Welfare analysis of matching models

April 10, 2018

1 Can unemployment be e¢ cient?

In matching models, unemployment is a productive input into the creation
of new jobs. Thus there is no presumption that the optimal unemployment
level is zero.
To have a low unemployment rate, vacancies must be high. But vacancies
consume resources, so that potentially we can have too many vacancies.
On the other hand, the unemployed are also costly because they do not
produce. It is also possible to have too much unemployment.
There are two central market failures in the matching model: congestion
externalities and appropriability problems.

1.1 Congestion externalities

The congestion externalities are as follows. An increase in search exerts posi-
tive externalities on the other side of the market. Unemployment reduces the
length of vacancies; conversely vacancies reduce the length of unemployment
Because m() is concave, search activity exerts a negative externality on
the same side of the market: More unemployment reduces the job …nding
rate.

1.2 Appropriability
The nature of the appropriability problem is as follows. A surplus arises
from turnover costs, and part of it surplus is appropriated by the worker in
the bargaining process. The …rm only appropriates a fraction of the surplus
created by a vacancy, but pays the entire vacancy cost. Hence the incentives

1
to post vacancies are too low. This is the hold-up problem discussed by
Grout (Econometrica, 1984). A symmetrical issue arises on the worker’s
side when evaluating the value of being unemployed: only a fraction of the
value of future jobs will be appropriated, hence the value of looking for a job
(=opportunity cost of working now) is underestimated.

2 Computing the optimum

We now compare the equilibrium derived in the preceding lecture notes with
the optimum. For simplicity we assume no shocks. I will denote by the
market value of labor market tightness and by its optimal value. Recall
that the equilibrium was characterized by three key conditions:

A job creation condition which is

c
= J("u ) = W ("u )(1 ') (1)
q( )

An evolution equation for the joint match surplus:

Z "u
rW (") = " q( )'W ("u ) F ("d )W + (W (x) _:
W ("))f (x)dx + W
"d
(2)

A job destruction condition

W ("d ) = 0:

These equations determine ; W; and "d :

Remember that these equations imply

(" "d )
W (") = ; (3)
r+
so that (2) can be reexpressed as

2
 Z "u
rW (") = " q( )'W ("u ) F ("d )W + (x _
")f (x)dx + W
r+ "d
= " q( )'W ("u ) F ("d )W (4)
+ I("d ) "(1 _;
F ("d )) + W
r+
where Z "u
I(z) = xf (x)dx;
z
The social planner’s problem is to maximize the PDV of output net of vacancy
costs:
Z +1
max (yt c(1 nt ) t )e rt dt;
f t ;"t g 0

subject to the law of motion of employment nt

n_ t = t q( t )(1 nt ) F ("d )nt ;

and an evolution equation for output:

y_ t = yt + t q( t )(1 nt ) "u + nt I("d ):

This equation keeps track of the units of output that are destroyed and
created by shocks and hirings. If a shock does not lead to job destruction, it
is decomposed between an average unit of output being destroyed and a new
unit of output drawn from the distribution of " truncated at " > "d – that
is, per unit of time, n jobs are "destroyed", destroying y units of output,
and n(1 F ("d )) are "recreated", creating nI("d ) units of output.
The state variables are output and the unemployment rate, and the con-
trol variables are the vacancy rate and the job destruction rate. The Hamil-
tonian is

Ht = (yt c(1 nt ) t )e rt + t e rt ( t q( t )(1 nt ) F ("d )nt )

+ t e rt ( yt + t q( t )(1 nt ) "u + nt I("d )) :

The …rst order conditions are

3
 @H
= 0 () c = ( + "u )(q + q 0 ): (5)
@

@H
= 0 () f ("d )n + n "d f ("d ) = 0 (6)
@"d
, + "d = 0

@H rt
= ( _ + r )e () c ( q + F ("d )) "u q + I("d ) = r _
@n
(7)
@H rt
= ( _ + r )e () 1 =r _ (8)
@y

Equation (5) can be written as

c
= !; (9)
q + q0
where

–
q + q 0 = @m=@v
is the marginal productivity of vacancies in job creation, and
therefore q+c q0 is the marginal social cost of job creation,
–
!= + "u (10)
is the marginal social value of a job, which is the sum of the mar-
ginal social value of an extra employed worker and the marginal
social value of the extra output generated by a new job "u ; itself
equal to the marginal social value of output times the output
produced by a new job "u :

Equation (6) states that at the margin of job destruction, the social
value of the foregone output from releasing the least productive worker,
"d ; equates the social value of having one additional worker available

4
 for job search, : Note that < 0 : conditional on a given level of total
output, I want as many workers to be engaged in job search as possible,
or equivalently I want the employed workers to be as productive as
possible.

From (8), we get that in steady state the marginal social value of 1 unit
of output is
1
= ; (11)
r+
the discount rate of 1 unit of output is augmented by ; re‡ecting our
convention that upon a shock, output is destroyed and a new unit of
output is generated from our existing pool of employed workers [This
contribution therefore appears on the RHS of (7), not that of (8)].
Substituting (6) and (11) into (10), we get that

!= ("u "d ); (12)

r+

It follows from (6) and (10) that

"d
= : (13)
r+
The value of having one extra worker employed while output is unchanged,
equates minus the value of that person’s job search. At the optimum, the
latter coincides with the PDV of the output produced by the marginal worker,
since job destruction is determined optimally. That value is precisely equal
to r+"d ; which explains the RHS of (13).

By adding (8) multiplied by "u to (7), we get a law of motion for ! :

r! !_ = "u + c q! F ("d ) + I("d ) "u : (14)

Now we are are going to rearrange this expression to make it look as

similar as possible to (4):

– Using (5), we see that c q! = !( q 0 2 )

– Since F ("d ) = ! F ("d ) + (! ) F ("d ) = ! F ("d ) +
" F ("d ); the last three terms in (14) are equal to ! F ("d ) +
r+ u

r+
(I("d ) "u (1 F ("d )):

5
 – Consequently (14) can be rewritten as

r! = "u !( q 0 2 ) F ("d )! + (I("d ) "u (1 F ("d )) + !;

_
r+
(15)

this is to be compared with (4) written at " = "u ; the law of motion for
the private value of a newly created job in the decentralized equilibrium:

rW ("u ) = "u q( )'W ("u ) F ("d )W + _ ("u )

(I("d ) "u (1 F ("d ))+W
r+
(16)
0 2
Note that q( ) = m(1= ; 1): Therefore q ( ) = 1= :@m=@u; implying
m1 = q 0 2 :

2.1 The job creation decision

The market equilibrium condition is

c = q( )W (1 '):

That is, vacancy cost [c] = probability of …lling a vacancy * value to the
…rm of a …lled job =probability of …lling a vacancy [q( ) = m(u;v)
v
] * value to
the match of a …lled job [W ]*fraction of that value going to the …rm [1 '].
In contrast, the socially optimal job creation condition is
@
c=! m(u; v): (17)
@v
That is, vacancy cost = marginal e¤ect of extra vacancy on jobs created
@
[ @v * marginal social value of a job [ ].
m(u; v)]
We note that:

The private …rm only cares on the average probability of …lling a va-
cancy, m(u;v)
v
in making its decision, while the social planner cares about
the marginal increment in the number of matches generated by an ex-
@
tra vacancy, @v m(u; v): Because of decreasing marginal returns, we have
that @v m(u; v) < m(u;v)
@
v
: This is the congestion externality at work and
it leads to too many vacancies.

6
 The social planner considers the whole social value of a job !; while
the private …rm only internalizes a fraction 1 ' of the private value of
the match created by its vacancy. This is the appropriability problem
and it leads to too few vacancies.

2.2 The asset valuation equation

Comparing (15) and (16), we see that the private opportunity cost of la-
bor is q( )'W = m(u;v) u
'W; while the social opportunity cost is ! @m@u
: The
worker/…rm pair considers that if the worker were instead unemployed, he
would …nd a job with a probability equal to the average job …nding rate,
m(u;v)
u
; and would appropriate a fraction ' of the private surplus of the job
he would …nd. By contrast, the social planner considers that if one extra
worker were unemployed, he would generate a ‡ow of new jobs equal to the
marginal product of labor in the matching function, @m @u
: Furthermore, the
full social gain of those extra jobs ! is taken into account by the social plan-
ner in computing the social opportunity cost of a job. Again we see the two
market failures:

Workers fail to internalize the fact that should they look for a job,
they generate extra jobs at a rate lower than their own job …nding
probability. This is the congestion externality which would lead to too
much worker search, i.e. too much unemployment.

Workers only appropriate a fraction ' of the private value of the jobs
they …nd. This is the appropriability problem which leads to too little
worker search, i.e. too little unemployment.

3 When is the equilibrium optimal?

Let us compute necessary conditions for the optimum and equilibrium to
coincide. For this we need that "d and are the same in the optimum and
the equilibrium. From (12) and (3) it must be that for "d to be the same,

W ("u ) = !:

7
 From (1) and (9) we need that, for to be the same

W ("u )(1 ') = !(1 + q 0 =q):

Thus we need ' = q 0 =q = m1 =( q) = um1 =m:

That is

u @m
@u
'= = u; (18)
m
which, given constant returns in matching, is equivalent to

v @m
@v
1 '= = v:
m
That is, the share of workers (…rms) in the surplus of a match is equal
to the elasticity of the matching function with respect to the corresponding
search input. These conditions are called the Hosios conditions.
Conversely, assume that (18) holds at the decentralized equilibrium. Given
(16), ! = W ("u ) is solution to (15). It then follows that (9) holds, implying
satis…es the optimality condition for job creation. Since (3) holds, (12) also
holds, and it is then straightforward to check that (7) coincides with (15)
for and de…ned by (13) and (11). These values trivially satisfy (6), (8),
and (10), from which it follows that (5) holds. Therefore, all the optimality
conditions hold.
The Hosios conditions mean that the appropriability and congestion prob-
lems exactly balance each other.
There is another way to express the Hosios conditions, which will prove
useful later. If the equilibrium is optimal, then
m @
c = q( )(1 ')W = (1 ')W = W m(u; v);
v @v
and
m @
rVu = q( )'W = 'W = W m(u; v):
u @u
Therefore, for an equilibrium to be optimal we need that

the cost of a vacancy c; is equal to the marginal product of a vacancy

in the matching function, times the net surplus of the match, and

8
 the opportunity cost of job search, rVu ; is equal to the marginal product
of unemployment in the matching function, times the net surplus of the
match.

We note that if ' > u ; there is too much unemployment: the appropri-
ability problem dominates the congestion externality on the …rms’side, and
the converse on the workers’ side. Both e¤ects push unemployment above
the optimum.