Econ 2010d: Heterogeneity in Macroeconomics

Lecture 3: Search Externalities and Policy in DMP

Adrien Bilal
Spring 2024
Harvard
Introduction
Recap: The DMP Model

• Built a model that speaks to unemployment volatility, inflows and outflows

• Today study its efficiency and normative properties

• When study a policy, need to specify what the market failure is!
I Unless about pure redistribution
I Without market failure, welfare theorems hold
I No need for any intervention (except redistribtuion)

1 / 28
Setup
Constrained Efficiency

• Same setup as in Lecture 2 (linear costs and Nash bargaining)

• We want to study how a benevolent planner would allocate resources

• If the planner could get rid of search frictions, would do so

I But if search frictions are part of technology, planner should internalize them

• Use concept of constrained efficiency

I Study planner’s problem taking search frictions as given

2 / 28
Planning Problem
Planner’s Problem (1/2)

∞
X
max E0 β t Ct
{Ct ,vt }t ,
t=0
s.t. Ct = zt (1 − ut ) + ut b − cvt
ut+1 − ut = s(1 − ut ) − f (vt /ut )ut , u0 given

• Here specify b as home production

• With linear preferences, maximizing consumption ≡ maximizing output
I Transfers immaterial: everyone has same marginal utility of consumption

• Stock-flow equation and matching function as constraints

I Constrained efficiency

3 / 28
Planner’s Problem (2/2)

• Planner problem simplifies to standard dynamic optimization problem

∞
X
β t zt (1 − ut ) + ut b − cvt
 
max E0
{vt }t ,
t=0
s.t. ut+1 − ut = s(1 − ut ) − M(ut , vt ) , u0 given

• Can solve with

I Lagrangian method
I Dynamic programming (define planner’s value Ω(u))

4 / 28
Planner’s Optimality Conditions (1/2)
• Use value function of planner:

Ω(u, z) = max z(1 − u) + ub − cv + βEu,z Ω(u 0 , z 0 )

v
s.t. u 0 = u + s(1 − u) − M(u, v )

• Obtain
 
∂M ∂Ω 0 0
c = −β (ut , vt )Ez (u , z )
∂v ∂u
   
∂Ω ∂M ∂Ω 0 0
(u, z) = b − z + β 1 − s − (ut , vt ) Ez (u , z )
∂u ∂u ∂u

• Using Cobb-Douglas formula for matching function, obtain

c = βq(1 − α)Ez [−∂u Ω0 ]

−∂u Ω = z − b + β (1 − s − αf ) Ez [−∂u Ω0 ]

5 / 28
Planner’s Optimality Conditions (2/2)
• Define the planner’s surplus from a job as S(u, z) = −∂u Ω(u, z)

• The planner’s surplus satisfies the Bellman equation

St = zt − b + β(1 − s − αft )Et St+1

• Vacancy creation becomes

c = (1 − α)βqt Et St+1

• Recall similar equations in decentralized equilibrium

St = zt − b + β(1 − s − γft )Et St+1

c = (1 − γ)βqt Et St+1

• Planner and equilibrium share same stock-flow equation

• Find the difference?

6 / 28
Efficiency
(In)Efficient Vacancy Creation
• Differences between planner and equilibrium are:
I The vacancy creation condition

c = (1 − α)βqt Et St+1 vs. c = (1 − γ)βqt Et St+1

I The surplus equation because of the option value of search

St = zt − b + β(1 − s − αft )Et St+1 vs. St = zt − b + β(1 − s − γft )Et St+1

• If α = γ, the decentralized equilibrium is efficient: Hosios condition (1990)

• If α 6= γ, the decentralized equilibrium is inefficient

• Intuition
I Both firms (and workers) congest the matching function for other searchers
v ∂M
I For firms: M ∂v
=1−α
I How much firms internalize congestion depends on surplus share 1 − γ
I Efficient when firms’s surplus share equals their technological congestion effect

7 / 28
When is Unemployment Too Low or Too High? (1/2)
• Focus again on steady-state zt ≡ z and approximation s, r  f :
1−γ z −b 1−αz −b
θDE ≈ vs. θSP ≈
γ c α c
s
and recall u ≈ mθ 1−α

• Ratio between equilibrium and efficient unemployment rates:

1−α
u DE

γ 1−α
≈
u SP 1−γ α
I Increasing in γ

• Unemployment is too high iff γ > α

I Worker bargaining power is high, firm bargaining power is low
I Firms receive too little surplus and so create too few vacancies

• To leading order, unemployment volatility unaffected by (in)efficiency

du DE du SP
≈
u DE u SP
8 / 28
When is Unemployment Too Low or Too High? (2/2)

• Many estimates find α ∼ 0.5, γ ∼ 0.2

• Imply that steady-state unemployment rate is too low (by half)!

I Somewhat surprising
I But recall that planner can redistribute between risk-neutral agents
I So planner just maximizes output, all about vacancy creation here

• Recall that we consider constrained efficiency

9 / 28
Optimal Policy
Labor Taxes
• Now return to the decentralized equilibrium

• Introduce a lump-sum labor tax τt paid by the firm

• Values of workers and vacant firm unchanged

• Value of filled firm in decentralized equilibrium now

Jt = zt − τt − wt + β(1 − s)Et Jt+1

• Solve for wages and values using bargaining solution as before
I Previous equations hold after replacing zt by zt − τt

• Seek tax τt such that in all states of the world

(1 − γ)St (τt ) = (1 − α)St , θtDE = θtSP

where now we use the alternative expression for the surplus (see Lecture 2)
St = zt − τt − b + θtDE c + β(1 − s − ftDE )Et St+1
St = zt − b + θtSP c + β(1 − s − ftSP )Et St+1
10 / 28
Optimal Labor Tax

• Optimal tax
α−γ 
τt = zt − b + θtDE c
1−γ

I When γ > α and unemployment too high, subsidize labor

I When γ < α and unemployment is too low, tax labor

• In practice need to
I Know α, γ
I Know zt , b, c
I Measure θt

• Can also use proportional wage tax, similar information requirements

11 / 28
Taking Stock

• We have seen the efficiency criteron of the DMP model

I Each side be compensated by how much it congests the matching function
I Hosios condition: γ ≶ α. Unemployment too high iff γ > α
I Can be corrected with labor or wage tax/subsidy

• So far considered a model with homogeneous workers and firms

• Now introduce heterogeneity and study how it affects efficiency

12 / 28
Composition Externalities
Setup

• Suppose now there are many types of jobs/firms i ∈ {1, 2, ..., I }

I zi+1,t > zi,t a.s.
I Measure Mi of each type

• Keep workers homogeneous

I Could have worker heterogeneous too, same logic

• Suppose that all firm types face random matching

I If firm i posts vi vacancies, it meets qvi workers, q = P M
i Mi vi

I All firm types bundled into same matching function

• Revert to convex cost formulation from lecture 2

I Important to ensure that all firm types are active in equilibrium
I cit = Ci0 (vit ) ≡ ci (vit ) = c0 viν

13 / 28
Values
• Value of unemployment now
 
X
Ut = b + βEt ft ωj (vv t )Ej,t+1 + (1 − ft )Ut 
j

where
M i vi
ωi (vv ) = P
j M j vj

is firm i’s vacancy share

• Other Bellman equations remain unchanged (except adding an i subscript)

• Firm type-specific surplus

 
X
Si,t = zit − b + β(1 − s − γft )Et Si,t+1 + βγft Et Sit+1 − ωj (vv t )Sjt+1 
| {z } j
as in homogeneous case | {z }
heterogeneity correction: worker’s outside option

14 / 28
Job Creation

• Job creation for each type

cit = (1 − γ)βqt Et Si,t+1

• In steady-state
X
Si = zi − b + β(1 − s)Si − βγf ωj Sj
j
ci = βq(1 − γ)Si

• Heterogeneity in wages across firms i !

 
X
wi = (1 − γ) b + βγf ωj Sj  + γzi .
j

15 / 28
Planner’s Problem
• Planner’s state variable is now employment at each type
I In homogeneous model could equivalently use unemployment
I Denote vectors with bold font: (n1 , ..., nI ) ≡ n and (v1 , ..., vI ) ≡ v

• Use value function of planner:

!
X  X
Ω(nn , z ) = max zi ni − Mi Ci (vi ) + 1− ni b + βEz Ω(nn 0 , z 0 )
v
i i
P !
j Mj vj
s.t. ni0 = (1 − s)ni + Mi vi q P for all i
1− j nj

• Obtain
X Mj vj 0
n0 , z 0 ) + β n0 , z 0 )
   
ci (vi ) = βq(θ)Ez ∂ni Ω(n q (θ)Ez ∂nj Ω(n
j
u
 0 0 
n, z )
∂ni Ω(n = zi − b + β (1 − s) Ez ∂ni Ω(n n ,z )
X Mj vj 0
n0 , z 0 )
 
+ q (θ)θEz ∂nj Ω(n
j
u

16 / 28
Planner’s Solution

• Define the planner’s suplus of a job at firm i as

Sit ≡ ∂ni Ω(nn t , z t )

• In steady-state:
X
Si = zi − b + β (1 − s) Si − βαf ωj Sj
j
n X o
ci = βq Si − α ωj Sj
j

17 / 28
Surpluses and Job Creation
• Compare steady-state match surplus in eq. Si vs. in planner’s solution Si
X
Si = zi − b + β(1 − s)Si − βγf ωj Sj
j
X
Si = zi − b + β (1 − s) Si − βαf ωj Sj
j

I Under Hosios condition α = γ, both coincide given f

• Compare vacancy creation conditions

c(vi ) = βq(1 − γ)Si

X
c(υi ) = βq(1 − α)S i + α ωj (S i − S j )
j

• Even w/ Hosios condition α = γ, heterogeneity ⇒ composition externality

I Planner internalizes that creation of job i congests matching fct for job j

18 / 28
Direction of Composition Externality
• Compare job creation conditions under Hosios condition

c(vi ) = βq(1 − α)Si

X
c(υi ) = βq(1 − α)S i + α ωj (S i − S j )
j
I Planner internalizes that creation of job i congests matching fct for job j

• Consider only two firm types i ∈ {1, 2}

• Then z2 > z1 and so S2 > S1 , S2 > S1

• Can show with some algebra that

ω2SP > ω2DE , ω1SP < ω1DE

I If f , q, ωi the same in SP vs. DE, SP increases vacancies at productive firm

I Too many bad jobs in equilibrium! (Acemoglu 2001)

19 / 28
Taking Stock: Composition Externalities

• With heterogeneous prod., search frictions lead to composition externalities

• Workers randomly meet firms ⇒ low-prod. firms “free-ride” labor market

I Planner diverts job creation away from low-prod. firms
I Happens despite endogenous job creation
I Always leads to excess job creation at unproductive firms
I If heterogeneous workers instead: too much search effort by low-prod. workers

• High-prod. firms would like to operate in a separate labor market

I Bonus slides: segmented markets
I Back to Hosios condition, no composition externalities

• Solution: Endogenous markets!

I Moen (1997): competitive/directed search
I Will see that it removes both the Hosios and the composition externalities

20 / 28
Competitive Search
Endogenous Labor Market Segmentation

• For now return to homogeneous firms and no aggregate shocks zt ≡ z

• Relax exogenous markets

• Markets now segmented by wage offer w

I Firms now have commitment power
I They post a wage offer w
I One labor market per wage offer
I Let firms open their own labor market/join an existing market

• Matching function applies market by market

• Workers observe wage offers and decide in which market to search

• Firms internalize that posting a higher wage may attract more applicants

21 / 28
Endogenous Labor Market Segmentation
• Value of unemployment if search in market with wage w

U(w ) = b + β [f (w )E (w ) + (1 − f (w ))U(w )]

• E (w ) as before

• Ex-ante worker arbitrage between (active) markets

U(w ) ≡ U ≡ max
0
U(w 0 ) =⇒ f (w ) [E (w ) − U] ≡ Λ
w

I When w increases, E (w ) rises

I Worker indifference implies that f (w ) must fall
I From matching function, q(w ) rises: firms attract more workers
I Trade-off between wage and recruiting speed across markets

• Surplus still independent from wage

S = z − b + β(1 − s)S − βΛ
22 / 28
Firm’s Problem (1/2)

• Vacant firms now choose which wage to offer

max q(w )J(w ) s.t. f (w )[S − J(w )] = Λ

w

• Trade off
I Higher wage/lower profits (J(w ) ↓)
I Hiring workers faster (q(w ) ↑)

• Can re-write as choosing J directly

1 α
max (mΛ−α ) 1−α J (S − J) 1−α
J

I Used matching function to relate q and f

I No need to express J(w ) = z − w ...

23 / 28
Firm’s Problem (2/2)

1 α
max (mΛ−α ) 1−α J (S − J) 1−α
J

• Solution

J = (1 − α)S , c = β(1 − α)qS

• “As if” bargaining solution with bargaining power γ = α!

I All firms identical here, so all choose same wage and split of value

• Commitment =⇒ firms internalize their congestion through wages

I Just as the planner does
I In equilibrium, only one wage is posted
I But firms internalize off-equilibrium changes in their market’s tightness

24 / 28
Bringing Back Heterogeneity

• Now introduce again firm types zi , i = 1...I as before

• Workers now indifferent between

I Different wages within firm types
I Different firm types

• Calculations in homogeneous case hold conditional on Λ ≡ fi (w )[Ei (w ) − U]

• Solution

Ji = (1 − α)Si , ci = β(1 − α)qi Si

I Coincides with planning solution when markets are segmented

I As in homogeneous case, surplus is shared efficiently

• Where did composition externalities go?

25 / 28
Composition Externalities

• With competitive search, opening a new labor market is part of technology

• So the planner can also open new markets

I Never optimal to mix firm types inside same market if can avoid it
I Otherwise bad firm crowds out applications to good firm

• Thus, we are back to the planner’s solution for segmented markets

c(υi ) = β(1 − α)qi Si

• The decentralized equilibrium is efficient

• No composition externalities

26 / 28
Taking Stock
• Competitive search makes search economy efficient
I Irrespectively of homogeneity or heterogeneity
I Also called directed search

• Key ingredients for efficiency to obtain

I Commitment to wage offers
I Workers perfectly observe all wage offers
I Possibility of opening new markets
I Undoes “Hosios” externality: firms internalize wage-congestion trade-off
I Undoes composition externalities: firms optimally segment across markets

• Competitive search sometimes very useful

I For writing models with frictions...
I ...But where the goal is to emphasize different market failure

• But interpretation harder than with random search

I If workers know everything about the labor market, what are search frictions?
I Search frictions usually a short-hand for incomplete information, screening, etc.
27 / 28
Conclusion
Conclusion: Unemployment and the DMP Model
• Unemployment facts
I St.dev. of unemployment rate ≈ 1.5 p.p. over the cycle
I Job finding rate accounts for ≈ 60 − 90% of that variation
I Vacancy creation strongly procyclical

• The DMP model: positive analysis

I Captures the basic forces that determine unemployment
I Generates income risk at worker level
I Need tweaking to match empirical u-rate volatility (high b, rigid wages)

• The DMP model: normative analysis

I Hosios condition: efficiency and direction of policy in plain vanilla model
I With heterogeneity also get composition externalities
I Competitive search (vs. random search) brings back efficiency

• Next: Implication of income risk for consumption patterns

28 / 28
Bonus: Segmented Markets
Segmented Markets

• Suppose now that labor markets are also segmented by firm type i
I Now can use either linear or convex job creation costs

• There is an identical matching function for every market i

I i-specific tightness θi , unemployment rate ui

• Unemployed workers choose in which market to search

I Can only search in one market at a time
I Value of unemployment equalized across markets i

29 / 28
Equilibrium Values with Segmented Markets
• Value of unemployment if search in market i

Uit = b + βEt [fit Ei,t+1 + (1 − fit )Uit ]

• Worker arbitrage between markets

Uit ≡ Ut =⇒ fit Et [(Ei,t+1 − Ut )] ≡ Λt

• Surplus satisfies for all markets i

γfit Et Si,t+1 = Λt

I High productivity, high surplus markets have low job-finding rate

I Workers enter market i, lowers job-finding rate, until indifferent with market j

• Surplus Bellman equation

Si,t = zi,t − b + β(1 − s − γfit )Et Si,t+1

30 / 28
Planning Problem

• Similar planning problem as without segmented markets

• Key difference: law of motion of employment

 
0 Vi
ni = (1 − s)ni + vi q
1 − ni

I Tightness is specific to market i

 P V 
j j
I Before had q 1−P nj
j

I No possibility of cross-type congestion ⇒ removes composition externalities

• Derive planning solution as before

31 / 28
Efficiency With Segmented Markets

• Surplus equations in steady-state

Si = zi − b + β(1 − s − γfi )Si , fi Si = constant

Si = zi − b + β(1 − s − αfi )Si , fi Si = constant

• Free entry/job creation conditions

c(vi ) = βq(1 − γ)Si

c(υi ) = βq(1 − α)Si

• Efficiency: back to Hosios condition α = γ

32 / 28
Taking Stock

• With segmented markets composition externalities vanish

• But imposes strong assumption

• Why can’t firm j operate in market i?

• If firm j less productive than i, incentive to enter market i

33 / 28