Content uploaded by Jonathan Benchimol
Author content
All content in this area was uploaded by Jonathan Benchimol on Aug 16, 2022
Content may be subject to copyright.
Do Expert Experience and Characteristics
A¤ect In‡ation Forecasts?
IAAE 2022 Annual Conference
International Association for Applied Econometrics
Jonathan Benchimol,1Makram El-Shagi,2and Yossi Saadon1
This pre senta tion do es not n ecess arily re‡ect the v iews of th e Bank o f Israel
June 23, 2022
1 / 50
Expert Forecasts and the Fed
ITo decide, decision-makers rely a lot on expert forecasts.
IFOMC meeting minutes: "forecaster" appears 6 times over the
pre-GFC decade (88 meetings) with a more than fourfold
increase over the post-GFC decade (25 times over 82
meetings).
IInterest rate announcements: "survey" does not appear in the
pre-GFC decade, while it appears 29 times during the
post-GFC decade.
IChairman’s speeches: "forecaster" appears only 8 times in the
216 governor speeches during the pre-GFC decade but 58 times
in only 160 governor speeches over the post-GFC decade.
IWho are the best and bolder experts, and why?
2 / 50
Motivation
IAgents rely on expert forecasts (Piotroski and Roulstone, 2004).
IExpert forecasts are generally better than market-based
forecasts (Adeney et al., 2017; Benchimol and El-Shagi, 2020).
IEconomic agents mostly rely on the average of expert
forecasts (Genre et al., 2013; Budescu and Chen, 2015).
IForecasts are aggregated in simplistic ways and policymakers
rank them without considering experts’characteristics
(Alessi et al., 2014; Coibion et al., 2020).
ITime-varying relative forecast performance of several forecasts
helps to select appropriate ones and generate better forecast
combinations (Giacomini and Rossi, 2010).
IExpert forecasts are subject to a plethora of factors that
psychologically and rationally a¤ect the forecaster:
Iattention to economic variables (Gabaix, 2019, 2020),
Ibehavioral biases (Thomas, 1999; Davis and Lleo, 2020),
Iasymmetric information (Keane and Runkle, 1990),
Iothers (Lim, 2001; Coibion and Gorodnichenko, 2015).
3 / 50
Labor and Behavioral Economics
IMincer (1974)
IThe wage is a function of the number of years of education
and experience.
ILaster, Bennett and Geoum (1999)
IThe wage of a forecaster is a function of his behaviors
(performance and herding).
IForecasting and opportunity costs.
IThomas (1999)
ICharacteristics and expert-speci…c non-rational expectations.
ITerence (2001), Capistran and Timmermann (2009), Coibion and
Gorodnichenko (2015)
IExpectation formation of analyst and expert forecasts.
IMalmendier and Nagel (2016) and Malmendier, Nagel and Yan
(2021)
IHigh in‡ation experience and in‡ation forecasts.
4 / 50
Understanding Expert Forecasts is Crucial
IPeople who experienced low stock market returns invest less
in the stock market: personal experience shapes optimism
and pessimism (Malmendier and Nagel, 2011).
IPolicymakers’in‡ation experience a¤ects their in‡ation
forecast (Malmendier and Nagel, 2016; Malmendier et al., 2021):
high in‡ation periods experienced by experts may in‡uence
their forecasting performance and boldness.
IDeterminants of experts’over- and underreaction in
predicting in‡ation (Barberis et al., 1998; Daniel et al., 1998).
5 / 50
Models
ITheoretical model
IIndividual characteristics
=)private and public information interpretation
=)forecast performance.
IEmpirical models
IIndividual characteristics
=)forecast performance and boldness.
6 / 50
Methodology
IDatabase 1: expert behaviors about the U.S. CPI by name
and a¢ liation (Bloomberg).
IPerformance (error).
IBoldness (herding)
IDatabase 2: forecaster characteristics (LinkedIn + websites).
IEducation: level, type, and quality.
IExperience: place, type, and time.
IA¢ liation: type and place.
IGender
IAge.
ITest: survival, panel estimations, forecasting ability,
pessimism.
7 / 50
Take-away
IEducation quality and type in‡uence expert’s survival.
IThe expert’s location, experience type, gender, and institution
location and type are critical determinants of his forecasting
behaviors.
IThe experience duration, in a CB, or with high in‡ation
periods signi…cantly in‡uences the expert’s sentiment and
under-reaction behaviors.
8 / 50
Data: CPI Expert Forecasts
IAnyone–including the experts–can know the published expert
forecasts at each point in time.
IEach expert can publish or update his forecasts to Bloomberg
until about 15 days before the publication of the e¤ective
CPI in‡ation data.
IAn expert not updating his forecasts according to additional
information prior to the e¤ective data publication is not
rational.
IConsequently, all the collected expert forecasts are the best
ones for each forecaster.
9 / 50
Dataset in a Nutshell (1)
I151 experts and 164 institutions published at least one US
CPI in‡ation forecast.
I31 forecasts per expert on average.
IInstitution types:
Iprivate …nancial, academic, CB, or other.
IPast job experience (type and duration): academic experience,
CB experience, and experience in …nancial institutions.
IThe expert’s experience duration corresponds to the time since
the …rst experience in forecasting according to the expert’s CV.
IAbout 23 years of experience in forecasting on average.
ILocation (local or foreign), education (level, type, and
quality), gender (male or female), and a¢ liation (type and
location).
IAbout 41% of the US CPI in‡ation experts are Americans.
IFemales represent about 12% of the experts.
10 / 50
Dataset in a Nutshell (2)
I77% of the experts have at least one degree (BA, MA, or
PhD) in Economics
IExperts have 218 months of experience in …nancial institutions
on average and only ten and eight months of experience in
academia and CB, respectively.
IThe highest degree of about 8% is a BA, 52% a MA, 26% a
PhD, and the rest has no identi…ed highest degree.
IAbout 39% of the US CPI in‡ation experts are graduated from
a top-ranked university (BA, MA, or PhD).
IThe date of birth is generally not reported on LinkedIn
IWe scrapped the available CVs from the internet and
Bloomberg to gather birth dates (less available but more
precise data).
IThe measure of age used in the paper is based on the available
date of birth.
IExperts that faced high in‡ation periods during their youth or
adolescence (born before 1973).
11 / 50
Dataset Summary
Summary Statistics
Average Std. Dev. Average
% of the sample
Forecasts per forecaster 43.4 36.6 Local forecaster 44.7
Forecast error 0.14 0.12 Local institution 30.6
Forecast boldness 0.08 0.09 Financial institution 92.9
Experience: Central bank (m) 10 31 Experience: Central bank 16.5
Experience: Academia (m) 10 26 Experience: Academia 20.0
Experience: Total (m) 185 143 Experience: High In‡ation 54.1
Gender 12.9
Education: MA 58.8
Education: PhD 28.2
Education: Economics 71.8
Education: Finance 14.1
Ranking: Economics 28.2
Ranking: Finance 29.4
Note: (m) indicates the number of months. This table presents a summary of
the data at our disposal.
12 / 50
Theoretical Model: Setting
IEven using a straightforward form of learning (Bayesian
learning), experience has nonlinear e¤ects depending on initial
conditions.
IConsider a forecaster exposed to both public (su
t) and private
(sr
t) signals aiming to forecast in‡ation π.
IBoth signals are normally distributed, isolated and mutually
independent, and unbiased but noisy:
su
t N πt,τ1
u,(1)
and
sr
t N πt,τ1
r,(2)
IThe precision of the public (private) signal, τu(τr), is known
and constant (unknown to the forecaster initially).
IThe expert learns about the quality (precision) of his private
signal over time through Bayesian learning.
13 / 50
Theoretical Model: Distributions
IFor simplicity: the prior regarding the precision of the private
signal is Gamma distributed, where the initial prior has a
mean of ˜
τ0and a variance of σ2
τ,0.
IThe mean of new distribution is given by
˜
τt=˜
τ2
t1/σ2
τ,0+1/2
˜
τt1/σ2
τ,t1+1/2sr
t1πt12,(3)
and the variance by
σ2
t=˜
τ2
t1/σ2
τ,0+1/2
˜
τt1/σ2
τ,t1+1/2sr
t1πt122.(4)
14 / 50
Theoretical Model: Weighted Forecast
IThe Bayesian point estimate for the average forecaster who
starts forecasting at t=0 and does so every period at time t
is
˜
τt=˜
τ2
0/σ2
τ,0+(t1)/2
˜
τ0/σ2
τ,0+t/2τr
.(5)
IA forecaster maximizing the expected precision of his forecast,
will then provide a weighted forecast
ft=τr
τr+˜
τt
su
t+˜
τt
τr+˜
τt
sr
t,(6)
implying a precision of
τt= τr
τr+˜
τt2
τ1
u+˜
τt
τr+˜
τt2
˜
τ1
t!1
.(7)
15 / 50
Theoretical Model: Nonlinearities and Interactions
1. Unless ˜
τ0=τu, the forecast performance will improve over
time, since any misconception (optimistic or pessimistic)
regarding the quality of the private signal will lead to a
suboptimally weighted forecast.
2. Precision converges monotonically to τ=τu+τr.
IA range of factors can drive forecast quality τrand the initial
optimism (pessimism) regarding one’s own forecast quality ˜
τ0.
IThe di¤erent trajectories of performance in response to time
are visible: factors that drive forecast performance–actual
quality and suboptimal weighting–also a¤ect the
performance’s response to experience (here shown as time).
IThe model predicts that both heterogeneity in prior
expectations and understanding of news across forecasters
may drive dispersion in forecasts.
IHeterogeneous interpretation of the incoming information
might exacerbate the dispersion of forecasts.
16 / 50
Outcome measures
IPerformance (RMSE) is
Li,t=πtEi
tτ[πt]2,(8)
where iis a speci…c expert, and tτis a point in time before t
but clearly after t1, since forecasters can update until the last
moment.
IBoldness (herding) is
Bi,t=Ei
tτ[πt]Mi,tEi
tτ[πt]2,(9)
where Mi,tis the median operator over all expert forecasts iat
time t.
17 / 50
Survival Analysis: Theory
IHow likely an expert will submit another forecast in the future
based on the current (and past) performance?
IThis hazard model predicts the risk of being removed from
the sample, i.e., being …red, retired (or forced to), or
transferred to another job.
IProportional hazards model (Cox, 1972)
hit =h0(t)eλ0f(Yit 1)+XiΛ,(10)
where Yit1is the past history of yit ,h0(t)corresponds to
the time-speci…c e¤ect, Λis a vector of regression
coe¢ cients, and λ0is a regression coe¢ cient.
I3 di¤erent transformations of f(Yit 1)describing slightly
di¤erent possibilities of how performance review might be
conducted by the hiring institutions: Raw, MA5, and wMA5.
18 / 50
Survival Estimates: Nonlinear Model
Performance Boldness
Raw MA5 wMA5 Raw MA5 wMA5
Dependent variable 2.509 8.14*** 5.173** 3.912 11.612** 7.651*
Age >60 years old -0.089 -0.124 -0.103 -0.115 -0.137 -0.13
Local forecaster (LF) -0.674* -0.668* -0.668* -0.682* -0.69* -0.685*
Local institution (LI) 0.173 0.199 0.19 0.196 0.176 0.196
Financial institution (FI) -0.767* -0.711* -0.754* -0.779** -0.744* -0.765**
Experience: Central bank (CB) 0.4 0.425 0.408 0.398 0.4 0.399
Experience: Academia (AC) 0.759* 0.799** 0.791** 0.737* 0.739* 0.747*
Gender (G) 0.319 0.38 0.349 0.335 0.376 0.355
Education: MA (MA) -0.196 -0.276 -0.231 -0.218 -0.249 -0.236
Education: PhD (PhD) -0.588 -0.697 -0.641 -0.576 -0.596 -0.591
Education: Economics (EC) -1.126** -1.11** -1.086** -1.151*** -1.157*** -1.141***
Education: Finance (EF) -0.528 -0.534 -0.483 -0.565 -0.539 -0.534
Ranking: Economics (RE) -0.07 -0.076 -0.076 -0.086 -0.094 -0.093
Ranking: Finance (RF) 0.057 0.067 0.057 0.063 0.069 0.064
Interactions
MA RF -0.537* -0.562** -0.55* -0.54* -0.558** -0.548**
PhD RE -0.598** -0.602** -0.596** -0.58** -0.578** -0.571*
Note: ***, **, and * indicate signi…cance at the 1%, 5%, and 10% levels,
respectively. Raw stands for the last period dependent variable, MA5 for
…ve-period moving average (backward-looking) and wMA5 for weighted moving
average where the most recent periods have a higher weight.
19 / 50
Survival Analysis: Results
IA degree in economics immunizes the expert; a degree in
…nance does not o¤er the same protection.
IExperts with EC may provide more convincing explanations to
back their information processing and subsequent forecasts of
the US CPI, an economic variable par excellence, than experts
with EF, increasing their survival rate.
IGraduated from a top-ranked university in economics or
…nance: improves the expert’s survival rate.
IExperience in academia: deteriorates expert’s survival rate.
IThe expert’s survival rate seems to be determined more by
past rather than recent performances. Low average
performance (MA5) signi…cantly decreases the expert’s
survival rate, while herding in‡uences the expert’s survival
rate less signi…cantly.
ISeveral theories of reputation and herd behavior indicate that
agents’performance and boldness may vary with career
concerns (Scharfstein and Stein, 1990; Zwiebel, 1995).
20 / 50
Survival Analysis: Interpretation
IIncreased reputational capital may increase the labor market
attractiveness of top-ranked pro…les:
IExperts originating from/targeting academia tend to leave the
profession more frequently than other experts.
IMight explain that these experts are …red or moved to another
occupation (in the same or another institution) more often.
Ie.g., they are hired based on higher or di¤erent expectations
than others, which they may not meet.
IMA/PhD graduation from a top institution in
…nance/economics improves the expert’s survival.
IContrasts with an education in …nance that does not
signi…cantly in‡uence the expert’s survival.
IExpert’s institution is not more an intermediary than a factor:
an expert with a PhD might increase his chances of working in
a better institution, with better information-gathering
functions, than an expert without a PhD.
21 / 50
Survival Estimates: Firm and Information
IFinancial institutions may have better information-gathering
functions, data, and private information access.
IExperts with a PhD may be attracted by these institutions,
and these institutions may prefer recruiting people with a PhD,
making a case for …rm-e¤ects that are both an intermediary
and a factor.
INot enough companies with more than one forecaster in the
sample (otherwise we would be able to control for …rm
…xed-e¤ects): it is unlikely that the institution is an
intermediary rather than a factor.
IAll our experts have access to a Bloomberg terminal, so we
can reasonably assume they all have access to almost the
same information and necessary equipment to build their
forecasts. Hence, the e¤ect of the PhD is more prevalent than
…rm-e¤ects.
22 / 50
Random E¤ects: Theory
IWe estimate
yit =ΓXit +ΛZi,t+ui+vt+εi,t,(11)
where Xand Zdenote forecaster and institution-speci…c
characteristics, respectively, and Γand Λthe corresponding
vectors of coe¢ cients. ui,vt, and εit represent the
forecaster-random e¤ect, time-…xed e¤ect, and idiosyncratic
component of the error term, respectively.
IWe split our sample (1997:Q1-2017:Q4) into two subsamples
Ipre-GFC (1997:Q1-2008:Q1)
Ipost-GFC (2008:Q1-2017:Q4)
23 / 50
Individual Random E¤ects: Nonlinear Model
Performance Boldness
Full sample Pre-GFC Post-GFC Full sample Pre-GFC Post-GFC
Experience -0.0003 -0.0156** 0.0017 -0.0018** -0.0182*** -0.0004
Age >60 years old 0.0016 0 -0.0015 0.0042 0 0.0022
Local forecaster (LF) 0.0037 -0.0145 0.0137 0.0068 -0.0097 0.007
Local institution (LI) -0.0114 -0.0014 -0.0241** -0.008* -0.0113 -0.01**
Financial institution (FI) -0.0067* -0.0662*** -0.004 -0.0016 -0.0479*** -0.001
Experience: Central bank (CB) -0.0044 -0.0167 -0.0036 -0.0016 -0.0203 -0.0001
Experience: Academia (AC) -0.0117*** -0.057*** -0.0082** -0.0069*** -0.0512*** -0.0048**
Gender (G) 0.0014 0.0686*** 0.0039 0.0023 0.0733*** 0.002
Education: MA (MA) -0.0047 0 -0.0128** -0.0032 0 -0.0069**
Education: PhD (PhD) 0.0037 0.0267 0.0031 -0.0023 0.0332 -0.0025
Education: Economics (EC) 0.0147 0.0172 0.0139* 0.0062 0.013 0.0009
Education: Finance (EF) 0.0092 0.0355 0.0156* -0.0068 0.0479 -0.0049
Ranking: Economics (RE) 0.0024 -0.0062 0.0018 0.0020** -0.0077 0.0018**
Ranking: Finance (RF) -0.0007 0.0097 -0.0003 -0.0003 0.0101 -0.0001
Constant 0.0372 0.0329 0.0781*** 0.0028 -0.0316 0.0242**
Two-way interactions
MA RE -0.0174*** 0 -0.0152** -0.0093*** 0 -0.0079***
MA RF 0.0139** 0 0.0104** 0.0081** 0 0.0058**
PhD RE -0.0119* 0.0196 -0.0109* -0.009** 0.0214 -0.0089***
PhD RF 0.0075 -0.0298* 0.0077 0.0038 -0.0271* 0.0039
MA EC -0.0098 -0.0221 0 -0.015 0.0119 0
FI LI 0.0276*** 0.131*** 0.0224*** 0.0094** 0.0955*** 0.0061*
GEF -0.0434*** -0.184*** 0 -0.0218** -0.173*** 0
MA LI 0.0497*** 0.138*** 0.0677*** 0.0337*** 0.103** 0.0306***
MA LF -0.0508*** -0.169*** -0.0769*** -0.036*** -0.165*** -0.0372***
Observations 2829 324 2505 2832 324 2508
R20.422 0.402 0.418 0.167 0.192 0.152
24 / 50
Random E¤ects: Performance
IWorking in a …nancial institution (FI) improves the expert’s
forecasting ability more before the GFC than after.
IHaving graduated with a PhD from a top university (RF)
appears to improve expert forecasts before the GFC.
IAfter the GFC and over the full sample, having a PhD from a
top university (RE) seems to improve expert forecasts.
IBefore the GFC, male experts appeared to provide better
forecasts than females (not true after the GFC/full sample).
IThe fact that male experts were more numerically dominant
prior to the GFC, mitigates this result.
IWomen bene…t less than men from connections in job
performance, herding behavior, and subjective evaluation by
others (Fang and Huang, 2017).
IWomen who graduated in economics have better performances
than men who graduated in economics.
25 / 50
Random E¤ects: Boldness
IWorking at a FI decreased the pre-GFC expert’s boldness.
IComing from a top-ranked institution in economics increased
boldness before the GFC.
IFemale experts with a degree in economics herd more than
male experts before the GFC.
IHaving graduated with a Master’s degree from a top university
in …nance reduced performance and boldness vs. econ.
26 / 50
Random E¤ects: Contribution
IPerformance and boldness results are often similar (Clarke and
Subramanian, 2006), con…rming that signi…cant
underperformers are more likely to issue bolder forecasts and
vice-versa.
ILike …nancial analysts who also tend to exhibit herding
behavior, which sometimes compromises accuracy, our results
suggest that social forces (ranking, institution type, and
location), education (type and level), and experience (type and
duration) in‡uence an expert’s rational economic logic and
cognitive biases (Christo¤ersen and Stæhr, 2019).
IThe di¤erence between pre- and post-GFC may re‡ect the
change in the expert’s attention or biases induced by the crisis
shock on characteristics’e¤ects (Andrade and Le Bihan, 2013;
Christo¤ersen and Stæhr, 2019).
27 / 50
Fixed E¤ects
ITime …xed e¤ects included in all our panel regressions to
control for the possibility of time-varying uncertainty in our
sample period.
IGiven that there is some competition between forecasting
institutions that should erode persistent quality di¤erentials,
this model has some plausibility in particular for forecasts
provided by private institutions.
IIt allows to test for the impact of total job experience, or
experience in a speci…c profession, measured in months–rather
than the number of forecasts provided.
ISince we do not include experience …xed e¤ects, we have to
control the initial experience obtained from the CV data.
IWe add a range of nontime-varying, expert and
institution-speci…c, controls.
28 / 50
Time and Forecaster Fixed E¤ects
IExperience is the only time-varying trait we consider =)
simple univariate panel regression with time (t) and forecaster
(i) speci…c e¤ects
yit =β0ln xit +ui+vt+εit ,(12)
where yit is one of our two loss functions (performance and
boldness) and xit is one of our experience measures.
IPotential endogeneity issues: the ability to gain experience
might depend on forecast performance.
IWhile the evidence for the existence of such an e¤ect is mixed
or moderately sized, we correct for it in a robustness test by
including a dummy for ‡agging the last …ve forecasts
submitted by any forecaster.
IFollows the spirit of selection estimators (Heckman, 1979):
yit =β0ln xit +β11lt2last5(i)+ui+vt+εit ,(13)
where last5(i)is the set of the last …ve periods in which
forecaster isubmits a forecast to our dataset. 29 / 50
Time and Forecaster Fixed E¤ects
Performance Boldness
Full sample Pre-GFC Post-GFC Full sample Pre-GFC Post-GFC
Experience -0.176*** -2.29 -0.271*** -0.137*** -1.027 -0.127***
Age >60 years old -0.03*** 0 -0.032*** -0.009 0 -0.005
last50.002 -0.042 -0.005 0.001 -0.011 -0.004
Two-way interactions with experience
Local forecaster (LF) 3.428*** -1.059 0.281* 1.512*** -0.27 0.188***
Local institution (LI) -4.409*** -1.634 0.105 -2.009*** -1.309 -0.101
Financial institution (FI) 1.409*** -0.279 0.085 0.662*** 0.381 0.09
Experience: Central bank (CB) 0.816*** 0.365 0.814*** 0.326*** 0.061 0.334***
Experience: Academia (AC) 2.351*** -0.006 -1.359*** 0.944*** -0.157 -0.6***
Experience: High In‡ation (HI) -0.018 -1.949** 0.034 -0.046** -1.121 -0.006
Gender (G) -0.145 -0.625 -0.007 -0.067 -0.76 0.063
Education: MA (MA) -2.681*** -0.438 -3.147*** -1.327*** -0.195 -1.462***
Education: PhD (PhD) -3.3*** 0 -3.29*** -1.552*** 0 -1.443***
Education: Economics (EC) 0.738*** -1.473* -0.859*** 0.36*** -0.815 -0.331**
Education: Finance (EF) -3.467*** 4.805 0.462 -1.582*** 7.649*** 0.193
Ranking: Economics (RE) 1.098*** -0.203 0.467*** 0.504*** -0.131 0.221***
Ranking: Finance (RF) -0.041*** 0.366 -0.028 0.004 -0.144 0.008
30 / 50
Time and Forecaster Fixed E¤ects - Interactions
Performance Boldness
Full sample Pre-GFC Post-GFC Full sample Pre-GFC Post-GFC
Three-way interactions with experience
FI RE -0.016 1.19 -0.003 0.013 0.46 0.017
MA RE -7.103*** -0.33 -6.857*** -3.449*** -0.296 -3.151***
PhD RE -7.117*** 0 -6.927*** -3.462*** 0 -3.201***
EC RE -3.573*** -0.727 0.262 0.544*** -0.466 0.067
FI RF -0.062 -1.275 -0.013 -0.027 -0.593 -0.028
GRF -0.162*** 0.427 -0.041 -0.111*** 0.492 0.02
MA RF -0.043** 0.083 -0.093** -0.031*** 0.089 -0.054**
EF RF 0.007 -3.304 -0.098 0.154*** -6.123*** -0.034
AC LI -15.052*** 2.069 7.791*** -6.39*** 1.104 3.248***
MA LI 21.214*** -4.6 20.684*** 10.349*** -3.378 9.555***
PhD LI 21.361*** 0 20.852*** 10.459*** 0 9.598***
FI LF -6.89*** 0 -6.908*** -3.278*** 0 -3.156***
GLF 6.441*** 4.387 6.725*** 2.776*** 2.463 2.956***
AC LF 14.892*** 0 -7.965*** 6.347*** 0 -3.207***
MA LF 0.023 2.564 0 0.06** 2.541 0
EC LF 7.453*** 0 -0.48* 3.392*** 0 -0.014
GFI -15.893*** 0 -15.794*** -7.647*** 0 -7.248***
EC FI -0.279 0 7.323*** 0.167 0 3.21***
GCB 7.149*** 0 7.017*** 3.322*** 0 3.019***
EC AC 3.749*** 00 1.641*** 00
EC MA -3.367*** 00 -1.357*** 00
Observations 2829 324 2505 2832 324 2508
R20.052 0.231 0.041 0.057 0.146 0.035
31 / 50
Fixed E¤ects: Results
IDecrease performance after the GFC or over the full sample:
being a local forecaster, working in a …nancial institution, or
having experience at a CB or in academia.
IIncrease performance: experience, working in a local
institution, or being educated in …nance after the GFC or over
the full sample.
ISimilar to analysts’…rm experience, which is strongly and
positively associated with analysts forecast boldness (Clarke
and Subramanian, 2006; Huang et al., 2017): we complement
these …ndings for CPI experts by di¤erentiating the experience
type, size, and somehow the reputation accumulated through
their education.
IHaving a Master’s degree or a PhD from a top university in
economics and …nance rankings (MA RE, PhD RE, and
MA RF), working in a FI and being a local forecaster (FI
LF), or having experience in academia and working in a local
institution (AC LI), improve forecasting performance.
32 / 50
Fixed E¤ects: No Interactions
IPutting aside interactions, performance and boldness results
are similar except for several interesting instances:
IHaving a degree in …nance led to greater boldness before the
GFC but to more herding behaviors after the GFC and over the
full sample.
IAs a result of cognitive biases and an intuitive reaction to
uncertainty and …nancial instability, experts with lower risk
tolerance may herd more (Christo¤ersen and Stæhr, 2019).
IBeing a local forecaster seems to increase performance and
boldness behaviors, complementing Clarke and Subramanian
(2006).
IA notable di¤erence between herding and performance results
is that high in‡ation periods experienced by experts seem to
decrease boldness without in‡uencing performance, which
contradicts Clarke and Subramanian (2006).
IHigh in‡ation experience may contribute to a speci…c fear to
remain far from the consensus.
33 / 50
Fixed E¤ects: Interactions
IHaving a degree in economics and having a Master’s degree
a¤ect forecasting performance and herding behaviors
di¤erently.
IWe demonstrate that:
IThe likelihood of boldness increases with the expert’s
forecasting performance and experience, and is in‡uenced by
institution (Clement and Tse, 2005).
IThat experience (Hong et al., 2000; Mikhail et al., 2003) and
education (De Franco and Zhou, 2009) in‡uence social
interactions, cognitive biases, and intuitive reaction to
uncertainty, an interpretation partially shared with
Christo¤ersen and Stæhr (2019).
34 / 50
Interpretation: Experience
IForecaster characteristics in‡uence performance and boldness,
in line with the literature about …nancial analysts (Hong et
al., 2000; Clement and Tse, 2003).
IThe relationship between performance, boldness, and the
educational level and …eld aligns with the recent literature
linking cognitive sciences to …nancial analysts’background
and behaviors (Shapiro, 2006; Poore et al., 2014).
IForecasts that deviate widely from the consensus— observable
by the forecaster— potentially carry career-related rewards and
reputational risks.
ILow-skilled …nancial analysts exhibit larger increases in
deviations from consensus than high-skilled analysts (Evgeniou
et al., 2013).
IThe high-skilled experts feel more con…dent with a high level,
or quality, of education than others, changing their professional
behaviors and outcomes.
35 / 50
Interpretation: Behaviors
IThe institutional type and education explain why …nancial
institutions and mutual fund managers, who lag behind their
competitors in a given period, increased the riskiness of their
portfolios in the next period in attempts to outperform
competitors (Chevalier and Ellison, 1999).
IDenrell and Fang (2010): non-performing …nancial analysts
make more extreme predictions and thus are overrepresented
ex-post among those who can see the next big thing.
IClear demonstration of the Skin in the Game phenomenon
(Taleb, 2018): even when all forecasters have identical
information and incentives, their e¤orts to maximize their
expected wages will lead many of them to anti-herding,
possibly biasing their projections to di¤erentiate their views
from the consensus.
IForecasters working in industries that o¤er the greatest relative
reward for publicity will make courageous predictions,
especially after the GFC (Laster et al., 1999).
36 / 50
Forecast Spread: In‡ation and De‡ation
IUnlike low in‡ation, de‡ation is almost universally considered
highly problematic in economics, allowing us to use those
periods to assess the question of optimism vs. pessimism.
-3,0
-2,0
-1,0
0,0
1,0
2,0
3,0
4,0
5,0
6,0
7,0
%
Note: The gray area represents the spread of forecasts (disagreement). The
blue shaded background highlights the situations where both de‡ation and
in‡ation were considered possible by forecasters.
37 / 50
Pessimism: Theory
IFor those subsamples we estimate a panel probit model
explaining the probablity that the forecast Eit (πit)would be
below zero, taking the form:
p(Eit (πit)<0)=Φ(ψ0ln xit +XiΨ),(14)
where Φis the cumulative distribution function of the
standard normal distribution, and Ψis a vector of regression
coe¢ cients.
IWe then split this sample even further into the subsamples
when de‡ation was observed (i.e., when "pessimism" was
justi…ed, or–in other terms–the lack of de‡ation expectation
was overly optimistic) vs. periods where no de‡ation was
observed, i.e., when expecting in‡ation can genuinely be seen
as overly pessimistic.
IProbit panel regressions for a general situation (i.e., during
both in‡ationary and de‡ationary periods) and during only
de‡ationary and in‡ationary periods.
38 / 50
Pessimism: Probit Estimates
All De‡ation
Estimates Marginal E¤ects Estimates Marginal E¤ects
Experience -0.886*** -0.32 -0.933*** -0.37
Experience: Central bank (CB) -0.642** -0.23 -0.61* -0.24
Experience: Academia (AC) 0.613** 0.22 0.884*** 0.35
Education: Economics (EC) -0.858** -0.31 -0.697 -0.28
Ranking: Economics (RE) -2.083 -0.02 -1.444 -0.02
Experience: High In‡ation (HI) 0.938** 0.34 1.135*** 0.45
Observations 359 257
Note: ***, **, and * indicate signi…cance at the 1%, 5%, and 10% levels,
respectively. All stands for both In‡ation and De‡ation periods, and De‡ation
stands for only de‡ation periods. Both samples (All ) are restricted to periods
where there was disagreement, i.e., some experts predicted de‡ation while
others did not. The second sample (De‡ation) is a subsample of All, where the
pessimists turned out to be correct.
39 / 50
Pessimism: Interpretation
IExperts with more experience or with CB experience are less
likely to predict de‡ation.
IExperts with CB experience are less pessimistic, but this is
mitigated when pessimism turns out to be justi…ed.
IExperts with CB experience are optimistic, but people with
high in‡ation experience are pessimistic.
IThis is mostly driven by actual de‡ation periods, experts with
CB experience do not (want to) see de‡ation coming, but
people with any high in‡ation experience do.
40 / 50
Over- under-reaction: Theory
IWe augment a standard Mincer and Zarnowitz (1969)
regression to test for under- and over-reaction for di¤erent
types of forecasters rather than just looking at the sample as
a whole (Barberis et al., 1998; Daniel et al., 1998).
IWe estimate
Ei
t1[πt]πt=α0+ α1+
k
∑
n=1
αn+1xn,it !πt+ui+εt,
(15)
where xn,it is the nth characteristic of forecaster iat time t.
IWe account for unobserved heterogeneity through random
e¤ects (ui).
41 / 50
Over- and Under-reaction: Random E¤ects
Forecast Error
CPI -0.0356**
Experience (E) 0.0001*
Experience: Central bank (CB) 0.0079**
Age >60 years old -0.0007**
Constant 0.0093*
Observations 2904
Note: ***, **, and * indicate signi…cance at the 1%, 5%, and 10% levels,
respectively. Unlike our previous exercises, these results are obtained from non
demeaned data, and the age variable used here is the age of the exp ert rather
than its dummy. A Hausman test justi…es the use of a random e¤ects model.
42 / 50
Over- and Under-reaction: Interpretation
IExperts underreact on average.
IExperts with CB experience or more (accumulated)
experience in forecasting tend to underreact less than the
average.
IOlder experts (over 60 years old) underreact more than the
average.
IExperts with CB experience understand in‡ation better than
the other experts on average, and interpret private information
better than others. This characteristic seems more robust and
impactful than accumulating experience in forecasting.
IThe misinterpretation of genuine new private information
a¤ects forecasting performance (Daniel et al., 1998):
overcon…dence about private information =)stronger return
predictability in …rms with the greatest information
asymmetries =)whether overcon…dence of investors and
traders can be identi…ed with speci…c characteristics.
43 / 50
Forecasting Ability: Tests
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
-2
0
2Local forecaster
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
-2
0
2Local institution
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
-2
0
2Financial institution
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
-2
0
2Experience: central bank
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
-5
0
5Education: MA
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
-5
0
5Education: Ph.D.
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
-2
0
2Gender
Note: The red dashed lines represent the critical values of the predictive ability
test.
44 / 50
Forecasting Ability: Results
IThe test rejects only for education level characteristics (MA
and PhD): due to the low number of experts holding neither
an MA nor a PhD, those two tests capture almost the same
information.
IThus, it is unsurprising that the rejection is driven by the
same period in the mid-2000s when PhD educated experts
underperformed the competition (mostly MAs), and MAs
outperformed their competition (mostly PhDs).
IThe test does not reject for the remaining characteristics:
experts’characteristics may have a signi…cant role in their
out-of-sample forecasting outcomes even in unstable
environments.
45 / 50
Forecasting Ability: Interpretation
IThese …ndings may have implications for policymakers.
IThey may select forecasters according to their characteristics
to maximize forecasting precision and optimize their decisions.
IAlternative to conventional forecast combination
methods in the literature: while the bias-adjusted combination
method is found to work well in practice (Capistrán and
Timmermann, 2009, JBES), we demonstrate that a
characteristics-based forecast combination is potentially more
desirable than equal-weighted or bias-adjusted forecast
combination methods.
IIn‡ation forecasts from the SPF are biased, present positive
serial correlation in forecast errors, cross-sectional dispersion,
and predictability patterns depending on in‡ation variance.
IAs we control for time …xed-e¤ects, we interpret experts’
shifts in performance not explained by asymmetric loss and
rational expectations (Capistrán and Timmermann, 2009, JMCB)
through their characteristics rather than in‡ation variance.
46 / 50
Conclusion (1)
IIn line with our theoretical model, our original databases
about expert characteristics and outcomes over the two last
decades allow us to assess which expert and institution
characteristics matter in their performance, boldness,
forecasting ability, and sentiment (optimism or pessimism).
IThe education of the forecaster (quality, level, or …eld), the
environment (type of institution or localization), or the
experience (type, duration, high in‡ation) are essential
factors in‡uencing expert forecasting.
IUnderperforming experts are more likely to no longer be part
of our expert database, while boldness does not signi…cantly
in‡uence the experts’survival rate.
IRelated to career concerns and institutional labor market
expectations, degrees in …nance or economics do not o¤er the
same protection, while having graduated from a top
university decreases the expert’s survival rate.
47 / 50
Conclusion (2)
IThe GFC changed both …nancial institutions and the expert’s
labor market.
IAfter the GFC, expert’s experience, location, institution type,
or education …eld or quality change their forecasting
performance.
IBefore the GFC, only the education …eld and quality matter
for forecasting performance and boldness under …xed e¤ects.
More characteristics play a signi…cant role under random
e¤ects.
IThe expert’s location, institution location and type, and
experience type a¤ect the expert’s forecasting ability.
IThe expert’s past experience or previous experience in a
CB signi…cantly in‡uence the expert’s sentiment.
48 / 50
Policy implications
IWe interpret our results as evidence of a characteristics
e¤ect in in‡ation expert outcomes.
IExperts’characteristics and experience matter for
policymakers as long as expert forecasts are considered in
their decision-making process.
IDecisionmakers should select the right subsample of experts
according to their characteristics (use and compare with
median and average).
IInstitutions can also use our results for their management
and human resource decision processes.
IOur study allows these institutions to select the best
candidates among their characteristics for a targeted
profession, in‡ation forecaster.
IThe localization and environment of these (…nancial)
institutions is essential for their information processing, future
outcomes and reputation.
49 / 50
Thanks
IThank you for your attention.
IPaper forthcoming @ JEBO.
IComments: jonathan@benchimol.name
IOther papers: JonathanBenchimol.com/Research
50 / 50