Econ 2010d: Heterogeneity in Macroeconomics

Lecture 2: The Search and Matching Model

Adrien Bilal
Spring 2024
Harvard
Introduction
Recap: Unemployment Facts

• St.dev. of unemployment rate ≈ 1.5 p.p. over the cycle

• Job finding rate accounts for ≈ 60 − 90% of that variation

• Vacancy creation strongly procyclical (Beveridge curve, elasticity - 1)

• Job finding rate positively related to V/U ratio (elasticity 0.4)

I Build model that speaks to those facts

I The Diamond-Mortensen-Pissarides (DMP) model
I Sometimes just called the “Search and Matching model”
I So that we can evaluate labor market policies

1 / 31
Some History
From Keynes to the DMP Model
• Keynes hypothesized that low aggregate (goods) demand leads to low output
I Low labor demand & high unemployment

• Two difficulties
I In the NK model, only (mostly) intensive margin and need very high Frisch
F Can use employment lotteries ⇒ Do we observe them in practice?
I No simultaneous flows in and out of unemployment

• Led to development of the DMP model

I Structures the movements in and out of unemployment & wage-setting
I Initially abstract from wage rigidity ⇒ will help with empirical performance!
I Diamond, Mortensen, Pissarides, Hall, Shimer

I
2 / 31
Setup
Preferences
• Discrete time t = 0, 1, ..., ∞

• Unit measure of ex-ante identical risk-neutral workers, discount factor β < 1

I Maximize present value of income stream It
"∞ #
X t
E0 β It
t=0

I Equivalently can be hand-to-mouth

I Risk-neutrality shuts down Euler equation at the heart of RBC/NK models
I Instead focus on cyclical patterns of labor market

• Risk-neutrality implies βR = 1 in equilibrium: β also discount factor of firms

• Employed workers receive an equilibrium wage wt (more on this later)

• Unemployed workers receive unemployment benefits/utility from leisure b

3 / 31
Technology: Production

• A firm uses one worker to produce zt units of output

• zt ∈ {z 1 , ..., z Z } follows a Markov chain with transition matrix T

I Can think about it as proxying aggregate demand
I Though in this simple model analogy is imperfect

• Alternatively can have CRS firms that employ many workers

4 / 31
Technology: Search and Matching
• A vacant firm (i.e. without a worker) incurs a vacancy cost c every period

• A vacant firm posts one vacancy per period (again immaterial under CRS)
I Denote vt the equilibrium measure of vacancies
I A competitive fringe of firms stands ready to enter and post vacancies

• Denote ut the measure of unemployed workers = unemployment rate

• Vacancies and unemployed workers meet according to a matching function:

Mt = M(ut , vt )
I Mt matches formed every period
I Today use Cobb-Douglas

M(u, v ) = mu α v 1−α

I All qualitative results extend to any CRS matching function

5 / 31
Matching in the Labor Market (1/2)
• Useful to define labor market tightness
vt
θt =
ut
• Probability of unemployed finding a job in current period
Mt
ft = −→ f (θ) = mθ1−α
ut
I We assumed random matching
I Matches distributed randomly to unemployed workers

• Probability that a vacant firm finds a worker

Mt
qt = −→ q(θ) = mθ−α = f (θ)/θ
vt
• Suppose m small enough that ft , qt ∈ [0, 1]
• Probability that a job terminates s is exogenous
I Can interpret as probability that zt falls below b
I Was less important than job finding rate for aggregate fluctuations
I Will endogenize in problem set 6 / 31
Matching in the Labor Market (2/2)

• Matching function summarizes the complex functioning of the labor market

I Imperfect information
I Time to locate job offers and apply
I Screening
I Processing times

• Can write micro-foundations of matching function

I Can sometimes imply different functional forms than CB

• Recall Lecture 1: log f was close to linear in v /u = θ

7 / 31
Lecture 1: Vacancies and the Job Finding Rate

1
.75
.5
log UE (de-meaned)
.25
0
-.25
-.5
-.75
-1

-1 -.75 -.5 -.25 0 .25 .5 .75 1

log v/u (de-meaned)

2001Q1-2008Q4 2009Q1-2020Q1
log UE = 0.38 * log v/u

• Recall Lecture 1: log f was close to linear in v /u = θ

I Regression log ft = (1 − α) log θt + log mt implies 1 − α = 0.4 and so α = 0.6
I If can interpret relationship as causal!
I Any issues with that?

• For aggregate data: CB good approximation

8 / 31
Value Functions
Value Functions

• Value of being unemployed

h i
Ut = b + βEt ft Et+1 + (1 − ft )Ut+1

• Value of being employed

h i
Et = wt + βEt sUt+1 + (1 − s)Et+1

• Value of vacant job/firm

h i
Vt = −c + βEt qt Jt+1 + (1 − qt )Vt+1

• Value of filled job/firm

h i
Jt = zt − wt + βEt sVt+1 + (1 − s)Jt+1

9 / 31
Surpluses
• The worker surplus is E − U and satisfies
 
Et − Ut = wt − b + βft Et [Et+1 − Ut+1 ] + β(1 − s)Et [Et+1 − Ut+1 ]
| {z }
foregone search value

• The firm surplus is J − V

I Impose free entry: Vt = 0
I So the firm surplus is just J and satisfies:

Jt = zt − wt + β(1 − s)Et [Jt+1 ]

• The match surplus is S = E + J − (U + V ) = E + J − U and satisfies

 
St = zt − b + βft Et [Et+1 − Ut+1 ] + β(1 − s)Et [St+1 ]

I Joint surplus independent from wages: Transferable utility!

I Pervasive property of search models with linear payoffs
I Useful for wage determination
10 / 31
Free Entry

• Free entry of firms implies Vt = 0 and so:

c = βqt Et [Jt+1 ]

I Vacancy creation adjusts until the cost equals expected future benefits
I Equilibrating force through qt = m (vt /ut )−α

11 / 31
Wage Determination
Non Competitive Labor Market

• The labor market is not competitive

I Competitive market: reallocation between buyers/sellers is instantaneous
F Arbitrages away any price differences
I Here reallocation takes time: labor market frictions (matching function)

• Cannot use the usual notion of labor market clearing to determine wages

• Instead, workers and firms face a bilateral monopoly when they meet
I If firm walks away, worker loses wage this period and must search again
I If worker walks away, firm loses profits this period and must search again

• Need to specify a resolution of the bilateral monopoly

I Several approaches, start with most common one today

12 / 31
Generalized Nash Bargaining
• Firms have no commitment power: cannot announce wage and stick to it

• Instead wage is determined after meeting occurs

• Renegotiated every period t

• Wage such that surplus is split between worker and firm with shares γ, 1 − γ:

wt = argmax(Et (w ) − Ut )γ Jt (w )1−γ

I γ called bargaining power of worker

I Microfoundation through alternative offer game à la Rubinstein (1980)

• Equilibrium wage satisfies the surplus-sharing equation

Et (wt ) − Ut = γSt , Jt (wt ) = (1 − γ)St

I Useful that joint surplus independent from wage

13 / 31
Equilibrium Wage
• Surplus-sharing rule implies:
 
wt = (1 − γ) b + βft γEt [St+1 ] + γzt

Weighted average between

I Worker’s flow outside option b plus foregone option value of search
I Firm’s flow output zt

• Can use free-entry to express βft (1 − γ)Et [St+1 ] = θt c:

wt = (1 − γ)b + γ(zt + θt c)

Equivalently weighted average between

I Worker’s flow outside option b
I Firm’s flow output zt plus cost-saving from not advertising jobs

• Wages are flexible because renegotiated every period

I Search frictions and price rigidities are distinct departures from RBC
14 / 31
Equilibrium: Tightness and Unemployment

• Given surplus sharing, re-write Bellman equation for joint surplus

St = zt − b + β(1 − s − γft )Et [St+1 ]

• Free-entry determines tightness θt :

c = βqt (1 − γ)Et [St+1 ]

• Unemployment follows the stock-flow equation

ut+1 − ut = s(1 − ut ) − ft ut

15 / 31
Comparative Statics
Steady-State

• Impose zt ≡ z and ut ≡ u at all time periods

I Surplus equation: S = z − b + β(1 − s − γf )S
I Free-entry: c = βq(1 − γ)S

• Combine to summarize steady-state by

I Job creation condition
r + s + γf (θ) z −b
= , r ≡ β −1 − 1
(1 − γ)q(θ) c
I Stock-flow identity/Beveridge curve
s
u=
s + f (θ)

16 / 31
Steady-State: First Pass (1)
• To gain intuition, log-linearize w.r.t. z:

1−γ dz
dθ =
γ + α(r + s)/f c

I Log-linearization as in RBC/NK with Marty

• Can also log-linearize w.r.t. r , s  f (θ):

1−γ z −b
θ ≈
γ c
s 1 s α
log u ≈ log − (1 − α) log θ ⇐⇒ log v ≈ log − log u
m 1−α m 1−α

I Here did not log-linearize w.r.t. aggregate shocks

I Instead log-linearized w.r.t. parameters r , s, m
F Motivated by the data: r ∼ 0.004, s ∼ 0.01, f ∼ 0.2 monthly
I Sometimes easier and/or leads to cleaner expressions

17 / 31
Steady-State: First Pass (2)

• We use the second log-linearization:

1−γ z −b 1 s α
θ≈ , log v ≈ log − log u
γ c 1−α m 1−α

• More vacancies relative to unemployed workers if

I Firm profit share 1 − γ is large
I Productivity z relative to unemployment value b is large
I Vacancy cost c is low

18 / 31
The Beveridge Curve: Data & DMP

9
8.5
log vacancies
8
7.5

1 1.5 2 2.5 3

log unemployment rate

2001Q1-2008Q4 2009Q1-2020Q1

1 s α
log v ≈ log − log u
1−α m 1−α
• With α ≈ 0.6 Beveridge curve has slope ≈ −1 as in data
I Is this surprising? Not so much
I Beveridge curve & matching fct regressions fit combinations of (log v , log u)
• Shift of Beveridge curve during Great Recession: m ↓
19 / 31
Steady-State: Comparative Statics w.r.t. Productivity
• Log-linearize w.r.t. z:
du r + s + γf z dz
= − (1 − u) (1 − α)
u α(r + s) + γf z −b z
| {z }
≈1 when r ,sf

• Log-linearize w.r.t r , s  f :
du z dz
u = cste × (z − b)−(1−α) =⇒ = −(1 − α)
u z −b z
• Steady-state comparative statics also informative about effect of shocks
I Recall that transitions in stock-flow model are virtually instantaneous

• In data
du
I
u
∼ 20 − 30%
Y dz
I If use labor productivity z = N
, z
∼ 2%

• To match data, need

z b
∼ 20 ⇐⇒ ∼ 95%
z −b z
20 / 31
The Unemployment Volatility Puzzle

• Shimer (2005): for plausible values of b, DMP model fails quantitatively

du
I From unemployment insurance b/z ∼ 0.4 ⇒ u
∼ 1%

21 / 31
A First Resolution?

• Hagedorn Manovskii (2008): high b reconciles DMP model and data

1
I Relies heavily on functional form and divergence of 1−b/z
close to 1

• Chodorow-Reich Karabarbounis (2016): careful measurement → b/z ∼ 0.05

I Incorporate unemployment insurance take-up, expiration, etc.
I Look at effective rates rather than statutory rates

22 / 31
Unpacking the Unemployment Volatility Puzzle

• Unemployment varies too little because tightness reacts too little

1
θ ≈ × (1 − γ)(z − b)
γc

I And so vacancies also react too little relative to data

• Key to vacancy creation are

I Free entry: ≈
I Firm’s profits: (1 − γ)(z − b)

• Explore role of both

23 / 31
Job Creation
Free Entry and Job creation Costs (1/2)

• Now suppose fixed measure M of firms enter each period

• Instead of linear vacancy cost cv , face strictly convex cost C (v ) = c0 v 1+ν

1+ν
I When C (v ) is linear (ν = 0) get back previous model
I With linear costs M vs. V irrelevant

• Previous equations hold after replacing c with c(v ) = C 0 (v )

I After adjusting for M in tightness

• Free entry now

c0 vtν = βqt (1 − γ)Et St+1

24 / 31
Free Entry and Job creation Costs (2/2)

• Key results becomes, to leading order when r , s  f (θ):

1−γ z −b du 1−α z dz
θ1+αν ≈ cste , ≈−
γ c0 u 1 + αν z − b z

I Tightness/unemployment even less responsive

I In expansions, hiring cost rises, dampening job creation
I In downturns, hiring cost declines, stimulating job creation

• Departures from linear cost do not resolve the unemployment volatility puzzle
I Convex costs make tightness even less responsive to shocks
I Needed tightness more responsive to shocks!

25 / 31
Wage Rigidity
Wage Rigidity in the DMP Model

• Return to linear cost model

• Recall so far, search frictions 6= wage rigidity

I Search ⇒ bilateral monopoly, resolved with bargaining
I Wage flexibly renegotiated every period
I Implications rely heavily on generalized Nash bargaining

• Introducing wage rigidity amounts to dropping/changing bargaining protocol

I Hall (2005)

26 / 31
Another Wage Setting Mechanism
• Any agreed upon wage must make
I Workers better off than being unemployed
I Firms better off than being vacant

• Any wage w such that

Et (w ) ≥ Ut
Jt (w ) ≥ 0

satisfies workers’ and firms’ participation constraints

I Under any such wage, matches are formed and no surplus is left on the table

• Any constant wage w is participation compatible if

b ≤ w ≤ inf (1 − β(1 − s))J t , where J t = zt + β(1 − s)Et [J t+1 ]

t

I Follows from simple algebraic manipulations of value functions

I If zt is constant get b ≤ w ≤ z
27 / 31
Hall’s Resolution of the Unemployment Volatility Puzzle

1. Choose a “long-run” wage w

I Hall (2005) uses the steady-state symmetric Nash bargaining wage (γ = 0.5)
I Idea: long-run level of wages determined by negotiations
I Obtains w /z = 0.96 in steady-state

2. Impose that wages fixed in the short-run: wt ≡ w

I Even in response to productivity shocks dzt
I Different from generalized Nash bargaining each period
I Wages are rigid

du
3. Obtains u ∼ 0.2 as in the data

28 / 31
Why Do Rigid Wages Work?
• No “surplus sharing” out of steady-state: Jt 6= γSt

• Can solve for Jt given fixed wage w

• Job creation now c = βqt Et Jt+1

• Job creation + Beveridge curve now imply when r , s  f

!
du 1 − α z dz du z dz
≈− v.s. ≈ −(1 − α)
u α z −w z u z −b z

• Two main differences

I 1/α: because worker’s wage does not depend on outside option anymore
F Bargained wage ⇒ ↑ in booms ⇒ profits ↓ ⇒ dampens vacancy creation
z z
I
z−w
instead of z−b
F Most of the action
F By fixing wage, relevant profits are z − w instead of (1 − γ)(z − b)
F Highlights that failure of bargaining model highly dependent on functional form

• Hall Milgrom (2008) microfound “rigid” wages with different bargaining

29 / 31
Taking Stock

• DMP model (with bargaining)

I Qualitatively speaks to facts about unemployment
I Quantitatively struggles to match unemployment volatility w/ prod. shocks

• Saw two resolutions today

I b/z ≈ 1: hard to justify
I Rigid wages: more promising
F Relevant if applies to wage of new hires, not incumbent workers
F Non-trivial to micro-found (Hall Milgrom 2008)
F Non-trivial to reconcile with wage cyclicality in data

• Next lecture: will study efficiency properties of DMP model

30 / 31
Bonus
Other resolutions

• Different bargaining (Hall Milgrom 2008)

I Observe that standard alternating offer solution is not a PBE
I Going back to unemployment is not credible while bargaining
I Alternative bargaining that reduces wage dependence on outside options
F “Microfounded” wage rigidity

• On-the-job search (Fujita Ramey 2012)

I Anchors expected hiring cost to existing wages ≈ wage rigidity

• Marginal vs. average workers (Elsby Michaels 2013)

I Firms have DRS, choose size to equate marginal product of labor to b
I As if b/z ≈ 1 endogenously

• Stock market valuations shocks rather than productivity (Kehoe et al. 2021)
I z should reflect firm’s valuation of jobs, stock market fluctuates much more

31 / 31