CHAPTER 5

Search, matching, and unemployment

55
56 5. SEARCH, MATCHING, AND UNEMPLOYMENT

Exercise 5.1. Being unemployed with only a chance of an offer

An unemployed worker samples wage offers on the following terms. Each period,
with probability φ, 1 > φ > 0, she receives no offer (we may regard this as a
wage offer of zero forever). With probability (1 − φ) she receives an offer to work
for w forever, where w is drawn from a cumulative distribution function F (w).
Successive drawings across periods are independently and identically distributed.
The worker chooses a strategy to maximize
!∞
E β t yt , where 0 < β < 1,
t=0

yt = w is the worker is employed, and yt = c is the worker is unemployed. Here c is

unemployment compensation, and w is the wage at which the worker is employed.
Assume that, having once accepted a job offer at wage w, the worker stays in the
job forever. "
Let v(w) be the expected value of ∞ t
t=0 β yt for an unemployed worker who has
offer w in hand and who behaves optimally. Write Bellman’s functional equation
for the worker’s problem.

Solution
"
Let v(w) be the expected value of ∞ t
t=0 β yt for an unemployed worker who has
offer w in hand and who behaves optimally.
# $ %
w ′ ′
(27) v(w) = max , c + φβv(0) + (1 − φ)β v(w )dF (w ) .
A,R 1−β
Here the maximization is over the two actions: accept the offer to work forever
at wage w, or reject the current offer and take a chance on drawing a new offer
next period.

Exercise 5.2. Two offers per period

Consider an unemployed worker who each period can draw two independently and
identically distributed wage offers from the cumulative probability distribution
function F (w). The worker will work forever at the same wage after having once
accepted an offer. In the event of unemployment during a period, the worker
receives unemployment
" compensation c. The worker derives a decision rule to
maximize E ∞ t=0 β t
y t , where yt = w or yt = c, depending
"∞ on whether she is
t
employed or unemployed. Let v(w) be the value of E t=0 β yt for a currently
unemployed worker who has best offer w in hand.
a. Formulate Bellman’s equation for the worker’s problem.
b. Prove that the worker’s reservation wage is higher than it would be had
the worker faced the same c and been drawing only one offer from the same
distribution F (w) each period.

Solution
 5. SEARCH, MATCHING, AND UNEMPLOYMENT 57

a. Note that the event max{w1 , w2 } < w is the event (w1 < w) ∩ (w2 < w).
Therefore prob{max(w1 , w2 ) < w} = F (w)2 . The worker will evidently limit his
choice to the larger of the two offers each period. Bellman’s equation is therefore
# $ %
w ′ 2 ′
v(w) = max , c + β v(w )d(F )(w ) ,
1−β
where w is the best offer in hand.
b. The reservation wage obeys the following equation:
$ ∞
β
(w̄2 − c) = (w′ − w̄2 )d(F 2 )(w′ ).
1 − β w̄2
Using the usual integration by part argument, one obtains the equation:
$ B
h2 (w̄2 ) ≡ (1 − β)w̄2 − β (1 − F (w ′ )2 )dw′ = 0.
w̄2
Observe that h2 is an increasing function. When the worker is given only one
offer, the reservation wage solves :
$ B
h1 (w̄1 ) ≡ (1 − β)w̄1 − β (1 − F (w ′ ))dw′ = 0.
w̄1
Since F (w)2 ≤ F (w), we have h2 (w) ≤ h1 (w). Therefore:

0 = h1 (w̄1 ) = h2 (w̄2 ) ≤ h1 (w̄2 ).

Since h2 is increasing if follows that

w̄1 ≤ w̄2 .
The intuition underlying this result is as follows: the worker could choose always
to ignore the second offer. This policy, possibly suboptimal, would leave the
worker with a decision problem that is formally identical to the standard one-
offer problem. The value of the objective function of the true problem is at least
as high as the value of the objective function under the artificially restricted
problem. Because the reservation wage has the property of equating the value
of accepting a job, w/(1 − β), with the value of rejecting, c + βEv(w ′ ), a higher
value of Ev(w ′ ), which results in the two-offer case, requires a higher reservation
wage.

Exercise 5.3. A random number of offers per period

An unemployed worker is confronted with a random number, n, of job offers
each period. With probability π"n ,∞the worker receives n offers in a given period,
where πn ≥ 0 for n ≥ 1, and n=1 πn = 1 for N < +∞. Each offer is drawn
independently from the same distribution F (w). Assume that the number of
offers n is independently distributed across time. The worker works forever at
wage w after having accepted a job and receives unemployment compensation
58 5. SEARCH, MATCHING, AND UNEMPLOYMENT

of "
c during each period of unemployment. He chooses a strategy to maximize
E ∞ t
t=0 β yt where yt = c if he is unemployed, yt = w if he is employed.
Let v(w) be the value of the objective function of an unemployed worker who has
best offer w in hand and who proceeds optimally. Formulate Bellman’s equation
for this worker.

Solution
& N $ '
w !
v(w) = max ,c + πn v(w′ )d(F n )(w′ ) .
1−β n=1
In effect, the worker is confronted with a lottery with probabilities π n over dis-
tributions F n (w), from which he will sample next period. As in Exercise 2.1, w
is the highest offer in hand.

Exercise 5.4. Cyclical fluctuations in number of job offers

Modify Exercise 5.3 as follows. Let the number of job offers n follow a Markov
process, with

prob {number of offers next period = m|number of offers this

(28) period = n} = πmn , m = 1, . . . , N, n = 1, . . . , N
"N
m=1 πmn = 1 for n = 1, . . . , N.
Here [πmn ] is a “stochastic matrix” generating a Markov chain. Keep all other
features of the problem as in Exercise 2.3. The worker gets n offers per period,
where n is now generated by a Markov chain so that thenumber of offers is possibly
correlated over time.
"
a. Let v(w, n) be the value of E ∞ t
t=0 β yt for an unemployed worker who has
received n offers this period, the best of which is w. Formulate Bellman’s equation
for the worker’s problem.
b. Show that the optimal policy is to set a reservation wage w̄(n) that depends
on the number of offers received this period.

Solution
a. The Bellman equation for the worker’s problem is
& N $ '
w !
(29) v(w, n) = max ,c + πm,n v(w′ , m)d(F m )(w′ ) .
accept,reject 1−β m=1

b. From equation (29), we see that the right branch of the right side of the
functional equation is evidently a function only of n. The argument in the text
applies for each n and implies a reservation wage that is a function of n.

Exercise 5.5. Choosing the number of offers

 5. SEARCH, MATCHING, AND UNEMPLOYMENT 59

An unemployed worker must choose the number of offers n to solicit. At a cost of

k(n) the worker receives n offers this period. Here k(n + 1) > k(n) for n ≥ 1. The
number of offers n must be chosen in advance at the beginning of the period
" and
cannot be revised during the period. The worker wants to maximize E ∞ t=0 β t
yt .
Here yt consists of w each period she is employed but not searching, [w − k(n)]
the first period she is employed but searches for n offers, and [c − k(n)] each
period she is unemployed but solicits and rejects n offers. The offers are each
independently drawn from F (w). The worker who accepts an offer works forever
at wage w.
Let Q be the value of the problem for an unemployed worker who has not yet
chosen the number of offers to solicit. Formulate Bellman’s equation for this
worker.
Solution
$ # %
w
Q = max max − k(n), −k(n) + βQ d(F n )(w).
n accept
reject
1−b
The worker proceeds sequentially each period, first choosing n, then deciding
whether to accept or reject the best offer.

Exercise 5.6. Mortensen externality

Two parties to a match (say, worker and firm) jointly draw a match parameter θ
from a c.d.f. F (θ). Once matched, they stay matched forever, each one deriving
a benefit of θ per period from the match. Each unmatched pair of agents can
influence the number of offers received in a period in the following way. The
worker receives n offers per period, with n = f (c1 + c2 ), where c1 is the resources
the worker devotes to searching and c2 is the resources the typical firm devotes
to searching. Symmetrically, the representative firm receives n offers per period
where n = f (c1 + c2 ). (We shall define the situation so that firms and workers
have the same reservation θ so that there is never unrequited love.) Both c1 and
c2 must be chosen at the beginning of the period, prior to searching during the
period. Firms and workers have the same preferences, given by the expected
present value of the match parameter θ, net of search costs. The discount factor
β is the same for worker and firm.
a. Consider a Nash equilibrium in which party i chooses ci , taking cj , j ̸= i, as
given. Let Qi be the value for an unmatched agent of type i before the level of ci
has been chosen. Formulate Bellman’s equation for agents of type 1 and 2.
b. Consider the social planning problem of choosing c1 and c2 sequentially so
as to maximize the criterion of λ times the utility of agent 1 plus (1 − λ) times
the utility of agent 2, 0 < λ < 1. Let Q(λ) be the value for this problem for
two unmatched agents before c1 and c2 have been chosen. Formulate Bellman’s
equation for this problem.
c. Comparing the results in (a) and (b), argue that, in the Nash equilibrium, the
optimal amount of resources has not been devoted to search.
60 5. SEARCH, MATCHING, AND UNEMPLOYMENT

Solution
a. $ # %
θ n
Q1 = max max − c1 , −c1 + βQ1 d(F )(θ) ,
c1 accept,reject 1−β
subject to n = f (c1 + c2 ), c2 given
$ # %
θ
Q2 = max max − c2 , −c2 + βQ2 d(F n )(θ)
c2 accept,reject 1−β
subject to n = f (c1 + c2 ), c1 given.
b.
() ( * +
θ θ
Q(λ) = maxc1 ,c2 maxaccept,reject λ 1−β − λc1 + (1 − λ) 1−β − c2 ,
'
,
−λc1 − (1 − λ)c2 + βQ(λ) d(F n )(θ)

subject to n = f (c1 + c2 ).
c. The Nash equilibrium is a (c1 , c2 ) pair that solves the two functional equations
in (a). In general, this (c1 , c2 ) pair will not solve the functional equation in (b)
because each agent in (a) neglects the effects of his choice of cj on the welfare of
the other agent. In general, there will be too little search in the Nash equilibrium
if f (c1 + c2 ) is increasing in (c1 + c2 ).

Exercise 5.7. Variable labor supply

An unemployed worker receives each period a wage offer w drawn from the distri-
bution F (w). The worker has to choose whether to accept the job – and therefore
to work forever – or to search for another offer and collect c in unemployment
compensation. The worker who decides to accept the job must choose the number
of hours to work in each period. The worker chooses a strategy to maximize
!∞
E β t u(yt , lt ), where 0 < β < 1,
t=0

and yt = c if the worker is unemployed, and yt = w(1 − lt ) if the worker is

employed and works (1 − lt ) hours; lt is leisure with 0 ≤ lt ≤ 1.
Analyze the worker’s problem. Argue that the optimal strategy has the reserva-
tion wage property. Show that the number of hours worked is the same in every
period.

Solution
Let s be the state variable. We choose s = (w, 0), where w is the wage offer and
0 = E if the worker is employed, and 0 = U if she is unemployed. Consider first
the situation of an employed worker. Bellman’s equation is
v(w, E) = max {u[w(1 − l), l] + βv(w, E)}.
l
 5. SEARCH, MATCHING, AND UNEMPLOYMENT 61

Then it follows that

u (w(1 − l(w)), l(w))
v(w, E) = ,
1−β
where l(w) ≡ argmax u(w(1 − l), l).
l
Let’s show that v(w, E) is increasing in w. Consider w1 < w2 . We have :

u (w1 (1 − l(w1 )), l(w1 )) ≤ u (w2 (1 − l(w1 )), l(w1 ))

≤ maxl u (w2 (1 − l), l)
≡ u (w2 (1 − l(w2 )), l(w2 )) .
Intuitively, a worker receiving w2 > w1 has the option work 1 − l(w1 ) hours paid
w2 , that yields a higher utility than working 1 − l(w1 ) hours paid w1 . Its optimal
choice 1 − l(w2 ) necessarily yields an even higher utility.

Now consider an unemployed worker. Bellman’s equation is

# $ %
′ ′
v(w, U ) = max V (w, E), u(c, 1) + β v(w , U )dF (w ) .
accept,reject

The outside maximization is over two actions: accept the offer (in which case the
worker chooses l optimally) or reject the offer, collect unemployment compensa-
tion, and wait for a new offer next period. The first term is incresing in w and
the second is independent of w. Therefore the optimal policy is to accept offers
offers that are at least equal to some w̄. Once an offer has been accepted, hours
worked are constant and equal to l(w).

Exercise 5.8. Wage growth rate and the reservation wage

An unemployed worker receives each period an offer to work for wage wt forever,
where wt = w in the first period and wt = φt w after t periods in the job. Assume
φ > 1, that is, wages increase with tenure. The initial wage offer is drawn from
a distribution F (w) that is constant over time (entry-level wages are stationary);
successive drawings across periods are independently and identically distributed.
The worker’s objective function is to maximize
!∞
E β t yt , where 0 < β < 1,
t=0

and yt = wt if the worker is employed and yt = c if the worker is unemployed,

where c is unemployment compensation. Let v(w) be the optimal value of the
objective function for an unemployed worker who has offer w in hand. Write
Bellman’s equation for this problem. Argue that, if two economies differ only in
the growth rate of wages of employed workers, say φ1 > φ2 , the economy with
the higher growth rate has the smaller reservation wage.
Note. Assume that φi β < 1, i = 1, 2.

Solution
62 5. SEARCH, MATCHING, AND UNEMPLOYMENT

If the worker accepts employment at wage w, the sequence {yt } is given by yt =

w, yt+1 = φw . . . , yt+j ="φj w . . .. Therefore the value of the objective function
if the worker accepts is ∞ j
j=0 β yt+j = w/(1 − βφ). Bellman’s equation for the
worker’s problem is
# $ %
w ′ ′
v(w) = max , c + β v(w )dF (w ) .
1 − βφ
Using the same argument as when studying McCall’s model, one shows that that
the optimal policy is to accept all offers to work with an initial wage higher than
a reservation wage w̄.
# w̄
1−φβ
w ≤ w̄
v(w) = w
1−φβ
w ≥ w̄.
Because, at w = w̄, we have
$ B
w̄
=c+β v(w′ )dF (w ′ ),
1 − φβ 0

we get, after substituting for v(w) its expression,

$ w̄ $ B
w̄ c+β ′ β
= w̄ dF (w ) + w′ dF (w′ ).
(1 − φβ) 1 − φβ 0 1 − φβ w̄
This equation can be rearranged to give
$ B
(1 − β)w̄ − β (w′ − w̄)dF (w ′ ) = (1 − βφ)c.
w̄

It is easy to see (using the Leibniz rule) that the left-hand side is increasing in w̄.
Therefore, if φ1 > φ2 , that is, (1 − βφ1 )c < (1 − βφ2 )c, it must be that w̄1 < w̄2 .
The intuition behind this result is simple: for any given offer w, the value of
accepting the offer is higher, the higher the growth rate of wages φ. Therefore,
the sooner an offer is accepted, the sooner the benefits of the growth in wages
are realized. This pattern makes some job offers more attractive even though the
initial wage is not very high.

Exercise 5.9. Search with a finite horizon

Consider a worker who lives two periods. In each period the worker, if unem-
ployed, receives an offer of lifetime work at wage w, where w is drawn from a
distribution F . Wage offers are identically and independently distributed over
time. The worker’s objective is to maximize E{y1 + βy2 }, where yt = w if the
worker is employed and is equal to c – unemployment compensation – if the
worker is not employed.
Analyze the worker’s optimal decision rule. In particular, establish that the
optimal strategy is to choose a reservation wage in each period and to accept
any offer with a wage at least as high as the reservation wage and to reject offers
below that level. Show that the reservation wage decreases over time.
 5. SEARCH, MATCHING, AND UNEMPLOYMENT 63

Solution
We first analyze the worker’s problem in the second period of life. We consider
an unemployed worker; an employed worker does not have to solve any decision
problem. Let v2 (w) be the optimal value of the problem for an unemployed
worker with offer w in hand. Then v2 (w) = max{w, c}. It follows that the
optimal strategy is to accept offers that are at least c and to reject all others.
The second-period reservation wage, w̄2 , is equal to c. In the first period if the
worker is faced with a wage w and accepts the offer, the value of the objective
function is w(1 + β). If the worker rejects he gets c in the first period and v 2 (w′ ),
a random variable,) in the following period. The expected value of rejecting the
∞
offer is thus c + β 0 v2 (w′ )dF (w ′ ).
Therefore the optimal value of the objective function for a worker with offer w in
hand is given by
# $ B %
′ ′
v1 (w) = max w(1 + β), c + β v2 (w )dF (w ) .
0

Notice that the second term in brackets is constant, whereas the first is increasing
in w. It follows that the optimal policy is of the reservation wage form. There
exists a w̄1 such that, for w ≤ w̄1 , the second term is higher, and therefore the
optimal strategy is to reject the job offer and to remain unemployed. Similarly,
when w > w̄1 , the first term is higher and the optimal strategy is to accept the
job. As usual w̄1 satisfies :
$ B
w1 (1 + β) = c + β v2 (w′ )dF (w ′ ).
0

Observe that v2 (w) = max{w, c} ≥ c. In words, a worker who is unemployed

in
) B the second period get at least c, the unemployement compensation. Thus
0
v2 (w )dF (w ′ ) = E(v2 (w′ )) ≥ c, with a strict inequality if Pr(w ′ ≥ c) > 0.
′

This inequality implies :

$ B
w1 (1 + β) = c + β v2 (w′ )dF (w ′ ) ≥ (1 + β)c.
0
The reservation wage decreases as the retirement date approaches. The intuition
underlying this result is that, the shorter the horizon, the smaller the benefits
of “waiting to see if next period the wage offer is really high” (the option value
of waiting) because those benefits cannot be enjoyed for a long period. The
implication is hence that the alternative to waiting – that is, accepting a job –
becomes more attractive. This aspect is reflected in the model by a decrease in
the reservation wage, which in fact corresponds to an increase in the percentage
of job offers that are accepted.

Exercise 5.10. Finite horizon and mean-preserving spread

64 5. SEARCH, MATCHING, AND UNEMPLOYMENT

Consider a worker who draws every period a job offer to work forever at wage
w. Successive offers are independently and identically distributed drawings from
a distribution Fi (w), i = 1, 2. Assume that F1 has been obtained from F2 by a
mean-preserving spread (see Section 2.4). The worker’s objective is to maximize
T
!
E β t yt , 0 < β < 1,
t=0

where yt = w is the worker has accepted employment at wage w and is zero

otherwise. Assume that both distributions, F1 and F2 , share a common upper
bound, B.
a. Show that the reservation wages of workers drawing from F1 and F2 coincide
at t = T and t = T − 1.
b. Argue that for t ≤ T − 2 the reservation wage of the workers that sample
wage offers from the distribution F1 is higher than the reservation wage of the
workers that sample from F2 . c. Now introduce unemployment compensation:
the worker who is unemployed collects c dollars. Prove that the result in (a) no
longer holds, that is, the reservation wage of the workers that sample from F 1 is
higher than the one corresponding to workers that sample from F2 for t = T − 1.

Solution
a. Let vti (w) be the optimal value of the objective function of an unemployed
worker at time t who has offer w in hand and draws wage offers from the distri-
bution Fi , i = 1, 2. Then it is clear that vTi (w) = max{0, w} = w. Therefore
)B i )B
0
vT (w)dFi (w) = 0 wdFi (w) = Ew, i = 1, 2. Clearly the reservation wage
at time T is zero: the worker accepts every offer. At time (T − 1), Bellmans’
equation for the worker’s problem is
( )B i ′ ,
i ′
vT −1 (w) = max w(1 + β), β 0 vT (w )Fi (dw )
max {w(1 + β), βEw}.
It is then clear that the worker will accept the offer if w(1 + β) ≥ βEw and will
reject it otherwise. Therefore the reservation wage w̄T −1 is βEw/(1+β). Because
the expectation of w is the same no matter whether w is drawn from F1 or F2 , it
follows that both types of workers have the same reservation wage.
b. We prove this point by induction. Assume that at t+1 the optimal policy under
both distribution is of the reservation wage form. Also, assume that wt+1 (1), the
reservation wage under c.d.f. F1 , is greater than wt+1 (2), the reservation wage
under c.d.f. F2 . Observe that those two assumptions are true at time T . The
Bellman equation at time t is:
# $ B %
1 − β T −t+1 ′ ′
vti (w) = max w ,β i
vt+1 (w )dFi (w ) ,
1−β 0
 5. SEARCH, MATCHING, AND UNEMPLOYMENT 65

T −t+1
where w 1−β1−β is the value of working at wage w in periods t, t + 1, . . . T . The
first term is increasing in w while the second one is constant. It follows that, at
time t, the optimal policy is also of the reservation wage form. Furthermore, the
time t reservation wage wt (i) solves the usual indifference condition:

T −t+1 )B
wt (i) 1−β1−β = β v i (w′ )dFi (w′ )
T −t+1 )0wt+1t+1
(i) 1−β T −t )B 1−β T −t ′
wt (i) 1−β1−β = β 0 w t+1 (i)dF i (w ′
) + β w dFi (w′ )
T −t+1 T −t
* )
1−β w t+1 (i)
)
1−β +
wt+1 (i) B ′
wt (i) 1−β1−β = β 1−β
1−β 0
(w t+1 (i) − w ′
) dF i (w ′
) + 0
w dF i (w ′
) .
Integrating the first term by part and rearanging yields :
-$ .
wt+1 (i)
β − β T −t+1 ′ ′
wt (i) = Fi (w )dw + Ei (w) .
1 − β T −t+1 0

Observe that E1 (w) = E2 (w) by assumption. Also, by definition of a mean

preserving spread and since wt+1 (1) ≥ wt+1 (2), we have:
) wt+1 (1) ) wt+1 (2) ) wt+1 (1)
0
F1 (w′ )dw′ ≥ F1 (w′ )dw′ + F2 (w′ )dw′
)0wt+1 (2) )wwt+1 (2)
t+1 (1)
≥ 0
F2 (w′ )dw′ + wt+1 (2)
F2 (w′ )dw′ .
Therefore wt (1) ≥ wt (2).
c. The value of the problem at t = T is vTi (w) = max{w, c}, i = 1, 2. Then
w̄T1 = w̄T2 = c. If we use the same argument as in (b), however, it follows directly
)B )B
that 0 max{w, c}dF1 (w) ≥ 0 max{w, c}dF2 (w), or EvT1 ≥ EvT2 . On the other
hand, the reservation wage at (T − 1) satisfies w̄Ti −1 = β/(1 + β)EvTi . Therefore
w̄T1 −1 ≥ w̄T2 −1 .

Exercise 5.11. Pissarides’ Analysis of Taxation and Variable Search Inten-

sity
An unemployed worker receives each period a zero offer (or no offer) with prob-
ability [1 − π(e)]. With probability π(e) the worker draws an offer w from the
distribution F . Here e stands for effort – a measure of search intensity – and π(e)
is increasing in e. A worker who accepts a job offer can be fired with probability
α, 0 < α < 1. The worker chooses a strategy, that is, whether to accept an offer
or not and how much effort to put into search when unemployed, to maximize
! ∞
E β t yt , 0 < β < 1,
t=0

where yt = w if the worker is employed with wage w and yt = 1 − e + z if the

worker spends e units of leisure searching and does not accept a job. Here z is
unemployment compensation. For the worker who searched and accepted a job,
yt = w − e − T (w); that is, in the first period the wage is net of search costs.
Throughout, T (w) is the amount paid in taxes when the worker is employed. We
66 5. SEARCH, MATCHING, AND UNEMPLOYMENT

assume that w − T (w) is increasing in w. Assume that w − T (w) = 0 for w = 0,

that, if e = 0, π(e) = 0 – that is, the worker gets no offers – and that π ′ (e) > 0,
π ′′ (e) < 0.
a. Analyze the worker’s problem. Establish that the optimal strategy is to choose
a reservation wage. Display the condition that describes the optimal choice of e,
and show that the reservation wage is independent of e.
b. Assume that T (w) = t(w − a) where 0 < t < 1 and a > 0. Show that
an increase in a decreases the reservation wage and increases the level of effort,
increasing the probability of accepting employment.
c. Show under what conditions a change in t has the opposite effect.

Solution
a. Let the state variable that completely summarizes current and future oppor-
tunities be x = (w, e, s), where w is the wage, e is the effort, and s = E if the
worker is employed and s = U if he is unemployed. Recall that, if the worker is
employed, then e = 0. Let Q be the expected value of the objective function for
an unemployed worker who behaves optimally before getting an offer. Then if
the worker is employed, the value of the objective function is given by
v(w, 0, E) = w − T (w) + β(1 − α)v(w, 0, E) + βαQ,
or
w − T (w) βαQ
v(w, 0, E) = + .
1 − β(1 − α) 1 − β(1 − α)
If the worker is unemployed, has an offer w in hand, and spent e > 0 units of
leisure searching this period, the value of the objective function is
v(w, e, U ) = max {w − T (w) − e + β(1 − α)v(w, 0, E)
+βαQ, 1 − e + z + βQ} ,
where the first term reflects the value of accepting employment and the second
the value of rejecting the offer. Using the expression we found for v(w, 0, E), we
get (
w−T (w)
v(w, e, U ) = max 1−β(1−α) −e
,
βαQ
+ 1−β(1−α) , 1 − e + z + βQ .
Then, using a standard argument, we see from the above equation that the opti-
mal strategy is to accept offers greater than or equal to w̄ and to reject all others;
w̄ is such that it makes the worker indifferent between accepting or rejecting the
job offer; that is, w̄ solves
w̄ − T (w̄) βαQ
−e+ = 1 − e + z + βQ,
1 − β(1 − α) 1 − β(1 − α)
or

(30) w̄ − T (w̄) = [1 − β(1 − α)](1 + z + βQ) − βαQ.

 5. SEARCH, MATCHING, AND UNEMPLOYMENT 67

Notice that we cannot use this expression for w̄ to compute the reservation wage,
because Q must be determined endogenously. It is clear, however, that, if Q is
independent of e (as we will show that it is), then w̄ does not depend on e.
Because we established that the optimal policy is of the reservation wage variety,
we can compute v(w, e, U ). This function is given by
# w−T (w) βαQ
1−β(1−α)
− e + 1−β(1−α) w ≥ w̄
v(w, e, U ) =
1 − e + z + βQ w ≤ w̄.
)∞
Let Φ(e) = Ev(w, e, U ) = 0 v(w, e, U )F (dw),
Φ(e) = (1 + z + βQ)F
) ∞ (w̄)
1
+ 1−β(1−α) w̄
[w − T (w)]F (dw) − e
βαQ
+[1 − F (w̄)] 1−β(1−α) .
Because we have shown that
βαQ w̄ − T (w̄)
= (1 + z + βQ) − ,
1 − β(1 − α) 1 − β(1 − α)
we have, after some substitution, that
1
Φ(e) =
)∞
1−β(1−α)
· w̄ ([w − T (w)] − [w̄ − T (w̄)])F (dw) + 1 + z + βQ − e.
Now consider Φ(0). Recall that, if e = 0, the worker gets no offers, and hence
v(w, 0, U ) = 1 + z + βQ. This expression is independent of w, and so Φ(0) =
1 + z + βQ. Therefore
1
Φ(e) =
)∞
1−β(1−α)
· w̄ ([w − T (w)] − [w̄ − T (w̄)])F (dw) + Φ(0) − e.
To simplify notation let (w − T (w)) − (w̄ − T (w̄)) ≡ ∆Y (w). Then the above
expression becomes
$ ∞
1
Φ(e) = ∆Y (w)F (dw) + Φ(0) − e.
1 − β(1 − α) w̄
If the worker chooses to spend e units of effort, he gets an offer with probability
π(e) and expected value Φ(e). With probability [1 − π(e)] he gets no offers. This
alternative has value Φ(0) − e.
Then the value of the problem for an unemployed worker who behaves optimally
is given by Q, where Q satisfies
Q ≡ max0≤e≤1 {π(e)Φ(e) + [1 − π(e)][Φ(0) − e]}
(31) {π(e)[Φ(e) − Φ(0) + e] + Φ(0) − e}
Q ≡ max0≤e≤1 ( ,
π(e) )∞
Q = max0≤e≤1 1−β(1−α) w̄
∆Y (w)F (dw) + 1 − e + z + βQ
The right-hand side defines a mapping from Q into the reals. To guarantee that
the problem is well behaved, we want to show that one such Q exists. This is
not a trivial problem: Q affects w̄ and ∆Y (w), so that the mapping is highly
nonlinear. In any case, it is clear that Q, and therefore w̄, are independent of e.
68 5. SEARCH, MATCHING, AND UNEMPLOYMENT

Let H be the mapping defined by the right-hand side of (31). Because π(e) is
increasing in e, we have that
1
)∞
HQ ≤ 1−β(1−α) ∆Y (w)F (dw) + 1 + z + βQ
w̄
1
)∞
≤ H̄Q ≡ 1−β(1−α) 0 [w − T (w)]F (dw) + 1 + z + βQ.
Therefore, if Q1 is such that Q1 = H̄Q1 (such a Q1 is easy to compute directly),
it follows that, for all Q ≥ Q1 , Q ≥ H̄Q. Thus ∀Q ≥ Q1 , HQ ≤ Q. On the other
hand, ( ,
π(0) )∞
HQ ≥ 1−β(1−α) w̄
∆Y (w)F (dw) + 1 + z + βQ
= 1 + z + βQ ≡ HQ.
Then we have that, for all Q ≥ 0, HQ ≤ HQ ≤ H̄Q and H0 > 0. Hence we have
established that H0 > 0 and that there exists Q1 < ∞ such that HQ ≤ Q for
Q ≥ Q1 .
Inasmuch as H is a continuous function of Q [this follows because w̄ is continuous
in Q, as is ∆Y (w)], we establish that there exists a Q̄ such that H Q̄ = Q̄.
We next prove that Q̄ is unique. To do so it suffices to show that the mapping
H is monotone in Q. A sufficient condition is that
/$ ∞ 0
∂ ∂ w̄
0≤ ∆Y (w)F (dw) + β < 1.
∂ w̄ w̄ ∂Q
)∞
Still, (∂/∂ w̄) w̄ ∆Y (w)F (dw) is (using the Leibniz rule) equal to −[1−(∂T /∂w)(w̄)]
[1−F (w̄)]. From the equation determining w̄, we get that [1−(∂T /∂w)(w̄)](∂ w̄/∂Q) =
β(1 − β)(1 − α). Because −[1 − F (w̄)]β(1 − β)(1 − α) + β ∈ (0, 1), however, H
is increasing. Next we use (31) to characterize the optimal choice of e. It is clear
that it satisfies
$ ∞
′ 1
(32) π (e) ∆Y (w)F (dw) = 1,
1 − β(1 − α) w̄
if the solution is interior. We assume that the distribution of w has sufficient
mass in the tail to make search attractive – that is, we assume that the solution
is interior. It is being claimed that it is possible to make assumptions about the
deep parameters of the model, F (w), α, β, z, π(e), that will guarantee that the
optimal choice of e is e > 0. We focus on this case only because the other is
trivial.
From (31) it is clear that the optimal Q satisfies
/ $ ∞ 0
−1 π(ē)
Q̄ = (1 − β) ∆Y (w)F (dw) + 1 − ē + z .
1 − β(1 − α) w̄
Using this equation in equation (30), we obtain another, more familiar character-
ization of the optimal reservation wage,
)∞
w̄ − T (w̄) = (1 + z) − β(1 − α)ē + β(1−α)π(ē)
1−β(1−α) w̄
{[w − T (w)]
(33)
− [w̄ − T (w̄)]} F (dw).
Then equations (33) and (32) summarize the determination of the endogenous
variables, e and w̄.
 5. SEARCH, MATCHING, AND UNEMPLOYMENT 69

b. Assume that T (w) = t(w − a). To explore the effect of a change in a, we

differentiate completely (33) and (32) with respect to a. We start with (33).
β(1−α)
(1 − t) ∂∂aw̄ + t = −β(1 − α) ∂a ∂ē
+ 1−β(1−α)
)∞ ∂e
· w̄ ∆Y (w)F (dw)π ′ (ē) ∂a
− β(1−α)π(ē)
1−β(1−α)
(1 − t)[1 − F (w̄)] ∂∂aw̄ .
)∞
Using equation (32) to eliminate 1/[1 − β(1 − α)] w̄ ∆Y (w)F (dw)π ′ (ē) = 1, we
get / 0
β(1 − α)π(ē)[1 − F (w̄)] ∂ w̄
(1 − t) 1 + = −t.
1 − β(1 − α) ∂a
Then (∂ w̄/∂a) < 0.
The intuition underlying this result is that an increase in a makes the income tax
more progressive, as it increases the subsidy to low-income workers. Because taxes
are paid (and the subsidy is received) only if the worker is employed, the increased
attractiveness of low-income jobs is reflected by a reduction in the minimum wage
at which an unemployed worker is willing to accept an offer. Notice that the term
(∂e/∂a) disappears in the above equation. This is just another consequence of
the property that e does not affect the choice of the reservation wage.
We next explore the effect on e. From (32) we get
π ′′ (ē) )∞ ∂e
1−β(1−α) w̄
∆Y (w)F (dw) ∂a
π ′ (ē)
= 1−β(1−α)
(1 − t) ∂∂aw̄ [1 − F (w̄)]
or
π ′′ (ē) ∂e (1−t)[1−F (w̄)] ∂ w̄
π ′ (ē)2 ∂a
= 1−β(1−α) ∂a
.
Because π ′′ (e) < 0, we have that (∂e/∂a) > 0, that is, effort is increased. Notice
that the increase in e increases π(ē), and hence the probability of getting an
acceptable offer π(ē)[1 − F (w̄)] rises. To fix the notation, let p = π(e)[1 − F (w̄)].
Then
∂p ∂e ∂ w̄
= [1 − F (w̄)]π ′ (ē) − F ′ (w̄)π(e) ,
∂a ∂a ∂a
and our results show that(∂p/∂a) > 0.
c. Next we analyze the effects of changing the marginal tax rate t. We follow
exactly the same method of totally differentiating (33) and (32) to get, from (33),
( ,
∂ w̄ β(1−α)π(ē)[1−F (w̄)]
(1 − t) ∂t 1 + 1−β(1−α)
β(1−α)π(ē) ) ∞
= w̄ − a − 1−β(1−α) w̄ (w − w̄)F (dw).
From (33), however, we got that
)∞
w̄ − β(1−α)π(ē)
1−β(1−α) w̄
(w − w̄)F (dw)
−1
= (1 − t) [(1 + z) − a − β(1 − α)ē].
Then
∂ w̄
sign = sign [(1 + z) − a − β(1 − α)ē].
∂t
70 5. SEARCH, MATCHING, AND UNEMPLOYMENT

From (32), after we substitute into the expression for (∂ w̄/∂t), we get
π ′′ (ē) ∂e π ′ (ē)
π ′ (ē) ∂t
=
1 )∞
[1−β(1−α)+β(1−α)π(ē)[1−F (w̄)]2
· w̄ + w̄ (w − w̄)F (dw) .

Therefore (∂e/∂t) < 0 unambiguously.

Notice that, in this case, an increase in t reduces the returns of being employed
and therefore makes working less attractive. Consequently, it is optimal for the
unemployed worker to reduce the level of effort, decreasing the probability of
finding a job. On the other hand, it is possible for the reservation wage to
decrease, that is, for some wage offers to be acceptable to the worker after the
increase in the tax rate. Such a decrease becomes more likely as a grows larger.
In this case, the increase in the marginal rate can actually increase payments to
the worker when w − a < 0. This higher subsidy makes working more attractive,
consequently reducing the reservation wage.

Exercise 5.12. Search and nonhuman wealth

An unemployed worker receives every period an offer to work forever at wage

w, where w is drawn from the distribution F (w). Offers are independently and
identically distributed. Every agent has another source of income, which we
denote ϵt , and that may be regarded as nonhuman wealth. In every period all
agents get a realization of ϵt , which is independently and identically distributed
over time, with distribution function G(ϵ). We also assume that wt and ϵt are
independent. The objective of a worker is to maximize
∞
!
E β t yt , 0 < β < 1,
t=0

where yt = w + φϵt if the worker has accepted a job that pays w, and yt = c + ϵt
if the worker remains unemployed. We assume that 0 < φ < 1 to reflect the fact
that an employed worker has less time to engage in the collection of nonhuman
wealth. Assume 1 > prob{w ≥ c + (1 − φ)ϵ} > 0.
Analyze the worker’s problem. Write down Bellman’s equation and show that
the reservation wage increases with the level of nonhuman wealth.

Solution

If the worker accepts a job that pays w, her total utility is given by
∞
! β
w + θϵt + E β j (w + φϵt+j ) = w + φϵ + (w + φEϵ).
j=1
1−β
 5. SEARCH, MATCHING, AND UNEMPLOYMENT 71

Then let v(w, ϵ) be the optimal value of the objective function for an unemployed
worker who has an offer w in hand and nonhuman wealth equal to ϵ. Then
&
β
v(w, ϵ) = max w + φϵ + 1−β
(w + φEϵ)
'
))
c+ϵ+β v(w′ , ϵ′ )dF (w ′ )dG(ϵ′ ) .

The second term in the bracketed expression does not depend on w. Therefore,
for each ϵ, the optimal strategy is to choose a reservation wage. To see how the
reservation wage w̄(ϵ) varies with ϵ, write the indifference condition :

β
w̄(ϵ) + φϵ + (w̄(ϵ) + φE(ϵ)) = c + ϵ + βQ,
1−β
))
where Q ≡ β v(w′ , ϵ′ )dF (dw ′ )dG(ϵ′ ). Rearanging gives :

w̄(ϵ) β
= c + (1 − φ)ϵ + βQ − φE(ϵ).
1−β 1−β
Since 0 < φ < 1, the above equation implies that w̄(ϵ) is an increasing function
of ϵ.

Exercise 5.13. Search and asset accumulation

A worker receives, when unemployed, an offer to work forever at wage w, where
w is drawn from the distribution F (w). Wage offers are identically and indepen-
dently distributed over time. The worker maximizes
∞
!
E β t u(ct , lt ), 0 < β < 1,
t=0

where ct is consumption and lt is leisure. Assume Rt is i.i.d. with distribution

H(R). The budget constraint is given by
at+1 ≤ Rt (at + wt nt − ct )
and lt + nt ≤ 1 if the worker has a job that pays wt . If the worker is unemployed,
the budget constraint is at+1 ≤ Rt (at +z −ct ) and lt = 1. Here z is unemployment
compensation. It is assumed that u(·) is bounded and that at , the worker’s asset
position, cannot be negative. This corresponds to a no borrowing assumption.
Write down Bellman’s equation for this problem.

Solution
A natural choice for the state variable in this problem is the vector (w, a, R, s),
where s = E if the worker is employed and s = U if the worker is unemployed.
72 5. SEARCH, MATCHING, AND UNEMPLOYMENT

We first analyze the problem faced by an employed worker. This problem is

# $ %
′ ′ ′
v(w, a, R, E) = max′ u(c, l) + β v(w, a , R , E)dH(R ) ,
c,l,n,a

subject to a′ ≤ R(a + wn − c), l + n ≤ 1.

If the worker is unemployed, the value function is given by
(
v(w, a, R, U ) = max v(w, a, R, E),
3 )) 4,
max u(c, 1) + β v(w′ , a′ , R′ , U )F (dw ′ )dH(R′ ) ,

subject to a′ ≤ R(a + z − c), where the first term in brackets reflects the value
of accepting the job, whereas the second represents the value of remaining unem-
ployed. In each case the asset position is chosen optimally. It is possible to argue
that the optimal strategy is to set a reservation wage w̄(a, R) that depends on
both the asset position and the rate of interest R.

Exercise 5.14. Temporary unemployment compensation

Each period an unemployed worker draws one, and only one, offer to work forever
at wage w. Wages are i.i.d. draws from the" c.d.f. F , where F (0) = 0 and
F (B) = 1. The worker seeks to maximize E ∞ t=0 β t
y t , where yt is the sum of the
worker’s wage and unemployment compensation, if any. The worker is entitled
to unemployment compensation in the amount γ > 0 only during the first period
that she is unemployed. After one period on unemployment compensation, the
worker receives none.
a. Write the Bellman equations for this problem. Prove that the worker’s optimal
policy is a time-varying reservation wage strategy.
b. Show how the worker’s reservation wage varies with the duration of unem-
ployment.
c. Show how the worker’s “hazard of leaving unemployment” (i.e., the probability
of accepting a job offer) varies with the duration of unemployment.
Now assume that the worker is also entitled to unemployment compensation if
she quits a job. As before, the worker receives unemployment compensation in
the amount of γ during the first period of an unemployment spell, and zero during
the remaining part of an unemployment spell. (To requalify for unemployment
compensation, the worker must find a job and work for at least one period.)
The timing of events is as follows. At the very beginning of a period, a worker
who was employed in the previous period must decide whether or not to quit.
The decision is irreversible; that is, a quitter cannot return to an old job. If
the worker quits, she draws a new wage offer as described previously, and if she
accepts the offer she immediately starts earning that wage without suffering any
period of unemployment.
 5. SEARCH, MATCHING, AND UNEMPLOYMENT 73

d. Write the Bellman equations for this problem. [Hint: At the very beginning
of a period, let v e (w) denote the value of a worker who was employed in the
previous period with wage w (before any wage draw in the current period). Let
v1u (w′ ) be the value of an unemployed worker who has drawn wage offer w ′ and
who is entitled to unemployment compensation, if she rejects the offer. Similarly,
let v+u
(w′ ) be the value of an unemployed worker who has drawn wage offer w ′
but who is not eligible for unemployment compensation.]
e. Characterize the three reservation wages, w̄ e , w̄1u , and w̄+
u
, associated with
the value functions in part d. How are they related to γ? (Hint: Two of the
reservation wages are straightforward to characterize, while the remaining one
depends on the actual parameterization of the model.)

Solution
a. Let vu1 (w) (v+
u
(w)) be the value function of an unemployed worker with wage
w in hand in the first (after the first) period of unemployment and who behaves
optimally. The Bellman equation are :
# $ B %
u w + ′ ′
v1 (w) = max ,γ + β v (w )dF (w )
{A,R} 1−β 0
# $ B %
u w + ′ ′
v+ (w) = max ,β v (w )dF (w ) .
{A,R} 1−β 0
In each of the two periods the problem is a standard one, leading to a reservation
wage policy. If the unemployed is in her fist period of unemployment then the
optimal policy is accept for w ≥ w 1 and reject otherwise. The )associated value
w w1 B
function is v1u (w) = 1−β for w ≥ w 1 and v1u (w) = 1−β = γ + β 0 v1u (w′ )dF (w ′ )
for w < w 1 . After one (or more) period(s) of unemployment the optimal policy
is to accept when w ≥ w + and to reject otherwise.
b. To show that w 1 > w+ , just write the two indifference conditions satisfied by
the two reservation wages :
w1
)B u ′
1−β
= γ+ β 0
v+ (w )dF (w ′ )
w + ) B u
1−β
= β 0 v+ (w′ )dF (w ′ ).
Clearly, w 1 > w+ . Note that w 1 − w+ = (1 − β)γ. This equality has the
following interpretation. Suppose that an unemployed worker in the first period
of unemployment receive an offer w. If he accepts it, her payoff is :
w
.
1−β
If, on the other hand, she rejects it, then her payoff is made of two terms. The
first term is the unemployment compensation, γ. The second term is the option
w+
value of waiting, which is equal to 1−β . Thus, the worker accepts whenever :

w ≥ γ(1 − β) + w + .
74 5. SEARCH, MATCHING, AND UNEMPLOYMENT

in term of “average payoff” per period, the worker accepts whenever the wage
exceed the reservation wage w + plus the annuity value of receiving unemployment
compensation today.
c. The workers probability of finding a job is determined by P [w > w i ], i = +, 1.
Since w1 > w+ , the probability of accepting a job is higher after one period of
unemployment: P [w > w + ] ≥ P [w > w 1 ].
d. v1u (w) and v+u
are defined as in question a. Let v e (w) be the value of an
employed worker with wage w in hand and who behaves optimally. The three
value functions are solution of the following system of Bellman equations :

# $ B %
(34) e
v (w) = max w + βv (w),e
v1u (w′ )dF (w ′ )
stay,quit 0
# $ B %
u e u ′ ′
(35) v1 (w) = max w + βv (w), γ + β v+ (w )dF (w )
accept,reject 0
# $ B %
u e u ′ ′
(36) v+ (w) = max w + βv (w), β v+ (w )dF (w ) .
accept,reject 0
)B )B
e. To simplify notations, define first Q1 ≡ 0 v1u (w′ )dF (w ′ ) and Q+ ≡ 0
u
v+ (w′ )dF (w ′ ).
The characterization goes in several steps.

Step 1 : Characterizing v e (w)

From equation (34) it is clear that if an employed worker decides to stay in a

given period, he will decide to stay in all the subsequent periods. Thus we can
rewrite equation (34) as :
# %
e w
v (w) = max , Q1 .
stay,quit 1−β
Furthermore, the above expression shows that the optimal policy of an employed
worker is of the reservation wage form. Specificaly, there exists we such that for
all w ≤ we the worker quits her job and v e (w) = Q1 . For all w > we the worker
w
stays at her job and v e (w) = 1−β . Lastly, we solves the following indifference
condition :
we
= Q1 .
1−β
Step 2 : Unemployed workers have reservation wage policies

Since v e (w) is increasing, it follows from the Bellman equations (35) and (36)
that unemployed workers have reservation wage policies. Let w 1 and w+ be the
corresponding reservation wages.
 5. SEARCH, MATCHING, AND UNEMPLOYMENT 75

Step 3 : Q1 ≥ Q+

Equations (35) and (36) imply that v1u (w) ≥ v+ u

(w) for all w. In word, unem-
ployed workers are better off in the first period of unemployment because they
receive the benefit γ. Integrating with respect to dF (w ′ ) implies that Q1 ≥ Q+ .

Step 4 : w+ = 0

This is an intuitive fact. After the first period of unemployment a worker does not
receive any benefit. Accepting an offer and quitting is as least as good as rejecting
an offer and drawing again next period. Let’s prove it formally. Assume w + > 0.
Then w+ + βv+ u
(w+ ) = βQ+ < Q1 = we + βv e (we ). Since w + v e (w) is weakly in-
creasing this implies that w + < we . Thus v e (w+ ) = Q1 . Thus w+ + βQ1 = βQ+ ,
implying that w + = β(Q+ − Q1 ) < 0. A contradiction.

Step 5 : γ > 0 ⇒ w 1 > 0

This is also an intuitive result. Since an unemployed workers receives benefits in

its first period of unemployment and, she surely refuses to work when the wage
offer is small enough. Formally, assume that w1 = 0. Then v1u (w) = v+ u
(w) for all
w. This implies that Q1 = Q+ . Also, since w1 = 0, accepting wage offer 0 is at
least as good as rejecting it. Thus :

0 + βv e (0) = βQ1 ≥ γ + βQ+ = γ + βQ1 .

When we use the fact that v e (0) = Q1 . The above implies γ < 0. A contradiction.

Step 6 : w1 ≤ we and w1 = γ − β(Q1 − Q+ )

The Bellman equation (35) implies that v1u (w) ≥ γ + βQ+ for all w. Integrating
with respect to dF (w ′ ) gives Q1 ≥ γ + βQ+ . From the indifference conditions
defining w e and w1 , this is equivalent to :

we + βv e (we ) ≥ w1 + βv e (w1 ).
Since v e is weakly increasing, it implies that w e ≥ w1 . Thus v e (w1 ) = Q1 . Using
this equality to rewrite the indifference condition defining w 1 gives :

w1 + βQ1 = γ + βQ+ ⇒ w1 = γ + β(Q+ − Q1 ).

The above manipulations show that 0 = w + ≤ w1 ≤ we and w1 = γ +β(Q+ −Q1 ).

Note that w1 is strictly less than γ. This reflect the fact that, when an agent
reject, she receives unemployment compensation this period but also loose the
right to receive it next period. On the other hand,if she accepts, she keeps the
right to receive unemployment compensation next period.
Lastly, we cannot tell whether or not w e is smaller or greater than gamma.
76 5. SEARCH, MATCHING, AND UNEMPLOYMENT

Step 6 : Dependence on γ

First note that the value functions are weakly increasing in γ. To see why
this is the case consider the optimization problem when the compensation is
γ ′ = γ + ∆γ > γ. A posible decision rule for the agents is to use the same reser-
vation wage policy as when the compensation is γ. Payoffs are the same as before
except for an additional ∆γ in the first period of unemployment. Therefore, the
value of using this decision rule has increased. Now, under the optimal decision
rule, the value is necessarily even larger.
)B
Since v1u (w, γ) is weakly increasing in γ, Q1 = 0 v1u (w′ , γ)dF (w ′ ) is also weakly
increasing in γ. Thus we = (1 − β)Q1 is weakly increasing in γ.

Exercise 5.15. Seasons, I

"
An unemployed worker seeks to maximize E ∞ t
t=0 β yt , where β ∈ (0, 1), yt is her
income at time t, and E is the mathematical expectation operator. The person’s
income consists of one of two parts: unemployment compensation of c that she
receives each period she remains unemployed, or a fixed wage w that the worker
receives if employed. Once employed, the worker is employed forever with no
chance of being fired. Every odd period (i.e., t = 1, 3, 5, . . .) the worker receives
one offer to work forever at a wage drawn from the c.d.f. F (W ) = prob(w ≤ W ).
Assume that F (0) = 0 and F (B) = 1 for some B > 0. Successive draws from F
are independent. Every even period (i.e., t = 0, 2, 4, . . .), the unemployed worker
receives two offers to work forever at a wage drawn from F . Each of the two
offers is drawn independently from F .
a. Formulate the Bellman equations for the unemployed person’s problem.
b. Describe the form of the worker’s optimal policy.

Solution
a. The Bellman equations for the even periods V e (w) and for the odd periods
V o (w) are:
# $ B %
o w e ′ 2 ′
V (w) = max ,c + β V (w )dF (w )
{A,R} 1−β 0
# $ B %
e w o ′ ′
V (w) = max ,c + β V (w )dF (w ) .
{A,R} 1−β 0

b. The workers optimal policy will be a reservation wage in each period below
which the worker refuses the best offer outstanding and above which she accepts.
For the odd periods, this reservation wage obeys the equation
 5. SEARCH, MATCHING, AND UNEMPLOYMENT 77

$ w̄e $ B
w̄o w̄e ′ w′
= c+β 2
dF (w ) + β dF 2 (w′ )
1−β 0 1 − β w̄ e 1 − β
$ w̄e o $ B o
w̄ − β w̄ e
2 ′ (w̄ − βw′ ) 2 ′
c = dF (w ) + dF (w )
0 1−β w̄e 1−β
$ B
5 6
o e 2 e o
c(1 − β) = (w̄ − β w̄ ) F (w̄ ) + w̄ 1 − F (w̄ ) − β2 e
w′ dF 2 (w′ ),
w̄e
and
$ w̄o $ B
w̄e w̄o ′ w′
= c+β dF (w ) + β dF (w′ )
1−β 0 1 − β w̄o 1 − β
$ w̄o e $ B e
w̄ − β w̄ o
w̄ − βw′
c = dF (w′ ) + dF (w′ )
0 1 − β w̄ o 1 − β
$ B
5 6
e o o e 2
c(1 − β) = (w̄ − β w̄ ) F (w̄ ) + w̄ 1 − F (w̄ ) − β o
w′ dF (w′ ).
w̄o
Equating the two expressions for c(1 − β):
$ B $ B
′ ′
o e 2 e
w̄ − β w̄ F (w̄ ) − β 2 e o o
w dF (w ) = w̄ − β w̄ F (w̄ ) − β w′ dF (w′ )
w̄e w̄o
/ $ B 0 / $ B 0
o o o ′ ′ e e 2 e ′ 2 ′
w̄ + β w̄ F (w̄ ) + w dF (w ) = w̄ + β w̄ F (w̄ ) + w dF (w ) .
w̄o w̄e
)B )B
For a given w̄, we know that w̄F 2 (w̄) + w̄ w′ dF 2 (w′ ) ≥ w̄F (w̄) + w̄ w′ dF (w′ ).
)B
Furthermore, using Leibnitz rule, we know that w̄F (w̄)+ w̄ w′ dF (w′ ) and w̄F 2 (w̄)+
)B ′ 2 ′
w̄
w dF (w ) are increasing in w̄. Using these two facts, the above equality can-
not hold for w̄ o < w̄e , because both terms on the left hand side would be less
than the corresponding terms on the right hand side. We conclude that w̄ o ≥ w̄e .
The intuition is that in odd periods, the unemployed worker’s outside option (re-
ject and two draws next period) is better than his outside option in even periods
(reject and sample once next period). That makes him want a higher reservation
wage in odd periods.

Exercise 5.16. Seasons, II

Consider the following problem confronting an unemployed worker. The worker
wants to maximize ∞
!
E0 β t yt , β ∈ (0, 1),
0
where yt = wt in periods in which the worker is employed and yt = c in periods
in which the worker is unemployed, where wt is a wage rate and c is a constant
level of unemployment compensation. At the start of each period, an unemployed
worker receives one and only one offer to work at a wage w drawn from a c.d.f.
F (W ), where F (0) = 0, F (B) = 1 for some B > 0. Successive draws from F
78 5. SEARCH, MATCHING, AND UNEMPLOYMENT

are identically and independently distributed. There is no recall of past offers.

Only unemployed workers receive wage offers. The wage is fixed as long as the
worker remains in the job. The only way a worker can leave a job is if she is
fired. At the beginning of each odd period (t = 1, 3, . . .), a previously employed
worker faces the probability of π ∈ (0, 1) of being fired. If a worker is fired, she
immediately receives a new draw of an offer to work at wage w. At each even
period (t = 0, 2, . . .), there is no chance of being fired.
a. Formulate a Bellman equation for the worker’s problem.
b. Describe the form of the worker’s optimal policy.

Solution
a. Let vUe (w), (vUo (w)) be the value an uemployed worker who has just received
an offer w at the start of an even (odd) period and proceeds optimally. Similarly,
let vEe (w), (vEo (w)) be the value an employed with wage w the beginning of an
even (odd) period. The Bellman equation for vUe is :

(37) # / $ 0 %
B $ B
vUe (w) = max w + β π vUo (w′ )dF (w ′ ) + (1 − π)vEo (w) ,c + β vUo (w′ )dF (w ′ ) .
0 0
Similarly, the Bellman equation for vUo is :
# $ B %
(38) vUo (w) e
= max w + vE (w), c + β vUe (w′ )dF (w ′ ) .
0
Bellman equations for employed workers are :

(39) vEo (w) = w + βvEe (w)

$ B
(40) e
vE (w) = w + βπ vUo (w′ )dF (w ′ ) + (1 − π)vEo (w).
0
There is no “max” in the two Bellman equations because it was assumed that the
only way an employed worker can leave a job is by being fired. These 4 Bellman
equations fully describe the worker’s dynamic choice problem.
b. First, solve for vEo (w) and vEe (w) in terms of the optimum value functions.
This produces:
)
e w (1 + (1 − π)β) + βπ vUo (w′ )dF (w ′ )
vE (w) =
1 − (1 − π)β 2
)
o w (1 + β) + β 2 π v o (w′ )dF (w ′ )
vE (w) = .
1 − (1 − π)β 2
Replacing those expressions in (37) and (38) shows that a reservation wage policy
is optimal in this setting. Let the w e (wo ) be the reservation wage in even (odd)
periods. The indifference conditions are :
 5. SEARCH, MATCHING, AND UNEMPLOYMENT 79

/ $ B 0 $ B
e
w +β π vUo (w′ )dF (w ′ ) + (1 − π)vEo (we ) =c+β vUo (w′ )dF (w ′ ),
0 0
which implies that:

1 + β(1 − π) β 2π
we + Qo = c + βQo ,
1 − (1 − π)β 2 1 − (1 − π)β 2
)B
where Qo =≡ 0
vUo (w′ )dF (w ′ ). Some simple algebra yields:

1 − (1 − π)β 2 β [1 − (1 − π)β 2 ] Qo − β 2 πQo

(41) we = c+
1 + β(1 − π) 1 + β(1 − π)
2
1 − (1 − π)β β [1 − (π + β(1 − π))β 2 ] o
(42) = c+ Q.
1 + β(1 − π) 1 + β(1 − π)
Similarly, we know from (38) that, at the reservation wage w o :

(43) wo + β [vEe (wo )] = c + βQe ,

which in turn implies that:

(1 − β) β 2π
wo + Qo = c + βQe .
1 − (1 − π)β 2 1 − (1 − π)β 2
Some simple algebra yields:

o 1 − (1 − π)β 2 β [1 − (1 − π)β 2 ] Qe − β 2 πQo

(44) w = c+ .
1−β 1−β
Now, we can compare the reservation wages in even and odd periods by comparing
eqs. (41) and (44). It is easy to verify that w o ≥ we whenever Qe ≥ Qo ,
since β ∈ (0, 1) and π ∈ (0, 1) . This means that, if the optimum value of the
unemployed (or, equivalently, just fired) worker is higher at the start of an even
period than at the start of an odd period, then the worker sets a higher reservation
wage in the odd period, because he is quite willing to wait another period while
being unemployed (in order to receive Qe ). Also note that the reverse statement
is not necessarily true.

Exercise 5.17. Gittins indices for beginners

At the end of each period, a worker can switch between two jobs, A and B, to
begin the following period at a wage that will be drawn at the beginning of next
period from a wage distribution specific to job A or B, and to the worker’s history
of past wage draws from jobs of either type A or type B. The worker must decide
to stay or leave a job at the end of a period after his wage for this period on
his current job has been received, but before knowing what his wage would be
next period in either job. The wage at either job is described by a job-specific
80 5. SEARCH, MATCHING, AND UNEMPLOYMENT

n-state Markov chain. Each period the worker works at either job A or job B.
At the end of the period, before observing next period’s wage on either job, he
chooses which job to go to next period. We use lowercase letters (i, j = 1, . . . , n)
to denote states for job A, and uppercase letters (I, J = 1, . . . n) for job B. There
is no option of being unemployed.
Let wa (i) be the wage on job A when state i occurs and wb (I) be the wage on
job B when state I occurs. Let A = [Aij ] be the matrix of one-step transition
probabilities between the states on job A, and let B = [Bij ] be the matrix for job
B. If the worker leaves a job and later decides to returns to it, he draws the wage
for his first new period on the job from the conditional distribution determined
by his last wage working at that job.
The worker’s objective
"∞ ist to maximize the expected discounted value of his life-
time earnings, E0 t=0 β yt , where β ∈ (0, 1) is the discount factor, and where yt
is his wage from whichever job he is working at in period t.
a. Consider a worker who has worked at both jobs before. Suppose that wa (i)
was the last wage the worker receives on job A and wb (I) the last wage on job B.
Write the Bellman equation for the worker.
b. Suppose that the worker is just entering the labor force. The first time
he works at job A, the probability distribution for his initial wage is π a =
(πa1 , . . . , πan ). Similarly, the probability distribution for his initial wage on job B
is πb = (πb1 , . . . , πbn ) Formulate the decision problem for a new worker, who must
decide which job to take initially. [Hint: Let va (i) be the expected discounted
present value of lifetime earnings for a worker who was last in state i on job A
and has never worked on job B; define vb (I) symmetrically.]

Solution
a. First we consider a worker who has worked at both jobs before. Suppose
that wa (i) was the last wage the worker receives at job A and wb (I) was the last
wage he received at job B.
Let v(i, I) be the optimum value, starting from next period, of a worker currently
active in job A at wage wa (i) who has also worked at job B (at some point in
the past) at a wage wB (I). Again, this worker is at the end of the current period
and has to decide where to go in the next period before having observed next
period’s wage on either job. Similarly, let v(I, i) be the optimum value, starting
from next period, of a worker currently active in job B at wage wb (I) who has
also worked at job A (at some point in the past) at a wage wA (i).
The Bellman equation for the first worker is given by:

& n n
'
! !
(45) v(i, I) = max Aij [wA (j) + βv(j, I)] , BIJ [wB (J) + βv(J, i)] ,
A,B
j=1 J=1
 5. SEARCH, MATCHING, AND UNEMPLOYMENT 81

while the Bellman equation of the second worker is given by:

& n n
'
! !
(46) v(I, i) = max Aij [wA (j) + βv(j, I)] , BIJ [wB (J) + βv(J, i)] .
A,B
j=1 J=1

Notice how v(i, I) = v(I, i) by comparing the r.h.s of eq. (45) and eq. (46).
This implies we can let v(i, I) denote the optimum value of a worker whose last
wage at job A was wa (i) and at job B was wb (I). (Let’s agree on making the first
argument of the value function the last wage at A).
& n n
'
! !
(47) v(i, I) = max Aij [wA (j) + βv(j, I)] , BIJ [wB (J) + βv(i, J)] .
A,B
j=1 J=1

b. Next, we turn to consider the problem facing a worker who is just about to
enter the labor force. Working backwards, we first examine the case of a worker
who has only worked on one job. Let va (i) denote the optimum value of a worker
at job A, making a wage wa (i), who has never worked on job B before and let
vb (I) denote the same value for a worker earning a wage wb (I) at B, who has
never worked job A before. Then we know that the Bellman equation of a worker
who has only worked at A is:
& n n
'
! !
va (i) = max Aij [wa (j) + βva (j)] , πb (J) [wb (J) + βv(i, J)] ,
A,B
j=1 J=1

while the Bellman equationof a worker who has only worked at B is given by:
& n n
'
! !
vb (I) = max πa (j) [wa (j) + βv(j, J)] , BIJ [wb (J) + βvb (J)] ,
A,B
j=1 J=1

Finally, consider the problem facing a worker who is about to enter the labor
force; naturally, she starts at the job that yields the highest expected lifetime
utility:
& n n
'
! !
Q = max πa (i) [wa (i) + βva (i)] , πa (I) [wb (I) + βvb (I)] .
A,B
i=1 I=1
Now we have exhaustively described the worker’s problem, proceeding back-
wards, which is the only way to solve this type of problem.

Exercise 5.18. Jovanovic (1979b)

An employed worker in the tth period of tenure on the current job receives a
wage wt = xt (1 − φt − st ) where xt is job-specific human capital, φt ∈ (0, 1) is the
fraction of time that the worker spends investing in job-specific human capital,
82 5. SEARCH, MATCHING, AND UNEMPLOYMENT

and st ∈ (0, 1) is the fraction of time that the worker spends searching for a
new job offer. If the worker devotes st to searching at t, then with probability
π(st ) ∈ (0, 1) at the beginning of t + 1 the worker receives a new job offer to
begin working at new job-specific capital level µ′ drawn from the c. d. f. F (·).
That is, searching for a new job offer promises the prospect of instantaneously
reinitializing job-specific human capital at µ′ . Assume that π ′ (s) > 0, π ′′ (s) < 0.
While on a given job, job-specific human capital evolves according to
xt+1 = G(xt , φt ) = g(xt φt ) − δxt ,
′ ′′
where g (·) > 0, g (·) < 0, δ ∈ (0, 1) is a depreciation rate, and x0 = µ where
t is tenure on the job, and µ is the value of the “match” parameter drawn at
the" start of the current job. The worker is risk neutral and seeks to maximize
E0 ∞ τ
τ =0 β yτ , where yτ is his wage in period τ .
a. Formulate the worker’s Bellman equation.
b. Describe the worker’s decision rule for deciding whether to accept an offer µ ′
at the beginning of next period.
c. Assume that g(xφ) = A(xφ)α for A > 0, α ∈ (0, 1). Assume that π(s) =
s.5 . Assume that F is a discrete n-valued distribution with probabilities fi ; for
example, let fi = n−1 . Write a Matlab program to solve the Bellman equation.
Compute the optimal policies for φ, s and display them.
Solution
a. Let v(x) be the optimum value at the start of the current period of an employed
worker who has accumulated a total amount x of job-specific capital and who
proceeds optimally. We know the worker will accept the new draw µ′ at the start
of next period whenever µ′ exceeds next period’s capital on the old job x′ . This
means her Bellman equation is given by:
/ $ 0
′ ′ ′ ′
v(x) = max x(1 − φ − s) + β ((1 − π(s))v(x )) + π(s) max(v(µ ), v(x ))dF (µ )
φ,s

/ /$ 00
′ ′ ′ ′ ′
v(x) = max x(1−φ−s)+β ((1 − π(s))v(x )) + π(s) v(µ )dF (µ ) + F (x )v(x )
φ,s x′
,
where x′ = G(x, φ) = g(xφ) − δx
b. The question is answered in part a.
c. The matlab code is in zia.stanford.edu/public/sarg/webdocs/teaching/econ210/
in files jova.m and readjova.txt