Monetary Policy with Heterogeneous Agents:

Insights from TANK models ∗

Davide Debortoli Jordi Galí
UPF, CREI and Barcelona GSE CREI, UPF and Barcelona GSE

September 2017
Preliminary and Incomplete

Abstract

Heterogeneous agents New Keynesian (HANK) models are shown to differ from their repre-
sentative agent (RANK) counterparts along two dimensions: differences in average consump-
tion at any point in time between constrained and unconstrained households, and consumption
heterogeneity within the subset of unconstrained households. These two factors are captured
in a simple way by two "wedges" that appear in an aggregate Euler equation, and whose
behavior can be traced in response to any aggregate shock, allowing us to assess their quanti-
tative significance. A simple two-agent New Keynesian (TANK) model abstracts completely
from heterogeneity within unconstrained agents, but is shown to capture reasonably well the
implications of a baseline HANK model regarding the effects of aggregate shocks on aggregate
variables. We discuss the implications of our findings for the design of optimal monetary
policy.
JEL Classification Numbers: E21, E32, E50.
Keywords: heterogeneous agents, new keynesian models, monetary policy.

∗
We thank Cristina Manea for excellent research assistance. We have benefitted from comments by participants
at CREI Macro Lunch, and 2017 NBER Summer Institute. Davide Debortoli acknowledges financial support from
the European Union’s Horizon 2020 Program under grant agreement 649396 (ADEMU).

1
1 Introduction

A growing literature has emerged in recent years that aims at re-examining some important macro
questions through the lens of monetary models with heterogenous agents. Models in this literature
commonly assume the presence of idiosyncratic shocks to individuals’ income, together with the
existence of incomplete markets and borrowing constraints. Those features are combined with
the kind of nominal rigidities and monetary non-neutralities that are the hallmark of New Key-
nesian models. Following Kaplan et al. (2016), we refer to those models as HANK models (for
“Heterogenous Agent New Keynesian"models).

Two key lessons can be drawn from this literature. Firstly, taking into account agents’ het-
erogeneity is important in order to understand the transmission of monetary policy, including the
relative contribution of direct and indirect effects (Kaplan et al. (2016)) or its redistributive ef-
fects across income groups (Auclert (2016)). Secondly, the transmission of monetary policy and its
aggregate effects may vary significantly depending on the prevailing fiscal policy, as the latter de-
termines how the implementation of monetary policy affects the distribution of individual income
and wealth among agents with different marginal propensities to consume.

As is well known, solving for the equilibria of HANK models requires the use of nontrivial
computational techniques, given the need to keep track of the distribution of wealth, and the
hurdles arising from the presence of occasionally binding borrowing constraints. The reliance on
numerical techniques for the analysis of those models often presents a challenge when it comes to
understanding the mechanisms underlying some of the findings, and may thus limit their use in
the classroom or as an input in policy institutions.

The purpose of the present paper is to assess the merits of HANK models for our understanding
of an economy’s aggregate behavior, relative to a simpler alternative that assumes the existence
of two types of consumers –"Ricardian" and "Keynesian"– with constant shares in the popula-
tion, while allowing only for aggregate shocks (i.e. disregarding idiosyncratic shocks). Ricardian

1
consumers are assumed to have full access to financial markets (including markets for stocks and
bonds), while Keynesian consumers are assumed to behave in a "hand-to-mouth"fashion, consum-
ing their current labor income at all times. This will be the case if they do not have access to
financial markets, find themselves continuously against a binding borrowing borrowing constraint,
or have a pure myopic behavior. Following Kaplan et al. (2016), we refer to those models as TANK
models (short for "Two Agent New Keynesian").

HANK and TANK models share a key feature that is missing in representative agent models,
namely, the fact that at any point in time a fraction of agents face a binding borrowing constraint
(or behave as if they did), and thus do not adjust their consumption in response to changes in
interest rates or variables other than current income. This fact generally implies that the economy’s
response to shocks and the possible role of monetary policy in shaping that response will differ
from the standard NK model with a representative agent (henceforth, RANK, for short).

On the other hand, TANK models differ from HANK models in three important ways. First,
the fraction of agents who are subject (or act as if subject) to a binding borrowing constraint does
not change over time in response to shocks, whereas in HANK models that fraction is endogenous
and may vary over time, as a result of the interaction of aggregate shocks and the distribution and
composition of wealth at any point in time. Secondly, TANK models assume away the impact on
agents’ current decisions of the likelihood of being financially constrained in the future, whereas
the latter possibility-–typically associated to the presence of idiosyncratic shocks—is a source of
(time-varying) precautionary savings in HANK models. Finally, the analysis of TANK models is
highly simplified relative to their HANK counterparts for there is no need to keep track of the
wealth distribution and its changes over time. In fact, as we show below, the implied equilibrium
conditions of a baseline TANK model can be reduced to a system of difference equations isomorphic
to of the standard NK model.

A first goal of this paper is to assess the extent to which TANK models can be viewed as

2
a tractable framework that captures well the main predictions of HANK models regarding the
economy’s response to aggregate shocks, monetary as well as non-monetary. In case of a favorable
answer, TANK models would be an appealing option as a laboratory for the study of the effects
of aggregate shocks, and the role of monetary and fiscal policy in shaping those effects, in the
presence of heterogenous agents.

With that goal in mind, we propose a simple framework, related to Werning (2015), that allows
us to identify two dimensions of heterogeneity that explain the differential behavior of a HANK
economy relative to its RANK counterpart. The first dimension of heterogeneity is given by the
difference in average consumption between constrained and unconstrained households at any point
in time. The dispersion in consumption within the subset of unconstrained households at any point
in time constitute the second relevant source of heterogeneity. HANK models generally account
for both dimensions of heterogeneity, whereas TANK models only capture the first dimension.
Furthermore, and as we show below, these two dimensions of heterogeneity are captured in a
simple way by two "wedges" that appear in an Euler equation for aggregate consumption, and
whose behavior can be traced in response to any aggregate shock, allowing us to assess their
quantitative significance.

Using that framework as a reference, we seek to understand the differences and similarities
in the predictions of HANK, TANK and RANK models with regard to the aggregate effects of
aggregate shocks.

A key finding of our analysis is that a simple TANK model approximates well, both from a
qualitative and a quantiative viewpoint, the aggregate dynamics of a canonical HANK model in
response to aggregate shocks, monetary and non-monetary. This is because of two reasons. On
the one hand, for standard calibrations of a HANK model, consumption heterogeneity between
constrained and unconstrained households fluctuates significantly in response to aggregate shocks,
and its fluctuations are well captured by a TANK model. On the other hand, consumption het-

3
erogeneity within the subset of unconstrained households remains roughly constant, since those
agents are able to limit consumption fluctuations by borrowing and saving. Thus, ignoring this
second form of heterogeneity –as in a TANK model– is largely inconsequential for determening the
behaviour of aggregate variables. This result is robust to alternative calibrations of the borrowing
limits, and specifications of the fiscal transfers, which in turn are among the main drivers of the
differences between HANK and RANK models.

The previous finding suggests that a TANK model may be used to obtain analytical results
that provide useful insights on the role of heterogeneity in more general HANK models. A first
insight in that regard pertains to the role of heterogeneity for the transmission of monetary policy
shocks. In particular, we show analytically that households’ heterogeneity may amplify or dampen
the effects of aggregate shocks depending on the size of constrained agents and the cyclicality of
fiscal transfers, among other factors, but independently of the magnitude of nominal rigidities, or
the way monetary policy is conducted.

A second insight pertains to the implications of heterogeneity for the design of monetary policy.
We use a TANK model to show that central banks face a non-trivial policy trade-off between sta-
bilizing inflation and a measure of consumption heterogeneity. That trade-off is, of couse, assumed
away in standard RANK models. In our TANK model, the objective function of a benevolent cen-
tral bank can be approximated with a simple (quadratic) loss function that penalizes fluctuations
in consumption heterogeneity, in addition to fluctuations in inflation and output gap. A policy
trade-off emerges whenever fluctuations in (the natural level of) output are not proportionally
distributed across households. In that case, it is not possible to simultaneously stabilize inflation,
the output-gap and the heterogeneity index —i.e. the “divine coincidence" does not hold. The
central bank may then tolerate some deviation from inflation and output from their respective tar-
gets in order to avoid too large fluctuations in consumption dispersion. However, we find that for
standard calibrations of the TANK model, the optimal policy still implies minimimal fluctuations

4
in inflation and the output-gap, and is thus nearly identical to the one that would prevail in a
RANK model.

The paper is related to two main strands of the literature. On the one hand, the emerging
literature introducing New-Keynesian features into heterogeneous agent models with idiosyncratic
risk and incomplete markets. Some examples are the works of McKay et al. (2016), Gornemann
et al. (2016), Kaplan et al. (2016), McKay and Reis (2016), Werning (2015), Auclert (2017), and
Ravn and Sterk (2014), among many others. The main difference with respect to that literature is
the development of a simple TANK model with no idiosyncratic shocks, which admits an analytical
solution, and the emphasis on the distinction between constrained and unconstrained households.
On the other hand, the paper builds on the earlier literature on two-agent models, such as Campbell
and Mankiw (1989), Galí et al. (2007), Bilbiie (2008), Bilbiie and Straub (2013) and Broer et.
al. (2016). The main difference with respect to that literature is the comparison, both from a
theoretical and a quantitative viewpoint, with more general heterogenous agent models.1 Our
analysis of optimal monetary policy in a TANK model is related to the previous works of Cúrdia
and Woodford (2010) and Nisticó (2016), who neverthless emphasize the importance of financial
frictions, rather than consumption heterogeneity as we do here. Finally, Bilbiie and Ragot (2017)
study optimal monetary policy in an TANK model where money is used to provide liquidty to
financially constrained agents. Differently from that work, we study the monetary policy trade-
offs in a cashless economy.

The paper is organized as follows. In section 1 we introduce our proposed framework and
discuss its potential usefulness. In section 2 we lay out a baseline TANK model and derive the
corresponding equilibrium conditions. In section 3 we derive the main predictions of the model
regarding the effects of three aggregate shocks –monetary policy, demand and technology– under
1
In indedendent work, Bilbiie (2017) uses a TANK model to illustrate the "direct" and "indirect" effects of
monetary policy shocks emphasized by Kaplan et al. (2016) in a more general HANK model. Also, Ravn and Sterk
(2017) build a tractable heterogeneous agent model with nominal rigidities and labor market frictions, giving rise
to endogenous unemployment risk.

5
alternative assumptions regarding the transfer policies in place, and discuss the main differences
with respect to the predictions of a baseline RANK model. In section 4 we carry out a quantitative
comparison of the predictions the baseline TANK model to those of a HANK model. Section 5
analyzes the optimal monetary policy in a TANK model, while section 6 summarizes our main
findings and concludes.

2 Heterogeneity and Aggregate Consumption: An Organiz-

ing Framework

We start out by describing an equilibrium condition for aggregate consumption that we use as a
simple organizing device to think about the implications of heterogeneity for aggregate demand.
Our approach is related to Werning (2015), but differs from the latter in the emphasis we attach
to the distinction between constrained and unconstrained households in our interpretation of the
deviations from the standard Euler equation of a representative household model.

Consider an economy with a continuum of heterogeneous households, indexed by s ∈ [0, 1].

Each household seeks to maximize utility E0 ∞ t s s
P
t=0 β U (Ct , Nt ; Zt ), where Zt is a preference shifter

following a stationary process. We specialize the utility function to be of the form U (C, N ; Z) ≡
 1−σ 
C −1 N 1+ϕ
1−σ
− 1+ϕ
Z. Let Θt ⊂ [0, 1] denote the set of households that in period t have unconstrained
access to a market for one-period bonds yielding a (gross) riskless real return Rt .2

Next we derive a generalized Euler equation for aggregate consumption for economies with
heterogeneous households. We proceed in two steps. First we derive an Euler equation in terms
of aggregate consumption for unconstrained households. Then we rewrite that Euler equation in
terms of aggregate consumption.
2
In the present section, we assume a riskless real bond for notational convenience. The analysis carries over to
the case of a nominally riskless bond, as assumed below.

6
2.1 A Generalized Euler Equation for Unconstrained Households’ Con-
sumption

For any household s ∈ Θt the following Euler equation is satisfied:

Zt (Cts )−σ = βRt Et Zt+1 (Ct+1

s
)−σ .

(1)

Averaging over all s ∈ Θt ,

Z Z 
Zt s −σ βRt s −σ
(C ) ds = Et Zt+1 (Ct+1 ) ds (2)
1 − λt s∈Θt t 1 − λt s∈Θt
where λt is the measure of constrained households in period t, i.e. households for which (1) is not
satisfied.
1
R
Let CtR ≡ 1−λt s∈Θt
Cts ds denote the average consumption of unconstrained households in
period t. Taking a second order Taylor expansion of the left hand side of (2) around CtR we obtain
(see Appendix for details):

Z  
Zt σ(1 + σ)
(Cts )−σ ds
' Zt (CtR )−σ
1+ s
vars|t {ct }
1 − λt s∈Θt 2
1
R
where vars|t {cst } ≡ 1−λt
(cs − cR
s∈Θt t
2 s s R R 3
t ) ds, with ct ≡ log Ct and ct ≡ log Ct . Similarly,

Z  
Zt+1
−σ σ(1 + σ)
s
(Ct+1 )−σ ds ' R
Zt+1 Ct+1 Φmean
t 1+ s
vars|t {ct+1 }
1 − λt s∈Θt 2
 R −σ
1
R C 1
R
where vars|t {cst+1 } ≡ 1−λt s∈Θt
(cst+1 −cR 2 mean
t+1 ) ds and Φt ≡ Ct+1|t
R
R
, with Ct+1|t ≡ 1−λ t
C s ds.
s∈Θt t+1
t+1

Combining the above results, we can write an approximate Euler equation in terms of average
consumption for unconstrained households as:

Zt (CtR )−σ = βRt Et Zt+1 (Ct+1

R −σ

) Vt+1 (3)
1
3
(cs − cR 2
R
Note that, strictly speaking 1−λ t s∈Θt t t ) ds is the cross-sectional variance of (log) consumption only
approximately, since cR s
t is the cross-sectional mean of ct only up to a second order approximation.

7
where
Vt+1 ≡ Φmean var
t+1 Φt+1

2+σ(1+σ)vars|t {cst+1 }
with Φvar
t+1 ≡ 2+σ(1+σ)vars|t {cst }
.

Notice that Φmean

t and Φvar
t and, as a result, the wedge Vt , capture different aspects of the
dynamics of heterogeneity within unconstrained households. If the set of households that are
unconstrained is time invariant and consumption is equalized among those households, then
Φmean
t = Φvar
t = 1 and, hence, Vt = 1 for all t. The previous condition is trivially satisfied in
a TANK model like the one analysed below.

More generally, in economies with more complex heterogeneity dynamics, one would expect Vt
to differ from unity and to vary over time in response to all kinds of shocks. To the extent that
R
consumption of unconstrained households is not lower than that of constrained ones, then Ct+1|t ≤
R
Ct+1 and Φmean
t+1 ≥ 1. Similarly, one would expect the dispersion of consumption across households

that are unconstrained in period t, given by vars|t {cst }, to be no larger than the corresponding
dispersion in t + 1 for the same subset of households, given by vars|t {cst+1 }, since in period t + 1 a
fraction of those households may have become constrained. In that case Φvar
t+1 ≥ 1 will obtain. If

one of the previous conditions is satisfied, as is likely, with strict inequality, then Vt+1 > 1, which
captures a precautionary savings motive among unconstrained households, i.e. it leads to lower
consumption today, for any given interest rate and expected consumption next period.

Yet, it is an open question, both empirical and theoretical, whether deviations of the sequence
{Vt } from unity are quantitatively significant or not and, more precisely, whether they imply
R −σ R −σ
 
deviations between Et Zt+1 (Ct+1 ) Vt+1 and Et Zt+1 (Ct+1 ) in response to shocks that are
R
of the same order of magnitude as Ct+1 /CtR .

2.2 A Generalized Euler Equation for Aggregate Consumption

R1
Next we proceed to rewrite (3) in terms of aggregate consumption Ct ≡ 0
Cts ds as follows:

8
 Zt Ht−σ (Ct )−σ = βRt Et Zt+1 Ht+1
−σ
(Ct+1 )−σ Vt+1

(4)

where Ht is an index of heterogeneity between constrained and unconstrained households defined

as:
CtR
Ht ≡ .
Ct
Consider a steady state with no aggregate shocks, Z = 1, constant Y, H and V and, hence,
βRV = 1. Log-linearization of (4) around that steady state combined with the goods market
clearing condition Yt = Ct yields:
1 1 1
yt = Et {yt+1 } − rbt − Et {∆zt+1 } + Et {∆ht+1 } − Et {b
vt+1 },
σ σ σ
where yt ≡ log Yt , ht ≡ log Ht , zt ≡ log Zt , rbt ≡ log(βV Rt ), and vbt ≡ log(Vt /V ).

Assuming limT →∞ Et {ct+T } = c, limT →∞ Et {ht+T } = h, and limT →∞ Et {zt+T } = 0, we can

derive the aggregate demand equation:
1 L

ybt = − rbt − zt − b
ht − fbt , (5)
σ
where rbtL and ft are defined recursively as

rbtL = Et {bL
rt+1 } + rbt

1
fbt = Et {fbt+1 } + Et {b
vt+1 }.
σ
Thus, output is seen to depend on four factors: (i) the exogenous demand shifter zt (ii) current
and expected real interest rates, as summarized by rbtL , and whose responses to shocks of all kinds
will be determined by the monetary policy rule in place, (iii) the h-index of heterogeneity between
constrained and unconstrained households, and (iv) variable fbt which is a function of expected
future distributions of consumption within unconstrained households.

9
 ht = vbt = 0 for all t. Thus, the extent to which
Note that in the representative household model, b
aggregate demand and output in a HANK model behave differently from their RANK counterparts,
conditional on the monetary policy stance (as summarized by rbtL ) will depend on the endogenous
ht and vbt to different aggregate shocks. In particular, an increase (decrease) in the
response of b
h-index in response to a shock that raises output will dampen (amplify) that response relative to
a RANK economy, ceteris paribus (i.e. conditional on a given real rate response). In other words,
the aggregate effects of a given shock is seen to depend on the extent to which it leads towards
a redistribution towards/from those unconstrained households. The extent of that redistribution
may in turn be a consequence of (i) differential characteristics of unconstrained households (e.g.
their participation in labor markets and/or ownership of firms) and/or (ii) the nature of the
transfer/tax regime in place. An analogous argument can be made about variable ft , though in
the latter case only the extent of redistribution within the subset of unconstrained households at
any point in time is relevant.

Given the above considerations, the analysis of the determinants of the dynamics of ht and vt
and their response to aggregate shocks should contribute to our understanding of the implication
of heterogeneity. Next we put that idea into practice when analyzing the properties of a particular
instance of a HANK model, which we refer to as the baseline TANK model, and for which all
deviations from its RANK counterpart are the result of variations in ht , since the assumption of a
representative unconstrained household implies that vt = 0 for all t.

3 A Baseline TANK Model

We consider an economy with two types of households, a continuum of firms, a monetary authority
and a fiscal authority.4 This simple model is used as a laboratory to study the implications of
4
Campbell and Mankiw (1989) introduced the two-agent framework to account for empirical deviations from
the permanent-income hypothesis. Gal that framework into a New Keynesian model in order to re-examine the
conditions for equiligrium uniqueness, and to account for the effects of government purchases on consumption. See

10
agents’ heterogeneity for the behavior of aggregate variables in response to aggregate shocks. As
in Werning (2015), our focus is on understanding the determinants of aggregate consumption. The
supply side of the model is kept relatively simple, and such that it can be directly embedded in
the richer HANK models considered below.

3.1 Households

All households are assumed to have identical preferences, given by

∞
X
E0 β t U (Cts , Nts ; Zt )
t=0

R 1
  −1
p
1 1− p
where Cts ≡ 0
Cts (i) p di is a consumption index, with Cts (i) denoting the quantities
consumed of good i ∈ [0, 1]. Nts denotes hours of work. The superindex s ∈ {R, K} specifies
the household type (“Ricardian” or “Keynesian”), as discussed below. β ∈ (0, 1) is the discount
factor. Zt is an exogenous preference shock, whose evolution is described by the log-normal process
zt = ρz zt−1 + εzt , where zt ≡ log Zt and εzt ∼ N (0, σz2 ). Period utility is specialized to be of the
form:
C 1−σ − 1 N 1+ϕ
 
U (C, N ; Z) ≡ − Z (6)
1−σ 1+ϕ
A constant measure 1 − λ of households, which we refer to as " Ricardian," have unconstrained
access to financial markets. In particular, they can trade two types of assets: one-period nominally
riskless bonds and shares in a fund that owns all firms. Their period budget constraint is given by

1 1 BtR R
Z
R R Bt−1 (1 + it−1 )
Pt (i)Ct (i)di + + Qt St = + Wt NtR + (Dt + Qt )St−1
R
+ TtR (7)
Pt 0 Pt Pt

where BtR and StR denote, respectively, holdings of bonds and shares in a stock market fund. TtR
denotes lump-sum tranfers (or taxes, if negative). Qt is the price of a share in the stock fund,
also Bilb of a version of TANK closer to the one considered here (i.e. without physical capital).

11
while Dt denoting the corresponding dividends (all of them in terms of the consumption bundle).
R  1−1
1 1−p p
it the nominal yield on bonds. Wt is the (real) wage. Pt ≡ 0 Pt (i) di is a price index.

A fraction λ of households, henceforth referred to as “Keynesian” just consume their labor

income net of taxes (or transfers) each period, possibly (but not necessarily) because they do not
have access to financial markets. Formally

Z 1
1
Pt (i)CtK (i)di = Wt NtK + TtK (8)
Pt 0

where CtK , NtK and TtK denote, respectively, consumption, hours worked and transfers for Keyne-
sian households.

Importantly, and in contrast with many of the HANK models in the literature, the two types
of agents do not face any form of idiosyncratic uncertainty. Furthermore, we assume that they
take the wage as given and are happy to supply as much labor as demanded by firms (i.e. they
also take NtR and NtK as given, more on this below).
R1
Optimal allocation of expenditures across goods implies Cts (i) = (Pt (i)/Pt )−p Cts and 0
Pt (i)Cts (i)di =
Pt Cts for s ∈ {R, K}. In addition, the intertemporal optimality conditions for Ricardian households
take the form  
1 Pt
= Et Λt,t+1 (9)
1 + it Pt+1

Qt = Et {Λt,t+1 (Qt+1 + Dt+1 )} (10)

R

−σ
where Λt,t+1 ≡ β(Zt+1 /Zt ) Ct+1 /CtR is the stochastic discount factor for one-period ahead
(real) payoffs.

As emphasized above, our focus is on aggregate demand dynamics, so we keep the supply side
of the model as simple as possible. In particular we assume a wage schedule

Wt = Mw Ctσ Ntϕ (11)

12
for all j ∈ [0, 1], where Mw > 1 can be interpreted as the gross average wage markup. Such a wage
schedule would arise in a version of the RANK model with a continuum of labor types, with wages
set in a decentralized way by each type of labor (or the union representing it) in a way consistent
with household utility maximization and given an isoelastic demand for labor. Throughout we
assume Wt ≥ (CtR )σ Ntϕ ≥ (CtK )σ Ntϕ for all t, so that all households will be willing to supply the
work hours demanded by firms.5

3.2 Firms

The supply side is standard, and consists of a continuum of identical monopolistically competitive
firms i ∈ [0, 1], each producing a differentiated product, using a technology

Yt (i) = At Nt (i)1−α

where Yt (i) denotes output, Nt (i) is labor input, and At is a technology parameter. The latter
evolves according to the process at = ρa log at−1 + εat , where at ≡ log At and εat ∼ N (0, σa2 ). The
demand for each good variety is isoelastic, and given by the function Yt (i) = (Pt (i)/Pt )− Yt , where
Yt denotes aggregate output.

As in Rotemberg (1983) we assume that each firm faces a quadratic price adjustment cost given
 2
by C (·) ≡ 2ξ Yt PPt−1
t (i)
(i)
− 1 where ξ > 0, and in terms of a CES goods bundle with elasticity of
substitution p . Since all firms are identical, profit maximization implies that all firms choose the
same price, i.e. Pt (i) = Pt for all i ∈ [0, 1], and hence produce the same output Yt with the same
Pt
labor input Nt . As a result, the (gross) price inflation rate Πt ≡ Pt−1
evolves according to
     
Yt+1  1 1
Πt (Πt − 1) = Et Λt,t+1 Πt+1 (Πt+1 − 1) + − (12)
Yt ξ Mpt Mp
5
We have analyzed a version of such a model for a TANK economy, where each union maximizes a weighted
average of the utilities of its Ricardian and Keynesian members, and derived a wage schedule slightly different
from (11), where the consumption of the two types enters separately, complicating the algebra without yielding any
significant additional insights on the issues of interest.

13
where Mpt ≡ (1 − α)At Nt−α /Wt is the average gross markup, and with the latter’s optimal value
p
in the absence of price adjustment costs given by Mp ≡ p −1
.

3.3 Fiscal and Monetary Policy

The cyclical behavior of taxes and transfers plays a crucial role in the determination of the effects of
aggregate shocks. To illustrate that role we consider an environment where transfers to Keynesian
households are given by
TtK = τ0 D + τd (Dt − D) (13)

where 0 ≤ τ0 ≤ 1, 0 ≤ τd ≤ 1, and where a balanced budget condition holds

λTtK + (1 − λ) TtR = 0. (14)

The previous specification nests as special cases laissez-faire (τ0 = τd = 0) and full steady
state redistribution (τ0 = 1 and τd = 0). In addition to the previous two cases, below we also
consider a " realistic" case with partial redistribution, both in steady state and period by period
(0 < τ0 < 1 and 0 < τd < 1). The case of full redistribution period by period (τ0 = 1 and τd = 1)
can be checked to be equivalent to the RANK model.

Regarding the conduct of monetary policy, we consider alternative scenarios. We start by

assuming that the central bank follows a simple interest rate rule, potentially including terms
related to income heterogeneity. Then, in Section 5, we derive the optimal monetary policy.

3.4 Equilibrium

Goods market clearing requires

Yt (i) = Ct (i) + Xt (i)

for all i ∈ [0, 1], where Xt (i) = (Pt (i)/Pt )−p (ξ/2)Yt (Πt − 1)2 captures the demand for good i to
meet price adjustment costs. Noting that in equilibrium all firms set the same prices, thus implying

14
Yt (i) = Yt and Ct (i) = Ct for all i ∈ [0, 1], we can write

Ct = Yt ∆p (Πt ) (15)

where ∆pt (Πt ) ≡ 1 − (ξ/2) (Πt − 1)2 and Ct ≡ (1 − λ)CtR + λCtK denotes aggregate consumption.
Furthermore, since all Ricardian households are identical, market clearing in the bonds and stock
markets implies that BtR = 0 and StR = 1/(1 − λ).

Aggregate employment is given by

1
Nt = (Yt /At ) 1−α (16)

and is assumed to be distributed uniformly among household types, so that NtR = NtK = Nt for
all t.

The Euler equation for Ricardian households can be rewritten in terms of aggregate consump-
tion as:
1 = β(1 + it )Et (Ct+1 /Ct )−σ (Ht+1 /Ht )−σ (Zt+1 /Zt )Π−1

t+1 (17)

where, as above, Ht ≡ CtR /Ct .

In equilibrium, and given the goods market clearing condition (15), we can write:

1 Dt TtR
 
1 Wt Nt
Ht = + +
∆p (Πt ) Yt 1 − λ Yt Y
   t 
1 1−α 1 − λτd Dt λ(τ0 − τd ) D
= + −
∆p (Πt ) Mpt 1−λ Yt 1 − λ Yt
     
1 λ(1 − τd ) p 1−α λ(τ0 − τd ) 1−α Y
= 1+ p ∆ (Πt ) − − 1− (18)
∆ (Πt ) 1−λ Mpt 1−λ Mp Yt

where the derivation of the second and third equalities makes use of the transfer and balanced
budget rules (13) and (14), the accounting identity Yt ∆p (Πt ) = Wt Nt + Dt , and the definition of
the average price parkup introduced earlier.

15
 Note that as long as τd < 1 (i.e. less than full redistribution) the heterogeneity index Ht is
increasing in the markup, given output, for a rise in the latter implies a redistribution of income
towards the unconstrained. The same is true for a rise in output, given the markup, as long as
τ0 > τd , since in that case a rise in output also implies a redistribution towards the unconstrained.

Equations (15) through (18), combined with the inflation equation (12) and a monetary policy
rule determining the interest rate it describe the equilibrium conditions of the model. In the next
sections we characterize the perfect foresight steady state with zero inflation and the log-linearized
equilibrium dynamics around that steady state.

3.4.1 Steady State

In a perfect foresight, zero inflation steady state for the above economy we have A = Z = 1,
Π = 1, ∆p (1) = 1 (implying C = Y ), and Mpt = Mp . Hence, evaluating (11) at that steady state
and combining it with the aggregate production function (16) yields the following expression for
steady state output:
 1−α
 σ(1−α)+ϕ+α
1−α
Y =
M
where M ≡ Mw Mp denotes the steady state "composite" markup. On the other hand, evaluating
(18) at the zero inflation steady state we obtain a closed-form expression for H:
 
λ(1 − τ0 ) 1−α
H =1+ 1− ≥ 1.
1−λ Mp

3.4.2 Equilibrium Dynamics around a Steady State

Note that in a neighborhood of the zero inflation steady state, and up to a first order approximation,
∆p,t ' 1 for all t. Accordingly, we henceforth approximate the goods market equilibrium condition
as
ybt = b
ct (19)

16
for all t, where ybt ≡ log(Yt /Y ) and b
ct ≡ log(Ct /C). Log-linearization of (12) around the zero
inflation steady state yields the inflation equation

µpt
πt = βEt {πt+1 } − δb (20)

bpt ≡ log(Mpt /Mp ) and δ ≡ p /(ξMp ). The deviations of the (log) price markup from
where µ
steady state can be written as:
   
1+ϕ α+ϕ
bpt
µ = at − σ + ybt . (21)
1−α 1−α

bpt = 0 all t we can solve for the natural (i.e. flexible price) level of output,
Note that by setting µ
ybtn (expressed in log deviations from steady state). In the illustrative model analyzed here, the
latter variable is as a function of technology only, given by

1+ϕ
ybtn = at
σ(1 − α) + α + ϕ
≡ ψ a at .

More generally, however, ybtn may also depend on other exogenous shocks that may shift the
markup-output schedule (21), including labor supply shocks and shocks to the desired markup,
among others. Independently of the number and nature of the shocks affecting ybtn , the following
relation will generally hold:  
α+ϕ
bpt
µ =− σ+ yet (22)
1−α
where yet ≡ yt − ytn is the output gap, i.e. the log deviation of output from its flexible price
counterpart. Substituting (22) into (20) we obtain a version of the New Keynesian Phillips curve
(NKPC, henceforth):
πt = βEt {πt+1 } + κe
yt (23)
α+ϕ


where κ ≡ δ σ + 1−α
. Thus, under the assumptions of our baseline TANK model, the supply
side is not affected by the presence of heterogeneity. This allows us to focus on its impact on

17
aggregate demand (which coincides with aggregate consumption in our simple model) in the spirit
of Werning (2015).

ht ≡ log(Ht /H):
Note that log-linearization of (18) yields, after letting b
   
λ (1 − α)(1 − τd ) p 1−α
ht =
b bt + 1 −
µ (τ0 − τd )b
yt
(1 − λ)H Mp Mp
= χy yet + χn ybtn (24)

where     
λ 1−α (1 − τd )(1 − α) α+ϕ
χy ≡ (τ0 − τd ) 1 − − σ+
(1 − λ)H Mp Mp 1−α

 
λ(τ0 − τd ) 1−α
χn ≡ 1− .
(1 − λ)H Mp
Log-linearizing the Euler equation for aggregate consumption (17), and imposing goods market
clearing we obtain:

1 b 1 n
yet = Et {e
yt+1 } − (it − Et {πt+1 }) + Et {∆b
ht+1 } + (1 − ρz )zt + Et {∆b
yt+1 }
σ σ

which can be combined with (24) to yield the Dynamic IS equation (DIS):

1
yet = Et {e
yt+1 } − (bit − Et {πt+1 } − rbtn ) (25)
σ(1 + χy )
where rbtn ≡ (1 − ρz )zt + σ (1 + χn ) Et {∆bn
yt+1 }.

Thus we see that the equilibrium conditions describing the non-policy block of the baseline
TANK model can be reduced to two equations –the NKPC and the DIS– in terms of the output
gap and inflation, that are isomorphic to those describing the equilibrium of the standard RANK
model. Under our assumptions the NKPC is identical to that in the RANK model. The DIS, on
the other hand, is affected by heterogeneity in two ways. First, the sensitivity of the output gap to
interest rate changes is now given by 1/[σ(1+χy )], which is smaller or larger than its counterpart in

18
the RANK model (1/σ) depending on the sign of χy , the elasticity of h with respect to the output
gap. Secondly, the impact of expected natural output growth on the natural rate of interest is
potentially amplified by the presence of the term χn , the elasticity of h with respect to natural
output.

Note that coefficients χy and χn are functions of the three "structural" parameters determining
heterogeneity in the model, namely, λ, τ0 , and τd . Trivially, χy = χn = 0 if λ = 0. Conditional on
λ > 0, two extreme cases are worth mentioning. Trivially, in the case of full redistribution (τ0 =
τd = 1) we also have χy = χn = 0, and, hence, equivalence to the RANK model, independently
of λ. At the other extreme, in the "laissez-faire" case (τ0 = τd = 0), we have χn = 0 and χy < 0,
i.e. the natural rate is not affected by the presence of heterogeneity, but output is more sensitive
to changes in interest rates, with that sensitivity increasing in the share of Keynesian households.
More generally, in an intermediate (and more realistic) case of less than full redistribution (0 <
τd ≤ τ0 < 1 ), the natural rate is more sentitive to changes in natural output growth (χn > 0),
while the impact of heterogeneity on the interest rate sensitivity of output is ambiguous (χy ≷ 0),
with χy > 0 if and only if

τ0 − τd σ(1 − α) + ϕ + α
> .
1 − τd Mp − 1 + α
Intuitively, when Keynesian agents are high enough constant transfers (i.e. τ0 is high), their
income is in not much affect by changes in interest rates. As a result, aggregate output would
be less sensitive to interest rates than in a representative agent economy (i.e. where all agents
are Ricardian). Viceversa, when constant transfers are low, income of Keynesian agents fluctuates
more, and thus the response of aggregate output is amplified. Thus, fiscal policy plays an important
role in determining the effects of interest rate changes. On the other hand, note that χy and χn
do not depend on the presence of nominal rigidities, or on the conduct of monetary policy.

The parameters χy and χn could alternatively be interpreted as reduced-form coefficients de-

19
scribing the cyclical behavior of the h-index, as implied by eq. (24). Under such interpretation, the
two parameters could be calibrated to match the observed cyclicality of heterogeneity between con-
strained and unconstrained households, without the need of explicitly modeling the actual sources
of consumption heterogeneity —e.g. labor income, employment status, wealth distribution, porto-
folio choices, etc. The resulting values could then be used to assess the effects of heterogeneity on
the sensitivity of the output gap to interest rate changes.

In addition to the two equations describing the non-policy block of the TANK economy, (23)
and (25), an additional equation describing how monetary policy is conducted is needed in order
to close the model. Following convention, in the analysis below we assume a Taylor-type rule of
the form:
bit = φπ πt + φy ybt + νt

where φπ ≥ 0, φy ≥ 0, and where νt is an exogenous monetary policy shock, following an AR(1)

process with autoregressive coefficient ρν ∈ [0, 1].

3.4.3 Equilibrium Dynamics: A Quantitative Illustration

To provide a quantitative illustration of the role of heterogeneity for the effects of monetary policy
shocks, we consider a calibration of the model where the share of Keynesian agents is set to 0.21,
in the middle of range of values typically used in the HANK literature.6 The remaining parameters
are set to standard values.7

Figure 1 shows that in a TANK model with no transfers (solid line with “o” markers) the re-
sponse of output to a monetary policy shock is amplified with respect to a RANK model, i.e. a
6
For instance, the fraction of constrained agents equals 13% in McKay et. al. (2016), while Kaplan et. al. (2014)
report that about one third of the U.S. households are hand-to-mouth, of which about one third is poor and two
thirds are wealthy.
7
Furthermore, we set the discount factor β = 0.9925, the coefficient of risk aversion σ = 1 (log-utility), the
(inverse) Frisch elasticity of substitution φ = 1, the labor income share 1 − α = 3/4, the elasticity of substitution
among good varieties  = 9 and the adjstument cost parameter ξ = 372.8 (implying the same slope of the Phillips
curve as a model with sticky prices a la Calvo with an average price duration of 1-year).

20
Figure 1: The Effects of Monetary Policy Shock in the TANK model under Alternative Transfer
Policies
Output Real Rate
0.6 0

0.4 -0.25

0.2 -0.5

0 -0.75
0 2 4 6 8 0 2 4 6 8
h-index Nominal Rate
0 0

-0.1 -0.2

-0.2 -0.4

-0.3 -0.6
0 2 4 6 8 0 2 4 6 8
Inflation Monetary Policy Shock
0.3 0

0.2
-0.125
0.1

0 -0.25
0 2 4 6 8 0 2 4 6 8

No Transfers Constant Transfers Full Redistribution (RANK)

case where there is full redistribution (dashed line). In the absence of transfers, firms’ profits con-
stitute the only source of earnings heterogeneity. Since profits move countercyclically in response
to demand shocks (a standard feature of New-Keynesian models), the income share of Ricardian
agents –the firms’ owners– is countercyclical, and thus the h−index is also countercyclical (i.e.
χy < 0). As a result, consumption of Keynesian agents responds more than consumption of Ri-
cardian agents, and the aggregate effects are thus amplified with respect to an economy where all
agents are Ricardian (as in RANK). A similar amplification effect also arises under in a case with
constant transfers set to eliminate steady state inequality (solid lines).

Figure 2 analyzes the response to a monetary policy shocks when varying the fraction Keynesian
agents λ from 0.1 to 0.3, while keeping the remaining parameters unchanged, and assuming a fiscal

21
Figure 2: The Effects of a Monetary Policy Shock in the TANK model under Alternative Size of
Keynesian Agents

Output Real Rate

0.6 0

0.4 −0.2
−0.4
0.2
−0.6
0
0 2 4 6 8 0 2 4 6 8
h−index Nominal Rate
0 0

−0.2
−0.2
−0.4

−0.4 −0.6
0 2 4 6 8 0 2 4 6 8
Inflation Monetary Policy Shock
0

0.2 −0.1
0.1
−0.2
0
0 2 4 6 8 0 2 4 6 8
λ = 0.1 λ = 0.2 λ = 0.3

policy with constant transfers. As expected, the figure shows that the larger is the fraction of
Keynesian agents, the larger is the response of output to a monetary policy shock.

4 HANK vs TANK: A Quantitative Comparison

The simple TANK model introduced in the previous section can at most account for the hetero-
geneity between constrained and unconstrained agents, but it ignores some key features of more
realistic HANK models, such as the presence of idiosyncratic risk and the associated dynamics of
the wealth distribution. Clearly, those features could potentially alter the aggregate implications

22
of monetary policy shocks, or of aggregate shocks in general. In this section, we thus compare the
predictions of our simple TANK model with those of a canonical HANK model, restricting our
attention to the responses of aggregate variables to aggregate shocks.

The HANK model considered is a standard Bewley-Aiyagari-Huggett economy, populated by

a continuum of ex-ante identical agents s ∈ [0, 1] with preferences given by eq. (6), and receiving
in each period a labor income exp(est )Wt Nt , where est denotes an idiosyncratic shock. Following
McKay et. al. (2016) and Auclert (2016), it is assumed that the idiosyncratic income shocks follows
an exogenous AR(1) process, with the persistence parameter ρ = 0.966 and standard deviation
σ = 0.017. Agents cannot perfectly insure against idiosyncratic risk, since the only asset available
to the agents is a riskless one-period nominal bond which is zero net supply, and each agent faces a
(exogenous) borrowing limit a = ψ Ȳ . In our baseline calibration we set ψ = 100%, which implies
that 21% of the agents are borrowing constrained in steady state. We also consider alternative
calibrations with tighter (ψ = 50%) and looser (ψ = 200%) borrowing limits, implying a fraction
of constrained agents of 36% and 11%, respectively.

The supply side is identical to the one described for the TANK model in Section 3. We instead
consider several alternative transfer policies. In our baseline calibration, as for the TANK model,
it is assumed that transfers to constrained agents are given by eq. (13).8 As an alternative, we
consider a more general transfer policy to take into account alternative degree of progressivity or
cyclicality of transfers. In particular, and following Auclert (2016), we consider a formulation for
transfers given by
Tts
 
D Dt D
= + − [γ + (1 − γ) exp(es )]
Yt Y Yt Y
and where the parameter γ measures the degree of progressivity of fiscal policy —i.e. a positive
8
The implicit assumption is that the government takes all profits and redistribute them across households
according to the specified transfer rule. Since the distribution of firms’ profits is fully determined by the transfer
policies, one could abstract from the households’ portfolio choice between debt and equity. An alternative (and
equally tractable) formulation is to assume that agents invest in a mutual funds choosing the asset allocation
between households’ credit and firms’ equity –see e.g. Gornemann et. al. (2016).

23
 Table 1: The Real Interest Rate Elasticity of Output (Baseline Transfer Policy)
Transfer Regime Borrowing Limit
τ0 , τd , ψ (1,0,1) (1,0.5,1) (1,1,1) (1,0,0.5) (1,0,2)
RANK 1.00 1.00 1.00 1.00 1.00
TANK 1.64 1.24 1.00 5.74 1.23
HANK 1.70 1.23 0.97 5.30 1.46
Implied λ 0.21 0.21 0.21 0.36 0.11

(negative) γ implies that the fiscal system is progressive (regressive). For instance, when γ = 1
all agents receive the same transfers in every period, regardless of the their labor income. Instead,
when γ = 0 the transfers are proportional to the idiosyncratic shock. Finally, when γ is negative
agents with high labor income receive a disproportionately larger share of profits. Clearly, this
type of fiscal policy cannot be mapped directly into a TANK model with no idiosyncratic shocks.
For this reason, and in order to compare the predictions with a TANK model, we first solve and
simulate the HANK model, and then use the resulting series to estimate the parameters τ0 and τd
in (13) for the TANK model.

As in previous studies in the HANK literature, we first calculate the real interest rate elasticity
of output ∂y/∂r on impact in response to a monetary policy shock.

Table I reports the results under the baseline transfer policy, under alternative values for the
parameters describing fiscal transfers (columns 1-3) and the borrowing limit (columns 4 and 5),
and where the response in the RANK model is normalized to one. For instance, in an economy
with constant transfers (τ0 = 1 and τd = 0, column 1) the response of output to a change in the
real interest rate in HANK is about 70% larger than in a RANK economy. In other words, in
this economy heteroegeneity has important implications for the dynamics of aggregate variables.
Notably, the differences between HANK and RANK are well captured by the simple TANK model,
where the output response is 64% larger than in RANK.9 Similarly, the TANK model provides a
9
This suggests that the large “indirect” (income) effects of monetary policy found by Kaplan et. al. (2016) in

24
 Table 2: The Real Interest Rate Elasticity of Output (Alternative Fiscal Policy)
Degree of Progressivity
γ = −1 γ = 0 γ = 1
RANK 1.00 1.00 1.00
TANK 1.44 1.17 1.00
HANK 1.50 1.18 0.97
τ0 0.43 0.72 1
τd 0.35 0.67 1
2
R 0.75 0.98 1

good approximation to the richer HANK model also under alternative parametrization for fiscal
transfers (columns 2 and 3), or under alternative values for the borrowing limit parameters (last
two columns). Finally, as shown in Table 2, the same results also hold under an alternative transfer
policy, for different value of tax progressivity (and cyclicality of income risk) as implied by the
parameter γ in the HANK model, and approximated by the resulting estimates of τ0 and τd in the
TANK model.

A more complete description of the model dynamics is provided in Figures 3-5 containing,
respectively, the impulse responses to monetary policy, preference and technology shocks, under
our baseline calibration. As it can be seen in Figure 3, the response of output to monetary policy
shocks is remarkably similar in TANK and HANK models, and different from the RANK model.
The main reason for the similarity between TANK and HANK can be understood looking at the
evolution of the h-index and v-index (second row). The TANK model approximates well the
heterogeneity between constrained and unconstrained agents (h-index). Instead, the heterogeneity
within unconstrained agents (v-index), which is constant by construction in a TANK model,
remains nearly constant in the HANK model. Similar considerations also holds in response to
preference shocks (Figure 4) and technology shocks (Figure 5), even though in the latter case the
an HANK model, are also captured by our simple TANK economy. Indeed the differences between TANK and
RANK are entirely due to the presence of hand-to-mouth agents, who are not affected by the “direct” intertemporal
substitution effect of monetary policy. See Bilbiie (2017) for a more thorough discussion on this issue.

25
differences between HANK and TANK are slightly larger.

In summary, our quantitative exercise shows that a simple TANK model may provide a good
approximation of a prototypical HANK model, even when the latter generates predictions that
are sizeably different from its RANK counterpart. Needless to say, TANK models might not
constitute a good approximation for richer HANK models, or more importantly the actual data.
This is likely to be the case for economies where aggregate shocks have large effects on consumption
heterogeneity, and in particular on the heterogeneity within unconstrained households. That could
be the case in economies with endogenous unemployment risk (e.g. Ravn and Sterk (2017)), or
where financial market participants have heterogeneous portfolios of assets (as in Kaplan et. al.
(2016), Bayer et. al. (2015) and Lütticke (2017)). Yet, assessing the quantitative significance of
fluctuations in consumption heterogeneity, both between and within different groups of agents,
remains a largely open question both from a theoretical and an empirical viewpoint.

5 Optimal Monetary Policy in a TANK model

Most of the existing literature on HANK models has focused on studying the aggregate effects of
monetary policy in the presence of households’ heterogeneity. However, and partly because of the
computational difficulties associated with HANK models, an open question is whether the presence
of households heterogeneity changes the prescriptions for the design of monetary policy resulting
from standard RANK models. For instance, it is subject to debate whether central banks should
be concerned or not with measures of heterogeneity. In what follows we analyze the implications
heterogeneity for the design of optimal monetary policy within a TANK model, which given the

26
 Figure 3: The Effects of a Monetary Policy Shock: HANK vs TANK

Output Real Rate

0.5 0

−0.25
0.25
−0.5

0 −0.75
0 2 4 6 8 0 2 4 6 8
h−index v−index
0.2 0.2

0 0

−0.2 −0.2
0 2 4 6 8 0 2 4 6 8
Inflation Monetary Policy Shock
0.25 0

0.125 −0.125

0 −0.25
0 2 4 6 8 0 2 4 6 8
HANK RANK TANK

27
 Figure 4: The Effects of a Preference Shock: HANK vs TANK

Output Real Rate

0.5 0.5

0.25 0.25

0 0
0 2 4 6 8 0 2 4 6 8
h−index v−index
0.2 0.2

0 0

−0.2 −0.2
0 2 4 6 8 0 2 4 6 8
Inflation Pref. Shock
0.25 0.5

0.125 0.25

0 0
0 2 4 6 8 0 2 4 6 8
HANK RANK TANK

28
 Figure 5: The Effects of a Technology Shock: HANK vs TANK

Output Real Rate

0
0.6

0.4 −0.5

0.2 −1
0 2 4 6 8 0 2 4 6 8
h−index v−index
0.5 0.5

0 0

−0.5 −0.5
0 2 4 6 8 0 2 4 6 8
Inflation Tech. Shock
−0.4
−0.6 1

−0.8
0.5
−1
−1.2 0
0 2 4 6 8 0 2 4 6 8
HANK RANK TANK

29
results of the previous section, could be viewed as a parsimonious representation of a richer HANK
model.

In particular, we consider a utilitarian central bank, namely one assigning equal weight to the
utility of all the agents. It can be shown that a second-order approximation to the central bank’s
welfare criterion around an efficient steady state (with no inequality) gives the central bank’s loss
function
∞
1 X tn 2 o
Lt ' E0 β πt + αy ỹt2 + αh ĥ2t + t.i.p. (26)
2 t=0
φ+α

1
and αh ≡ σξ 1−λ


where αy = σ + 1−α ξ λ
and t.i.p. indicates all the terms independent of policy.
The only difference with respect to the loss function of a standard RANK model is the presence
of the term αh ĥ2t . That term indicates that fluctuations in consumption heterogeneity generate
welfare losses for the central bank. Intuitively, a benevolent central bank would like to spread to
costs of fluctuations equally across all the agents, which implies no fluctuations in consumption
heterogeneity.

The optimal monetary policy is then the solution to the following problem
∞
1 X tn 2 2 2
o
min E0 β πt + αy ỹt + αh ĥt (27)
2 t=0
s.t. πt = βEt {πt+1 } + κe
yt (28)

ĥt = χy yet + χn ybtn . (29)

Notably, eq. (29) signals the presence of a policy trade-off related to heterogeneity. In the
absence of heterogeneity, the optimal policy with a representative agent would be to stabilize both
inflation and the output-gap (i.e. πt = ỹt = 0). In our TANK model, the central bank would
also like to stabilize heterogeneity. However, this might not be possible. This is because as long
as χn 6= 0, fluctuations in the natural level of output leads to fluctuations in the heterogeneity
index ĥt . As a result, a central bank that wishes to stabilize output at its natural level would have

30
to tolerate some fluctuations in heterogeneity. Similarly, a central bank that wishes to stabilize
heterogeneity would have to tolerate fluctuations in inflation and the output-gap.

Using the optimality conditions to the above problem, the optimal monetary policy can be
charactized by the following price level targeting rule

1 
p̂t ≡ pt − p−1 =− αy ỹt + χy αh ĥt .
κ

The latter equation, differently from a RANK model, indicates that the price level should respond
to changes in heterogeneity. For example, in a situation where χy < 0 (as under the fiscal policies
considered above), the price level should increase (decrease) whenever there is an increase (de-
crease) in the h-index. In other words, the central bank should tolerate some inflation whenever
consumption of Ricardian agents increases more than average consumption.

We finally examine the magnitude of the effects of hetereogeneity on the optimal policy. To
that end, Figure 6 illustrates the impulse response to a 1% technology shock under the optimal
monetary policy in a TANK model (solid lines) in comparison to a RANK model (dashed line),
under the baseline calibration of Section 3. Clearly, in a RANK model inflation and the output
gap remain constant in every period. Instead TANK model, we do observe an increase in both
inflation and output gap, but of a neglible magnitude (less than 0.05%), as well as an increase in
the h-index. Also, the response of the nominal rate is almost identical in a RANK and in a TANK
model. Overall, these results suggest that from a quantitative viewpoint, heterogeneity seems to
alter only minimally the monetary policy prescriptions obtained in a standard RANK model.

The main reason for that result is that, under our baseline calibration, the optimal weight
on heterogeneity ah in eq. (26) is very small and is less than 1% of the weight that should be
assigned to inflation stabilization. Clearly, if central banks attached a higher weight to measures of
heterogeneity, heterogeneity would have a larger quantitative effects on interest rates and inflation,
and other variables.

31
Figure 6: Optimal Monetary Policy in TANK - The Effects of Technology Shocks

Output−Gap Real Rate

0.04 0

0.02 −0.2

0 −0.4

0 2 4 6 8 0 2 4 6 8
h−index Nominal Rate
0.15 0

−0.2
0.075

−0.4
0
0 2 4 6 8 0 2 4 6 8
−3
x 10 Inflation Technology Shock
20 1

10
0.5

0
0 2 4 6 8 0 2 4 6 8
RANK (λ = 0) TANK: λ = 0.2 TANK: λ = 0.3

32
6 Conclusions

We identify two key dimensions of heterogeneity that explain the differential behavior of a Hetero-
geneous agents New-Keynesian economy (HANK) relative to its Representative Agent (RANK)
counterpart: (i) the differences in average consumption between constrained and unconstrained
households and (ii) consumption heterogeneity within the subset of unconstrained households. A
tractable two-agent model (TANK), which only captures the first dimension, approximates reason-
ably well the implications of a baseline HANK model regarding the effects of aggregate shocks on
aggregate variables. We then use the TANK model as a laboratory to study the design of optimal
monetary policy, and show that heterogeneity introduces a non-trivial policy trade-off, but that
under standard calibrations price stability is nearly optimal, as in standard RANK models.

Our results should not be interpreted as implying that existing HANK models are not useful,
nor that households’ heterogeneity is not important for the conduct of monetetary policy. On the
contrary, HANK models seem essential to describe economies with large fluctuations in consump-
tion heterogeneity within financial market participants. Also, our analysis highlights some aspects
of heterogeneity that are relevant for monetary policy, but that are not captured by simple models,
nor are well understood from an empirical viewpoint. For this reason, we hope that our results
will constitute a useful input for future studies on monetary policy with heterogeneous agents.

The current TANK model can be extended along several dimensions. First, to introduce some
form of idiosyncratic risk between Ricardian and Keynesian agents. One possibility is to consider
that in each period, a fraction of Ricardian agent might become Keynesian, and viceversa, as
e.g. in Nisticó (2016) and Bilbiie (2017). That extension of the model could be useful to address
the so-called “forward-guidance” puzzle inherent in representative agent models, and that are also
present in the current version of our TANK model. Second, it sthe comparison between TANK
and HANK models, could be extended to alternative frameworks –e.g. with capital, government
debt and other assets (liquid and illiquid)– to understand to what extent TANK models might

33
be able to capture some of the defining features of the richer HANK models (e.g. Kaplan et. al.
(2016)).

APPENDIX

A1. Some Approximations

1
R
Using the definition CtR ≡ 1−λt s∈Θt
Cts ds, a second-order approximation to the left hand side
of (2) is given by
Z Z  s −σ
Zt s −σ Zt R −σ

Ct
(Ct ) ds = Ct ds
1 − λt s∈Θt 1 − λt s∈Θt CtR
(Z "  s   s 2 # )
Zt −σ C t σ (1 + σ) C t
CR


' 1−σ −1 + −1 ds
1 − λt t s∈Θt C R
t 2 C R
t
" Z  s 2 #
−σ σ (1 + σ) 1 C t
= Zt CtR


1+ − 1 ds
2 1 − λt s∈Θt CtR
 
R −σ

σ (1 + σ) s
= Z t Ct 1+ vars|t {ct } .
2
1
R
2
where vars|t {cst } ≡ 1−λ t s∈Θt
cst − cR
t ds and where in the last step we have use the fact that up
 s 2  2
Ct s R 2

Cts
to a second order approximation C R − 1 ' ct − ct ≡ log C R .
t t
1
R
R s
Similarly, using the definition Ct+1|t ≡ 1−λt s∈Θt Ct+1 ds we can write the term inside the
expectations on the right hand size of (2) as
!−σ
Z Z s
Ct+1
Zt+1 Zt+1 “ R ”−σ
s
(Ct+1 )−σ ds ' Ct+1|t R
ds
1 − λt s∈Θt 1 − λt s∈Θt Ct+1|t
R !−σ 8Z 2 ! !2 3 9
Ct+1|t s
Ct+1 s
Ct+1
Zt+1 “ R ”−σ <
41 − σ σ (1 + σ) =
= Ct+1 R R
−1 + R
− 1 5 ds
1 − λt Ct+1 : s∈Θt Ct+1|t 2 Ct+1|t ;
”−σ » –
“
R σ(1 + σ)
= Zt+1 Ct+1 Φmean
t 1+ vars|t {cst+1 }
2

2  R
−σ
Ct+1|t

1
R
where vars|t {cst+1 } ≡ 1−λt s∈Θt
cst+1 − cR
t+1|t ds and Φmean
t+1 ≡ R
Ct+1
.

Combining these results, we get the Euler equation

Zt (CtR )−σ = βRt Et Zt+1 (Ct+1

R −σ

) Vt+1

34
where
Vt+1 ≡ Φmean var
t+1 Φt+1

2+σ(1+σ)vars|t {cst+1 }
where Φvar
t+1 ≡ 2+σ(1+σ)vars|t {cst }
, which corresponds to eq. (3) in the main text.

35
 REFERENCES

Auclert, Adrien (2016): " Monetary Policy and the Redistribution Channel," mimeo.

Bayer, Christian, Ralph Lütticke, Lien Pham-Do, Volker Tjaden (2015): “Precautionary Sav-
ings, Illiquid Assets, and the Aggregate Consequences of Shocks to Household Income Risk, CEPR
Discussion Paper No. DP10849.

Bilbiie, Florin (2008): “Limited Asset Market Participation, Monetary Policy and (Inverted)
Aggregate Demand Logic,” Journal of Economic Theory, 140(1), 162-196.

Bilbiie, Florin and Ragot, Xavier (2017): “Optimal Monetary Policy and Liquidity with Het-
erogeneous Households,” CEPR Discussion Paper No. DP11814.

Bilbiie, Florin (2017): "The New Keynesian Cross: Understanding Monetary Policy with Two
Agents," mimeo.

Broer, Tobias, Per Krusell, Niels-Jakob Hansen, and Erik Oberg (2016): “The New Keynesian
Transmission Channel: A Heterogeneous-Agent Perspective,” NBER working paper no. 22418.

Campbell, John Y. and N. Gregory Mankiw (1989), “Consumption, Income, and Interest Rates:
Reinterpreting the Time Series Evidence,” in O.J. Blanchard and S. Fischer (eds.), NBER Macroe-
conomics Annual 1989, 185-216, MIT Press.

Galí, Jordi, David López-Salido, and Javier Vallés (2007): "Understanding the Effects of Gov-
ernment Spending on Consumption,” Journal of the European Economic Association, 5(1), 227-270.

Gornemann, Nils, Keith Kuester and Makoto Nakajima, (2016): “Doves for the Rich, Hawks
for the Poor? Distributional Consequences of Monetary Policy”, mimeo.

Kaplan, Greg, Giovanni L. Violante, and Justin Weidner (2014): " The Wealthy Hand-to-
Mouth," Brookings Papers on Economic Activity 2014 (Spring), 77–138.

Kaplan, Greg, Benjamin Moll and Giovanni L. Violante (2016), “Monetary Policy According
to HANK”, NBER Working paper No. 21897.

36
 Lütticke, Ralph (2017): “Transmission of Monetary Policy with Heterogeneity in Household
Portfolios”, mimeo.

McKay, Alisdair and Ricardo Reis (2016) “The Role of Automatic Stabilizers in the U.S. Busi-
ness Cycle, Econometrica 84(1) 141–94.

McKay, Alisdair, Emi Nakamura and Jon Steinsson (2016): " The Power of Forward Guidance
Revisited," American Economic Review, 106(10), 3133-3158.

Nisticò, Salvatore (2016): "Optimal Monetary Policy and Financial Stability in a non-Ricardian
Economy,” Journal of the European Economic Association, 14(5), 1225-1252.

Ravn, Morten and Vincent Sterk (2014): "Job Uncertainty and Deep Recessions," Discussion
Papers 1501, Centre for Macroeconomics (CFM).

Ravn, Morten and Vincent Sterk (2017): “Macroeconomic Fluctuations with HANK & SAM:
an Analytical Approach,” mimeo.

Werning, Iván (2015): "Incomplete Markets and Aggregate Demand,"mimeo.

37