Ijcb 23 Q 2 A 3

Central Bank Credibility and Monetary Policy∗
Kwangyong Park
Bank of Korea
In this paper, a numerical measure of central bank credibil-

ity is proposed that can be incorporated into a New Keynesian
model under bounded rationality. This measure arises due to
the existence of the changes in private beliefs, which are differ-
ent from those of the central bank. It is shown that central bank
credibility matters for macroeconomic stability. Specifically, as
credibility increases, macroeconomic variables vary less since
private agents’ expectations are more anchored. Through this
channel, the model generates endogenous volatility changes.
Finally, the credibility of the Federal Reserve and the European
Central Bank are computed based on the proposed method.
JEL Codes: E3, E52, E58, D8.
1. Introduction
Over the past 20 years, many central banks in advanced countries

have introduced inflation targeting as their way of conducting mon-
etary policy. By introducing inflation targeting, the notion of central
bank credibility became more important for policymakers. Despite
the importance of credibility, there has been little research quantify-
ing time-varying central bank credibility that is determined endoge-
nously by the central bank’s actions and economic outcomes in a
∗
This paper is a revised version of the first chapter of my Ph.D. dissertation
at Indiana University. I thank Boragan Aruoba, the co-editor, and two anony-
mous referees for their valuable comments and suggestions that substantially
improved the article. I am also grateful to Cosmin Ilut, Jinill Kim, Boreum Kwak,
Eric Leeper, Jongho Park, Sungho Park, Bruce Preston, Todd Walker, and semi-
nar and conference participants at Indiana University, Bank of Korea, Chonnam
National University, and Spring 2017 Midwest Macro Meeting for helpful com-
ments, suggestions, and discussions. All errors are mine. The views expressed in
this paper are solely those of the author, and do not necessarily reflect those of
the Bank of Korea. Author contact: Economic Research Institute, Bank of Korea,
55 Namdaemun-ro, Jung-gu, Seoul, Republic of Korea. E-mail: [email protected].
145
146 International Journal of Central Banking June 2023
New Keynesian framework, which has been a workhorse for central

banking studies in recent decades. In stylized New Keynesian mod-
els, it is implicitly assumed that the central bank is fully credible, in
the sense that private agents believe that the central bank follows
predetermined rules and tries to achieve the policy target, even if
the rules and/or target are unknown to the public. Therefore, the
results derived from previous studies might be too optimistic and
overestimate the efficiency of monetary policy.1
In this paper, a numerical measure of central bank credibility is
proposed using a version of a New Keynesian model with imperfect
knowledge. Then, implications of credibility on macroeconomic sta-
bility are analyzed. The proposed credibility measure is defined as
the tendency of private agents to rely on the central bank’s forecast
announcements in forming expectations of future endogenous vari-
ables, and is determined by the relative performance of the central
bank’s forecasts compared with those of private agents. That is, we
consider central bank communications, especially announcements of
forecasts, as important sources for shaping credibility. As private
agents cannot observe internal procedures of the central bank, such
as the actual targets and policy rules, the central bank’s announce-
ments are the sole sources for measuring its credibility.
The formulation of credibility proposed here reflects the views of
policymakers. For instance, Svensson interpreted a large gap between
the Riksbank’s announced repo rate path and market expectations
as being low credibility.2 Yellen (2006) also described credible mon-
etary policy as a situation in which “market participants correctly
anticipate the actions that the Fed will make in response to eco-
nomic news and shocks.” In addition, Blinder (2018) pointed out
that ignoring the messages that were sent by the central bank is
almost the same thing as not believing the central bank.
In this paper, we find that perpetual assessments of central bank
credibility can generate endogenous changes in volatility of endoge-
nous variables, such as inflation and the output gap. Specifically,
the baseline model that is estimated based on the U.S. economy
1
Goodfriend and King (2015) also pointed this out: “Forecasts, and policy,
should not be based solely on forecasts from a model that assumes full credibility
in the stated policy path.” See Goodfriend and King (2015) for details.
2
See Goodfriend and King (2015).
Vol. 19 No. 2 Central Bank Credibility and Monetary Policy 147
generates a 33 percent higher standard deviation of inflation com-

pared with the associated rational expectations model. In addition,
we also show that volatility of endogenous variables increases as
credibility deteriorates. Specifically, comparing volatility of infla-
tion between periods with the credibility level above and below the
median, inflation varies 81 percent more when the credibility level
is below the median. As the central bank builds a credible reputa-
tion, private expectations become more anchored around the zero
steady state, which is also the target of the central bank, and this
helps to stabilize the economy by preventing vicious feedback cycles
produced by the self-referential effects of expectations formation.
This paper contributes to the monetary policy credibility liter-
ature. Various researchers have provided different definitions of the
credibility of central banks. One of the widely used definitions is
related to imperfect information and/or knowledge of private agents
regarding monetary policy (Cukierman and Meltzer 1986; Ball 1994,
1995; Erceg and Levin 2003; Schaumburg and Tambalotti 2007).
This stream of research describes imperfect credibility as a status
wherein private agents cannot directly observe the objectives of the
central bank, such as the inflation target, or as the possibility that
the central bank may renege on a pre-announced policy path. Our
research differs from those, as we allow endogenous changes in the
level of central bank credibility evaluated by private agents, based
on the central bank’s announcements and economic outcomes within
a single framework.
On the other hand, many previous studies in the literature have
measured the credibility of a central bank and/or analyzed how
credibility affects macroeconomic stability. These studies share the
same concept of credibility, which is either a deviation of (long-
term) inflation expectations from the central bank’s target rate or
a range between the realized inflation and the target rate (Johnson
1997, 1998; Bomfim and Rudebusch 2000; Demertzis, Marcellino,
and Viegi 2010). Ours differs from those in some aspects. First,
we explicitly specify the belief structure and build a microfounded
model that is consistent with the underlying belief. Second, the cred-
ibility measure they suggest is not bounded, so it is difficult to inter-
pret and compare. However, the credibility measure proposed in this
paper is bounded to between 0 and 1, so it is easier to interpret and
compare across countries and periods.
There have been studies that connect bounded rationality to

credibility issues similar to ours. Gibbs and Kulish (2017) study
disinflation using a structural model that is close to ours. The cred-
ibility measure therein represents the proportion of households that
form expectations rationally. Our approach differs from it in some
ways. First, the level of credibility is fixed in their model, making it
impossible to study the interaction between credibility and macro-
economic outcomes. In addition, they do not specify explicitly the
underlying belief structure that can result in the expectation forma-
tion used in their study. Hommes and Lustenhouwer (2019) examine
the stability of a heterogeneous expectations model under a bounded
rationality assumption. Although they incorporate an endogenous
credibility measure that is similar to ours, the expectation forma-
tions in their model are too restrictive: some stick to policy targets
and the others rely on past observations to forecast future variables.
Hence, their credibility measure and model cannot account for the
forward-looking behavior.
This paper is organized as follows. The microfoundations of the
model are explained in Section 2. Belief structures and the credibility
measure are proposed in Section 3. Section 4 estimates the model.
Section 5 depicts historical credibility of the Federal Reserve and
European Central Bank. Section 6 discusses implications of credibil-
ity on macroeconomic stability. Finally, Section 7 concludes.
2. Model
This section develops a variant of the canonical New Keynesian

model that shares microfoundations with many other models. Details
of the microfoundations can be found in Preston (2005). In this
model, near rationality is assumed, and households hold subjective
beliefs. The formation of expectations will be discussed in detail in
the subsequent section.
A continuum of households, i, on the unit interval maximize their
lifetime utility
∞

T −t cT (i)1−σ
Êti ψT β − χnT (i) , (1)
1−σ
T =t
where β ∈ (0, 1) is the discount factor, χ > 0 measures disutility of

labor, and σ is the relative risk aversion parameter, subject to the
following budget constraint:
1 + it−1
ct (i) + bt (i) ≤ bt−1 (i) + wt nt (i) + Γt (i) (2)
1 + πt
and the no-Ponzi condition

−1
−t 1 + it+j
lim Êti ΠTj=0 Bt (i) ≥ 0. (3)
T →∞ 1 + πt+j+1
The variables ct (i), nt (i), bt (i), it , πt , wt , and Γt (i) denote con-

sumption, labor supply, real bond holdings, nominal interest rate,
net inflation rate, real wage, and real dividends from firms. ψt is
the exogenous preference shifter. The operator Êti denotes private
agents’ subjective expectations based on information up to time t.
A continuum of monopolistically competitive firms maximize
profits,
∞

Êti αT −t Qt,T [pt (i)yT (i) − PT wT nT (i)] , (4)
T =t
subject to the linear production technology, yT (i) = nT (i), and the

demand function derived from the households’ problem
−θT
pt (i)
nT (i) = yT (i) = YT , (5)
PT
where α is the Calvo (1983) parameter and denotes the probability

of not being able to reset prices in subsequent periods, and pt (i)
and Pt are the prices charged by firm i and the aggregate price
level. In addition, yt (i) and Yt present the output produced by firm
i and the aggregate output, respectively. θt > 1 depicts the elastic-
ity of demand across differentiated goods and follows an exogenous
process. The stochastic discount factor Qt,T is given as
Pt Ytσ
Qt,T = β T −t . (6)
PT YTσ
In a symmetric equilibrium, private agents share the same sub-

jective beliefs; thus, aggregate subjective expectations are the same
as individual expectations, although private agents are not aware of
this. The log-linear approximation around the zero-inflation steady
state gives the following decision rules for consumption and price
streams:

1
∞
ĉt (i) = Êti β T −t (1 − β)ŵT +1 − ı̂T − π̂T +1 − β ψ̂T − ψ̂T +1
σ
T =t
(7)
∞

p̂t (i) = Êti (αβ)T −t [(1 − αβ)(ŵT + u∗T ) + αβ π̂T +1 ] , (8)
T =t
where the hatted variables are the log deviations from their steady
states, except p̂t (i) = ln(pt (i)/Pt ), ı̂t = ln[(1 + it )/(1 + ı̄)], and
u∗t = ln(θt /θ̄). In the remainder, the hat notations that denote the
log deviations from the steady states are dropped for simplicity, as
there is no confusion that arises from this notational simplification.
We define the output gap as xt = yt − ytn = σ −1 wt . That is,
the output gap is the difference between the actual output and the
natural output, which is the level of output in a flexible-price envi-
ronment. Aggregating and imposing market clearing conditions to
Equations (7) and (8) yields the following equations, which are coun-
terparts of the dynamic IS and Phillips curve in a canonical New
Keynesian model.3
∞
T −t 1
xt = Êt β (1 − β)xT +1 − (iT − πT +1 − rT )n
(9)
σ
T =t
∞

πt = Êt (αβ)T −t [κxT + (1 − α)βπT +1 + uT ] (10)
T =t
3
The decision rules are comparable with the other models with bounded
rationality, for instance, Gabaix (2020), which extends a canonical New Keynesian
model by incorporating the sparsity-based limited attention suggested in Gabaix
(2014, 2016), as the current consumption gap is determined by the streams of
future consumption gaps and real interest rates in both models. However, expec-
tation formation differs as Gabaix (2020) assumes a specific term structure of
attention allocations while we explicitly specify subjective beliefs held by private
agents.
The slope of the Phillips curve is given by κ = (1 − αβ)(1 − α)/α.

ut is the autoregressive cost-push shock due to variations in a firm’s
markup reflecting fluctuations in elasticity of demand θt . An exoge-
nous disturbance rtn = β(ψ̂ − Êt ψ̂t+1 ) arises due to changes in house-
hold preferences, and it can be interpreted as fluctuations in the
natural interest rate.4
The model is closed with a version of the Taylor rule that
describes the behavior of the central bank.
it = φπ πt + φx xt + mt (11)
The central bank reacts to inflation and the output gap. mt is the
monetary policy shock and follows AR(1) process.
3. Beliefs, Forecasts, and Credibility
In this research, we rely on a near-rationality assumption and a

learning mechanism, as private agents have imperfect knowledge
about how the central bank conducts its policy. In addition, as a
byproduct, this assumption sidesteps the technical problem that
arises in rational expectations models due to the presence of higher-
order beliefs. However, this departure from rational expectations
is minimized by assuming that credibility only matters in non-
fundamental drift terms in the perceived law of motion (PLM)
and that the other parts of the model share the same structure as
the rational expectation model without the credibility measure, as
shown in detail in the subsequent sections.
3.1 Belief Structures

Private agents have the prediction model expressed below:
zt = Hāt−1 + Ωst + e1t

(12)
āt = F āt−1 + νt ,
4
Equations (9) and (10) can be reduced to the ordinary Euler equation and
dynamic Phillips curve if subjective expectations are formed rationally. This can
be proven by leading the equation and applying the law of iterated expectations.
See Preston (2005) for details.
where zt = [xt πt it ] is a vector containing endogenous variables that

private agents need to predict and st = [rtn ut mt ] denotes a vector
that observable exogenous disturbances are stacked. e1t is the predic-
tion error of private agents’ prediction model. āt indicates a vector of
na unobserved time-varying terms, possibly random-walk drifts. Fol-
lowing Eusepi, Giannoni, and Preston (2015), it is assumed that the
number of the underlying driving forces of drift terms, na , is equal to
two and denote aπ and ax , respectively.5 These terms are labeled as
the nominal and real factor. The nominal factor reflects uncertainty
about the inflation target of the central bank, and the real factor
represents fundamental uncertainty about long-term technological
advance.6 This specification is supported by empirical analysis in
Eusepi, Giannoni, and Preston (2015). It is also intuitively plausible
based on the Fisher equation. Movements in long-term expectations
of the interest rate cannot be decoupled from those of inflation pro-
vided that the real interest rate does not deviate substantially from
its steady state. Under this assumption, the drift term attached to
the interest rate prediction equation can be expressed as λ1 ax +λ2 aπ .
Following Eusepi, Giannoni, and Preston (2015), we assume λ2 = 1
and interpret λ1 as the relative contribution of the real factor to the
nominal interest rate.
Exogenous disturbances follow the stationary AR(1) process
st = Φst−1 + εt , (13)
where εt are i.i.d. shocks. It is also assumed that parameters that

govern reactions to fundamental disturbances, Ω, are known to pri-
vate agents and coincide with their rational expectation counter-
parts.7 This assumption simplifies analysis and helps place greater
focus on the dynamics of long-term expectations, whereas it mini-
mizes deviation from the rational expectations model. Hence, private
agents only need to learn about the unobserved drift terms.
5
a denotes the estimate of ā, which is unobservable to private agents.
6
The real factor can be also interpreted as a shock to the higher-order belief,
or to private agent sentiments, à la Angeletos and La’O (2013). For example,
persistent positive waves of the real factor can be considered as strong optimism
for the real activities in the economy, as perceived by private agents.
7
This assumption does not change the qualitative results of this paper, since
learning procedures for constants and coefficients attached to structural distur-
bances are totally separated.
Based on the estimated time-varying drift terms at−1 up to time

t − 1 and the beliefs given above, private agents compute their own
forecasts as
ÊtP [zT ] = HF T −t at−1 + ΩΦT −t st , (14)
where ÊtP denotes the private forecast based on the information up

to time t. If time-varying drift terms ā are more persistent than
structural disturbances st , long-term forecasts are driven more by
time-varying drift terms than by disturbances. For example, if ā
have unit roots, then infinite-horizon long-term forecasts are simply
given by the current time-varying drift terms.
lim ÊtP [zT ] = Hat−1 (15)

T →∞
Following Eusepi, Giannoni, and Preston (2015), at is updated

using the steady-state Kalman filter recursion
at = F at−1 + K(zt − Hat−1 − ΩΦst−1 ), (16)
where the time-invariant Kalman gain matrix is
K = P H (HP H + R)−1 . (17)
P is given as E[(āt − at )(āt − at ) ] and R denotes private agents’

prior beliefs about the covariance matrix of the observation error
terms e1t in their prediction equation (12).8
The central bank is also near rational and has its own PLM,
which is used to predict future endogenous variables. The PLM of
the central bank is given as follows:
zt = Ωst + e2t , (18)
which coincides with the rational expectations solution to the model.

Here, e2t is the prediction error of the central bank’s prediction
model. Compared with that of private agents, the PLM of the
8
Since the Kalman gain is determined by the private agents’ prior beliefs and
is time invariant, the steady-state Kalman filter can be understood as a ver-
sion of the constant-gain learning process, which is widely used in the learning
literature—for instance, Eusepi and Preston (2011).
central bank clearly shows that the central bank’s forecasts are
well anchored, as there are no time-varying terms ā. This for-
mulation seems reasonable, as many central banks predict their
economy by using country-specific dynamic stochastic general equi-
librium (DSGE) models with rational expectations. In addition,
Mokhtarzadeh and Petersen (2021) show that central banks inter-
ested in maintaining inflation stability should communicate their
predictions solely based on rational expectations in their labora-
tory experiment. On the other side, the actual law of motion nests
that of the rational expectations under this assumption, as shown
later. This characterization sets a natural comparison of the model,
thereby facilitating interpretation of the results.
The central bank announces its own forecasts of endogenous
variables after observing the realization of disturbances
ÊtCB [zT ] = ΩΦT −t st , (19)
where ÊtCB denotes the forecast of the central bank based on infor-
mation up to time t.
The belief structure can be justified under the assumption that
no commitment device exists. This is a reasonable assumption, as
there is no credible commitment device. It is neither allowed nor
possible to inform the public credibly of the central bank’s internal
decisionmaking procedure, including objectives and rules. A notion
of credibility arises due to this point. Even if a central bank deliber-
ately conducts monetary policy and follows its target, private expec-
tations can diverge from the target because the central bank cannot
commit to the target and communicate its policy credibly. Private
agents continuously evaluate the central bank’s resolutions and abil-
ities to achieve the target based on the history of outcomes. The
formal definition of credibility used in this paper is presented below.
3.2 Credibility Measure

The tendency of private agents to rely on the central bank’s
announced forecasts in forming expectations of future endogenous
variables defines the credibility.
Definition 1 (Credibility). Central bank credibility ξt is the
time-varying relative weight attached to the announced forecasts
made by the central bank, other than private agents’ own forecasts,
when private agents derive ensemble forecasts and form expectations.
Specifically, private agents use the following subjective expectations:
Êt zT = ξt−1 ÊtCB [zT ] + (1 − ξt−1 )ÊtP [zT ]. (20)
As the proposed measure determines the relative weight among

forecasts, it can be determined naturally based on the relative accu-
racy of two forecasts. To be precise, the credibility measure, ξt , is
determined by the following dynamic predictor selection problem,
modified from that of Brock and Hommes (1997) and a possibly
sluggish law of motion.

P
exp δ 1 U
ξ˜t = 1 − t
× Dt , (21)
exp δ1 UtP + exp δ1 UtCB
where
∞

Utk =− ω j
zt−j − k
Êt−j−1 [zt−j ] W zt−j − k
Êt−j−1 [zt−j ]
j=0
(22)

Dt = 1 − exp −δ2 CB
Êt−1 [zt ] − Êt−1
P
[zt ] CB
Êt−1 [zt ] − Êt−1
P
[zt ]
(23)
for k ∈ {CB, P } and
ξt = ξt−1 + η(ξ˜t − ξt−1 ). (24)
U k is the fitness measure of central bank (k = CB) or private

(k = P ) prediction. The fitness measure is a discounted weighted
sum of the negative of the past squared prediction errors. ω ∈ [0, 1]
is the memory parameter that controls the degree of discounting of
past prediction errors in measuring the fitness. The fitness measure
becomes more persistent when ω approaches one. D ∈ [0, 1] meas-
ures the distance between two predictors and approaches zero as two
predictors become similar. W is a weighting matrix and δ1 and δ2
are the intensity of choice parameters. η ∈ [0, 1] controls inertia of
the credibility measure. As it approaches one, inertia does not exist

and the law of motion becomes ξt = ξ˜t . This sluggish process allows
slow adjustments, rather than drastic changes, in credibility.
The credibility measure is determined by two parts: the distance
between private and central bank forecasts (Dt ), and the relative
accuracy of these forecasts (U CB , U P ). Firstly, if the central bank’s
CB
forecasts are close to those of private agents (Êt−1 [zt ] − Êt−1
P
[zt ] →
0, Dt → 0), there is no reason for private agents to disregard the
forecasts announced by the central bank, because they believe that
the central bank shares their views on the economy. In that case,
private agents will perceive the central bank to be credible and
will make use of it (ξ˜t → 1).9 At the same time, if two predic-
tors differ (Dt → 1) and private forecasts are more accurate than
those of the central bank (U P > U CB ), private agents will doubt
the credibility of the announced policy targets and forecasts of the
central
bank, and will place
more
reliance
on their own forecasts
(exp δ1 UtP / exp δ1 UtP + exp δ1 UtCB → 1, ξ˜t → 0). Therefore,
poor central bank forecasts decrease a central bank’s credibility.
Kocherlakota (2011) stated the same point:
I’ve been emphasizing the importance of communication and

communication matters greatly. But, ultimately, the public’s
beliefs about the FOMC’s inflation objective will also depend
on inflationary outcomes. If annual inflation averages less than
1.5 percent for more than three or four years, onlookers will
begin to suspect that the FOMC’s true objective for inflation
is lower than its declared “two percent or a bit under.” Cor-
respondingly, if inflation is persistently higher than 2 percent,
then the public will begin to believe that the FOMC’s true
objective for inflation is higher than 2 percent. In either case,
inflation expectations could become unmoored, and the FOMC
could lose control of inflation itself. Communication can only
be effective if the FOMC also retains credibility.
9
Some policymakers acknowledged this point in their speeches. For example,
Bernanke (2004) argued that “The . . .way in which clear and open communication
enhances the effectiveness of monetary policy. . . is by helping to align financial-
market participants’ expectations about the future course of monetary policy
more closely with the policy committee’s own plans and projections.”
This is also in line with Plosser (2010)’s prescription of achieving

credibility.
One advantage of this formulation is that the aggregate economic
outcomes do not necessarily influence credibility directly. That is,
higher inflation realizations than the announced target do not nec-
essarily decrease credibility. If the economy is hit by a severe exoge-
nous shock, aggregate variables can deviate from the pre-announced
policy targets. Since private agents understand this possibility and
make the same mistakes in predicting macro variables, they take
a lenient stance in relation to these deviations. Put differently, this
case translates into larger losses in both U P and U CB , and credibility
ξt does not change much.10
Although this formulation seems similar to those of previous
studies, such as Bomfim and Rudebusch (2000) and Hommes and
Lustenhouwer (2019), it differs from them in several ways. First,
our formulation does not employ past realizations as predictors to
make expectations behave adaptively. In aforementioned research,
the credibility measure is simply the degree of forward-lookingness.
On the contrary, both predictors used in this paper are forward look-
ing. Second, it is more plausible to account for the central bank’s
real-time communications. In Hommes and Lustenhouwer (2019),
the credibility measure represents the proportion of households that
use policy targets as their future forecasts. However, as the interim
10
One might raise the question as to whether central bank credibility can be
independent of the central bank’s ability of keeping inflation stable if the central
bank’s staff become very good at predicting inflation and gross domestic product
(GDP), to a point where the private sector’s expectations totally rely on the cen-
tral bank’s forecasts. In this model, central bank credibility is not independent
of the central bank’s ability to keep inflation stable in the longer-run perspec-
tive. Definitely, credibility rises if the central bank’s predictions are good enough.
However, it is necessary to understand what makes the central bank’s predictions
better. In general, a central bank’s predictions become more accurate when non-
fundamental beliefs (at ) that affect the private predictions are suppressed. These
non-fundamental beliefs emerge when inflation and the output gap diverge from
the pre-announced targets continuously beyond the private agents’ tolerance.
While temporary deviations due to large shocks are tolerated, continuing devia-
tions may trigger divergent beliefs and drops in credibility. Therefore, deliberate
policymaking for stabilizing inflation and the output gap is required to achieve
high credibility and to make private agents rely more on the central bank’s pre-
dictions. For this reason, the proposed measure is related to a central bank’s
ability to stabilize the economy in the longer-run horizon.
targets may not coincide with pre-announced policy targets, it is

plausible that households would just use the central bank announce-
ments in predictions, as in this paper, rather than rely on the long-
run policy targets.
The proposed measure is also close to the one used by Gibbs
and Kulish (2017). In their paper, two forecasts formed by rational
expectations and adaptive learning are used to derive the ensemble
forecast, and credibility is described as the proportion of households
that have the rational expectations. Their measure is closely related
to ours, as past realizations matter as well through the learning
process but the weight placed on adaptive learning is not changing
over time endogenously.
Compared with the credibility measures used in the previous
research, the proposed measure possesses several advantages. First,
our credibility measure is ex ante bounded in the unit interval, mak-
ing it easier to interpret and compare. Second, what matters in
determining credibility is the relative accuracy of forecasts, not the
distance between either the actual or the predicted inflation and pre-
announced inflation target. Hence, the deviation of inflation from the
target, which arises due to large shocks, does not necessarily damage
credibility because the public understands that it is inevitable and
that their own predictors also perform badly.
3.3 Equilibrium
Combining expectations formation equations (14), (19), and (20)
with policy rule equations (9), (10), and (11), the actual laws of
motion (ALM) for this economy can be obtained.11
zt = C0 at−1 + Ωst (25)

C0 is the factor loading matrix, and this changes endogenously, as
the credibility measure ξt−1 is contained in this. The ALM and Equa-
tions (21), (22), and (24) governing the dynamics of the credibility
measure fully characterize the model.
One advantageous feature of the model is that we can recover
the ordinary rational expectations model if central bank credibility
11
See Appendix A for details.
is fixed at the maximum level in any of the cases. That is, when
the central bank credibility measure ξt is exogenously imposed to
be one, the ALM become identical to those in the rational expec-
tations model. Hence, the ordinary rational expectations model can
be considered as a special case, in which the central bank is fully
credible under any circumstances. The ALM, however, diverge away
from that of the rational expectations model, and the effect of the
subjective beliefs magnifies as the credibility measure moves away
from one.
4. Historical Credibility of Central Banks
The credibility measure suggested above can be readily computed

based on the observed data using Equations (20)–(24) if data exist
about the economic forecasts of private agents and the central bank.
To understand the credibility measure better, we construct the his-
torical credibility series of the Federal Reserve (the Fed) and the
European Central Bank (ECB). In this exercise, we only use the
one-quarter-ahead inflation forecasts for data compatibility.12 For
the Fed, we compute credibility for 1968:Q4 to 2014:Q4 using Green-
book and SPF forecasts.13 Similarly, credibility of the ECB is com-
puted by the European Survey of Professional Forecasters and the
ECB’s own announcements on economic projections.
Figure 1 shows the historical series of the Federal Reserve’s credi-
bility computed by the SPF and Greenbook forecasts and the names
of the Fed’s chairperson. It shows that the Fed’s credibility has fluc-
tuated mostly between 0.3 and 1, but retains relatively high levels.
While there have been substantial drops in credibility in 1972, from
1979 to 1980, in the mid-2000s, and during the global financial crisis,
mean credibility is higher than 0.85 for the entire period. The shaded
12
The choice of values of ω, δ1 , and δ2 are discussed in Section 5. In the esti-
mation, it is shown that considering only inflation forecast errors results in the
similar outcomes compared with the case that takes all three variables—output
gap, inflation, and interest rate—into account.
13
There are multiple numbers of Greenbook forecasts in a given quarter. In this
procedure, we choose the first Greenbook forecast in the quarter to comply with
the assumption that forecasts are announced at the beginning of the period in
the model. Lastly, the median forecast from the SPF is used for private forecasts.
Figure 1. Credibility Measure of the

Federal Reserve, 1970:Q1–2014:Q4
areas represent recessions identified by the National Bureau of Eco-

nomic Research (NBER). This figure shows that shifts in credibility
do not show any clear relationship with the business cycle.
Although it is hard to identify one specific factor that deter-
mines credibility, since that is determined collectively by various
macroeconomic outcomes, the computed credibility of the Fed hints
that major changes in the ways of conducting monetary policy are
particularly important. In the first few years of the 1970s, frequent
modifications in monetary policymaking took place as the techniques
required for setting and pursuing money targets were developed. The
figure also shows that the Fed’s credibility was at a low level at the
beginning of Volcker’s tenure, but the Fed built credibility slowly
over the 1980s. The reason behind this shift might be attributed to
a dramatic change in policy course, which attempted to tame per-
sistently high inflation. While credibility remained relatively high
until the early 2000s, it shows a persistent drop in the mid-2000s.
The Fed’s credibility shifts rapidly during the global financial cri-
sis period. Since the Fed introduced new policies sequentially dur-
ing this period, that might drag credibility downward, as will be
Figure 2. Credibility of the European Central Bank,

2005:Q1–2019:Q3 (left) and Credibility of the Fed
during the Global Financial Crisis (right)
presented in details below. Experiencing drops in credibility after

introducing new policies seems reasonable. As private beliefs are
persistent, realizing the expected effects takes time. Hence, credibil-
ity may decrease initially. As the expected effects emerge, the central
bank starts to gain credibility.
Lastly, we provide two examples to which our measure of cred-
ibility may apply. To do so, we expand the analysis to credibility
of the ECB and to the impacts of unconventional monetary policies
conducted during the global financial crisis in the United States.
The left panel in Figure 2 depicts the credibility of the ECB. One
notable feature is that the ECB’s credibility has been persistently
lower than that of the Fed. Specifically, the average credibility of the
ECB is more than 15 percent lower than that of the Fed. It provides
an example of the way in which the proposed measures can be used
to compare credibility across economies, and calls for deeper future
research on this matter.
The right panel provides the Fed’s credibility and major policy
events during the global financial crisis. The black and red arrows
indicate policy changes regarding quantitative easing and forward
guidance, respectively. It is difficult to identify the factors that drive
credibility, but we may find some stylized facts regarding the influ-
ence of monetary policy on credibility. For instance, as is postulated
above, major policy shifts precede drops in credibility. To go into
more depth, the figure shows that the Fed’s credibility tends to
decrease following the announcements regarding quantitative easing.
Interestingly, the impacts of forward guidance seem unclear. This

finding is also valid in the case of the ECB. The announcement that
the ECB began purchasing government bonds (quantitative easing,
or QE) preceded the largest drop in credibility.
If private beliefs are persistent and influential concerning eco-
nomic outcomes, we can provide an explanation for the above find-
ings. First, policies that only affect private beliefs or expectations,
such as forward guidance, may have small impacts in credibility. If
credibility is low, forward guidance may not be able to exert mean-
ingful influence, hence it would not be able to enhance credibility.
On the other hand, if credibility is high, private agents will change
their beliefs immediately following the projections announced by the
central bank. Therefore, credibility would not show drastic changes
in this case either. Second, it will take a long time to show intended
outcomes after major policy changes if economic outcomes are sub-
stantially affected by private beliefs, which evolve relatively slowly.
This may temporarily damage credibility, as the policy effects come
with a delay. We are going to show that these requirements seem to
be satisfied in the estimated model.
5. Estimation
As this paper focuses on quantitatively measuring the influence of

central bank credibility on the economy, it is necessary to estimate
the model to discipline the parameters. As the model is highly non-
linear, we estimate the model by implementing a bootstrap parti-
cle filter Metropolis-Hastings algorithm to evaluate the likelihood
functions and to derive posterior distributions of parameters.14
5.1 Data
We include six observables to estimate the model. For inflation and
short-term interest rate measures, the GDP deflator and the federal
funds rate (FFR) are used. For the output gap measure, the output
gap published by the Congressional Budget Office is used. We also
include a private forecasts series on the level of GDP, GDP deflator,
and FFR from the Federal Reserve Bank of Philadelphia’s Survey
14
See Herbst and Schorfheide (2015) for theoretical discussion on this method.
of Professional Forecasters (SPF) to match the model-generated pri-

vate forecasts. The sample period covers 1981:Q3 to 2007:Q4. Since
the model lacks the zero lower bound constraint, we only use data
from before the financial crisis.15
5.2 Calibrated Parameters

We calibrate some parameters before estimating the model. This set
of parameters includes the subjective time discount factor, β; two
intensity of choice parameters, δ1 and δ2 ; weights on past predic-
tion errors, ω; and the weighting matrix, W . The subjective time
discount factor β is set to 0.995. δ1 and δ2 are chosen to minimize
the distance between the credibility measure estimated in the model
and the one calculated directly from the private forecasts and the
Fed’s forecasts in Figure 1. As a result, δ1 = 1.62 and δ2 = 23 are
chosen.16 The parameter that governs weights on past prediction
errors, ω, is set to zero for tractability. While there is no memory
regarding the past loss due to prediction errors, the persistence of
credibility can be still captured by the sluggish evolution of credi-
bility measure which is controlled by η. The weighting matrix W is
set as the identity matrix. This means that private agents put the
same weight on the prediction errors on the output gap, inflation,
and interest rate.17
5.3 Method
We estimate the model by constructing a Metropolis-Hastings par-
ticle filter (MHPF). The proposal parameters are drawn from a
Markov chain repeatedly for 60,000 iterations. We discard the first
10,000 draws to remove any influence from the initial condition.
15
Extending the sample period beyond 2007 using the shadow rate proposed
by Wu and Xia (2016) does not change the estimates substantially.
16
As long as δ1 and δ2 are sufficiently large, the predictor selection problem
mimics the classical choice behavior. That is, the agents always choose the pre-
dictor that is more accurate, and the weight put on the other predictor becomes
negligible. Hence, the quantitative results do not change considerably even if
these parameters are changed. See Appendix D.
17
The case where private agents only consider prediction errors on inflation
results in similar outcomes. Specifically, the estimated credibility and private
beliefs are almost identical to the one from the baseline case. See Appendix D.
Table 1. Estimated Parameters
Posterior
Parameters Distribution Prior Mean S.D. Mode [0.05, 0.95]
α Beta 0.7 0.08 0.73 [0.72,0.73]

σ Gamma 2 0.5 1.77 [1.76,1.78]
φx Gamma 0.12 0.1 0.17 [0.17,0.17]
φπ Gamma 2 0.15 2.12 [2.12,2.13]
ρr Beta 0.7 0.12 0.73 [0.72,0.73]
ρu Beta 0.7 0.12 0.73 [0.73,0.74]
ρm Beta 0.7 0.12 0.79 [0.78,0.79]
λ1 Normal 0 0.5 –0.25 [–0.28,–0.25]
f1 Uniform 0 1 0.74 [0.73,0.75]
f2 Uniform 0 1 0.99 [0.99,0.99]
σεr IGamma 0.1 2 0.58 [0.58,0.58]
σεu IGamma 0.1 2 0.15 [0.14,0.16]
σεm IGamma 0.1 2 0.50 [0.49,0.51]
K11 Uniform –1 1 0.34 [0.34,0.36]
K12 Uniform –1 1 0.12 [0.12,0.12]
K13 Uniform –1 1 –0.28 [–0.30,–0.28]
K21 Uniform –1 1 –0.09 [–0.09,–0.08]
K22 Uniform –1 1 0.30 [0.29,0.31]
K23 Uniform –1 1 0.14 [0.13,0.15]
η Beta 0.7 0.12 0.66 [0.64,0.66]
Marginal Log- –1002.9
Likelihood
Note: The priors and posteriors, Para(1) and Para(2), correspond to the mean and
standard deviation of the Normal, Gamma, Inverse Gamma, and Beta distributions
and to the lower and upper bounds for the Uniform distribution.
Next, we collect one draw in every five draws to thin out the chain
to reduce the autocorrelation of the chain. Then we construct mar-
ginal posterior densities from the remaining 10,000 draws. The scal-
ing parameter is set to vary along the iteration so that it guarantees
achieving the acceptance rate in between 0.2 and 0.4. The resulting
total acceptance rate is around 0.28. Lastly, the number of particles
is set to 10,000.18
5.4 Estimation Outputs

Table 1 shows the prior and posterior distributions for the model
parameters. We choose diffuse priors for f1 , f2 , K, η, and λ1 , as
18
See Appendix B for details regarding the measurement and transition equa-
tions and iteration process.
Figure 3. Comparison between Estimated

and Directly Computed Credibility
they are model specific, and so there is no consensus on the choice

of these prior distributions. Other priors are quite general in the
literature. In the next section, we use the modes of posteriors when
simulating the model.
Some results are worth mentioning. The results for f1 and f2
show that the long-term inflation forecasts are highly persistent,
and not anchored, but that the perception regarding the long-term
business outlook is less persistent. In addition, the estimate of η
shows that the credibility measure is moderately persistent.
Next, we compare the estimated credibility and the directly com-
puted credibility derived in the previous section to validate the esti-
mation. Figure 3 presents the credibility series directly computed by
the forecast data (blue solid line) and the estimated credibility the
non-linear filter (red dashed line). While the estimated credibility
shows more sluggish movement, the two series move in a similar man-
ner. Numerically, the correlation coefficient between the two series
is 0.48. The correlation reaches 0.57 when the empirical credibility
series is smoothed by a moving average with a five-quarter window.
Finally, before analyzing the model, it is necessary to check
whether the model results in stable dynamics under the given
parameters. Appendix C analyzes the model stability and confirms

that the model is stable under the current calibration.
6. Quantitative Results
In this section, we examine the role of central bank credibility on

the overall macroeconomic stability by simulating the model. To this
end, the model is simulated 50,000 times, for 1,200 periods each, to
gauge the effects of endogenously evolving central bank credibility
on macroeconomic stability. The initial 1,000 periods are discarded
to remove any effects from initial conditions. The model parame-
ters are set to their posterior modes derived from the estimation
procedure.
6.1 Endogenous Volatility Changes

First, we compare the macroeconomic volatility obtained in the base-
line model to that computed in the rational expectations model
to examine whether endogenous central bank credibility can cre-
ate additional volatility in macroeconomic variables.19 To this end,
standard deviations in the output gap and inflation are computed
and compared with those from the rational expectations model.
The results are summarized in Table 2. This table contains the
mean standard deviations obtained from the simulations across mod-
els and subperiods. In general, introducing the credibility measure in
a New Keynesian model raises the volatility of both the output gap
and inflation. Specifically, the standard deviation of inflation is 33
percent higher in our model, whereas that of the output gap changes
by only 7 percent. We derive two observations from the above results.
First, introducing central bank credibility can endogenously gener-
ate additional volatility in the economy. Second, shifts in central
bank credibility can substantially affect the volatility of inflation,
but that of the output gap is affected much less.
Next, the standard deviations of the macro variables are calcu-
lated for different levels of credibility to analyze the relationship
19
The rational expectations model is evaluated at the same parameter values
with the exception of the belief structure.
Table 2. Volatility of Macro Variables
Baseline
RE Baseline RE
σ(x) 1.37 1.47 1.07

σ(xh ) 1.36 1.37 1.01
σ(xl ) 1.37 1.54 1.12
σ(xl )/σ(xh ) 1.01 1.12
σ(π) 0.61 0.81 1.33
σ(πh ) 0.54 0.54 1.00
σ(πl ) 0.66 0.98 1.48
σ(πl )/σ(πh ) 1.22 1.81
Note: The average standard deviations are obtained from the simulations. Subscripts
h indicate standard deviations computed from periods with credibility higher than
the median and l with lower than the median.
between credibility and macroeconomic volatility. To this end, sim-

ulated series are divided into two groups of subperiods. One contains
variables at periods when credibility is higher than the median, and
the other when credibility is lower than the median.20 For better
contrast of the results, we also report standard deviations derived
by simulating the same sets of shocks under the rational expecta-
tions model since there might be the possibility that higher volatility
is caused by larger shocks rather than the endogenous credibility
channel.21
Table 2 clearly shows that the variations of endogenous vari-
ables decrease in credibility. The standard deviations of the output
gap and inflation are 12 percent and 81 percent higher in the low-
credibility periods. It is appropriate to ask whether this difference is
fully caused endogenously by introducing credibility concerns in the
model, because some portions of the increments can be caused just
by selecting low-volatility realization periods as high-credibility peri-
ods. This concern can be resolved by comparing the σ(xl )/σ(xh ) and
σ(πl )/σ(πh ) ratios between the baseline and rational expectations
20
The median is around 0.63.
21
Since there is no change in credibility in the rational expectations (RE)
model, we report σ(xh ), σ(πh ), σ(xh ), and σ(πh ) of the RE model based on
the credibility level of the baseline model calculated from the same shocks.
models. As shown in Table 2, accounting for credibility of the central

bank significantly increases the ratio for inflation. The endogenous
evolution of credibility generates an additional 59 percent increment
in the standard deviation of inflation (81 percent minus 22 percent)
while that of the output gap shows 11 percent rise in volatility com-
pared with the rational expectations model. Thus, we can conclude
that an increase in the standard deviation of inflation and the output
gap is largely generated by introducing endogenous credibility into
the model. One thing worth noticing is that volatility is almost the
same across the models when credibility is in the upper half. This
is reasonable, as our model converges on the rational expectations
model as credibility enhances.22
Next, it is still not clear whether there is a monotonic nega-
tive relationship between the credibility and volatility of endoge-
nous variables. To answer this question, we divided the simulated
series into finer credibility bins, as in Figure 4. It clearly shows that
the standard deviations of the output gap and inflation increase
monotonically as credibility decreases. In particular, the variability
of inflation increases exponentially as credibility decreases. To be
precise, the increments in the volatility of inflation rise from 2 per-
cent to 55 percent as credibility decreases, while the increments in
the output gap increase only from 3 percent to 10 percent. This sug-
gests that most of the benefits of higher credibility accrue in the low
credibility region. This result is in line with the results discussed in
Schaumburg and Tambalotti (2007).
The quantitative results can be connected to the recent exper-
imental studies on central bank credibility (Ahrens, Lustenhouwer,
and Tettamanzi 2017; Mokhtarzadeh and Petersen 2021). In these
studies, central bank credibility is measured by the fraction of fore-
casters that have the same (or close) projected value as the one
announced by the central bank. As predicted in this paper, it is
22
It is well known that introducing a learning procedure instead of rational
expectations assumption produces an additional layer of interaction between
economic outcomes and monetary policy and results in more volatile macroeco-
nomic dynamics compared with rational expectations models as shown above. For
instance, Orphanides and Williams (2004) find that policies that fail to maintain
control over inflation are vulnerable to episodes in which the public’s expecta-
tions become decoupled from the policy objectives under imperfect knowledge
environment using a perpetual learning model.
Figure 4. Standard Deviations of Output Gap

and Inflation across Credibility Quartiles
shown that credibility decreases when the central bank makes larger
prediction errors in these experiments. The experimental results can
explain the theoretical results provided in this paper reasonably well.
First, when credibility is higher, volatility is almost identical to that
which arises in the rational expectations model. Second, when cred-
ibility is lower, the standard deviations (or similarly, mean square
deviations from the targets) of the output and inflation increase.
Specially, volatility of the output increases by quantitatively small
amounts whereas that of inflation increases substantially. These
results are surprisingly in line with the theoretical predictions pro-
vided in this model.
Combining the above results with the estimated credibility of
the Fed, we can postulate that shifts in the Fed’s credibility is
one possible explanation for the Great Moderation, referring to the
period of low macroeconomic variability between the mid-1980s and
the global financial crisis. Since the Fed’s credibility during the
Great Moderation was higher than during other periods (Table 3),
additional volatility injected because of the lower credibility had
Table 3. Average Credibility of the Fed across the Periods
Period 1970–1984 1985–2007 2008–2014

Mean Credibility 0.83 0.91 0.86
been suppressed. This might have contributed to relatively stable

macroeconomic developments. For this reason, lower macroeconomic
volatility during the Great Moderation might be partially attributed
to an increase in the Fed’s credibility.
6.2 Underlying Mechanism: Feedback Effect of Private Beliefs

In this subsection, we analyze the underlying mechanism that cre-
ates additional volatility in the economy by examining the impulse
response of the model. Before proceeding with the results, it is nec-
essary to emphasize that the impulse response is not unique, and
depends on a realization of history due to the non-linearity of the
model. In particular, the level of credibility and the real and nomi-
nal factors that occur when a shock hits the economy are important
determinants of the impulse response. For this reason, we derive
impulse response functions with different initial conditions for two
factors, ax and aπ , and for credibility ξ.23 Since two initial val-
ues for subjective factors are required to generate impulse response
functions, we assume a−1 = a−2 for simplicity. Finally, it is also
noteworthy that impulse response functions obtained in case a = 0
and ξ = 1 is almost identical to those in the rational expectations
model.
Impulse responses are calculated by differencing simulated series
with and without a specific temporary shock. Finally, we present
impulse responses to the preference shock in the main text, and
those to the cost-push and monetary policy shock are delegated to
Appendix E, as the underlying mechanism is the same.
In Figure 5, it is assumed that the private agents perceive
that the long-term inflation is higher than the zero inflation
23
Alternatively, we may derive generalized impulse responses as in much of the
empirical literature. However, we do not follow this strategy because our method
helps to understand better the transmission mechanisms of the model.
Figure 5. Impulse Responses to a One-Standard-Deviation

Preference Shock When aπ Is Positive
Note: The blue solid line represents impulse responses from ξ−1 = 1, aπ−1 = 0.01
and the red dashed line represents impulse responses from ξ−1 = 0.4, aπ−1 = 0.9.
target (aπ−1 > 0) and the exact values differ across initial credi-
bility. To be exact, aπ−1 is assumed to be 0.01 and 0.9 for the initial
credibility level 1 (high credibility), and 0.4 (low credibility) cases.24
When ξ−1 = 1, subjective factors ax and aπ do not affect the
output gap and inflation, even if these factors have non-zero val-
ues. Therefore, their effects on endogenous variables are limited in
subsequent periods. For this reason, impulse responses are similar
to those of the rational expectations model when initial credibility
is equal to one: the output gap, inflation, and interest rate increase
and return to the steady state monotonically.
24
We choose these values for realistic model simulations based on the estimated
nominal factor for each credibility level for the U.S. economy.
As credibility deteriorates, however, the model dynamics change

considerably. When aπ is positive, a positive preference shock
reduces credibility, as a positive aπ makes private forecasts rela-
tively more accurate compared with the central bank’s forecasts.
Specifically, private forecasts become more accurate as credibility
deteriorates since private agents anticipate higher inflation than the
central bank at time zero due to the positive perception on the nom-
inal factor. Thus, credibility drops more than 0.2 points immediately
in the low credibility case and remains at the lower level for longer
periods.
Similarly, the nominal factor is revised upward and returns to
zero slowly in all cases, as realized inflation is higher than expected.
This shift in the nominal factor again feeds back to even higher infla-
tion and this feedback effect gets larger as credibility decreases since
the influence of the nominal factor becomes more substantial. As
credibility decreases, private expectations of future inflation increase
because of a higher aπ . This increases inflation instantly and feeds
into a higher nominal factor. This stimulates inflation again and
the self-referential cycle continues. For this reason, inflation even
overshoots in the low-credibility case.
Initially, the output gap increases sharply due to the posi-
tive preference shock. However, as credibility drops, private agents
expect higher nominal interest rates in future periods, and the cen-
tral bank actually increases the nominal interest rate to stabilize
higher inflation. Therefore, the output gap shows a more contracted
path compared with the case with a higher credibility.
While the initial real factor is zero, it is revised upward due to the
positive reaction of the output gap. Then, as private agents update
their perceptions on the real factor, it comes back to zero while the
deviation is larger in the low-credibility case.
6.3 Improving Credibility

Finally, in this subsection, we discuss how a central bank can
enhance its credibility. There are many things that can affect central
bank credibility. For instance, even a single word spoken by policy-
makers could affect central bank credibility. Hence, it is not an easy
task to discuss every possible option for improving credibility. For
Table 4. Monetary Policy Rules and Average Credibility
2.12
φπ 1.12 (–1) 1.62 (–0.5) (Benchmark) 2.62 (+0.5) 3.12 (+1)
Mean 89 95 100 101 102

Credibility
0.17
φx 0.07 (–0.1) (Benchmark) 0.27 (+0.1) 0.37 (+0.2)
Mean 100 100 101 102

Credibility
Note: The average credibility in the benchmark model is normalized to 100. Therefore,
the average credibility under different specifications can be interpreted as the percentage
changes compared to the benchmark case.
this reason, we only focus on the role of the central bank’s sys-
tematic reactions to economic developments in shaping credibility.
To this end, we analyze how the average credibility changes as we
vary the parameters governing the monetary policy rule specified in
Equation (11).
Table 4 presents how the average credibility changes as the mon-
etary policy reaction function varies. The upper panel shows the
changes in credibility when the central bank’s response to inflation
changes, while the lower panel presents the changes in credibility as
the reaction to the output gap changes. Compared with the bench-
mark case that represents the current policy practice, stronger reac-
tions to both inflation and the output gap result in a higher mean
credibility, though the increments are quite small. This is reason-
able since stronger responses to inflation and the output gap make
them easier to forecast by pushing them closer to their respective
targets.25
25
However, it is uncertain whether stronger responses to inflation and the out-
put gap are welfare improving, since a stronger reaction to inflation results in
a more volatile output gap, while a stronger response to the output gap leads
to more volatile inflation. This suggests that the optimal monetary policy may
depend on central bank credibility. Although we do not analyze the optimal mon-
etary policy under the credibility restriction since that is out of the scope of this
paper, we believe that it might be an interesting future research topic.
On the contrary, weaker responses to inflation sharply reduce

credibility, while weaker reactions to the output gap do not change
credibility substantially. This non-linear relationship between mon-
etary policy reactions and credibility suggests that the Fed is effi-
ciently conducting its policy so that it achieves high credibility
without creating excessively volatile macroeconomic responses to
monetary policy.
7. Conclusion
In this research, a numerical measure of central bank credibility is

proposed and its effects on macroeconomic stability are examined.
The main contributions of this paper are to show to what extent
accounting for credibility affects macroeconomic stability. Specif-
ically, it is shown that volatility in the output gap and inflation
increase as credibility deteriorates due to the self-referential effect
of private beliefs. This model can generate endogenous volatility
changes based on the shifts in credibility without relying on exoge-
nous volatility regime changes. This result theoretically confirms the
idea that maintaining a credible reputation helps to anchor private
expectations and to achieve macroeconomic stability. Despite their
importance, these results have not yet been fully analyzed in a New
Keynesian framework with endogenous central bank credibility con-
cerns, and this paper provides a useful benchmark that can be easily
analyzed.
The findings derived in this research have important implications
for many issues in monetary policy. For instance, the definition of
credibility used in this paper is related to the forward guidance, espe-
cially the Delphic effects of forward guidance, as studied in, among
others, Campbell et al. (2012, 2017). A central bank’s forecasts as
described in this model are closely related to forward guidance, as
they hint at a future course for the economy and communicate infor-
mation held by the central bank. Most analyses of forward guidance,
however, implicitly assume full credibility so that private agents
believe what the central bank says and focus on information flow
from the central bank to the public. If the private agents do not
think that the central bank is fully credible, the Delphic effects that
were described might disappear. Therefore, this research suggests
that it is necessary to take the credibility issue into account when

the effects of forward guidance are examined.26
Although central bank credibility has been considered an impor-
tant feature that shapes the efficacy of monetary policy, it has not
been sufficiently taken into account in quantitative monetary pol-
icy studies. Nonetheless, credibility matters in many cases, as doc-
umented in this paper, hence more serious research regarding this
issue is warranted.
Appendix A. Actual Laws of Motion
First, obtain a minimum state variable (MSV) solution for the

rational expectations model below by method of undetermined
coefficients.
xt = Et xt+1 − σ −1 (it − Et πt+1 − rtn )
πt = κxt + βEt πt+1 + ut (A.1)
it = φπ πt + φx xt + mt ,
A unique and bounded solution exists if κ(φπ − 1) + (1 − β)φx > 0

holds. Note that this condition is satisfied in a current calibration.
The MSV solution is given as
xt = a11 rtn + a12 ut + a13 mt

πt = a21 rtn + a22 ut + a23 mt
it = a31 rtn + a32 ut + a33 mt ,
where
1
a21 =
1−βρr
κ ((1 − ρr )σ + φx ) − ρr + φπ
1 − βρr
a11 = a21
κ
a31 = φπ a21 + φx a11
26
In their experiment paper, Ahrens, Lustenhouwer, and Tettamanzi (2017)
emphasize the importance of credibility in shaping the effectiveness of forward
guidance.
1 − ρu + φσx
a22 =
φπ −ρu
κ σ + 1−βρ
κ
u
1 − ρu + φx
σ
(1 − βρu ) 1
a12 = a22 −
κ κ
a32 = φπ a22 + φx a12
−1
a23 =
1−βρm
κ (σ(1 − ρm ) + φπ ) + φπ − ρm
1 − βρm
a13 = a23
κ
a33 = φπ a23 + φx a13 + 1.
Combining Equations (14), (19), and (20), obtain future expec-

tations on endogenous variables
Êt xT = (1 − ξt−1 )f1T −t axt−1 + a11 ρTr −t rtn + a12 ρTu −t ut + a13 ρTm−t mt
Êt πT = (1 − ξt−1 )f2T −t aπt−1 + a21 ρTr −t rtn + a22 ρTu −t ut + a23 ρTm−t mt
Êt iT = (1 − ξt−1 )(λ1 f1T −t axt−1 + λ2 f2T −t aπt−1 ) + a31 ρTr −t rtn
+ a32 ρTu −t ut + a33 ρTm−t mt .
Inserting these expectations into policy rules, Equations (9), (10),

and (11), gives the following system of equations:

it (1 − ξt−1 )f1 λ1 β
xt + = 1−β− axt−1
σ 1 − βf1 σ
(1 − ξt−1 )f2
+ (1 − λ2 β)aπt−1
σ(1 − βf2 )
σ(1 − β)ρr a11 + ρr a21 − βρr a31 + 1 n
+ rt
σ(1 − βρr )
σ(1 − β)ρu a12 + ρu a22 − βρu a32
+ ut
σ(1 − βρu )
σ(1 − β)ρm a13 + ρm a23 − βρm a33
+ mt
σ(1 − βρm )
κ(1 − ξt−1 )αβf1 x (1 − α)β(1 − ξt−1 )f2 π

πt − κxt = at−1 + at−1
1 − αβf1 1 − αβf2
ρr (καβa11 + (1 − α)βa21 ) n
+ rt
1 − αβρr
καβρu a12 + (1 − α)βρu a22 + 1
+ ut
1 − αβρu
ρm (καβa13 + (1 − α)βa23 )
+ mt
1 − αβρm
it − φx xt − φπ πt = mt . (A.2)
In matrix form, this system can be expressed as follows:

⎡ ⎤
⎡ ⎤ ⎡ ⎤ axt−1
1 0 1 ⎡ ⎤ d11 d12 d13 d14 d15 ⎢ π ⎥
σ xt
⎢ −κ 1 0⎥ ⎢d21 d22 d23 d24 d25 ⎥ ⎢at−1 ⎥
⎢ ⎥ ⎣π ⎦ = ⎢ ⎥ ⎢ rtn ⎥ .
⎣−φx −φπ 1⎦ t ⎣ 0 0 0 0 1 ⎦⎢⎣ ut ⎦
⎥
it
mt
(A.3)
By inverting the leading matrix in the left-hand side, we can obtain
the actual laws of motion,
⎡ ⎤
xt
⎣ πt ⎦ =
it
⎡ ⎤
⎡ σd11 −φπ d21 σd12 −φπ d22 ⎤ axt−1
h h a11 a12 a13 ⎢ π ⎥
⎢ κσd11 +(φx +σ)d21 κσd12 +(φx +σ)d22
a a a ⎥⎢at−1 ⎥
⎢ h h 21 22 23 ⎥⎢ rtn ⎥ ,
⎣ σ(φx +κφπ )d11 +φπ σd21 σ(φx +κφπ )d12 +φπ σd22 a ⎦⎢ ⎥
h h 31 a32 a33 ⎣ u ⎦
t
mt
(A.4)
where h = φx + κφπ + σ. Note that the actual laws of motion (ALM)

and future expectations are identical to those of the rational expec-
tations model without credibility when ξt−1 converges to one. To
see this, note that dij = 0 when ξt−1 = 1 for all i and j. Therefore,
ALM reduces to an MSV solution of underlying rational expectations
model as ξt−1 converges to one.
Appendix B. Estimation Procedure
The measurement equation is given as

⎡ ⎤
xt
⎢ πt ⎥
⎡ ⎤ ⎢ ⎥
⎢ it ⎥
xt ⎢ x ⎥
⎢ ⎥ ⎢ at ⎥

⎢ ⎥
πt
⎢ ⎥ ⎢ aπ ⎥
⎢ it ⎥ I3 O3×8 ⎢ t ⎥
⎢ P ⎥ ⎢ ξ ⎥ + ot ,
⎢Êt [xt+1 ]⎥ = O3×3 M3×8 ⎢ xt ⎥
(B.1)
⎢ ⎥ ⎢at−1 ⎥
⎣ÊtP [πt+1 ]⎦ ⎢ π ⎥
⎢at−1 ⎥
ÊtP [it+1 ] ⎢ rn ⎥
⎢ t ⎥
⎣ u ⎦
t
mt
where
⎡ ⎤
0 0 0 0 0 f1 0 ρr a11 ρu a12
ρm a13
M3×8 = ⎣0 0 0 0 0 0 f2 ρr a21 ρm a23 ⎦ .
ρu a22
0 0 0 0 0 λ1 f1 f2 ρr a31 ρu a32
ρm a33
(B.2)
The transition equations consist of laws of motion which are
derived in the main text and deterministic identity equations.
The estimation procedure can be summarized by the following:
(i) Draw V from the Markov chain (proposal parameter).
(ii) Generate particles of exogenous disturbances z.
(iii) For t = 1 : T ,
– Propagate state variables xt given particles zt and initial
states xt−1 .
– Evaluate likelihood functions p(yt | xt , zt , V).
– Resample the particles weighted by their likelihoods.
– Approximate the time t likelihood p̂(yt | V) weighted by

likelihoods of each particle.
(iv) Approximate the likelihood function p̂(y1:T | V).


1:T |V)p(V)
(v) With probability α(V | θi−1 ) = min 1, p̂(yp̂(y
1:T |θ
i−1 )p(θ i−1 ) ,
set θi = V otherwise, θi = θi−1 where p(θ) is the prior
distribution.
We set the standard deviations of observation errors to 20 per-

cent of the standard deviations of corresponding actual data series
following Herbst and Schorfheide (2015).
Appendix C. Model Stability
Combining Equation (16) and Equation (25) gives
at = (F + KC0 − KH)at−1 + KΩεt . (C.1)
As explained in Eusepi, Giannoni, and Preston (2015), the self-

referentiality of beliefs can lead to instability. This instability arises if
any eigenvalue of the matrix F +KC0 −KH lies outside the unit cir-
cle. This approach is, however, not possible in this case, because the
model stability depends on the evolution of the credibility measure ξ.
Specifically, C0 contains ξ, so the evolution of credibility affects the
dynamics of beliefs. For this reason, we analyze a Jacobian matrix of
this non-linear system following Hommes and Lustenhouwer (2019)
and Branch and McGough (2010), among many others.
ξ˜t is determined by the following:
P

exp δ 1 U
ξ˜t = 1 − t

exp δ1 UtP + exp δ1 UtCB
2
× 1 − exp −δ2 Êt−1 πt − Êt−1 πt
CB P
. (C.2)
We introduce an auxiliary variable qt which is defined as below.

1 − exp(δ1 UtCB − δ1 UtP ) δ1 CB
qt = = tanh − (Ut − Ut ) P
(C.3)
1 + exp(δ1 UtCB − δ1 UtP ) 2
Using this auxiliary variable and evaluating the distance between
two predictions, ξ˜t can be simplified as below.
qt + 1
ξ˜t = 1 − 1 − exp(−δ2 f22 (aπt−2 )2 ) (C.4)
2
The model can be summarized as a system of non-linear

equations.
zt = M1 (at−1 , ξt−1 , st )
at = M2 (at−1 , zt , st−1 )
ξt = M3 (qt , at−2 , ξt−1 )
qt = M4 (at−2 , st−1 , zt ) (C.5)
at−1 = I(at−1 )
st+1 = M5 (st )
st = I(st )
Before analyzing the stability, we show that there is a steady

state in this system.
Proposition 1. A steady state of the model exists and this steady
state satisfies x = 0, π = 0, i = 0, a = 0, s = 0, q = 0, and ξ = 1.
Proof. It is easy to show that x = 0, π = 0, i = 0, a = 0, s = 0,
q = 0, and ξ = 1 solve the system of equations.
As this steady state with x = 0, π = 0, and i = 0 coincides with
that of a rational expectations model, we label this steady state as
the fundamental steady state.
The next proposition provides the global stability result of the
model. The main idea behind this result is that, as zero credibility
is the most de-stabilizing condition, the system is globally stable if
the eigenvalues of the Jacobian are inside the unit circle when the
credibility measure ξ is fixed at zero for all t.
Proposition 2. The fundamental steady state is globally stable
under the baseline calibration.
Proof. If full credibility, ξ = 1 for all t, is imposed, the econ-
omy has a globally stable steady state, which is the fundamental
steady state, as the belief terms a do not affect the dynamics, and
the steady state is exactly the same with that of the underlying
rational expectations model. As ξ moves away from one to zero, the
influence of the belief terms a on the system increases and the eco-
nomic dynamics become more unstable. Therefore, if the economic
system is stable under zero credibility, it is globally stable. Under
the assumption that ξ = 0 for all t, the system can be written as

follows:
at = M2 (at−1 , M1 (at−1 , st ), st−1 )
st+1 = M5 (st ) (C.6)
st = I(st ).
Then, we can obtain the following Jacobian matrix:

⎡ ⎤
f1 − kf11 + Δ11 −kf12 + Δ12 ka11 ka12 ka13 −ka11 ρr −ka12 ρu −ka13 ρm
⎢ −kf21 + Δ21 f2 − kf22 + Δ22 ka21 ka22 ka23 −ka21 ρr −ka22 ρu −ka23 ρm ⎥
⎢ ⎥
⎢ 0 0 ρr 0 0 0 0 0 ⎥
⎢ ⎥
⎢ 0 0 0 ρ 0 0 0 0 ⎥
⎢ u ⎥
⎢ 0 0 0 0 ρ 0 0 0 ⎥,
⎢ m ⎥
⎢ 0 0 1 0 0 0 0 0 ⎥
⎢ ⎥
⎢ 0 0 0 1 0 0 0 0 ⎥
⎢ ⎥
⎣ 0 0 0 0 1 0 0 0 ⎦
0 0 0 0 0 0 0 0
(C.7)
where
σdk1j − φπ dk2j κσdk1j + (φx + σ)dk2j
Δij = Ki1 + Ki2
φx + κφπ + σ φx + κφπ + σ
σ(φx + κφπ )dk1j + φπ σdk2j
+ Ki3
φx + κφπ + σ

f1 λ1 β f2 1 − λ2 β
dk11 = 1−β− , dk12 =
1 − βf1 σ 1 − βf2 σ
καβf1 (1 − α)βf2
dk21 = , dk22 = ,
1 − αβf1 1 − αβf2
where kfij = fj Kij + λj fj Ki3 , kaij = Ki1 a1j + Ki2 a2j + Ki3 a3j , and
Kij denotes (i, j) element of the Kalman gain matrix. It has three
zero and five non-zero eigenvalues, which are all inside the unit circle
under baseline calibration. Therefore, the fundamental steady state
is globally stable.
Appendix D. Alternative Calibration
In this appendix, we provide evidence that using alternative cali-

brations for δ1 , δ2 , and W does not change the quantitative result
considerably. To be precise, we illustrate this robustness by showing
Figure D.1. Estimated Credibility with Different

Calibrations (left) and Estimated aπ
with Different Calibrations (right)
the estimated credibility and private beliefs across different calibra-

tions. Figure D.1 presents the estimated results. The blue solid lines
represent the estimated series obtained from the baseline calibra-
tion. The red dotted lines and the green dashed lines are estimated
results with different calibrations. Specifically, “Alternative 1” shows
the case that δ1 = δ2 = 15 while “Alternative 2” stands for the case
with different W shown below.
⎡ ⎤
000
W = ⎣0 1 0⎦
000
We choose δ1 = δ2 = 15 as an alternative calibration in this exer-

cise, but choosing any values with sufficiently large δ2 , say greater
than 2, produces very similar result. Under “Alternative 2” assump-
tion, the private agents only care about the forecast errors associated
with inflation. The estimated output shows that the results are quite
similar to the baseline case.
Appendix E. Additional Impulse Responses
Figure E.1. Impulse Responses to

One-Standard-Deviation Preference
Shock When aπ Is Negative
Note: The blue solid line represents impulse responses from ξ−1 = 1, aπ−1 =
−0.01 and the red dashed line represents impulse responses from ξ−1 = 0.4,
aπ−1 = −0.9. The green dotted line represents impulse responses from ξ−1 = 0.6,
aπ−1 = −0.3.

Shock When ax Is Negative
Note: The blue solid line represents impulse responses from ξ−1 = 1, ax−1 =
ax−1 = −0.9. The green dotted line represents impulse responses from ξ−1 = 0.6,
ax−1 = −0.3.

Shock When ax Is Positive
Note: The blue solid line represents impulse responses from ξ−1 = 1, ax−1 = 0.01
and the red dashed line represents impulse responses from ξ−1 = 0.4, ax−1 = 0.9.
The green dotted line represents impulse responses from ξ−1 = 0.6, ax−1 = 0.3.

One-Standard-Deviation Cost-Push
Shock When aπ Is Positive
The green dotted line represents impulse responses from ξ−1 = 0.6, aπ−1 = 0.3.

aπ−1 = −0.3.

ax−1 = −0.3.


One-Standard-Deviation Monetary Policy
Shock When aπ Is Positive
The green dotted line represents impulse responses from ξ−1 = 0.6, aπ−1 = 0.3.

aπ−1 = −0.3.

ax−1 = −0.3.

Appendix F. Scatter Plots of Simulated Series
We provide additional evidence that a combination of lower credibil-

ity and shifting private beliefs undermines macroeconomic stability
based on the simulated time series. Figure F.1 shows the relation-
ship between credibility and economic outcomes. The upper panels
present the relationship between private beliefs and credibility while
the bottom panels illustrate the co-movements of macro variables
and credibility. The nominal factor spreads out as credibility dete-
riorates. This leads to more volatile realizations of inflation that is
caused by a self-referential effect examined above. The real factor
shows the same pattern as in the nominal factor. It suggests that
the private beliefs are closely related to central bank credibility as
asserted. However, the output gap does not show any distinctive pat-
tern across the different levels of credibility. This is in line with the
above result that the volatility of the output gap is not significantly
affected by credibility.
Figure F.1. Scatter Plots of Simulated Series
References
Ahrens, S., J. Lustenhouwer, and M. Tettamanzi. 2017. “The Stabi-

lizing Role of Forward Guidance: A Macro Experiment.” Working
Paper.
Angeletos, G.-M., and J. La’O. 2013. “Sentiments.” Econometrica
81 (2): 739–79.
Ball, L. 1994. “Credible Disinflation with Staggered Price-Setting.”
American Economic Review 84 (1): 282–89.
———. 1995. “Disinflation with Imperfect Credibility.” Journal of

Monetary Economics 35 (1): 5–23.
Bernanke, B. S. 2004. “Central Bank Talk and Monetary Policy.”
Speech to the Japan Society, New York, October 7.
Blinder, A. S. 2018. “Through a Crystal Ball Darkly: The Future of
Monetary Policy Communication.” American Economic Review:
Papers and Proceedings 108 (5): 567–71.
Bomfim, A., and G. Rudebusch. 2000. “Opportunistic and Deliber-
ate Disinflation under Imperfect Credibility.” Journal of Money,
Credit and Banking 32 (November): 707–21.
Branch, W., and B. McGough. 2010. “Dynamic Predictor Selection
in a New Keynesian Model with Heterogeneous Expectations.”
Journal of Economic Dynamics and Control 34 (8): 1492–1508.
Brock, W. A., and C. H. Hommes. 1997. “A Rational Route to
Randomness.” Econometrica 65 (5): 1059–95.
Calvo, G. A. 1983. “Staggered Prices in a Utility-Maximizing Frame-
work.” Journal of Monetary Economics 12 (3): 383–98.
Campbell, J. R., C. L. Evans, J. D. M. Fisher, and A. Justini-
ano. 2012. “Macroeconomic Effects of Federal Reserve Forward
Guidance.” Brookings Papers on Economic Activity (1, Spring):
1–80.
Campbell, J. R., J. D. M. Fisher, A. Justiniano, and L. Melosi.
2017. “Forward Guidance and Macroeconomic Outcomes Since
the Financial Crisis.” NBER Macroeconomics Annual 2016,
Vol. 31, ed. M. Eichenbaum, E. Hurst, and J. A. Parker, 283–357.
Chicago: University of Chicago Press.
Cukierman, A., and A. Meltzer. 1986. “A Theory of Ambigu-
ity, Credibility, and Inflation under Discretion and Asymmetric
Information.” Econometrica 54 (5): 1099–1128.
Demertzis, M., M. Marcellino, and N. Viegi. 2010. “Anchors for
Inflation Expectations.” Working Paper.
Erceg, C., and A. Levin. 2003. “Imperfect Credibility and Inflation
Persistence.” Journal of Monetary Economics 50 (4): 915–44.
Eusepi, S., M. Giannoni, and B. Preston. 2015. “The Limits of Mon-
etary Policy with Long-term Drift in Expectations.” Working
Paper.
Eusepi, S., and B. Preston. 2011. “Expectations, Learning, and Busi-
ness Cycle Fluctuations.” American Economic Review 101 (6):
2844–72.
Gabaix, X. 2014. “A Sparsity-based Model of Bounded Rationality.”

Quarterly Journal of Economics 129 (4): 1661–1710.
———. 2016. “Behavioral Macroeconomics via Sparse Dynamic Pro-
gramming.” Working Paper.
———. 2020. “A Behavioral New Keynesian Model.” American Eco-
nomic Review 110 (8): 2271–2327.
Gibbs, C. G., and M. Kulish. 2017. “Disinflations in a Model
of Imperfectly Anchored Expectations.” European Economic
Review 100 (November): 157–74.
Goodfriend, M., and M. King. 2015. “Review of the Riksbank’s
Monetary Policy 2010-2015.”
Herbst, E. P., and F. Schorfheide. 2015. Bayesian Estimation of
DSGE Models. Princeton University Press.
Hommes, C., and J. Lustenhouwer. 2019. “Inflation Targeting and
Liquidity Traps under Endogenous Credibility.” Journal of Mon-
etary Economics 107 (November): 48–62.
Johnson, D. 1997. “Expected Inflation in Canada 1988-1995: An
Evaluation of Bank of Canada Credibility and the Effect of Infla-
tion Targets.” Canadian Public Policy 23 (3): 233–58.
———. 1998. “The Credibility of Monetary Policy: International
Evidence Based on Surveys of Expected Inflation.” In Price Sta-
bility, Inflation Targets and Monetary Policy, 361–95. Proceed-
ings of a conference held by the Bank of Canada, May 1997.
Kocherlakota, N. 2011. “Communication, Credibility and Implemen-
tation: Some Thoughts on Past, Current and Future Monetary
Policy.” Speech in Carlson School of Management, University of
Minnesota, September 6.
Mokhtarzadeh, F., and L. Petersen. 2021. “Coordinating Expecta-
tions through Central Bank Projections.” Experimental Econom-
ics 24 (3): 883–918.
Orphanides, A., and J. Williams. 2004. “Imperfect Knowledge,
Inflation Expectations, and Monetary Policy.” In The Inflation-
Targeting Debate, ed. B. S. Bernanke and M. Woodford, 201–46.
University of Chicago Press.
Plosser, C. I. 2010. “Credible Commitments and Monetary Policy
After the Crisis.” Speech to the Swiss National Bank Monetary
Policy Conference, Zurich, Switzerland, September 24.
Preston, B. 2005. “Learning about Monetary Policy Rules when

Long-Horizon Expectations Matter.” International Journal of
Central Banking 1 (2, September): 81–126.
Schaumburg, E., and A. Tambalotti. 2007. “An Investigation of
the Gains from Commitment in Monetary Policy.” Journal of
Monetary Economics 54 (2): 302–24.
Wu, J. C., and F. D. Xia. 2016. “Measuring the Macroeconomic
Impact of Monetary Policy at the Zero Lower Bound.” Journal
of Money, Credit and Banking 48 (2–3): 253–91.
Yellen, J. L. 2006. “Enhancing Fed Credibility.” Speech to
the Annual Washington Policy Conference, sponsored by the
National Association for Business Economics (NABE), Wash-
ington, DC, March 13.

Ijcb 23 Q 2 A 3

Uploaded by

Copyright:

Available Formats

Ijcb 23 Q 2 A 3

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ijcb 23 Q 2 A 3

Uploaded by

Copyright:

Available Formats

Central Bank Credibility and Monetary Policy∗

In this paper, a numerical measure of central bank credibil-

Over the past 20 years, many central banks in advanced countries

New Keynesian framework, which has been a workhorse for central

generates a 33 percent higher standard deviation of inﬂation com-

There have been studies that connect bounded rationality to

This section develops a variant of the canonical New Keynesian

where β ∈ (0, 1) is the discount factor, χ > 0 measures disutility of

and the no-Ponzi condition

The variables ct (i), nt (i), bt (i), it , πt , wt , and Γt (i) denote con-

subject to the linear production technology, yT (i) = nT (i), and the

where α is the Calvo (1983) parameter and denotes the probability

In a symmetric equilibrium, private agents share the same sub-

The slope of the Phillips curve is given by κ = (1 − αβ)(1 − α)/α.

3. Beliefs, Forecasts, and Credibility

In this research, we rely on a near-rationality assumption and a

3.1 Belief Structures

zt = Hāt−1 + Ωst + e1t

where zt = [xt πt it ] is a vector containing endogenous variables that

where εt are i.i.d. shocks. It is also assumed that parameters that

Based on the estimated time-varying drift terms at−1 up to time

ÊtP [zT ] = HF T −t at−1 + ΩΦT −t st , (14)

where ÊtP denotes the private forecast based on the information up

lim ÊtP [zT ] = Hat−1 (15)

Following Eusepi, Giannoni, and Preston (2015), at is updated

at = F at−1 + K(zt − Hat−1 − ΩΦst−1 ), (16)

where the time-invariant Kalman gain matrix is

K = P H  (HP H  + R)−1 . (17)

P is given as E[(āt − at )(āt − at ) ] and R denotes private agents’

zt = Ωst + e2t , (18)

which coincides with the rational expectations solution to the model.

ÊtCB [zT ] = ΩΦT −t st , (19)

3.2 Credibility Measure

Êt zT = ξt−1 ÊtCB [zT ] + (1 − ξt−1 )ÊtP [zT ]. (20)

As the proposed measure determines the relative weight among

for k ∈ {CB, P } and

ξt = ξt−1 + η(ξ˜t − ξt−1 ). (24)

U k is the ﬁtness measure of central bank (k = CB) or private

the credibility measure. As it approaches one, inertia does not exist

I’ve been emphasizing the importance of communication and

This is also in line with Plosser (2010)’s prescription of achieving

targets may not coincide with pre-announced policy targets, it is

zt = C0 at−1 + Ωst (25)

4. Historical Credibility of Central Banks

The credibility measure suggested above can be readily computed

Figure 1. Credibility Measure of the

areas represent recessions identiﬁed by the National Bureau of Eco-

Figure 2. Credibility of the European Central Bank,

presented in details below. Experiencing drops in credibility after

Interestingly, the impacts of forward guidance seem unclear. This

As this paper focuses on quantitatively measuring the inﬂuence of

of Professional Forecasters (SPF) to match the model-generated pri-

5.2 Calibrated Parameters

Table 1. Estimated Parameters

α Beta 0.7 0.08 0.73 [0.72,0.73]

5.4 Estimation Outputs

Figure 3. Comparison between Estimated

they are model speciﬁc, and so there is no consensus on the choice

parameters. Appendix C analyzes the model stability and conﬁrms

where zt = [xt πt it ] is a vector containing endogenous variables that

K = P H (HP H + R)−1 . (17)

P is given as E[(āt − at )(āt − at ) ] and R denotes private agents’