8085

2020
February 2020

Shocks, Frictions, and Ine-

quality in US Business Cycles
Christian Bayer, Benjamin Born, Ralph Luetticke
Impressum:

CESifo Working Papers

ISSN 2364-1428 (electronic version)
Publisher and distributor: Munich Society for the Promotion of Economic Research - CESifo
GmbH
The international platform of Ludwigs-Maximilians University’s Center for Economic Studies
and the ifo Institute
Poschingerstr. 5, 81679 Munich, Germany
Telephone +49 (0)89 2180-2740, Telefax +49 (0)89 2180-17845, email office@cesifo.de
Editor: Clemens Fuest
www.cesifo-group.org/wp

An electronic version of the paper may be downloaded

· from the SSRN website: www.SSRN.com
· from the RePEc website: www.RePEc.org
· from the CESifo website: www.CESifo-group.org/wp
 CESifo Working Paper No. 8085

Shocks, Frictions, and Inequality in

US Business Cycles

Abstract

How much does inequality matter for the business cycle and vice versa? Using a Bayesian
likelihood approach, we estimate a heterogeneous-agent New-Keynesian (HANK) model with
incomplete markets and portfolio choice between liquid and illiquid assets. The model enlarges
the set of shocks and frictions in Smets and Wouters (2007) by allowing for shocks to income
risk and taxes. We find that adding data on inequality does not materially change the estimated
shocks and frictions driving the US business cycle. The estimated shocks, however, have
significantly contributed to the evolution of US wealth and income inequality. The systematic
components of monetary and fiscal policy are important for inequality as well.
JEL-Codes: C110, D310, E320, E630.
Keywords: Bayesian estimation, business cycles, income inequality, incomplete markets,
monetary and fiscal policy, wealth inequality.

Christian Bayer
University of Bonn / Germany
christian.bayer@uni-bonn.de

Benjamin Born Ralph Luetticke

Frankfurt School of Finance & Management University College London
Frankfurt / Germany London / United Kingdom
b.born@fs.de r.luetticke@ucl.ac.uk

January 28, 2020

We would like to thank Gregor Boehl, Edouard Challe, Miguel Leon-Ledesma, Johannes
Pfeifer, Mu-Chun Wang, and participants at various conferences and seminars for helpful
comments and suggestions. Christian Bayer gratefully acknowledges support through the ERC-
CoG project Liquid-House-Cycle funded by the European Union’s Horizon 2020 Program under
grant agreement No. 724204. We would like to thank Sharanya Pillai for excellent research
assistance.
Codes are available at https://github.com/BenjaminBorn/HANK_BusinessCycleAndInequality.
1 Introduction
A new generation of monetary business cycle models has become popular featuring het-
erogeneous agents and incomplete markets (known as HANK models). This new class of
models implies new transmission channels of monetary1 and fiscal2 policy, as well as new
sources of business cycle fluctuations working through household portfolio decisions.3 Much
of this literature so far has focused on specific channels of transmission, shocks, or puzzles.
In contrast, the present paper takes a more encompassing approach and asks how our view
of the business cycle and of inequality dynamics changes when we bring this model to the
data. In particular, we aim to answer two questions: First, do data on inequality change the
estimated shocks and frictions driving the US business cycle? Second, how important are
business cycle shocks for the evolution of US inequality?
For this purpose, we study the business cycle using a technique that has become standard
at least since Smets and Wouters’ (2007) seminal paper, extending this technique to the
analysis of HANK models: We estimate an incomplete markets model by a full information
Bayesian likelihood approach using the state-space representation of the model. Specifically,
we estimate an extension of the New-Keynesian incomplete markets model of Bayer et al.
(2019). We add features such as capacity utilization, a frictional labor market with sticky
wages, and progressive taxation, as well as the usual plethora of shocks that drive business
cycle fluctuations in estimated New-Keynesian models: aggregate and investment-specific
productivity shocks, wage- and price-markup shocks, monetary- and fiscal-policy shocks,
risk premium shocks, and, as two additional incomplete-market-specific ones, shocks to the
progressivity of taxes and shocks to idiosyncratic productivity risk.
In this model, precautionary motives play an important role for consumption-savings
decisions. Since individual income is subject to idiosyncratic risk that cannot be directly
insured and borrowing is constrained, households structure their savings decisions and port-
folio allocations to optimally self-insure and achieve consumption smoothing. In particular,
we assume that households can either hold liquid nominal bonds or invest in illiquid physical
capital. Capital is illiquid because its market is segmented and households participate only
from time to time. This portfolio-choice component, which gives rise to an endogenous liq-
uidity premium, and the presence of occasional hand-to-mouth consumers leads the HANK
model to have rich distributional dynamics in response to aggregate shocks.
1
Auclert (2019) analyzes the redistributive effects of monetary policy, Kaplan et al. (2018) show the
importance of indirect income effects, and Luetticke (2018) analyzes the portfolio rebalancing channel of
monetary policy. McKay et al. (2016) study the effectiveness of forward guidance.
2
Auclert et al. (2018) and Hagedorn et al. (2019) discuss the fiscal multiplier, McKay and Reis (2016)
discuss the role of automatic stabilizers.
3
Bayer et al. (2019) quantify the importance of shocks to idiosyncratic income risk, and Guerrieri and
Lorenzoni (2017) look at the effects of shocks to the borrowing limit.

1
 To infer the importance of inequality for the business cycle, we estimate the HANK model
with and without data on inequality. We first estimate the model on the same observables
as in Smets and Wouters (2007) (plus proxies for income risk and taxes) covering the time
period of 1954 to 2015.4 We then re-estimate the model with two additional observables for
the shares of wealth and income held by the top 10% of households in each dimension, which
are taken from the World Inequality Database. We focus on the top 10% shares because this
measure is most consistent across alternative data sources such as the Survey of Consumer
Finances, where available.5 With respect to the first question, we find that the addition of
distributional data does not change what we infer about the aggregate shocks and frictions
driving the US business cycle. The answer to the second question rationalizes this result.
We find that business cycle shocks generate very persistent movements in wealth and income
inequality that are consistent with the U-shaped evolution of US inequality over 1954-2015.
In the HANK model, even transitory shocks have very persistent effects on inequality,
because wealth is a slowly moving variable that accumulates past shocks and thus business
cycle shocks persistently redistribute across households with different portfolios. To our
knowledge, this paper is the first to quantify the distributional consequences of all standard
business cycle shocks and estimate their importance in explaining US inequality. The impulse
response functions show that wealth inequality responds in particular to technology, fiscal,
and markup shocks. Wage and price markups directly affect the distribution of income
shares, while technology and fiscal shocks affect the return spread between illiquid capital
and liquid bonds. For income inequality, income risk shocks are important as well. The
drivers of consumption inequality are a mixture of the drivers of current income and wealth.
The historical decomposition of US inequality reveals that changing markups and tech-
nology are the main contributors to the rise of wealth and income inequality from the 1980s
to today. Yet, fiscal policies also play their role both by providing liquid assets for self-
insurance through government deficits and by changing the incentives to self-insure through
progressive taxation. Both move liquidity premia and thus affect the savings incentives of
the rich and the poor differentially. Quantitatively, we find deficits to be less important
than changes in progressive taxation. For consumption and income inequality, fluctuations
in income risk play a significant role and this role goes beyond increasing the dispersion of
income once the higher risk is realized. Wealth poor, and thus badly insured, households
react to an increase in uncertainty by cutting consumption particularly strongly, while for
well-insured households, which are already consumption rich, behavior changes little. Con-
sequently, these shocks account for 20% of the cyclical variations in consumption inequality.
They also account for 20% of aggregate consumption fluctuations in US recessions.
4
We use the estimates of income risk for the US provided by Bayer et al. (2019) and top marginal tax
rates as a proxy for progressivity.
5
See Kopczuk (2015) and Bricker et al. (2016) for detailed comparisons of all available data sources.
2
 Given the estimated shocks, we assess the importance of policy rules in shaping inequality
over the business cycle. We find that policy rules are more important in shaping inequality
than policy mistakes. Broadly speaking, output stabilization is key to reducing fluctuations
in inequality. A more hawkish monetary policy, i.e., a stronger reaction to inflation, would
have increased inequality in the 1970s and today. Both periods – through the lens of our
model – are characterized by high markups such that hawkish policy leads to output losses
and increases inequality. Countercyclical fiscal policy affects inequality not only by stabilizing
output, but also through its effect on returns. Larger and more persistent deficits after the
Great Recession would have depressed the liquidity premium and reduced wealth inequality.
To our knowledge, our paper is the first to provide an encompassing estimation of shocks
and frictions using a HANK model with portfolio choice. Most of the literature on monetary
heterogeneous-agent models has used a calibration approach.6 Auclert et al. (2020) and
Hagedorn et al. (2018) go beyond calibration but use one-asset HANK models. The latter
provide parameter estimates based on impulse-response function matching, while the former
estimate the model using the MA-∞ representation in the sequence space. Using a state-
space approach is key for us, as we need to deal with mixed-frequency data.
Our findings provide new insights into the literature on the drivers of inequality that
focuses on long-run trends such as the rising skill premium or changes in taxes.7 Kaymak
and Poschke (2016) and Hubmer et al. (2019), which are most closely related to our approach,
use quantitative models to study permanent changes in the US tax and transfer system and
the distribution of income. They find that these changes can explain a significant part of the
recent increase in wealth inequality. We complement their findings by showing that business
cycles have very persistent effects on inequality and can account for 50% of the rise in US
wealth inequality from 1980 to 2015. In our estimation, tax progressivity is of secondary
importance for wealth inequality relative to changes in markups and technology. In terms
of methods, these papers solve for steady-state transitions of calibrated models, while we
estimate our model on US macro and micro time series data.
In the sense that it estimates a state-space model of both distributional (cross-sectional)
data and aggregates is also the paper by Chang et al. (2018) is also related. They find that,
in an SVAR sense, shocks to the cross-sectional distribution of income have only a mild
impact on aggregate time series. Our finding of structural estimates being relatively robust
to the inclusion or exclusion of cross-sectional information resembles their results.8
6
See, for example, Auclert et al. (2018); Ahn et al. (2018); Bayer et al. (2019); Broer et al. (2019); Challe
and Ragot (2015); Den Haan et al. (2017); Gornemann et al. (2012); Guerrieri and Lorenzoni (2017); McKay
et al. (2016); McKay and Reis (2016); Ravn and Sterk (2017); Sterk and Tenreyro (2018); Wong (2019).
7
There is a growing literature on inequality dynamics. On the theory side, see, e.g., Gabaix et al. (2016).
On the empirical side, see, e.g., Heathcote et al. (2010), Piketty and Saez (2003) or Saez and Zucman (2016).
8
Our approach is different and simpler than the method suggested by Liu and Plagborg-Møller (2019),

3
 Focusing on the methodological contribution, Auclert et al. (2019) provide a fast estima-
tion method for heterogeneous-agent models that requires a sequence space representation
of the model and thus does not allow us to deal with missing or mixed frequency data as we
need to do here, when combining cross-sectional and aggregate data. Since this is the setup
we are facing, we build on the solution method of Reiter (2009) using the dimensionality
reduction approach of Bayer and Luetticke (2018) to make this feasible for estimation. We
further exploit the fact that only a small fraction of the Jacobian of the non-linear difference
equation that represents the model needs to be re-calculated during the estimation.
The remainder of this paper is organized as follows: Section 2 describes our model econ-
omy, its sources of fluctuations, and its frictions. Section 3 provides details on the numerical
solution method and estimation technique. Section 4 presents the parameters that we cali-
brate to match steady-state targets and prior distributions for the remaining parameters that
we estimate. It also gives an overview over the data we employ in our estimation. Section
5 discusses the estimated shocks and frictions driving the US business cycle. Section 6 does
so for US inequality. Section 7 concludes. An Appendix follows.

2 Model
We model an economy composed of a firm sector, a household sector, and a government
sector. The firm sector comprises (a) perfectly competitive intermediate goods producers
who rent out labor services and capital; (b) final goods producers that face monopolistic
competition, producing differentiated final goods out of homogeneous intermediate inputs;
(c) producers of capital goods that turn consumption goods into capital subject to adjustment
costs; (d) labor packers that produce labor services combining differentiated labor from (e)
unions that differentiate raw labor rented out from households. Price setting for the final
goods as well as wage setting by unions is subject to a pricing friction à la Calvo (1983).
Households earn income from supplying (raw) labor and capital and from owning the firm
sector, absorbing all its rents that stem from the market power of unions and final goods
producers, and decreasing returns to scale in capital goods production.
The government sector runs both a fiscal authority and a monetary authority. The fiscal
authority levies taxes on labor income and distributed profits, issues government bonds, and
adjusts expenditures to stabilize debt in the long run and aggregate demand in the short
run. The monetary authority sets the nominal interest rate on government bonds according
to a Taylor rule.
which includes full cross-sectional information in the estimation of a heterogeneous-agent DSGE model. We,
in contrast, only use the model to fit certain generalized cross-sectional moments.

4
2.1 Households
The household sector is subdivided into two types of agents: workers and entrepreneurs. The
transition between both types is stochastic. Both rent out physical capital, but only workers
supply labor. The efficiency of a worker’s labor evolves randomly exposing worker-households
to labor-income risk. Entrepreneurs do not work, but earn all pure rents in our economy
except for the rents of unions which are equally distributed across workers. All households
self-insure against the income risks they face by saving in a liquid nominal asset (bonds) and
a less liquid asset (capital). Trading illiquid assets is subject to random participation in the
capital market.
To be specific, there is a continuum of ex-ante identical households of measure one,
indexed by i. Households are infinitely lived, have time-separable preferences with time-
discount factor β, and derive felicity from consumption cit and leisure. They obtain income
from supplying labor, nit , from renting out capital, kit , and from earning interest on bonds,
bit , and potentially from profits or union transfers. Households pay taxes on labor and profit
income.

2.1.1 Productivity, labor supply and labor income

A household’s gross labor income wt nit hit is composed of the aggregate wage rate on raw
labor, wt , the household’s hours worked, nit , and its idiosyncratic labor productivity, hit .
We assume that productivity evolves according to a log-AR(1) process with time-varying
volatility and a fixed probability of transition between the worker and the entrepreneur
state:
  
h


 exp ρ h log h̃it−1 + it with probability 1 − ζ if hit−1 6= 0,

h̃it = 1 with probability ι if hit−1 = 0, (1)



0 else,

with individual productivity hit = R h̃h̃it di such that h̃it is scaled by its cross-sectional average,
it
R
h̃it di, to make sure that average worker productivity is constant. The shocks hit to produc-
tivity are normally distributed with time-varying variance that follows a log-AR(1) process
with endogenous feedback to aggregate effective hours Nt+1 (hats denote log-deviations from
the steady state):

2
σh,t = σ̄h2 exp sbt , (2)
bt+1 + σ ,
sbt+1 = ρs sbt + ΣY N (3)
t

5
i.e., at time t households observe a change in the variance of shocks that drive the next
period’s productivity. With probability ζ households become entrepreneurs (h = 0). With
probability ι an entrepreneur returns to the labor force with median productivity. An en-
trepreneur obtains a fixed share of the pure rents (aside from union rents), ΠFt , in the
economy (from monopolistic competition in the goods sector and the creation of capital).
We assume that the claim to the pure rent cannot be traded as an asset. Union rents, ΠUt are
distributed lump-sum across workers, leading to labor-income compression. For tractability,
we assume union profits to be taxed at the average income tax rate of the economy.
This modeling strategy serves two purposes. First and foremost, it generally solves
the problem of the allocation of pure rents without distorting factor returns and without
introducing another tradable asset.9 Second, we use the entrepreneur state in particular – a
transitory state in which incomes are very high – to match the income and wealth distribution
following the idea by Castaneda et al. (1998). The entrepreneur state does not change the
asset returns or investment opportunities available to households.
With respect to leisure and consumption, households have Greenwood et al. (1988) (GHH)
preferences and maximize the discounted sum of felicity:10
∞
X
E0 max β t u [cit − G(hit , nit )] . (4)
{cit ,nit }
t=0

The maximization is subject to the budget constraints described further below. The felic-
ity function u exhibits a constant relative risk aversion (CRRA) with risk aversion parameter
ξ > 0,
1
u(xit ) = x1−ξ ,
1 − ξ it
where xit = cit − G(hit , nit ) is household i’s composite demand for goods consumption cit
and leisure and G measures the disutility from work. Goods consumption bundles varieties
9
There are basically three possibilities for dealing with the pure rents. One attributes them to capital
and labor, but this affects their factor prices; or one introduces a third asset that pays out rents as dividends
and is priced competitively; one distributes the rents in the economy to an exogenously determined group of
households. The latter has the advantage that factor supply decisions remain the same as in any standard
New-Keynesian framework and still avoids the numerical complexity of dealing with three assets.
10
The assumption of GHH preferences is mainly motivated by the fact that many estimated DSGE models
of business cycles find small aggregate wealth effects in the labor supply; see, e.g., Born and Pfeifer (2014).
It also simplifies the numerical analysis somewhat. Unfortunately, it is not feasible to estimate the flexible
form of preference of Jaimovich and Rebelo (2009), which also encompasses King et al. (1988) preferences.
This would require solving the stationary equilibrium in every likelihood evaluation, which is substantially
more time consuming than solving for the dynamics around this equilibrium.

6
j of differentiated goods according to a Dixit-Stiglitz aggregator:
Z ηt −1
 η η−1
t
t
ηt
cit = cijt dj .

Each of these differentiated goods is offered at price pjt , so that for the aggregate price level,
R 1−ηt
1−η 1
Pt = pjt dj t , the demand for each of the varieties is given by

 −ηt
pjt
cijt = cit .
Pt

Assuming a (progressive) income-tax schedule (which we borrow from Benabou, 2002;

Heathcote et al., 2017), a household’s net labor income, yit , is given by

P
yit = (1 − τtL )(wt hit nit )1−τt , (5)

where wt is the aggregate wage rate and τtL and τtP determine the level and the progressivity
of the tax code. Given net labor income, the first-order condition for labor supply is

∂G(hit , nit ) P −τ P yit

= (1 − τtP )(1 − τtL )(wt hit )1−τt nit t = (1 − τtP ) . (6)
∂nit nit

Assuming that G has a constant elasticity w.r.t. n, ∂G(h it ,nit )

∂nit
= (1 + γ) G(hnitit,nit ) with γ > 0,
we can simplify the expression for the composite consumption good xit making use of this
first-order condition (6) and substitute G(hit , nit ) out of the individual planning problem:

1 − τtP
xit = cit − G(hit , nit ) = cit − yit . (7)
1+γ

When the Frisch elasticity of labor supply is constant and the tax schedule has the form
(5), the disutility of labor is always a fraction of labor income, constant across households.
Therefore, in both the budget constraint of the household and its felicity function only
after-tax income enters and neither hours worked nor productivity appears separately.
What remains to be determined is individual and aggregate effective labor supply. With-
P n1+γ
out further loss of generality, we assume G(hit , nit ) = hit1−τ̄ 1+γ
it
, where τ̄ P is the stationary
equilibrium level of progressivity of the tax code. This functional form simplifies the house-
hold problem in the stationary equilibrium as hit drops out from the first-order condition and
all households supply the same number of hours nit = N (wt ). Total effective labor input,
R R
nit hit di, is hence also equal to N (wt ) because we normalized hit di = 1.
Importantly, this means that we can read off average productivity risk directly from the

7
estimated income risk series of Bayer et al. (2019). Without scaling the labor disutility
by productivity we would need to translate productivity risk to income risk through the
endogenous hour response. When tax progressivity does not coincide with its stationary
equilibrium value, individual hours worked differ across agents and are given by

τ̄ P −τ P 1−τ P
 1 P γ+τ Pt γ+τtP
nit = (1 − τtP )(1 − τtL ) γ+τt hit t wt t ,

(8)

such that aggregate effective hours are given by

Z t 1−τ P Z γ+τ̄ P
 P
 1
L γ+τtP γ+τtP γ+τtP
Nt = nit hit = (1 − τt )(1 − τt ) wt hit . (9)
| {z }
:=Ht

Here Ht measures how the tax progressivity influences the (hours-weighted) average labor
productivity. Scaling of the disutility of labor by hit1−τ̄ is thus a normalization of Ht to one
in the stationary equilibrium. It implies that, despite the progressive income tax, changes
in the distribution of productivity h have no first-order effect on effective labor supply and
thus, as in Bayer et al. (2019), shocks to income risk do not directly move effective labor.
Household after-tax labor income, plugging in the optimal supply of hours, is then:

1−τtP 1+γ γ+τ̄ P

1+γ (1−τtP ) (1−τtP )
P γ+τtP γ+τtP
τtL )(wt hit nit )1−τt τtL ) γ+τt τtP ) γ+τt
P P
zit = (1 − = (1 − (1 − wt hit . (10)

2.1.2 Consumption, savings, and portfolio choice

Given this labor income, households optimize intertemporally subject to their budget con-
straint:

R(bit ,Rtb ,At )

cit + bit+1 + qt kit+1 = bit πt
+ (qt + rt )kit + zit + Ihit 6=0 ΠUt + Ihit =0 ΠFt ,
kit+1 ≥ 0, bit+1 ≥ B ,

where ΠUt is union profits, ΠFt is firm profits (both net of taxes), bit is real bond holdings, kit is
the amount of illiquid assets, qt is the price of these assets, rt is their dividend, πt = PtP−P t−1
t−1

is realized inflation, and R is the nominal interest rate on bonds, which depends on the
portfolio position of the household and the central bank’s interest rate Rtb , which is set one
period before. All households that do not participate in the capital market (kit+1 = kit ) still
obtain dividends and can adjust their bond holdings. Depreciated capital has to be replaced
for maintenance, such that the dividend, rt , is the net return on capital. Holdings of bonds

8
have to be above an exogenous debt limit B, and holdings of capital have to be non-negative.
1−τ P
Substituting the expression cit = xit + 1+γt yit for consumption, we obtain the budget
constraint for the composite leisure-consumption good:

R(bit ,Rtb ,At ) τtP +γ

xit + bit+1 + qt kit+1 = bit πt
+ (qt + rt )kit + 1+γ it
z + Ihit 6=0 ΠUt + Ihit =0 ΠFt , (11)
kit+1 ≥ 0, bit+1 ≥ B.

Households make their savings choices and their portfolio choice between liquid bonds
and illiquid capital in light of a capital market friction that renders capital illiquid because
participation in the capital market is random and i.i.d. in the sense that only a fraction, λ,
of households is selected to be able to adjust their capital holdings in a given period.
What is more, we assume that there is a wasted intermediation cost that drives a wedge
between the government bond yield Rtb an the interest paid by/to households Rt . This wedge,
At , is given by a time-varying term plus a constant, R, when households resort to unsecured
borrowing. This means, we specify:

R b A if bit ≥ 0
t t
R(bit , Rtb , At ) =
R b A + R if bit < 0.
t t

The extra wedge for unsecured borrowing creates a mass of households with zero unsecured
credit but with the possibility to borrow, though at a penalty rate. The intermediation
efficiency wedge At can be thought of as a cost of a banking sector turning government
bonds into deposits. This cost follows an AR(1) process in logs and fluctuates in response to
shocks, A t . If At goes down, households will implicitly demand less government bonds and
find it more attractive to save in (illiquid) real capital, akin to the “risk-premium shock” in
Smets and Wouters (2007).
Since a household’s saving decision will be some non-linear function of that household’s
wealth and productivity, inflation and all other prices will be functions of the joint distribu-
tion, Θt , of (b, k, h) in t. This makes Θ a state variable of the household’s planning problem
and this distribution evolves as a result of the economy’s reaction to aggregate shocks. For
simplicity, we summarize all effects of aggregate state variables, including the distribution
of wealth and income, by writing the dynamic planning problem with time-dependent con-
tinuation values.
This leaves us with three functions that characterize the household’s problem: value
function V a for the case where the household adjusts its capital holdings, the function V n

9
for the case in which it does not adjust, and the expected envelope value, EV , over both:

Vta (b, k, h) = max

0 0
u[x(b, b0a , k, k 0 , h)] + βEt Vt+1 (b0a , k 0 , h)
k ,ba

Vtn (b, k, h) = max u[x(b, b0n , k, k, h)] + βEt Vt+1 (b0n , k, h) (12)
b0n

Et Vt+1 (b0 , k 0 , h) =Et λVt+1

 a 0 0  n
(b0 , k, h)
 
(b , k , h) + Et (1 − λ)Vt+1

Expectations about the continuation value are taken with respect to all stochastic processes
conditional on the current states, including time-varying income risk. Maximization is sub-
ject to the corresponding budget constraint.

2.2 Firm Sector

The firm sector consists of four sub-sectors: (a) a labor sector composed of “unions” that
differentiate raw labor and labor packers who buy differentiated labor and then sell labor
services to intermediate goods producers, (b) intermediate goods producers who hire labor
services and rent out capital to produce goods, (c) final goods producers who differentiate
intermediate goods and then sell them to goods bundlers, who finally sell them as consump-
tion goods to households, and to (d) capital goods producers, who turn bundled final goods
into capital goods.
When profit maximization decisions in the firm sector require intertemporal decisions
(i.e. in price and wage setting and in producing capital goods), we assume for tractability
that they are delegated to a mass-zero group of households (managers) that are risk neutral
and compensated by a share in profits.11 They do not participate in any asset market and
have the same discount factor as all other households. Since managers are a mass-zero group
in the economy, their consumption does not show up in any resource constraint and all but
the unions’ profits go to the entrepreneur households (whose h = 0). Union profits go lump
sum to worker households.

2.2.1 Labor Packers and Unions

Worker households sell their labor services to a mass-one continuum of unions indexed by j,
each of whom offers a different variety of labor to labor packers who then provide labor ser-
vices to intermediate goods producers. Labor packers produce final labor services according
11
Since we solve the model by a first-order perturbation in aggregate shocks, the assumption of risk-
neutrality only serves as a simplification in terms of writing down the model. With a first-order perturbation
we have certainty equivalence and fluctuations in stochastic discount factors become irrelevant.

10
to the production function
Z ζt −1
 ζ ζ−1
t
t
ζt
Nt = n̂jt dj , (13)

out of labor varieties n̂jt . Cost minimization by labor packers implies that each variety of
labor, each union j, faces a downward-sloping demand curve
 −ζt
Wjt
n̂jt = Nt ,
WtF

where Wjt is the nominal wage set by union j and WtF is the nominal wage at which labor
packers sell labor services to final goods producers.
Since unions have market power, they pay the households a wage lower than the price
at which they sell labor to labor packers. Given the nominal wage Wt at which they buy
labor from households and given the nominal wage index WtF , unions seek to maximize their
discounted stream of profits. However, they face a Calvo-type (1983) of adjustment friction
with indexation with the probability λw to keep wages constant. They therefore maximize

∞
( −ζt )
WF t t

X Wjt π̄W Wt Wjt π̄W
E0 β t λtw t Nt F
− F , (14)
t=0
Pt Wt Wt WtF

by setting Wjt in period t and keeping it constant except for indexation to π̄W , the steady-
state wage inflation rate.
Since all unions are symmetric, we focus on a symmetric equilibrium and obtain the
linearized wage Phillips curve from the corresponding first-order condition as follows, leaving
out all terms irrelevant at a first-order approximation around the stationary equilibrium:
W
     
πtW πt+1 wt 1
log π̄W
= βEt log π̄W
+ κw wtF
− µW
, (15)
t

WtF wtF
with πtW := F
Wt−1
= F
wt−1
πtY being wage inflation, wt and wtF being the respective real
wages for households and firms, and µ1W = ζtζ−1 t
being the target mark-down of wages
t
the unions pay to households, Wt , relative to the wages charged to firms, WtF and κw =
(1−λw )(1−λw β)
λw
. This target fluctuates in response to markup shocks, µW
t , and follows a log
12
AR(1) process.
12
Including the first-order irrelevant terms, the Phillips curve reads
 W h  W  F i  
π π 1−τt+1 ζt+1 Wt+1 Pt Nt+1 wt 1
log π̄tW = βEt log π̄t+1 W 1−τt ζt W F Pt+1 Nt + κw wF − µW
t t t

where τt is the average income tax.

11
2.2.2 Final Goods Producers

Similar to unions, final goods producers differentiate a homogeneous intermediate good and
set prices. They face a downward-sloping demand curve

yjt = (pjt /Pt )−ηt Yt

for each good j and buy the intermediate good at the nominal price M Ct . As we do for
unions, we assume price adjustment frictions à la Calvo (1983) with indexation.
Under this assumption, the firms’ managers maximize the present value of real profits
given this price adjustment friction, i.e., they maximize:

∞
( −ηt )1−τtP
pjt π̄Yt pjt π̄ t

X 1−τ P M Ct
E0 β t λtY (1 − τtL )Yt t − , (16)
t=0
Pt Pt Pt

with a time constant discount factor.

The corresponding first-order condition for price setting implies a Phillips curve
π  π   
t t+1 1
log = βEt log + κY mct − µY
, (17)
π̄ π̄ t

where we again dropped all terms irrelevant for a first-order approximation and have κY =
(1−λY )(1−λY β)
λY
. Here, πt is the gross inflation rate of final goods, πt := PPt−1
t
, mct := MPCt t is
the real marginal costs, π̄ is steady-state inflation and µYt = ηtη−1t
is the target markup. As
for the unions, this target fluctuates in response to markup shocks, µY , and follows a log
AR(1) process.

2.2.3 Intermediate Goods Producers

Intermediate goods are produced with a constant returns to scale production function:

Yt = Zt Ntα (ut Kt )(1−α) ,

where Zt is total factor productivity and follows an autoregressive process in logs, and ut Kt
is the effective capital stock taking into account utilization ut , i.e., the intensity with which
the existing capital stock is used. Using capital with an intensity higher than normal results
in increased depreciation of capital according to δ (ut ) = δ0 + δ1 (ut − 1) + δ2 /2 (ut − 1)2 ,
which, assuming δ1 , δ2 > 0, is an increasing and convex function of utilization. Without loss
of generality, capital utilization in the steady state is normalized to 1, so that δ0 denotes the
steady-state depreciation rate of capital goods.

12
 Let mct be the relative price at which the intermediate good is sold to final goods pro-
ducers. The intermediate goods producer maximizes profits,

mct Zt Yt − wtF Nt − [rt + qt δ(ut )] Kt ,

where rtF and qt are the rental rate of firms and the (producer) price of capital goods re-
spectively. The intermediate goods producer operates in perfectly competitive markets, such
that the real wage and the user costs of capital are given by the marginal products of labor
and effective capital:
 1−α
ut Kt
wtF = αmct Zt , (18)
Nt
 α
Nt
rt + qt δ(ut ) = ut (1 − α)mct Zt . (19)
ut Kt

We assume that utilization is decided by the owners of the capital goods, taking the
aggregate supply of capital services as given. The optimality condition for utilization is
given by  α
Nt
qt [δ1 + δ2 (ut − 1)] = (1 − α)mct Zt , (20)
ut Kt
i.e., capital owners increase utilization until the marginal maintenance costs equal the marginal
product of capital services.

2.2.4 Capital Goods Producers

Capital goods producers take the relative price of capital goods, qt , as given in deciding
about their output, i.e., they maximize

∞
( "  2 # )
X φ It
E0 β t It Ψt qt 1− log −1 , (21)
t=0
2 It−1

where Ψt governs the marginal efficiency of investment à la Justiniano et al. (2010, 2011),
which follows an AR(1) process in logs and is subject to shocks Ψ
t .
13

Optimality of the capital goods production requires (again dropping all terms irrelevant
13
This shock has to be distinguished from a shock to the relative price of investment, which has been shown
in the literature (Justiniano et al., 2011; Schmitt-Grohé and Uribe, 2012) to not be an important driver of
business cycles as soon as one includes the relative price of investment as an observable. We therefore focus
on the MEI shock.

13
up to first order)
    
It It+1
Ψt qt 1 − φ log = 1 − βEt Ψt+1 qt+1 φ log , (22)
It−1 It

and each capital goods producer will adjust its production until (22) is fulfilled.
Since all capital goods producers are symmetric, we obtain as the law of motion for
aggregate capital
"  2 #
φ It
Kt − (1 − δ(ut )) Kt−1 = Ψt 1− log It . (23)
2 It−1

The functional form assumption implies that investment adjustment costs are minimized
and equal to 0 in the steady state.

2.3 Government
The government operates a monetary and a fiscal authority. The monetary authority con-
trols the nominal interest rate on liquid assets, while the fiscal authority issues government
bonds to finance deficits, chooses both the average tax rate in the economy as well as tax
progressivity, and adjusts expenditures to stabilize debt in the long run and output in the
short run.
We assume that monetary policy sets the nominal interest rate following a Taylor-type
(1993) rule with interest rate smoothing:

b
ρR  (1−ρR )θY
Rtb
 
Rt+1 πt (1−ρR )θπ Yt
= R
t . (24)
R̄b R̄b π̄ Yt∗

The coefficient R̄b ≥ 0 determines the nominal interest rate in the steady state. The coeffi-
cients θπ , θY ≥ 0 govern the extent to which the central bank attempts to stabilize inflation
and the output gap, where the gap, YYt∗ , is defined relative to what output would be at
t
stationary equilibrium markups, Yt∗ . ρR ≥ 0 captures interest rate smoothing.
We assume that the government runs a budget deficit and hence accumulates debt gov-
erned by a rule (c.f. Woodford, 1995):
 −γB   γY
Bt+1 Bt πt γπ Yt ρG G
= Dt , Dt = Dt−1 t , (25)
Bt B̄ π̄ Yt∗

where Dt is a persistent shock to the government’s structural deficit. Besides issuing bonds,
the government uses tax revenues Tt , defined below, to finance government consumption, Gt ,

14
and interest on debt. The parameters γB , γY , and γπ measure, respectively, how the deficit
reacts to outstanding debt, the ouput gap, and inflation.
The government sets the average tax rate in the economy according to a similar rule
 (1−ρτ )γBτ  (1−ρτ )γYτ
τt  τt−1 ρτ Bt Yt
= τt . (26)
τ̄ τ̄ B̄ Yt∗

The parameter τtP that governs the progressivity of the tax schedule evolves according to

P
ρP
τtP

τt−1
= Pt . (27)
τ̄ P τ̄ P

The level parameter of the tax code τtL adjusts such that the average tax rate on income
equals this target level:


τ P
Et wt nit hit + Ihit =0 ΠFt − τtL Et wt nit hit + Ihit =0 ΠFt t
τt = , (28)
Et wt nit hit + Ihit =0 ΠFt

where Et is the expectation operator, which here gives the cross-sectional average. Total taxes


Tt are then Tt = τt wt nit hit + Ihit 6=0 ΠUt + Ihit =0 ΠFt and the government budget constraint
determines government spending residually: Gt = Bt+1 + Tt − Rtb /πt Bt .
There are thus four shocks to government rules: monetary policy shocks, R t , tax pro-
P τ
gressivity shocks t , tax level shocks t , and structural deficit, i.e., government spending,
shocks, Gt . We assume these shocks to be log normally distributed with mean zero.

2.4 Goods, Bonds, Capital, and Labor Market Clearing

The labor market clears at the competitive wage given in (18). The bond market clears
whenever the following equation holds:

Bt+1 = B d (Rtb , At , rt , qt , ΠFt , ΠUt , wt , πt , τt , τtP , Θt , Vt+1 ) := Et λb∗a,t + (1 − λ)b∗n,t ,

 
(29)

where b∗a,t , b∗n,t are functions of the states (b, k, h), and depend on how households value
asset holdings in the future, Vt+1 (b, k, h), and the current set of prices (and tax rates)
(Rtb , At , rt , qt , ΠFt , ΠUt , wt , πt , τt , τtP ). Future prices do not show up because we can express
the value functions such that they summarize all relevant information on the expected future
price paths. Expectations in the right-hand-side expression are taken w.r.t. the distribu-
tion Θt (b, k, h). Equilibrium requires the total net amount of bonds the household sector
demands, B d , to equal the supply of government bonds. In gross terms there are more liquid

15
assets in circulation as some households borrow up to B.
Last, the market for capital has to clear:

Kt+1 = K d (Rtb , At , rt , qt , ΠFt , ΠUt , wt , πt , τt , τtP , Θt , Vt+1 ) := Et [λkt ∗ + (1 − λ)k] , (30)

where the first equation stems from competition in the production of capital goods, and the
second equation defines the aggregate supply of funds from households – both those that
trade capital, λkt∗ , and those that do not, (1 − λ)k. Again kt∗ is a function of the current
prices and continuation values. The goods market then clears due to Walras’ law, whenever
labor, bonds, and capital markets clear.

2.5 Equilibrium
A sequential equilibrium with recursive planning in our model is a sequence of policy func-
tions {x∗a,t , x∗n,t , b∗a,t , b∗n,t , kt∗ }, a sequence of value functions {Vta , Vtn }, a sequence of prices
{wt , wtF , ΠFt , ΠUt , qt , rt , Rtb , πt , πtW , τt , τtP }, a sequence of stochastic states At , Ψt , Zt and shocks
G P τ A Z Ψ µW µY
R σ
t , t , t , t , t , t , t , t , t , t , aggregate capital and labor supplies {Kt , Nt }, distribu-
tions Θt over individual asset holdings and productivity, and expectations Γ for the distri-
bution of future prices, such that
1. Given the functional Et Vt+1 for the continuation value and period-t prices, policy func-
tions {x∗a,t , x∗n,t , b∗a,t , b∗n,t , kt∗ } solve the households’ planning problem, and given the
policy functions {x∗a,t , x∗n,t , b∗a,t , b∗n,t , kt∗ }, prices, and the value functions {Vta , Vtn } are a
solution to the Bellman equation (12).

2. Distributions of wealth and income evolve according to households’ policy functions.

3. The labor, the final goods, the bond, the capital, and the intermediate goods markets
clear in every period, interest rates on bonds are set according to the central bank’s
Taylor rule, fiscal policies are set according to the fiscal rules, and stochastic processes
evolve according to their law of motion.

4. Expectations are model consistent.

16
3 Numerical Solution and Estimation Technique
We solve the model by perturbation methods. We choose a first-order Taylor expansion
around the stationary equilibrium following the method of Bayer and Luetticke (2018). This
method replaces the value functions with linear interpolants and the distribution functions
with histograms to calculate a stationary equilibrium. Then it performs dimensionality
reduction before linearization but after calculation of the stationary equilibrium. The di-
mensionality reduction is achieved by using discrete cosine transformations (DCT) for the
value functions and perturbing only the largest coefficients of this transformation.and by
approximating the joint distributions through distributions with a fixed copula and flexible
marginals. We solve the model originally on a grid of 80x80x22 points for liquid assets,
illiquid assets, and income, respectively. The dimensionality-reduced number of states and
controls in our system is roughly 900.
Approximating the sequential equilibrium in a linear state-space representation then boils
down to the linearized solution of a non-linear difference equation

Et F (xt , Xt , xt+1 , Xt+1 , σΣt+1 ), (31)

where xt is “idiosyncratic” states and controls: the value and distribution functions, and Xt
is aggregate states and controls: prices, quantities, productivities, etc. The error term t
represents fundamental shocks. Importantly, we can also order the equations in a similar
way. The law of motion for the distribution and the Bellman equations describe a non-
linear difference equation for the idiosyncratic variables, and all other optimality and market
clearing conditions describe a non-linear difference equation for the aggregate variables. By
introducing auxiliary variables that capture the mean of b, k, and h, we make sure that the
distribution itself does not show up in any aggregate equation other than in the one for the
summary variables. Yet, these equations are free of all model parameters.
This helps substantially in estimating the model. For each parameter draw, we need to
calculate the Jacobian of F and then use the Klein-algorithm (2000) (see also Schmitt-Grohé
and Uribe, 2004) to obtain a linear state-space representation, which we then feed into a
Kalman filter to obtain the likelihood of the data given our model. However, most model
parameters do not show up in the Bellman equation. Only ρh , σ̄h , λ, β, γ, and ξ do, but these
parameters we do not estimate but calibrate from the stationary equilibrium.14 Therefore,
the Jacobian of the “idiosyncratic equations” is unaltered by all parameters that we estimate
and we only need to calculate it once. Similarly, “idiosyncratic variables” (i.e., the value
14
Note that the scaling of idiosyncratic risk, st , shows up in the Bellman equation, but similar to a price
and not as a parameter.

17
functions and the histograms) only affect the aggregate equations through their parameter-
free effect on summary variables, such that this part of the Jacobian also does not need to
be updated during the estimation. This leaves us with the same number of derivatives to be
calculated for every parameter draw during the estimation as in a representative-agent model.
Still, solving for the state-space representation and evaluating the likelihood are substantially
more time consuming and computing the likelihood of a given parameter draw takes roughly
4 to 5 seconds on a workstation computer, 90% of the computing time goes into the Schur
decomposition, which is still much larger because of the many additional “idiosyncratic”
states (histograms) and controls (marginal value functions) the system contains.
We use a Bayesian likelihood approach as described in An and Schorfheide (2007) and
Fernández-Villaverde (2010) for parameter estimation. In particular, we use the Kalman
filter to obtain the likelihood from the state-space representation of the model solution15
and employ a standard random walk Metropolis-Hastings algorithm to generate draws from
the posterior likelihood. Smoothed estimates of the states at the posterior mean of the
parameters are obtained via a Kalman smoother of the type described in Koopman and
Durbin (2000) and Durbin and Koopman (2012).

4 Calibration, Data, and Priors

We follow a two-step procedure to estimate the model. First, we calibrate or fix all pa-
rameters that affect the steady state of the model. Second, we estimate by full-information
methods all parameters that only matter for the dynamics of the model, i.e., the aggregate
shocks and frictions. Table 1 summarizes the calibrated and externally chosen parameters
and Table 3 lists the estimated parameters. One period in the model refers to a quarter of
a year.

4.1 Calibrated Parameters

We fix a number of parameters either following the literature or targeting steady-state ratios;
see Table 1 (all at quarterly frequency of the model). For the household side, we set the
relative risk aversion to 4, which is common in the incomplete markets literature; see Kaplan
et al. (2018). We set the Frisch elasticity to 0.5; see Chetty et al. (2011). We take estimates
15
The Kalman filter allows us to deal with missing values and mixed frequency data quite naturally. For
a one-frequency data set without missing values, one can speed up the estimation by employing so-called
“Chandrasekhar recursions” for evaluating the likelihood. These recursions replace the costly updating of the
state variance matrix by multiplications involving matrices of much lower dimension (see Herbst, 2014, for
details). This is especially relevant for the two-asset HANK model as the speed of evaluating the likelihood
is dominated by the updating of the state variance matrix, which involves the multiplication of matrices that
are quadratic in the number of states.

18
 Table 1: External/calibrated parameters (quarterly frequency)

Parameter Value Description Target

Households
β 0.981 Discount factor see Table 2
ξ 4 Relative risk aversion Kaplan et al. (2018)
γ 2 Inverse of Frisch elasticity Chetty et al. (2011)
λ 0.065 Portfolio adj. prob. see Table 2
ρh 0.98 Persistence labor income Storesletten et al. (2004)
σh 0.12 STD labor income Storesletten et al. (2004)
ζ 1/5000 Trans.prob. from W. to E. see Table 2
ι 1/16 Trans.prob. from E. to W. Guvenen et al. (2014)
R̄ 1.65% Borrowing penalty see Table 2
Firms
α 0.68 Share of labor 62% labor income
δ0 1.75% Depreciation rate 7.0% p.a.
η̄ 11 Elasticity of substitution Price markup 10%
ζ̄ 11 Elasticity of substitution Wage markup 10%
Government
τ̄ L 0.12 Tax rate level G/Y = 20%
P
τ̄ 0.18 Tax progressivity Heathcote et al. (2017)
R̄b 1.004 Nominal rate 1.6% p.a.
π̄ 1.00 Inflation 0% p.a.

for idiosyncratic income risk from Storesletten et al. (2004), ρh = 0.98 and σ̄h = 0.12.
Guvenen et al. (2014) provide the probability that a household will fall out of the top 1%
of the income distribution in a given year, which we take as the transition probability from
entrepreneur to worker, ι = 1/16.
Table 2 summarizes the calibration of the remaining household parameters. We match
4 targets: 1) average illiquid assets (K/Y=11.44), 2) average liquidity (B/Y=1.58), 3) the
fraction of borrowers, 16%, and 4) the average top 10% share of wealth, which is 67%.
This yields a discount factor of 0.981, a portfolio adjustment probability of 6.5%, borrowing
penalty of 1.65% quarterly (given a borrowing limit of two times average quarterly income),
and a transition probability from worker to entrepreneur of 1/5000.16
For the firm side, we set the labor share in production, α, to 68% to match a labor
income share of 62%, which corresponds to the average BLS labor share measure over 1954-
2015. The depreciation rate is 1.75% per quarter. An elasticity of substitution between
differentiated goods of 11 yields a markup of 10%. The elasticity of substitution between
labor varieties is also set to 11, yielding a wage markup of 10%. All are standard values in
16
Detailed data sources can be found in Appendix A.

19
 Table 2: Calibrated parameters

Targets Model Data Source Parameter

Mean illiquid assets (K/Y) 11.44 11.44 NIPA Discount factor

Mean liquidity (B/Y) 1.58 1.58 FRED Port. adj. probability
Top10 wealth share 0.67 0.67 WID Fraction of entrepreneurs
Fraction borrowers 0.16 0.16 SCF Borrowing penalty

the literature.
The government taxes labor and profit income using a non-linear tax schedule that ap-
proximates the progressivity of the US tax system; see Heathcote et al. (2017). The progres-
sivity parameter, τ P = 0.18, is taken from Heathcote et al. (2017). The level of taxes, τ L ,
is set to clear the government budget constraint that corresponds to a government share of
G/Y = 20%. The policy rate is set to an annualized rate of 1.6%. This corresponds to the
average federal funds rate in real terms over 1954-2015. We set steady-state inflation to zero
as we have assumed indexation to the steady-state inflation rate in the Phillips curves.

4.2 Estimation Data

We use quarterly US data from 1954Q3 to 2015Q4 and include the following eight observable
time series: the growth rates of per capita GDP, private consumption, investment, federal
tax receipts, and wages, all in real terms, the logarithm of the level of per capita hours
worked, the log difference of the GDP deflator, and the (shadow) federal funds rate. We add
more data with shorter and/or non-quarterly availability. Idiosyncratic income uncertainty
(estimated from panel data in the Survey of Income and Program Participation (SIPP) in
Bayer et al., 2019) is available as quarterly series from 1983Q1 to 2013Q1 and included in
log-levels. We proxy the progressivity of the US tax and transfer system by the highest
bracket of the US individual income tax. This series is available at annual frequency from
1954 to 2015. Wealth and income shares of the top 10% are included at annual frequency
and available from 1954 to 2014 from the World Inequality Database (drawing on work by
Piketty, Saez, and Zucman; see, e.g., Piketty and Saez (2003) or Saez and Zucman (2016)).

20
4.3 Prior Distributions
Columns 1-4 of Table 3 presents the parameters we estimate and their assumed prior dis-
tributions. The posterior distribution is discussed in the next section. Where available, we
use prior values that are standard in the literature and independent of the underlying data.
Following Justiniano et al. (2011), we impose a gamma distribution with prior mean of 5.0
and standard deviation of 2.0 for δ2 /δ1 , the elasticity of marginal depreciation with respect
to capacity utilization, and a gamma prior with mean 4.0 and standard deviation of 2.0 for
the parameter controlling investment adjustment costs, φ. For the slopes of price and wage
Phillips curves, κY and κw , we assume gamma priors with mean 0.1 and standard deviation
0.02, which corresponds to price and wage contracts having an average length of one year.
Following Smets and Wouters (2007), the autoregressive parameters of the shock processes
are assumed to follow a beta distribution with mean 0.5 and standard deviation 0.2. The
standard deviations of the shocks follow inverse-gamma distributions with prior mean 0.1%
and standard deviation 2%. The only exception is the uncertainty shock, where, given the
evidence in Bayer et al. (2019), we use a higher prior mean of 1.0. The employment feedback
parameter in the uncertainty process is assumed to follow a normal prior with large variance.
Regarding policy, for the inflation and output feedback parameters in the Taylor-rule, θπ
and θY , we impose normal distributions with prior means of 1.7 and 0.13, respectively, while
the interest rate smoothing parameter ρR has the same prior distribution as the persistence
parameters of the shock processes. In the bond rule, the debt-feedback parameter γB is
assumed to follow a gamma distribution with mean 0.10 and standard deviation 0.08, such
that the prior for the autocorrelation of debt is centered around 0.9, implying a half-life
of a deviation in debt of between one and eight years. The parameters governing feedback
to inflation and output, γπ and γY , follow standard normal distributions. Similarly, the
autoregressive parameters, in the tax rules, ρi where i ∈ {P, τ }, are assumed to follow beta
distributions (with mean 0.5 and standard deviation 0.2), while the feedback parameters, γYτ
and γBτ , follow standard normal distributions.
The standard deviations of the measurement errors are assumed to have inverse-gamma
prior distributions. For the uncertainty series, we set a relatively high prior mean of 5.00%,
while for all other measurement errors we set lower prior means of 0.05%. As the top marginal
tax rate does not directly map into how we model progressivity, we allow for a linear scaling
parameter ∝τ (Boivin and Giannoni, 2006), following a standard normal prior, in addition
to the measurement error.

21
 Table 3: Prior and posterior distributions of estimated parameters

Parameter Prior Posterior HANK Posterior HANK*

Distribution Mean Std. Dev. Mean Std. Dev. 5% 95 % Mean Std. Dev. 5% 95%

Frictions

δs Gamma 5.00 2.00 1.483 0.137 1.265 1.717 1.420 0.031 1.371 1.472
φ Gamma 4.00 2.00 0.233 0.032 0.182 0.288 0.218 0.036 0.162 0.278
κ Gamma 0.10 0.02 0.101 0.014 0.080 0.126 0.105 0.014 0.083 0.129
κw Gamma 0.10 0.02 0.128 0.017 0.100 0.157 0.133 0.019 0.103 0.164

Debt and monetary policy rules

ρR Beta 0.50 0.20 0.800 0.016 0.774 0.825 0.803 0.014 0.779 0.826
σR Inv.-Gamma 0.10 2.00 0.265 0.015 0.241 0.291 0.266 0.016 0.242 0.293
θπ Normal 1.70 0.30 2.603 0.125 2.404 2.817 2.614 0.053 2.520 2.695
θy Normal 0.13 0.05 0.086 0.020 0.054 0.119 0.078 0.018 0.048 0.107
22

γB Gamma 0.10 0.08 0.144 0.018 0.116 0.174 0.157 0.021 0.119 0.190
γπ Normal 0.00 1.00 -1.134 0.031 -1.185 -1.083 -1.175 0.025 -1.215 -1.135
γY Normal 0.00 1.00 -0.716 0.030 -0.765 -0.668 -0.697 0.026 -0.741 -0.655
ρG Beta 0.50 0.20 0.988 0.008 0.972 0.997 0.992 0.005 0.981 0.998
σG Inv.-Gamma 0.10 2.00 0.260 0.015 0.237 0.286 0.263 0.016 0.238 0.289

Tax rules

ρτ Beta 0.50 0.20 0.550 0.029 0.502 0.597 0.552 0.058 0.457 0.648
στ Inv.-Gamma 1.00 2.00 0.127 0.156 0.022 0.537 0.122 0.171 0.021 0.544
τ
γB Normal 0.00 1.00 0.777 0.062 0.674 0.883 0.833 0.075 0.697 0.953
γYτ Normal 0.00 1.00 2.646 0.176 2.360 2.943 2.496 0.134 2.303 2.755
ρP Beta 0.50 0.20 0.986 0.006 0.974 0.995 0.988 0.006 0.978 0.996
σP Inv.-Gamma 1.00 2.00 2.220 0.629 1.354 3.422 2.014 0.532 1.289 3.027
∝τ Normal 0.00 1.00 2.226 0.513 1.503 3.177 2.408 0.510 1.648 3.309
 Table 3: Prior and posterior distributions of estimated parameters - continued

Parameter Prior Posterior HANK Posterior HANK*

Distribution Mean Std. Dev. Mean Std. Dev. 5% 95 % Mean Std. Dev. 5% 95%

Structural Shocks

ρA Beta 0.50 0.20 0.977 0.011 0.957 0.992 0.983 0.009 0.965 0.995
σA Inv.-Gamma 0.10 2.00 0.160 0.014 0.139 0.184 0.159 0.013 0.140 0.181
ρZ Beta 0.50 0.20 0.995 0.002 0.991 0.999 0.992 0.003 0.986 0.996
σZ Inv.-Gamma 0.10 2.00 0.601 0.028 0.558 0.649 0.608 0.029 0.564 0.658
ρΨ Beta 0.50 0.20 0.976 0.008 0.963 0.988 0.965 0.009 0.950 0.978
σΨ Inv.-Gamma 0.10 2.00 2.723 0.229 2.362 3.107 2.531 0.214 2.197 2.895
ρµ Beta 0.50 0.20 0.889 0.022 0.852 0.923 0.900 0.021 0.863 0.933
σµ Inv.-Gamma 0.10 2.00 1.695 0.149 1.471 1.958 1.645 0.139 1.435 1.889
ρµw Beta 0.50 0.20 0.909 0.020 0.873 0.938 0.909 0.020 0.875 0.938
23

σµw Inv.-Gamma 0.10 2.00 5.355 0.513 4.605 6.272 5.216 0.541 4.441 6.158

Income Risk Process

ρs Beta 0.50 0.20 0.657 0.032 0.602 0.708 0.639 0.038 0.577 0.699
σs Inv.-Gamma 1.00 2.00 63.184 4.257 56.561 70.559 61.443 4.177 55.144 68.804
ΣN Normal 0.00 100.00 0.834 0.127 0.615 1.038 0.634 0.086 0.469 0.756

Measurement Errors

σsme Inv.-Gamma 5.00 10.00 4.267 3.208 1.217 11.126 4.076 3.164 1.195 10.412
σTme Inv.-Gamma 0.05 0.01 3.695 0.170 3.427 3.985 3.707 0.174 3.433 3.999
στme
p Inv.-Gamma 0.05 0.01 0.056 0.065 0.012 0.193 0.038 0.027 0.012 0.091
me
σW I Inv.-Gamma 0.05 0.01 – – – – 3.772 0.402 3.168 4.475
me
σII Inv.-Gamma 0.05 0.01 – – – – 8.074 0.819 6.845 9.514

Notes: The standard deviations of the shocks and measurement errors have been transformed into percentages by multiplying by 100. HANK and
HANK* denote posterior estimates for the model without and with observable inequality series, respectively.
5 US Business Cycles
One key advantage of HANK models is that we can use them to understand the distribu-
tional consequences of business cycle shocks and policies. This raises three questions. First,
does the inclusion of measures of inequality change what the model infers about shocks and
frictions in business cycles? Second, to what extent do business cycle shocks explain the
movements in inequality measures? Third, how would inequality have developed if govern-
ment business cycle policies had been different?
To answer these questions, we estimate the HANK model with and without additional
observables (plus measurement error) for the shares of wealth and income held by the top
10% of households in each dimension, which are taken from the World Inequality Database.
The reason we focus on the top 10% wealth and income share is that this measure is most
consistent across alternative, but less frequently available, data sources such as the Survey
of Consumer Finances (SCF); see Kopczuk (2015).17
In this section, we answer the first question by comparing parameter estimates, variance
decompositions, and historical decompositions of US business cycles for the estimated HANK
model with and without data on inequality. We postpone the model implications for US
inequality to Section 6.

5.1 Parameter Estimates w/ and w/o Inequality

Table 3 reports the posterior distributions across the two main estimation variants: HANK
without data on inequality and HANK* with data on inequality.18 Strikingly, the parameter
estimates with and without data on inequality are basically the same; none of the estimated
parameters is substantially different across the two estimations. Only some parameters of the
fiscal policy rule (γBτ , γYτ ) have significantly different posterior distributions. Using inequality
data leads the model to view tax policy as slightly more aggressive in stabilizing government
debt.
This implies that both inequality measures provide little additional identification of busi-
17
We abstain from including other cross-sectional data in the estimation, such as the Panel Study of Income
Dynamics (PSID) or the Survey of Consumer Finances (SCF) to avoid dealing with two measurements of
the same model variable. Since the income risk data we use for both sets of estimation exercises do contain
some cross-sectional information on income inequality (it is constructed from SIPP data), we also consider
a third estimation exercise where we also drop the income risk data. Results are available in Appendix E
and are in general highly similar to the other two sets of estimation results.
18
The estimation is conducted with 5 parallel RWMH chains started from an over-dispersed target distri-
bution after an extensive mode search. After burn in, 300,000 draws from the posterior are used to compute
the posterior statistics. The acceptance rates across chains are between 26% and 29%. Appendix F pro-
vides Gelman and Rubin (1992) convergence statistics. Traceplots of individual parameters are available on
request.

24
ness cycle shocks and frictions. In the next section on US inequality, we show that already
the model estimated only on aggregate data implies a U-shaped evolution of inequality from
1950 to 2015 in line with the data. This explains why adding data on inequality has little
effect on the estimated parameters. The estimated shocks and frictions do a good job in
matching the evolution of wealth and income inequality over the last 60 years. We will
explore this result in detail in the next section.
The parameter estimates are broadly in line with the representative-agent literature
(which corresponds to our priors that are taken from this literature). Real frictions are
an exception. They are up to one order of magnitude smaller in our estimation. In par-
ticular, investment adjustment costs are substantially smaller. This reflects the portfolio
adjustment costs at the household level that generate inertia in aggregate investment. Our
estimates for nominal frictions are standard and close to the priors, with price and wage
stickiness being less than 4 quarters on average. In terms of shocks, the estimated persis-
tence and variance for the seven “standard” shocks are comparable to the results of Smets
and Wouters (2007). The persistence ranges from 0.995 for TFP to 0.889 for wage markups.
The variance ranges from 0.2% for risk premium shocks to 5% for wage-markup shocks.
Idiosyncratic income risk as a driver of portfolio allocations and consumption demand has
been highlighted in Bayer et al. (2019). Our estimate for the shock series mostly coincides
with the observed series for income risk, i.e., we find income risk shocks to be slightly less
persistent than the risk series itself ρσ = 0.66 and the variance slightly larger at 63%. While
our estimates imply that there is endogenous amplification of uncertainty, the feedback is
negligible in economic terms; see Appendix C for the historical time series of income risk
implied by the model. It is important to note here that the income process we use focuses
on the uncertainty of persistent shocks to income. As we do not model job-search and
unemployment, we abstract from the short-run and transitory income risk fluctuations that
others have discussed (e.g., Ravn and Sterk, 2017; Den Haan et al., 2017).
In terms of the estimated policy coefficients, the estimated Taylor rule is in line with the
literature. The coefficients on inflation and output deviations are 2.6 and 0.1, and there is
substantial inertia 0.8. The fiscal rule that governs deficits and hence government spending
exhibits a countercyclical response to inflation and output deviations, −1.1 and −0.7, and
features persistence as well. The tax rule that governs average taxation has similar properties.
Average tax rates rise when output or debt is high, but not very persistently. Changes to
tax progressivity, by contrast, are very persistent, 0.99.

25
 Figure 1: Variance decompositions: Output growth and its components

Growth of output Yt Growth of consumption Ct

HANK

HANK
HANK*

HANK*
0 20 40 60 80 100 0 20 40 60 80 100

Growth of investment It Growth of government exp. Gt

HANK

HANK
HANK*

HANK*
0 20 40 60 80 100 0 20 40 60 80 100

Notes: Conditional variance decompositions at a 4-quarter forecast horizon. HANK* [HANK]

corresponds to the estimated HANK model with inequality data [w/o inequality data].

5.2 Variance Decompositions of Business Cycles

Next, we show what the estimated parameters imply for our view of US business cycles
by looking at variance decompositions at business cycle frequency.19 Figure 1 shows these
decompositions for the growth rates of output, consumption, investment, and government
spending. Unsurprisingly, we find very similar decompositions for the estimations with
and without using inequality data. As in the representative-agent literature, TFP and
investment-specific technology shocks are the most important drivers of output growth each
explaining roughly 25% of its conditional variance at business cycle frequency. All together,
supply side shocks (the two markup and the two productivity shocks)20 account for almost
75% of output volatility.
Demand side shocks (i.e., shocks to uncertainty, fiscal policy, monetary policy, and the
19
We report the conditional variances at 4 quarters.
20
The grouping of the shocks is based on the question of whether they directly affect the Phillips curve
as the relevant aggregate supply function or primarily work through affecting the bond-market clearing
condition, as the aggregate demand function.

26
 Table 4: Contribution of shocks to US recessions

Output growth Consumption growth

Shocks HANK HANK* HANK HANK*

TFP, Z -0.23 -0.27 -0.16 -0.18

Inv.-spec. tech., Ψ -0.25 -0.22 0.14 0.15
Price markup, µY -0.03 -0.03 -0.10 -0.11
Wage markup, µW -0.31 -0.31 -0.52 -0.50
Risk premium, A -0.27 -0.27 -0.39 -0.38
Monetary policy, R 0.12 0.13 0.23 0.23
Structural deficit, G 0.02 0.02 -0.03 -0.07
Tax level, τ 0.00 0.00 0.00 0.00
P
Tax progressivity, τ 0.01 0.01 0.00 0.00
Income risk, σ -0.06 -0.05 -0.18 -0.14

Notes: The table displays the average contribution of the various shocks
during an NBER-dated recession that result from our historical shock de-
composition. Values are calculated by averaging the value of each shock
component over all NBER recession quarters. To improve readability,
we normalized the size of the overall contraction to −1%. In the data,
the average is −1.24% for output and −0.5% for consumption.

risk premium), by contrast, explain around 55% of consumption volatility. A new source
of fluctuations in consumption is income risk shocks. They explain 10% of the volatility of
consumption growth. This is so because income risk is mostly exogenous.

5.3 Historical Decompositions of US Business Cycles

While the variance decompositions help us understand the average cycle implied by the
model, a historical decomposition tells us how the model views the actual cycles that the
US economy has gone through. We summarize the historical decomposition of NBER-dated
recessions in Table 4.21 During the average NBER-dated recession quarter, 0.82% of a 1.0%
decline in output results from technology and markup shocks. In particular, investment-
specific technology shocks are important with a contribution of 0.25%. Risk premium and
income risk shocks explain another 0.33% of a 1.0% decline in output. Monetary policy
21
We report historical decompositions for output, consumption, investment, and government expenditures
from 1954 to 2015 in Appendix C.

27
 Figure 2: US inequality – data vs. model

10

30 Data
HANK
HANK* 5
20

10
0

0
Data
-5
HANK
-10
HANK*

-20 -10
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

(a) Top 10% income share (b) Top 10% wealth share
Notes: Data (crosses) correspond to log-deviations of the annual observations of the share of
pre-tax income and wealth held by the top 10% in each distribution in the US taken from the
World Inequality Database. HANK* (solid) [HANK (dashed)] corresponds to the smoothed
states of both implied by the estimated HANK model with inequality data [w/o inequality
data]. Shaded areas correspond to NBER-dated recessions.

shocks are estimated to be, on average, slightly more expansionary during recessions than
the Taylor rule implied rate would suggest – policy shocks contribute positively to output
growth in recessions.
This fact is even stronger for consumption. Similarly, investment-specific technology
shocks stabilize consumption in recessions as households invest less upon an adverse investment-
specific technology shock. In consequence, the decline in consumption during recessions is
to a larger extent driven by risk premia and income risk – 0.6% out of a 1% decline.

6 US Inequality
Now, we show that the estimated business cycle shocks can explain the movements of wealth
and income inequality in the US. Figure 2 plots the inequality data and the model implied
smoothed states for the estimation with and without inequality data. Both data series are
available on an annual basis almost throughout our whole sample period (1954-2014). The
top 10% wealth and income shares are both U-shaped and trough around 1980 in the data.
The model implied top 10% wealth and income shares match the data well. In the data,
the top 10% wealth share increases by 12 percentage points from 1980 to 2015, and the model
gets 50% of this increase. The top 10% (pre-tax) income share increases by 32 percentage
points over the same time period in the data, and the model predicts an increase by almost

28
40 percentage points.22 Business cycle shocks can move inequality along the lines of what
we observe in the data. This matching of the distributional data, on top of the “standard”
macroeconomic time series, does not change significantly what we infer about shocks and
frictions; see the previous section. Section 6.2 will provide a more detailed account of the
driving forces behind this. The business cycle analysis requires that both wage markups
and price markups be increasing in lockstep up until the mid 1970s. The decade after, both
markups fall, leading to the protracted boom of the 1980s. As a result of this comovement,
income inequality remains roughly constant and below its long-term average. The result is
a fall in wealth inequality. The increase in inequality over the most recent three decades is
estimated to be a result of rising price markups that are not accompanied by higher wage
markups this time. In addition, income risks, i.e., idiosyncratic productivity risks, have been
increasing over the last three decades.

6.1 Propagation of Inequality

Why is the model able to explain the slow-moving inequality dynamics? Our model implies
that business cycle shocks have very persistent effects on the wealth distribution, as Figure
3 shows, as an example, for markup and income risk shocks.23 The response to either shock
is the least persistent for income inequality, is more persistent for consumption inequality
and most persistent for wealth inequality. Consider, for example, a price-markup shock.
This drives up the income of entrepreneurs, the income richest households in our model.
However, because of sticky prices, the increase in inequality is staggered. Therefore, we
see initially a greater rise in consumption than in income inequality because entrepreneurs
foresee their future incomes increasing and dissave. Once markups reach their now increased
target, entrepreneurs save part of their higher income to smooth consumption. Consequently,
consumption inequality peaks later than income inequality, and the rise in markups slowly
translates into wealth inequality, which then peaks last. This makes it possible for transitory
business cycle shocks to explain persistent deviations in inequality.
Income risk shocks move the consumption Gini immediately and strongly. Poor house-
holds cut back their consumption to accumulate liquidity in response to the shock. As a
result, wealth becomes initially more equally distributed. Yet, as demand falls and markups
increase, income inequality goes up immediately, too. Later, when the increase in income risk
leads to more dispersed realizations of income, income inequality remains high and wealth
inequality overshoots its long-run level.
22
The fact that the model fits the actual movement of income inequality well reaffirms our modeling choices
regarding the distribution of income from various sources – in particular to attribute rents to the income

29
 Figure 3: Impulse responses of inequality

3.0 0.6 0.4

2.0 0.4 0.2

1.0 0.2 0.0

0.0 -0.2
0.0 0 10 20 30 40 0 10 20 30 40
0 10 20 30 40

(a) Top 10% income share (b) Consumption Gini (c) Top 10% wealth share
Notes: Response (in percent) of the top 10% pre-tax income share, the consumption Gini, and the
top 10% wealth share to a one standard deviation price-markup shock (solid, black) and income-
risk shock (dashed, red). The response of the Gini coefficient/top 10% shares is calculated by
including them as a generalized moment in the linearized model.

6.2 Historical Decompositions of US Inequality

To dig into the details of the evolution of inequality, Figure 4(a) plots the historical decom-
position of the top 10% income share. The decomposition of the level of income inequality
shows that medium-term trends of income inequality primarily result from markup shocks
and fluctuations in income risk. Looking at growth rates reveals that income risk is also an
important driver of income inequality at business cycle frequency, and in the Great Recession
in particular.24
With respect to particular historical episodes, our decomposition suggests the following.
Rising wage markups and low idiosyncratic productivity risks are mainly responsible for
the decrease in income inequality throughout the 1960s until the 1970s. The 1980s are
seen as a period of liberalization through the lens of our model (both in terms of output
cycles and inequality). Wage markups fell, which increased income inequality, but this was
partly offset by falling price markups. This picture changes throughout the 1990s but most
clearly from the early 2000s onward. Through the lens of our model, it is larger income
risks and sharply increasing price markups that best explain aggregate fluctuations and the
sharp rise in income inequality these years have witnessed. Interestingly and despite the
rich.
23
In Appendix D, we report the IRFs of wealth, income, and consumption inequality for all 10 shocks.
24
This finding is consistent with other papers that study the Great Recession in particular; see, e.g., Perri
and Steinberg (2012) or McKay (2017).

30
 Figure 4: Historical decompositions of US inequality

40

20 4

0
0

-2
-20

-4

-40 -6
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

(a) Top 10% income share (b) Top 10% wealth share

10

-5

-10

1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

(c) Gini coefficient of consumption

Notes: Historical decomposition of the log-deviations of the top 10% share in pre-tax income
(top left), top 10% wealth share (top right), and the Gini coefficient of consumption (bottom).
These inequality measures are treated as generalized moments that are included as controls into
the state-space representation of the model. Shaded areas correspond to NBER dated recessions.

use of completely different data sources, the historical decomposition thus is in line with the
evidence by De Loecker and Eeckhout (2020) on the evolution of markups in the US.25
For the evolution of wealth inequality other shocks are important as well. Figure 4(b)
shows the historical decomposition of the top 10% wealth share. Wealth inequality fell in
the first half of the sample and then increased. The pattern is similar in shape to income
inequality, but smoother. Yet, the drivers of wealth inequality are not the same as the drivers
of income inequality. The decomposition shows that up until the end of the 1970s, wage-
markup, investment-specific technology, and monetary shocks are the strongest downward
drivers of wealth inequality. From the 1980s on, it is then mainly shocks to investment-
25
There is a growing literature on the rise of markups; see, e.g., Karabarbounis and Neiman (2019), Barkai
(2019), Hall (2018), or Kehrig and Vincent (2018).

31
 Table 5: Contribution of shocks to US inequality 1980-2015

Shock Top 10% Income Top 10% Wealth Gini Consumption

TFP, Z -0.30 -1.14 -0.33

Inv.-spec. tech., Ψ 1.32 2.63 1.62
Price markup, µY 11.99 2.44 2.50
Wage markup, µW 8.83 0.90 3.46
Risk premium, A -0.51 0.06 1.72
Monetary policy, R 6.81 0.72 2.10
Structural deficit, G -0.84 -0.94 1.55
Tax level, τ 0.00 0.00 0.00
Tax progressivity, τ P 2.06 1.70 2.23
Income risk, σ 8.99 0.07 2.35
Sum of shocks 39.56 6.25 17.40
Notes: The table displays the contribution (in p.p.) of the various shocks to the
increase in the top 10% share of pre-tax income, top 10% share of wealth, and the
Gini coefficient of consumption from 1980 to 2015 based on our historical shock
decompositions.

specific technology and fiscal policy (deficits and tax progressivity) that drive up wealth
inequality. Investment-specific technology and fiscal deficits matter for wealth inequality
because they affect the return spread between the liquid and illiquid asset. Only since the
2000s have rising price markups become a strong positive contributor to wealth inequality.
Finally, Figure 4(c) plots the historical decomposition of the Gini coefficient of consump-
tion. Income risk is the most important driver of short-run fluctuations in consumption
inequality. The long-run trend in consumption inequality is primarily due to markups and
fiscal policy. The reason why income risk is an important driver of consumption inequality
lies in the portfolio choice problem of the households. In general, poor households react more
strongly to changes in income risk when rebalancing their portfolios (both in the data and
in the model; see Bayer et al., 2019). This means that when income risk goes up, the poor
more severely cut back consumption to acquire more liquid funds. Therefore, an increase in
income risk decreases the consumption of the poor more strongly than the consumption of
the wealthy.
Table 5 summarizes the driving forces behind the increase in all three inequality measures
from 1980 to 2015. The business cycle shocks in our model capture virtually all the observed

32
 Figure 5: Variance decompositions: Inequality
Cond.

Cond.
Uncond.

Uncond.
0 20 40 60 80 100 0 20 40 60 80 100

(a) Top 10% income share (b) Top 10% wealth share
Cond.
Uncond.

0 20 40 60 80 100

(c) Gini Consumption

Notes: Unconditional and conditional (4-quarter horizon) variance decompositions of the top
10% share of pre-tax income, top 10% share of wealth, and the Gini coefficient of consumption.

increase in income inequality and roughly half of the increase in wealth inequality. The
estimated model somewhat misses the trough in the 1980s and does not fully capture the
rising wealth inequality after the Great Recession. The likely main reason seems that our
model misses the full complexity of household portfolios and the late divergent returns on
houses and other forms of capital; see Kuhn et al. (2020). Figure 5 shows that, in general,
our findings from the historical decompositions also hold true for the average business cycle
in terms of variance decompositions.
These findings challenge the prevailing literature on the drivers of inequality that focuses
on long-run trends such as the rising skill premium or changes in the tax and transfer system
(see, e.g., Kaymak and Poschke (2016) or Hubmer et al. (2019)). We add to this literature
by showing that the business cycle has very persistent effects on inequality and can account
for up to 50% of the rise in US wealth inequality from 1980 to 2015. In the estimation of our
model, we allow for very persistent shocks to tax progressivity that capture the decline in
progressivity from the 1970s onward.26 We find that the decline of tax progressivity explains
1.7 p.p. of the increase in wealth inequality since the 1980s; see Table 5. This makes the
design of the tax system more important for the historical decomposition of wealth inequal-
ity than government spending or monetary shocks. Changes in tax progressivity are the
26
See Appendix C for the implied time series.

33
third most important driver of wealth inequality after price markups and investment-specific
technology, which account for 2.5 p.p. and 2.1 p.p respectively. One reason for the muted
effect on inequality is that lower progressivity leads to a higher capital stock through its in-
teraction with portfolio heterogeneity, because wealthy households have a higher propensity
to invest in capital; see Luetticke (2018). This reduces dividends and increases wages, which
mitigates the effect on inequality.

6.3 Inequality Counterfactuals

How important are the estimated policy coefficients for the evolution of inequality? To
understand the role of systematic business cycle policies in shaping inequality, we run a set
of counterfactual monetary and fiscal policy experiments based on the estimated model. The
results of these experiments are displayed in Figures 6 and 7. In detail, the figures display the
difference in the evolution of output, income inequality, wealth inequality and consumption
inequality between running the estimated shock sequence through our baseline estimate and
through the solution with the counterfactually set policy parameters.
First, we consider an experiment where the Fed reacts very aggressively to inflation; see
Figure 6. This creates large output losses after markup shocks, but stabilizes very effectively
after demand shocks. Given the series of shocks, output would have been lower in the 70s,
and income, wealth, and consumption inequality would have been substantially higher. This
reflects the fact that our model attributes a substantial fraction of the fluctuations of the 70s
to markup (cost-push) shocks. In the 1980s and especially the 1990s, the same policy would
have led to higher output, and lower inequality; however, because markups were falling and,
importantly, a substantial fraction of shocks during this time are demand shocks. For the
Great Recession, which we estimate to be followed by a persistent increase in the demand for
government bonds (risk premium shock) alongside an increase in markups, the more hawkish
monetary policy would have helped the recovery and lowered inequality. Interestingly, the
commitment to a more aggressive policy reaction requires a smaller interest rate movement
in response to the increase in the demand for bonds. Second, we consider a dovish policy
where we triple the monetary policy response to output fluctuations. This leads in general
to more stable markups and output at the expense of higher inflation volatility; see also
Gornemann et al. (2012). It is not fully the mirror image of the hawkish policy we looked at
before because this experiment changes the response to fluctuations in output, not inflation.
In fact, it generates output fluctuations in the 1990s, while it does little to change all series
in the 1970s. For the Great Recession, a more dovish policy stance would have led to an
earlier recovery and in particular lower income inequality. The effects on wealth inequality
are milder.

34
 Figure 6: Counterfactual evolution of output and income, wealth, and consumption inequality: Monetary policy

Aggregates
1 1

0.2

0 0

0.0

-1 -1

-0.2
Hawkish Monetary Policy
-2 Dovish Monetary Policy -2

-0.4
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

(a) Output (b) Nominal rate (quarterly) (c) Marginal costs

35

Inequality
6
2.0

0.6
4 1.5

0.4 1.0
2

0.2 0.5

0
0.0
0.0

-2 -0.5
-0.2
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

(d) Top 10% income share (e) Top 10% wealth share (f) Consumption Gini
Notes: The panels display the evolution of output, the nominal interest rate, and marginal costs as well as wealth, income, and consumption
inequality that the model would counterfactually predict had the government policies been different, feeding the smoothed sequence of
shocks (as in Figure 4) through the model. The lines represent the difference (in p.p.) in the evolution compared to feeding the same shocks
through the baseline model. The solid line corresponds to a setup where we double the inflation response θπ . The dashed line reflects the
counterfactual where we double the estimated response to output, θy . Shaded areas correspond to NBER-dated recessions.
 Figure 7: Counterfactual evolution of output and income, wealth, and consumption inequality: Fiscal policy

Aggregates
0.75
3
15

0.50
2
10

1 0.25

5
0
0.00
0
-1
-0.25
-2 -5

1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

Output Debt Liquidity premium (quarterly)

36

Inequality
1.00

1.0 2

0.75
0.5 1

0.50
0.0 0

0.25
-0.5 -1

-1.0 0.00 -2
Deficit Stimulus
Tax Stimulus
-1.5 -0.25 -3

1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

Top 10% income share Top 10% wealth share Consumption Gini
Notes: The panels display the evolution of output, government bonds, and the liquidity premium as well as, wealth, income, and consumption
inequality that the model would counterfactually predict had the government policies been different, feeding the smoothed sequence of shocks
(as in Figure 4) through the model. The lines represent the difference (in p.p.) in the evolution compared to feeding the same shocks through
the baseline model. The solid line corresponds to a setup where we allow for persistence deviations of government debt by setting γB = 0.1
and γB τ = 0.4. The dotted line reflects the counterfactual where we double the estimated tax response, γ τ . Shaded areas correspond to
Y
NBER-dated recessions.
 Finally, we consider alternative fiscal policy scenarios; see Figure 7. First, we assume
more aggressive deficit (spending) policies. In particular, we allow debt to increase more
persistently, by lowering γB and γBτ such that the average tax rate path is roughly kept as in
the baseline. Second, we consider a policy that adjusts taxes more heavily, leaving the overall
deficit and hence debt as in the baseline. That is, we consider a policy that lowers taxes
rather than raises government consumption when fighting a recession with a government
deficit.
The two alternative policies fare particularly differently in their response to the Great
Recession. In the first scenario, the more active deficit scenario, government debt rises
almost by another 50% and output is initially more stable after 2008 until the government
starts to raise taxes to bring back the government debt to its steady-state level. In fact, the
aggressive fiscal policy lifts the nominal rate (not displayed) by almost 1 percentage points
(annualized), and inflation up by 0.5 percentage points, because households are willing to
hold the extra liquidity only at higher returns. The equilibrium result is a substantially
lower liquidity premium. This, in turn, reduces wealth inequality substantially, but drives
up income and consumption inequality. The depressed liquidity premium means that there
is redistribution to wealth-poor households, which predominantly save in liquid assets at
the expense of wealth-rich households, which mostly hold illiquid assets. Cutting taxes
more aggressively during the Great Recession while holding the deficit constant would have
stabilized output and hence consumption inequality more strongly. However, this would have
increased the liquidity premium, putting further downward pressure on nominal rates, such
that our abstraction from the effective lower bound becomes even more binding.

7 Conclusion
How much does inequality matter for the business cycle and vice versa? To shed light on this
two-way relationship, this paper estimates a state-of-the-art New-Keynesian business cycle
model with household heterogeneity and portfolio choice on macro and micro data. We find
household income risk to be an important driver of output and consumption, in particular in
US recessions. Otherwise, we find that household heterogeneity and the inclusion of micro
data in the estimation do not materially alter the shocks and frictions in US business cycles.
However, we find that business cycles are important to understand the evolution of US
inequality. We show that business cycle shocks and policy responses can account for 50%
of the increase in US wealth inequality and virtually all of the increase in income inequality
since the 1980s. The reason behind this is that wealth (inequality) is a slowly moving variable
that accumulates past shocks. Our analysis suggests that price markups have substantially

37
increased over the last two decades. This has driven down output and has increased income,
consumption and wealth inequality. A more expansionary fiscal policy that would have
allowed government debt to increase substantially more after the Great Recession would
have had a positive impact on interest rates and thus helped the economy to escape the
effective lower bound earlier and boosted the recovery. At the same time, this evolution of
government debt would have eroded the return difference between illiquid and liquid assets,
helping in particular poor households to accumulate wealth, driving down wealth inequality.
These findings suggest that future research on inequality should take business cycles into
account. A synthesis of the previous literature that focuses on permanent changes, e.g. in
the tax and transfer system or the skill premium, with the forces that we highlight will be
an important area of research. Our findings further suggest exploring the role of shocks that
affect household insurance for the business cycle. Including a micro-foundation for income
risk, as, e.g., via search and matching, is of first order-importance to understand how the
business cycle and policies work differently by affecting income risk itself.

References
Ahn, S., Kaplan, G., Moll, B., Winberry, T., and Wolf, C. (2018). When inequality matters
for macro and macro matters for inequality. NBER Macroeconomics Annual, 32(1):1–75.

An, S. and Schorfheide, F. (2007). Bayesian analysis of DSGE models. Econometric Reviews,
26(2-4):113–172.

Auclert, A. (2019). Monetary policy and the redistribution channel. American Economic
Review, 109(6):2333–67.

Auclert, A., Bardóczy, B., Rognlie, M., and Straub, L. (2019). Using the sequence-space
Jacobian to solve and estimate heterogeneous-agent models. NBER Working Papers 26123.

Auclert, A., Rognlie, M., and Straub, L. (2018). The intertemporal Keynesian Cross. NBER
Working Papers 25020.

Auclert, A., Rognlie, M., and Straub, L. (2020). Micro jumps, macro humps: Monetary
policy and business cycles in an estimated HANK model. NBER Working Papers 26647.

Barkai, S. (2019). Declining labor and capital shares. Journal of Finance (forthcoming).

Bayer, C. and Luetticke, R. (2018). Solving heterogeneous agent models in discrete time
with many idiosyncratic states by perturbation methods. CEPR Discussion Papers 13071.

38
Bayer, C., Luetticke, R., Pham-Dao, L., and Tjaden, V. (2019). Precautionary savings,
illiquid assets, and the aggregate consequences of shocks to household income risk. Econo-
metrica, 87(1):255–290.

Benabou, R. (2002). Tax and education policy in a heterogeneous-agent economy: What

levels of redistribution maximize growth and efficiency? Econometrica, 70(2):481–517.

Boivin, J. and Giannoni, M. (2006). DSGE Models in a Data-Rich Environment. NBER

Working Papers 12772.

Born, B. and Pfeifer, J. (2014). Policy risk and the business cycle. Journal of Monetary
Economics, 68:68–85.

Bricker, J., Henriques, A., Krimmel, J., and Sabelhaus, J. (2016). Measuring income and
wealth at the top using administrative and survey data. Brookings Papers on Economic
Activity, 2016(1):261–331.

Broer, T., Hansen, N.-J. H., Krusell, P., and Öberg, E. (2019). The New Keynesian trans-
mission mechanism: A heterogenous-agent perspective. The Review of Economic Studies,
87(1):77–101.

Calvo, G. A. (1983). Staggered prices in a utility-maximizing framework. Journal of Mone-

tary Economics, 12(3):383–398.

Castaneda, A., Diaz-Gimenez, J., and Rios-Rull, J.-V. (1998). Exploring the income distri-
bution business cycle dynamics. Journal of Monetary Economics, 42(1):93 – 130.

Challe, E. and Ragot, X. (2015). Precautionary saving over the business cycle. The Economic
Journal, 126(590):135–164.

Chang, M., Chen, X., and Schorfheide, F. (2018). Heterogeneity and aggregate fluctuations.
Mimeo, University of Pennsylvania.

Chetty, R., Guren, A., Manoli, D., and Weber, A. (2011). Are micro and macro labor supply
elasticities consistent? A review of evidence on the intensive and extensive margins. The
American Economic Review, 101(3):471–475.

De Loecker, J. and Eeckhout, J. (2020). The rise of market power and the macroeconomic
implications. The Quarterly Journal of Economics (forthcoming).

Den Haan, W. J., Rendahl, P., and Riegler, M. (2017). Unemployment (fears) and deflation-
ary spirals. Journal of the European Economic Association, 16(5):1281–1349.

39
Durbin, J. and Koopman, S. J. (2012). Time Series Analysis by State Space Methods. Oxford
University Press.

Fernández-Villaverde, J. (2010). The econometrics of DSGE models. SERIEs, 1(1):3–49.

Gabaix, X., Lasry, J.-M., Lions, P.-L., and Moll, B. (2016). The dynamics of inequality.
Econometrica, 84(6):2071–2111.

Gelman, A. and Rubin, D. B. (1992). Inference from iterative simulation using multiple
sequences. Statistical Science, 7(4):457–472.

Gornemann, N., Kuester, K., and Nakajima, M. (2012). Monetary policy with heterogeneous
agents. Philadelphia Fed Working Paper 12-21.

Greenwood, J., Hercowitz, Z., and Huffman, G. W. (1988). Investment, capacity utilization,
and the real business cycle. American Economic Review, 78(3):402–417.

Guerrieri, V. and Lorenzoni, G. (2017). Credit crises, precautionary savings, and the liquidity
trap. The Quarterly Journal of Economics, 132(3):1427–1467.

Guvenen, F., Kaplan, G., and Song, J. (2014). How risky are recessions for top earners?
The American Economic Review, 104(5):148–153.

Hagedorn, M., Manovskii, I., and Mitman, K. (2018). Monetary policy in incomplete market
models: Theory and evidence. Mimeo, University of Pennsylvania.

Hagedorn, M., Manovskii, I., and Mitman, K. (2019). The fiscal multiplier. NBER Working
Papers 25571.

Hall, R. E. (2018). New evidence on the markup of prices over marginal costs and the role
of mega-firms in the US economy. NBER Working Papers 24574.

Heathcote, J., Perri, F., and Violante, G. L. (2010). Unequal we stand: An empirical analysis
of economic inequality in the United States, 1967–2006. Review of Economic Dynamics,
13(1):15–51.

Heathcote, J., Storesletten, K., and Violante, G. L. (2017). Optimal tax progressivity: An
analytical framework. The Quarterly Journal of Economics, 132(4):1693–1754.

Herbst, E. (2014). Using the “Chandrasekhar Recursions” for likelihood evaluation of DSGE
models. Computational Economics, 45(4):693–705.

40
Hubmer, J., Krusell, P., and Smith Jr, A. A. (2019). Sources of U.S. wealth inequality: Past,
present, and future. Mimeo, University of Pennsylvania.

Jaimovich, N. and Rebelo, S. (2009). Can news about the future drive the business cycle?
American Economic Review, 99(4):1097–1118.

Justiniano, A., Primiceri, G. E., and Tambalotti, A. (2010). Investment shocks and business
cycles. Journal of Monetary Economics, 57(2):132 – 145.

Justiniano, A., Primiceri, G. E., and Tambalotti, A. (2011). Investment shocks and the
relative price of investment. Review of Economic Dynamics, 14(1):101–121.

Kaplan, G., Moll, B., and Violante, G. L. (2018). Monetary policy according to HANK.
American Economic Review, 108(3):697–743.

Karabarbounis, L. and Neiman, B. (2019). Accounting for factorless income. NBER Macroe-
conomics Annual, 33(1):167–228.

Kaymak, B. and Poschke, M. (2016). The evolution of wealth inequality over half a century:
The role of taxes, transfers and technology. Journal of Monetary Economics, 77:1–25.

Kehrig, M. and Vincent, N. (2018). The micro-level anatomy of the labor share decline.
NBER Working Papers 25275.

King, R. G., Plosser, C. I., and Rebelo, S. T. (1988). Production, growth and business cycles:
I. The basic neoclassical model. Journal of Monetary Economics, 21(2):195 – 232.

Klein, P. (2000). Using the generalized Schur form to solve a multivariate linear rational
expectations model. Journal of Economic Dynamics and Control, 24(10):1405–1423.

Koopman, S. J. and Durbin, J. (2000). Fast filtering and smoothing for multivariate state
space models. Journal of Time Series Analysis, 21(3):281–296.

Kopczuk, W. (2015). What do we know about the evolution of top wealth shares in the
united states? Journal of Economic Perspectives, 29(1):47–66.

Kuhn, M., Schularick, M., and Steins, U. I. (2020). Income and wealth inequality in America,
1949-2013. Journal of Political Economy (forthcoming).

Liu, L. and Plagborg-Møller, M. (2019). Full-information estimation of heterogeneous agent

models using macro and micro data. Mimeo, Princeton University.

41
Luetticke, R. (2018). Transmission of monetary policy with heterogeneity in household
portfolios. CFM Discussion Paper Series, (CFM-DP2018-19).

McKay, A. (2017). Time-varying idiosyncratic risk and aggregate consumption dynamics.

Journal of Monetary Economics, 88:1–14.

McKay, A., Nakamura, E., and Steinsson, J. (2016). The power of forward guidance revisited.
The American Economic Review, 106(10):3133–3158.

McKay, A. and Reis, R. (2016). The role of automatic stabilizers in the US business cycle.
Econometrica, 84(1):141–194.

Perri, F. and Steinberg, J. (2012). Inequality and redistribution during the Great Recession.
Economic Policy Paper 12-1, Federal Reserve Bank of Minneapolis.

Piketty, T. and Saez, E. (2003). Income inequality in the United States, 1913–1998. The
Quarterly Journal of Economics, 118(1):1–41.

Ravn, M. O. and Sterk, V. (2017). Job uncertainty and deep recessions. Journal of Monetary
Economics, 90:125–141.

Reiter, M. (2009). Solving heterogeneous-agent models by projection and perturbation.

Journal of Economic Dynamics and Control, 33(3):649–665.

Saez, E. and Zucman, G. (2016). Wealth inequality in the United States since 1913: Evidence
from capitalized income tax data. The Quarterly Journal of Economics, 131(2):519–578.

Schmitt-Grohé, S. and Uribe, M. (2004). Solving dynamic general equilibrium models using
a second-order approximation to the policy function. Journal of Economic Dynamics and
Control, 28(4):755–775.

Schmitt-Grohé, S. and Uribe, M. (2012). What’s news in business cycles. Econometrica,

80(6):2733–2764.

Smets, F. and Wouters, R. (2007). Shocks and frictions in US business cycles: A Bayesian
DSGE approach. American Economic Review, 97(3):586–606.

Sterk, V. and Tenreyro, S. (2018). The transmission of monetary policy through redistribu-
tions and durable purchases. Journal of Monetary Economics, 99:124–137.

Storesletten, K., Telmer, C. I., and Yaron, A. (2004). Cyclical dynamics in idiosyncratic
labor market risk. Journal of Political Economy, 112(3):695–717.

42
Taylor, J. B. (1993). Discretion versus policy rules in practice. In Carnegie-Rochester
Conference Series on Public Policy, volume 39, pages 195–214. Elsevier.

Wong, A. (2019). Refinancing and the transmission of monetary policy to consumption.

Mimeo, Princeton University.

Woodford, M. (1995). Price-level determinacy without control of a monetary aggregate.

Carnegie-Rochester Conference Series on Public Policy, 43(Supplement C):1 – 46.

Wu, J. C. and Xia, F. D. (2016). Measuring the macroeconomic impact of monetary policy
at the zero lower bound. Journal of Money, Credit and Banking, 48(2-3):253–291.

43
A Data
A.1 Data for Calibration
Mean illiquid assets. Fixed assets (NIPA table 1.1) over quarterly GDP (excluding
net exports; see below), averaged over 1954-2015.
Mean liquidity. Gross federal debt held by the public as percent of GDP (FY-
PUGDA188S). Available from 1954-2015.
Fraction of borrowers. Taken from the Survey of Consumer Finances (1983-2013); see
Bayer et al. (2019) for more details.
Average top 10% share of wealth. Source is the World Inequality Database (1954-
2015).

A.2 Data for Estimation

The observation equation describes how the empirical times series are matched to the corre-
sponding model variables:
 
∆ log (Yt )
 

 ∆ log (Ct ) 


 ∆ log (It ) 



∆ log wtF
 
 
 

 ∆ log
 (T
t)


 

 log N̂t 

OBSt = 
  
b

 log R̂t+1


 

 log (π̂t ) 


 log (ŝt ) 

 τ P



 ∝
 × log τ̂ t 


ˆ wealth
 log p90p100

t


  income
 
ˆ t
log p90p100

where ∆ denotes the temporal difference operator and the hats above the variables denote
relative deviations from the steady state.
Unless otherwise noted, all series available at quarterly frequency from 1954Q3 to 2015Q4
from the St.Louis FED - FRED database (mnemonics in parentheses).
Output. Sum of gross private domestic investment (GPDI), personal consumption ex-
penditures for nondurable goods (PCND), durable goods (PCDG), and services (PCESV),

44
and government consumption expenditures and gross investment (GCE) divided by the GDP
deflator (GDPDEF) and the civilian noninstitutional population (CNP16OV).
Investment. Sum of gross private domestic investment (GPDI) and personal consump-
tion expenditures for durable goods (PCDG) divided by the GDP deflator (GDPDEF) and
the civilian noninstitutional population (CNP16OV).
Consumption. Sum of personal consumption expenditures for nondurable goods (PCND)
and services (PCESV) divided by the GDP deflator (GDPDEF) and the civilian noninstitu-
tional population (CNP16OV).
Federal tax receipts. Federal government current tax receipts (FEDT) divided by the
GDP deflator (GDPDEF) and the civilian noninstitutional population (CNP16OV).
Real wage. Hourly compensation in the nonfarm business sector (COMPNFB) divided
by the GDP deflator (GDPDEF).
Inflation. Computed as the log-difference of the GDP deflator (GDPDEF).
Nominal interest rate. Quarterly average of the effective federal funds rate (FED-
FUNDS). From 2009Q1 till 2015Q4 we use the Wu and Xia (2016) shadow federal funds
rate.
Hours worked. Nonfarm business hours worked (COMPNFB) divided by the civilian
noninstitutional population (CNP16OV).
Idiosyncratic income risk. Based on Bayer et al. (2019) and available from 1983Q1
till 2013Q1.
Tax progressivity. US Individual Income Tax: Tax Rates for Regular Tax: Highest
Bracket in Percent (IITTRHB). Available annually 1954 to 2015.
Wealth inequality. p90p100 of US net personal wealth from the World Inequality
Database. Available annually 1954 to 2014.
Income inequality. p90p100 of US pre-tax national income from the World Inequality
Database. Available annually 1954 to 2014.

45
B Inspecting Model Mechanisms
B.1 Impulse Response Functions of the Estimated HANK* Model

46
 Figure 8: IRFs to structural deficit and monetary policy shocks

struct. deficit shock output consumption investment

0.3 0.2 0.0 0.0
0.2 0.0 -0.2 -0.3
0.1 -0.2 -0.4 -0.6
0.0 -0.4 -0.6 -0.9
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

spending nominal rate(p.p) inflation(p.p) dividend

2 0.06 0.06 0.006
1 0.04 0.04 0.003
0 0.02 0.02 0.000
-1 0.00 0.00 -0.003
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

capital price wages hours capital

0.025 0.10 0.25 0.2
0.000 0.05 0.00 0.0
-0.025 0.00 -0.25 -0.2
-0.050 -0.05 -0.50 -0.4
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

bonds utilization marginal costs liquidity premium

2 0.2 0.10 0.025
1 0.1 0.05 0.000
0 0.0 0.00 -0.025
-1 -0.1 -0.05 -0.050
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

mon. policy shock output consumption investment

0.4 0.3 0.2 1
0.2 0.0 0.0 0
0.0 -0.3 -0.2 -1
-0.2 -0.6 -0.4 -2
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

spending nominal rate(p.p) inflation(p.p) dividend

2 0.2 0.1 0.025
1 0.1 0.0 0.000
0 0.0 -0.1 -0.025
-1 -0.1 -0.2
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 -0.050 0 10 20 30 40

capital price wages hours capital

0.2 0.2 0.3 0.025
0.0 0.0 0.0 0.000
-0.2 -0.2 -0.3 -0.025
-0.4 -0.4 -0.6 -0.050
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

bonds utilization marginal costs liquidity premium

2 0.2 0.25 0.25
1 0.0 0.00 0.00
0 -0.2 -0.25 -0.25
-1 -0.4 -0.50 -0.50
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

Notes: Top: IRF to a structural deficit shock. Bottom: IRF to a monetary policy
shock.

47
 Figure 9: IRFs to markup shocks

price markup output consumption investment

2 0.0 0.2 2
1 -0.2 0.0 0
0 -0.4 -0.2 -2
-1 -0.6 -0.4 -4
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

spending nominal rate(p.p) inflation(p.p) dividend

2 0.10 0.2 0.025
1 0.05 0.1 0.000
0 0.00 0.0 -0.025
-1 -0.05 -0.1 -0.050
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

capital price wages hours capital

0.2 0.0 0.0 0.1
0.0 -0.5 -0.2 0.0
-0.2 -1.0 -0.4 -0.1
-0.4 -1.5 -0.6 -0.2
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

bonds utilization marginal costs liquidity premium

2 0.5 1 0.05
1 0.0 0 0.00
0 -0.5 -1 -0.05
-1 -1.0 -2 -0.10
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

wage markup output consumption investment

6 0.0 0.0 0
3 -0.5 -0.5 -1
0 -1.0 -1.0 -2
-3 -1.5 -1.5 -3
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

spending nominal rate(p.p) inflation(p.p) dividend

1 0.2 0.2 0.025
0 0.1 0.1 0.000
-1 0.0 0.0 -0.025
-2 -0.1 -0.1
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 -0.050 0 10 20 30 40

capital price wages hours capital

0.05 1.0 0.0 0.2
0.00 0.5 -0.5 0.0
-0.05 0.0 -1.0 -0.2
-0.10 -0.5 -1.5 -0.4
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

bonds utilization marginal costs liquidity premium

4 0.2 0.50 0.2
2 0.0 0.25 0.1
0 -0.2 0.00 0.0
-2 -0.4 -0.25 -0.1
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

Notes: Top: IRF to a price-markup shock. Bottom: IRF to a wage-markup shock.

48
 Figure 10: IRFs to technology shocks

tfp output consumption investment

0.7 1.0 0.7 2.5
0.6 0.8 0.6 2.0
0.5 0.6 0.5 1.5
0.4 0.4 0.4 1.0
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

spending nominal rate(p.p) inflation(p.p) dividend

1.5 0.00 0.00 0.015
1.0 -0.05 -0.05 0.010
0.5 -0.10 -0.10 0.005
0.0 -0.15 -0.15 0.000
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

capital price wages hours capital

0.2 0.75 0.4 1.0
0.1 0.50 0.2 0.5
0.0 0.25 0.0 0.0
-0.1 0.00 -0.2 -0.5
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

bonds utilization marginal costs liquidity premium

1 0.4 0.25 0.05
0 0.2 0.00 0.00
-1 0.0 -0.25 -0.05
-2 -0.2 -0.50 -0.10
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

inv.-spec. tech. output consumption investment

3 0.8 1.0 6
2 0.7 0.5 4
1 0.6 0.0 2
0 0.5 -0.5 0
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

spending nominal rate(p.p) inflation(p.p) dividend

2 0.3 0.2 0.050
0 0.2 0.1 0.025
-2 0.1 0.0 0.000
-4 0.0 -0.1 -0.025
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

capital price wages hours capital

0 0.8 0.6 2
-1 0.6 0.4 1
-2 0.4 0.2 0
-3 0.2 0.0 -1
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

bonds utilization marginal costs liquidity premium

2 2 0.4 1
0 1 0.2 0
-2 0 0.0 -1
-4 -1 -0.2 -2
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

Notes: Top: IRF to a TFP shock. Bottom: IRF to an MEI shock.

49
 Figure 11: IRFs to risk premium and income risk shocks

risk premium output consumption investment

0.20 0.0 0.0 0.0
0.15 -0.2 -0.2 -0.5
0.10 -0.4 -0.4 -1.0
0.05 -0.6 -0.6 -1.5
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

spending nominal rate(p.p) inflation(p.p) dividend

3 0.0 0.0 0.025
2 -0.1 -0.1 0.000
1 -0.2 -0.2 -0.025
0 -0.3 -0.3
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 -0.050 0 10 20 30 40

capital price wages hours capital

0.2 0.2 0.0 0.1
0.0 0.0 -0.2 0.0
-0.2 -0.2 -0.4 -0.1
-0.4 -0.4 -0.6 -0.2
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

bonds utilization marginal costs liquidity premium

2 0.2 0.2 0.25
1 0.0 0.0 0.00
0 -0.2 -0.2 -0.25
-1 -0.4 -0.4 -0.50
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

income risk output consumption investment

100 0.1 0.2 0.2
50 0.0 0.0 0.0
0 -0.1 -0.2 -0.2
-50 -0.2 -0.4 -0.4
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

spending nominal rate(p.p) inflation(p.p) dividend

1.0 0.025 0.025 0.002
0.5 0.000 0.000 0.000
0.0 -0.025 -0.025 -0.002
-0.5 -0.050 -0.050 -0.004
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

capital price wages hours capital

0.025 0.1 0.1 0.050
0.000 0.0 0.0 0.025
-0.025 -0.1 -0.1 0.000
-0.050 -0.2 -0.2 -0.025
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

bonds utilization marginal costs liquidity premium

0.4 0.1 0.1 0.1
0.2 0.0 0.0 0.0
0.0 -0.1 -0.1 -0.1
-0.2 -0.2 -0.2 -0.2
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

Notes: Top: IRF to a risk premium shock. Bottom: IRF to an income risk shock.

50
 Figure 12: IRFs to tax shocks

tax level output consumption investment

0.050 0.025
0.2 0.025
0.1 0.025 0.000 0.000
0.0 0.000 -0.025 -0.025
-0.1 -0.025 0 10 20 30 40 -0.050
0 10 20 30 40 -0.050 0 10 20 30 40 0 10 20 30 40

spending nominal rate(p.p) inflation(p.p) dividend

0.06 0.004 0.004 0.00050
0.03 0.002 0.002 0.00025
0.00 0.000 0.000 0.00000
-0.03 -0.002 -0.002 -0.00025
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

capital price wages hours capital

0.050 0.050
0.001 0.002
0.000 0.025 0.025 0.000
-0.001 0.000 0.000 -0.002
-0.002 -0.025 0 10 20 30 40 -0.025 0 10 20 30 40 -0.004
0 10 20 30 40 0 10 20 30 40

bonds utilization marginal costs liquidity premium

0.050 0.050
0.025 0.006
0.000 0.025 0.025 0.003
-0.025 0.000 0.000 0.000
-0.050 -0.025 0 10 20 30 40 -0.025 0 10 20 30 40 -0.003
0 10 20 30 40 0 10 20 30 40

tax prog. output consumption investment

2.5 0.1 0.2 0.0
2.0 0.0 0.1 -0.2
1.5 -0.1 0.0 -0.4
1.0 -0.2 -0.1 -0.6
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

spending nominal rate(p.p) inflation(p.p) dividend

0.0 0.06 0.050 0.002
-0.1 0.04 0.025 0.000
-0.2 0.02 0.000 -0.002
-0.3 0.00 -0.025 -0.004
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

capital price wages hours capital

0.025 0.10 0.1 0.2
0.000 0.05 0.0 0.0
-0.025 0.00 -0.1 -0.2
-0.05 -0.2 -0.4
-0.050 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

bonds utilization marginal costs liquidity premium

0.4 0.05 0.10 0.050
0.2 0.00 0.05 0.025
0.0 -0.05 0.00 0.000
-0.2 -0.10 -0.05 -0.025
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

Notes: Top: IRF to tax level shock. Bottom: IRF to a tax progessivity shock.

51
 Figure 13: Historical decompositions: Taxes and income risk

30 20

20
10

10
0

0
-10

-10
-20

-20
-30
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

(a) Average tax rate (b) Tax progressivity

300

200

100

-100

-200

-300
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

(c) Income Risk

Notes: Historical decompositions of the log-deviations of the average tax rate (top left), tax
progressivity (top right), and income risk (bottom left). Shaded areas correspond to NBER-
dated recessions.

C Further Historical Decompositions

Figure 13 shows the historical decomposition of taxes and income risk. The historical decom-
position shows that income risk is mostly driven by exogenous shocks and not endogenous
feedback. Figure 14 shows the historical decomposition of the growth rates of output, con-
sumption, investment, and government spending for the HANK* model (with inequality
data). Figure 15 shows the historical decomposition of the log-level of output, consumption,
investment, and government spending for the HANK* model. Figure 16 shows the historical
decomposition of the nominal rate, inflation, marginal costs, and government debt.

52
Figure 14: Historical decompositions: Output, consumption, investment and government
spending (growth rates)

2
2

0
0

-2 -2

-4
-4

1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

(a) Output growth (b) Consumption growth

15
10
10

5
0
0

-5
-10
-10

-15
-20

1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

(c) Investment growth (d) Government spending growth

Notes: Historical decompositions of the log-deviations of output growth (top left), consumption
growth (top right), investment growth (bottom left), and government spending growth (bottom
right). Shaded areas correspond to NBER-dated recessions.

53
Figure 15: Historical decompositions: Output, consumption, investment and government
spending (log levels)

20

10
10

0
0

-10 -10

-20 -20
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

(a) Output (b) Consumption

40

40
20

20
0

0
-20

-20
-40

1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

(c) Investment (d) Government spending

Notes: Historical decompositions of the log-deviations of output (top left), consumption (top
right), investment (bottom left), and government spending (bottom right). Shaded areas cor-
respond to NBER-dated recessions.

54
Figure 16: Historical decompositions: Nominal rate, inflation, marginal costs and govern-
ment debt (log levels)

2 2

0
0

-2 -1

-2

-4
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

(a) Nominal rate (quarterly) (b) Inflation

75

5
50

0 25

0
-5

-25
-10

1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

(c) Marginal costs (d) Government debt

Notes: Historical decompositions of the log-deviations of the nominal bond rate (top left),
inflation (top right), marginal costs (bottom left), and government debt (bottom right). Shaded
areas correspond to NBER-dated recessions.

55
D Further Results on Inequality Dynamics
Figure 17 presents the estimated impulse responses of inequality on all 10 shocks. The
general picture is that income inequality shows the most transitory movements. Wealth
inequality moves most persistently after all shocks because it is driven by accumulation
decisions. Consumption, driven by both income and wealth, shows both short- and long-run
dynamics.
Monetary policy and risk premium shocks both increase inequalities through a rise in
price markups that benefits mostly the (wealthy) entrepreneurs. The effects on income and
consumption inequality are rather transitory, while wealth inequality increases persistently
(more so for the risk premium shock).
Turning to fiscal policy, increasing the structural deficit lowers income and wealth in-
equality but increases consumption inequality. On the tax side, a shock to the average tax
rate has only very small overall effects on inequality, but tends to lower income inequality
(at least in the very short run) but increases consumption and wealth inequality marginally.
A shock to the progressiveness of the tax schedule, on the other hand, causes persistently
lower income, consumption, and wealth inequality.
Price and wage markup shocks both have persistent effects on income, consumption, and
wealth inequality; however, they differ markedly in the sign of their effects. While price
markups increase inequality because they raise profits that go to entrepreneurs, a rise in the
target wage markup lowers inequalities by increasing wage compression.
Higher income risk leads to a quick increase in income and consumption inequality as
poorer households over-proportionally increase their liquid asset holdings for insurance pur-
poses. Wealth inequality initially falls because poor households react strongly by accumulat-
ing extra, mostly liquid, assets. After about 2.5 years, wealth inequality increases persistently
when higher income risks have been realized and have led to more dispersed incomes.
The response of wealth inequality to TFP and investment-specific technology shocks looks
very similar. In both cases, wealth inequality initially increases but eventually declines,
and persistently so, after about 5 years. While both shocks lead to a persistent decline
in consumption inequality, the investment-specific technology shock initially raises it for
the first year or so. Both shocks also raise income inequality, the TFP shock by raising
productivity and therefore the returns to capital held by the wealthy households, and the
investment-specific technology shock by making investment in capital more efficient, which
benefits the wealthy as well.

56
 Figure 17: Impulse responses of inequality

Monetary shock Structural deficit shock

1.0 Top 10 Income 0.050 Top 10 Wealth Gini
0.15 Consumption 0.1 Top 10 Income 0.1 Top 10 Wealth 0.3Gini Consumption

0.5 0.025 0.10 0.0 0.0 0.2

0.0 0.000 0.05 -0.1 -0.1 0.1

-0.5 -0.025 0.00 -0.2 -0.2 0.0

0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

Average tax shock Tax progressivity shock

0.025Top 10 Income 0.002 Top 10 Wealth Gini
0.004 Consumption -0.05 Top 10 Income 0.05 Top 10 Wealth Gini
-0.050 Consumption

0.000 0.001 0.002 -0.10 0.00 -0.075

-0.025 0.000 0.000 -0.15 -0.05 -0.100

-0.050 -0.001 -0.002 -0.20 -0.10 -0.125

0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

TFP shock MEI shock

2 Top 10 Income 0.2 Top 10 Wealth 0.0Gini Consumption 0.5 Top 10 Income 0.2 Top 10 Wealth 0.2Gini Consumption

1 0.1 -0.2 0.0 0.1 0.0

0 0.0 -0.4 -0.5 0.0 -0.2

-1 -0.1 -0.6 -1.0 -0.1 -0.4

0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

Price markup shock Income risk shock

4 Top 10 Income 0.4 Top 10 Wealth 0.20Gini Consumption 3 Top 10 Income 0.025 Top 10 Wealth 0.6Gini Consumption

2 0.2 0.15 2 0.000 0.4

0 0.0 0.10 1 -0.025 0.2

-2 -0.2 0.05 0 -0.050 0.0

0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

Wage markup shock Risk premium shock

0 Top 10 Income 0.2 Top 10 Wealth 0.0Gini Consumption 1.0 Top 10 Income 0.025 Top 10 Wealth Gini
0.3 Consumption

-1 0.0 -0.2 0.5 0.000 0.2

-2 -0.2 -0.4 0.0 -0.025 0.1

-3 -0.4 -0.6 -0.5 -0.050 0.0

0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40

Notes: The figures display the impulse responses of income, consumption, and wealth inequality in
response to the shocks labeled above. Parameter estimates from HANK*. See main text for further
details.

57
 Figure 18: US inequality – data vs. model

10

30 Data
HANK
HANK* 5
20

10 0

0
Data
-5
HANK
-10 HANK*

-10
-20
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

(a) Top-10% income share (b) Top-10% wealth share

Notes: Data (crosses) corresponds to log-deviations of the annual observations of the share
of pre-tax income and wealth held by the top 10% in each distribution in the US taken from
the World Inequality Database. HANK* (solid) [HANK (dashed)] corresponds to the smoothed
states of both implied by the estimated HANK model with inequality data [w/o inequality data
and income risk shocks]. Shaded areas correspond to NBER dated recessions.

E Estimation Without Income Risk Data

We re-estimate the model without movements in income risk as this measure already contains
some cross-sectional information about income inequality. For this purpose, we do not allow
for shocks to income risk, i.e., we keep the variance of idiosyncratic income risk at its steady-
state value, and drop the time series for income risk from Bayer et al. (2019) from the
estimation. Figure 18 shows that the model without movements in income risk (denoted by
HANK) still implies a U-shaped evolution of income and wealth inequality. The fit is less
good then for the model with all inequality data (HANK*). More persistent fluctuations in
markups partly compensate for the missing movements in income risk.

58
F MCMC Diagnostics

Table 6: Gelman and Rubin (1992) convergence diagnostics

HANK HANK*
Parameter PSRF 97.5% PSRF 97.5%
δs 1.042 1.107 1.011 1.030
φ 1.022 1.055 1.011 1.028
κ 1.004 1.011 1.007 1.018
κw 1.005 1.010 1.007 1.018
ρA 1.002 1.004 1.003 1.007
σA 1.003 1.006 1.001 1.003
ρZ 1.003 1.006 1.002 1.005
σZ 1.001 1.002 1.000 1.000
ρΨ 1.001 1.003 1.004 1.010
σΨ 1.013 1.033 1.013 1.035
ρµ 1.002 1.005 1.001 1.002
σµ 1.004 1.010 1.003 1.008
ρµw 1.004 1.009 1.005 1.013
σµw 1.002 1.005 1.011 1.020
ρs 1.023 1.062 1.007 1.008
σs 1.006 1.017 1.002 1.004
ρR 1.011 1.029 1.002 1.004
σR 1.002 1.005 1.001 1.003
θπ 1.026 1.070 1.019 1.049
θy 1.003 1.008 1.002 1.003
γB 1.011 1.029 1.025 1.055
σG 1.021 1.057 1.016 1.039
γπ 1.001 1.002 1.009 1.019
γY 1.034 1.087 1.005 1.012
ρτ 1.032 1.081 1.013 1.032
στ 1.276 1.895 1.428 2.980
γBτ 1.013 1.032 1.039 1.090
γYτ 1.021 1.058 1.035 1.080
ρP 1.000 1.001 1.001 1.003
σP 1.010 1.024 1.019 1.050
ΣN 1.041 1.103 1.032 1.074
∝τ 1.010 1.025 1.015 1.040
ρG 1.002 1.004 1.012 1.024
σsme 1.013 1.025 1.026 1.046
σTme 1.003 1.007 1.007 1.020
στme
p 1.317 2.149 1.121 1.255
me
σW I – – 1.001 1.003
me
σII – – 1.001 1.003
Note: Gelman and Rubin (1992) potential scale reduction factor (PSRF) and 97.5% quantile.

59