncertainty and disagree-

ment and a richer characterization of their features that are critical for the evaluation

of models of expectations formation.

The individual forecast data allow us to make several contributions to the existing

literature, two of which warrant special mention. First, we construct individual mea-

sures of uncertainty and disagreement, with the latter involving the introduction of

new respondent-specic measures of disagreement derived from both point and den-

sity forecasts. Importantly, our metrics of the divergence between density forecasts

allow us to extend the notion of disagreement beyond its conventional association

with differences in point forecasts. We also demonstrate how a density-based mea-

sure of disagreement can be more informative than its point-based counterpart.

Second, we investigate the properties of the individual measures of uncertainty

and disagreement across the three forecast variables. Specically, we examine the

cross-sectional behavior of uncertainty and disagreement, as well as their movements

over time. This analysis provides a useful background for our formal investigation

into the issues of heterogeneity and persistence. Moreover, and in contrast to studies

conducted at the aggregate level, we consider the roles of respondent and time effects

in the behavior of uncertainty and disagreement.

The empirical analysis yields several ndings of note. The results indicate forecast-

ers’ uncertainty and disagreement display substantial heterogeneity and persistence,

with the latter feature challenging a key prediction of expectations models that em-

phasize information frictions. In particular, this class of models—sticky information

models and noisy information models—can generate heterogeneity in forecasters’

behavior, but not differences that are systematic in nature.1The results also indicate

individual disagreement and individual uncertainty are associated, respectively, with

prominent time effects and prominent respondent effects, indicating that the extent of

forecasters’ personal disagreement varies over time much more than their predictive

condence. Importantly, these divergentproperties not only can explain our evidence

documenting an economically insignicant relationship at the individual level be-

tween uncertainty and both the point- and density-based disagreement measures, but

also the evidence from several studies at the aggregate levelindicating that disagree-

ment is not a reliable proxy for uncertainty. Taken together, these empirical features

provide a basis for the further development of expectations models.

1. See Mankiw and Reis (2002) for sticky information models and Sims (2003), Woodford (2003),

and Mackowiak and Wiederholt (2009) for noisy information models.