Centre de recherche sur l'emploi et les

uctuations economiques (CREFE )

Center for Research on Economic Fluctuations and Employment (CREFE)
Universite du Quebec a Montreal

Cahier de recherche/Working Paper No. 64

Voting on Unemployment Insurance Generosity 

Stephane Pallage
CREFE - Universite du Quebec a Montreal
Christian Zimmermann
CREFE - Universite du Quebec a Montreal

July 1998

||||||||||||||||||||{
Pallage: Department of Economics, UQAM, CP 8888 Succ. Centre-Ville, Montreal, QC, H3C 3P8, email:
pallage.stephane@uqam.ca.
Zimmermann: Department of Economics, UQAM, CP 8888 Succ. Centre-Ville, Montreal, QC, H3C 3P8,
email: zimmermann.christian@uqam.ca.
 This paper circulated previously under the title \Moral Hazard and Optimal Unemployment Insurance
in an Economy with Heterogeneous Skills" and was presented at the University of Miami, the Institute for
International Economic Studies at Stockholm University, Universite Paris I, Concordia University, HEC
Montreal, Universitat Bern, the annual meeting of the Society for Economic Dynamics in Oxford, the
Canadian Macro Study Group in Toronto and the Tow Conference at the University of Iowa. We beneted
greatly from comments by Dave Andolfatto, Paul Gomme, Gary Hansen, Ayse I_mrohoroglu, Vctor Ros-Rull,
Randy Wright and conference participants at the above institutions.
Abstract:
In this paper, we show that in a dynamic general equilibrium economy, the presence of
moral hazard need not induce large cuts in optimal unemployment insurance benets when
agents are asked to vote on these benets. We nd that it takes quite a large proportion
of \shirkers" to bend the generosity of the unemployment insurance program desired by the
median voter. This result stands in sharp contrast to the extant literature. In addition to
their dierences in asset holdings, agents in our economy are heterogeneous in skills and may
dier in their wage as well as in their transition in and out of unemployment. Our results
have important implications in the light of recent unemployment insurance reforms.

Keywords:
Unemployment insurance, shirking, heterogeneity, voting.
JEL classication: J65, E24, D7
1 Introduction
In this paper, we analyze the eect of moral hazard on the generosity of an unemployment
insurance system when heterogeneous agents are asked to vote on the benets.
An unemployment insurance program is a mechanism by which society protects its members
from adverse idiosyncratic employment shocks. Because society cannot perfectly dierentiate
those hit and those pretending to be hit by such shocks, the optimal unemployment insurance
[UI] is necessarily less generous than under full information. The economic literature is
unanimous on this qualitative impact of moral hazard. It remains an open question, however,
how much less generous the optimal unemployment insurance should be under moral hazard.
In other words, what are the quantitative consequences of the possibility of shirking? This
question has a great social signicance since, in the name of moral hazard, many governments
have recently proposed a reform of their UI program aiming at a reduction of expenditures.
The few quantitative studies to date all seem to agree that moral hazard matters a lot.
Hansen & I_mrohoroglu (1992) question the opportunity of an unemployment insurance pro-
gram when a small proportion of agents can abuse the system and go undetected. Using a
general equilibrium model of the United States economy, they show that even little moral
hazard can cause the optimal replacement ratio to drop dramatically. Zhang (1995) and
Wang & Williamson (1996), in models with unobservable search and job{retention eorts,
draw similar conclusions.
In this paper, we revisit these results by passing to voters the choice of UI generosity. When
confronted with a certain level of moral hazard in the economy, how does a majority arise
in favor of a certain unemployment insurance program? If agents are heterogeneous in
their asset holdings but also in their chances of nding a job and in their expected income,
would they rather vote for a public UI system or a private, self{nanced one? In previous
studies, the model is tted to the United States economy under the assumption that skills
are homogeneous accross agents. Yet, in the United States people who never nished high
school account for 12% of the labor force, high school graduates represent a 61% share,
while only 27% of the labor force holds a college degree.1 Furthermore, it appears that these
groups have very dierent average incomes and unemployment rates. An individual without
a high school degree earns on average 3.4 times less in the United States than a university
graduate2 and is 4.3 times as likely to be jobless.1
Our model is a dynamic general equilibrium model with indivisible labor and a storage
technology like that of Hansen & I_mrohoroglu (1992).3 Agents in our economy are categorized
along four skill types which determine their income level and the matrix of their transition
probabilities in and out of employment opportunities. They choose how much to consume,
whether to work or not if oered a job, and how much to save. Their skill is common
1 Source: Statistical Abstract of the United States, based on Current Population Survey, 1993, Bureau of
the Census.
2 Based on the Survey of Income and Program Participation for 1990 (US Bureau of the Census).
3 Gomes, Greenwood, & Rebelo (1997) use a similar model, but add aggregate uncertainty since their
focus is the study of the unemployment rate over the business cycle.

1
knowledge, but the reason for unemployment is imperfectly monitored: as a result, some
agents may choose to shirk by rejecting oers and still try to collect UI benets.
Aside from the works of Hansen & I_mrohoroglu (1992), Zhang (1995) and Wang & Willi-
amson (1996), a good number of studies also try to assess the welfare implications of UI
programs. Baily (1978) was one of the rst to estimate the optimal replacement rate. Follo-
wing Shavell & Weiss (1979), Hopenhayn & Nicolini (1997) characterize the unemployment
insurance as an optimal contract between a risk neutral principal (the state) minimizing cost
and a risk-averse agent (the jobless worker). They determine the ecient size of benets
over time as a function of the unemployment spell. Since their notion of eciency is one of
minimum cost, it says little about the eects of the UI program on output and welfare in
general. The merit of their study, however, is to be precise about the terms of the ecient
contract, while we take the contract as given.
The potential welfare gains of an unemployment insurance scheme can be of three types.
First, a UI program allows for better consumption smoothing. Second, by providing insu-
rance it can allow workers to be more selective in their job search. This in turn may lead to
better job matches. Third, it may have redistributive eects. Part of the literature can be
divided in the way these eects are considered. For instance, Andolfatto & Gomme (1996),
focussing on the welfare impact of smoother consumption, show that the 1971 reform of the
Canadian UI system, giving way to broader benets, was welfare improving. Their model,
however, is deprived of saving technologies, hence agents cannot self{insure. Such a model
is likely to over{estimate the welfare gains of a generous UI program. An empirical study by
Gruber (1994) also focussing on the consumption smoothing eect in a model without moral
hazard, suggests that current replacement rates in the US are too high. Valdivia (1996)
nds similar results in a calibrated search model. The job{matching eect is analyzed by
Acemoglu & Shimer (1997) who nd it non{negligeable within a search model without moral
hazard. In this paper we bring a novel focus on redistribution while also analyzing consump-
tion smoothing eects. Our methodology does not permit us to account for the last source
of welfare gains. But the job{matching eect would only strengthen our results.
Only Wright (1986) and Hassler & Rodrguez Mora (1997), however, analyze the process
of voting for a UI system. Wright shows a time inconsistency in the votes of agents. Two
identical people would vote dierently whether they are employed or unemployed. We show
that this inconsistency does not necessarily carry through in our economy with moral hazard
and incentive eects. Hassler & Rodrguez Mora, in a model similar to Wright's, except for
the possibility of revoting, show that low turnover has a positive impact on the generosity
of the UI program chosen by the median voter.
We proceed as follows. In the coming section, we outline the model economy, that we later
parametrize to the United States economy in Section 3. We discuss our results in Section 4,
and conclude in Section 5.

2
2 The Model Economy
Our model economy is the economy of Hansen & I_mrohoroglu (1992) augmented with skill
heterogeneity. The economy is made of a continuum of innitely{lived agents. The conti-
nuum is divided in four groups j 2 f1; : : :; 4g of invariant measure j .4 The four groups
dier by one characteristic, the level of education. This heterogeneity in population allows
us to account for the dierences in transition probabilities and income between population
groups observed in the United States.
Agents have preferences over consumption c and leisure l. One good is produced in this
economy through labor. It can be consumed or saved using a storage technology. The
quantity of that good that was saved to date represents an agent's asset m. Labor is indi-
visible. If a worker, individual i in group j works for ^h hours out of his un it{endowment
of time (lij = 1 ^h). This worker produces yj units of output, where yj re ects his group's
productivity. Job opportunities occur randomly in a way that is described below.5
The sequence of events and actions is as follows. Every period t, agents learn whether
they are given a job opportunity or not. They carry with them their savings mijt to date
and a \certicate" of employment in the previous period (tij = 1 if a job opportunity was
taken at t 1, 0 otherwise). Whoever is (not) oered a job at t is indexed by st = e (u).
Employment opportunities follow Markov transition probabilities pj (st+1jst). A person who
is currently oered a job can choose to take it or leave it. In the latter case, he may receive
unemployment benets with a certain probability, itself depending on his previous state of
employment, (tij ) | that is whether he is a quitter (st = e, tij = 1, tij+1 = 0) or a
searcher (st = e, tij = 0, tij+1 = 0). Once agents have found out their job opportunity and
labor decisions have been taken, those who work are paid yj , those without job opportunity
receive UI benets and those who shirk learn if they have been successful at collecting
benets. Given their disposable income and their current assets, agents nally choose their
level of consumption and savings.
To sum up, each agent i in group j chooses a consumption{leisure{savings bundle (cij ; lij ; mij )
so as to maximize the following objective function:

1
X
U (cij ; lij ) = E0  tU (cijt ; ltij )
t=0

4 In other words there are no exchanges of population between the four groups. This assumption highly
simplies the computation of equilibrium since the distribution of agents between these groups does not
enter as a state variable.
5 Previous models of indivisible labor such as Hansen (1985), Greenwood & Human (1987) and Rogerson
(1988) all suered from the peculiarity that in equilibrium unemployed agents were better o than workers,
because they ended up consuming equal amounts. Rogerson & Wright (1988) have shown that non{separable
preferences prevent this from occurring. The model developed in Hansen & I_mrohoroglu (1992) and ours
use this type of preferences.

3
subject to the budget constraint:

mijt+1 + cijt = mijt + ytijd

where, U (:; :) is a strictly concave and increasing function of consumption and leisure, yijd
stands for the agent's disposable income and  his discount factor. Note that introducing an
interest payment on goods stored would not change the results, as Hansen & I_mrohoroglu
(1992) have argued.
Unemployment insurance is oered by a government which collects a tax ( ) to nance its
program. A public UI program in this set{up is a pair (;  ) specifying a replacement ratio 
common to each skill group and a tax rate  . The UI policy is taken as given by individual
agents when they solve their intertemporal expected utility maximization. The government
balances its budget every period and chooses the UI program that rallies the majority of
votes. The following agents are eligible for UI benets: Every unemployed agent (s = u),
that is anyone who did not receive an employment opportunity, but also every quitter or
searcher who refused a job and went undetected. The probability for a shirker to receive
benets (tij ) corresponds to the probability of not being caught when abusing the system.
It depends on the monitoring eort chosen by the government. We can view monitoring as
a lottery. Since there is a one{to{one relationship between (:), the success rate of shirking,
and the level of moral hazard, we will use (:) as an index of moral hazard.
Eligibility is governed by the following binary function:
8 8
>
>
> >
< with probability 1 if sijt = u
>
>
>
< 1 > with probability (0) if sijt = e; tij = 0; tij+1 = 0
ij
t = > :
>
with probability (1) if sijt = e; tij = 1; tij+1 = 0
>
>
>
>
:
0 otherwise.

For an agent in group j , eligibility to UI benets implies receiving a fraction  of a worker's

income yj in the same group. His date t disposable income ytijd depends on his employment
opportunity sijt 2 fe; ug, his acceptance or rejection of a job oer tij+1 = f0; 1g, his eligibility
to UI benets ijt 2 f0; 1g, the replacement ratio  and the tax rate  :
8
>
>
> yj (1
< j
 ) if sijt = e and tij+1 = 1
ytijd(sijt ; tij+1; ijt ; ;  ) = > y (1  ) if ijt = 1
>
:0
> if ijt = 0

The problem of the agents is recursive and admits the following Bellman equation, which
will prove very useful when we compute the equilibrium. Agents take as given the level of
moral hazard (0) and (1) as well as the UI policy. For ease of notations, we drop the time
4
subscript and use primes to denote future states. Call  the vector of exogenous variables {
i.e.  = (j; ; ; (0); (1)). Agents solve:

v(m; s; ; ) =
8
>
>
>
maxm U (m + (1  )yj m0; 1) +  Ps pj (u; s0)v(m0; s0; 0; )
0 0 if s = u
>
>
>
>
> 8 9
< >
>
>
>
maxm U (m + (1  )yj m0; 1 ^h) +  Ps pj (e; s0)v(m0; s0; 1; ) ;
0 0
>
>
>
>
> < =
>
>
> max > ()[max U (m + (1  )yj m0; 1) +  P pj (e; s0)v(m0; s0; 0; )] > if s = e
>
>
> > m >
> > s >
: +(1  ( ))[maxm U (m m0 ; 1) +  P pj (e; s0)v (m0; s0; 0; )]
0
> >
0
: ;
0
s 0

We will call \private", an insurance scheme in which agents of a given skill group nance
their own UI program, with the result that replacement ratios as well as tax rates (insurance
premia) may dier accross skill groups. By private, we therefore understand a UI program
that is compulsory and that can discriminate with an easily visible criterion, here education.
Eligibility to benets from either public or private programs will be governed by the same
rules.
In the next section, we parametrize this model to the United States economy and solve it
numerically.

3 Parametrization, computations and voting

All parameter choices are made so that measures aggregated over skills correspond to those of
Hansen & I_mrohoroglu (1992). Using 1993 data from the Statistical Abstract of the United
States,6 we divide the US labor force into four groups according to educational attainment.
The rst or high skill group (college graduates, HS) represents 27% of the labor force, the
second (some college but no degree, MS) 26%, High School graduates (mS) 35%, and people
without High School degree (LS) 12%. Mean monthly income for the four groups ranges
from $856 to $2,905.7 Unemployment rates consistently dier between the four categories
over a 10{year period: they range, on average over the period, from 2.47% for the HS group
to 10.65% for the LS group.6
When it comes to dierences in the duration of unemployment, Nickell (1979) provides
evidence that unemployment duration is fairly similar accross educational groups in the
United Kingdom. Mincer (1991, 1994) corroborates this nding for more recent United States
data. We therefore assume similar durations accross groups and follow Hansen & I_mrohoroglu
(1992) who estimate at 12 weeks the average time without a job oer. Transition probabilities
are determined using theu nemployment rates and duration. Normalizing aggregate income
6 Based on data from the Current Population Survey (US Bureau of the Census).
7 Based on the Survey of Income and Program Participation for 1990 (US Bureau of the Census).

5
to 1, the unemployment rate to 6% (to be consistent with Hansen & I_mrohoroglu (1992))
and choosing 6 weeks as the period length, we have the statistics summarized in Table 1.
We consider the following utility function:
 1 
ij ij cij 1t  lij t 1
U (ct ; lt ) = 1 
with  = 0:67 and  = 2:5 as in Hansen & I_mrohoroglu (1992) and Kydland & Prescott
(1982). The discount factor  is set to 0.995. It is very dicult to assess the extent of moral
hazard empirically. We therefore perform our computations for a wide set of (:){values.
To solve the model, for given replacement ratio () and levels of moral hazard [(0) and
(1)], we iterate on the agents value function over a grid of state variables. Our asset space
is composed of 301 points between 0 and an upper bound that is never chosen by agents
in our computations. We x this upper bound at 8 as in Hansen & I_mrohoroglu (1992).
Other state variables include the skill group, the current job oer status and previous period
employment. Since there is no transition from one skill group to another, we can analyze
each group separately thereby reducing the dimensions of our state space. Once the value
functions and the corresponding optimal decision rules are found, the invariant distribution
of the agents is obtained iteratively. This is performed with an initial guess on the tax rate,
which is then veried given the decision rules and the invariant distribution of each skill
group. When the tax rate balances the government budget, we have reached an equilibrium.
To assess the desirability of an equilibrium, we ask the agents in our economy to vote on ,
given a certain exogeneous level of moral hazard.8 Simple majority rule is applied to select
an equilibrium. A replacement ratio ~ that survives all alternatives is our optimal voting
level of benets. We contrast the results of the votes with the optimal level of benets ^
from the perspective of a planner maximizing a social welfare function with identical weight
on each individual agent's utility (average utility criterion), which is used in the literature.
The votes we consider involve steady{state to steady{state comparisons. Hence agents are
really asked under what system they would rather be born. The dynamics from one system to
another are not taken into account by the voters. In that sense, these are really \helicopter{
drop" votes. Krusell & Ros-Rull (1993) do present dynamic voting equilibria for an optimal
taxation problem. We restrict ourselves to steady{state votes, however, to ensure tractable
computations.
Two types of UI systems will be made available. On the one hand, a purely public system,
cross{nanced by the various skill groups. On the other hand, a private system, nanced
within each skill group and in which the insurer is able to discriminate by the skill level.
Agents will be asked to vote on the optimal level of benets within a system, be it private or
public, but also, in a separate vote, on whether they would rather switch to the alternative
system.
8 Note that there is a one{to{one relationship between the replacement ratio and the tax rate. Hence
voters choose along a single dimension.

6
4 Results
In this section, we present our results. Each subsection deals with one question.

4.1 Homogeneous skills: the benchmark

In an economy where agents do not dier in their skills, what UI program would they elect
to run?
Our model economy, when aggregated over skills, is exactly that of Hansen & I_mrohoroglu
(1992). We will take it as our benchmark for the purpose of comparison.
Depending on the level of moral hazard on searchers, measured as (0), agents vote for
dierent levels of benets ~. In absence of moral hazard, the median voter chooses ~ = :65
(see Table 2). When 10% of unemployed individuals rejecting oers can go undetected
[(0) = :10], the majority of agents still vote for the same replacement ratio. For higher
levels of moral hazard, the generosity is gradually reduced.
In Table 2, we also compare the choices of the median voter with the optimal replacement
ratios ^ that would emerge from the average utility criterion, used by Hansen & I_mrohoroglu
(1992).9 For low levels of moral hazard [(0)  :10], the median voter would choose lower
generosity than the social planner, but for (0)  :15, a more generous UI would be picked.
Moral hazard exerts a much stronger in uence on the optimal benets when average utility
is maximized, as can be seen from the very sharp drop in ^ for (0) between .15 and .20. The
votes essentially atten the functional ((0)), with the consequence that no sharp decline in
benets is observed passed a certain level of moral hazard, in contrast to the average utility
criterion.
Voting stresses the in uence of shirking on the median voter. With higher levels of moral
hazard, her utility is aected only through the higher tax rate needed to nance a UI system
distributing benets to more people. When using the average utility criterion, the lower em-
ployment due to shirking translates into lower overall income and consumption and reduces
overall utility sharply.
When searchers are perfectly monitored, but potential quitters are not, moral hazard matters
more in the benchmark than it does in the case of searchers. In Table 3, when the success
rate of shirking for quitters (1) goes from 0 to 10%, the replacement ratio chosen by the
median voter drops from 65% to 45%, while it was unaected in the case of searchers. As
9 Some of the results we present in Table 2 dier from those of Hansen & I_ mrohoroglu (1992). The reason
for this lies in the atness of the average utility function around the optimum. As a matter of fact, for an
agent to be indierent between a replacement ratio of .65 and the optimum (.75) only requires an increase in
consumption smaller than .025%. Consequently, a small change in the tax rate may have a dramatic impact
on the optimum. Determining the exact tax rate to balance the budget is therefore essential, but requires
extensive computing technology. Current technology allows us to be more precise about the convergence
criterion, which explains the dierences between our benchmark and the results of Hansen & I_mrohoroglu
(1992). Note that we replicate Hansen & I_mrohoroglu (1992) with their tax rate.

7
moral hazard is increased, ~ drops but less sharply than ^ and remains more generous for
high moral hazard levels. As quitters are more likely than unemployed agents to receive job
oers, they have a higher propensity to forego an oer, especially as they may receive UI
benets if monitoring is not perfect. Finally, the fact that the pool of potential quitters is
much larger than the pool of potential searchers weighs heavily in the welfare cost of moral
hazard.

4.2 A public UI in an economy with heterogeneous skills

If agents dier in skills, would the median voter alter her vote regarding the generosity of
the publicly nanced unemployment insurance? In other words, would anything change from
the benchmark outcome if agents do not have the same chances of nding a job or the same
income prospects?
We answer this question via Experiment 1: we compute the optimal voting replacement
ratios for all levels of moral hazard, when the four groups dier in both unemployment
probability and income.
The answer is `no', as is clear from Table 2: ~ is identical to the benchmark's for almost all
values of (0) (case of searchers). Skill types do not seem to matter for the outcome of the
vote. Indeed, the median voter appears to be quite similar to the one in the benchmark.
Notice, however, in Table 2, the distinction between actual votes and the choices ofa social
planner maximizing av erage utility. The latter would settle for very dierent replacement
ratios ^, with an emphasis on very generous benets for low levels of moral hazard. The case
of no moral hazard [(0) = 0] is strikingly more generous. Indeed, the insurance is perfect
[^ = 1] despite the fact that the unemployed individuals do not need full compensation
of income as they enjoy leisure in our model. If we allowed benets to go above 100% of
labor income in each group, the optimum, from the planner's point of view, would be at the
striking value of 305%. As can be seen in Figure 1, the welfare gains for the LS and mS
groups of a generous UI heavily outweigh the welfare losses for the MS and HS groups. This
is due to the concavity of the utility function. Of course, in the presence of moral hazard,
such a situation cannot happen because of the incentives to shirk.
Needless to say, there is a danger in maximizing average utility since the marginal utility of
the lower skilled agents dominates those of higher skilled agents for all or most values of the
replacement ratio.
In the case of quitters, as in the benchmark, moral hazard matters more (Table 3) than it
does for searchers. Indeed, when (1) goes from 0 to .10, the replacement ratio chosen by
the median voter drops from 65% to 45%. For similar reasons as in the benchmark, the drop
is sharper when the average utility criterion is used.
Interestingly, when 20% of quitters can collect benets [(1) = :20], the vote ping{pongs
between a replacement ratio of 25% and one of 35%. Under a status quo of  = :25, a very
feeble majority votes in favor of switching to  = :35: the alternative passes by a vote of
8
50.65%. In the reverse scenario where the status quo is  = :35, yet another feeble majority
(51.47% of voters) chooses to return to  = :25. Such ping{pong voting is due to a change
in the distribution of agents between both status quos.10 Those who make the dierence
in the ballot are the medium skilled (MS) quitters, who, since they quit do not mind a
generous insurance, and those who are genuinely without oers. As another peculiar voting
behavior, we even observe an increase in generosity when moral hazard on quitters goes from
(1) = :25 to (1) = :30. It turns out that when (1) = :25, voters are almost undecided
between replacement ratios of 20% and 30%. At a status quo of  = :20, the oer to switch
to  = :30 is rejected by a margin of 1.5%. Under a status quo of  = :30, agents accept to
switch to  = :20 by a margin of 1.36%.
In the rest of the paper, we will focus on searchers since one might argue it is much easier
to monitor quitters than it is to monitor searchers.

4.3 A private UI
If instead of a public, cross{nanced UI system, each skill group were to organize and nance
its own unemployment insurance program, for what replacement ratio would its members
vote?
To answer this question, we consider an economy with the same heterogeneity in skills as
above, but deprived of redistributive means across groups. In our Experiment 2, each skill
group runs its own UI program that balances its budget. Results are presented at the bottom
of Table 2.
Each skill group chooses a generosity similar to that of the public system, at least for low
levels of moral hazard. Of course, tax rates dier widely from the public system and take
their toll on generosity for higher levels of moral hazard, especially for groups that have
high unemployment incidence. From the results in Table 2, it appears that, regardless of the
moral hazard level, the tax rate of the high skill [HS] agents in Experiment 1 exceeds their
tax rate under the private system, while the opposite is true for the other skill groups. Since
the median voter belongs to the latter, it is not surprising to observe a high generosity of
the public UI program.
For low levels of moral hazard on searchers [(0)  :15], all skill groups vote for the same
benets ~ than a social planner maximizing the average utility within each skill group. For
higher levels of moral hazard, the median voter in virtually all skill groups chooses more
generous replacement ratios than her group{specic social planner would pick.

10Taking transitions into account in the voting process would probably lead to a single and stable optimal
replacement ratio.

9
4.4 Going private? Going public?
Next, we submit to popular vote the following proposition. Given the success rate of shirking
in the economy and a status quo consisting of the public UI program (Experiment 1 with ~
chosen in Table 2), would you consider switching to a privately run UI system (Experiment
2)?
For each level of (0), we successively oer voters alternative replacement ratios ranging
from 0 to 1. As appears in the second column of Table 4, the median voter, however, is
not eager to switch. Only in two instances would the private insurance be chosen: for
(0) = :15, the median voter accepts to switch provided her group specic private insurance
has a replacement ratio of 60%; similarly for (0) = :50, switching would happen if the
median voter obtains a replacement ratio of 15%. In the latter case, Table 2 provides the
explanation for this vote: with a public system, all groups are guaranteed a ratio  = :15.
The tax rate is then 1.12%. In the private insurance program, the MS group self{nances
the same replacement ratio with a tax rate of 1.11%. The slight dierence in tax rates is
sucient for the MS group to choose to go private. Detailed analysis of the votes would
reveal that the HS group is willing to switch for almost any private insurance program, while
LS and mS groups vote as a block against it, regardless of its generosity.
In the reverse scenario under which the status quo is a private UI system, where all groups
have settled for their optimal ~ of Experiment 2, would a majority of agents arise in favor
of a public insurance?
The third column of Table 4 answers this question. For all levels of moral hazard other than
15% and 50%, the switch from private to public UI programs rallies a majority of votes. We
never observe \ping{pong" voting as agents never consider \switching back" to the program
they chose to quit. Detailed analysis would show that, consistently with the above results,
high skill agents are categorically opposed to any public alternative.

4.5 Counting the votes

Votes by members of a skill group are essentially \bang{bang": agents vote almost always as
a block, whether they are unemployed or not. This result suggests that time consistent votes
can happen when skill group members share a suciently strong dierence with members of
other groups. In this case, they have very similar objectives compared to the other groups.
In Figures 2-5, we illustrate this phenomenon.
As we can see from the gures, the LS group tends to vote as a block. So does the mS group
in absence of moral hazard. In Wright (1986)'s framework, even if agents were identical,
except for their probability of nding employment, those at work were more likely to vote
dierently than those who were unemployed, the lower the discount factor of agents. In
our economy, the discount factor is very close to 1, and agents dier not only in their
probability of employment, but also in their income prospects in case of employment. As
a consequence, groups act as a whole, except maybe for the HS group, whose unemployed
10
members sometimes favor a more generous system than the members currently employed.

4.6 The impact of moral hazard on employment

Moral hazard raises a natural question: Among the skill groups in this model, does anyone
choose to reject oers more often than the others?
Remember that moral hazard is not the only reason why one may choose torefuse an oer
in this model . An individual with high asset holdings may nd it preferable to enjoy leisure
rather than work. Higher replacement ratios mean higher tax rates, which make leisure
less costly. Beyond a certain threshold, some people choose not to work. For Experiment
1 without moral hazard, Figure 6 (North-Western corner) shows that all skill groups start
refusing oers at the same replacement ratio. However if all groups appear to have the same
threshold, that does not mean they stop working at the same rate. In fact, the lower the skill
level, the higher the propensity to refuse jobs (as appears from the slopes of the employment
curves). With high replacement ratios, the low skill group attains a certain level of assets
much more easily than with low replacement ratios. Since LS agents have a low expected
level of assets, they tend to use them sooner and drop out of labor.
When some moral hazard is introduced [(0) = :1] in Experiment 1, the same holds, with
a tendency for low skill agents to start shirking at lower replacement ratios than the other
groups (Figure 6, North-East). Indeed, they are the ones beneting most from the UI
program | they contribute little and receive large benets | while the others are not hurt
too much. However, for higher levels of moral hazard, no group seems to shirk more than
any other. See Figure 6 (South corners) for (0) = :3 and .5: For such levels of moral hazard,
no group can resist the temptation of shirking.

5 Conclusion
In the extant literature, moral hazard is reported to exert a large in uence on the optimal
unemployment insurance. We show that this is not necessarily the case. Acknowledging the
dierences in skills within the United States labor force, which translate into income and
unemployment rate disparities, we ask agents to vote for unemployment insurance benets.
We do so within a dynamic general equilibrium framework with a population heterogene-
ous along several dimensions: asset holdings, current and past employment status, income
prospects and employment transition probabilities. As we increase the level of moral ha-
zard in the economy, the median voter does not choose to cut benets as drastically as was
previously reported.
When oered the possibility to switch to a privately run insurance, nanced within each skill
group, voters in majority prefer to stick to their public, cross{nanced UI program for most
levels of moral hazard. If the status quo were a private UI system, the public alternative
would rally a majority of votes for most levels of shirking.
11
In contrast to Wright (1986), our voting equilibria suggest that votes within a skill group
can be time consistent when groups dier suciently strongly. Whether unemployed or not,
agents within skill groups tend to vote as a block.
Hansen & I_mrohoroglu (1992) suggested that when a small proportion of people could refuse
oers and still collect unemployment insurance benets, the optimal replacement ratio had
to be extremely small. Our ndings do not support such drastic conclusions. It takes quite
a big proportion of abusers to make large cuts in benets optimal.

12
References

Acemoglu, D. & Shimer, R. (1997). Ecient unemployment insurance. Manuscript. MIT-Princeton.

Andolfatto, D. & Gomme, P. (1996). Unemployment insurance and labor market activity in Canada.
Carnegie-Rochester Conference Series on Public Policy, 44.
Baily, M. (1978). Some aspects of optimal unemployment insurance. Journal of Public Economics, 10,
379{402.
Gomes, J., Greenwood, J., & Rebelo, S. (1997). Equilibrium unemployment. Manuscript. University of
Rochester.
Greenwood, J. & Human, G. W. (1987). A dynamic equilibrium model of in ation and unemployment.
Journal of Monetary Economics, 19, 203{228.
Gruber, J. (1994). The consumption smoothing benets of unemployment insurance. NBER Working Paper
#4750.
Hansen, G. D. (1985). Indivisible labor and the business cycle. Journal of Monetary Economics, 16, 309{18.
Hansen, G. D. & I_ mrohoroglu, A. (1992). The role of unemployment insurance in an economy with liquidity
constraints and moral hazard. Journal of Political Economy, 100 (1), 118{142.
Hassler, J. & Rodrguez Mora, J. V. (1997). Employment turnover and the public allocation of unemployment
insurance. Working Paper. Institute for International Economic Studies. Stockholm University.
Hopenhayn, H. A. & Nicolini, J. (1997). Optimal unemployment insurance. Journal of Political Economy,
105 (2), 412{438.
Krusell, P. & Ros-Rull, J.-V. (1993). Distribution, redistribution, and capital accumulation. Mimeo, Uni-
versity of Pennsylvania.
Kydland, F. E. & Prescott, E. C. (1982). Time to build and aggregate uctuations. Econometrica, 50 (6),
1345{1370.
Mincer, J. (1991). Education and unemployment. NBER Working Paper No 3838.
Mincer, J. (1994). Human capital: a review. In Labor Economics and Industrial Relations: Markets and
Institutions, ed. by Clark Kerr and Paul D. Staudohar, Harvard University Press: Cambridge.
Nickell, S. (1979). Education and lifetime patterns of unemployment. Journal of Political Economy, 87 (5),
117{131.
Rogerson, R. (1988). Indivisible labor, lotteries and equilibrium. Journal of Monetary Economics, 21 (3-16).
Rogerson, R. & Wright, R. (1988). Involuntary unemployment in economies with ecient risk sharing.
Journal of Monetary Economics, 22 (501-515).
Shavell, S. & Weiss, L. (1979). The optimal payment of unemployment insurance benets over time. Journal
of Political Economy, 87, 1347{1362.
Valdivia, V. H. (1996). Evaluating the welfare benets of unemployment insurance. Manuscript. North-
western University.
Wang, C. & Williamson, S. (1996). Unemployment insurance with moral hazard in a dynamic economy.
Carnegie-Rochester Conference Series on Public Policy, 44, 1{41.
Wright, R. D. (1986). The redistributive roles of unemployment insurance and the dynamics of voting.
Journal of Public Economics, 31, 377{399.
Zhang, G. (1995). Unemployment insurance analysis in a search economy. Manuscript.

13
 Table 1: Calibration
Skill group HS MS mS LS aggregate
measure .2691 .2606 .3514 .1189 1
income 1.6220 .8950 .7577 .4779 1
unempl. rate .0280 .0518 .0696 .1209 .06
Transition probabilities
HS MS mS LS aggregate
p(eje) .9856 .9727 .9623 .9224 .9681
p(uje) .0144 .0273 .0374 .0776 .0319
p(uju) .5000 .5000 .5000 .5000 .5000
p(eju) .5000 .5000 .5000 .5000 .5000
Note: This table summarizes the key elements of our calibration. All parameters are chosen so that on aggregate, we have the
same calibration as Hansen & I_ mrohoroglu (1992).

Table 2: Moral hazard on searchers only

Moral hazard (0) 0 .10 .15 .20 .25 .30 .50 1
Benchmark ~ .65 .65 .60 .40 .30 .25 .15 .05
~ .0398 .0398 .0369 .0292 .0221 .0190 .0130 .0040
^ .75 .70 .55 .25 .20 .15 .10 0
Experiment 1 ~ .65 .65 .55 .40 .35 .25 .15 .05
~ .0336 .0336 .0290 .0247 .0241 .0168 .0112 .0034
^ 1 .80 .60 .45 .35 .25 .15 .05
Experiment 2 ~HS .65 .65 .60 .30-.35 .25 .25 .15 .05
~HS .0184 .0184 .0170 .0086-.0116 .0086 .0110 .0078 .0021
~MS .60 .65 .60 .40 .30 .25 .15 .05
~MS .0317 .0343 .0317 .0256 .0196 .0169 .0111 .0037
~mS .65 .65 .55 .40 .35 .25 .15 .05
~mS .0464 .0464 .0395 .0340 .0312 .0216 .0140 .0046
~LS .65 .65 .35 .25 .30 .25 .15-.20 .10
~LS .0916 .0916 .0515 .0374 .0490 .0408 .0248-.0358 .0192
^HS .65 .65 .60 .35 .20 .15 .10 0
^MS .65 .65 .60 .25 .20 .15 .10 .05
^mS .65 .65 .55 .25 .20 .20 .10 .05
^LS .65 .65 .35 .25 .20 .15 .10 .05
Note: This table presents the optimal voting replacement ratio and tax rates (~ and ~) as a function of the level of moral hazard
on searchers (0), for respectively, an economy with homogeneous skills (the benchmark), an economy with heterogeneous skills
under a public UI system (Experiment 1) and a private discriminating UI system (Experiment 2). For each experiment, the
optimal replacement ratio from average utility maximization (^) is also displayed.

14
 Table 3: Moral hazard on quitters only
Moral hazard (1) 0 .10 .15 .20 .25 .30 .50
Benchmark ~ .65 .45 .40 .30 .25 .20 .10
^ .75 .35 .25 .15 .15 .10 0
Experiment 1 ~ .65 .45 .35 f.25, .35g .20 .25 .10
^ 1 .55 .45 .20 .15 .15 .05
Note: This table is similar to Table 2, but for moral hazard on quitters only ((1)  0 and (0) = 0).

Table 4: To switch or not to switch

(0) Private vs. Public Public vs. Private
0 Public Public:  2 [:60; :85]
.10 Public Public:  2 [:60; :70]
.15 Private:  = :60 Private
.20 Public Public:  2 [:30; :40]
.25 Public Public:  2 [:30; :35]
.30 Public Public:  = :25
.50 Private:  = :15 Private
1 Public Public:  = :05

Note: This table presents, for every level of moral hazard on searchers (0), the results of the votes on whether to switch from
a public to a private UI system (2nd column) or alternatively to switch from a private to a public UI program (3rd column).
Status quos are characterized by the optimal replacement ratios ~ of the corresponding UI program in Table 2. Switches have
been highlighted.

15
 Figure 1:
Experiment 1, no moral hazard
−1

−1.005 MS

−1.01
Normalized average utility

−1.015 HS LS

−1.02 mS

−1.025

−1.03

−1.035
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Replacement ratio

In this gure we have normalized average utilities so that -1 is the maximum in each
group.
Figure 2:
Experiment 1, Status quo: θ=.20, π(0)=π(1)=0
1

0.9
HS
0.8

0.7
Proportion of votes in favor

0.6
MS
0.5

0.4

0.3
mS
0.2

0.1
LS
0
0 0.2 0.4 0.6 0.8 1
Alternative θ
Note: Given the status quo above, this gure displays the proportion of voters in each
group who accept an alternative replacement ratio. HS, MS, mS and LS groups are
respectively of measure: 26.91%, 26.06%, 35.14% and 11.89%.

16
 Figure 3:
Experiment 1, Status quo: θ=.40, π(0)=π(1)=0
1

0.9

0.8
HS
0.7
Proportion of votes in favor

0.6
MS
0.5

0.4

0.3
mS
0.2

0.1
LS
0
0 0.2 0.4 0.6 0.8 1
Alternative θ
Note: See note of Figure 2.

Figure 4:
Experiment 1, Status quo: θ=.65, π(0)=π(1)=0
1

0.9

0.8

0.7
Proportion of votes in favor

0.6

0.5 MS

0.4

HS
0.3
mS
0.2

0.1
LS
0
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
Alternative θ
Note: See note of Figure 2.

17
 Figure 5:
Experiment 1, Status quo: θ=.65, π(0)=.1, π(1)=0
1

0.9

0.8

0.7
Proportion of votes in favor

0.6

0.5 MS

0.4

HS
0.3
mS
0.2

0.1
LS
0
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
Alternative θ
Note: See note of Figure 2.

Figure 6: Experiment 1, Employment vs. replacement ratio

pi(0)=0 pi(0)=.1
1 1

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6
0 0.5 1 0 0.5 1

pi(0)=.3 pi(0)=.5
1 1

0.8 0.8

0.6 0.6

0.4 0.4
0 0.2 0.4 0.6 0.8 0 0.5 1

For a given level of moral hazard, each gure plots employment participation against
replacement ratio for all skill groups in Experiment 1. The curves from top to bottom
represent HS, MS, mS and LS employment.

18