Faculty of Economics

CAMBRIDGE WORKING PAPERS IN ECONOMICS

JANEWAY INSTITUTE WORKING PAPERS

How do transfers and universal basic income

impact the labor market and inequality?
Christopher Marcelo R.
Rauh Santos
University of INSPER
Cambridge

Abstract
This paper studies the impact of existing and universal transfer programs on vacancy creation, wages, and
welfare using a search-and-matching model with heterogeneous agents and on-the-job human capital
accumulation. We calibrate the general equilibrium model to match key moments concerning
unemployment, wage and wealth distributions, as well as the distribution of EITC and transfers. In addition,
unemployment insurance benefits are related to pre-unemployment earnings and subject to exhaustion,
after which agents can only rely on transfers and savings. First, we show that existing transfers hamper
economic activity but provide sizeable welfare gains. Next, we show that a universal basic income of nearly
$12,500 to each household per year, which replaces all existing transfer programs and unemployment
benefits, can lead to small aggregate welfare gains. These welfare gains mostly accrue to less skilled
individuals despite their sizable fall in wages, and the overall rise in skill premia and wage inequality. Albeit
the extra burden of higher taxes to finance UBI, we show that the increased action in hiring is a key channel
though which outcomes for low education groups improve with the reform. However, if we keep the UI
benefits in place, the positive effects on job creation vanish and UBI does not improve upon the current
system.

Reference Details
2208 Cambridge Working Papers in Economics
2022/05 Janeway Institute Working Paper Series

Published 31 January 2022

Key Words Transfer programs, EITC, Means-tested transfers, Welfare programs, Labor supply, On-the-
job human capital accumulation, Life cycle, Inequality, Universal basic income, UBI,
Unemployment, General equilibrium

Websites www.econ.cam.ac.uk/cwpe
www.janeway.econ.cam.ac.uk/working-papers
 How do transfers and universal basic income
impact the labor market and inequality?
Christopher Rauh* Marcelo R. Santos*
University of Cambridge INSPER

January 29, 2022

Abstract

This paper studies the impact of existing and universal transfer programs on va-
cancy creation, wages, and welfare using a search-and-matching model with hetero-
geneous agents and on-the-job human capital accumulation. We calibrate the gen-
eral equilibrium model to match key moments concerning unemployment, wage and
wealth distributions, as well as the distribution of EITC and transfers. In addition, un-
employment insurance benefits are related to pre-unemployment earnings and subject
to exhaustion, after which agents can only rely on transfers and savings. First, we show
that existing transfers hamper economic activity but provide sizeable welfare gains.
Next, we show that a universal basic income of nearly $12,500 to each household per
year, which replaces all existing transfer programs and unemployment benefits, can
lead to small aggregate welfare gains. These welfare gains mostly accrue to less skilled
individuals despite their sizable fall in wages, and the overall rise in skill premia and
wage inequality. Albeit the extra burden of higher taxes to finance UBI, we show that
the increased action in hiring is a key channel though which outcomes for low educa-
tion groups improve with the reform. However, if we keep the UI benefits in place,
the positive effects on job creation vanish and UBI does not improve upon the current
system.

Keywords: Transfer programs; EITC; Means-tested transfers; Welfare programs; Labor sup-
ply; On-the-job human capital accumulation; Life cycle; Inequality; Universal basic income;
UBI; Unemployment; General equilibrium.

* E-mail: Rauh: [email protected]. Santos: [email protected]. We thank Rui Castro, Nezih Guner,
Baris Kaymak, and Elena Pastorino and the participants of the LuBraMacro workshop in Lisbon for their valu-
able comments and suggestions. Part of this research was conducted when Santos was visiting the CIREQ.
All errors are ours.

1
1 Introduction
In 2017 the United States spent nearly 2 trillion USD on means-tested transfer programs
(Salazar et al. 2017).1 Excluding health-related transfers, Guner et al. (2021) estimates that
the total expenditures in transfers to the working-age population at all levels amounted to
nearly 2.3% of GDP. As expected, most of these transfers go to poor households. Guner
et al. (2022) find that more than 50% of households in some point in a given year. Individu-
als without any income on average receive transfers worth more than $15,000 a year. More-
over, households with positive income receive considerable transfers, as even a household
with mean income receives nearly $1,000 a year. Given these magnitudes, transfer pro-
grams are likely to have first-order effects on labor supply, inequality, and savings.
Conditional transfer programs seek to restrict payments to those in need. Since it is
hard to observe individuals’ earnings capacity, targeting is usually based on income. This
feature imposes an implicit tax on earned income, and thus a key concern about condi-
tional transfers is the potential disincentive to (search for) work, especially at the bottom of
the earnings distribution.2 In light of the convoluted transfer programs and inefficiencies
accompanying eligibility checks, the concept of universal basic income has gained trac-
tion amongst certain politicians and interest groups.3 The underlying idea is that each
person receives the same pre-specified amount irrespective of a household’s or aggregate
economic circumstances.
While the effect of transfers on labor supply has been studied extensively, much less is
known regarding the labor-market equilibrium impact of these policies, particularly their
effects on vacancy creation, wages, and welfare. To fill this gap, we study the effects of
means-tested and universal transfer programs on equilibrium labor market allocations and
workers’ welfare, using a model that combines three key ingredients.
First, we consider a search-and-matching model along the lines of a Diamond-Mortensen-
Pissarides’ (D-M-P) model featuring endogenous search intensity.4 By doing so, we can
study how transfers influence vacancy creation and wage bargaining between firms and
workers. In this environment, employment opportunities are endogenous. They depend
on how hard individuals search for a job and the aggregate labor market conditions that
determine how easy it is to locate jobs per unit of search effort.
Second, we depart from the standard D-M-P setting with risk neutrality by combining
1
The US has seven major means-tested transfer programs: Temporary Assistance for Needy Families
(TANF - formerly AFDC), Supplemental Nutrition Assistance Program (SNAP - formerly foodstamps), Sup-
plemental Nutrition Program for Women, Infants, and Children (WIC), Supplemental Security Income (SSI),
Medicaid, housing assistance and the Earned Income Tax Credit (EITC).
2
See Marinescu (2018) for an overview.
3
For instance, Andrew Yang, a candidate in the 2020 Democratic Party presidential primaries, advocates
for the “freedom dividend” of $1,000 per month for all American adults. In our analysis, we do not factor in
the potential savings that could be made by replacing the administrative apparatus.
4
Diamond (1982), Mortensen (1982) and Pissarides (1985)

2
the search-and-matching framework with incomplete markets as in Krusell et al. (2010)
and introduce endogenous job separations as in Bils et al. (2011). Our motivation for do-
ing this is to study the welfare effects of transfers when workers care about consumption
smoothing. In addition, by allowing for diminishing marginal utility in consumption and
imperfect insurance, wealth affects workers’ reservation wages and thus the cross-sectional
distributions of wealth play a critical role in determining the aggregate labor response to
changes in the transfer system.
Third, working and earning a wage does not only have implications for the cross-
sectional distribution of income and wealth, but also for life-cycle profiles through human-
capital accumulation on the job. In particular, labor productivity is determined by two
components. First, agents are ex-ante heterogeneous concerning their ability, which can be
interpreted as pre-market skills such as innate ability or obtained through education. The
second component reflects human capital or experience, and accumulates in a learning-
by-doing fashion. In particular, we build on the work of Ljungqvist and Sargent (1998) in
assuming that worker’s productivity changes over time according to laws of motion that
depend on whether the worker is employed or not. Employed agents can experience pro-
ductivity increases, while the careers of idle workers stagnate.
We combine these key trade-offs by enhancing a Bewley-Huggett-Aiyagari incomplete-
markets model (Bewley 1986, Huggett 1993, Aiyagari 1994) with labor-market frictions.
The lifespan is uncertain and agents survive from one period to the next with a given prob-
ability. This stochastic OLG environment allows the model to accommodate the actual
life-cycle earnings growth in a very tractable way. Individuals may resort to self-insurance
to protect themselves against the uncertainty in labor income and thus savings will be, to
some extent, motivated by precautionary reasons. As in the standard D-M-P model, firms
enter by posting vacancies and match with workers bilaterally, with match probabilities
given by an aggregate matching function. In addition, firms can direct their search by post-
ing vacancies for agents of a given level of education.
We use microdata for the US to inform the model and proceed with a quantitative anal-
ysis. The model matches key facts concerning unemployment, and the wage and wealth
distributions, as well as the distribution of EITC and transfers. Unemployment insurance
(UI) benefits are related to pre-unemployment earnings and subject to exhaustion, after
which agents rely on transfers and their own savings. The calibrated model allows us to
conduct several counterfactual experiments to shed light on the impacts of existing and
new transfer regimes on labor market outcomes, inequality, and welfare.
We find that removing transfers stimulates economic activity through higher precau-
tionary savings, greater incentives to work and lower taxation, leading to an increase in
output of 5.2% and a reduction of the unemployment rate by more than a quarter. How-
ever, this boost to the economy would be accompanied by a large increase in consump-
tion inequality of 21%, and a 24% reduction in welfare as measured by the consumption

3
equivalent variation (CEV), which is largely felt by the unemployed (-46%) but also by the
employed (-9%). The main reasons that the employed also suffer is that transfers are not
restricted to the unemployed and the employed might also become unemployed in the fu-
ture. We find that removing the EITC has a mild negative impact on the economy with
the unemployment rate going up by 2.7% and output falling by 1.2%. We also find that
eliminating the EITC leads to a benign increase in income and consumption inequality and
an aggregate welfare loss of 1.9%.
So far there is only limited evidence on unconditional basic income, mostly from small
pilot schemes.5 The machinery of our calibrated model provides a laboratory for an in-
vestigation into the potential impact of removing all existing forms of public insurance
and transfers and replacing them with universal basic income (UBI), a lump sum transfer.
We consider two sets of counterfactual simulations in which each person receives a pre-
determined amount, irrespective of their income and asset levels or employment status. In
the first set, we remove transfers and the EITC. In the second set, we additionally remove
unemployment benefits. Within each case, we consider four levels of universal basic in-
come: 10%, 15%, 20%, and 25% of the benchmark average income. The costs of the scheme
are financed by adjusting the level of labor income taxation.
We find that in the steady state, a low level of UBI, i.e. 10% of average benchmark
income, leads to an even larger boost to the economy in terms of aggregate output and
reduced unemployment than removing transfers. However, this boost comes at a sizeable
welfare loss of 9% overall, albeit 41% of households still being better off than in the bench-
mark. As UBI becomes more generous, the aggregate economic impact dampens while
overall welfare increases. We find that welfare gains peak at 0.7% when a UBI of 20% of
average income – or nearly $12,500 to each agent per year – substitutes for all programs,
with nearly 75% of the population preferring it over the benchmark. Under this scenario,
even though aggregate capital falls by 1.9%, aggregate output is 0.5% larger than in the
baseline due to lower unemployment and higher on-the-job human capital accumulation,
which accounts for nearly one-fourth of the 1.6% increase in aggregate labor.
In addition, average wages fall but skill premia rise, leading to an increase in wage in-
equality. This is because a UBI of 20% provides less than what unskilled agents receive
from the combination of transfers, EITC, and UI, while the opposite happens for the skilled
ones and thus the fall in wages is larger for low educated groups. On the other hand, since
unemployment benefits are related to earnings, the removal of UI weakens workers’ hands
in bargaining, especially for less-educated agents. The changes in wages by level of edu-
cation go hand in hand with the changes observed in unemployment, with low education
groups experiencing a much larger drop compared to the highly educated ones. This ex-
plains our findings that those who have not completed college prefer higher levels of UBI,
5
See Hoynes and Rothstein (2019) for an overview.

4
in contrast to those who have completed college education.
The positive effects of UBI on vacancy creation are due to its universal nature, which
reduces the disincentives to work since employed and unemployed agents receive the same
amount. We show that the expected benefit of creating a vacancy increases for all education
groups and is larger for the unskilled. The reason for this is twofold. On the one hand, the
value of a filled vacancy is larger for poor agents compared to the benchmark. On the other
hand, we show that there is also a composition effect due to an increase in human capital
accumulation. In particular, there is an increase in the mass of workers in the upper end
of the distribution of human capital where the value of a job is much higher. Intuitively,
hiring workers is an investment activity in which costs are paid up front and benefits accrue
gradually as agents become more productive when employed. Better incentives to work
induce higher average tenure on the job and thus greater (expected) returns to job creation.
We conduct a decomposition exercise to quantify the importance of the different ingre-
dients of our model in shaping the UBI effects. Since a UBI of 20% is very costly, 8.5% of
output – as opposed to nearly 2.5% for the baseline transfers – tax distortions are likely
to have first order effects. We show that keeping the level of labor income taxation at the
benchmark would boost consumption by 19.5%, while the overall welfare gain from UBI
would increase by 20%.6 In addition, average wages would go up by nearly 7% instead
of falling 2.8% when taxes are allowed to change. Assuming that labor market tightness
is kept unchanged, a UBI of 20% that replaces all programs would lead to a drop in the
job finding rate of 4.5 percentage points instead of an increase of 17 percentage points in
the main experiment. The unemployment rate would increase by 0.4 percentage points,
leading to a fall in labor of 0.7% as opposed to an increase of 1.6% when tightness can be
adjusted. The fall in aggregate output would be much larger, -7%. All in all, these exercises
highlight that despite the extra burden of higher taxes to finance UBI, the increased action
in hiring makes up for the associated welfare losses.
We also show that a UBI policy reform would be harder to justify on welfare grounds if
paired with unemployment insurance. When UI is kept in place, we find that CEV peaks
at the same level of UBI but welfare gains attained in this case are nearly zero, 0.01%. The
reason for this is a much poorer performance in terms of job creation with UI. Under this
scenario, unemployment rises by 0.8 percentage points and aggregate output decreases by
6%. Unemployment increases since the presence of unemployment benefits exerts upward
pressure on the equilibrium wage, which lowers firms’ profits attained from filled jobs,
reducing the incentives to vacancy creation. In addition, the job finding rate also falls due
to lower search intensity. These effects are heterogeneous across education groups with
low educated agents experiencing a greater rise in unemployment since their reservation
wage are disproportionately more affected by the UI. In addition, wage inequality is lower
6
Since we keep the level of taxes at the benchmark, we allow the wasteful government expenditure to
adjust to balance the government budget.

5
but average wages fall more compared to the case where UI is also eliminated.
We contribute to the literature that studies the welfare and aggregate effects of taxes and
transfer programs using general-equilibrium models with heterogeneous agents. Lopez-
Daneri (2016) studies a revenue-neutral reform of the U.S. income tax and welfare system
that involves the adoption of a negative income tax. Wellschmied (2021) studies the sav-
ings effects of the asset means-test in US income support programs. Ortigueira and Siassi
(2021) study the effects of the U.S. anti-poverty system on savings, labor supply, and mar-
ital decisions of non-college-educated workers with children. Guner et al. (2020) study the
effects of conditional transfers on households’ labor supply with children.7
We also contribute to the emerging literature focusing on the impact of UBI using quan-
titative models in the Bewley-Huggett-Aiyagari tradition. Conesa et al. (2021) and Ludu-
vice (2021) find that it is hard to justify a UBI policy reform on welfare grounds. Guner
et al. (2021) find similar conclusions in a model that incorporates heterogeneity in gender
and marital status. Daruich and Fernández (2020) study the impact of UBI on parental
skill investments during early childhood and education decisions and find that UBI is not
a good idea when the welfare of future generations is taken into account. Ferriere et al.
(2021) build on the work in Heathcote et al. (2017b) to study the optimal negative relation
between transfers and income-tax progressivity using a Ramsey approach. They show that
most of the welfare gains in the benchmark plan can be attained by a UBI of $26,000 per
household.
We add to this literature by evaluating a UBI policy reform in a framework that com-
bines job search and job creation with incomplete markets. This is important because, on
the one hand, transfers affect the precautionary savings motive and wealth inequality. On
the other hand, redistribution policies also influence the labor market and the opportuni-
ties it provides. Thus, the efficiency-equity trade-off associated with transfers depends on
the endogenous relationship between vacancy creation, incentives to work, and savings
behavior. In addition, unlike the papers above in which the wage level is competitively de-
termined, in our model wages are determined by Nash bargaining and thus more generous
transfers affect workers’ outside options and firms’ incentives to create vacancies.
In related work, Jaimovich et al. (2021) also study the effects of UBI in a search-and-
matching model with incomplete markets. Our model and our analysis differ from this
paper in several ways. First, in their framework, pre-reform transfers are modeled as a
fixed amount received by agents with zero income. In contrast, we estimate transfers and
the Earned Income Tax Credit separately using microdata. Our baseline economy closely
matches the distribution of transfers by income in both cases. Second, agents in our econ-
7
Our work also relates to the literature that studies optimal tax progressivity (e.g., Conesa and Krueger
2006; Kindermann and Krueger 2020; Guner et al. 2016; Heathcote et al. 2017b), capital taxation (Golosov
et al. 2003; Conesa et al. 2009; Boar and Midrigan 2021) or age-dependent taxation (da Costa and Santos 2018;
Ndiaye 2017; Heathcote et al. 2017a).

6
omy are heterogeneous in terms of education. Since ex-ante heterogeneity is a key de-
terminant of inequality in earnings and wealth as well as in labor market outcomes, this
ingredient is important to evaluate the redistribute consequences of transfers programs.8
Third, we allow for on-the-job human capital accumulation, which implies that the returns
to forming a match are backloaded and thus more sensitive to transfers that affect the in-
centives to work and to separate. Fourth, in our model search intensity is endogenous.
Fifth, we study a reform in which UBI replaces not only the current transfer system but
also unemployment benefits and show that this is an important perspective one should
bear in mind when thinking about the consequences of the UBI policy.
The paper is organized as follows. In section 2, we present the model economy. In sec-
tion 3 we describe the parameterization and calibration of the benchmark economy as well
as how the different transfer programs and tax credits are estimated. Section 4 discusses
the properties of the benchmark economy. In section 5, we present the main findings of our
quantitative experiments, and section 6 concludes.

2 Model

2.1 Demography and preferences

Time is discrete and runs eternally. The economy is populated by overlapping generations
of individuals who face an exogenous probability, ν, of surviving to the next period. At
each period, 1 − ν new agents are born, and 1 − ν die, so that the population remains
constant at the normalized unit level. Individuals’ labor productivity is determined by
two components. The first one, denoted by e, is realized at birth and retained throughout
one’s life. It can be interpreted as innate ability and pre-market skills obtained through
education. The second one, z, is a stochastic component and its law of motion is explained
below.
Agents have preferences over random streams of consumption, ct , according to
" ∞ t
! #
X Y
E βet ν u(ct , `t , st ) , (1)
t=0 s=0

where E is the expectation operator conditional on information at birth and the intra-period
utility takes the following separable functional form:
!
c1−γ
t s1+φ
u(ct , `t , st ) = + `t d − χ t . (2)
1−γ 1+φ
8
For instance, Huggett et al. (2011) show that ex-ante heterogeneity is a key determinant of inequality in
earnings, wealth, and consumption, while Setty and Yedid-Levi (2021) show that it is important to consider
ex-ante heterogeneity to assess the redistributive consequences of unemployment insurance.

7
 The parameter d captures the utility from leisure when unemployed, while `t is an in-
dicator function which assumes value 1 when unemployed and zero otherwise. The val-
uation of leisure in (2) entails that the marginal rate of substitution between leisure and
consumption is decreasing in c, and thus the worker’s reservation match quality is increas-
ing in savings as in Bils et al. (2011). In addition, an unemployed worker looks for a job by
exertingchoosing search intensity s. The cost of searching for a job depends on how inten-
sively the worker searches. The parameter χ determines the cost of search effort, while φ is
an elasticity parameter that governs how search effort responds to a change in transfers.
We allow the discount factor βe to depend on individual’s ability as a way to generate an
empirically plausible cross-sectional wealth distribution, which has been already explored
in Krusell and Smith (1998) and more recently in Krueger et al. (2016), among others.9

2.2 Human capital accumulation

In line with Ljungqvist and Sargent (1998) and Kehoe et al. (2019), we think of z as repre-
senting the individual’s human capital, which is accumulated on the job in a leaning-by-
doing fashion. It captures the fact that while working, agents can practice their skills and
learn, thereby increasing productivity. The dynamics of z are stochastic and are character-
ized by shocks moving agents up and down the job ladder. In particular, the human capital
of an employed worker evolves according to the following law of motion:

log z 0 = (1 − ϕ)z̄ + ϕ log z + σε0 (3)

where ϕ determines the persistence, σ the volatility and z̄ the mean of the process, while ε
is a Gaussian disturbance with zero mean and unit variance.
While off the job, the worker has less opportunities to practice her skills and we assume
that those who transit from employment to non-employment maintain their most recent z
throughout the unemployment spell. The human capital of a non-employed agent who has
just found a job evolves according to a similar AR(1) process but with z̄ = 0, that is:

log z 0 = ϕ log z + σ0 . (4)

We represent below the Markov processes in (3) and (4) as Fw (z, z 0 ) and Fu (z, z 0 ). New-
born workers enter the economy endowed with a draw of z from the invariant distribution
associated with Fu (z, z 0 ). These laws of motion imply that a worker’s productivity increases
on average over time since it converges to exp(z̄) from below. The speed at which produc-
tivity and, as a consequence, the wage grow over the employment spell is determined by
the persistent parameter ϕ. For instance, the slope of the life-cycle earnings profile will be
steeper, the lower the value of ϕ.
9
See Falk et al. (2018) for experimental evidence on heterogeneity in discount rates.

8
 In addition, the higher the human capital of a worker who loses her job, the higher the
likelihood of her falling down the ladder since Fu (z, z 0 ) converges to exp(0) from above.
The model thus produces saw-toothed individual productivity profiles: growth while em-
ployed and decline in transitions from unemployment into employment.

2.3 Labor market frictions

Workers and firms come together via random search. We assume that a worker’s type is
observable and that firms can direct their search by posting vacancies for individuals of
a given level of education e. Let ue be the measure of nonemployed agents of type e, Se
be the aggregate search effort in market e, and ve the corresponding measure of vacancies
posted by firms for agents in submarket e. The flow of successful matches within a period
are given by the following matching function:

(Se ue )ve
M (Se , ve ) = 1 (5)
[(Se ue )η + veη ] η
where η determines the interaction between the measure of job searches Se ue and vacancies
ve .
We use this matching function, which was proposed by Haan et al. (2000), to ensure that
job finding rates are between 0 and 1. Specifically, dividing (5) by Se , the probability per
search effort that a nonemployed individual of type e matches with a vacancy in submarket
e is then

θe
m(θe ) = 1
(1 + θeη ) η
where θe = Sve eue is labor market tightness in the submarket e. A worker supplying search
effort s then finds a job with probability sm(θe ). The job filing rate in the submarket e can
similarly be obtained dividing (5) by ve :

1
q(θe ) = 1 .
(1 + θeη ) η

2.4 Asset markets

Consumers face idiosyncratic income shocks. Because markets are incomplete, they can-
not perfectly smooth consumption. Thus, savings may be precautionary and allow partial
insurance against shocks. Agents can accumulate two kinds of tangible assets: physical
capital, k, which is used as an input for production, and equity x, which is a claim for the
aggregate profit. Let r be the return to capital and div be the dividend paid to the holders of
equity. The total amount of equities is normalized to one. As there is no aggregate risk, the

9
equity price remains constant in equilibrium. The equity price p has to satisfy a standard
no-arbitrage condition, which implies that the returns on holding capital and equity are
equal:

div + p
p= (6)
1+r−δ
where δ is the depreciation rate of capital.
Since capital and equity both are riskless and provide the same return and therefore
are the same from the consumer’s viewpoint, we do not have to keep track of the asset
composition of the consumers. In the following, we define total financial resources as:

(1 + r − δ)(k + px) = (1 + r − δ)a (7)

and use a as the state variable for a consumer.

Asset holdings are subject to an exogenous lower bound. More precisely, for our main
exercise, we assume that agents are not allowed to contract debt, so that the amount of
assets carried over from one period to the next is such that a0 ≥ 0.

2.5 Government
The government levies taxes on capital income, consumption, and labor income. We as-
sume that consumption is taxed at a flat rate τc and capital income at a flat rate τk . In
addition, the government collects a non-linear and progressive tax schedule on labor in-
come according to the tax function suggested by Benabou (2002) and more recently used
by Heathcote, Storesletten, and Violante (2017):

T (w) = max{w − τw w1−ξ , 0} (8)

where w is the individual’s wage, and T (w) is tax paid. Parameters τw and ξ regulate the
level and progressivity of taxation, respectively. For instance, if ξ = 0 then the tax rate is
flat at 1 − τw , and the system is progressive if ξ > 0.
Government revenue is used to finance a stream of exogenously given government con-
sumption, G. In addition, there are two government-run programs in the economy. Newly
unemployed workers are eligible to receive unemployment insurance (UI) benefits. The
UI system consists of a replacement rate ϑ and a ceiling on the benefits b̄. As long as the
cap does not bind, the UI benefit is a fraction ϑ of the average wage w̄(e, z) earned by em-
ployed workers of type e and productivity z. In particular, the formula for the UI benefits

is given by b(e, z) = min ϑw̄(e, z), b̄ . Note that since a worker who transits from employ-
ment to unemployment maintains the most recent z throughout the unemployment spell,
the benefits are directly linked with the worker’s wage in the last job.
Hereafter, we use ι as an indicator function that takes value 1 if the unemployed worker

10
is collecting unemployment insurance and 0 otherwise. To economize on the state space,
we assume that the exhaustion of UI benefits are stochastic events, which is governed by
the following transition matrix:
 
1 0
Πι,ι0 =  . (9)
1−π π

Thus, for example, if an unemployed worker is not currently receiving UI benefits, ι = 0,

then she will not receive it next period, i.e. ι0 = 0 with probability one. On the contrary,
if ι = 1, the parameter π denotes the probability of UI benefits not being exhausted next
period.
The government also runs a welfare system designed to mimic the social programs in
the US economy. Households can receive two different types of transfers. The first one,
T r1 (y), can be considered welfare and is a function of individual’s total income, consist-
ing of wage or unemployment benefits, interest, and dividend income. In particular, y is
defined as follows:



 ra + w(a, e, z), if employed

y = ra + b(e, z), if unemployed with ι = 1


if unemployed with ι = 0.

ra,

The second type of transfer, T r2 (y), represents the EITC which is paid to low income em-
ployed agents and it is essentially a work subsidy. It is a function of total income y since
capital income is also part of the eligibility criteria. We assume that the amount of both
types of transfers are given by a Ricker function:

eρ1j eρj2 y y ρj3 if y > 0
T rj (y) = (10)
ρ j if y = 0.
0

That is, the transfers of an individual with no income are given by ρ00 and ρ10 , while
the transfers of a household with positive income are determined by the three parameters
(ρj1 , ρj2 , ρj3 ) each for j = 1, 2. Because EITC is contingent on work requirements, we have that
ρ20 is actually zero, which implies that agents at the bottom of the income distribution do
not necessarily receive the highest transfers due to the fact that many of these individuals
do not have a job.

2.6 Workers’ maximization problem

Let Vw (a, e, z) denote the value function of an employed worker with individual produc-
tivity (e, z) who owns a assets. Similarly, Vu (a, e, z, ι) denotes the value function of an un-

11
employed worker, where ι indicates the UI eligibility status as explained above. Workers
move between employment and unemployment according to the endogenous job-finding
rate mw (θe ), and the exogenous job separation rate ςe . Workers take both probabilities as
given. We can write the recursive problem of an employed agent as

Vw (a, e, z) = max : u(c, 0, 0)

a0 ≥0,c
h X i
+ βe ν (1 − ςe ) Fw (z, z 0 ) max{Vw (a0 , e, z 0 ), Vu (a0 , e, z 0 , ι = 1)} + ςe Vu (a0 , e, z 0 , ι = 1) (11)
z0

subject to the following budget constraint:

(1 + τc )c + a0 = [1 + (1 − τk ) r] a + w(a, e, z) − T (w(a, e, z)) + T r1 (y) + T r2 (y). (12)

Notice that the continuation value in (11) reflects the consumer’s survival probability, ν, the
type-specific exogenous separation probability ςe and the decision of whether to continue
a relationship. Equation (12) is the household’s budget constraint. The worker’s wage, w,
is determined through Nash bargaining between the firm and the worker every period as
explained below, and thus depends on her individual state (a, e, z).

Vu (a, e, z, ι) = max : u(c, 1, s)

a0 ≥0,c,s
X h X i
0 0 0 0 0 0 0
+βe ν Πι,ι sm(θe )
0 Fu (z, z ) max{Vw (a , e, z ), Vu (a , e, z, ι )}+(1−sm(θe ))Vu (a , e, z, ι )
ι0 z0
(13)

subject to
(1 + τc )c + a0 = [1 + (1 − τk ) r] a + ιb(e, z) + T r1 (y). (14)

2.7 Firms
On the other side of the market, there is a continuum of risk-neutral, infinitely-lived firms.
They maximize the expected value of the sum of profit streams and use the net real interest
rate r to discount the future. Firms use both capital and labor inputs to produce according
to a standard Cobb-Douglas production function f (k, n) = k α n1−α . Production can take
place only in a worker-job match and each match consists of one job and one worker. Labor
efficiency units, n, are thus given by ez.
To create a job, a firm first posts a vacancy. The flow cost of posting a vacancy is denoted
by κe . There is free entry of firms, so that the asset value of holding a vacant position is
always zero in equilibrium.

12
 q(θe ) X X λu (a, e, z, ι)
κe = F (z, z 0 ) max {J(a0 , e, z 0 ), 0} (15)
1 + r ι z0 ue

where a0 = au (a, e, z, ι).

A firm with a vacancy does not know what worker type it will meet next period.
The firm does know, however, the distribution of worker types among the unemployed.
The population of unemployed workers at state (a, e, z, ι) is given by λu (a, e, z, ι), so that
λu (a, e, z, ι)/ue is the conditional density function. Since the matching process is random,
the firm can be matched with any worker of type e in the current period unemployment
pool.
The free entry condition (15) pins down the vacancy-unemployment ratio in each sub-
market e, θe . Clearly, no vacancies are created in submarket e if the value of expected profits
conditional on matching is sufficiently low in that submarket. This may occur for all values
of e such that even if a vacancy leads to a match for a firm with probability one, expected
profits are lower than the vacancy posting cost.
Once a match is formed, firms choose the amount of capital employed in the job. The
value of a job filled by a type’s e worker with asset level a and productivity shock z is given
by
(
J(a, e, z) = max ψk α n1−α − (r + δ)k − w(a, e, z)
k
)
(1 − ςe )(1 − ν) X
+ Fw (z, z 0 ) max {J(a0 , e, z 0 ), 0} (16)
1+r z 0

where a0 = aw (a, e, z), meaning that the firm internalizes the worker’s next period asset
decision. The continuation value takes into account that a worker-firm pair is dissolved
exogenously with the worker’s death or with a per-period probability ςe , which depends
on the worker’s type. It also captures the possibility that a match no longer yields a positive
value to the firm and is thus destroyed.
Note that with a frictionless capital market, all firms pay the same rental rate r, im-
plying equal marginal products across firms. Thus, the same capital to labor ratio, k/n, is
employed at each filled job. In fact, the first order condition implies that
 1
 1−α
k α
k̄ = = (17)
n r+δ
α
α

1−α
and plugging (17) into (16), the flow profit can be written as π(a, e, z) = (1 − α) r+δ
n−
w(a, e, z).

13
2.8 Wage setting
Wages are determined, period by period and without commitment, using Nash bargaining
within each worker-firm pair. To define the Nash product, we assume that the outside
option of the worker is the value function of an unemployed worker collecting UI benefits,
Vu (a, e, z, ι = 1). We think this is relevant for a number of reasons. First, in standard
search-and-matching models, UI benefits are supposed to exert a push effect on wages
throughout the duration of employment. In order to generate this outcome, one should let
the incumbent worker use Vu (a, e, z, ι = 1) as her threat point to bargain with the firm. An
empirical justification for this is that fair and unfair dismissals cannot be distinguished,
but generally the burden of the proof that a dismissal was fair lies with the employer.
Therefore, we use Vu (a, e, z, ι = 1) to define the outside option of all workers and thus
the Nash bargaining solution solves the problem

max [Vw (a, e, z) − Vu (u, e, z, ι = 1)]ζ J(a, e, z)1−ζ (18)

w

where ζ ∈ (0, 1) is a parameter that represents the bargaining power of the worker. Sim-
ilarly to Krusell et al. (2010), the Nash solution generates a wage that is increasing in a
worker’s assets, reflecting that being unemployed is less painful for a worker with greater
assets. In turn, as can be seen in equation (15), this makes the vacancy creation decision to
depend on the unemployed asset holdings. To the extent that social insurance affects the
individual’s savings behavior, it establishes a channel through which transfers affect wage
and vacancy creation in the model.
In addition, marginal taxes and transfers affect wages and profits not only through their
influence on net payoffs, J and Vw − Vu , but also through the sharing rule. In fact, as has
been highlighted by Pissarides (1985), marginal taxes and transfers strengthen the firm’s
hand in the wage bargain since its share of the surplus from the job increases. Intuitively, a
small increment in the negotiated wage benefits the worker less since she attains a smaller
part of it. In contrast, unemployment benefits strengthen the worker’s hand in bargaining
since a small increment in the bargained wage would give her an extra benefit in case of
separation. The EITC has a similar effect for low levels of income as T r20 (ye ) > 0 in this case.
These effects are important for one to bear in mind when thinking about the consequences
of the UBI reforms considered below.

2.9 Equilibrium
A stationary equilibrium is a list of value functions (Vw (a, e, z), Vu (a, e, z, ι), J(a, e, z)), deci-
sion rules for asset holdings (aw (a, e, z), au (a, e, z, ι), and search intensity s(a, e, z, ι), a wage
function w(a, e, z), a population distribution across possible individuals’ states λe (a, e, z)
and λu (a, e, z, ι), a value of type-specific labor-market tightness θe , and a tax rate τw such

14
that:

1. Given the aggregate variable θe , the wage function w(a, e, z) and the policy parame-
ters, households solve the maximization problem in (11) and (13).

2. Given the wage schedule w(a, e, z) and the workers asset-holding decision rule, aw (a, e, z),
the value of a filled vacancy J(a, e, z) satisfies equation (16).

3. Given the asset value J(a, e, z), the asset-holding decision au (a, e, z, ι), the measure of
unemployed workers λu (a, e, z, ι), the number of vacancies posted in each submarket
e is consistent with equation (15).

4. The wage function w(a, e, z) is determined through Nash bargaining between the
firms and the workers according to (18).

5. Equilibrium distributions, λe (a, e, z) and λu (a, e, z, ι), satisfy the equilibrium stock-
flow equations across the different states of the economy implied by the sets of deci-
sion rules as well as the idiosyncratic shocks described above.

6. The tax parameter τw is such that the aggregate government’s budget constraint

Tw + τc C + τk rA = G + BU I + BT r (19)

is satisfied every period, where BU I and BT r are the aggregate UI benefits and trans-
fers that are due, and Tw is the aggregate revenue from labor income tax.

7. The dividend paid to equity owners every period is the sum of flow profits from all
matches, net of the expenditure on vacancies

X Z 
div = π(a, e, z)dλe (a, e, z) − κe ue θe
e

and p can be computed from (6).

8. Aggregate capital satisfies the asset market clearing condition:

1
k= A − p.
1 + (1 − τk )r

In addition, we assume that that newborn agents start off their lives in unemployment
without UI benefits, and with zero asset holding.

15
2.10 Welfare measure
The government is a benevolent Ramsey planner that fully commits to fiscal policy. The
planner maximizes social welfare by choosing a budget feasible level of transfers subject
to allocations being an equilibrium. We consider an Utilitarian social welfare criterion that
evaluates the ex-ante expected utility across all agents in the economy as in

∞ t
!
X Y
E βt ν u(ct , `t , st ) (20)
t=0 s=1

where E denotes the unconditional expectation operator with respect to all possible perma-
nent types and histories. This welfare criterion takes into account the concern of the policy
maker for redistribution and insurance against idiosyncratic shocks, as well as the distor-
tions the transfer system imposes on labor supply, job creation, and capital accumulation
decisions.
We compute the welfare change, ∆, as the amount of consumption that one would
have to remove or add in order to make the utilitarian welfare criterion equal between
a benchmark transfer system and some alternative policy. The welfare variation (CEV)
is calculated as follows: Let V (ω) denote the expected utility of an agent who enters the
economy with state ω under the transfer system we aim at evaluating. Then, define
" t
#
Y
0
V (ω) = E νUt,0 ((1 + ∆)ct , `t , st )
s=1

where Ut,0 (ct , `t , st ) is the flow utility attained by the agent under the benchmark at period t.
Our relevant measure of welfare variation is

CEV = min Eω V 0 (ω) − Eω V 1 (ω) .

 
(21)
∆

3 Parameterization and calibration

A model period is 1.5 months. We separate the parameters into two groups: the exoge-
nously given and parameters calibrated through the simulated method of moments. As
is customary, we associate the parameters with the target that provides the most intuition
for its value, but all parameters are determined jointly. The value of the parameters, their
sources, and targets are shown in Table 1.
Agents in our model represent household heads and we divide the labor force into four
productivity types according to levels of education: Less than high-school, high-school
graduates, some college, and bachelor’s degree. The population share of each group is
denoted by µe and, according to CPS data, corresponds to µ1 = 0.11, µ2 = 0.31, µ3 = 0.26

16
and µ4 = 0.32, respectively.

Preference parameters: We set risk aversion to 2 and the survival rate ν to 1−1/320, imply-
ing that workers stay in the market for an average of 40 years. We introduce heterogeneity
in discount factors across education groups in order to match differences in wealth between
the groups in the data. The SCF reports (every three years) the median family net worth
and provides a breakdown by education of household heads. We normalize the median
wealth of high school dropouts to 1 and compute the median wealth ratios by education
for each year. Our calibration targets are the averages of these premia for the years 1998 to
2013: 2.9, 3.3, 12.0 for high school graduates, some college, and bachelor’s degree and over,
respectively. The four discount factors, reported in the lower panel of Table 1, are chosen to
match the three premia. This procedure, together with the calibration of ν, yields an annual
interest rate of 4.8%.
The parameter d captures the utility from leisure when unemployed. Since d governs
the marginal rate of substitution between leisure and consumption, it also directly affects
the extent to which the worker’s reservation match quality is increasing in savings. Thus,
we calibrate d to approximate the correlation between wages and asset holdings observed
in the data. Figure A.1 shows the model replicates well the actual pattern of average wages
by wealth deciles.
The parameters of the search cost function, (χ, φ), are calibrated as follows. The value
of χ is chosen in such a way that the average time spent on job search is 3.8 percent of
the disposable time in line with Krueger and Mueller (2010), which reports that an unem-
ployed worker spends on average 32 minutes per day in job search activities. Since the
parameter φ governs how search effort responds to a change in benefits of unemployment,
we choose its value to match the average elasticity of unemployment duration with respect
to unemployment benefit generosity. The model counterpart of this moment is computed
by holding fixed the value of θe and thus it captures the response that would be observed
if only search intensity responds to UI. We use as a target the estimate obtained by Chetty
(2008), which indicates that a 10% increase in unemployment benefit level is associated
with a 3-5% increase in unemployment duration. Table 1 reports the calibrated parameters.

Labor productivity shocks: To calibrate ex-ante productivity, e, we use data on hourly

wages from the Panel Study of Income Dynamics (PSID). We restrict attention to male
household heads who are 25 years or older, continuously employed and have no miss-
ing data on wages or hours of work. We compute the hourly wage for each person every
year, assign people to the appropriate education group, and compute the median hourly
wage within an education group for a given year. We use the median wages to compute the
education premia in each year. The parameter values reported in the upper panel of Table
1 take into account the average wage premia over all years in our sample, while keeping
the weighted average of ex-ante productivity equal to one.

17
 The parameters that characterize the human capital dynamics are (z̄, ϕ, σ). We do not
have direct information about events that may change human capital on the job, such as
training or specific knowledge acquisition. Since (z̄, ϕ) are directly linked to life-cycle wage
growth, we use information on age-earnings profiles to identify these parameters. In par-
ticular, we choose (z̄, ϕ) to approximate the simulated mean earnings profile with the one
computed from a standard Mincer regression of log wages with standard controls, includ-
ing education. Then, we use the residual variance to pin down σ. The values that we obtain
are presented in Table 1. Our procedure implies that the difference in log wages between
workers with 30 years of experience and those just entering the labor market is 1.04, which
is comparable with the estimates in Elsby and Shapiro (2012) who use census data on full-
time workers and find a value of 1.19. To compute Fe (z, z 0 ) and Fu (z, z 0 ), we apply the
algorithm described in Tauchen (1986) to approximate the stochastic processes in (3) and
(4) by a first-order Markov chain with 41 points.
Figure A.2 shows how human capital evolves over the employment spell under the
baseline calibration. In the left panel, we show the evolution of human capital of a ran-
domly chosen agent who enters the market with mean human capital drawn from the in-
variant distribution of Fu (z, z 0 ). In the right panel, we show the average human capital of
a sample of agents who enter the market with mean human capital. Even though there is a
lot of volatility at the individual level, workers’ productivity, on average, grows over time.
The growth is steeper in the first few years and then slows down.

Technology: The values of the technological parameters are presented in Table 1. We

choose a value for α based on the US time series data from the National Income and Prod-
I/Y
uct Accounts (NIPA). The depreciation rate, in turn, is obtained by δ = K/Y − g. We set the
annual investment-product ratio I/Y equals to 0.22 and the annual capital-product ratio
K/Y equals 3. The economic growth rate, g, is constant and consistent with the average
growth rate of GDP over the second half of the last century. Based on data from Penn-
World Table, we set g equal to 2.3%, which yields a depreciation rate of nearly 5% in annual
terms. Finally, the scale parameter ψ is chosen so that the average wage in the benchmark
is equal to 1.

Recruiting cost and separation rate: The difference in unemployment rates among types
is directly related to the variation in the job separation rate. Thus, we compute the implied
separation rates such that the unemployment rate for each education group is consistent
with the data. According to the CPS data, the average unemployment rates decline with
skill and equal 8.7%, 6.3%, 4.6%, and 2.9%, respectively. The resulting values of ςe are
reported in Table 1. The values we find for ςe are in line with their counterpart in the data
3.3%, 2.1%, 1.6%, and 0.8%.
Next, we calibrate the recruiting cost parameters κe to match the job finding probability
in the data. We compute monthly job transitions using the CPS Merged Outgoing Rotation

18
Group from 1995-2019, while restricting the sample to household heads aged 25-65. The
resulting transition rates from unemployment to employment by level of education are
30.7%, 28.4%, 27.3%, and 27.6%, respectively. In particular, given the productivity levels
e, the expected individual productivity, and the separation rates, we use the free entry of
firms in 15 to solve for κe .

Matching function and worker’s bargaining power: The matching function parameter η
is chosen to be 1.60 in line with the value estimated by Schaal (2017). This value yields a
matching elasticity of nearly 0.6. We then use the Hosio’s condition to set the bargaining
weight to 0.60 following Shimer (2005) and many others.

Transfers and EITC: We estimate transfer functions as in Guner et al. (2022). In Figure
1 we can see the estimated function (black dashed line) as given by equation (10) using
microdata from the SIPP from 1998-2012. We can see that a household with zero income
in a given year on average receives 22% of mean household income in terms of transfers.
For a household with a positive income, the amount immediately drops to 17%, and the
received amount continues to drop after that.10

Figure 1: Transfers and EITC as a function of household income (normalized) - data vs. fit

(a) Transfers excluding EITC (b) EITC

.025
.2

.02
Transfers excluding EITC
.15

.015
EITC
.01
.1

.005
.05

0 .5 1 1.5 2
0

Normalized household income

0 .2 .4 .6 .8 1
Normalized household income Data Fit

Notes: The x-axis shows normalized household income, while the y-axis shows the normalized
amount of transfers received on average. The dots represent binned averages, while the gray line
shows the approximated schedule according to the estimated function (10) via OLS using microdata
from the SIPP.

We follow Guner et al. (2022) to compute the transfer schedules of the EITC, a refund-
able tax credit for low- to moderate-income households, in particular with children, which
has working requirements attached to it. The official schedule is characterized by three
intervals, the phase-in, plateau, and phase-out. This means that the credit is proportional
10
The presented amounts are unconditional on receipt. For details on take-up, we refer to Guner et al.
(2022).

19
 Table 1: Estimation and calibration of model parameters

External calibration
Parameter Description Values Source/target
γ Risk aversion 2 Standard
1
ν Death probability 1 − 320 40 years working life
α, δ Capital share, depreciation rate 0.3, 1.04% NIPA, I/Y ratio
τc , τk Consumption and capital tax 7%, 30% Fuster et al. (2007)
ξ Tax progressivity 0.098 Guner et al. (2014)
ϑ, 1 − π Repl. rate, prob. of UI expiring 0.40, 0.25 US policies
η Matching function curvature 1.60 Schaal (2017)
ζ Worker’s bargaining power 0.60 Hosio’s condition
e1 Permanent productivity <HS 0.65 Median earnings PSID
e2 Permanent productivity HS 0.83 Median earnings PSID
e3 Permanent productivity SC 1.00 Median earnings PSID
e4 Permanent productivity C 1.42 Median earnings PSID
Internal calibration
Parameter Description Values Target
β1 Discount factor <HS 0.9995 Wealth distribution
β2 Discount factor HS 0.9996 Wealth distribution
β3 Discount factor SC 0.9998 Wealth distribution
β4 Discount factor C 1.0001 Wealth distribution
κ1 Recruiting cost <HS 0.05 Job finding probability
κ2 Recruiting cost HS 0.11 Job finding probability
κ3 Recruiting cost SC 0.12 Job finding probability
κ4 Recruiting cost C 0.15 Job finding probability
ς1 Separation rate <HS 4.39% Unemployment rate
ς2 Separation rate HS 2.73% Unemployment rate
ς3 Separation rate SC 1.78% Unemployment rate
ς4 Separation rate C 0.99% Unemployment rate
z̄, ϕ Mean and persistence of prod. 1.55, 0.985 Mean earnings profile
σ Std of innovation 0.004 Residual inequality
τw Tax level 0.77 Balance gov. budget constraint
bmax Ceiling for UI benefits 0.24 48% of median wage
d Utility from leisure 0.06 Corr. between w and a
χ Cost of search 0.60 Time spent on job search
φ Cost of search 1.90 Elasticity of u duration w.r.t b
Notes: The internally calibrated parameters are estimated using the simulated method of moments (SMM) in which we minimize the sum
of the equally weighted squared distance between model and data moments.

20
 Table 2: Parameters of transfer and EITC schedules

Function parameters
Transfers EITC
ξ0 0.217 0.007
ξ1 -1.816 -1.873
ξ2 -4.290 -4.551
ξ3 -0.006 0.715
Notes: The parameters from the
transfer function (10) are estimated
using microdata from the SIPP
1998-2012.

to income up to a cutoff point before later declining with income, eventually reaching zero.
Froemel and Gottlieb (2016) rely on the government’s schedule to parameterize a relation
between income and EITC which they use in their model. However, even amongst eligible
households take-up has been found to be far below 100% (Currie 2004). Therefore, we use
microdata from 2001-2012 from the Survey of Income and Program Participation (SIPP) to
estimate EITC as a function of household income. In the right panel of Figure 1 we present
this relationship where both are normalized by average household income ($75,932.91).
We see that the average amount received plateaus at around 20% of normalized income at
which households on average receive about 2% of normalized income.11
The parameters of the Ricker model as in equation (10) for transfers and the EITC are
presented in Table 2.

Other government parameters: First, we set government consumption, G, to 17% of the

economy’s output under the baseline calibration. Following the literature, we assume a
consumption tax of 7% and a capital income tax rate of 30%.12 The parameter ξ, which
governs the progressivity of the labor income taxes, is calibrated in line with the estimates
provided by Guner et al. (2014). Marginal tax rates are chosen to raise enough revenue to
balance the government budget constraint. The value we find for τw is 0.77.13 The parame-
ters of the UI system are then calibrated as follows. We set the replacement rate ϑ at 40%,
which is a value typically used for the US economy (Krusell et al., 2010), and the ceiling
on unemployment benefits, b̄ = 0.24, corresponds to about 50% of the mean wage in the
model. This calibration is based on the mean of the maximum benefits by state as reported
by the U.S. Department of Labor. The probability π is set to 0.75 to make UI benefits expire
after 26 weeks – roughly four periods in the model – in line with US policies.
11
Given that the SIPP provides intervals with the highest category being $3,500 for 2001-2003, $4,000 for
2004-2007, and $5,000 as of 2008 this could be considered a lower bound. However, only 10% of household
receiving positive amounts claim to receive an amount in this category so the impact of the cap should be
rather small, especially when considering that the maximum amount that can be received according to federal
rules is $5,112 in 2011.
12
See, for example, Fuster et al. (2007).
13
Recall that τw (1 − ξ)w−ξ is the marginal retention rate, 1 − T 0 (y).

21
4 Benchmark economy
In Figure 2 we present the fit of the model (black bars) relative to the data (gray bars). In the
top left panel, we see the share of total wages that accrue to each wage quintile. The model
captures the distribution nicely with, for instance, 3.9% (3.7%) of wages being earned by
the bottom wage quintile in the model (data), and 48% (47%) by the top wage quintile. The
top right panel plots the same distribution for wealth quintiles. Again the model does an
excellent job capturing the distribution of wealth, which is visibly more unequal than it is
for labor income.
The bottom left panel shows that transfers are largely captured by the bottom income
quintile in the model (49%) and the data (49%). The top income quintile receives 4.1% of
transfers in the data, which is lower in the model (2.6%). This is because the calibration has
only limited power to change the distribution of transfers, given that they are governed
by exogenous parameters estimated outside the model. In the right panel in the bottom,
we see that an even larger fraction of EITC payments accrue to the bottom income quintile
in the data (61%) and the model (61%), while top income earners receive hardly any EITC
payments.

Figure 2: Benchmark model fit: Inequality and transfers

Wage distribution Wealth distribution
50 100
Model Model
40
Data 80 Data

60
30
(%)

(%)

40
20
20

10
0

0 -20
1 2 3 4 5 1 2 3 4 5
Wage quintile (%) Wealth quintile (%)
Transfers distribution EITC distribution
60 70
Model Model
50 Data 60 Data
50
40
40
(%)

(%)

30
30
20
20

10 10

0 0
1 2 3 4 5 1 2 3 4 5
Income quintile (%) Income quintile (%)

Notes: The figure shows the model fit by comparing the shares of total wages, wealth, transfers,
and EITC being held/received by each quintile in the model (black) and data (gray).

In Figure 3, we show the model fit in terms of labor market outcomes. In the top left

22
panel, we see that the model matches the patterns of the unemployment rates across edu-
cation groups closely. Out of those with less than high school education, 8.8% (8.7%) are
unemployed in the model (data), while for those with a college degree, the unemployment
rates in the model (data) are much lower at 2.9% (2.9%). The bottom right panel displays
the job finding probability across education groups. Those with less than high school edu-
cation face a 49% (50%) probability of transitioning from unemployment to employment in
the model (data), while for those with college education the probability is 44% (44%).
Finally, in the bottom graph of Figure 3, we show the unemployment rate by wealth
decile for the model and the data. The actual values are computed using the microdata from
the SIPP for household heads aged 25 to 65. It can be seen that, both in the model and in the
data, unemployment in the bottom decile is twice as high as for the second decile and four
times higher than the remaining deciles. Since these are not directly targeted moments in
our calibration, the figures provide an external validation by showing that our model repli-
cates well not only the relationship between wages and wealth inequality observed in the
data – as shown in the previous section – but also the relationship between unemployment
and wealth inequality. This is important because the implied efficiency-equity trade-off of
transfers depends on the endogenous relationship between vacancy creation, incentives to
work, and savings behavior. Our general equilibrium framework allows us to study how
these ingredients interact to determine the distributional and aggregate consequences of
transfers.

5 Policy experiments

5.1 The impact of transfers and EITC

In this section, we study how the current transfer system affects the labor market and in-
equality. Table 3 and Figure 4 display the findings for the case where we eliminate transfers,
T r1 (y), and the EITC, T r2 (y). The second column displays the results for the benchmark
economy. In the third column, we show the results when only transfers, T r1 (y), are re-
moved, while in the fourth, we eliminate only the EITC, T r2 (y). Finally, the fifth column
displays the findings when both types of transfers are removed. In the top panel, we show
the relevant aggregate variables, while the bottom panel displays the Gini coefficient for
gross and net income, assets, and consumption. Figure 4 displays the heterogeneous ef-
fects of the reforms.
We start by describing the effects on labor market outcomes. When transfers and EITC
are removed, the unemployment rate goes down by nearly 1.3 percentage points, which
is almost entirely driven by the elimination of transfers. In fact, as can be seen in Table
3, in the case that we eliminate only the EITC, the unemployment rate actually increases
by 0.14 percentage points. This is not surprising given that the EITC is a subsidy to work.

23
 Figure 3: Benchmark model fit: Labor market outcomes
Unemployment rate Job finding probability
10 60
Model Model
Data 50 Data
8

40
6
(%)

(%)
30
4
20

2
10

0 0
1 2 3 4 1 2 3 4
Education group Education group
Unemployment rate
14
Model
12 Data
10

8
(%)

0
1 2 3 4 5 6 7 8
Wealth decile (%)

Notes: In the top two panels the figure shows the model fit by comparing the unemployment rate
and job finding probability by level of education (1 is less than high school, 2 is high school, 3 is
some college, and 4 is college graduate) in the model (black) and data (gray). In the bottom left
panel the figure shows the non-targeted fit of the unemployment rate by wealth decile in the model
(black) and data (gray).

Note that unemployment goes up due to less job creation, which is explained by a lower
level of aggregate capital. On the contrary, when only transfers are eliminated, the model
predicts a sizeable increase in both job creation and search intensity, leading to an increase
in job finding probability of nearly 17 percentage points. In the middle panel of Figure 4,
we show that these effects are stronger among lower educated groups.
The model predicts a sizable increase of 7.7% in average wages when transfers and
the EITC are removed. Interestingly, even though the removal of transfers has a negative
effect on reservation wages for low-income groups, wages actually rise for all agents in
the counterfactual economy. This is so because the increase in the precautionary savings
motive leads to an increase of 14% in aggregate capital, raising the value of a job and push-
ing wages up. This effect, however, is heterogeneous, and labor income becomes more
disperse. In fact, one can see in the top left panel of Figure 4 that the bottom quintiles ex-
perience a fall in their relative share of total wages, while workers in the top quintile reap
the greatest share.
The increase in wage inequality is counteracted by a reduction in wealth inequality. In

24
 Table 3: Eliminating transfers and the EITC

Removing

Variable Benchmark Transfers EITC Both

Aggregate output Y 100 +5.21% -1.18% +4.79
Aggregate capital A 100 +14.38% -3.62% +12.99
Labor 100 +1.46% -0.14% +1.42
Consumption 100 +1.99% -2.08% +1.82%
Mean human capital 100 0.24% -0.02% 0.22%
Unemployment rate u 5.03% 3.71% 5.17% 3.75%
Mean job finding prob. 45.06% 61.92% 43.90% 61.34%
Mean search intensity 100 +21.03 +0.31% +21.45
Mean wage 100 +8.44% -2.59% +7.75%
Mean income 100 +4.49% -1.28% +4.26%
Interest rate r 0.59% 0.46% 0.64% 0.47%
τw 22.32% 15.42% 22.33% 14.91%
BT r1 /REV 16.01% 0.00% 16.44% 0.00%
BT r2 /REV 1.38% 1.91% 0.00% 0.00%
Gini
Gross income 0.480 0.483 0.493 0.484
Net income 0.386 0.441 0.401 0.447
Consumption 0.323 0.392 0.333 0.397
Assets 0.763 0.705 0.785 0.711
Welfare
CEV — -24.10% -1.93% -25.88%
CEV employed — -9.36% -1.88% -11.40%
CEV unemployed — -46.77% -2.06% -47.94%
Notes: The table presents changes to the economy when eliminating transfers, EITC, and both. Marked with ‘+/-’ are rel-
ative changes to the benchmark economy set to 100 for outcomes for which levels do not have an obvious interpretation.
For outcomes for which levels have a clear interpretation, we present actual levels indicated by the level in the benchmark
column.

fact, Table 3 shows that the Gini measuring wealth inequality drops from 0.76 to 0.71. In
addition, while the share of wealth held by the richest 20% falls from 81% to 63% – see
the right panel in the top of Figure 4 – the shares of the quintiles one, two, three, and four
increase by 0.05, 2.0, 5.8 and 7.2 percentage points, respectively.
This fall in the dispersion of asset holdings is explained by a stronger response of sav-
ings among low-income individuals when we move to the counterfactual economy. In
addition, the fall in the interest rate reduces wealth accumulation at the top, where a large
share of income is accounted for by capital income. These effects come almost entirely from
eliminating transfers rather than the EITC, which is due to the relative size of both types of
programs.
Although the removal of transfers has positive effects on economic activity, with aggre-

25
gate output increasing 5.2%, the model predicts a more dire picture when one considers the
impact on welfare. In fact, in the bottom of Table 3, we show that overall welfare as mea-
sured by the consumption equivalent variation (CEV) declines 24%, with the unemployed
bearing most of the loss, 47%, due not only to the loss of insurance but mainly to a rise in
search cost. Note that the social insurance provided by transfers is also highly valued by
employed workers, as is reflected by the sizeable welfare loss of 9.3% when this type of
transfer is removed. The welfare effect of eliminating EITC is relatively small, -1.9%, with
the impact on both groups being more similar.

5.2 Universal basic income

In the following, we present two sets of counterfactual simulations in which each person
receives a pre-determined fraction of average total income in the benchmark, irrespective
of their income level or employment status. In the first set, we remove transfers and the
EITC. In the second set, we additionally remove unemployment benefits.14 Within each set,
we present the aggregate effects for four levels of universal basic income: 10%, 15%, 20%,
and 25% of average income. The costs of the scheme are financed by adjusting the level of
taxation through τw in order for the government budget to be balanced in equilibrium. In
each panel of Figure 5 we present relative changes to the benchmark economy for outcomes
for which levels do not have an obvious interpretation, and we present levels for outcomes
with a clear interpretation.15 The black dashed lines represent the experiment for UBI is
paired with UI, while the solid gray line reflects the results with UBI only.
It can be seen that lower levels of UBI increase economic activity, but the effects are
much smaller in the presence of UI. In fact, when a UBI of 10% is paired with UI, aggregate
capital, labor and output go up by 2.89%, 0.58%, and 1.27%, respectively, while the figures
are 14.4%, 2.15% and 5.73% for the case where UI is also removed. The increase in labor
goes hand in hand with a drop in the unemployment rate: 0.5 percentage points in the for-
mer case and 1.8 in the latter. In addition, average human capital also goes up, accounting
for nearly 22% of the increase in labor when UBI substitutes for all programs as opposed to
10% with UI.
To understand the stronger effects on the economy without UI, note first that low levels
of UBI provide less than what unemployed agents receive from the combination of trans-
fers, EITC, and UI in the benchmark. This drop in social insurance boosts the precautionary
savings motive in good times and search effort in bad times, especially among agents in the
lower rungs of the income distribution. Given the higher levels of asset accumulation, firms
have more capital at their disposal, increasing the marginal returns to labor and, therefore
14
Since UBI is set as a fraction of the benchmark average income, individuals receive the same amount in
both cases.
15
In Appendix Tables A.1 and A.2 we present the statistics.

26
 Figure 4: Eliminating transfers and EITC
Wage distribution Wealth distribution
50 100

Benchmark Benchmark
40 w/o EITC 80 w/o EITC
w/o Transfers w/o Transfers
30 w/o Both 60 w/o Both
(%)

(%)
20 40

10 20

0 0
1 2 3 4 5 1 2 3 4 5
Wage quintile (%) Wealth quintile (%)
Unemployment rate Job finding probability
10 70

Benchmark 60
8 w/o EITC
w/o Transfers 50

6 w/o Both
40
(%)

(%)
30
4

20
2
10

0 0
1 2 3 4 1 2 3 4
Education group Education group
Change in avg. search intensity (%)
25

20

15

10

-5
1 2 3 4
Education group

Notes: The figure shows the the shares of total wages and wealth (top) going to each quintile in the
benchmark (black) and when eliminating EITC (dark gray), transfers (gray), and both (light gray).
Similarly, the remaining panels show the unemployment rate, job finding probability, and change
in mean search intensity for each education group (1 is less than high school, 2 is high school, 3 is
some college, and 4 is college graduate).

the incentives to create vacancies. The second effect acts on the supply side. Since em-
ployed and unemployed agents receive the same value, a universal basic income policy –
if not set too high – provides no ex-ante disincentive to work. This positive effect of UBI
is tampered out when UI is in place through its effect on the amount of effort devoted to
searching for a job and the reservation wage of the unemployed.
Both effects become clear when one compares the job finding probability in both economies.
In fact, a UBI of 10% causes the job finding probability to go up by 13% when UI is kept

27
 Figure 5: The impact of different levels of UBI with and without UI on the economy
Change relative to benchmark

Change relative to benchmark

Change relative to benchmark

Change relative to benchmark

Aggregate output Y Aggregate capital A Aggregate consumption C Aggregate labor L

.05

-.03-.02-.01 0 .01 .02

-.3 -.2 -.1 0 .1 .2
.05

0
0

-.05
-.1 -.05

-.1
0 .05 .1 .15 .2 .25 0 .05 .1 .15 .2 .25 0 .05 .1 .15 .2 .25 0 .05 .1 .15 .2 .25
UBI share mean benchmark income UBI share mean benchmark income UBI share mean benchmark income UBI share mean benchmark income
Change relative to benchmark

Change relative to benchmark

Change relative to benchmark

Mean income Mean wage Mean search intensity Interest rate r
.05

.005.006.007.008.009 .01
.1

.1 .2

Interest rate r
0

-.2 -.1 0
-.05

-.1
-.1

-.2

0 .05 .1 .15 .2 .25 0 .05 .1 .15 .2 .25 0 .05 .1 .15 .2 .25 0 .05 .1 .15 .2 .25
UBI share mean benchmark income UBI share mean benchmark income UBI share mean benchmark income UBI share mean benchmark income

Unemployment rate u Mean job finding probability Tax progressivity Share of Y spent on UBI
Mean job finding probability

Share of Y spent on UBI

.15
.03 .04 .05 .06 .07

.3 .4 .5 .6
Unemployment rate u

.3 .4 .5 .6 .7

Tax progressivity

.05 .1
.2

0
0 .05 .1 .15 .2 .25 0 .05 .1 .15 .2 .25 0 .05 .1 .15 .2 .25 0 .05 .1 .15 .2 .25
UBI share mean benchmark income UBI share mean benchmark income UBI share mean benchmark income UBI share mean benchmark income

with UI without UI

Notes: Each panel shows the effect of introducing different levels of UBI relative to the benchmark
economy without UI (gray solid line) and with UI (black dashed line). The outcomes are indicated
in the titles of the panels. The numeric results can also be found in Appendix Tables A.1 and A.2.

in place, while it jumps nearly 60% without UI. Note that this stark difference can not be
explained by the rise in search intensity, which increases only by 5 percentage points more
when all programs are removed. As UBI becomes more generous, the incentives to save
and search for a job reduce and the supply side effect mentioned above becomes dominant.
For instance, for a UBI of 20%, even though aggregate capital falls -1.89% and search in-
tensity is similar to the benchmark, the increase in the job finding probability is still large,
nearly 38%, if all programs are removed. In contrast, when a UBI of 20% is paired with
UI, the job finding probability collapses 15%. These patterns go hand in hand with the
ones observed for unemployment and aggregate labor. Even for a UBI of 25% – nearly
$15,600 annually – we find that unemployment (labor) is lower (higher) compared to the
benchmark if UI is also eliminated.
To understand how the interaction between UBI and UI affects job creation in the model,

28
the top left panel of Figure 8 shows the expected return of a match for a UBI of 20%.16 It can
be seen that the expected benefit of creating a vacancy increases without UI for all education
groups, while it decreases when UI is kept in place. The reason for this is twofold. On the
one hand, as is shown in the right top panel of Figure 8, the value of a filled vacancy is
larger for poor agents when UBI substitutes for all programs. For richer workers – see the
bottom left panel – the value of a filled vacancy without UI is still higher than with UI but
lower than the benchmark. On the other hand, there is also a composition effect due to
increased human capital accumulation. This is shown in the bottom right panel of Figure
8, which displays the change in the distribution of z. It can be seen that – without UI – there
is an increase in the mass of workers in the upper end of the distribution of z for which the
value of a job is much higher, while the opposite happens when UBI is paired with UI.
Intuitively, hiring workers is an investment activity in which costs are paid upfront and
benefits accrue gradually as individuals become more productive when employed. Better
incentives to work induce higher average tenure on the job and thus greater incentives to
job creation.
Note that, as expected, the cost of UBI is high due to its universal nature, requiring a
large increase in the tax level to balance the government budget. Even for a UBI of 10%
– nearly $6,250 annually – the program’s total cost amounts to 4% of output when UBI
replaces all programs, which corresponds to an increase of nearly 1.5 percentage points in
government transfer payments as a share of output compared to the benchmark. Moreover,
despite each individual receiving the same amount in both economies, UBI is more costly
when paired with UI. In fact, for a UBI of 20%, the total cost of UBI equals 8.5% of output
when UI is also eliminated as opposed to nearly 9.1% when UI is provided. The difference
increases as the level of UBI grows since labor market outcomes deteriorate faster in the
latter case.
It is important to highlight that average wages also increase initially and then fall steadily
as UBI becomes more generous. As is shown in Figure 5, for a UBI of 10%, the increase in
wages is much larger in the economy without UI (8.3% as opposed to 1.5%). For high lev-
els of UBI, both the drop in capital and the rise in the tax level reduce wages. Interestingly,
when one looks at the change in wage by education groups – see Figure A.3 – it can be seen
that the pattern is very different between the two economies. As UBI becomes more gen-
erous, the average wage falls much more for low education groups when all programs are
removed. In contrast, if we keep UI in place, the change in wage is more uniform across ed-
ucation levels and is smaller for less-educated workers. The reason for this is twofold. On
the one hand, unemployment insurance disproportionately affects the reservation wage of
low education groups, which prevents their bargaining wage from falling by as much as
their more educated counterparts. On the other hand, the lower average tenure when UI is
16
F (z, z 0 ) max {J(a0 , e, z 0 ), 0} λu (a,e,z,ι)
P P
This corresponds to the right hand side of equation 15: ι z0 ue .

29
in place affects more highly skilled agents for whom human capital accumulation is a more
important driver of earnings growth over the life cycle.
The changes in wages described in Figure A.3 imply that skill premium - and thus wage
inequality – increases compared to the benchmark when UBI replaces all programs. This
concentration of wages can also be seen in the top left panel of Figure 9, which displays the
distribution of wages by quintile. Note that the shares of wages captured by the bottom
four quintiles tend to be decreasing in the level of UBI, as the top quintile sees an increase.
Given the progressive nature of the tax schedule, the increase in wage inequality amplifies
the redistribution to the poor since a larger part of the costs is financed by high-income
earners.
In Figure 6 we show how inequality in other dimensions is affected by the same sets
of experiments. For the economy without UI, income inequality first falls slightly for low
levels of UBI and then increases as UBI becomes more generous. When UI is kept in place,
the fall in income inequality is initially more pronounced, but it increases faster for high
levels of UBI. In contrast, consumption inequality spikes and increases by 10% when a UBI
of 10% replaces all other programs. Increasing UBI further, consumption inequality drops
and falls below benchmark levels when reaching 20% of benchmark income. Interestingly,
for high values of UBI, the economy where UI is kept in place generates more net income
and consumption inequality than the economy where all programs are eliminated. The
reason for this is twofold. On the one hand, as discussed above, wages fall more in the
economy with UI due to larger disincentives to capital accumulation and job creation. On
the other hand, the associated increase in unemployment – see Appendix Figure A.4 – is
larger for low educated individuals.
In addition, Figure 6 also shows that asset inequality falls for low levels of UBI in both
cases and is always larger in the economy without UI. This is so because the absence of
unemployment insurance generates a stronger precautionary savings motive among agents
in the lower rungs of the wealth distribution. It can be seen in the top right panel of Figure 9
that the share of the top 20% exhibits a U-shaped pattern: it decreases nearly 9 percentage
points for a UBI of 15% and then increases as the UBI becomes more generous, staying
below the benchmark value for a UBI of 25%. The opposite pattern is observed for the
lower quintiles. For the economy in which UBI is paired with UI – see the top right panel
of Figure A.4 – the response is less pronounced when UBI is 15% but the share of the fifth
(forth) quintile increases (decreases) rapidly as UBI increases. Note that the larger rise in
interest rate also helps to explain the concentration of wealth in this case.
In Figure 7 we look at the share of the population that would favor UBI over the bench-
mark in a one-dimensional referendum. UBI combined with UI is favored over the bench-
mark at all levels of UBI, but peaks at 76% when UBI is 20% of benchmark income. Without
unemployment insurance, UBI can only convince 40% of the population at a 10% level but
also reaches an approval rating of 75% when it is 20% of benchmark income.

30
 Figure 6: The impact of different levels of UBI with and without UI on inequality

Gross income Net income

.42
.48 .49 .5 .51 .52 .53

.4
Gini

Gini
.38 .36
0 .05 .1 .15 .2 .25 0 .05 .1 .15 .2 .25
UBI share of mean benchmark income UBI share of mean benchmark income

Consumption Assets

.85
.36

.8
.32 .34
Gini

Gini
.75
.3

.7

0 .05 .1 .15 .2 .25 0 .05 .1 .15 .2 .25

UBI share of mean benchmark income UBI share of mean benchmark income

with UI without UI

Notes: The figure shows changes in the Gini coefficient relative to the benchmark when introducing
different levels of UBI (x-axis) with UI (black dashed) and without (gray solid).

As for welfare, we report in the right top panel of Figure 7 the aggregate consumption
equivalent variation (CEV) for both economies, while the bottom panels show the welfare
change by employment status. It can be seen that a relatively low level of UBI – 10% of
the benchmark average income – leads to a sizable welfare loss of 9% when all programs
are removed. Note that the loss is large even if UI is kept in place, 6%. As UBI becomes
more generous, aggregate welfare gains go up, increasing faster in the economy without
UI. Under this scenario, we find that welfare gains peak at 0.7% when UBI is 20% of average
income. In addition, all these gains are attained by employed workers as the unemployed
experience a welfare loss of -1.5%. For the economy with UI, we find that CEV peaks at the
same level of UBI, but welfare gains are nearly zero, 0.01%. In this case, winners and losers
flip side, and a CEV of 16.7% attained by the unemployed is compensated by a welfare loss
of -0.8% experienced by the employed.
Thus, our findings indicate that a UBI policy reform is more likely to improve welfare if
it substitutes for all programs, including UI. It is worth noting that the welfare-maximizing
level of UBI is in line with the magnitudes advocated in policy discussions. For instance,

31
 Figure 7: The impact of different levels of UBI with and without UI on welfare
Change relative to benchmark .

Change relative to benchmark

Political support CEV
.4 .5 .6 .7 .8

-.1 -.08-.06-.04-.02 0
.1 .15 .2 .25 0 .05 .1 .15 .2 .25
UBI share of mean benchmark income UBI share of mean benchmark income
Change relative to benchmark

Change relative to benchmark

CEV employed CEV unemployed

-.3 -.2 -.1 0 .1 .2

-.08 -.06 -.04 -.02 0

0 .05 .1 .15 .2 .25 0 .05 .1 .15 .2 .25

UBI share of mean benchmark income UBI share of mean benchmark income

with UI without UI

Notes: The figure shows changes in the share of the population preferring the new steady state rela-
tive to the benchmark (top left), and changes in welfare as measured using consumption equivalent
variation when introducing different levels of UBI (x-axis) with UI (black dashed) and without (gray
solid).

the policy proposal advocated by Andrew Yang, a candidate to the primaries of the Demo-
cratic Party for the presidential election of the United States in 2020 is to give every agent
in the economy a UBI that would amount to $12,000 per year, or $1,000 monthly, which
corresponds to nearly 20% of the average income in the U.S.17
In the bottom right panel of Figures 9 and A.4, we look at the impact of different levels
of UBI by education. We see that a UBI of 20% generates a positive CEV for all education
groups except for the highly educated ones. In general, those who have not completed high
school prefer higher levels of UBI, while those with college education prefer lower levels.
Note also that low (high) education groups benefit (lose) more (less) when UBI is 20% in
the economy without UI, which explains the much larger welfare gains attained under this
scenario.
To understand why low educated agents experience more improvements – and college
17
The average income in 2020 is nearly $62,500.

32
workers do not – without UI, one can look at the heterogeneous impacts on labor market
outcomes in both cases, which are also presented in Figures 9 and A.4. It can be seen that
unemployment exhibits a U-shaped pattern and the levels are always below the benchmark
values for all education groups in the economy without unemployment insurance. Note
that, for high levels of UBI, unemployment decreases more for unskilled workers than for
the skilled ones. In fact, for a UBI of 25%, the unemployment rate of workers with less
than high-school is 24% lower than its benchmark value, while it is 9% lower for the high-
est education group. On the contrary, when UI is kept in place, the unemployment rate
increases in the level of UBI, and it grows faster for college educated workers, being 55%
higher when UBI is 25% as opposed to 35% for those with less than high school education.
Note that high levels of UBI decrease search intensity in both cases, but the effect is bigger
for more educated individuals and much larger in the presence of UI. Interestingly, when
UBI replaces all programs, the increase in job creation compensates for the fall in search
intensity, leading to an increase in job finding rate as can be seen in the middle right panel
of Figure 9. Given that search intensity falls more for college workers, the increase in the
job finding rate is smaller for this group.
The above results indicate that general equilibrium effects acting through job creation
are important to understand the effects of UBI but are less informative about the quantita-
tive importance of the different channels at work. To this end, we conduct a decomposition
exercise to quantify the importance of i) taxes, ii) labor market tightness θe , and iii) the in-
terest rate r in shaping the UBI effects. Given that the welfare gains peak for a UBI around
20% without unemployment insurance, we narrow the analysis under this scenario. Since a
UBI of 20% is very costly, 8.5% of output, tax distortions are likely to have first-order effects.
We show in Table 4 that keeping the level of labor income taxation, τw , at the benchmark
would boost consumption by 19.5%. At the same time, the overall welfare gain from UBI
would increase by 20%.18 In addition, average wages would go up by nearly 7% instead of
falling 2.8% when τw is allowed to change.
Assuming that labor market tightness is kept unchanged, a UI of 20% that replaces all
programs would drop the job finding rate by 4.5 percentage points instead of an increase
of 17 percentage points in the main experiment. The unemployment rate would increase
by 0.4 percentage points, leading to a fall in aggregate labor by 0.7% instead of an increase
of 1.6% when tightness is allowed to adjust. The aggregate capital and output fall would
be much larger: -21% and -7%, respectively.
Finally, keeping the interest rate at the benchmark level would lead to an output growth
of 2.7%, largely explained by an increase in capital of 5.5%. Job finding probability would
not increase as much as the main experiment due to a fall of nearly 1% in search intensity.
All in all, these exercises highlight that despite the extra burden of higher taxes to fi-
18
Since we keep τw at the benchmark, we allow the wasteful government expenditure G to adjust to balance
the government budget.

33
 Figure 8: Understanding the impact of UBI on job creation
Expected benefit of creating a vacancy Value of a filled vacancy: poor agent
8 6
College
Without UI
6 With UI 5

4
4
(%)

2
2

0
1

-2 Less than HS
0
1 2 3 4 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2
Education group Productivity
Value of a filled vacancy: rich agent
5 1.5
College

4 1

3 0.5

2 0

1 -0.5

0 Less than HS -1

-1 -1.5
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2
Productivity Productivity

Notes: Top left panel shows the right hand side of the vacancy creation condition – equation 15 – by
education level. Top right panel shows the value of a filled vacancy, J(a, e, z), by z (log scale) and by
education level for a worker with low a. J(a, e, z) is shown for the benchmark (black line), for the
economy where UBI is 20% without UI (dashed dark gray) and for the economy where a UBI of 20%
is pared with UI (light gray). Bottom left panel shows J(a, e, z) when a is high. The bottom right
panel shows the percentage change in the distribution of z compared to the benchmark. Education
group: 1=less than high school, 2=high school, 3=some college, and 4=college graduate.

nance UBI, the increased action in hiring makes up for the associated welfare losses, and
it is a key channel through which less-educated groups improve with the reform despite
their sizable fall in wages and an overall rise in wage inequality. Moreover, although dis-
tributing the same amount to everyone, including the very rich, the UBI reform can still
result in redistribution to the poor since it is financed through progressive taxation.

34
 Figure 9: The impact of different levels of UBI without UI
Wage distribution Wealth distribution
60 100

Benchmark Benchmark
50
15% 80 15%
20% 20%
40
25% 60 25%
(%)

(%)
30
40
20

20
10

0 0
1 2 3 4 5 1 2 3 4 5
Wage quintile (%) Wealth quintile (%)
Unemployment rate Job finding probability
10 80

Benchmark
8 15%
60
20%
6 25%
(%)

(%)

40
4

20
2

0 0
1 2 3 4 1 2 3 4
Education group Education group
Change in avg. search intensity (%) CEV (%)
10 4

5
2

0
0
-5
-2
-10
15% 15%
-4
-15 20% 20%
25% 25%
-20 -6
1 2 3 4 1 2 3 4
Education group Education group

Notes: The figure shows the shares of total wages and wealth (top) going to each quintile in the
benchmark (black) and when UBI is 15% (dark gray), 20% (gray), and 25% (light gray) of mean
benchmark income. Similarly, the remaining panels show the unemployment rate, job finding prob-
ability, change in mean search intensity, and change in CEV for each education group (1 is less than
high school, 2 is high school, 3 is some college, and 4 is college graduate).

35
 Table 4: Decomposition - UBI 20% without UI

Keeping at the benchmark

Variable Benchmark UBI 20% Taxes θe r

Aggregate output Y 100 +0.54% +4.95 -7.06% +2.71
Aggregate capital A 100 -1.89% +12.63 -20.81% +5.52
Consumption 100 -1.00% +19.55% -17.37% +1.74%
Labor 100 +1.62% +1.79 -0.66% +1.51
Mean human capital 100 +0.39% +0.42% -0.13% +0.37%
Unemployment rate u 5.03% 3.65% 3.51% 5.45% 3.75%
Mean job finding prob. 45.06% 62.44% 65.09% 40.40% 60.66%
Mean search intensity 100 +0.16% +1.73 -6.28% -0.98
Mean wage 100 -2.82% +6.97% -15.77% +0.12%
Mean income 100 -0.55% +3.42% -9.84% +1.81%
Interest rate r 0.65% 0.70% 0.54% 0.95% 0.65%
τw 22.36% 41.90% 22.36% 45.87% 40.24%
BU BI /Y — 8.47% 8.12% 10.07% 8.29%
Gini

Gross income 0.495 0.515 0.504 0.590 0.499

Net income 0.398 0.377 0.378 0.412 0.367
Consumption 0.329 0.326 0.325 0.350 0.320
Assets 0.779 0.743 0.728 0.820 0.720

Welfare
CEV — 0.68% 19.92% -21.38% 4.44%
CEV employed — 0.85% 19.46% -22.11% 3.73%
CEV unemployed — -1.50% 7.88% -15.59% 2.10%
Notes: The table presents changes to the economy when eliminating transfers, EITC, and UI, while introducing UBI of 20% of mean bench-
mark income. In the column ‘Taxes’ we keep taxes, in ‘θe ’ labor market tightness, and ‘r’ the interest rate at the benchmark level. Marked
with ‘+/-’ are relative changes to the benchmark economy set to 100 for outcomes for which levels do not have an obvious interpretation. For
outcomes for which levels have a clear interpretation, we present actual levels indicated by the level in the benchmark column.

Conclusions
Transfers to poorer households help buffer shocks but can disincentivize precautionary
savings and work. The US, and many other developed countries, spend a considerable
amount of money on transfers. In this paper, we look at how transfers and the EITC affect
the US economy, inequality, and welfare. To do so, we calibrate a rich heterogeneous agent
search-and-matching model with incomplete markets to the US economy while incorpo-
rating transfer functions estimated using microdata.
When we shut down the EITC in a counterfactual exercise, we find benign negative

36
effects on the economy and welfare. On the other hand, removing transfers boosts the
economy but leaves the average person worse off due to the extremely low consumption
associated with unemployment.
Universal basic income has become a much discussed policy option with little real-
world evidence to draw from. We fill this gap by simulating an economy in which we
withdraw all transfer programs and replace them with different levels of UBI. Within these
experiments, we also investigate the differential impact when pairing UBI with unemploy-
ment insurance versus UBI without unemployment insurance.
When paired with unemployment insurance, we find that economic activity across all
dimensions drops rapidly as the level of UBI increases. The reason is that unemployment is
a relatively comfortable state with both unemployment insurance and a higher level of UBI
and precautionary savings cease to be necessary. We find that at 20% of mean benchmark
income, a world with UBI paired with unemployment insurance dominates the benchmark
economy for almost three-quarters of individuals. However, these welfare gains are heav-
ily concentrated amongst the less educated and the unemployed. When also removing
unemployment insurance, political support in a referendum compared to the benchmark
economy would be equally high, but gains would be much less concentrated. While the
college-educated would still on average be unhappy with such a policy given that they are
financing most of the handouts through increased taxes, the majority would prefer the as-
sured inflow of cash. Moreover, overall economic output would actually be higher than in
the benchmark.
One reason the overall economy does not contract is that, because of the universal na-
ture of lump-sum transfers, the disincentives to (search for) work are reduced. Therefore
a positive side effect of UBI, which is mostly overlooked in other models, is that vacancy
creation is also more attractive given the reduced frictions. In a decomposition exercise, we
find that the increase in labor market tightness is the key driver of the gains.
This project brings us a step closer to understanding the overall impact the introduction
of UBI would have. However, other proposals of UBI include even replacing social security.
This sort of interesting question is left to future research.

37
References
Aiyagari, S. R. (1994). Uninsured idiosyncratic risk and aggregate saving. The Quarterly
Journal of Economics 109(3), 659–684.

Bewley, T. (1986). Stationary monetary equilibrium with a continuum of independently

fluctuating consumers. Contributions to mathematical economics in honor of Gérard De-
breu 79.

Bils, M., Y. Chang, and S.-B. Kim (2011). Worker heterogeneity and endogenous separations
in a matching model of unemployment fluctuations. AEJ: Macro.

Boar, C. B. and V. Midrigan (2021). Nit picking: The macroeconomic effects of a negative
income tax. Working Paper, NYU..

Chetty, R. (2008). âmoral hazard versus liquidity and optimal unemployment insurance,.
Journal of Polotical Economics.

Conesa, J. C., S. Kitao, and D. Krueger (2009). Taxing capital? not a bad idea after all!
American Economic Review 99(1), 25–48.

Conesa, J. C. and D. Krueger (2006). On the optimal progressivity of the income tax code.
Journal of Monetary Economics 53(7), 1425–1450.

Conesa, J. C. C., B. Li, and Q. Li (2021). A quantitative evaluation of universal basic income.
Technical report, Working Paper, Stony Brook University.

Currie, J. (2004). The take up of social benefits. Technical report, National Bureau of Eco-
nomic Research.

da Costa, C. E. and M. Santos (2018). Age-dependent taxation with endogenous life-cycle

path for wages. International Economic Review.

Daruich, D. and R. Fernández (2020). Universal basic income: A dynamic assessment.

Technical report, National Bureau of Economic Research.

Diamond, P. A. (1982). Wage determination and efficiency in search equilibrium. The Review
of Economic Studies 49(2), 217–227.

Elsby, M. and M. Shapiro (2012). Why does trend growth aect equilibrium employment? a
new explanation of an old puzzle. American Economic Review.

Falk, A., A. Becker, T. Dohmen, B. Enke, D. Huffman, and U. Sunde (2018). Global evidence
on economic preferences. The Quarterly Journal of Economics 133(4), 1645–1692.

38
Ferriere, A., P. Grubener, G. Navarro, and O. Vardishvili (2021). Larger transfers financed
with more progressive taxes? on the optimal design of taxes and transfers.

Froemel, M. and C. Gottlieb (2016). The earned income tax credit: Targeting the poor but
crowding out wealth.

Fuster, L., A. Imrohoroglu, and S. Imrohoroglu (2007). Elimination of social security in a

dynastic framework. Review of Economic Studies.

Golosov, M., N. Kocherlakota, and A. Tsyvinski (2003). Optimal indirect and capital taxa-
tion. The Review of Economic Studies 70(3), 569–587.

Guner, N., R. Kaygusuz, and G. Ventura (2014). Income taxation of us households: Facts
and parametric estimates. Review of Economic Dynamics 17(4), 559–581.

Guner, N., R. Kaygusuz, and G. Ventura (2020). Child-related transfers, household labour
supply, and welfare. The Review of Economic Studies 87(5), 2290â2321.

Guner, N., R. Kaygusuz, and G. Ventura (2021). Rethinking the welfare state. Technical
report.

Guner, N., M. Lopez-Daneri, and G. Ventura (2016). Heterogeneity and government rev-
enues: Higher taxes at the top? Journal of Monetary Economics 80, 69–85.

Guner, N., C. Rauh, and G. Ventura (2022). Means-tested transfers in the us: Facts and
parametric estimates. Technical report.

Haan, W., G. Ramey, and J. Watson (2000). Job destruction and the experiences of displaced
workers. In Carnegie-Rochester Conference Series on Public Policy, vol. 52, ed. Bennett T.
McCallum and Charles I. Plosser, 87-128. Amsterdam: North- Holland..

Heathcote, J., K. Storesletten, and G. L. Violante (2017a). Optimal progressivity with age
-dependent taxatio. Technical report, Minneapolis Fed SR 551.

Heathcote, J., K. Storesletten, and G. L. Violante (2017b). Optimal tax progressivity: An

analytical framework. The Quarterly Journal of Economics 132(4), 1693–1754.

Hoynes, H. and J. Rothstein (2019). Universal basic income in the united states and ad-
vanced countries. Annual Review of Economics 11, 929–958.

Huggett, M. (1993). The risk-free rate in heterogeneous-agent incomplete-insurance

economies. Journal of economic Dynamics and Control 17(5-6), 953–969.

Huggett, M., G. Ventura, and A. Yaron (2011). Sources of lifetime inequality. American
Economic Review 101(December), 2923–2954.

39
Jaimovich, N., I. Saporta-Eksten, O. Setty, and Y. Yedid-Levi (2021). Universal basic income:
Inspecting the mechanisms.

Kehoe, P., V. Midrigan, and E. Pastorino (2019). Debt constraints and employment. Journal
of political Economy.

Kindermann, F. and D. Krueger (2020). High marginal tax rates on the top 1%? lessons
from a life cycle model with idiosyncratic income risk.

Krueger, A. B. and A. Mueller (2010). Job search and unemployment insurance: New evi-
dence from time use data. Journal of Public Economics.

Krueger, D., K. Mitman, and F. Perri (2016). Macroeconomics and household heterogeneity.
Handbook of Macroeconomics.

Krusell, P., A. Mukoyama, and A. Sahin (2010). Labour-market matching with precaution-
ary savings and aggregate fluctuations. The Review of Economic Studies 77(4).

Krusell, P. and T. Smith (1998). Income and wealth heterogeneity in the macroeconomy.
Journal of Political Economy.

Ljungqvist, L. and T. J. Sargent (1998). The european unemployment dilemma. Journal of

political Economy.

Lopez-Daneri, M. (2016). Nit picking: The macroeconomic effects of a negative income tax.
Journal of Economic Dynamics and Control 80, 1–16.

Luduvice, A. V. D. (2021). The macroeconomic effects of universal basic income programs.

Marinescu, I. (2018). No strings attached: The behavioral effects of us unconditional cash

transfer programs. National Bureau of Economic Research Working Paper Series (w24337).

Mortensen, D. T. (1982). Property rights and efficiency in mating, racing, and related games.
The American Economic Review 72(5), 968–979.

Ndiaye, A. (2017). Flexible retirement and optimal taxation.

Ortigueira, S. and N. Siassi (2021). The u.s. tax-transfer system and low-income households:
Savings, labor supply, and household formation. Review of Economic Dynamics, 1–44.

Pissarides, C. A. (1985). Short-run equilibrium dynamics of unemployment, vacancies, and

real wages. The American Economic Review 75(4), 676–690.

Salazar, J., B. Blom, T. Gullo, J. Holland, R. Sunshine, and K. Kelly (2017). Federal spending
for means tested programs, 2007 to 2027. Technical report, Congressional Budget Office
Washington United States.

40
Schaal, E. (2017). Uncertainty and unemployment. Econometrica 85(6), 1675–1721.

Setty, O. and Y. Yedid-Levi (2021). On the provision of unemployment insurance when

workers are ex-ante heterogeneous.

Shimer, R. (2005). The cyclical behavior of equilibrium unemployment and vacancies.

American economic review 95(1), 25–49.

Tauchen, G. (1986). Finite state markov-chain approximations to univariate and vector

autoregressions. Economics letters 20(2), 177–181.

Wellschmied, F. (2021). âthe welfare of asset means-testing income support. Quantitative

Economics, 12, 217–249.

41
A Additional tables and figures

Figure A.1: Average wages

Average wages

1.8 Model
Data
1.6

1.4

1.2

0.8

0.6

0.4

0.2
0 2 4 6 8 10
Wealth decile
Notes: Average wages are normalized to 1 both in the data and in the model.

Figure A.2: Life-cycle profile of human capital

(a) Human capital of a random agent (b) Mean human capital

2.5 1.2

2
1

1.5
0.8
1

0.6
0.5

0 0.4
0 50 100 150 200 0 50 100 150 200
Periods Periods
Notes: One model period on the x-axis is equivalent to 1.5 months. Panel a) shows the human
capital, ez , for a single random agent who enters the market with mean human capital. Panel b)
shows the average human capital of a sample of newborns who enters the market with mean human
capital.

42
 Figure A.3: Change in average wage by education
Without UI With UI
15 5
10%
10 15% 0
20%
25%
5 -5
(%)

(%)

0 -10

-5 -15

-10 -20
1 2 3 4 1 2 3 4
Education Education

Notes: The figure shows the change in average wages by education groups relative to the bench-
mark. The left panel shows the case where when all programs are replaced by UBI, while the right
panel shows the results when UI is kept in place. The change when UBI is 10% is represented by the
black bar, 15% by the dark gray bar, 20% by the gray bar, and 25% by the light gray bar.

43
 Table A.1: Universal basic income with unemployment insurance

Fraction of average income at the baseline

Variable Benchmark 10% 15% 20% 25%

Aggregate output Y 100 +1.27% -2.38% -6.16% -9.86%
Aggregate capital A 100 +2.89% -7.10% -16.83% -24.53%
Consumption 100 +0.97% -2.82% -2.44% -10.14%
Labor 100 +0.58% -0.26% -1.10% -2.62%
Mean human capital 100 +0.06% -0.09% -0.23% -0.60%
Unemployment rate u% 5.03 4.48% 5.16% 5.83% 7.11%
Mean job finding prob. 45.06% 51.07% 43.58% 38.29% 30.98%
Mean search intensity 100 +7.59% -2.85% -10.60 -20.19%
Mean wage 100 +1.50% -4.22% -11.17% -16.82%
Mean income 100 +1.17% -1.55% -5.39% -9.34%
Interest rate r 0.65% 0.63% 0.74% 0.88% 0.99%
τw 22.36% 28.76% 37.63% 44.88% 59.70%
BU BI /Y — 4.21% 6.55% 9.08% 11.82%

Gini

Gross income 0.495 0.484 0.493 0.523 0.533

Net income 0.398 0.395 0.379 0.378 0.361
Consumption 0.329 0.344 0.326 0.324 0.302
Assets 0.779 0.730 0.769 0.807 0.827

Welfare
CEV — -5.83% -2.55% 0.01% -3.29%
CEV employed — -4.94% -2.75% -0.78% -4.79%
CEV unemployed — -12.02% 3.21% 16.67% 23.85%
Political support — 57.38% 72.15% 75.96% 58.18%
Notes: The table presents changes to the economy when eliminating transfers and EITC, while introducing UBI of 10%, 15%, 20%, and 25%
of mean benchmark income. Marked with ‘+/-’ are relative changes to the benchmark economy set to 100 for outcomes for which levels do
not have an obvious interpretation. For outcomes for which levels have a clear interpretation, we present actual levels indicated by the level
in the benchmark column.

44
 Table A.2: Universal basic income without unemployment insurance

Fraction of average income at the baseline

Variable Benchmark 10% 15% 20% 25%

Aggregate output Y 100 +5.73% +4.20% +0.54% -4.08%
Aggregate capital A 100 +14.46% +9.54% -1.89% -15.55%
Consumption 100 +2.61% +2.25% -1.00% -6.31%
Labor 100 +2.15% +1.96% +1.62% +1.21%
Mean human capital 100 +0.48% +0.45% +0.39% +0.34%
Unemployment rate u 5.03% 3.20% 3.37% 3.65% 3.98%
Mean job finding prob. 45.06% 71.80% 68.05% 62.44% 56.58%
Mean search intensity 100 +12.86% +7.67% +0.16% -9.49%
Mean wage 100 +8.34% +4.67% -2.82% -6.25%
Mean income 100 +4.43% +2.69% -0.55% -1.23%
Interest rate r 0.66% 0.52% 0.57% 0.70% 0.89%
τw 22.36% 26.19% 33.00% 41.90% 51.75%
BU BI /Y — 4.03% 6.13% 8.47% 11.10 %

Gini

Gross income 0.495 0.492 0.487 0.515 0.515

Net income 0.398 0.410 0.381 0.377 0.355
Consumption 0.329 0.364 0.334 0.326 0.298
Assets 0.779 0.697 0.700 0.743 0.766

Welfare
CEV — -9.01% -0.19% 0.68% -0.74%
CEV employed — -8.14% -0.21% 0.85% -1.82%
CEV unemployed — -30.64% -11.67% -1.50% 2.22%
Political support — 40.63% 52.15% 74.91% 68.38%
Notes: The table presents changes to the economy when eliminating transfers, EITC, and UI, while introducing UBI of 10%, 15%, 20%, and
25% of mean benchmark income. Marked with ‘+/-’ are relative changes to the benchmark economy set to 100 for outcomes for which lev-
els do not have an obvious interpretation. For outcomes for which levels have a clear interpretation, we present actual levels indicated by
the level in the benchmark column.

45
 Figure A.4: The impact of different levels of UBI with UI

Wage distribution Wealth distribution

50 100

Benchmark Benchmark
40 15% 80 15%
20% 20%
30 25% 60 25%
(%)

(%)
20 40

10 20

0 0
1 2 3 4 5 1 2 3 4 5
Wage quintile (%) Wealth quintile (%)
Unemployment rate Job finding probability
12 60

Benchmark
10 50
15%
20%
8 40
25%
(%)

(%)

6 30

4 20

2 10

0 0
1 2 3 4 1 2 3 4
Education group Education group
Change in avg. search intensity (%) CEV (%)
0 4

-5 2

0
-10
-2
-15
-4
-20
-6
15% 15%
-25 20% -8 20%
25% 25%
-30 -10
1 2 3 4 1 2 3 4
Education group Education group
Notes: UBI substitutes for the welfare programs but UI is kept in place. The figure shows the shares
of total wages and wealth (top) going to each quintile in the benchmark (black) and when UBI is 15%
(dark gray), 20% (gray), and 25% (light gray) of mean benchmark income. Similarly, the remaining
panels show the unemployment rate, job finding probability, change in mean search intensity, and
change in CEV for each education group (1 is less than high school, 2 is high school, 3 is some
college, and 4 is college graduate).

46