Working Paper Series - Identification of systematic monetary policy

Working Paper Series

Identification of systematic monetary

policy

Lukas Hack, Klodiana Istrefi, Matthias Meier

Disclaimer: This paper should not be reported as representing the views of the European Central Bank

(ECB). The views expressed are those of the authors and do not necessarily reflect those of the ECB.

No 2851

Abstract

We propose a novel identiﬁcation design to estimate the causal eﬀects of systematic

monetary policy on the propagation of macroeconomic shocks. The design combines (i) a

time-varying measure of systematic monetary policy based on the historical composition of

hawks and doves in the Federal Open Market Committee (FOMC) with (ii) an instrument

that leverages the mechanical FOMC rotation of voting rights. We apply our design to

study the eﬀects of government spending shocks. We ﬁnd ﬁscal multipliers between two and

three when the FOMC is dovish and below zero when it is hawkish. Narrative evidence from

historical FOMC records corroborates our ﬁndings.

Keywords: monetary policy, FOMC, rotation, government spending.

JEL Codes: E32, E52, E62, E63, H56.

ECB Working Paper Series No 2851

Non-technical summary

Monetary policy decisions made by central banks are intentional responses to macroeconomic

conditions. These responses are known as systematic monetary policy. In theory, system-

atic monetary policy plays a crucial role in inﬂuencing the impact of macroeconomic shocks.

However, there is a lack of reduced-form empirical evidence that identiﬁes and quantiﬁes this

causal relationship. In this study, we ﬁrst introduce an identiﬁcation design to assess the causal

eﬀects of systematic monetary policy on the propagation of macroeconomic shocks. We then

use this design to study the interaction between government spending and the response of

systematic monetary policy. Our ﬁndings show that systematic monetary policy is a crucial

determinant of the eﬀectiveness of ﬁscal policy.

Our identiﬁcation design combines a measure of systematic monetary policy based on the histor-

ical composition of hawks and doves in the Federal Reserve’s Federal Open Market Committee

(FOMC) since the 1960s, along with an instrument that levers the mechanical rotation of voting

rights in the FOMC. The classiﬁcation of FOMC members as hawks or doves is based on narra-

tives from news archives, portraying them as either more concerned about inﬂation (hawks) or

more concerned about supporting employment and growth (doves), as in Istreﬁ (2019).

To account for changes in the composition of hawks and doves in the FOMC that are inﬂuenced

by economic and political developments, we construct an instrumental variable that takes advan-

tage of the mechanical rotation of voting rights in the FOMC. This rotation is a yearly process

that redistributes voting rights among the Federal Reserve Bank presidents. The mechanical

nature of the rotation renders it exogenous to economic or political factors, allowing us to iden-

tify the causal eﬀects of systematic monetary policy. To the best of our knowledge, our FOMC

rotation instrument is the ﬁrst instrument of systematic monetary policy.

To estimate the causal eﬀects of systematic monetary policy on the propagation of ﬁscal shocks

in the economy, we use a local projection model where macroeconomic variables such as GDP

or government consumption respond to the ﬁscal shock, the interaction between the ﬁscal shock

and the Hawk-Dove balance in the FOMC, the level of the Hawk-Dove balance, and potentially

other factors as well. The instrument vector is given by the vector of regressors when replacing

the Hawk-Dove balance with the FOMC rotation instrument. For the ﬁscal shock, we examine

military spending shocks as studied in previous works by Ramey (2011) and Ramey and Zubairy

(2018), for the period from 1960 to 2014.

Our ﬁndings show that the response of GDP to government spending shocks depends crucially

on the number of dovish and hawkish FOMC members. An increase in discretionary govern-

ment spending leads to a GDP expansion which is more pronounced when more dovish FOMC

members vote in the FOMC. Conversely, more hawks dampen the expansionary eﬀect of govern-

ment spending. Quantitatively, we ﬁnd that the peak GDP increase roughly doubles when there

are two more doves in the FOMC relative to the long-run sample average. In contrast, we ﬁnd

that GDP does not expand in response to additional government spending when there are two

more hawks in the FOMC.

A common metric to evaluate the eﬀectiveness of ﬁscal spending is the ﬁscal multiplier: the

increase in GDP per additional government spending. We ﬁnd a strong and highly signiﬁcant

dependence of the ﬁscal multiplier on systematic monetary policy. Under a hawkish FOMC,

ECB Working Paper Series No 2851

the multiplier is insigniﬁcant, with point estimates at or below 0. Under a dovish FOMC,

the multiplier is highly statistically signiﬁcant and ranges between 2 and 3. An additional

important result is that when we consider systematic monetary policy in our analysis, the

average multipliers are larger and more precisely estimated compared to a linear model that

ignores this relationship.

Upon examining the mechanism underlying the state-dependent eﬀects of government spending

shocks, we observe distinct patterns in nominal interest rates depending on the hawkishness

of the FOMC. Under a hawkish FOMC, nominal interest rates tend to rise substantially. On

the other hand, under a dovish FOMC, nominal interest rates initially decline and experience a

delayed increase. This suggests that a hawkish FOMC hikes rates in response to ﬁscal expansion

to contain inﬂationary pressures. Indeed, we ﬁnd that a hawkish FOMC is more successful in

containing inﬂation expectations and inﬂation.

It is important to note that drawing the conclusion that the government should increase spending

when central banks have committees with dovish members in the majority could be misleading.

This is because such changes in government spending would not be random shocks (what we

studied) but predictable policy decisions. The Lucas critique applies if there are structural

changes in the conduct of ﬁscal policy. To avoid misleading conclusions, a promising direction

for future research is to utilize our ﬁndings to inform micro-founded models that study optimal

ﬁscal stabilization policies.

Finally, while our identiﬁcation design is speciﬁc to U.S. monetary policy, a promising avenue

for future research is to study other countries or currency areas in which committees decide

monetary policy. In fact, since 2015 the European Central Bank’s Governing Council allocates

voting rights to its members through a rotation mechanism. Investigating these contexts can

provide valuable insights into the interaction between systematic monetary policy and ﬁscal

shocks in diﬀerent settings.

ECB Working Paper Series No 2851

1 Introduction

Monetary policy is not random but a purposeful response to macroeconomic conditions. This

response represents systematic monetary policy. Fundamentally, the systematic response reﬂects

the preferences of the policymakers, e.g., concerning price stability and employment, which

change over time as the policymakers change. As a consequence, the eﬀects of macroeconomic

shocks diﬀer across time, depending on systematic monetary policy. In theory, systematic

monetary policy is well-known to be important for the propagation of macroeconomic shocks.

However, there is no direct evidence on the causal eﬀects of the Fed’s systematic monetary

policy.

The main contribution of this paper is an identiﬁcation design to estimate the causal eﬀects

of systematic monetary policy on the propagation of macroeconomic shocks. We use historical

ﬂuctuations in the composition of hawks and doves in the Federal Open Market Committee

(FOMC) to measure time variation in systematic monetary policy. To address the concern

that these ﬂuctuations are endogenous to economic and political developments, we propose an

instrument that exploits the mechanical rotation of voting rights in the FOMC. To the best of

our knowledge, our FOMC rotation instrument is the ﬁrst instrument for systematic monetary

policy. We then apply the identiﬁcation design to government spending shocks and ﬁnd that

ﬁscal multipliers signiﬁcantly depend on systematic monetary policy. When the FOMC is

dovish, it delays tightening in response to an expansionary ﬁscal spending shock, and ﬁscal

multipliers are between two and three. Conversely, multipliers can be negative under a hawkish

FOMC that tightens faster and more aggressively.

We measure time variation in systematic U.S. monetary policy building on the narrative clas-

siﬁcation of FOMC members by Istreﬁ (2019) which uses news archives to classify members of

the FOMC as hawks and doves, for the period 1960 to 2023. Hawks are more concerned about

inﬂation, while doves are more concerned about supporting employment and growth. Following

Istreﬁ (2019) and Bordo and Istreﬁ (2023), we aggregate individual FOMC member preferences

into an aggregate Hawk-Dove balance for each FOMC meeting.

The Hawk-Dove balance is an

appealing measure of systematic monetary policy because it reﬂects the aggressiveness of the

FOMC towards fulﬁlling one or the other leg of the dual mandate without having to specify a

policy reaction function or the policy tools.

Identifying the causal eﬀects of systematic monetary policy, independent of how it is measured,

is challenging because of endogeneity. For example, systematic monetary policy may change in

response to unemployment or inﬂation (Davig and Leeper, 2008). Similarly, the appointment

of central bankers can depend on economic and political circumstances, e.g., as documented

A vast empirical literature estimates the eﬀects of monetary policy shocks (e.g., the pioneering work by

Romer and Romer, 1989; Bernanke and Blinder, 1992; Cochrane and Piazzesi, 2002). These shocks are commonly

understood as deviations from a policy rule, whereas most policy variation is due to systematic monetary policy,

i.e., the rule itself. While evidence on monetary policy shocks may (indirectly) be informative about the eﬀects

of systematic monetary policy under certain assumptions (e.g., McKay and Wolf, 2022), we propose to directly

estimate the causal eﬀects of systematic monetary policy.

Istreﬁ (2019) constructs the FOMC Hawk-Dove balance and shows that these preferences match with narra-

tives on monetary policy, preferred interest rates, dissents, and forecasts of FOMC members. Bordo and Istreﬁ

(2023) study what forms these preferences and how the FOMC composition aﬀects decision making by estimating

a Taylor rule augmented by the Hawk-Dove balance. We go beyond their analysis by estimating the dynamic

causal eﬀects of systematic monetary policy on the propagation of macroeconomic shocks.

ECB Working Paper Series No 2851

for the Nixon administration (Abrams, 2006; Abrams and Butkiewicz, 2012). We discuss this

identiﬁcation challenge through the lens of a New Keynesian model in which the coeﬃcients of

the monetary policy rule ﬂuctuate in response to macroeconomic shocks. The model dynamics

can be represented as a state-dependent local projection. The OLS estimates of the local

projection will fail to identify the causal eﬀects of systematic monetary policy because they are

contaminated by unobserved shocks that change the monetary policy rule. Instead, we show

that an instrument that captures exogenous variation in systematic monetary policy achieves

identiﬁcation.

We propose an instrument that levers exogenous variation in the Hawk-Dove balance arising

from the FOMC rotation. The rotation is an annual mechanical scheme that shuﬄes four voting

rights among eleven Federal Reserve Bank presidents.

Speciﬁcally, our FOMC rotation instru-

ment is the Hawk-Dove balance of the subset of FOMC members with temporary voting rights

through the rotation. Importantly, the mechanic nature of the rotation renders it orthogonal

to economic and political developments.

Our identiﬁcation design combines the measure of systematic monetary policy and the instru-

ment in a state-dependent local projection for a macroeconomic shock of interest. Speciﬁcally,

we regress an outcome of interest on the shock, the shock interacted with the Hawk-Dove

balance, the Hawk-Dove balance in levels, and possibly further controls. The instrument vector

is given by the vector of regressors when replacing the Hawk-Dove balance with the FOMC

rotation instrument. This local projection is in line with the dynamics of a New Keynesian

model with time-varying systematic monetary policy. However, diﬀerent from a New Keynesian

model, our design identiﬁes the eﬀects of systematic monetary policy without imposing strong

structural assumptions.

We apply our identiﬁcation design to study the eﬀects of government spending shocks. The

response of monetary policy to ﬁscal policy is widely considered to be crucial for the eﬀectiveness

of ﬁscal policy, both in the policy debate (e.g., Blinder, 2022) and in New Keynesian theory

(e.g., Woodford, 2011; Farhi and Werning, 2016). Notwithstanding the perceived importance of

this type of ﬁscal-monetary interaction, there is no direct evidence on the causal eﬀects of the

Fed’s systematic monetary policy for the propagation of government spending shocks.

We focus on the military spending shocks in Ramey (2011) and Ramey and Zubairy (2018) for

the period 1960-2014.

We ﬁnd that the real GDP response signiﬁcantly depends on systematic

monetary policy. The GDP response increases in the share of dovish FOMC members, and

decreases in the share of hawks. When the Hawk-Dove balance exceeds the sample average

by two doves, quarterly GDP increases by up to 0.7% in response to a military shock, which

is expected to raise cumulative military spending by 1% of GDP over the next ﬁve years.

Conversely, quarterly GDP falls by up to 0.3% when the Hawk-Dove balance exceeds the sample

average by two hawks.

The rotation is considered important by Fed watchers in the media. Each year before the rotation, they

discuss its implications for the direction of monetary policy. Relatedly, Ehrmann et al. (2022) study how voting

rights aﬀect the communication of Federal Reserve Bank presidents and market reaction to this communication.

In the post-Korean War sample that we study, Ramey (2011) shows that these shocks are poor instruments

and the average spending multiplier is imprecisely estimated. In contrast, we show that accounting for time-

varying systematic monetary policy strongly improves the precision of the estimated multiplier.

For comparison, an increase of the Hawk-Dove balance by two doves or two hawks roughly corresponds to

ECB Working Paper Series No 2851

The negative (and signiﬁcant) dependence of the GDP response on the FOMC’s Hawk-Dove

balance is in line with commonly used New Keynesian models in which a more aggressive central

bank response to ﬁscal shocks leads to smaller GDP eﬀects. We see this evidence as supporting

the usefulness of our identiﬁcation design. In contrast, the OLS estimate substantially under-

estimates the role of systematic monetary policy on the GDP response to spending shocks.

A common metric to assess the eﬀectiveness of ﬁscal spending is the spending multiplier, the

dollar increase of real GDP per additional dollar of real government spending. We estimate the

two- and four-year cumulative spending multipliers and ﬁnd strong dependence on systematic

monetary policy. While multipliers under a hawkish FOMC are typically insigniﬁcant with point

estimates at or below 0, we ﬁnd that dovish multipliers are between 2 and 3 and statistically

signiﬁcant. Moreover, the average multipliers are larger and much more precisely estimated

when accounting for systematic monetary policy compared to a linear model that omits this

state dependency. These results are robust to various modeling choices, as we show in an

extensive sensitivity analysis.

We further inspect the mechanism behind the state-dependent eﬀects of government spending

shocks. We show that nominal interest rates rise under a hawkish FOMC, while they initially

fall and rise only with substantial delay under a dovish FOMC. When the Hawk-Dove balance

exceeds the sample average by two hawks, the federal funds rate (FFR) starts to increase within

one year and rises up to 50 basis points beyond the pre-shock level around two years after the

shock. Conversely, the FFR falls below the pre-shock level for more than two years after the

shock, and then sharply rises toward a 50 basis point increase three years after the shock, when

there are two more doves in the FOMC. The diﬀerent interest rate responses are consistent

with the ﬁscal multiplier estimates across hawkish and dovish FOMCs. Moreover, we ﬁnd that

hawkish policy is more successful in containing inﬂation and that the monetary policy response

primarily transmits to real GDP through private consumption.

Finally, we complement our quantitative analysis with narrative evidence from the historical

records of the FOMC meetings. These records reveal that FOMC members and staﬀ frequently

discuss changes in (military) government spending, their potential impact on the economy

and inﬂation, and the FOMC’s policy response. We further provide detailed case studies of

two important military spending buildup events in the 1960s, associated with the U.S. Space

Program and the Vietnam War. We show that a hawkish FOMC indeed tightens faster after

military buildups, whereas a dovish FOMC delays action.

Relation to literature. This paper contributes to a literature that aims to identify the eﬀects

of systematic monetary policy on the propagation of macroeconomic shocks. Closely related

are McKay and Wolf (2022) and Barnichon and Mesters (2022) which diﬀer in the required

structural assumptions and observational requirements from our identiﬁcation design.

Their

approach assumes a linear (structural) model to identify the eﬀects of monetary policy rules

using multiple monetary policy (news) shocks. In contrast, our approach exploits the historical

one standard deviation in the change of the Hawk-Dove balance.

McKay and Wolf (2022) focus on constructing policy counterfactuals, whereas Barnichon and Mesters (2022)

uses a similar approach to study optimal policy. Relatedly, Wolf (2023) uses the approach of McKay and Wolf

(2022) to provide ﬁscal policy shock counterfactuals for a strict inﬂation targeting central bank.

ECB Working Paper Series No 2851

variation in systematic monetary policy in a non-linear model to directly estimate its causal

eﬀects. In fact, our evidence shows that the non-linearity with respect to systematic mone-

tary policy is statistically signiﬁcant and economically large. This suggests that avoiding the

linearity assumption is important, underscoring the importance of our identiﬁcation design. A

more traditional approach constructs monetary policy counterfactuals via a sequence of mone-

tary policy shocks (e.g., Bernanke et al., 1997; Kilian and Lewis, 2011; Benati, 2021).

Yet,

this approach is subject to the Lucas critique (Sargent, 1979), except for the special case of

modest shocks (Leeper and Zha, 2003). Our identiﬁcation design is not subject to this Lucas

critique because we explicitly model and estimate how the dynamics depend on systematic

monetary policy. Another closely related paper is Cloyne et al. (2021), which estimates the

role of systematic monetary policy for the propagation of ﬁscal consolidation shocks. Whereas

Cloyne et al. (2021) leverages time-invariant cross-country diﬀerences in systematic monetary

policy estimated from Taylor rule regressions, we leverage exogenous variation in systematic

monetary policy in the U.S. over time.

Another approach of estimating the eﬀects of time-varying systematic monetary policy uses

non-linear VAR models (e.g., Primiceri, 2005; Sims and Zha, 2006). A key advantage of our

approach is that it requires weaker identifying assumption and addresses the potential endo-

geneity of systematic monetary policy. Our paper also relates to a literature studying macroe-

conomic models with exogenous changes in systematic monetary policy (e.g., Davig and Leeper,

2007; Bianchi, 2013; Leeper et al., 2017) or endogenous changes (e.g., Davig and Leeper, 2008;

Barthélemy and Marx, 2017). Our time-series approach requires fewer structural assumptions

and provides moments to discipline such models.

Finally, our paper relates to a large empirical literature that estimates the government spending

multiplier. Most empirical estimates ﬁnd an average ﬁscal spending multiplier between 0.5 and

1.5 (e.g., Blanchard and Perotti, 2002; Mountford and Uhlig, 2009; Barro and Redlick, 2011;

Ramey, 2011). Closely related are recent papers which study how the eﬀects of government

spending shocks diﬀer at the zero lower bound (e.g., Ramey and Zubairy, 2018; Miyamoto

et al., 2018). While the zero lower bound reﬂects a speciﬁc monetary policy regime, this regime

is endogenous to the business cycle which means the estimates may reﬂect both the regime and

the shocks leading to it. Instead, we isolate the causal eﬀects of monetary policy. Another

related paper is Nakamura and Steinsson (2014), which estimates relative regional multipliers

that diﬀerence out the response of monetary policy. Our paper also relates to many recent

papers that have estimated state-dependencies of the multiplier (other than systematic monetary

policy), e.g., depending on the economy being in recession (Auerbach and Gorodnichenko, 2012;

Jordà and Taylor, 2016; Ramey and Zubairy, 2018; Ghassibe and Zanetti, 2022); sign of the

shock (Barnichon et al., 2022; Ben Zeev et al., 2023); exchange-rate regime, trade openness,

and public debt (Ilzetzki et al., 2013); foreign holdings of debt (Broner et al., 2022); and tax

progressivity (Ferrière and Navarro, 2018).

The paper is organized as follows: Section 2 provides a simple New Keynesian model to

discuss the identiﬁcation challenge. Section 3 introduces the identiﬁcation design for system-

A further related paper on the intersection of shocks and systematic policy is Arias et al. (2019) which

identiﬁes monetary policy shocks via sign restrictions on systematic monetary policy.

ECB Working Paper Series No 2851

atic U.S. monetary policy. Section 4 contains the main empirical results on the eﬀects of ﬁscal

spending shocks. Section 5 provides evidence for understanding the mechanism. Section 6

provides a narrative of the FOMC records in the 1960s. Section 7 concludes.

2 Identiﬁcation challenge

In this section, we present a stylized non-linear New Keynesian model in which systematic

monetary policy may ﬂuctuate endogenously. We use the model to expound the challenge of

empirically identifying the eﬀects of systematic monetary policy on the propagation of macroe-

conomic shocks.

A New Keynesian model. The model is a textbook New Keynesian model (e.g., Galí,

2015) except for a monetary policy rule with time-varying coeﬃcients. Households choose

consumption, labor and bond holdings to maximize E

∞

t=0



log C

− N

1+φ



subject to

budget constraints. Intermediate good ﬁrms produce variety goods using Y

= x

where

is exogenous productivity. The price of the variety good can be reset with a constant

probability 1 − θ. Final good ﬁrms produce the ﬁnal good Y



(ϵ−1)/ϵ



ϵ/(ϵ−1)

. A

ﬁscal policy authority ﬁnances government spending G

= γY x

with lump-sum taxes where

γ ∈ [0, 1), Y is steady-state output, and x

denotes exogenous changes in ﬁscal spending. Goods

market clearing requires Y

= C

+ G

. The exogenous variables follow stable AR(1) processes

log x

= ρ

log x

t−1

+ ε

with ε

∼ (0, σ

) for k = a, s respectively. A monetary policy rule

closes the model. Letting lower case letters denote (log) deviations from the steady state, the

monetary authority sets nominal interest rates i

according to

, (2.1)

where

∈ (1, ∞) is systematic monetary policy, which varies over time according to a stable

AR(1) process

= ρ

t−1

+ ζ

+ η

, (2.2)

where

= ϕ+ϕ

and ϕ denotes the unconditional mean of

. Importantly, we allow systematic

monetary policy to be endogenous, as ϕ

may respond to macroeconomic shocks (ε

, ε

Such

endogeneity creates an empirical identiﬁcation challenge as we discuss toward the end of this

section. In addition, we allow for exogenous changes in systematic monetary policy, captured

by the exogenous policy shifter η

. We assume that ε

, ε

, and η

are mutually independent

and identically distributed over time.

Accounting for the non-linear eﬀects of systematic monetary policy ϕ

, the approximate equi-

librium dynamics of GDP are given by

= a + b

+ b

+ c

+ dϕ

, (2.3)

For DSGE models with exogenous changes in the Taylor rule coeﬃcients, see, e.g., Davig and Leeper (2007)

and Bianchi (2013). For endogenous changes in the Taylor rule coeﬃcients, see, e.g., Davig and Leeper (2008)

and Barthélemy and Marx (2017).

ECB Working Paper Series No 2851

where a, b

, b

, c

, d are coeﬃcients that depend on the deep structural parameters of the

model. Appendix A.1 provides details on the derivation.

Identiﬁcation challenge. We next discuss the challenge of identifying the eﬀects of system-

atic monetary policy from a regression when y

is generated by (2.3). Without loss of generality,

and in anticipation of our main empirical question, we focus our discussion on the ﬁscal spending

shock. Consider an econometrician who observes y

, ε

, ϕ

, and estimates the state-dependent

local projection

t+h

= α

+ β

+ γ

+ δ

+ v

t+h

, (2.4)

for h = 0, . . . , H forecast horizons. For h = 0, the residual v

t+h

contains lagged spending shocks,

contemporaneous and lagged technology shocks, and the interaction of these shocks with ϕ

For h > 0, the residual further contains shocks (ε

, ε

) and policy shifter (η

) occuring between

t and t + h. The estimands in (2.4) are

= b

(ρ

)

, γ

= c

(ρ

)

, δ

= d(ρ

)

. (2.5)

Both β

, the average eﬀect of the spending shock, and γ

, the diﬀerential eﬀect associated with

, diminish in the forecast horizon h.

We next ask whether the OLS estimates of (β

, γ

, δ

) are consistent, i.e., whether they asymp-

totically recover the estimands in (2.5).

In general, consistency holds under the strong exogeneity

assumption ζ

= ζ

= 0, that is if ϕ

is independent of the macroeconomic shocks. In contrast,

if ϕ

correlates with at least one of the shocks, the OLS estimates do not consistently esti-

mate (β

, γ

, δ

). If, for example, ϕ

responds to a technology shock, the OLS estimator will

be contaminated by the response of GDP to the technology shock.

For further details, see

Appendix A.2.

Now suppose the econometrician observes an instrument ϕ

that is correlated with ϕ

(rele-

vance), but uncorrelated with all past, present, and future macroeconomic shocks ε

and ε

and that is uncorrelated with all past and future policy shifters η

(exogeneity). Consider the

IV estimates of (β

, γ

, δ

) when using (ε

, ε

, ϕ

) as instrument vector for the regres-

sors (ε

, ε

, ϕ

). The IV estimator consistently estimates (β

, γ

, δ

), even when ϕ

ﬂuctu-

ates endogenously in response to macroeconomic shocks (ζ

, ζ

̸= 0). For further details, see

Appendix A.2. This result guides the remainder of our paper in which we propose an instrument

for systematic monetary policy and use it to estimate the causal eﬀects of systematic monetary

policy.

Illustration. To illustrate the eﬀects of systematic monetary policy and the identiﬁcation

challenge, we focus on a special case of our economy in which ρ

= ρ

= 0. To understand

We explicitly include δ

in the vector of coeﬃcients because including the (endogenous) control variable ϕ

in the regression is important for identiﬁcation, as ϕ

is correlated with ε

and ε

in general.

If the econometrician observes and includes all shocks and corresponding interaction terms in the regression

according to equation (2.3), then the OLS estimates will be consistent without the exogeneity assumption. In

practice, this is infeasible as many shocks are (partially) unobserved.

ECB Working Paper Series No 2851

Figure 1: GDP response and systematic monetary policy

Notes: The solid line shows the model solution for the GDP response to a spending shock as a function of systematic

monetary policy (ϕ

), i.e., b

+ c

, with b

and c

given by (2.6) and the parametrization: β = 0.99, θ = 0.75, ϵ = 9,

φ = 2, γ = 0.2,

ϕ = 1.5, ζ

= 1, ζ

= 0.25, σ

= σ

= 1. The dashed line shows the OLS estimate

OLS

+ ˆγ

OLS

based

on a regression of (2.3) when the terms in u

are unobserved. The implied coeﬃcients are β

= 0.164 and γ

= −0.017,

and the large-sample OLS estimates are

OLS

= 0.164 and ˆγ

OLS

= −0.002.

how ϕ

aﬀects the GDP response to ﬁscal spending shock ε

, we need to know

= γ (1 + λϕ) ω

−1

, c

= −γ(1 − γ)λφω

−2

, (2.6)

where ω = 1 + λ (φ(1 − γ) + 1) ϕ, λ = (1 − θ)(1 − βθ)/θ. Since β

> 0 and γ

< 0 (under

standard parameter restrictions), the GDP response falls in the strength of the monetary policy

reaction to inﬂation. This is the monetary oﬀset (e.g., Woodford, 2011; Christiano et al., 2011).

The solid line in Figure 1 illustrates the monetary oﬀset. The dashed line illustrates the OLS

bias in the estimated GDP response to the spending shock. In our example, the OLS estimate

understates the role of systematic monetary policy.

3 Identiﬁcation design

In this section, we propose an identiﬁcation design to study how systematic monetary policy

in the U.S. shapes the propagation of macroeconomic shocks. Our identiﬁcation design relies

on three crucial elements: (i) a measure of systematic monetary policy, (ii) an instrument for

systematic monetary policy, and (iii) a state-dependent local projection regression that combines

(i) and (ii) to tackle the identiﬁcation challenge discussed in the preceding section.

3.1 Hawk-Dove balance in the FOMC

In the following, we build on the classiﬁcation of Federal Open Market Committee (FOMC)

members into hawks and doves by Istreﬁ (2019) and argue that the Hawk-Dove balance captures

ECB Working Paper Series No 2851

well variation in systematic monetary policy over time.

The FOMC. The FOMC is the committee of the Federal Reserve that sets U.S. monetary

policy. The FOMC consists of 12 members: the seven members of the Board of Governors of

the Federal Reserve System, including the Federal Reserve Chair, the president of the Federal

Reserve Bank (FRB) of New York, and four of the remaining 11 FRB presidents, who serve

one-year terms on a rotating basis. The seven FRB presidents temporarily without FOMC

membership participate in the FOMC meetings as non-voters.

Individual policy preferences. To measure the policy preferences of FOMC members we

use the Istreﬁ (2019) classiﬁcation of FOMC members as hawks and doves, for the period 1960-

2023.

Underlying this classiﬁcation are more than 20,000 real-time media articles from over

30 newspapers and business reports of Fed watchers (available in news archives like ProQuest

Historical Newspapers and Factiva) mentioning individual FOMC members. Istreﬁ (2019) uses

these articles to categorize individual FOMC members as hawks or doves for each FOMC

meeting based on the news information available up until the meeting. So, the Hawk-Dove

classiﬁcation is a panel that tracks FOMC members over time, at FOMC meeting frequency.

Hawks are perceived to be more concerned with inﬂation, while doves are more concerned with

employment and growth.

Through the lens of our model in Section 2, we can think about

hawks as preferring a larger inﬂation coeﬃcient ϕ

than doves. However, the Hawk-Dove clas-

siﬁcation we use is not tied to assuming a speciﬁc policy rule.

Overall, 129 of the 147 FOMC members between 1960 and 2023 are classiﬁed as hawk or dove.

The news coverage for the remaining 18 members does not allow classiﬁcation (as hawk or dove)

for any meeting, as some served in the early 1960s with sparse media coverage and others are

very recent appointments in the FOMC. The majority (95) of the classiﬁed FOMC members

are consistently hawks or doves over time while the rest switches camps at least once. Swings

are equally split in either direction and quite uniformly distributed over time. On average, the

34 swinging FOMC members switch camps at only 1.8% of the member-meeting pairs.

While true policy preferences are unobserved, Istreﬁ (2019) shows that perceived preferences

match well with policy tendencies that are unknown in real-time to the public, as expressed

by preferred interest rates, with forecasting patterns of individual FOMC members, and with

dissents. In addition, Bordo and Istreﬁ (2023) show that the FOMC members’ educational

background, e.g., whether they graduated from a university related to the Chicago school of

economics, and early life experience, i.e., whether they grew up during the Great Depression,

predicts the Hawk-Dove classiﬁcation. The long lasting eﬀect of the early life experience in the

formation of policy preferences is consistent with the very few swings in our sample.

The data in Istreﬁ (2019) covers 1960 through 2014. The data is currently extended up to the ﬁrst meeting

of 2023. Thus, our sample covers all 634 (scheduled) FOMC meetings between 1960 and 2023.

A typical example of a newspaper quote used to categorize a hawk reads: “Volcker leans toward tight-money

policies and high interest rates to retard inﬂation”, New York Times, 2 May 1975. For a dove: “The weakness

of Treasury prices and higher yields was seen reﬂecting the view that Bernanke will be ‘pro-growth’ and perhaps

less hawkish on inﬂation, said John Roberts, managing director at Barclays Capital in New York”, Dow Jones

Capital Markets Report, 24 October 2005.

ECB Working Paper Series No 2851

Aggregate Hawk-Dove balance. To measure variation in systematic monetary policy over

time, we aggregate the cross-section of individual FOMC member preferences into an aggregate

Hawk-Dove balance for each meeting (cf. Istreﬁ, 2019). We do so because the nature of monetary

policy-making by committee involves the aggregation of diverse individual policy preferences in

a collective decision.

We adopt a symmetric numerical scale for the qualitative Hawk-Dove

classiﬁcation in order to aggregate the preferences. We deﬁne Hawk

iτ

as the policy preference

of FOMC member i at FOMC meeting τ :

Hawk

iτ











+1 Consistent hawk

Swinging hawk

±0 Preference unknown

−

Swinging dove

−1 Consistent dove

(3.1)

A consistent hawk is an FOMC member that has not been categorized as a dove in the past.

In contrast, a swinging hawk has been a dove at some point in the past. The deﬁnition of a

consistent dove and a swinging dove is analogous. We assign a lower weight to swingers as they

are often perceived as ‘middle-of-the-roaders’ with more moderate leanings to the hawkish or

dovish side (Istreﬁ, 2019).

Finally, we assign Hawk

iτ

= 0 when the policy preference of the

FOMC member is (yet) unknown.

We next aggregate the individual policy preferences in (3.1). We compute the aggregate Hawk-

Dove balance by

Hawk

i∈M

Hawk

iτ

(3.2)

where M

denotes the set of FOMC members at meeting τ. A full FOMC consists of |M

| = 12

members but |M

| is occasionally below 12 because of absent members or vacant positions.

The Hawk-Dove balance in (3.2) is the arithmetic average across individual preferences. This

is our baseline aggregation of the Hawk-Dove balance in the FOMC and conforms well with the

consensual mode in which the FOMC typically operates.

1617

In Section 4.5, we show that our

Relatedly, Blinder (1999) writes: While serving on the FOMC, I was vividly reminded of a few things all of

us probably know about committees: that they laboriously aggregate individual preferences; that they need to be led;

that they tend to adopt compromise positions on diﬃcult questions; and–perhaps because of all of the above–that

they tend to be inertial.

Our empirical ﬁndings are robust to not distinguishing between consistent and swinging preferences, see

Section 4.5.

When a substitute temporarily replaces an absent FOMC member, we assume the substitute acts in the

interest of the original FOMC member and assign the same policy preference, see Appendix B for details. This

assumption aﬀects less than one percent of all observations and is not important for our results.

Riboni and Ruge-Murcia (2010) argue that a consensus model ﬁts actual policy decisions of the Federal

Reserve. In addition, Riboni and Ruge-Murcia (2022) provide evidence suggesting that policy proposals of the

Fed Chair are the result of a compromise, reﬂecting a balance of power within the FOMC.

Cieslak et al. (2022) construct a Hawk-Dove score based on the language in FOMC meeting transcripts.

In contrast to our measure which captures FOMC members preferences about monetary policy, their measure

captures (a hawkish or dovish) sentiment on current direction of policy changes. Furthermore, Ferguson et al.

(2023) classify central bank governors in 80 countries as hawks and doves, with respect to ﬁnancial sector support,

for the periods preceding banking crises.

ECB Working Paper Series No 2851

Figure 2: Hawk-Dove balance in the FOMC

Notes: The solid red line shows the quarterly time series of the aggregate Hawk-Dove balance of the FOMC

(Hawk

) from 1960 until 2023. The dashed red line shows the aggregate Hawk-Dove balance of the subgroup

of rotating FRB presidents with voting right in period t, the FOMC rotation instrument (Hawk

). Grey

bars indicate NBER dated recessions.

empirical ﬁndings are robust to alternatively using the median of preferences or putting a higher

weight on the Fed Chair’s preference. Finally, we aggregate Hawk

from meeting frequency to

quarterly frequency. We compute the Hawk-Dove balance Hawk

for quarter t as the average

balance in the ﬁrst month of the quarter. If the ﬁrst month is without a meeting, we use the

ﬁrst preceding month with a meeting.

We present the evolution of the Hawk-Dove balance from 1960 to 2023 as the solid line in

Figure 2. There is considerable variation in this balance, featuring both hawkish and dovish

majorities. The variation reﬂects the turnover of rotating FOMC members, the turnover of non-

rotating FOMC members, and changes in policy preferences of incumbent FOMC members. We

discuss the importance of these components for Hawk

ﬂuctuations in Subsection 3.2.

Systematic monetary policy. The aggregate Hawk-Dove balance Hawk

represents our

measure of systematic U.S. monetary policy. It accounts for the diversity of views within

the FOMC on how policy should be adjusted to promote both, price stability and maximum

employment. This diversity is usually expressed in FOMC meetings through diﬀerent forecasts

of individual members, through dissents, and in public through speeches. While the Fed’s

response to macroeconomic shocks is more sophisticated and depends on various economic

factors, we argue that our Hawk-Dove balance matches well with narratives of monetary policy

in the U.S. (Istreﬁ, 2019). For example, the dovish leaning of Hawk

in the mid-1960s coincides

with a period of delays and hesitation from the FOMC to take anti-inﬂationary action (Meltzer,

2005). The hawkish majorities in the 1970s might be surprising given the high inﬂation rates

in this period. Yet it is consistent with monetary policy being misguided by an underestimated

ECB Working Paper Series No 2851

natural rate of unemployment (DeLong, 1997; Romer and Romer, 2002) and persistence of

inﬂation (Primiceri, 2006). In particular, Orphanides (2004) argues that for the periods before

and after Paul Volcker’s appointment in 1979, policy was broadly similar and consistent with

a strong reaction to Greenbook inﬂation forecasts.

During the 1980s, the perception of a

less hawkish FOMC reﬂects nominations of dovish Board members by President Reagan. In

addition, it is consistent with the imperfect credibility of hawkish policy during the Volcker

disinﬂation, as observed in persistently elevated long-term interest rates (indicative of inﬂation

expectations) in this period (Goodfriend and King, 2005). Overall, this suggests that the Hawk-

Dove balance captures important aspects of the Fed’s systematic policy-making.

Our approach of measuring systematic policy via Hawk

has several advantages to alternative

approaches such as calibrating or estimating policy rules (e.g., Clarida et al., 2000; Bauer et al.,

2022). Importantly, we do not have to specify a particular reaction function, nor do we need to

restrict the analysis to speciﬁc policy instruments or communication strategies.

We further

avoid the well-known identiﬁcation issues that plague the estimation of monetary policy rules

(Cochrane, 2011; Carvalho et al., 2021). Independently of the policy tool or policy rule, our

measure reﬂects the aggressiveness of the FOMC towards fulﬁlling one or the other leg of

the dual mandate. In addition, the Hawk-Dove balance reﬂects public beliefs, in real-time,

about monetary policymakers. In contrast, ex-post estimates of systematic monetary policy

may inadvertently use ex-post information not available at the time of the policy decision,

potentially giving rise to misleading conclusions (Orphanides, 2003).

Comparability over time. A potential concern with the classiﬁcation of FOMC members

into hawks and doves is that the meaning of being a hawk or dove might have changed over

time. We argue this is likely no major concern. First, Istreﬁ (2019) has classiﬁed each member

as a hawk or dove based on a common and time-invariant deﬁnition, that is the policy leaning

with regard to the dual mandate of the Fed: maximum employment and stable prices. Second,

given that preferences tend to be stable, we would expect many swings whenever the meaning of

hawks or doves changes. However, swings in measured preferences are rare suggesting that the

meaning of being a hawk or dove is relatively stable over time. Third, the fact that we observe

large and persistent ﬂuctuations in Hawk

is incompatible with the Hawk-Dove classiﬁcation

being a relative ranking, according to which hawks are those FOMC members which are more

hawkish than the contemporaneous average policy preference among FOMC members, and

analogously for doves. Finally, in a robustness exercise in Section 4, we show that our results

are robust to using an alternative Hawk-Dove balance which accounts for potential trends in

the meaning of hawks and doves.

Relation to monetary policy shocks. Empirically identiﬁed monetary policy shocks are

often considered to reﬂect changes in central bank preferences (Christiano et al., 1999; Ramey,

Moreover, Orphanides (2003) shows that a dovish Taylor rule with a suﬃciently large weight on the output

gap would have resulted in substantially higher inﬂation.

For a summary of alternative policy rules that the FOMC consults, see here:

https://www.federalreserve.gov/monetarypolicy/policy-rules-and-how-policymakers-use-them.htm. Policy

instruments have been changing over our sample, from targeting monetary aggregates to targeting the Fed Funds

rate, conducting balance sheets policy, and through forward guidance communication.

ECB Working Paper Series No 2851

2016). Hence, they may be related to the Hawk-Dove balance, our measure of systematic mone-

tary policy. In Appendix D, we characterize this relationship based on the Romer and Romer

(2004) identiﬁcation strategy of monetary policy shocks. Because their identiﬁcation strategy

assumes a time-invariant policy rule, the identiﬁed monetary policy shocks may indeed capture

time variation in systematic monetary policy. However, the relationship between identiﬁed

monetary policy shocks and systematic monetary policy is non-linear and also depends on the

state of the economy (e.g., the inﬂation rate). Instead, our Hawk-Dove balance provides a

cleaner measure of systematic monetary policy.

3.2 FOMC Rotation Instrument

We next propose and discuss a novel FOMC rotation instrument that allows us to identify the

eﬀects of systematic monetary policy, even if monetary policy is endogenous to the state of the

economy (cf. Section 2).

Potential endogeneity. Systematic monetary policy may change depending on the state of

the economy. For example, the Federal Reserve may become more dovish in response to high

unemployment, or more hawkish in response to high inﬂation (cf. Davig and Leeper, 2007).

More fundamentally, some FOMC members may become hawkish or new appointments may

increase the number of hawks in the FOMC. Changes in individual policy preferences could be

intrinsic responses to changes in the macroeconomic environment. They could also be driven by

external pressure from lobbies or the government. Relatedly, Abrams (2006) and Abrams and

Butkiewicz (2012) document the inﬂuence of the Nixon administration on the FOMC in the

period leading up to the 1972 election. In addition, which type of central banker gets appointed

may depend on the state of the economy. In this context, note that members of the Board of

Governors and the Fed Chair require a nomination from the U.S. President for their ﬁrst and

any subsequent term. This may render both extensive margin and intensive margin changes in

the Hawk-Dove balance endogenous.

FOMC rotation instrument. To address the endogeneity of the Hawk-Dove balance we

propose an instrument which leverages exogenous variation in Hawk

that arises from the

annual FOMC rotation. Each year, four FOMC memberships rotate among eleven FRB pres-

idents following a mechanical scheme that has been in place since the early 1940s. According

to the scheme, some FRB presidents become FOMC members every second year (Cleveland

and Chicago) and others every third year (Philadelphia, Richmond, Boston, Dallas, Atlanta,

St. Louis, Minneapolis, San Francisco and Kansas City). As the rotation of voting rights is

independent of the state of the economy, it induces exogenous variation in Hawk

. To leverage

the variation from the FOMC rotation we propose a novel instrument, which we refer to as

FOMC rotation instrument. Formally, the instrument is given by

Hawk

i∈R

Hawk

iτ

, (3.3)

ECB Working Paper Series No 2851

where R

denotes the set of rotating FOMC members at FOMC meeting τ. A full set of rotating

members consists of |R

| = 4 members.

We aggregate the FOMC rotation instrument to

quarterly frequency analogously to the Hawk-Dove balance.

In Figure 2, the dashed line presents the FOMC rotation instrument over time. On average, the

rotating presidents are more hawkish than the overall FOMC Hawk-Dove balance, reﬂecting the

fact that FRB presidents tend to be more hawkish than governors (Chappell et al., 2005; Istreﬁ,

2019; Bordo and Istreﬁ, 2023). Both series display sizable variation over time, but ﬂuctuations

in the instrument Hawk

are more short-lived, with a year-over-year autocorrelation of 0.20

compared to 0.66 for Hawk

, see Table 1.

Table 1: Summary statistics

Mean Median SD Autocorr Corr Min Max T

Hawk

0.04 0.09 0.35 0.66 - -0.80 0.67 253

Hawk

0.28 0.33 0.45 0.20 0.64 -0.75 1.00 253

Notes: Summary statistics for the quarterly time series from 1960 until 2023. Hawk

is the average Hawk-

Dove balance of the FOMC. Hawk

is the FOMC rotation instrument. Autocorr refers to the year-over-year

ﬁrst-order autocorrelation. Corr refers to the correlation with Hawk

Relevance of instrument. Our instrument Hawk

aggregates the policy preferences of one-

third of the FOMC members, capturing a signiﬁcant part of the variation in the overall Hawk-

Dove balance Hawk

. In fact, the correlation between Hawk

and Hawk

is 0.64. Formal

weak instrument tests require a fully speciﬁed regression model and are therefore delegated to

Section 4. However, we can estimate a stylized ﬁrst-stage regression to study the explanatory

power of the FOMC rotation instrument. We regress Hawk

on Hawk

and a constant. This

regression has an R

of 0.41 and an eﬀective F-statistic (Montiel Olea and Pﬂueger, 2013)

for joint signiﬁcance of 46.13, well above the common threshold of 10 for weak instruments

(Andrews et al., 2019).

We further provide a decomposition of Hawk

into intensive margin changes of incumbent

FOMC members’ policy preferences and extensive margin changes in the composition of the

FOMC due to entry and exit, see Appendix C for details. We ﬁnd that extensive margin

changes in the FOMC composition due to the rotation account for 53% of the variance in yearly

changes of Hawk

. The turnover of non-rotating FOMC members accounts for another quarter

of the variance, and the remainder is due to preference changes of incumbent FOMC members

and various covariance terms. Both the ﬁrst-stage regression and the variance decomposition

strongly suggest that our instrument is relevant for Hawk

Finally, the rotation is considered important by Fed watchers in the media. Each year before

the rotation, they discuss its implications for monetary policy. A typical media discussion, here

an article in The New York Times from January 1, 2011, reads as follows:

In our sample, |R

| = 4 for 625 out of 634 FOMC meetings and |R

| = 3 for the remaining nine meetings

because of an absent member.

ECB Working Paper Series No 2851

As the Federal Reserve debates whether to scale back, continue or expand its $600

billion eﬀort to nurse the economic recovery, four men will have a newly prominent

role in inﬂuencing the central bank’s path. The four men are presidents of regional

Fed banks, and under an arcane system that dates to the Depression, they will become

voting members in 2011 on the Federal Open Market Committee, [...] the change

in voting composition is likely to give the committee a somewhat more hawkish cast.

This could amplify anxieties about unforeseen eﬀects of Bernanke’s policies [...]. Two

of the four new voters are viewed as hawkish on inﬂation, meaning that they tend

to be more worried about unleashing future inﬂation than they are about reducing

unemployment in the short run.

Exogeneity of instrument. We next argue that variation in Hawk

is quasi-exogenous.

First, the rotation scheme is mechanical and time-invariant and therefore unrelated to the state

of the economy. Second, new appointments of FRB presidents are relatively infrequent and

unlikely to be inﬂuenced by the federal government. FRB presidents are appointed by the

Board of Directors of the respective Federal Reserve district. The directors are to represent

the broader public and ﬁnancial institutions located in the district. In contrast, members of

the Board of Governors (including the Fed Chair) are nominated by the U.S. president and

conﬁrmed by the Senate. Furthermore, the average tenure of an FRB president is eleven years

but only seven years for a governor in our sample. Relatedly, Bordo and Istreﬁ (2023) show that

diﬀerent from governors, there is no correlation between the preferences of the FRB presidents

and the U.S. president’s party at the time of their appointment. In addition, some regional

FRBs have persistent leanings toward either the dovish or the hawkish camp. For example, the

Cleveland FRB president is typically a hawk whereas the president of the San Francisco FRB

is typically a dove.

Third, swings of preferences are likely a negligible threat to the exogeneity of our instrument.

For rotating FOMC members, swings occur only in 1.3% of member-meetings pairs, and not all

swings are endogenous to the state of the economy.

In addition, we ﬁnd that swings account

for a negligible fraction of the variance of the rotation instrument. In particular, we decompose

Hawk

into intensive margin changes of preferences (swings) and extensive margin changes

of the composition of rotating FOMC members due to either the rotation or appointments,

see Appendix C for details. The rotation accounts for 93% of the variance in yearly changes

of Hawk

, appointments for 8% and swings for 1%.

Fourth, Hawk

displays relatively

short-lived time series ﬂuctuations that are unlikely to be correlated with slow-moving macroe-

conomic trends, such as increasing market power, female labor force participation, and various

technological innovations. Similarly, Hawk

is uncorrelated with business cycle ﬂuctuations.

For example, the correlation between Hawk

and yearly real GDP growth is -0.02 and statis-

tically insigniﬁcant. In contrast, the correlation between Hawk

and GDP growth is 0.15 and

Bordo and Istreﬁ (2023) discuss three major swing waves in the FOMC during 1960-2014. The ﬁrst wave is

a hawkish wave inﬂuenced by inﬂation dynamics in the late 1960s to early 1970s. The second wave is a hawkish

swing in the early 1990s, related to the discussion on inﬂation targeting inspired by the announcements of the

Reserve Bank of New Zealand and Bank of Canada. Finally, the third swing wave is a dovish one in the late

1990s, following a new understanding of the economy.

Our empirical results are robust to excluding swingers from our instrumental variable, see Section 4.5.

ECB Working Paper Series No 2851

signiﬁcant at the 5% level.

Overall, the above arguments support the validity of our FOMC rotation instrument for iden-

tifying the causal eﬀects of systematic monetary policy. To the best of our knowledge, this

paper is the ﬁrst to propose an instrument for systematic monetary policy. We believe this is a

substantial contribution to the literature which opens up myriad research questions.

A validation exercise for Hawk

and Hawk

. Given our deﬁnition of hawkish policy

makers and conventional wisdom about hawkish monetary policy, we should expect a hawkish

FOMC to respond more aggressively to inﬂation. As validation exercise, we empirically test

this correlation via a dynamic Taylor rule regression. We use Hawk

as instrument in a

local projection of the federal funds rate on the Greenbook inﬂation forecast interacted with

Hawk

. We ﬁnd that a hawkish FOMC indeed raises the federal funds rate signiﬁcantly more

aggressively in the presence of higher inﬂation forecasts. For more details on the exercise, the

results, and a weak instrument test, see Appendix E. Overall, this exercise suggests that Hawk

and Hawk

capture important variation in systematic monetary policy.

3.3 Local projection framework

Finally, we propose to combine Hawk

and Hawk

in a state-dependent local projection

framework that permits causal identiﬁcation of how systematic monetary policy shapes the

propagation of various macroeconomic shocks. The setup of the local projection is consistent

with the New Keynesian model discussed in Section 2.

We regress an outcome variable of interest, x

t+h

, on a macroeconomic shock of interest, ε

, the

interaction of the shock with the Hawk-Dove balance Hawk

, as well as Hawk

in levels and a

vector of additional control variables Z

. Formally,

t+h

= α

+ β

+ γ

(Hawk

− Hawk) + δ

(Hawk

− Hawk) + ζ

+ v

t+h

, (3.4)

for h = 0, . . . , H forecast horizons. Hawk denotes the arithmetic sample mean of Hawk

. To

address the potential endogeneity of Hawk

, we use the instrument vector

1, ε

, ε



Hawk

− Hawk





Hawk

− Hawk



, Z

(3.5)

for the regressors in (3.4). The two key coeﬃcients in (3.4) are β

and γ

, which capture the

average response, when the Hawk-Dove balance of the FOMC is at its sample average, and

the diﬀerential response, when the FOMC is more or less hawkish than the sample average.

Based on Section 2, the IV estimator is consistent if the instrument Hawk

is orthogonal to

all macroeconomic shocks (both observed shocks ε

and other unobserved shocks) at all lags

and leads. In the next section, we discuss whether the identifying assumptions are satisﬁed in

the context of a government spending shock.

In general, this framework can be used to study the propagation of any shock through systematic

U.S. monetary policy. Our framework permits revisiting a range of important empirical ques-

tions, such as the role of systematic monetary policy for the eﬀects of oil-related shocks (e.g.,

Bernanke et al., 1997; Kilian and Lewis, 2011), technology shocks (e.g., Galí et al., 2003), news

ECB Working Paper Series No 2851

shocks (e.g., Barsky and Sims, 2011), ﬁscal spending shocks (e.g., Ramey and Zubairy, 2018),

and tax shocks (e.g., Romer and Romer, 2010). Moreover, our framework allows the estimation

of a new set of moments that can be used to discipline structural models with time variation

in systematic monetary policy, such as regime-switching models (e.g., Davig and Leeper, 2007;

Bianchi, 2013; Bianchi and Ilut, 2017).

4 Government spending and monetary policy

In this section, we use our identiﬁcation design to estimate how the eﬀects of U.S. government

spending shocks depend on systematic monetary policy. We ﬁnd that a hawkish FOMC signif-

icantly dampens the expansionary eﬀects of increased government spending on GDP, while

a dovish FOMC supports it. Relatedly, we ﬁnd sizeable diﬀerences in the ﬁscal multiplier

depending on the hawkishness or dovishness of the FOMC. We further provide evidence on the

strength of our instrument, and perform an extensive sensitivity analysis considering alternative

Hawk-Dove balances, an alternative spending shock, varying sample periods, and the inclusion

of additional control variables.

4.1 Data and identifying assumptions

We next discuss the data (in addition to Hawk

and Hawk

) and the identifying assumptions

for our analysis of government spending shocks.

Variables. We ﬁrst specify the local projection framework (3.4)-(3.5). Our baseline shock of

interest, ε

in (3.4), is the military spending shock constructed by Ramey (2011) and Ramey

and Zubairy (2018), based on a narrative approach to identify surprise build-ups (or build-

downs) in U.S. military spending. The shock is constructed as the present value of expected

changes in real defense spending over the next years, typically up to a horizon of ﬁve years,

and expressed relative to real potential GDP. The two outcome variables of interest, x

t+h

(3.4), are real GDP and real government spending, both expressed relative to real potential

GDP.

Finally, the vector of control variables, Z

in (3.4), includes four lags of real GDP and

real government spending, both relative to potential output and four lags of the ﬁscal spending

shock. If we restrict γ

= δ

= 0, our speciﬁcation of (3.4) corresponds to equation (1) of

Ramey and Zubairy (2018). This facilitates the comparability of our results with the literature.

Sample. Our baseline sample covers the period from 1960Q1 to 2014Q4, which is the longest

possible sample for which the Hawk-Dove balance and the ﬁscal spending shocks are available.

Our sample includes important military spending shocks, e.g., the Vietnam War, the Carter-

Reagan military buildup, and 9/11. On the other hand, our sample excludes WWII and the

Korean War which are important events in Ramey (2011) and Ramey and Zubairy (2018).

Detrending by potential GDP is the so-called Gordon and Krenn (2010) transformation. Compared to using

log variables, this avoids using an ex-post multiplication with the GDP/G ratio, which substantially varies over

time, to obtain the ﬁscal spending multiplier.

Ramey (2011) shows that excluding the Korean War renders military spending shocks a weak instrument

for contemporaneous government spending. In general, it is not surprising that military spending shocks are a

ECB Working Paper Series No 2851

the context of studying the response of monetary policy to ﬁscal spending shocks, however, it

may be desirable to exclude these events because monetary policy was less autonomous from

ﬁscal policy prior to the Treasury-Fed Accord in 1951. Between 1942 and 1951, the Fed was

constrained to support government bond prices by pegging short-term interest rates.

Identifying assumptions. Two key identifying assumptions are necessary for the causal

interpretation of the estimates of β

and γ

in (3.4).

The ﬁrst assumption is that military spending shocks are random shocks. In particular,

the distribution of military spending shocks does not depend on systematic monetary policy.

According to Ramey and Shapiro (1998) and Ramey and Zubairy (2018), military spending

shocks are unanticipated changes in spending plans triggered by geopolitical events and are

therefore exogenous to the economy. This argument similarly applies when conditioning on

systematic monetary policy. We provide three additional arguments as to why the military

spending shocks are independent of systematic monetary policy: (i) the response of military

spending to the shock does not depend on systematic monetary policy, see Section 5.2; (ii)

the news quotes used to construct military spending shocks as described in the supplementary

appendix to Ramey and Zubairy (2018) do not mention monetary policy, the Federal Reserve,

or the FOMC for our sample; and (iii) the Hawk-Dove balance does not predict future spending

shocks. The speciﬁc concern the last point addresses is that military spending shocks might be

timed to episodes with a more dovish FOMC. To test this concern we regress future military

spending shocks on Hawk

and use Hawk

as an instrument. We ﬁnd no signiﬁcant eﬀects of

the Hawk-Dove balance on contemporaneous or future military spending shocks, see Figure G.1

in Appendix H. If anything, we ﬁnd expansionary shocks when the FOMC is hawkish, inconsis-

tent with the timing hypothesis above.

The second assumption is that the FOMC rotation instrument is orthogonal to other macroe-

conomic shocks at all leads and lags. This is plausible for various reasons as discussed in

Section 3.2. More speciﬁcally, given that ﬂuctuations in Hawk

are relatively short-lived and

uncorrelated with real GDP growth, it is unlikely that our estimates capture diﬀerences in the

response across booms and busts (e.g., Auerbach and Gorodnichenko, 2012; Ramey and Zubairy,

2018). It is similarly unlikely that Hawk

correlates with changes in systematic ﬁscal policy,

which tends to be persistent.

4.2 GDP and government spending

We next present our empirical estimates of the causal eﬀects of systematic monetary policy on

the responses of real GDP and real government spending to ﬁscal spending shocks. We ﬁnd

that expansionary spending shocks raise GDP signiﬁcantly more strongly when the FOMC is

more dovish.

Baseline IV estimates. Figure 3 shows the responses of real GDP and real government

spending (G) to a military spending shock conditional on systematic monetary policy (Hawk

weak instrument for contemporaneous government spending because the shocks largely pertain to future spending.

Therefore, we do not use military spending shocks as an instrument but as shocks in our local projection framework

(3.4) and ﬁnd a signiﬁcant dynamic government spending response, see Section 4.2.

ECB Working Paper Series No 2851

Figure 3: Responses to spending shocks conditional on monetary policy

(a) Average GDP (β

) (b) Average G (β

)

) (d) Diﬀerential G (γ

)

(e) State-dependent GDP (β

± γ

) (f) State-dependent G (β

± γ

)

Notes: The ﬁgure shows responses of real GDP and real government spending (G) to an expansionary military spending

shock, corresponding to one percent of GDP, conditional on systematic monetary policy (Hawk

). We show IV estimates

based on the local projection framework (3.4)-(3.5) as speciﬁed in Section 4.1. The β

captures the responses when Hawk

equals its sample average. The γ

captures the diﬀerential responses when Hawk

exceeds the sample average by two

hawks. The β

± γ

shows the state-dependent responses when Hawk

exceeds the sample average either by two hawks

(+2 Hawks) or by two doves (+2 Doves). The shaded areas indicate 68% and 95% conﬁdence bands using Newey-West

standard errors.

The estimates are based on the local projection framework (3.4)-(3.5) as speciﬁed in Section 4.1.

The solid lines show the point estimates and the shaded areas indicate 68% and 95% conﬁdence

ECB Working Paper Series No 2851

bands using Newey-West standard errors.

All estimates of β

and γ

are normalized to

correspond to an expansionary shock that raises the expected present discounted value of future

military spending by one percent of GDP.

Panels (a) and (b) show the IV estimates of β

for GDP and G, which capture the responses

when Hawk

equals its sample average. The average responses of both GDP and G are positive

and signiﬁcantly diﬀerent from zero at most horizons beyond the ﬁrst year. Both responses

build up gradually and exceed 0.15% for GDP and 0.11% for G after one year.

Panels (c) and (d) show the estimates of γ

, which capture the diﬀerential responses of GDP

and G when the FOMC exceeds the average Hawk-Dove balance by two hawks. Speciﬁcally, γ

is scaled to capture an increase in (Hawk

− Hawk) of 2/12. This means, for example, that

two FOMC members with unknown preferences are replaced by two consistent hawks, or that

two FOMC members swing from dovish to hawkish. An increase in Hawk

by 2/12 exceeds

one standard deviation of the change in Hawk

which is 0.15. Importantly, the GDP response

is lower after a ﬁscal expansion when the FOMC is more hawkish. This eﬀect is statistically

signiﬁcant at the 5% level until three years after the shock. The estimated magnitudes are

sizeable. Between two and three years after the shock the GDP response is more than 0.4%

lower under a more hawkish FOMC. Conversely, the GDP response is 0.4% higher when there

are two more doves in the FOMC. The diﬀerential response of government spending (G) is also

negative at horizons until three years after the shock, albeit smaller in absolute terms and less

signiﬁcant. In Section 5.2, we show that the diﬀerential G response is driven by non-military

spending, whereas we ﬁnd no meaningful diﬀerential eﬀect for military spending.

Panels (e) and (f) of Figure 3 show β

± γ

, the state-dependent responses when Hawk

exceeds the sample average either by two hawks (+2 Hawks) or by two doves (+2 Doves). The

GDP response strongly varies between the dovish and the hawkish FOMC. The dovish FOMC

supports the GDP expansion while the hawkish FOMC undoes the GDP expansion. Quantita-

tively, GDP increases by up to 0.68% under the dovish FOMC, but falls by up to 0.35% under

the hawkish FOMC. The former response is highly statistically signiﬁcant, whereas the latter

response is less precisely estimated.

Overall, our evidence suggests that monetary oﬀset of ﬁscal spending shocks is not a constant

feature of monetary policy but varies strongly with the Hawk-Dove balance in the FOMC.

In contrast to the GDP response, government spending displays smaller and less signiﬁcant

diﬀerences in the state-dependent responses.

Comparison with OLS. We compare our IV estimates presented above with the OLS coun-

terparts. Figure 4 shows the OLS and IV estimates of cumulative GDP responses to a military

shock as a function of the FOMC’s Hawk-Dove balance. At horizons of around one year, the

OLS estimates substantially understate the dependence of the GDP response on the Hawk-Dove

balance. In contrast, at long horizons of around four years, the OLS bias seems negligible.

For the Newey-West standard errors, we set the bandwidth to h + 1, where h is the horizon in (3.4). A

truncation parameter rule (Lazarus et al., 2018) or automatic bandwidth selection leads to similar results.

Normalizing the responses to a shock size of 1% of GDP approximately normalizes to one standard deviation

of the shock series, which is 1.17% of GDP.

Figure G.3 in the Appendix presents the cumulated responses at intermediate horizons of two and three

years. Figure G.2 presents the OLS estimates of β

and γ

ECB Working Paper Series No 2851

Figure 4: Cumulative GDP responses for OLS and IV

(a) 1-year response (b) 4-year response

Notes: The ﬁgure shows the cumulative real GDP response to an expansionary military spending shock, corresponding to

one percent of GDP, conditional on systematic monetary policy (Hawk

). We show IV and OLS estimates based on the

local projection framework (3.4)-(3.5) as speciﬁed in Section 4.1. The displayed estimates are computed as

h=0

[β

(Hawk

− Hawk

)] for H = 4 quarters (Panel a) and H = 16 quarters (Panel b).

This comparison suggests that ignoring the endogeneity of Hawk

leads to biased conclusions

about the role of systematic monetary policy for ﬁscal spending shocks.

4.3 Weak instruments

A common concern with IV estimates is the strength of the instrument. We provide evidence

supporting the strength of our instruments, including weak instrument tests and weak instrument-

robust inference, reinforcing the contribution of our identiﬁcation design.

First-stage results. Our local projection framework (3.4) contains two endogenous regres-

sors, ε

(Hawk

− Hawk) and (Hawk

− Hawk). The estimates of the two associated ﬁrst-stage

regressions are shown in Table G.1 in the Appendix. We ﬁnd that the instrumental variable

(Hawk

− Hawk

) has a positive eﬀect on the endogenous variable ε

(Hawk

− Hawk)

and is signiﬁcant at the one percent level. Similarly, (Hawk

− Hawk

) has a positive and

highly signiﬁcant eﬀect on (Hawk

− Hawk). In both regressions, the R

increases by about 0.4

when including the instruments as regressors and we also ﬁnd large jumps up in the associated

Kleibergen-Paap F-statistics. Taken together, this suggests that our instruments are strong

(Bound et al., 1995).

Weak instrument tests. We further use three statistical tests to assess the strength of our

instrument more formally. First, we use the Montiel Olea and Pﬂueger (2013) test of weak

instruments, which is popular in time series settings because it is robust to autocorrelation

and heteroskedasticity. Formally, we test whether the relative weak instrument bias for the IV

estimates of γ

exceeds 10%, 20%, or 30%.

Panel (a) of Figure 5 shows the p-values of the

We apply the test to γ

because it is our main coeﬃcient of interest (together with β

), and because the

Montiel Olea and Pﬂueger (2013) test can only be applied to a single endogenous regressor. For the other

endogenous regressor, (Hawk

− Hawk) in levels, we estimate the ﬁrst stage separately and plug in the ﬁtted

values in the second stage used to test the interaction term. If we alternatively replace Hawk

level term by

ECB Working Paper Series No 2851

Figure 5: Weak instrument tests

(a) Montiel Olea and Pﬂueger (2013) (b) Lewis and Mertens (2022)

Notes: The ﬁgure shows p-values for rejecting the null of weak instruments for the responses of real GDP, based on the

local projection framework (3.4)-(3.5) as speciﬁed in Section 4.1. The Montiel Olea and Pﬂueger (2013) test evaluates the

null of the bias in γ

exceeding a threshold τ. Similarly, the Lewis and Mertens (2022) test evaluates the null of the ℓ

norm of the bias in γ

and δ

exceeding a threshold τ . For the former, the endogenous regressor Hawk

is not tested

but directly replaced by its ﬁrst stage ﬁtted value. The critical values and associated p-values are based on Newey-West

standard errors.

weak instrument tests for the diﬀerential GDP response. At all horizons, even a relatively small

10% bias (τ = 0.1) can be rejected at signiﬁcance levels below 2%.

The second weak instrument test we apply was recently developed by Lewis and Mertens (2022)

and generalizes Montiel Olea and Pﬂueger (2013) to allow for multiple endogenous regressors.

We apply this test to jointly evaluate whether the average relative bias across γ

and δ

exceeds

some threshold τ and report the results in Panel (b) of Figure 5. A small average bias of 10%

can be rejected at signiﬁcance levels below 10% for most horizons. Moreover, we can reject a

bias of 20% at the two percent level for all horizons. For government spending, both tests lead

to the same conclusion, see Figure G.4 in Appendix G.

Lastly, we test for weak instruments via the reduced form of our regression framework. Following

Chernozhukov and Hansen (2008), the hypothesis test of the reduced form estimates of γ

against zero is equivalent to testing whether the instrument has zero relevance. Figure G.5 in

the Appendix shows that the reduced-form estimates for γ

are signiﬁcant, as in Figure 3. To

summarize, all three tests indicate that our instruments are not weak.

Weak instrument-robust inference. To address residual concerns about instrument strength,

we further provide inference that is robust to weak instruments and allows for multiple endoge-

nous regressors based on Andrews (2018). We ﬁnd robust conﬁdence sets for the diﬀerential

GDP and G responses similar to our baseline intervals, see Figure G.6 in the Appendix. This

provides additional support for the strength of our instruments.

4.4 Fiscal spending multiplier

A key object for the design and evaluation of ﬁscal policies is the ﬁscal spending multiplier. We

use our framework to estimate how the ﬁscal spending multiplier depends on the hawkishness

Hawk

we obtain very similar results.

ECB Working Paper Series No 2851

of the FOMC. We ﬁnd that a dovish FOMC leads to substantially larger multipliers, relative

to an average or a more hawkish FOMC composition.

Deﬁnition and estimation. The multiplier is deﬁned as the dollar amount by which GDP

increases per dollar increase in ﬁscal spending (both in real terms). A common procedure is

to compute the multiplier as the cumulative response of GDP to a spending shock divided

by the cumulative response of government spending to the same shock over some horizons of

interest (e.g., Mountford and Uhlig, 2009; Ramey and Zubairy, 2018). To study how systematic

monetary policy shapes the ﬁscal multiplier, we deﬁne the monetary policy-dependent ﬁscal

multiplier as

F M

(χ) =

h=0

GDP

+ γ

GDP

h=0

+ γ

(4.1)

where H is the maximal considered forecast horizon, β

and γ

are the average and diﬀerential

responses of outcome i ∈ {GDP, G} to a spending shock, and χ indicates some level of the

Hawk-Dove balance in deviation from the sample mean (Hawk

− Hawk).

We estimate the cumulative average and diﬀerential responses,

h=0

and

h=0

, in one

step by replacing the left-hand side of the local projection in (3.4) by the cumulative outcome

between h = 0 and H. Otherwise, we exactly follow Section 4.2 and use the speciﬁcation of

the local projection framework in Section 4.1. We construct valid standard errors for F M

(χ)

by accounting for the covariance between the estimates in the numerator and denominator of

(4.1). Appendix F provides further details and a comparison of our multiplier estimation with

the approach in Ramey and Zubairy (2018).

Results. Table 2 presents the IV estimates of the ﬁscal spending multipliers F M

(χ) for

both a two-year and a four-year horizon. For an average Hawk-Dove balance, χ = 0, the

cumulative spending multiplier is 1.3 at both horizons, and signiﬁcantly diﬀerent from zero at

the 10% level. Analogous to Figure 3, we consider a range of χ from −2/12 to +2/12. As

the FOMC becomes more dovish than average, the multiplier increases from 1.3 to 2.3 for one

additional dove (χ = −1/12), and to 3 for two additional doves (χ = −2/12). The diﬀerence

between the average and the dovish multipliers are similar across the two horizons. Moreover,

the diﬀerence is statistically signiﬁcant at the 5% level for the four-year horizon, see Table G.3

in Appendix G. Conversely, as the FOMC becomes more hawkish, the multiplier F M

(χ) drops

to zero or below and is insigniﬁcantly diﬀerent from zero. The diﬀerences in F M

(χ) across χ

are mainly driven by diﬀerences in the cumulative GDP response rather than the G response.

The diﬀerences in the GDP response across χ are larger in magnitude and more signiﬁcant, see

Table G.2. This result is analogous to the ﬁndings in Figure 3.

Comparison with linear model. We explicitly estimate how the ﬁscal spending multiplier

depends on systematic monetary policy, whereas much of the related literature has estimated a

single ‘average’ ﬁscal spending multiplier (e.g., Blanchard and Perotti, 2002; Ramey, 2016). To

compare our results with this tradition in the literature, we estimate an average ﬁscal spending

ECB Working Paper Series No 2851

Table 2: Government spending multipliers and monetary policy

Baseline model Linear

Outcome +2 Hawks +1 Hawk Average +1 Dove +2 Doves model

Two-year horizon

Multiplier -4.825 -0.476 1.348 2.351 2.986 0.860

(5.229) ( 1.418) (0.708) (0.934) (1.239) (1.427)

GDP (cum) -1.689 -0.282 1.124 2.531 3.937 0.616

(0.989) (0.768) (0.649) (0.689) (0.865) (1.057)

G (cum) 0.350 0.592 0.834 1.076 1.319 0.716

(0.250) (0.300) (0.395) (0.510) (0.634) (0.338)

Four-year horizon

Multiplier -1.790 -0.001 1.308 2.307 3.095 0.838

(2.637) (0.862) (0.475) (0.808) (1.162) (1.449)

GDP (cum) -2.735 -0.002 2.731 5.465 8.198 1.494

(2.498) (1.557) (0.842) (1.045) (1.892) (2.747)

G (cum) 1.528 1.808 2.088 2.368 2.649 1.782

(1.010) (0.804) (0.734) (0.848) (1.079) (0.689)

Notes: The table shows IV estimates of the cumulative ﬁscal spending multipliers F M

(χ) in equation (4.1) for H = 8

(top panel) and H = 16 quarters (bottom panel), as well as the cumulative GDP response (numerator of F M

(χ)) and the

cumulative G response (denominator of F M

(χ)). The coeﬃcients are estimated using a cumulative version of the local

projection framework (3.4)-(3.5) as speciﬁed in Section 4.1. For our baseline model, the columns present diﬀerent states of

the Hawk-Dove balance between “+2 Hawks” (χ = +2/12), “Average” (χ = 0), and “+2 Doves” (χ = −2/12). The linear

model in the last column presents the estimates when we restrict γ

= δ

= 0 in the local projection (3.4). Driscoll-Kraay

standard errors are in parenthesis, see Appendix F for details.

multiplier in a linear version of our framework. To be precise, we estimate (3.4) using the same

data but restricting γ

= δ

= 0. We then use the estimates of β

from the linear model

to compute the ﬁscal multiplier

F M

= (

h=0

GDP

)/(

h=0

). The resulting estimates

are presented in the last column of Table 2. We ﬁnd average multipliers of about 0.85 at both

horizons. While this estimate is relatively close to the multiplier estimates in Ramey and Zubairy

(2018) which range from 0.66 to 0.71 (see their Table 1), it is substantially below the multiplier

of 1.3 for an average FOMC composition (F M

(0)) in our baseline model. In addition, the

standard errors for the multiplier in the linear model are substantially larger than the standard

errors of F M

(0). This comparison suggests that accounting for systematic monetary policy

is important for the magnitude and precision of multiplier estimates. Moreover, one potential

reason for the broad range of multiplier estimates in the literature is not accounting for time

variation in systematic monetary policy.

ECB Working Paper Series No 2851

4.5 Sensitivity analysis

In this section, we provide an extensive sensitivity analysis to assess the robustness of our base-

line results. We investigate alternative Hawk-Dove balances, an alternative spending shock,

varying sample periods, and the inclusion of additional control variables. The multiplier esti-

mates for all speciﬁcations are provided in Appendix G.

Alternative Hawk-Dove balances. We address potential concerns regarding the aggrega-

tion of individual policy preferences and the comparability of preferences over time.

While our baseline Hawk

aggregates individual preferences by an unweighted arithmetic average,

we consider four alternative aggregation schemes. First, we use the median policy preference

across FOMC members. Second, we use an arithmetic average but double the weight of the

Fed Chair. Third, we use the arithmetic average but do not distinguish between consistent and

swinging FOMC members when deﬁning Hawk

iτ

in (3.1). We estimate impulse responses and

multipliers similar to the baseline, albeit smaller ones for the median aggregation, see Table G.3.

In a fourth alternative aggregation, we consider the role of strong majorities in the FOMC. We

construct an alternative Hawk-Dove balance which equals -1 if Hawk

falls below the ﬁrst

quartile or tertile of the distribution of Hawk

over time, +1 above the highest quartile or

tertile, and zero otherwise. The estimated average and diﬀerential eﬀects remain quite similar

in terms of the shapes and signiﬁcance and also roughly align with the baseline multipliers,

see Table G.4. Finally, we construct an alternative instrument by setting the preferences of

swinging FRB presidents to zero before aggregating them to the FOMC rotation instrument.

This yields somewhat larger multiplier estimates, albeit less precisely estimated, see Table G.3.

This suggests that swings in the instrument are not driving or amplifying our results.

Another potential concern is that the meaning of being a hawk or dove might have changed

over time, see the discussion in Section 3.1. To account for trends in the Hawk-Dove balance,

we consider an alternative Hawk-Dove balance which subtracts from the baseline Hawk

its

backward-looking 5, 10, or 15-year moving average. The estimated average and diﬀerential

responses are very similar to our baseline estimates. In addition, the average and dovish multi-

pliers have similar magnitudes as the baseline while the hawkish multiplier is similarly imprecise,

see Table G.3 in the Appendix. Overall, our results reinforce the arguments in Section 3.1 that

the classiﬁcation of hawks and doves is indeed comparable over time.

Alternative spending shock. Our baseline shock is speciﬁc to military spending. We inves-

tigate the external validity of our results by using an alternative ﬁscal spending shock, which

is identiﬁed from a timing restriction on total government spending as suggested by Blanchard

and Perotti (2002), henceforth BP. They assume that only government spending shocks can

aﬀect government spending contemporaneously.

We ﬁnd that GDP and G respond more swiftly compared to our baseline. This is in line with

the nature of the BP shock. More importantly, we ﬁnd that a hawkish FOMC signiﬁcantly

dampens the expansionary eﬀect on GDP. The average ﬁscal multiplier is around 1.4 for the

For the impulse responses of GDP and G associated to the sensitivity analysis, see Appendix G of the CEPR

Discussion Paper version (Hack et al., 2023, see https://cepr.org/publications/dp17999).

ECB Working Paper Series No 2851

four-year horizon, see Table G.3, which is remarkably similar to our baseline multiplier. The

state-dependent multiplier ranges from 0.88 to 1.74 between the hawkish and dovish FOMC

(χ = ±2/12). While the variation in the multiplier is more compressed compared to the

baseline, it is similarly signiﬁcant.

Great Recession and ZLB. Our baseline results are estimated using the sample from

1960Q1 to 2014Q4 which includes the Great Recession (GR) and the subsequent ZLB period.

We investigate the sensitivity of our results on a sample that ends either in 2007Q4 to exclude

the GR and ZLB period or in 2008Q4 to exclude the ZLB period. For both of these subsamples,

our estimates are highly similar to the baseline for average and diﬀerential responses and the

corresponding multipliers in Table G.3.

Additional control variables. Finally, we investigate the sensitivity of our results to adding

potentially important co-variates to the baseline speciﬁcation of our local projection framework.

The additional control variables are short-term and long-term interest rates, inﬂation, and the

primary surplus. While the estimates are similar to the baseline, we naturally give up some

statistical power, see Table G.3. Nevertheless, we estimate dovish multipliers around 2 which

substantially exceeds the average multiplier, consistent with our baseline results. We further

add non-linear controls by including interactions of Hawk

with the control variables. The

results are remarkably close to the baseline, see Table G.3.

5 Inspecting the mechanism

In this section, we inspect the mechanism behind our ﬁndings in the previous section. We show

that in response to an expansionary spending shock, nominal and real interest rates rise and

inﬂation is dampened under a hawkish FOMC. Conversely, interest rates initially fall and rise

only with substantial delay under a dovish FOMC, supporting a crowd in of consumption

and investment and an increase in non-military government spending. We argue that the

ﬁscal multiplier estimates across hawkish and dovish monetary regimes are plausible given the

diﬀerent interest rate responses.

5.1 Interest rates and inﬂation

Conventional wisdom says that monetary policy tightens in response to higher government

spending in order to mitigate the inﬂationary pressure. The Federal Reserve can use a range

of tools, including the target federal funds rate, the discount rate, balance sheet policies and

communication including forward guidance. These tools can aﬀect short- and long-term interest

rates, and hence inﬂation.

Nominal interest rates. We study the response of the federal funds rate (FFR) and the

annualized yield on 1-year and 10-year Treasury securities to government spending shocks by

using our local projection framework (3.4)-(3.5) with interest rates as outcome variable x

t+h

We follow the speciﬁcation in Section 4.1 but include four lags of the FFR, 1-year and 10-year

ECB Working Paper Series No 2851

Figure 6: Responses of nominal interest rates

(a) Average FFR (β

) (b) State-dependent FFR (β

± γ

)

) (d) State-dependent 1-year rate (β

± γ

)

(e) Average 10-year rate (β

) (f) State-dependent 10-year rate (β

± γ

)

Notes: The ﬁgure shows responses of the federal funds rate (FFR), as well as the 1-year and 10-year treasury yields to an

expansionary military spending shock, corresponding to one percent of GDP, conditional on systematic monetary policy

(Hawk

). All outcomes are annualized interest rates. We show IV estimates based on the local projection framework

(3.4)-(3.5) as speciﬁed in Section 5.1. The β

captures the responses when Hawk

equals its sample average. The β

± γ

shows the state-dependent responses when Hawk

exceeds the sample average either by two hawks (+2 Hawks) or by two

doves (+2 Doves). The shaded areas indicate 68% and 95% conﬁdence bands using Newey-West standard errors.

Treasury yields, and CPI inﬂation as additional control variables to control for pre-trends in

these outcomes.

Panels (a), (c) and (e) of Figure 6 show the IV estimates of β

, the average response of the

three nominal interest rates when Hawk

equals its sample average. The average FFR response

ECB Working Paper Series No 2851

appears muted in the ﬁrst year, after which it gradually increases and reaches 30 basis points

at horizons beyond two years. The average responses of the 1-year and 10-year yields feature

similar shapes, albeit at lower magnitudes. Panels (b), (d) and (f) show the IV estimates of

± γ

, the state-dependent interest rate responses when Hawk

exceeds the sample average

either by two hawks (+2 Hawks) or by two doves (+2 Doves). All interest rates increase faster

and more strongly under a hawkish FOMC. Compared to the average response, the peak in

the FFR is reached one year earlier and is almost double in size (about 56 basis points). In

contrast, under a dovish FOMC, the FFR falls for almost two years and a reversion to a higher

FFR is observed only three years after the shock. Similarly, both 1-year and 10-year Treasury

yields increase after two years under a dovish FOMC, suggesting that the monetary regimes

also diﬀer in their eﬀects on expected future policy at long horizons.

The delayed FFR response is consistent with the initial uncertainty surrounding the military

spending shock and the gradually evolving macroeconomic eﬀects of the shock, see Figure 3.

Section 6 provides narrative evidence from the FOMC historical records suggesting that indeed

the FOMC delays action until some uncertainty about the spending plans and their potential

eﬀect on the economy and inﬂation is resolved. Furthermore, a delayed diﬀerential policy

response that extends for several quarters beyond the term of the FOMC and the associated

rotation present at the time of the shock, is consistent with the decision dynamics in the FOMC.

For example, Laurence Meyer, member of the Board of Governors from 1996 to 2002, describes

these dynamics during his term at the Fed as follows:

So was the FOMC meeting merely a ritual dance? No. I came to see policy deci-

sions as often evolving over at least a couple of meetings. The seeds were sown at

one meeting and harvested at the next. [...] Similarly, while in my remarks to my

colleagues it sounded as if I were addressing today’s concerns and today’s policy deci-

sions, in reality I was often positioning myself, and my peers, for the next meeting.

Laurence Meyer (2004), A Term at the Fed: An Insider’s View, Harper Business

Consistent with Meyer’s view that it takes time to inﬂuence policy strategies in the FOMC,

we ﬁnd that the FOMC rotation (Hawk

) is more important for the policy response to the

spending shock and its real eﬀects when the shock occurs closer to the beginning of the FOMC

rotation, which takes place in the ﬁrst quarter of the year. When we drop spending shocks in

the second half of the year, we obtain similar ﬁndings compared to the baseline, see Figures H.1-

H.2 in Appendix H. Conversely, the dependence on monetary policy becomes weaker and less

signiﬁcant when dropping spending shocks in the ﬁrst half of the year.

Inﬂation rates. We further asses the eﬀects of military spending on inﬂation expectations,

CPI core inﬂation (excluding food and energy prices), and CPI headline inﬂation.

We esti-

mate the inﬂation responses using the speciﬁcation of our local projection framework (3.4)-(3.5)

for nominal interest rates and additionally control for four lags of the inﬂation measure under

consideration. Overall, the inﬂation responses are not precisely estimated. The average response

We use one-year inﬂation expectations based on the CPI forecasts from the Livingston Survey of the Federal

Reserve Bank of Philadelphia. It is the oldest continuous survey on the expectations of economists from industry,

government, banking, and academia. For details, see Appendix B.

ECB Working Paper Series No 2851

of expected inﬂation tends to be positive, while the evidence is mixed for core and headline inﬂa-

tion. Turning to the dependence on the Hawk-Dove balance, we ﬁnd that inﬂation expectations

increase sluggishly under a dovish FOMC and peak at about three years. In contrast, inﬂation

expectations tend to fall under a hawkish FOMC, suggesting that the FOMC is successful in

containing inﬂation expectations. The response of core inﬂation follows a similar but even more

sluggish pattern, suggesting that policy tightening is successful in containing inﬂationary pres-

sures. Compared to the interest rate responses, the inﬂation response appear delayed by one

to two years, broadly in line with the lags in the transmission of monetary policy. Finally, the

results for headline inﬂation are more mixed, possibly due to larger transitory ﬂuctuations in

energy and food prices.

Real interest rates. In a large class of models, the real eﬀects of monetary policy depend

on its ability to aﬀect real interest rates. Under a hawkish FOMC, the response of nominal

rates is larger, while the response of inﬂation is smaller. Hence, the implied response of real

interest rates is larger. In response to a government spending shock, real interest rates increase

by more if the FOMC is hawkish and by less if the FOMC is dovish. We obtain similar

results when directly estimating the real interest rate response. We consider real interest rates

constructed by subtracting the expected CPI inﬂation from the three nominal interest rates

considered in Figure 6. Figure H.3 in Appendix H presents the IV estimates of the average and

state-dependent responses.

Relation to ﬁscal multipliers in the literature. The interest rate responses allow us

to relate our ﬁscal spending multiplier estimates in Table 2 with the ﬁndings in the related

literature. Our spending multiplier is between two and three under the dovish FOMC which

is associated with a weak negative response of the nominal (and real) FFR for the ﬁrst two

years. In theory, the multiplier may be far above one (or negative) depending on the response

of interest rates (Woodford, 2011; Farhi and Werning, 2016). In an estimated medium-scale

DSGE model, Christiano et al. (2011) ﬁnd multipliers between two and four at the ZLB when

the short-run nominal interest rate does not respond.

Our ﬁndings also relate to an empirical literature that estimates ﬁscal spending multipliers.

For example, Nakamura and Steinsson (2014) estimate two-year regional multipliers for the

U.S. of approximately 1.5. To the extent that regional multipliers correspond to the aggregate

multiplier when nominal interest rates do not respond, we can compare their estimates to our

two-year multiplier estimates. In particular, we construct a spending multiplier for the case in

which the nominal FFR is unresponsive by choosing the Hawk-Dove balance (χ) that minimizes

the squared distance of the FFR response from zero in the ﬁrst two years.

This requires a

χ slightly below the “+1 Dove” case in Table 2. The associated two-year spending multiplier

is 1.9, which is similar to the range of estimates in Nakamura and Steinsson (2014). While

our identiﬁcation design allows us to estimate the spending multiplier when monetary policy is

non-responsive, it also allows us to study many other monetary policy scenarios.

Formally, we solve min

h=0

(β

F F R

− χ · γ

F F R

)

, where χ indicates a level of the Hawk-Dove balance in

deviation from the sample mean (Hawk

− Hawk).

ECB Working Paper Series No 2851

Figure 7: Responses of inﬂation rates

(a) Average expected inﬂation (β

) (b) State-dependent expected inﬂation (β

± γ

)

) (d) State-dependent core inﬂation (β

± γ

)

(e) Average headline inﬂation (β

) (f) State-dependent headline inﬂation (β

± γ

)

Notes: The ﬁgure shows responses of expected inﬂation, CPI core, and CPI headline inﬂation to an expansionary military

spending shock, corresponding to one percent of GDP, conditional on systematic monetary policy (Hawk

). All outcomes

are annualized inﬂation rates. We show IV estimates based on the local projection framework (3.4)-(3.5) as speciﬁed in

Section 5.1. The β

captures the responses when Hawk

equals its sample average. The β

± γ

shows the state-dependent

responses when Hawk

exceeds the sample average either by two hawks (+2 Hawks) or by two doves (+2 Doves). The

shaded areas indicate 68% and 95% conﬁdence bands using Newey-West standard errors.

We further compare our estimates with the estimate of the aggregate spending multiplier when

monetary policy is constrained at the ZLB. Ramey and Zubairy (2018) ﬁnds a ZLB multiplier of

1.6 after two years (when excluding WWII), while Miyamoto et al. (2018) ﬁnd a ZLB multiplier

well above 1.5 for Japan. Notwithstanding the endogeneity of a binding ZLB, our multiplier

ECB Working Paper Series No 2851

of 1.9 under a non-responsive FFR is similar to the ZLB multipliers in the literature. Overall,

our multiplier estimates and the associated interest rate path are broadly similar to previous

quantitative and empirical ﬁndings.

5.2 Decomposing the GDP and G responses

In the following, we examine the underlying components of the responses of real GDP and real

government spending. We ﬁnd that the diﬀerential GDP eﬀects are primarily driven by private

consumption and somewhat less by private investment. Moreover, the diﬀerential government

spending response is almost entirely driven by non-military expenses, consistent with our iden-

tifying assumptions.

Investment and consumption. The ﬁscal spending multiplier can be above one when GDP

components other than G are crowded in by the spending shock. Conversely, crowding out may

lead to multipliers below one. Therefore, we estimate the responses of private investment and

private consumption.

The results are provided in Figure H.4 in Appendix H. For the average

Hawk-Dove balance, we ﬁnd a mild but insigniﬁcant crowding out of private consumption and

crowding-in of private investment in the short run. In contrast, the crowding out of consumption

is strong and signiﬁcant under a hawkish FOMC. For investment, we ﬁnd a similar albeit

smaller and less signiﬁcant pattern. Overall, the strong state-dependence of our ﬁscal multipliers

appears to be mainly driven by private consumption.

Military and non-military government spending. We further estimate the respective

responses of military and non-military government spending to the military spending shock, see

Figure H.5 in Appendix H. On average, we ﬁnd a modest and insigniﬁcant decline in non-military

spending. However, the government cuts non-military spending strongly and signiﬁcantly under

a hawkish FOMC. Fiscal policy responds to higher interest rates by lowering non-military

spending, consistent with higher costs of servicing federal debt and lower tax revenues. This

further contributes to the negative GDP response under a hawkish FOMC in addition to the

contributions by consumption and investment. For military spending, we ﬁnd the opposite.

While the average response is large and signiﬁcant, there is no meaningful dependence on the

Hawk-Dove balance for military expenditures. This suggests that the shock is unrelated to

systematic monetary policy, corroborating our key identifying assumptions, as discussed in

Section 4.1.

6 Historical FOMC records

Interviewer: What would have happened, do you think, if the Fed had not raised

the discount rate?

Chairman Martin: A golden opportunity to stop inﬂation in its tracks would have

been lost.

Interviewer: It was primarily the projection of Vietnam spending; is that correct?

For details on the deﬁnition of consumption and investment, see Appendix B.

ECB Working Paper Series No 2851

Chairman Martin: Right. I kept telling him we could not have guns and butter.

Interviewer: When you talked to Lyndon Johnson about this projection, what did

he say? Did he disagree with it or did he agree with it?

Chairman Martin: He disagreed. He thought we could have guns and butter.

We complement our quantitative analysis with narrative evidence from the records of discussions

and decisions at FOMC meetings. This evidence serves two purposes. First, it conﬁrms that

the FOMC members discuss changes in government defense spending, assessing the impact on

economic activity and inﬂation as well as the FOMC’s policy response. Second, it shows that

the policy response depends on the composition of the FOMC.

To illustrate the FOMC discussion around military spending shocks, the FOMC composition,

and the corresponding policy response, we focus on two important events during the 1960s:

the acceleration of the U.S. Space Program in 1961 and the Vietnam ground war starting in

1965. The corresponding military shocks are both large while the FOMC composition appears

on average hawkish in the ﬁrst part of the 1960s and dovish in the second part, see Figure 2.

In this period, the Fed was headed by William McChesney Martin, a consistent hawk whose

tenure as chairman from 1951 to 1970 was the longest in history.

For both events, we identify three phases of the FOMC’s reaction to military defense spending

from the historical FOMC records. First, there is uncertainty about the extent to which the

spending plans will be realized and about their impact on the economy. Second, the eﬀects of

higher spending on the economy become visible while inﬂation appears unresponsive, therefore

they wait until “all the evidence was in”. Third, the eﬀects on inﬂation become visible but the

FOMC delays action. The ﬁrst two are common for hawkish and dovish committees while the

third phase is more pronounced under a dovish one, broadly in line with our empirical ﬁndings.

We summarize the key aspects of each case study below.

The sources for our narrative evidence

are the FOMC Historical Minutes until 1967 and the Memoranda of Discussion thereafter.

6.1 The U.S. Space Program

In the ﬁrst half of 1961, Ramey and Zubairy (2018) identify two expansionary shocks related to

President Kennedy’s defense spending plans, including the Space Program to “go to the Moon”.

In the FOMC meeting of August 1, 1961, the staﬀ presents the following assessment:

On top of substantial increases in expenditures to ﬁnance space exploration and

longer-run defense measures [...] the President has found it necessary to recommend

an increase of $3-1/2 billion in current defense expenditures [...]. More important,

the President accompanied his recommendations with a very ﬁrm statement regarding

his intentions with respect to the 1963 budget. These factors have certainly tended

to minimize the immediate inﬂationary expectations and the urgency of the need for

counter-measures. As of this moment in time, actual developments do not seem to

call for any change in monetary policy. (p.8)

Former Fed Chairman William McChesney Martin: Oral History, Interview I by Michael L. Gillette in 1987,

LBJ Library Oral History Collection. The interviewer refers to the decision of the Federal Reserve to raise the

discount rate on December 1965. Lyndon B. Johnson was the President of the United States from 1963 to 1969.

The complete case studies can be found in Appendix I of the CEPR Discussion Paper version (Hack et al.,

2023, see https://cepr.org/publications/dp17999).

ECB Working Paper Series No 2851

The majority of the FOMC members argued similarly for no change in policy because the eﬀects

could not yet be evaluated. Hawkish FOMC members suggested the need for alertness to avoid

getting into an inﬂationary situation while agreeing to no policy change in this meeting. In this

regard, New York Fed ﬁrst-vice president, William Treiber noted: If expenditures and related

private spending result in an upsurge of activity with inﬂationary aspects, we may have to modify

our policy of basic monetary ease sooner than we would otherwise have done. In the coming

period undue ease should be avoided. (p.22-23)

FOMC members started to acknowledge the expansionary impact on employment and business

sentiment in defense-related industries by the end of 1961 and later in 1963 on prices. On

May 7, 1963, the FOMC voted to ﬁrm policy as a preemptive move against inﬂation.

In this

meeting, Chairman Martin said:

If the Committee waited too long, however, it might have to deal with an active

problem of inﬂationary pressures. In his opinion, there was already a good bit of

pressure in some areas that could build up rapidly. If one waited until after the

resulting price movements actually occurred, he might wonder why he had not done

something about it before. It would be too late at that juncture. (p.61)

In this period, the FOMC composition was hawkish on average. This helped the hawkish

Chairman Martin to reach a consensus for tighter policy to act preemptively against inﬂationary

pressures.

6.2 The Vietnam War

In 1965, the U.S. entered the ground war in Vietnam leading to a series of expansionary military

spending shocks lasting until 1967Q1. In the FOMC meeting of August 10, 1965, the staﬀ’s

presentation explicitly accounted for the intended increase of military spending:

Further stimulus to the economy will come from expanded Government procure-

ment for Vietnam hostilities. [...] the increases in spending and in the armed forces

now proposed do not appear signiﬁcant enough to touch oﬀ [...] widespread price

increases. [...] The market response to Vietnam developments doesn’t suggest any

widespread fears of shortages, rationing, or inﬂation. On balance, then, the domestic

evidence isn’t clear enough to me to justify a signiﬁcant policy move in either direc-

tion at this juncture. (p. 28-29).

Several FOMC members agreed with the staﬀ’s assessment and argued for an unchanged policy

due to signiﬁcant uncertainties related to the developments in Vietnam. In contrast, few hawkish

FOMC members noted that the Vietnam hostilities were already aﬀecting industrial prices.

Two meetings later, on September 28, the dovish members dissented against the “status quo”,

arguing that, in their judgment, evidence of inﬂationary pressure was lacking and hence, they

preferred an easier policy. In contrast, Alfred Hayes (New York Fed), a hawk, argued in the

The FOMC shifted the emphasis of monetary policy toward slightly less ease and toward maintaining a

moderately ﬁrm tone in the money market in June 1962, mentioning balance-of-payments concerns. In this

period, FOMC members interested in a tighter, inﬂation-focused monetary policy often cited the balance-of-

payments criterion to bolster their case (Bordo and Humpage, 2014).

ECB Working Paper Series No 2851

meeting of October 12, 1965 that: Looking ahead, I think we have a real basis for concern about

potential inﬂationary pressures (p.25). Chairman Martin shared similar thinking on inﬂation

while sensing that he did not have a majority to ﬁrm policy:

While the evidence was not clear, he thought there were many signs of inﬂation and

of inﬂationary psychology in the economy. [...] But the Committee had a tendency

to feel that it was best to wait until all the evidence was in before making a policy

change. The diﬃculty was that when all the evidence was in it was likely to be too

late. [...] With a divided Committee and in face of strong Administration opposition

he did not believe it would be appropriate for him to lend his support to those who

favored a change in policy now. (p.68-69)

On December 5, 1965, the discount rate was raised with a narrow majority in order to prevent

the risk of inﬂation. However, the tightening signal by the Fed was not enough to contain

the buildup of inﬂationary pressures. While this had become clear for most members, the

U.S. President had promised an anti-inﬂationary ﬁscal program and the FOMC delayed action

in support of promised ﬁscal restraint. On September 13, 1966, Governor James Robertson

summarized the situation as follows: Inﬂationary pressures are persisting, as the staﬀ materials

have underlined. [...] To counter these inﬂationary pressures, we now have the promise of help

from a somewhat greater degree of ﬁscal restraint. (p.72).

Hoping on the legislative action to raise taxes in 1967, by the last quarter of 1966 and throughout

the ﬁrst part of 1967, the FOMC eased policy, despite two large expansionary military spending

shocks hitting in 1966Q4 and 1967Q1. In the FOMC meeting of September 12, 1967, Chairman

Martin acknowledged that tightening had been delayed for too long because of the tendency

to underestimate the strains being put on economic resources by the hostilities in Vietnam. A

“guns and butter” economy was not feasible; the country’s resources were not suﬃcient for that.

(p.73). The FOMC decided to tighten the policy on December 12, 1967. Once again, Chairman

Martin admitted delayed action as follows:

It was his feeling that the Committee had in a sense been caught in a trap [...]

From the standpoint of economic considerations alone, it would have been desirable

to adopt a ﬁrmer monetary policy a number of months ago. (p.96)

In the period between 1965 and 1967, the FOMC is categorized as dovish on average. Both,

the dovish committee and the political pressure against tighter policy made it more diﬃcult for

Chairman Martin to reach a consensus for ﬁrm policy within the FOMC. Indeed, we observe that

even when the expansionary eﬀects of military spending related to the Vietnam War became

evident, the FOMC initially hesitated, then tightened modestly but soon erred toward loose

policy.

Overall, the narrative evidence from the 1960s supports the important elements that we highlight

in this paper: the Fed’s reaction to military spending and the role of the FOMC composition

for this reaction.

ECB Working Paper Series No 2851

7 Conclusion

This paper proposes an identiﬁcation design to estimate the eﬀects of systematic monetary

policy for the propagation of macroeconomic shocks. Our design combines the narrative classi-

ﬁcation of FOMC members’ policy preferences from Istreﬁ (2019) with a novel FOMC rotation

instrument for systematic monetary policy. The identiﬁcation design opens up myriad research

opportunities, such as revisiting the eﬀects of various ﬁscal, technology, and oil shocks and their

dependence on systematic monetary policy.

We use our identiﬁcation design to study government spending shocks and ﬁnd that ﬁscal

spending multipliers depend strongly and signiﬁcantly on systematic monetary policy. We

inspect the mechanism behind our result and ﬁnd consistent interest rate and inﬂation responses.

Our results suggest that it is critically important for ﬁscal policy makers to consider the stance of

monetary policy in their decision making process. However, a potentially misleading conclusion

from our results is that the government should increase spending when the FOMC is dovish.

This could be misleading because such changes in government spending are not random shocks.

Moreover, the Lucas (1976) critique applies if the conduct of ﬁscal policy changes structurally.

To avoid such misleading conclusions, a promising avenue for future research is to use our results

to discipline micro-founded models to study optimal ﬁscal stabilization policy.

Finally, while our identiﬁcation design is speciﬁc to U.S. monetary policy, a promising avenue

for future research is to study other countries or currency areas in which committees decide

monetary policy. In fact, since 2015 the European Central Bank’s Governing Council allocates

voting rights to its members through a rotation mechanism.

References

Abrams, B. (2006): “How Richard Nixon Pressured Arthur Burns: Evidence from the Nixon

Tapes,” Journal of Economic Perspectives, 20, 177–188.

Abrams, B. A. and J. L. Butkiewicz (2012): “The Political Business Cycle: New Evidence

from the Nixon Tapes,” Journal of Money, Credit and Banking, 44, 385–399.

Andrews, I. (2018): “Valid Two-Step Identiﬁcation-Robust Conﬁdence Sets for GMM,” Review

of Economics and Statistics, 100, 337–348.

Andrews, I., J. H. Stock, and L. Sun (2019): “Weak Instruments in Instrumental Variables

Regression: Theory and Practice,” Annual Review of Economics, 11, 727–753.

Arias, J. E., D. Caldara, and J. F. Rubio-Ramirez (2019): “The systematic component

of monetary policy in SVARs: An agnostic identiﬁcation procedure,” Journal of Monetary

Economics, 101, 1–13.

Auerbach, A. and Y. Gorodnichenko (2012): “Measuring the Output Responses to Fiscal

Policy,” American Economic Journal: Economic Policy, 4, 1–27.

Baily, M. N., C. Hulten, and D. Campbell (1992): “Productivity Dynamics in Manufac-

turing Plants,” Brookings Papers on Economic Activity, 23, 187–267.

Barnichon, R., D. Debortoli, and C. Matthes (2022): “Understanding the Size of the

Government Spending Multiplier: It’s in the Sign,” Review of Economic Studies, 89, 87–117.

ECB Working Paper Series No 2851

Barnichon, R. and G. Mesters (2022): “A Suﬃcient Statistics Approach for Macro Policy

Evaluation,” Working Paper Series 2022, Federal Reserve Bank of San Francisco.

Barro, R. J. and C. J. Redlick (2011): “Macroeconomic Eﬀects from Government

Purchases and Taxes,” Quarterly Journal of Economics, 126, 51–102.

Barsky, R. and E. Sims (2011): “News shocks and business cycles,” Journal of Monetary

Economics, 58, 273–289.

Barthélemy, J. and M. Marx (2017): “Solving endogenous regime switching models,”

Journal of Economic Dynamics and Control, 77, 1–25.

Bauer, M. D., C. Pflueger, and A. Sunderam (2022): “Perceptions about Monetary

Policy,” NBER Working Paper 30480.

Bauer, M. D. and E. T. Swanson (2023): “An Alternative Explanation for the “Fed Infor-

mation Eﬀect”,” American Economic Review, 113, 664–700.

Ben Zeev, N., V. A. Ramey, and S. Zubairy (2023): “Do Government Spending Multipliers

Depend on the Sign of the Shock?” AEA Papers and Proceedings, forthcoming.

Benati, L. (2021): “Leaning against house prices: A structural VAR investigation,” Journal

of Monetary Economics, 118, 399–412.

Bernanke, B. S. and A. S. Blinder (1992): “The Federal Funds Rate and the Channels of

Monetary Transmission,” American Economic Review, 82, 901–921.

Bernanke, B. S., M. Gertler, M. Watson, C. A. Sims, and B. M. Friedman (1997):

“Systematic Monetary Policy and the Eﬀects of Oil Price Shocks,” Brookings Papers on

Economic Activity, 1997, 91–157.

Bianchi, F. (2013): “Regime Switches, Agents’ Beliefs, and post-World War II U.S. Macroe-

conomic Dynamics,” Review of Economic Studies, 80, 463–490.

Bianchi, F. and C. Ilut (2017): “Monetary/Fiscal policy mix and agents’ beliefs,” Review

of Economic Dynamics, 26, 113–139.

Blanchard, O. and R. Perotti (2002): “An Empirical Characterization of the Dynamic

Eﬀects of Changes in Government Spending and Taxes on Output,” Quarterly Journal of

Economics, 117, 1329–1368.

Blinder, A. S. (1999): Central Banking in Theory and Practice, vol. 1 of MIT Press Books,

The MIT Press.

——— (2022): A Monetary and Fiscal History of the United States, 1961-2021, Princeton

University Press.

Bordo, M. D. and O. F. Humpage (2014): “Federal Reserve policy and Bretton Woods,”

NBER Working Paper 20656.

Bordo, M. D. and K. Istrefi (2023): “Perceived FOMC: The making of hawks, doves and

swingers,” Journal of Monetary Economics, forthcoming.

Bound, J., D. A. Jaeger, and R. M. Baker (1995): “Problems with Instrumental Variables

Estimation When the Correlation Between the Instruments and the Endogenous Explanatory

Variable is Weak,” Journal of the American Statistical Association, 90, 443–450.

Broner, F., D. Clancy, A. Erce, and A. Martin (2022): “Fiscal Multipliers and Foreign

Holdings of Public Debt,” Review of Economic Studies, 89, 1155–1204.

Carvalho, C., F. Nechio, and T. Tristão (2021): “Taylor rule estimation by OLS,” Journal

ECB Working Paper Series No 2851

of Monetary Economics, 124, 140–154.

Chappell, H., R. R. McGregor, and T. A. Vermilyea (2005): Committee Decisions on

Monetary Policy: Evidence from Historical Records of the Federal Open Market Committee,

vol. 1, MIT Press, 1 ed.

Chaudhuri, S. and E. Zivot (2011): “A new method of projection-based inference in GMM

with weakly identiﬁed nuisance parameters,” Journal of Econometrics, 164, 239–251.

Chernozhukov, V. and C. Hansen (2008): “The reduced form: A simple approach to

inference with weak instruments,” Economics Letters, 100, 68–71.

Christiano, L., M. Eichenbaum, and S. Rebelo (2011): “When is the Government

Spending Multiplier Large?” Journal of Political Economy, 119, 78 – 121.

Christiano, L. J., M. Eichenbaum, and C. L. Evans (1999): “Monetary policy shocks:

What have we learned and to what end?” Handbook of Macroeconomics, 1, 65–148.

Cieslak, A., S. Hansen, M. McMahon, and S. Xiao (2022): “Policymakers’ Uncertainty,”

mimeo.

Clarida, R., J. Gali, and M. Gertler (2000): “Monetary Policy Rules and Macroeconomic

Stability: Evidence and Some Theory,” Quarterly Journal of Economics, 115, 147–180.

Cloyne, J. S., Ò. Jordà, and A. M. Taylor (2021): “Decomposing the Fiscal Multiplier,”

NBER Working Paper 26939.

Cochrane, J. H. (2011): “Determinacy and Identiﬁcation with Taylor Rules,” Journal of

Political Economy, 119, 565–615.

——— (2022): “The ﬁscal roots of inﬂation,” Review of Economic Dynamics, 45, 22–40.

Cochrane, J. H. and M. Piazzesi (2002): “The Fed and Interest Rates - A High-Frequency

Identiﬁcation,” American Economic Review, 92, 90–95.

Coibion, O. and Y. Gorodnichenko (2011): “Monetary Policy, Trend Inﬂation, and the

Great Moderation: An Alternative Interpretation,” American Economic Review, 101, 341–70.

Davig, T. and E. M. Leeper (2007): “Generalizing the Taylor Principle,” American

Economic Review, 97, 607–635.

——— (2008): “Endogenous Monetary Policy Regime Change,” in NBER International

Seminar on Macroeconomics 2006, National Bureau of Economic Research, NBER Chap-

ters, 345–391.

DeLong, B. J. (1997): “America’s Peacetime Inﬂation: The 1970s.” in Reducing Inﬂation:

Motivation and Strategy, ed. by C. Romer and D. Romer, University of Chicago Press.

Ehrmann, M., R. Tietz, and B. Visser (2022): “Voting Right Rotation, Behavior of

Committee Members and Financial Market Reactions: Evidence from the U.S. Federal Open

Market Committee,” IMF Working Papers 2022/105, International Monetary Fund.

Farhi, E. and I. Werning (2016): “Fiscal Multipliers: Liquidity Traps and Currency Unions,”

in Handbook of Macroeconomics, Elsevier, vol. 2, 2417–2492.

Ferguson, N., M. Kornejew, P. Schmelzing, and M. Schularick (2023): “The Safety

Net: Central Bank Balance Sheets and Financial Crisis, 1987-2020,” Economics working

paper 23102, Hoover Institution.

Ferrière, A. and G. Navarro (2018): “The Heterogeneous Eﬀects of Government Spending

: It’s All About Taxes,” International Finance Discussion Papers, 1237.

ECB Working Paper Series No 2851

Galí, J. (2015): Monetary Policy, Inﬂation, and the Business Cycle: An Introduction to the

New Keynesian Framework and Its Applications, Princeton University Press.

Galí, J., J. López-Salido, and J. Vallés (2003): “Technology shocks and monetary policy:

assessing the Fed’s performance,” Journal of Monetary Economics, 50, 723–743.

Ghassibe, M. and F. Zanetti (2022): “State dependence of ﬁscal multipliers: the source of

ﬂuctuations matters,” Journal of Monetary Economics, 132, 1–23.

Goodfriend, M. and R. G. King (2005): “The incredible Volcker disinﬂation,” Journal of

Monetary Economics, 52, 981–1015.

Gordon, R. J. and R. Krenn (2010): “The End of the Great Depression: VAR Insight on

the Roles of Monetary and Fiscal Policy.” NBER Working Paper 16380.

Hack, L., K. Istrefi, and M. Meier (2023): “Identiﬁcation of Systematic Monetary Policy,”

CEPR Discussion Paper 17999 (https://cepr.org/publications/dp17999).

Ilzetzki, E., E. G. Mendoza, and C. A. Végh (2013): “How big (small?) are ﬁscal

multipliers?” Journal of Monetary Economics, 60, 239–254.

Istrefi, K. (2019): “In Fed Watchers’ Eyes: Hawks, Doves and Monetary Policy,” Banque de

France Working Paper.

Jordà, Ò. and A. M. Taylor (2016): “The Time for Austerity: Estimating the Average

Treatment Eﬀect of Fiscal Policy,” The Economic Journal, 126, 219–255.

Kilian, L. and L. T. Lewis (2011): “Does the Fed Respond to Oil Price Shocks?” Economic

Journal, 121, 1047–1072.

Lazarus, E., D. J. Lewis, J. H. Stock, and M. W. Watson (2018): “HAR Inference:

Recommendations for Practice,” Journal of Business & Economic Statistics, 36, 541–559.

Leeper, E. M., N. Traum, and T. B. Walker (2017): “Clearing up the Fiscal Multiplier

Morass,” American Economic Review, 107, 2409–54.

Leeper, E. M. and T. Zha (2003): “Modest policy interventions,” Journal of Monetary

Economics, 50, 1673–1700.

Lewis, D. J. and K. Mertens (2022): “A Robust Test for Weak Instruments with Multiple

Endogenous Regressors,” Staﬀ reports, no. 1020, Federal Reserve Bank of New York.

Liu, Y. and J. C. Wu (2021): “Reconstructing the yield curve,” Journal of Financial

Economics, 142, 1395–1425.

Lucas, R. E. (1976): “Econometric policy evaluation: A critique,” Carnegie-Rochester Confer-

ence Series on Public Policy, 1, 19–46.

McKay, A. and C. K. Wolf (2022): “What Can Time-Series Regressions Tell Us About

Policy Counterfactuals?” NBER Working Paper 30358.

Meltzer, A. H. (2005): “Origins of the Great Inﬂation,” Federal Reserve Bank of St. Louis

Review, March/April, 145–176.

Miyamoto, W., T. L. Nguyen, and D. Sergeyev (2018): “Government Spending Multi-

pliers Under the Zero Lower Bound: Evidence from Japan,” American Economic Journal:

Macroeconomics, 10, 247–77.

Montiel Olea, J. L. and C. Pflueger (2013): “A Robust Test for Weak Instruments,”

Journal of Business & Economic Statistics, 31, 358–369.

Mountford, A. and H. Uhlig (2009): “What are the eﬀects of ﬁscal policy shocks?” Journal

ECB Working Paper Series No 2851

of Applied Econometrics, 24, 960–992.

Nakamura, E. and J. Steinsson (2014): “Fiscal Stimulus in a Monetary Union: Evidence

from US Regions,” American Economic Review, 104, 753–92.

Orphanides, A. (2003): “The quest for prosperity without inﬂation,” Journal of Monetary

Economics, 50, 633–663.

——— (2004): “Monetary Policy Rules, Macroeconomic Stability, and Inﬂation: A View from

the Trenches,” Journal of Money, Credit and Banking, 36, 151–175.

Primiceri, G. (2006): “Why Inﬂation Rose and Fell: Policymakers’ Beliefs and US Postwar

Stabilization Policy,” Quarterly Journal of Economics, 867–901.

Primiceri, G. E. (2005): “Time Varying Structural Vector Autoregressions and Monetary

Policy,” Review of Economic Studies, 72, 821–852.

Ramey, V. (2016): “Macroeconomic Shocks and Their Propagation,” Handbook of Macroeco-

nomics, 2, 71 – 162.

Ramey, V. A. (2011): “Identifying Government Spending Shocks: It’s all in the Timing,”

Quarterly Journal of Economics, 126, 1–50.

Ramey, V. A. and M. D. Shapiro (1998): “Costly capital reallocation and the eﬀects of

government spending,” Carnegie-Rochester Conference Series on Public Policy, 48, 145–194.

Ramey, V. A. and S. Zubairy (2018): “Government Spending Multipliers in Good Times

and in Bad: Evidence from US Historical Data,” Journal of Political Economy, 126, 850–901.

Riboni, A. and F. J. Ruge-Murcia (2010): “Monetary policy by Committee: Consensus,

Chairman dominance, or simple majority?” Quarterly Journal of Economics, 125, 363–416.

——— (2022): “The power of the Federal Reserve Chair,” International Economic Review.

Romer, C. and D. Romer (1989): “Does Monetary Policy Matter? A New Test in the Spirit

of Friedman and Schwartz,” in NBER Macroeconomics Annual 1989, Volume 4, National

Bureau of Economic Research, Inc, 121–184.

Romer, C. D. and D. H. Romer (2002): “The Evolution of Economic Understanding and

Postwar Stabilization Policy,” NBER Working Paper 9274.

——— (2004): “A New Measure of Monetary Shocks: Derivation and Implications,” American

Economic Review, 94, 1055–1084.

——— (2010): “The Macroeconomic Eﬀects of Tax Changes: Estimates Based on a New

Measure of Fiscal Shocks,” American Economic Review, 100, 763–801.

Sargent, T. J. (1979): “Estimating Vector Autoregressions Using Methods not Based on

Explicit Economic Theories,” Quarterly Review, 3.

Sims, C. A. and T. Zha (2006): “Were There Regime Switches in US Monetary Policy?”

American Economic Review, 96, 54–81.

Wolf, C. K. (2023): “Fiscal Stimulus and the Systematic Response of Monetary Policy,” AEA

Papers and Proceedings, forthcoming.

Woodford, M. (2011): “Simple Analytics of the Government Expenditure Multiplier,” Amer-

ican Economic Journal: Macroeconomics, 3, 1–35.

ECB Working Paper Series No 2851

Appendix A New Keynesian model

A.1 Equilibrium dynamics

We derive equation (2.3). Denoting by lower case letters (log) deviations from steady state, we

obtain the following three equilibrium conditions for the model described in Section 2: A New

Keynesian Phillips Curve

= βE

[π

t+1

] + λ



φ +

1 − γ



−

λγ

1 − γ

− λ(1 + φ)x

, (A.1)

where λ = (1 − θ)(1 − βθ)/θ, a dynamic IS equation

= E

t+1

] − (1 − γ)(i

− E

[π

t+1

]) + γ (1 − ρ

) x

, (A.2)

and the Taylor rule

, (A.3)

where

= ϕ + ϕ

follows

= ρ

t−1

+ ζ

+ η

, |ρ

| < 1.

We assume the macroeconomic shocks (ε

, ε

) and the exogenous shifter η

are mutually inde-

pendent and identically distributed over time. We ﬁrst rewrite the IS equation by plugging in

the Taylor rule and Phillips Curve to obtain

1 − γ

1 + λ (φ(1 − γ) + 1) ϕ



t+1

]

1 − γ

+ (1 − βϕ

[π

t+1

]

1 − γ

(ϕ

λ + (1 − ρ

)) x

+ ϕ

λ (φ + 1) x



(A.4)

Combining (A.1) and (A.4), the model dynamics follow Y

= A(ϕ

) E

t+1

] + B(ϕ

) X

, with

= (y

, π

)

′

, X

= (x

, x

)

′

and A(ϕ

), B(ϕ

) depending only on model parameters. A ﬁrst-

order approximation around ϕ

= 0 yields

= AE

t+1

] + BX



∂

t+1

] + AE

[∂

t+1

] + ∂



, (A.5)

where A ≡ A(0), B ≡ B(0), ∂

(·) denotes a derivative with respect to ϕ

that is evaluated at

= 0, and we omit the approximation error. Consider the guess for the law of motion that

reads Y

= A + BX

+ CX

+ Dϕ

. It is straightforward to verify that the guess satisﬁes (A.5).

The coeﬃcients of the guess depend on the deep structural parameters of the model and can be

determined via the method of undetermined coeﬃcients. This fully describes the approximate

state-dependent model dynamics with respect to systematic monetary policy ϕ

and provides

equation (2.3) in the main text, where a = A

, b

= B

, b

= B

, and analogously for C and

D. In the special case ρ

= ρ

= 0, the coeﬃcients in (2.3) are given by (2.6).

ECB Working Paper Series No 2851

A.2 Identiﬁcation

We next describe the identiﬁcation results in Section 2 in some more detail. The key step is to

determine the residual in the state-dependent local projection (2.4). Using (2.2), (2.3), and the

laws of motion for x

and x

, we obtain

t+h

= F

· z

t+h

where F

is a coeﬃcient vector and z

t+h

is the following vector of variables:

t+h

t−1

, {ε

t+i

}

i=1

, x

t−1

t+h

, ε

{η

t+i

}

i=1

, ε

{ε

t+i

}

i=1

, ε

{ε

t+i

}

i=1

{ε

t+i

t+h

}

i=1

, {η

t+i

}

i=1

, {ε

t+i

}

i=1

, x

t+h

, x

t+h

′

where {ε

t+i

}

i=1

denotes the vector of all ε

t+i

for i = 1 through i = h, and analogously for all

terms in braces. Deﬁning the vector of regressors (excluding the intercept) in (2.4) by

, ε

, ϕ

′

a suﬃcient condition for the consistency of the OLS estimates of (β

, γ

, δ

) requires

E[X

t+h

)

′

] = 0,

where 0 denotes a zero matrix with conforming dimension. In general, this orthogonality condi-

tion is satisﬁed if ζ

= ζ

= 0.

We next turn to the IV estimator of (β

, γ

, δ

). Consider an instrument ϕ

with the following

properties:

E[ϕ

t+i

] = E[ϕ

t+i

] = 0 ∀i

E[ϕ

] ̸= 0, E[ϕ

t+i

] = 0 ∀i ̸= 0

Deﬁning as instrument vector

, ε

, ϕ

′

consistency of the IV estimator requires

E[Q

t+h

)

′

] = 0.

This condition is generally satisﬁed in our model. Hence, the IV estimator consistently estimates

(β

, γ

, δ

) even in the absence of strong exogeneity assumptions for systematic monetary policy

ECB Working Paper Series No 2851

Appendix B Data

This section explains the data sources and data preparation procedures used to compile the

data set used in our analysis.

B.1 Narrative account

We use Istreﬁ’s (2019) data set which is a panel of FOMC members where the time dimension

refers to FOMC meetings. The panel contains the policy preferences of voting members at each

FOMC meeting for 1960-2023. We code the numerical variable Hawk

iτ

∈ [−1, 1] as explained

in the main text.

Accounting for missing data. The information on some FOMC members during the ﬁrst

ﬁve years of our sample is relatively sparse, leaving us with many unclassiﬁed FOMC members

in this period. For example, we observe the preferences for only 115 out of 195 member-meeting

pairs in 1960. The share of observed preferences increases gradually. From 1966 onward, we

reach an average share of 88 percent which is satisfactory. Before 1966 we employ the following

imputation scheme. We assume that the unobserved preferences coincide with the ﬁrst observed

preference of the respective FOMC member. Formally, let t(i) be the ﬁrst meeting for which

Hawk

iτ

is not missing for member i. Then, we assume that Hawk

iτ

= Hawk

it(i)

for all τ < t(i).

Accounting for short-term substitutes. Occasionally, voting FOMC members do not

attend the meetings but instead, get replaced by a substitute. This introduces noise in the

data under the assumption that the short-term substitutes act in the best interest of the person

that is substituted. This assumption is appropriate as the short-term substitutes are often

direct subordinates of the original voting member. Thus, we explicitly assume that short-term

substitutes act as if the original member attended the meeting if the following criteria hold:

(i) the substitution period is not longer than 6 months when the substitute is from the same

Federal Reserve bank, (ii) the substitution period is not longer than 3 months if the substitute

is not from the same Federal Reserve bank, (iii) the substitution does not take place at the

beginning or the end of a rotation cycle within a rotation group.

Formally, let Hawk

iτ

the preference of a substitute that appears in meeting τ. Then, we set Hawk

iτ

= Hawk

jτ −k

where j is the original member and we use the most recent observation of her policy preferences

from meeting τ − k. However, many times it holds that the preferences of the substitute and

the original voter coincide which implies that the procedure above does not change the data.

It turns out that we change less than 1 percent of the data points, and that our results are

insensitive to the choices made above.

For example, consider the rotation between the Chicago and Cleveland Federal Reserves. Suppose that

the Chicago president had the voting right until meeting τ and the Cleveland president thereafter. If Chicago

exercises the voting right in τ + 1 on behalf of Cleveland, we would not conduct any corrections because it is not

clear whether the Chicago president acts in the best interest of the Cleveland president or not.

ECB Working Paper Series No 2851

B.2 Macroeconomic data

All aggregate series are explained below. If applicable, we put the data identiﬁers of the respec-

tive data source in parentheses.

Ramey and Zubairy (2018). We take the series for potential output (rgdp_pott6 ), real

GDP (rgdp), nominal government spending (ngov), the GDP deﬂator (pgdp) and the military

spending news shock (news) from the replication package of Ramey and Zubairy (2018). We

follow their data preparation steps to create the aggregate series as in their paper.

FRED Economic Data. We use headline CPI (CPIAUCSL) and CPI core (CPILFESL)

inﬂation deﬁned as the year-over-year growth rate of the respective price index, and the eﬀective

federal funds rate (DFF). The 10-year treasury market yield (DGS10 ) starts only in 1962q1

and is therefore combined with the very same variable from Romer and Romer (2010) to obtain

a series that starts in 1960q1. Similarly, we use the 1-year market yield from Liu and Wu (2021)

and impute the ﬁrst four observations (1960q1 to 1960q4) with a similar 1-year treasury market

yield from Fred (DTB1YR). Personal consumption expenditures (PCE), gross private domestic

investment (GPDI ), and federal government defense expenditures (FDEFX ) is divided by the

GDP deﬂator and by real potential GDP, both taken from Ramey and Zubairy (2018), see

above. We compute non-military government spending by subtracting the defense spending

from total government spending. Variables are averaged to quarterly frequency, if applicable.

Livingston survey. We use inﬂation expectations from the Livingston survey. Our measure

of inﬂation expectation is the annualized expected growth rate of CPI forecasts from 6 to 12

months ahead. Because the survey is biannual, we assume that inﬂation expectations remain

constant in quarters in which no new data is available. Formally, we let π

= π

t−1

, whenever

there is no survey conducted in quarter t. The (ex-ante) real rates are computed as i

= i

− π

where i

is a nominal rate of interest.

Additional series. The validation exercise in Appendix E is based on forecasts from the

Fed’s Greenbook. We use the average of the one- and two-quarter ahead inﬂation forecast,

following Coibion and Gorodnichenko (2011). We use additional series for some exercises in our

sensitivity analysis. For the Blanchard and Perotti (2002) shock, we account for anticipation

in government spending by including the one-quarter projected growth rate of government

spending from Ramey’s (2011) data.

As an additional control variable we use the primary

surplus (svt_q) from Cochrane (2022), seasonally adjusted via X-13 ARIMA-SEATS procedure

from the U.S. Census Bureau.

The ﬁscal shock is computed as news

/(pgdp

t−1

× rgdp_pott6

t−1

) × 100. Detrended real GDP is

rgdp

/rgdp_pott6

× 100 and detrended real government spending is ngov

/(pgdp

× rgdp_pott6

) × 100.

The survey of professional forecasters provides the government spending forecasts only from 1981q3 onward.

Thus, Ramey (2011) imputes the government spending forecasts with defense spending forecasts to extend the

sample until 1968q4.

ECB Working Paper Series No 2851

Appendix C Hawk-Dove decompositions

In this section, we propose a decomposition of ﬂuctuations in the aggregate Hawk-Dove balance

Hawk

and the FOMC rotation instrument Hawk

. We ﬁnd that the FOMC rotation is a

key source of variation in Hawk

. Variation in the instrument Hawk

is largely due to the

rotation of incumbent FOMC members.

Decomposition of Hawk

. We derive a decomposition of the aggregate Hawk-Dove balance

similar to the aggregate productivity decomposition in Baily et al. (1992). We ﬁrst rewrite the

aggregate Hawk-Dove balance in equation (3.2) as

Hawk

i∈M

Hawk

, s

. (C.1)

We deﬁne a decomposition over p-period changes in the balance:

∆

Hawk

= Hawk

− Hawk

t−p

i∈M

Hawk

−

i∈M

t−p

Hawk

it−p

(C.2)

We next partition the set M

into the set of “surviving” FOMC members S

present in t − p

and t, the set of entering FOMC members E

present in t but not in t − p, and the set of exiting

FOMC members X

present in t − p but not in t to rewrite:

∆

Hawk

i∈S

Hawk

− s

t−p

Hawk

it−p

) +

i∈E

Hawk

−

i∈X

t−p

Hawk

it−p

i∈S

t−p

(Hawk

− Hawk

it−p

) +

i∈S

− s

t−p

)Hawk

i∈E

Hawk

−

i∈X

t−p

Hawk

it−p

(C.3)

The ﬁrst term captures changes in preferences of surviving FOMC members, the second term

captures changes in the number of FOMC members, the third term captures entry into the

FOMC, and the last term captures exit from the FOMC.

Finally, we further distinguish between the rotating and non-rotating FOMC members in the

set of entering and exiting FOMC members, denoted E

, E

, X

and X

to obtain our

decomposition of interest:

∆

Hawk

i∈S

t−p

(Hawk

− Hawk

it−p

) +

i∈S

− s

t−p

)Hawk

i∈E

Hawk

−

i∈X

t−p

Hawk

it−p

i∈E

Hawk

−

i∈X

t−p

Hawk

it−p

(C.4)

The second row captures changes in the aggregate Hawk-Dove balance due to the entry and

exit of rotating FOMC members, while the third row captures the contribution of entry and

exit of non-rotating FOMC members.

ECB Working Paper Series No 2851

We use the decomposition of the aggregate Hawk-Dove balance in equation (C.4) to quan-

tify the statistical importance of three factors (corresponding to the three rows of the equa-

tion): intensive-margin changes in preferences, extensive margin changes of non-rotating FOMC

members, and extensive margin changes due to the rotation. We focus on the yearly changes

in the quarterly aggregate Hawk-Dove balance (i.e., we set p = 4) to capture well the changes

due to the annual rotation.

The variance in yearly changes of the aggregate Hawk-Dove balance is 0.083. The variance of

the ﬁrst term in the ﬁrst row of (C.4), which captures intensive margin changes of preferences,

corresponds to 9% of the total variance. Changes in the weights, the second term in the

ﬁrst row, are negligible in size. The variance of the second row of (C.4), capturing extensive

margin changes of non-rotating FOMC members, corresponds to 22% of the total variance.

The variance of the third row of (C.4), capturing extensive margin changes of rotating FOMC

members, corresponds to 53% of the total variance. Finally, the covariances between these terms

account for 15% of the total variance. The results diﬀer little for quarterly changes (p = 1).

Notably, extensive margin changes of rotating FOMC members still account for 52% of the total

variance.

Decomposition of Hawk

. Analogously, we propose a decomposition for the FOMC rota-

tion instrument. We ﬁrst rewrite Hawk

in equation (C.7) as

Hawk

i∈R

Hawk

, s

. (C.5)

After a little algebra, we obtain

∆

Hawk

i∈S

t−p

(Hawk

− Hawk

it−p

) +

i∈S

− s

t−p

)Hawk

i∈E

Hawk

−

i∈X

t−p

Hawk

it−p

, (C.6)

where S

denotes the set of surviving rotating FOMC members. Finally, we further distinguish

between the rotating entering FOMC members whose appointments start or end (A), i.e., they

enter for the ﬁrst time after their appointment as regional FRB president or appear the last

time as such, and incumbent (I) regional FRB presidents, denoted E

, E

, X

∆

Hawk

i∈S

t−p

(Hawk

− Hawk

it−p

) +

i∈S

− s

t−p

)Hawk







i∈E

Hawk

−

i∈X

t−p

Hawk

it−p













i∈E

Hawk

−

i∈X

t−p

Hawk

it−p







(C.7)

We use the decomposition of the FOMC rotation instrument in equation (C.7) to quantify

ECB Working Paper Series No 2851

the statistical importance of three factors (corresponding to the three rows of the equation):

intensive-margin changes in preferences, extensive margin changes of rotating FOMC members

whose appointment starts or ends, and extensive margin changes of incumbent rotating FOMC

members.

For yearly changes in the rotation instrument, we ﬁnd that 93% of the variance is due to the

rotation of incumbent members, while 7% is due to appointments starting or ending. All other

variances and covariances are negligible in size. Note that yearly changes mechanically mute the

importance of intensive margin changes, because current rotating FOMC members are typically

not still FOMC members a year later. Therefore, we also study quarterly changes (p = 1).

Intensive margin changes now explain 4% of the variance, appointments account for 23%, and

rotations of incumbent members account for 71%.

Appointments become relatively more important for p = 1 because only every fourth quarter

of ∆

Hawk

features a rotation. Compared to ∆

Hawk

for which the rotation aﬀects all

quarters, we mechanically lower the importance of rotations and the overall variance for p = 1.

Appendix D Relation to monetary policy shocks

In this section, we show how empirically identiﬁed monetary policy shocks relate to changes in

systematic monetary policy. While the model in Section 2 makes a sharp distinction between

systematic monetary policy (ϕ

) and monetary policy shocks (that would enter the Taylor rule as

additive extra terms), empirically identiﬁed monetary policy shocks often blur this distinction.

We revisit the seminal identiﬁcation strategy proposed by Romer and Romer (2004), RR hence-

forth. They identify monetary policy shocks as the residual from a regression of changes in the

target federal funds rate on various Greenbook forecasts. To interpret their regression through

the lens of our model in Section 2, we consider a stylized version of the RR regression

= ϕ

+ ε

, (D.1)

in which π

denotes the Greenbook inﬂation forecast before a change in monetary policy, and

is the RR monetary policy shock.

For simplicity, we put estimation and identiﬁcation

concerns aside. We further assume the following data-generating policy rule

= ϕ

+ ε

, (D.2)

where ε

is a true monetary policy shock and systematic monetary policy satisﬁes ϕ

> 1.

Combining this with the RR regression yields the RR shock

= (ϕ

− ϕ

)π

+ ε

. (D.3)

Hence, the empirical shock ε

captures variation in systematic monetary policy ϕ

, variation

The stylized regression omits any lags or leads from the original regression in Romer and Romer (2004).

This is inconsequential if the DGP features iid ﬂuctuations. We further omit unemployment and output growth

from the original regression, because they are absent from the policy rule in the model.

For simplicity, we deﬁne the Taylor rule over π

, instead of π

as in Section 2. The insight in this section

does not change under the original rule, except that ε

also depends on the forecast error π

− π

ECB Working Paper Series No 2851

in inﬂation forecasts π

, and monetary policy shocks ε

. The empirical shock captures the

model shock ε

= ε

in the special case when systematic monetary policy is time-invariant.

In general, the empirical shock also captures joint time-variation in systematic monetary policy

and inﬂation forecasts π

, where the latter naturally depends on the state of the economy.

Finally, high-frequency identiﬁed monetary policy shocks may reﬂect changes in systematic

monetary policy in a similar fashion (Bauer and Swanson, 2023).

Appendix E Validation exercise

We use the Hawk-Dove balance and the FOMC rotation instrument to estimate the federal

funds rate (FFR) response to inﬂation forecasts as a function of the hawkishness of the FOMC.

In support of our identiﬁcation design, we ﬁnd that a hawkish FOMC is associated with a more

pronounced hike of the federal funds rate in the face of inﬂationary pressure. We estimate

a state-dependent local projection speciﬁcation that is akin to a forward-looking Taylor rule.

Formally, we estimate a set of regressions

F F R

t+h

= α

+ β

ˆπ

+ γ

ˆπ

(Hawk

− Hawk) + ζ

+ v

t+h

, (E.1)

for h = 0, 1, ..., H, and F F R

t+h

and ˆπ

denote the federal funds rate and the average of the one-

and two-quarter ahead Greenbook inﬂation forecast, respectively. The control vector includes

four lags of the federal funds rate and the inﬂation forecast. The data is at a quarterly frequency

and the sample runs from 1969 to 2008, due to the availability of inﬂation forecasts and the

reaching of the zero lower bound in 2008.

Figure E.1 presents IV estimates where we use the FOMC rotation instrument interacted with

the inﬂation forecast as an instrument for the interaction term in the speciﬁcation above. We

show estimates that are normalized to represent the inﬂation forecast being one percentage

point above the sample average. The left panel displays the response under the average FOMC

(β

). The right panel displays the diﬀerential response (γ

) when there are 2 more hawks in

the FOMC relative to the average composition.

On average, the FOMC reacts with a federal funds rate hike. The response is statistically

signiﬁcant at the ﬁve percent level for six quarters. The response builds up over time, consistent

with interest rate smoothing. Incidentally, it satisﬁes the Taylor principle for a prolonged period

of almost 2 years and peaks at 1.48 percentage points. The response turns stronger when the

FOMC is more hawkish, as indicated by the diﬀerential eﬀects in Panel (b). The estimates

of the interaction coeﬃcient γ

are hump-shaped and peak after 2 years at 0.92 percentage

points. The response is signiﬁcant at ﬁve percent for almost 2 years. This result suggests that a

more hawkish FOMC is associated with a stronger and more persistent federal funds rate hike.

Conversely, a more dovish FOMC implies a substantially weaker response.

Finally, this validation exercise lends itself to assessing the relevance condition of our instrument

Consider the high-frequency identiﬁed monetary policy shock ε

HFI

= i

− E

t−∆

], where E

t−∆

denotes

expectations shortly before the meeting. Combining the shock with the monetary policy rule in (D.2) yields

HFI

= ε

+ϕ

−E

t−∆

[ϕ

]. Hence, ε

HFI

convolutes monetary policy shocks ε

with changes in systematic

monetary policy ϕ

and inﬂation forecasts π

ECB Working Paper Series No 2851

Figure E.1: FFR response to inﬂation and the FOMC hawkishness

Average FFR (β

) Diﬀerential FFR (γ

)

Notes: The ﬁgure shows responses of the federal funds rate to an inﬂation Greenbook forecast that is one percentage point

above its sample average, conditional on systematic monetary policy (Hawk

). We show IV estimates based on (E.1). The

captures the responses when Hawk

equals its sample average. The γ

captures the diﬀerential responses when Hawk

exceeds the sample average by two hawks. The shaded areas indicate 68% and 95% conﬁdence bands using Newey-West

standard errors.

more formally. We use the weak instruments test from Montiel Olea and Pﬂueger (2013).

can reject the null of weak instruments. More formally, we compute p-values for the bias

exceeding 10% percent of the benchmark, see Montiel Olea and Pﬂueger (2013) for details.

The p-values are bounded from above by 0.055 and are below the 0.05 level at most horizons.

Moreover, for a test of whether the bias exceeds 20%, we can reject the null at 1% for all

horizons.

Overall, we show that the federal funds rate response to inﬂation correlates positively with the

hawkishness of the FOMC, Hawk

. The responses are consistent with our measurement of the

stance of systematic monetary policy and are further in line with Bordo and Istreﬁ (2023). We

see this result as a validation that our measurement of systematic monetary policy, through

Hawk

, captures important aspects of the Federal Reserve’s monetary policy-making.

Appendix F Estimating ﬁscal multipliers

This section elaborates on how we obtain the ﬁscal multiplier estimate and on how to conduct

valid inference.

Cumulated local projection. Summing up equation (3.4) for horizon 0 to H yields

h=0

t+j

= ˜α

+ ˜γ

(Hawk

− Hawk) +

(Hawk

− Hawk) +

+ ˜v

t+j

, (F.1)

where

h=0

, ˜γ

h=0

, and analogously for ˜α

. The outcome is

either cumulated GDP or cumulated government spending (G). With this, one can estimate the

In our setting with a single endogenous regressor, this test is equivalent to the test by Lewis and Mertens

(2022).

ECB Working Paper Series No 2851

numerator and the denominator of the ﬁscal multiplier from equation (4.1) in one step.

One-step multiplier estimation. We employ an estimation procedure akin to seemingly

unrelated regressions to estimate equation (F.1) for GDP and G in a single step. This allows us

to conduct standard asymptotic inference with respect to the implied ﬁscal multipliers. First,

we deﬁne the regressor matrix w and instrument matrix q where each column is a vector of

data, i.e. w

1·

= (w

, .., w

)

′

and T denotes the sample size.

w = (1, s, s(Hawk − Hawk), (Hawk − Hawk), Z)

′

(F.2)

q = (1, s, s(Hawk

− Hawk

), (Hawk

− Hawk

), Z)

′

(F.3)

The associated vector of coeﬃcients is θ

= (˜α

, ˜γ

)

′

. Let us further deﬁne W = I

⊗w

and Q = I

⊗ q where I

∈ R

2x2

denotes the identity matrix. Finally, we deﬁne the outcome

as X

= (Y

, .., Y

, G

, .., G

)

′

∈ R

2T ×1

, where each entry is given by X

j=0

t+j

, and

Y, G refer to GDP and government spending respectively. The instrumental variable estimate

of Θ

= (θ

′

, θ

′

)

′

follows from the standard formula,

= (Q

′

W )

−1

′

. (F.4)

We use Driscoll-Kraay standard errors to allow for serial correlation and cross-correlation

between cumulated GDP and cumulated government spending. The multiplier estimate is

a non-linear function of the estimate

. Standard errors can be computed with the delta

method as the above estimation procedure allows to compute an estimate of the full covariance

matrix.

Discussion. This procedure has an important advantage compared with the one-step estima-

tion procedure used by Ramey and Zubairy (2018). It admits multiplier estimation in one step

without using the ﬁscal shocks as an instrument for (cumulated) government spending. This

is important as Ramey and Zubairy (2018) have to use both the military spending shock and

the Blanchard-Perotti shock to obtain a suﬃciently strong instrument. Thus, their multiplier

estimates do not only hinge on the exogeneity assumption for the military shocks but also on

the exogeneity assumption for the Blanchard-Perotti shock. Our approach remains valid even

in samples in which the military spending shock has less explanatory power for government

spending. This is crucial when working with the military spending shocks in a sample that

starts after the Korean War (Ramey, 2011).

Appendix G Additional results for Section 4

This appendix contains additional ﬁndings discussed in the main text as well as the results of

our sensitivity analysis.

Note that the point estimates are identical to the natural plug-in estimator that one obtains when estimating

(F.1) in two-separate regressions.

ECB Working Paper Series No 2851

Table G.1: Responses of GDP and government spending, incl. ﬁrst-stage

GDP responses G responses First-stage results

Regressors (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

0.142 0.166 0.185 0.283 0.092 0.140 0.157 0.152 0.050 0.010

(0.096) (0.095) (0.085) (0.130) (0.047) (0.056) (0.051) (0.054) (0.039) (0.007)

(Hawk

− Hawk

) -1.672 -3.099 -2.485 -0.873 -0.342 -0.401 -0.030 0.220

(0.775) (0.841) (1.433) (1.174) (0.209) (0.258) (0.416) (0.653)

Hawk

− Hawk

-2.770 -3.698 -4.247 -4.562 -0.593 -0.985 -1.389 -0.948

(1.220) (1.728) (2.216) (2.217) (0.322) (0.650) (1.020) (1.135)

(Hawk

− Hawk

) 0.290 -0.019

(0.053) (0.021)

Hawk

− Hawk

-0.008 0.402

(0.017) (0.042)

t−1

0.024 0.057 0.086 0.245 0.044 0.076 0.092 0.124 0.007 0.011

(0.157) (0.216) (0.221) (0.153) (0.033) (0.046) (0.043) (0.043) (0.003) (0.006)

t−2

0.110 0.035 0.078 0.150 0.032 0.052 0.063 0.092 -0.012 0.007

(0.125) (0.185) (0.205) (0.160) (0.030) (0.030) (0.041) (0.049) (0.011) (0.008)

t−3

0.045 0.036 0.126 0.188 0.038 0.036 0.037 0.073 -0.000 0.008

(0.149) (0.163) (0.153) (0.144) (0.018) (0.028) (0.045) (0.052) (0.006) (0.008)

t−4

0.001 0.033 0.152 0.224 0.023 0.037 0.060 0.139 -0.018 0.004

(0.141) (0.125) (0.117) (0.144) (0.022) (0.027) (0.041) (0.038) (0.012) (0.010)

GDP

t−1

1.314 0.777 0.424 0.037 0.033 0.103 0.135 0.124 -0.000 -0.012

(0.182) (0.243) (0.252) (0.282) (0.053) (0.075) (0.100) (0.121) (0.013) (0.017)

GDP

t−2

-0.406 -0.166 -0.110 0.149 0.006 0.060 0.035 0.039 -0.016 0.013

(0.190) (0.209) (0.159) (0.197) (0.054) (0.072) (0.077) (0.094) (0.020) (0.014)

GDP

t−3

-0.240 -0.012 -0.093 0.081 0.062 0.034 0.044 0.084 0.004 -0.005

(0.203) (0.180) (0.223) (0.171) (0.055) (0.068) (0.068) (0.062) (0.016) (0.010)

GDP

t−4

-0.164 -0.440 -0.284 -0.444 -0.103 -0.218 -0.279 -0.355 0.003 -0.026

(0.183) (0.267) (0.313) (0.333) (0.051) (0.095) (0.138) (0.167) (0.007) (0.011)

t−1

-0.639 0.012 0.336 0.864 1.340 1.311 1.121 1.138 0.028 -0.073

(0.714) (1.012) (0.909) (0.940) (0.195) (0.241) (0.308) (0.387) (0.022) (0.055)

t−2

1.177 0.596 0.194 -0.223 0.008 -0.042 0.078 0.097 0.008 0.062

(0.617) (0.734) (0.602) (0.479) (0.195) (0.220) (0.277) (0.278) (0.039) (0.047)

t−3

-0.391 -0.347 -0.233 -0.519 -0.079 -0.106 -0.136 -0.138 0.017 -0.091

(0.651) (0.706) (0.618) (0.491) (0.203) (0.248) (0.273) (0.272) (0.054) (0.042)

t−4

0.022 0.020 0.018 0.162 -0.346 -0.308 -0.268 -0.343 -0.039 0.140

(0.920) (0.911) (0.888) (0.791) (0.202) (0.314) (0.437) (0.486) (0.049) (0.060)

Observations 196 196 196 196 196 196 196 196 196 196

0.577 0.347 0.201 0.138 0.934 0.843 0.730 0.646 0.452 0.547

excl. IVs 0.036 0.154

F-statistic 16.398 4.243 3.418 2.630 94.688 22.683 11.316 15.287 43.691 28.077

F-statistic excl. IVs 4.935 5.804

Notes: The table shows responses of real GDP and real government spending (G) to an expansionary military spending

shock, corresponding to one percent of GDP, conditional on systematic monetary policy (Hawk

). We show IV estimates

based on the local projection framework (3.4)-(3.5) as speciﬁed in Section 4.1. Columns (1) to (4) and (5) to (8) display the

one, two, three, and four-year ahead responses, respectively. Regressor ε

captures the responses when Hawk

equals its

sample average and ε

(Hawk

− Hawk

) captures the diﬀerential responses. Columns (9) and (10) display the ﬁrst-stage

results for ε

(Hawk

− Hawk

) and (Hawk

− Hawk

), respectively. The shaded areas indicate 68% and 95% conﬁdence

bands using Newey-West standard errors.

ECB Working Paper Series No 2851

Figure G.1: Responses of military spending shocks to systematic monetary policy

(a) Baseline model (δ

) (b) Linear model (δ

)

Notes: The ﬁgure shows responses of the military spending shock to systematic monetary policy (Hawk

). We show IV

estimates based on the local projection framework (3.4)-(3.5) as speciﬁed in Section 4.1. The δ

captures the diﬀerential

response when Hawk

exceeds the sample average by two hawks. Panel (a) shows the results for our baseline model whereas

Panel (b) shows the results when we restrict β

= γ

= 0 in the local projection (3.4). The shaded areas indicate 68% and

95% conﬁdence bands using Newey-West standard errors.

Table G.2: Testing for diﬀerences across regimes, p-values

Outcome

+2 Hawk

vs.

Average

+1 Hawks

vs.

Average

vs.

+1 Dove

Average

vs.

+2 Doves

Two-year horizon

Multiplier 0.223 0.119 0.102 0.104

GDP (cum) 0.000

G (cum) 0.080

Four-year horizon

Multiplier 0.245 0.122 0.041 0.041

GDP (cum) 0.008

G (cum) 0.448

Notes: The table shows p-values corresponding to statistical tests for whether the ﬁscal multiplier or its components

are signiﬁcantly diﬀerent across monetary regimes (Hawk

). The tests are based on the multiplier estimates reported in

Table 2 in Section 4.4, using Driscoll-Kraay standard errors, see Appendix F for details.

ECB Working Paper Series No 2851

Figure G.2: Responses of GDP and government spending, OLS

(a) Average GDP (β

) (b) Average G (β

)

) (d) Diﬀerential G (γ

)

Notes: The ﬁgure shows responses of real GDP and real government spending (G) to an expansionary military spending

shock, corresponding to one percent of GDP, conditional on systematic monetary policy (Hawk

). We show OLS estimates

based on the local projection framework (3.4) as speciﬁed in Section 4.1. The β

captures the responses when Hawk

equals its sample average. The γ

captures the diﬀerential responses when Hawk

exceeds the sample average by two

hawks. The shaded areas indicate 68% and 95% conﬁdence bands using Newey-West standard errors.

Figure G.3: Cumulated GDP responses for OLS and IV

2-year cumulated response (b) 3-year cumulated response

Notes: The ﬁgure shows the cumulative real GDP response to an expansionary military spending shock, corresponding to

one percent of GDP, conditional on systematic monetary policy (Hawk

). We show IV and OLS estimates based on the

local projection framework (3.4)-(3.5) as speciﬁed in Section 4.1. The displayed estimates are computed as

h=0

[β

(Hawk

− Hawk

)] for H = 8 quarters (Panel a) and H = 12 quarters (Panel b).

ECB Working Paper Series No 2851

Figure G.4: Weak instrument tests

(a) Montiel Olea and Pﬂueger (2013) (b) Lewis and Mertens (2022)

Notes: The ﬁgure shows p-values for rejecting the null of weak instruments for the responses of real government spending

(G), based on the local projection framework (3.4)-(3.5) as speciﬁed in Section 4.1. The Montiel Olea and Pﬂueger (2013)

test evaluates the null of the bias in γ

exceeding a threshold τ. Similarly, the Lewis and Mertens (2022) test evaluates

the null of the ℓ

norm of the bias in γ

and δ

exceeding a threshold τ . For the former, the endogenous regressor Hawk

is not tested but directly replaced by its ﬁrst stage ﬁtted value. The critical values and associated p-values are based on

Newey-West standard errors.

Figure G.5: Diﬀerential responses of GDP and government spending, reduced-form

(a) Diﬀerential GDP (γ

) (b) Diﬀerential G (γ

)

Notes: The ﬁgure shows diﬀerential responses of real GDP and real government spending (G) to an expansionary military

spending shock, corresponding to one percent of GDP, conditional on systematic monetary policy (Hawk

). We show

reduced-form estimates based on the local projection framework (3.4)-(3.5) as speciﬁed in Section 4.1. The γ

captures the

diﬀerential responses when Hawk

exceeds the sample average by two hawks. Moreover, testing whether γ

is statistically

signiﬁcant from zero is equivalent to testing for zero relevance of the instrument, as explained in the main text. The shaded

areas indicate 68% and 95% conﬁdence bands using Newey-West standard errors.

ECB Working Paper Series No 2851

Figure G.6: Responses of GDP and government spending, robust inference

(a) Diﬀerential GDP (γ

) (b) Diﬀerential G (γ

)

Notes: The ﬁgure shows diﬀerential responses of real GDP and real government spending (G) to an expansionary military

spending shock, corresponding to one percent of GDP, conditional on systematic monetary policy (Hawk

). We show IV

estimates based on the local projection framework (3.4)-(3.5) as speciﬁed in Section 4.1. The γ

captures the diﬀerential

responses when Hawk

exceeds the sample average by two hawks. The shaded areas indicate 68% and 95% conﬁdence

bands using Newey-West standard errors. The dashed bands provide 95% conﬁdence sets, robust to weak identiﬁcation

based on Andrews (2018), constructed via the reﬁned projection method from Chaudhuri and Zivot (2011).

ECB Working Paper Series No 2851

Table G.3: Cumulative 4-year government spending multipliers, Robustness

Multipliers across regimes

p-values for diﬀerences

across regimes

Speciﬁcation +2 Hawks Average +2 Doves

+2 Hawks

vs.

+2 Doves

+2 Hawks

vs.

Average

vs.

+2 Doves

Baseline -1.790 1.308 3.095

0.122 0.245 0.041

(2.637) (0.475) (1.162)

BP shock 0.880 1.359 1.735

0.082 0.096 0.068

(1.062) (0.837) (0.698)

Aggregation schemes

Median 0.421 1.412 2.225

0.043 0.065 0.036

(0.568) (0.543) (0.811)

Chair weight -1.676 1.531 3.462

0.070 0.175 0.077

(2.223) (0.661) (1.514)

Swinger weight -1.599 1.260 3.037

0.090 0.179 0.043

(2.165) (0.551) (1.138)

Drop swingers from IV 0.227 1.590 4.069

0.113 0.531 0.380

(2.262) (0.978) (3.113)

Accounting for trends

5-year MA -10.655 1.206 3.551

0.792 0.821 0.220

(53.169) (1.177) (1.966)

10-year MA -4.799 0.779 3.049

0.452 0.556 0.092

(10.138) (0.877) (1.162)

15-year MA -2.114 0.931 2.822

0.146 0.277 0.036

(3.010) (0.426) (0.869)

Accounting for the ZLB

End sample ’08 -2.002 1.299 3.088

0.108 0.225 0.032

(2.672) (0.510) (1.133)

End sample ’07 -3.380 0.911 3.005

0.203 0.343 0.031

(4.499) (0.527) (1.132)

(Table continues on the next page)

ECB Working Paper Series No 2851

Table G.3 (continued): Cumulative 4-year government spending multipliers, Robustness

Multipliers across regimes

p-values for diﬀerences

across regimes

Speciﬁcation +2 Hawks Average +2 Doves

+2 Hawks

vs.

+2 Doves

+2 Hawks

vs.

Average

vs.

+2 Doves

Additional controls

Interest rates 0.388 1.251 1.848

0.306 0.327 0.354

(1.272) (0.688) (0.755)

Interest rates,

inﬂation

0.743 1.254 2.033

0.335 0.315 0.386

(0.873) (0.646) (1.038)

Interest rates,

inﬂation, surplus

0.661 1.317 2.188

0.379 0.335 0.467

(1.111) (0.855) (1.425)

Non-linear controls

in t -1.043 1.414 2.813

0.208 0.357 0.068

(2.657) (0.554) (1.132)

in t, ..., t − 4 0.333 2.004 3.022

0.340 0.483 0.137

(2.298) (0.637) (1.154)

Notes: The table shows IV estimates of the cumulative ﬁscal spending multipliers F M

(χ) in equation (4.1) for H = 16

quarters. The last three columns show p-values corresponding to statistical tests for whether the ﬁscal multiplier is

signiﬁcantly diﬀerent across monetary regimes (Hawk

). The baseline coeﬃcients are estimated using a cumulative version

of the local projection framework (3.4)-(3.5) as speciﬁed in Section 4.1. The columns present diﬀerent states of the Hawk-

Dove balance between “+2 Hawks” (χ = +2/12), “Average” (χ = 0), and “+2 Doves” (χ = −2/12). Driscoll-Kraay

standard errors are in parenthesis, see Appendix F for details. BP shock: The shock is contemporaneous G, conditional

on controls that include four lags of real GDP and real government spending, as well as the projected growth rate of real

government spending. The projected growth rate is taken from the Survey of Professional Forecasters and is available from

1969 onward, which is the start of our sample, see Appendix B. Aggregation schemes: We use three variants of Hawk

Median: We aggregate the cross-section of FOMC members by the median, instead of the arithmetic average. Chair weight:

We assign the preferences of the Fed Chair twice the weight of an ordinary member when aggregating to Hawk

. Swinger

weight: We do not discriminate between swingers and consistent members. Finally, we use an alternative deﬁnition of the

instrumental variable Hawk

by setting swingers to zero before aggregating to a time series. Accounting for trends: We

use three variants of Hawk

where we subtract the backward-looking 5, 10, or 15-year moving average from Hawk

prior

estimation. Accounting for the ZLB: We use a sub-sample that ends either in 2008Q4 or 2007Q4 to exclude the ZLB, or

both the ZLB and the Great Recession. Additional controls: We augment the control vector Z

gradually by four lags of

treasury yields with 1-year and 10-year maturity, the fed funds rate (interest rates), CPI inﬂation, and the primary surplus

from Cochrane (2022). Non-linear controls: We augment the control vector Z

by interacting the baseline control vector

also with Hawk

and instrument these controls accordingly (in t). We further augment the control vector by including

and instrumenting lagged interaction terms, i.e. Hawk

t−i

× C

t−i

with i = 1, ..., 4 and C

referring to G, GDP, and ε

(in

t, ..., t − 4).

ECB Working Paper Series No 2851

Table G.4: Cumulative 4-year government spending multipliers, Discrete Hawk-Dove balance

Multipliers across regimes

p-values for diﬀerences

across regimes

Speciﬁcation Hawkish Average Dovish

Hawkish

vs.

Dovish

Hawkish

vs.

Average

vs.

Dovish

Quartiles -6.002 1.727 4.814

0.264 0.460 0.201

(10.343) (0.775) (2.774)

Tertiles -3.481 0.490 2.835

0.336 0.488 0.047

(6.227) (0.772) (1.083)

Notes: The table shows IV estimates of the cumulative ﬁscal spending multipliers F M

(χ) in equation (4.1) for H = 16

quarters. The last three columns show p-values corresponding to statistical tests for whether the ﬁscal multiplier is

signiﬁcantly diﬀerent across monetary regimes (Hawk

). The coeﬃcients are estimated using a cumulative version of the

local projection framework (3.4)-(3.5) as speciﬁed in Section 4.1. We use two discrete variants of Hawk

. We deﬁne that

the discrete Hawk

equals -1 if Hawk

falls below the ﬁrst quartile or tertile of the distribution of Hawk

over time, +1 if

above the highest quartile or tertile, and zero else. The columns present diﬀerent states of the Hawk-Dove balance between

“Hawkish” (χ within the last quartile or tertile), “Average” (χ between the ﬁrst and last quartile or tertile) “Dovish” (χ

within the ﬁrst quartile or tertile).

Appendix H Additional results for Section 5

This appendix contains additional ﬁndings discussed in the main text.

ECB Working Paper Series No 2851

Figure H.1: Responses of nominal interest rates, omit shocks at end of rotation cycle

(a) Average FFR (β

) (b) State-dependent FFR (β

± γ

)

) (d) State-dependent 1-year rate (β

± γ

)

(e) Average 10-year rate (β

) (f) State-dependent 10-year rate (β

± γ

)

Notes: The ﬁgure shows responses of the federal funds rate (FFR), as well as the 1-year and 10-year treasury yields to an

expansionary military spending shock, corresponding to one percent of GDP, conditional on systematic monetary policy

(Hawk

). All outcomes are annualized interest rates. We show IV estimates based on the local projection framework

(3.4)-(3.5) as speciﬁed in Section 5.1. The β

captures the responses when Hawk

equals its sample average. The β

± γ

shows the state-dependent responses when Hawk

exceeds the sample average either by two hawks (+2 Hawks) or by two

doves (+2 Doves). The shaded areas indicate 68% and 95% conﬁdence bands using Newey-West standard errors. We set

the military spending shocks occurring in Q3 or Q4 to zero.

ECB Working Paper Series No 2851

Figure H.2: Responses of GDP and government spending, omit shocks at end of rotation cycle

(a) Average GDP (β

) (b) State-dependent GDP (β

± γ

)

) (d) State-dependent G (β

± γ

)

Notes: The ﬁgure shows responses of real GDP and real government spending (G) to an expansionary military spending

shock, corresponding to one percent of GDP, conditional on systematic monetary policy (Hawk

). We show IV estimates

based on the local projection framework (3.4)-(3.5) as speciﬁed in Section 4.1. The β

captures the responses when Hawk

equals its sample average. The β

± γ

shows the state-dependent responses when Hawk

exceeds the sample average

either by two hawks (+2 Hawks) or by two doves (+2 Doves). The shaded areas indicate 68% and 95% conﬁdence bands

using Newey-West standard errors. We set the military spending shocks occurring in Q3 or Q4 to zero.

ECB Working Paper Series No 2851

Figure H.3: Responses of real interest rates

(a) Average real FFR (β

) (b) State-dependent real FFR (β

± γ

)

) (d) State-dependent real 1-year rate (β

± γ

)

(e) Average real 10-year rate (β

) (f) State-dependent real 10-year rate (β

± γ

)

Notes: The ﬁgure shows responses of the real federal funds rate (FFR), as well as the 1-year and 10-year real treasury yields

to an expansionary military spending shock, corresponding to one percent of GDP, conditional on systematic monetary

policy (Hawk

). We show IV estimates based on the local projection framework (3.4)-(3.5) as speciﬁed in Section 5.1.

All outcomes are annualized ex-ante real interest rates which we compute as nominal rate minus one-year ahead inﬂation

expectations according to the Livingston Survey, see Appendix B for details. The β

captures the responses when Hawk

equals its sample average. The β

± γ

shows the state-dependent responses when Hawk

exceeds the sample average

either by two hawks (+2 Hawks) or by two doves (+2 Doves). The shaded areas indicate 68% and 95% conﬁdence bands

using Newey-West standard errors.

ECB Working Paper Series No 2851

Figure H.4: Decomposing the GDP response, private spending

(a) Average consumption (β

) (b) State-dependent consumption (β

± γ

)

) (d) State-dependent investment (β

± γ

)

Notes: The ﬁgure shows responses of real private consumption and real private investment to an expansionary military

spending shock, corresponding to one percent of GDP, conditional on systematic monetary policy (Hawk

). We show IV

estimates based on the local projection framework (3.4)-(3.5) as speciﬁed in Section 4.1. The β

captures the responses

when Hawk

equals its sample average. The β

± γ

shows the state-dependent responses when Hawk

exceeds the sample

average either by two hawks (+2 Hawks) or by two doves (+2 Doves). The shaded areas indicate 68% and 95% conﬁdence

bands using Newey-West standard errors. We modify the control vector to include four lags of consumption, investment,

and government spending, as well as the shock and a residual component of GDP, which we compute as GDP minus

consumption, investment, and government spending.

ECB Working Paper Series No 2851

Figure H.5: Decomposing the GDP response, government spending

(a) Average non-military G (β

) (b) State-dependent non-military G (β

± γ

)

) (d) State-dependent military G (β

± γ

)

Notes: The ﬁgure shows responses of real government spending (for military and non-military purposes) to an expansionary

military spending shock, corresponding to one percent of GDP, conditional on systematic monetary policy (Hawk

). We

show IV estimates based on the local projection framework (3.4)-(3.5) as speciﬁed in Section 4.1. The β

captures the

responses when Hawk

equals its sample average. The β

± γ

shows the state-dependent responses when Hawk

exceeds

the sample average either by two hawks (+2 Hawks) or by two doves (+2 Doves). The shaded areas indicate 68% and

95% conﬁdence bands using Newey-West standard errors. We modify the control vector to include four lags of military and

non-military government spending as well as the shock and real GDP.

ECB Working Paper Series No 2851

Acknowledgements

An earlier version of this paper was titled “Systematic Monetary Policy and the Effects of Government Spending.”

We thank Francesco Fusari, Karel Mertens, Alexander Meyer-Gohde, and Filippo Palotti for insightful discussions, and Klaus Adam,

Regis Barnichon, Antoine Camous, Antonio Ciccone, John Cochrane, Davide Debortoli, Refet Gürkaynak, Peter Karadi, Michael

McMahon, Evi Pappa, Morten Ravn, Jon Steinsson and Christian Wolf, as well as participants at various seminars and conferences for

helpful comments. Lukas Hack and Matthias Meier acknowledge financial support from the German Research Foundation (DFG)

through CRC TR 224 (Project C05). Matthias Meier acknowledges financial support from the UniCredit & Universities Foundation.

The views expressed in this article are solely the responsibility of the authors, and should not be interpreted as reflecting the views of

the Banque de France or the Eurosystem.

Lukas Hack

University of Mannheim, Mannheim, Germany; email: lukas.hack@uni-mannheim.de

Klodiana Istrefi

European Central Bank, Frankfurt am Main, Germany; Centre for Economic Policy Research, London, United Kingdom;

email: klodiana.i[email protected]pa.eu

Matthias Meier

University of Mannheim, Mannheim, Germany; email: m.meier@uni-mannheim.de

Postal address 60640 Frankfurt am Main, Germany

Telephone +49 69 1344 0

Website www.ecb.europa.eu

electronically, in whole or in part, is permitted only with the explicit written authorisation of the ECB or the authors.

This paper can be downloaded without charge from www.ecb.europa.eu, from the Social Science Research Network electronic library or

from RePEc: Research Papers in Economics. Information on all of the papers published in the ECB Working Paper Series can be found

on the ECB’s website.

PDF ISBN 978-92-899-6216-2 ISSN 1725-2806 doi:10.2866/34644 QB-AR-23-088-EN-N