Lecture 11:

Firm Dynamics

Patrick Macnamara
ECON80301: Advanced Macroeconomics
University of Manchester

Fall 2024
Introduction

• Today, we will continue with more applications of dynamic programming.

• Let’s consider models with heterogeneity on the firm side.
In particular:
• Hopenhayn (1992): idiosyncratic productivity shocks + endogenous entry/exit
• Hopenhayn and Rogerson (1993): additional feature is labor adjustment costs.

• I’ll focus on Hopenhayn and Rogerson (1993), as Hopenhayn (1992) is a special

case of this paper.

2 / 37
Hopenhayn and Rogerson (1993)

3 / 37
Overview of Hopenhayn & Rogerson (1993)

• No aggregate uncertainty
• Perfect competition + Decreasing Returns to Scale (DRS)
• Firms are subject to idiosyncratic productivity shocks.
• Two costs:
1. Firms must pay a fixed cost to operate.
2. Firms pay a labor adjustment cost.
• Firms will decide optimally when to exit and how many workers to hire
(or lay off).
• While individual firms may be growing/contracting, entering/exiting, the
distribution of firms will be constant over time.
• Hopenhayn (1992) is a special case of this paper (i.e., when there are no labor
adjustment costs).

4 / 37
Production

• Labor is the only input to production:1

y = f (s, n′ )

where s is idiosyncratic productivity, and n′ is the labor input.

I’m following the notation of Hopenhayn and Rogerson (1993).
n′ is today’s labor input, chosen today.
• Assume decreasing returns to scale (DRS)
e.g., f (s, n) = snθ , where θ ∈ (0, 1).
With perfect competition (and no labor adjustment cost), we need DRS for
heterogeneity to exist in equilibrium.

1
The production function can easily be generalized to incorporate capital.
5 / 37
Profits

• Given previous level of employment, n, firms choose today’s labor input, n′ ,

taking as given the wage w . Current profits are then:
labor adjustment cost
π(s, n, n ; w ) = f (s, n ) − wn − cf − wg(n′ , n)
′ ′ ′

fixed operating cost

(generates exit)
• Assume the labor adjustment cost is:

g(n′ , n) = τ max(0, n − n′ )

Firm pays a tax when it reduces the level of employment (the tax revenue is
then rebated back to households as a lump-sum transfer).

6 / 37
Productivity

• Assume productivity s follows an AR(1) process:

ln s ′ = (1 − ρ)s̄ + ρ ln s + ε, ε ∼ N(0, σε2 )

• Some notation:
H(s ′ |s) is the conditional cdf for s ′ , given the current s h(s ′ |s) is the
conditional pdf for s ′ , given the current s
• In the numerical simulations, I discretized this process using Tauchen (1986).
However, for exposition purposes, let’s continue to assume s follows a
continuous AR(1) process.

7 / 37
Timing Assumptions

receive π(st , nt−1 , nt ; w )

learn st−1 learn st learn st+1

t −1 t t +1

entry/exit entry/exit entry/exit

decisions decisions decisions

• Exit decision is made prior to observing next period’s shock.

• Some firms operate today, knowing they will exit tomorrow.

8 / 37
Exit

• Let V (s ′ , n′ ; w ) denote the value of operating tomorrow, given tomorrow’s

productivity s ′ and initial employment n′ .
Notice that n′ is chosen today, but firm doesn’t learn s ′ until tomorrow.
• Firm exits iff
E [ V (s ′ , n′ ; w )| s] < −wg(0, n′ )
If firm exits, it must pay the adjustment cost to bring employment to zero.

9 / 37
Productivity Threshold for Exit

E[V(s',n')|s]
-wg(0,n')

exit cutoff
log s

Define a productivity threshold, s̄(n′ ), such that the firm operates iff s ≥ s̄(n′ ) and
exits iff s < s̄(n′ ):
E [ V (s ′ , n′ ; w )| s = s̄] = −wg(0, n′ )

10 / 37
Incumbent Firm’s Problem

• V (s, n; w ) solves the following Bellman equation:

V (s, n; w ) = max {π(s, n, n′ ; w ) + β max (E [ V (s ′ , n′ ; w )| s] , −wg(0, n′ ))}

n′
value of incumbent with tomorrow, exit decision made
prod. s and emp. n last period based on today’s productivity
• Associated policy functions:
Let n′ (s, n) be the policy function for employment.
Let χ(s, n) be the exit policy function:
(
0 if s ≥ s̄(n′ (s, n))
χ(s, n) =
1 if s < s̄(n′ (s, n))

• If τ = 0, the labor decision is static each period and the model reduces to
Hopenhayn (1992).

11 / 37
Entry
• Entry assumptions:
1. Potential entrant pays entry cost ce .
2. AFTER paying entry cost, entrant draws initial productivity s from distribution
with pdf ν(·).
• Let Me denote the mass of entrants.
• Free entry (FE) condition is imposed:

holds with equality

Z
V (s, 0; w )ν(s)ds − ce ≤ 0
if Me > 0
new entrant is like an incumbent
with productivity s and n = 0
• FE condition pins down the wage in equilibrium.2 Entry occurs until the value
of entry is zero.

2
This is a convenient property. If we add aggregate uncertainty, we don’t need to use Krusell
and Smith (1998).
12 / 37
Distribution of Firms

• Let µ(s, n) represent the distribution of firms in the current period, AFTER
entry/exit decisions AND after the realization of current productivity.
This defines the complete set of firms who operate today, including entrants
and the firms that will exit tomorrow.
• Given this distribution, we can compute aggregate statistics.
e.g., aggregate labor demand:
Z
N d (w ) = n′ (s, n; w )dµ(s, n)

13 / 37
Law of Motion for Distribution
• Let T be the law of motion for µ:

µ′ = T (µ, Me , n′ (s, n; w ), χ(s, n; w ))

where T is defined by
Z
µ′ (s ′ , n′ ) = Q(s ′ , n′ , s, n)dµ(s, n) + Me ν(s ′ )1 {n′ = 0} ∀(s ′ , n′ )

and Q(s ′ , n′ , s, n) is the probability of transitioning from (s, n) to (s ′ , n′ ):

Q(s ′ , n′ , s, n) = h(s ′ |s) (1 − χ(s, n)) 1 {n′ (s, n) = n′ }

• Given Me and the policy rules n′ (s, n; w ) and χ(s, n; w ), we can iterate on this
law of motion to solve for the stationary distribution (more on this later):

µ = T (µ, Me , n′ (s, n), χ(s, n))

14 / 37
Households

• There is a representative household with preferences

∞
β t [u(ct ) − ψNt ]
X

t=0

• In a stationary steady state with constant prices and an interest rate satisfying
β(1 + r ) = 1, the household’s optimization problem reduces to a static
optimization problem:

max u(c) − ψN subject to c = wN + Π + R

c,N

where Π is aggregate profits and R is aggregate tax receipts.

• Solve consumer’s problem to obtain N s = N s (w , Π + R).

15 / 37
Stationary Recursive Equilibrium, 1/2
A stationary recursive equilibrium is defined as
(i) the value function V (s, n),
(ii) the policy functions n′ (s, n) and χ(s, n),
(iii) the exit cutoff s̄(n′ ),
(iv) the mass of entrants Me ,
(v) the distribution µ(s, n), and
(vi) the wage w
such that
(1) Given the wage w , V (s, n) solves the firm’s Bellman equation, where n′ (s, n) is
the associated labor policy function. The exit rule, s̄(n′ ), satisfies
E [V (s ′ , n′ )|s̄] = −wg(0, n′ )
and the associated exit policy function χ(s, n) is given by:
(
0 if s ≥ s̄(n′ (s, n))
χ(s, n) =
1 if s < s̄(n′ (s, n))
16 / 37
Stationary Recursive Equilibrium, 2/2
(2) Given the wage w , the FE condition is satisfied:
Z
V (s, 0)ν(s)ds ≤ ce

This should hold with equality if Me > 0.

(3) µ is a stationary distribution, given Me and the policies n′ (s, n), χ(s, n):

µ = T (µ, Me , n′ (s, n), χ(s, n))

(4) The wage w clears the labor market:

Z
d
N (w ) = n′ (s, n)dµ(s, n) = N s (w , Π + R)

where
Z Z Z
Π+R = f (s, n′ (s, n))dµ − w n′ (s, n)dµ − cf dµ − Me ce

17 / 37
Solution Method, 1/4

• In practice, how do we solve for the equilibrium?

• Two important conditions:
Z the distribution of initial productivity draws for entrants.

V (s, 0; w )ν(s)ds = ce (FE)

Z
n′ (s, n; w )dµ(s, n) = N s (w , Π + R) (LM)

18 / 37
Solution Method, 2/4

LM

w∗ FE

Me∗ Me

• FE condition pins down equilibrium wage w ∗ .

• Given w ∗ , we can use LM condition to pin down the equilibrium mass of
entrants, Me∗ .
A higher Me increases total labor demand but reduces per-firm labor demand
(through competition and wage adjustments)

19 / 37
Solution Method, 3/4

• The first step is to solve for the wage w ∗ which satisfies the FE condition:
Z
V (s, 0; w ∗ )ν(s)ds = ce (1)

• To do this, we need to repeatedly solve the firm’s Bellman equation for different
values of the wage until we have found the wage w ∗ at which (1) is satisfied.
• Notice we can solve for the wage w ∗ without having to solve for the invariant
distribution of firms.

20 / 37
Solution Method, 4/4

• Given w ∗ and the policy functions n′ (s, n) and χ(s, n), we need to solve for the
mass of entrants Me and the distribution µ(s, n) such that
1. the labor market clears at the wage w ∗ :
Z
n′ (s, n)dµ(s, n) = N s (w ∗ , Π + R)

2. and µ is a stationary distribution:

µ = T (µ, Me , n′ (s, n), χ(s, n))

• Fortunately, we don’t need to repeatedly solve for the stationary µ for different
values of Me .

21 / 37
Solution Method: Solving for the Distribution, 1/2

• We can solve for the stationary distribution up to a scale factor.

• Define µ̂ as the stationary distribution given Me = 1 and n′ (s, n) and χ(s, n):

µ̂ = T (µ̂, Me = 1, n′ (s, n), χ(s, n))

• Claim: given Me , n′ (s, n) and χ(s, n), the stationary distribution is

µ = Me µ̂

Proof: see the next slide.

22 / 37
Solution Method: Solving for the Distribution, 2/2

• If µ̂ = T (µ̂, Me = 1, n′ (s, n), χ(s, n)), then:

Z
′ ′
µ̂(s , n ) = Q(s ′ , n′ , s, n)d µ̂(s, n) + 1 × ν(s ′ )1(n′ = 0)

• Suppose µ = Me µ̂ is today’s distribution and the mass of entrants is now Me .

Z
µ′ (s ′ , n′ ) = Q(s ′ , n′ , s, n)dµ(s, n) + Me × ν(s ′ )1(n′ = 0)
Z
= Me Q(s ′ , n′ , s, n)d µ̂(s, n) + Me ν(s ′ )1(n′ = 0)
Z 
= Me Q(s ′ , n′ , s, n)d µ̂(s, n) + ν(s ′ )1(n′ = 0)
= Me µ̂(s ′ , n′ ) = µ(s ′ , n′ )

• Therefore, µ = Me µ̂ is a stationary distribution for Me .

23 / 37
Solution Method: Solving for Me , 1/3

• We can solve for µ̂, and then use that µ = Me µ̂ to quickly solve for the Me
which clears the labor market.
Z
n′ (s, n)dµ(s, n) = N s (w ∗ , Π + R)

where N s is obtained by solving

max u(c) − ψN subject to c = wN + Π + R

c,N

• Example: suppose u(c) = ln c. Then the solution to the household’s problem is

1 Π+R
N s (w , Π + R) = −
ψ w

24 / 37
Solution Method: Solving for Me , 2/3

• Note also that

Π + R = Y − wN d − Mcf − Me ce

where
Z Z Z
′ d ′
Y = f (s, n (s, n))dµ, N = n (s, n)dµ, M= dµ

• We can factor out Me :  

Π + R = Me Π̂ + R̂
where
Π̂ + R̂ = Ŷ − w N̂ d − M̂cf − ce
Z Z Z
Ŷ = f (s, n′ (s, n))d µ̂, N̂ d = n′ (s, n)d µ̂, M̂ = d µ̂

25 / 37
Solution Method: Solving for Me , 3/3

• Labor market clearing condition:

Z
1 Π+R
n′ (s, n)dµ(s, n) = N s (w , Π + R) = −
ψ w
• Use µ = Me µ̂ and w = w ∗ :
Z
1 Π+R
Me n′ (s, n)d µ̂ =−
ψ w∗
1 Me (Π̂ + R̂)
Me N̂ d = −
ψ w∗
• Solve for Me to obtain
1/ψ
Me =
N̂ d + (Π̂ + R̂)/w ∗

26 / 37
Solution Method: Summary

(1) Solve for the wage w ∗ which satisfies the FE condition

Z
V (s, 0; w ∗ )ν(s)ds = ce

With this wage, we will have solved for the value function V (s, n) and the
associated policy functions n′ (s, n) and χ(s, n).
(2) Solve for the invariant distribution up to a scale factor.
i.e., taking as given n′ (s, n) and χ(s, n), assume Me = 1 and solve for the
invariant distribution µ̂.
(3) Solve for the Me which guarantees that w ∗ is the labor market clearing wage.
(4) Compute the invariant distribution as µ = Me µ̂.

27 / 37
Value Functions (τ > 0)

1500
low s
1400 medium s productivity
high s

1300

1200
v(s,n)

1100

1000
optimal current employment is
greater than n
n' > n
if initial n is high, n' < n,
900 incur adgjustment cost of lay-off

800
0 500 1000 1500 2000
n

28 / 37
 Optimal Labor Policies

1500
low s, > 0
no adjustment cost vs with cost
low s, = 0
high s, > 0
high s, = 0

1000
n'

n' < n
500 adjustment cost hold you from firing workers,
so the blue line has more workers

n' > n, hire more n = n'

adjustment cost does not make them hire more
because there is potential cost of laying off in the future
for high n, choose high n'

0
0 500 1000 1500
n

29 / 37
Distribution of Firms and Employment
only most productive firms survive
0.08
invariant dist, productivity large firms hire more workers
0.07 firm shares
employment shares
0.06

0.05

0.04

0.03

0.02

0.01

0
-0.5 0 0.5 1 1.5 2 2.5 3 3.5
log productivity

Selection increases average productivity of operating firms.

30 / 37
Main Results labor market protections reduce wages, consumption and output
productive firms become more cautious, so dont hire workers as much as they should, reduce employment

τ = 0 τ = 0.1 τ = 0.2
Wage 1.000 0.975 0.954
Consumption (output) 100 97.5 95.4
Average productivity 100 99.2 97.9
Average firm size 61.2 61.8 65.1
Job turnover rate3 0.30 0.26 0.22
Note: these numbers are taken from Table 3 in Hopenhayn and Rogerson (1993).

• Higher tax τ reduces aggregate employment, reduces average productivity, and

reduces job turnover rate.
• Tax on firing causes firms to be more cautious on hiring, reducing overall
employment.
• Resources are used less efficiently and there is misallocation.
3
The job turnover rate is either the job creation rate or the job destruction rate, which must be
the same in a stationary equilibrium. Job creation rate = total jobs created, as a fraction of total
employment. Job destruction rate = total jobs destroyed, as a fraction of employment.
31 / 37
Further Applications

32 / 37
International Trade

• Melitz (2003) adapts firms dynamics model of Hopenhayn (1992) to allow for
international trade.
• Key ingredients:
• Monopolistic competition, continuum of heterogeneous firms produce
differentiated varieties.
• All firms share the same fixed operation cost f > 0, but are heterogeneous over
productivity.
• Additional fixed cost to export, fx > 0 + iceberg trade cost.
• Endogenous entry and exit + exogenous exit. some fraction of firms are forced to exit and entry
• Exposure to international trade induces only the most productive firms to
export, while simultaneously forcing the least productive firms to exit. This
reallocates market shares toward the more productive firms and increases
aggregate productivity.

33 / 37
Financial Constraints, 1/2

• Hennessy and Whited (2005, 2007) develop a “dynamic trade-off” theory to

model the capital structure of firms.4
• Firms choose investment, financed by debt and equity.
• Firms make dynamic decisions whether to raise funds directly from shareholders
via external equity (which is costly), or distribute funds to shareholders as
dividends (which is taxed).
• Structurally estimate model via Simulated Method of Moments (SMM).
• Hennessy and Whited (2005) matches some empirical facts regarding firm
financing.
• Hennessy and Whited (2007) estimates the magnitude of external financing
costs via SMM (marginal equity flotation costs around 9.1%, larger for small
firms).

4
An alternative is pecking-order theory, which is an ad-hoc, non-maximizing theory whereby
firms first prefer internal equity, then debt, lastly raising external equity.
34 / 37
Financial Constraints, 2/2

• Khan and Thomas (2013) study the effect of a credit shock in an economy with
two frictions:
• Partial irreversibility in investment, whereby:

k ′ = (1 − δ)k + i if i ≥ 0
′
k = (1 − δ)k + i/θk with θk ∈ (0, 1) if i < 0
get a bad price for selling capital,
so plus exogeneous borrow constaint,
• Collateralized borrowing constraint, b ′ ≤ ζθk k can generate recession

They find that negative shock to ζ can generate large & persistent recession.
• Karabarbounis and Macnamara (2021) study misallocation of capital in a model
where firms face different types of financial constraints along the life cycle.
• Firms borrow short-term via collateral constraint when young (smaller), then via
long-term debt with no collateral when older (bigger).
• The two financial constraints have different effects on productive firms.
productive firms get the best price, less borrow constraint?

35 / 37
Monetary Policy

• Ottonello and Winberry (2020) add a New Keynesian component to a model of

basically look at heterogeneity at the firm side
firm dynamics to study to the effect of monetary policy (e.g., Heterogeneous
Agent New Keynesian (HANK) model with financial frictions).
• Key ingredients:
• idiosyncratic shocks to productivity and capital quality
• Individual firms face borrowing costs which reflect the probability of default.
• Exogenous exit + endogenous exit when firms default.
• Firms cannot issue external equity, but can build up net worth ⇒ rich
heterogeneity in net worth across firms.
• Key result: Investment of firms with low default risk are the most responsive to
monetary policy.
heterogeneity is important when we look at the economy

36 / 37
What’s Next?

• Homework 8 due at end of exam period, but do not wait until the last minute.

37 / 37