n?u=RePEc:red:sed016:740&r=sog

**Agency** **Business** **Cycles** ∗

Mikhail Golosov

Princeton University and NBER

Guido Menzio

University of Pennsylvania and NBER

November 2015

Abstract

We propose a new business cycle theory. Firms need to randomize over firing or

keeping workers who have performed poorly in the past, in order to give them an

ex-ante incentive to exert effort. Firms have an incentive to coordinate the outcome

of their randomizations, as coordination allows them to load the firing probability on

states of the world in which it is costlier for workers to become unemployed and, hence,

allows them to reduce overall agency costs. In the unique robust equilibrium, firms

use a sunspot to coordinate the randomization outcomes and the economy experiences

endogenous, stochastic aggregate fluctuations.

JEL Codes: D86, E24, E32.

Keywords: Unemployment, Moral Hazard, Endogenous **Business** **Cycles**.

∗ Golosov: Department of Economics, Princeton University, 111 Fisher Hall, Princeton, NJ 08544 (email:

golosov@princeton.edu); Menzio: Department of Economics, University of Pennsylvania, 3718 Locust Walk,

Philadelphia, PA 19104 (email: gmenzio@sas.upenn.edu). We are grateful to Paul Beaudry, David Berger,

Katarina Borovickova, V.V. Chari, Veronica Guerrieri, Christian Haefke, Kyle Herkenhoff, John Kennan,

Philipp Kircher, Narayana Kocherlakota, Ricardo Lagos, Rasmus Lentz, Igor Livschits, Nicola Pavoni, Thijs

van Rens, Guillaume Rocheteau, Karl Shell, Robert Shimer and Randy Wright for comments on earlier drafts

of the paper. We are also grateful to participants at seminars at the University of Wisconsin Madison, New

York University, University of Chicago, New York FRB, Minneapolis FRB, and at the Search and Matching

European Conference in Aix-en-Provence, the Search and Matching Workshop at the Philadelphia FRB, the

conference in honor of Chris Pissarides at Sciences Po, the Econometric Society World Congress in Montreal,

the NBER EFG Meeting at the New York FRB.

1

1 Introduction

We present a new business cycle theory, where aggregate fluctuations are caused by the

fact that the agents in the economy randomize over some individual decision in a correlated

fashion. We consider a model where firms need to randomize over firing or keeping workers

who have performed poorly in the past, in order to give them an ex-ante incentive to exert

effort. Firms have a desire to coordinate the outcome of their randomizations, as coordination

allows them to load the firing probability on states of the world in which it is costlier for

workers to become unemployed and, hence, it allows them to reduce the overall agency

costs. In the unique robust equilibrium of the model, firms use a sunspot to coordinate

outcome of their randomization and the economy experiences aggregate fluctuations that

are endogenous– in the sense that they are not driven by exogenous shocks to fundamentals

or by exogenous shocks to the selection of equilibrium– and stochastic– in the sense that

they follow a non-deterministic path. Our theory of business cycles implies a novel view

of recessions which is opposite to view of recessions as “rainy days” proposed by the Real

**Business** Cycle theory of Kydland and Prescott (1982).

The theory is cast in the context of a search-theoretic model of the labor market in the

spirit of Pissarides (1985) and Mortensen and Pissarides (1994). In particular, we consider

a labor market populated by identical risk-averse workers– who look for vacancies when

unemployed and produce output when employed– and by identical risk-neutral firms– that

attract new workers by posting vacancies and produce according to a technology that has

constant returns to scale in labor. Unemployed workers and vacant firms come together

through a frictional process that is summarized by a matching function. Matched workerfirm

pairs produce under moral hazard: the firm does not observe the worker’s effort but

only his output, which is a noisy signal of effort. Matched firm-worker pairs Nash bargain

over the terms of an employment contract that specifies the level of effort recommended to

the worker, the wage paid by the firm to the worker, and the probability with which the

worker is fired by the firm conditional on the output of the worker and on the realization of

a sunspot, an inherently meaningless signal that is observed by all market participants.

The theory builds on two assumptions. First, the theory needs some decreasing returns to

matching in the labor market. Decreasing returns to matching may either come directly from

decreasing returns to scale in the matching function, or they may come from a vacancy cost

that increases with the total number of vacancies in the market. Second, the theory needs

employment contracts to be incomplete enough that firing takes place along the equilibrium

path. In this paper, we simply assume that current wages are paid before observing output

2

and employment contracts are renegotiated period by period, so that firing is the only tool

that firms can use to give workers an incentive to exert effort.

In the first part of the paper, we characterize the properties of the optimal employment

contract. We show that the optimal contract is such that the worker is fired with positive

probability only when the output of the worker is low and the realization of the sunspot is

such that the gains from continued trade accruing to the worker are suffi ciently high relative

to the gains from continued trade accruing to the firm. The result is intuitive. Firing is

costly– as it destroyed a valuable firm-worker relationship– but it is necessary– as it is the

only way for the firm to give the worker an incentive to exert effort. However, when firing

takes place, it is only the value of the destroyed relationship that would have accrued to the

worker that gives incentives. The value of the destroyed relationship that would have accrued

to the firm is just “collateral damage.”The optimal contract minimizes the collateral damage

by loading the firing probability on the realizations of the sunspot for which the worker’s

continuation gains from trade are highest relative to the firm’s. In other words, the optimal

contract loads the firing probability on the states of the world where the cost to the worker

from losing the job is highest relative to the cost to the firm from losing the worker.

In the second part of the paper, we characterize the relationship between the realization

of the sunspot and firing in general equilibrium. We find that there is an equilibrium in

which firms fire all of their non-performing workers for some realizations of the sunspot,

and firms do not fire any of their non-performing workers for the other realizations. There

is a simple logic behind this finding. Suppose that firms load up the firing probability on

some realizations of the sunspot. In those states of the world, unemployment is higher

and, because of decreasing returns to matching, the job-finding probability of unemployed

workers is lower. In turn, if the job-finding probability is lower, the workers have a worse

outside option when bargaining with the firms and their wage is lower. If the wage is lower,

the workers’marginal utility of consumption relative to the firms’is lower and, according to

Nash bargaining, the workers’gains from trade relative to the firms’are higher. Hence, if the

other firms in the market load the firing probability on some realizations of the sunspot, an

individual firm has the incentive to load the firing probability on the very same states of the

world. In other words, firms have a desire to coordinate the outcome of the randomization

between firing and keeping their non-performing workers, and the sunspot allows them to

achieve coordination.

Naturally, alongside the equilibrium in which firms use the sunspot to coordinate on

firing or keeping non-performing workers, there are also equilibria in which firms (fully or

3

partially) ignore the sunspot and randomize on firing or keeping non-performing workers

independently. However, these equilibria are an artifact of the simplifying assumption that

all firms randomize simultaneously. Indeed, we find that, in a version of the model where

firms fire sequentially, only the equilibrium with perfect coordination survives. In this version

of the model, the perfect coordination equilibrium takes the form of a firing cascade where the

firing decisions of the first few firms uniquely determine the firing decision of all subsequent

firms.

In the equilibrium where firms coordinate on firing or keeping non-performing workers,

the economy experiences aggregate fluctuations. These aggregate fluctuations are endogenous.

Indeed, they are not caused by exogenous shocks to fundamentals, nor by exogenous

shocks to the selection of the equilibrium played by market participants. Instead, aggregate

fluctuations in our model are caused by coordinated randomization– i.e. every firm needs

to randomize on firing or keeping its non-performing workers, and different firms want to

coordinate their randomization outcomes. The aggregate fluctuations in our model are stochastic.

Indeed, the economy does not follow a deterministic limit cycle or a deterministic

chaotic map as in previous theories of endogenous business cycles. Instead, the economy follows

a stochastic process, in which the probability of a firing burst and, hence, of a recession

is an equilibrium outcome.

In the last part of the paper, we calibrate the model to measure the magnitude and

properties of **Agency** **Business** **Cycles** (ABC), i.e. the aggregate fluctuations experienced by

the economy in the equilibrium where firms coordinate on firing or keeping non-performing

workers. We find that ABC feature fluctuations in unemployment, in the rate at which

unemployed workers become employed (UE rate), and in the rate at which employed workers

become unemployed (EU rate) that are approximately half as large as those observed in the

US labor market and that– as it has been the case in the US labor market since 1984– are

uncorrelated with labor productivity.

We then test some of the distinctive features of ABC. First, in ABC, a recession starts

with an increase in the EU rate which leads to an increase in the unemployment rate. In

turn, the rise in the unemployment rate leads, because of decreasing returns to scale in

matching, to a fall in the UE rate. Hence, the EU rate leads both the unemployment rate

and the UE rate. We find that the US labor market features the same pattern of leads

and lags. Second, in ABC, the probability of a recession is endogenous and depends on the

aggregate state of the economy. Specifically, the lower is the unemployment rate, the higher

is the probability with which firms need to fire their non-performing workers in order to give

4

them an incentive to exert effort and, hence, the higher is the probability of a recession. We

find that the US labor market also features a negative relationship between unemployment

and the probability of a recession. Third, in ABC, a recession is a period when the value

of time in the market relative to the value of time at home is abnormally high. Using an

admittedly basic approach, we construct a time-series for the net value of a job to a worker

and we find this series to be strongly countercyclical. The finding is consistent with Davis

and Von Wachter (2011) who document that the cost to a worker from losing a job is strongly

countercyclical.

The first contribution of the paper is to advance a novel theory of business cycles, where

aggregate fluctuations are endogenous and stochastic and emerge because different market

participants have to randomize over some decision and find it optimal to coordinate the

randomization outcomes. In the business cycle literature, there are theories where aggregate

fluctuations are driven by exogenous shocks to the current value of fundamentals (e.g.,

Kydland and Prescott 1982 or Mortensen and Pissarides 1994), to the future value of fundamentals

(e.g., Beaudry and Portier 2004 or Jaimovich and Rebelo 2009), or to the stochastic

process of fundamentals (e.g., Bloom 2009). In our theory, all fundamentals are fixed. There

are theories where aggregate fluctuations are driven by exogenous shocks to the selection of

the equilibrium played by market participants (e.g., Heller 1986, Cooper and John 1988 or

Benhabib and Farmer 1994). In our theory, market participants always play the same, unique

equilibrium. There are theories where aggregate fluctuations emerge endogenously as limit

cycles (e.g., Diamond 1982, Diamond and Fudenberg 1989, Mortensen 1999 or Beaudry, Galizia

and Portier 2015) or as chaotic dynamics (e.g., Boldrin and Montrucchio 1986 or Boldrin

and Woodford 1990). In our theory, the economy follow a stochastic process. There are theories

where aggregate fluctuations are driven by common shocks to higher-order beliefs (e.g.,

Angeletos and La’O 2013). In our theory there are no such shocks.

The second contribution of the paper is to advance a new view of recessions. In theories

where business cycles are caused by fluctuations in productivity– such as in the Real **Business**

Cycle theory of Kydland and Prescott (1982) or in Mortensen and Pissarides (1994)– a

recession is a period when the value of time in the market relative to the value of time at

home is abnormally low. Indeed, in these theories, a recession is a period when the output of

a worker in the market is unusually low. In our theory, a recession is a period when the value

of time in the market relative to the value of time at home is abnormally high. Indeed, in our

theory, a recession is a period when the output of a worker in the market is not relatively low,

but the value of staying at home looking for a job is unusually low. At first glance, the data

says that the net value of employment for a worker is countercyclical. But if the net value

5

of employment is countercyclical, why is there more unemployment in recessions? And why

is the rate at which workers lose their job higher and the rate at which unemployed workers

find a job lower? That is, if recessions are times when the gains from trade are high, why is

there less trade? Our theory provides an answer to this puzzle: when the gains from trade

in the labor market are high, firms find it optimal to get rid of their non-performing workers

and this creates congestions in the labor market that lowers the speed at which unemployed

workers find jobs.

2 Environment and Equilibrium

2.1 Environment

Time is discrete and continues forever. The economy is populated by a measure 1 of identical

workers. Every worker has preferences described by P β t [υ(c t ) − ψe t ], where β ∈ (0, 1) is

the discount factor, υ(c t ) is the utility of consuming c t units of output in period t, and ψe t

is the disutility of exerting e t units of effort in period t. The utility function υ(·) is strictly

increasing and strictly concave, with a first derivative υ ′ (·) such that υ ′ (·) ∈ [υ ′ , υ ′ ], and a

second derivative υ ′′ (·) such that −υ ′′ (·) ∈ [υ ′′ , υ ′′ ], with υ ′ > υ ′ > 0 and υ ′′ > υ ′′ > 0. The

consumption c t is equal to the wage w t if the worker is employed in period t, and to the

value of home production b if the worker is unemployed in period t. 1

The coeffi cient ψ is

strictly positive, and the effort e t is equal to either 0 or 1. Every worker is endowed with

one indivisible unit of labor.

The economy is also populated by a positive measure of identical firms. Every firm has

preferences described by P β t c t , where β ∈ (0, 1) is the discount factor and c t is the firm’s

profit in period t. Every firm operates a constant returns to scale production technology

that transforms one unit of labor (i.e. one employee) into y t units of output, where y t is a

random variable that depends on the employee’s effort e t . In particular, y t takes the value y h

with probability p h (e) and the value y l with probability p l (e) = 1 − p h (e), with y h > y l ≥ 0

and 0 < p h (0) < p h (1) < 1. Production suffers from a moral hazard problem, in the sense

that the firm does not directly observe the effort of its employee, but only the output.

Every period t is divided into five stages: sunspot, separation, matching, bargaining and

production. At the first stage, a random variable, z t , is drawn from a uniform distribution

1 As the reader can infer from the notation, we assume that workers are banned from the credit market and,

hence, they consume their income in every period. The assumption is made only for the sake of simplicity.

6

with support [0, 1]. 2

The random variable is aggregate, in the sense that it is publicly

observed by all market participants. The random variable is a sunspot, in the sense that it

does not directly affect technology, preferences or any other fundamentals, although it may

serve to coordinate the behavior of market participants.

At the separation stage, some employed workers become unemployed.

An employed

worker becomes unemployed for exogenous reasons with probability δ ∈ (0, 1). In addition,

an employed worker becomes unemployed because he is fired with probability s(y t−1 , z t ),

where s(y t−1 , z t ) is determined by the worker’s employment contract and it is allowed to

depend on the output of the worker in the previous period, y t−1 , and on realization of the

sunspot in the current period, z t . For the sake of simplicity, we assume that a worker who

becomes unemployed in period t can search for a new job only starting in period t + 1.

At the matching stage, some unemployed workers become employed. Firms decide how

many job vacancies v t to create at the unit cost k > 0. Then, the u t−1 workers who were

unemployed at the beginning of the period and the v t vacant jobs that were created by the

firms search for each other. The outcome of the search process is described by a decreasing

return to scale matching function, M(u t−1 , v t ), which gives the measure of bilateral matches

formed between unemployed workers and vacant firms. We denote v t /u t−1 as θ t , and we

refer to θ t as the tightness of the labor market. We denote as λ(θ t , u t−1 ) the probability that

an unemployed worker meets a vacancy, i.e. λ(θ t , u t−1 ) = M(u t−1 , θ t u t−1 )/u t−1 . Similarly,

we denote as η(θ t , u t−1 ) the probability that a vacancy meets an unemployed worker, i.e.

η(θ t , u t−1 ) = M(u t−1 , θ t u t−1 )/θ t u t−1 . We assume that the job-finding probability λ(θ t , u t−1 )

is strictly increasing in θ t and strictly decreasing in u t−1 and that the job-filling probability

η(θ t , u t−1 ) is strictly decreasing in both θ t and u t−1 . That is, the higher is the labor market

tightness, the higher is the job-finding probability and the lower is the job-filling probability.

However, for a given labor market tightness, both the job-finding and the job-filling

probabilities are strictly decreasing in unemployment. 3

At the bargaining stage, each firm-worker pair negotiates the terms of a one-period employment

contract x t . The contract x t specifies the effort e t recommended to the worker

in the current period, the wage w t paid by the firm to the worker in the current period,

and the probability s(y t , z t+1 ) with which the firm fires the worker at the next separation

2 Assuming that the sunspot is drawn from a uniform distribution with support [0, 1] is without loss in

generality.

3 Given any constant returns to scale matching function, the job-finding and the job-filling probabilities

are only functions of the market tightness. Given any decreasing returns to scale matching function, the

job-finding and the job-filling probabilities are also (decreasing) functions of unemployment.

7

stage, conditional on the output of the worker in the current period and on the realization

of the sunspot at the beginning of next period. We assume that the outcome of the bargain

between the firm and the worker is the Axiomatic Nash Bargaining Solution.

At the production stage, an unemployed worker home-produces and consumes b units of

output. An employed worker chooses an effort level, e t , and consumes w t units of output.

Then, the output of the worker, y t , is realized and observed by both the firm and the worker.

A few comments about the environment are in order. We assume that the employment

contract cannot specify a wage that depends on the current realization of the worker’s output.

Hence, the firm cannot use the current wage to give the worker an incentive to exert effort.

We also assume that the employment contract is re-bargained every period. Hence, the firm

cannot use future wages to give the worker an incentive to exert effort. Overall, firing is the

only tool that the firm can use to incentivize the worker. These restrictions on the contract

space are much stronger than what we need. Indeed, our theory of business cycles only

requires that principals sometimes fire their non-performing agents along the equilibrium

path. As we know from Clementi and Hopenhayn (2006), equilibrium firing may obtain

under complete contracts as long as the agent is protected by some form of limited liability.

We assume that the matching function M(u, v) has decreasing returns to scale. From the

theoretical point of view, one can justify the assumption by noting that the classic urn-ball

matching function with finite urns and finite balls has decreasing returns to scale (see, e.g.,

Burdett, Shi and Wright 2001). From the empirical point of view, it is easy to justify the

assumption, since estimating a Cobb-Douglass matching function for the US economy reveals

that the exponents on unemployment and vacancies sum up to less than 1. 4 Moreover, the

assumption is not critical. Indeed, the equilibrium conditions of our model are identical to

those of a model in which the matching function has constant returns to scale but the cost

of a vacancy is strictly increasing in the aggregate number of vacancies. 5

4 Petrongolo and Pissarides (2000) show that some empirical studies on the matching function have found

increasing returns to scale, some have found constant returns to scale and others have found decreasing

returns to scale depending on the data and on the estimation method. Menzio and Shi (2011) show that

the estimates of the matching functions are biased if– as most of the studies reviewed by Petrongolo and

Pissarides (2000) do– one abstracts from the fact that both employed and unemployed workers search for

and match with vacancies.

5 As the assumption of an increasing marginal cost of a vacancy is more common in than the assumption

of a decreasing returns to scale matching function, the reader may be more comfortable with this alternative

interpretation of the equilibrium conditions.

8

2.2 Equilibrium

We now derive the conditions for an equilibrium in our model economy. Let u denote the

measure of unemployed works at the beginning of the bargaining stage. Let W 0 (u) denote

the lifetime utility of a worker who is unemployed at the beginning of the production stage.

Let W 1 (x, u) denote the lifetime utility of a worker who is employed under the contract

x at the beginning of the production stage. Let W (x, u) denote the difference between

W 1 (x, u) and W 0 (u). Let F (x, u) denote the present value of profits for a firm that, at the

beginning of the production stage, employs a worker under the contract x. Let x ∗ (u) denote

the equilibrium contract between a firm and a worker when unemployment is u. Finally, let

θ(u, ẑ) denote the labor market tightness at the matching stage of next period, when the

current unemployment is u and next period’s sunspot is ẑ. Similarly, let h(u, ẑ) denote the

unemployment at the bargaining stage of next period, when the current unemployment is u

and next period’s sunspot is ẑ.

The lifetime utility W 0 (u) of an unemployed worker is such that

W 0 (u) = υ(b) + βEẑ [W 0 (h(u, ẑ)) + λ(θ(u, ẑ), u)W (x ∗ (h(u, ẑ)), h(u, ẑ))] . (1)

In the current period, the worker home-produces and consumes b units of output. At the

matching stage of next period, the worker finds a job with probability λ(θ(u, ẑ), u) in which

case his continuation lifetime utility is W 0 (h(u, ẑ))+W (x ∗ (h(u, ẑ)), h(u, ẑ)). With probability

1 − λ(θ(u, ẑ), u), the worker does not find a job and his continuation lifetime utility is

W 0 (h(u, ẑ)).

The lifetime utility W 1 (x, u) of a worker employed under the contract x = (e, w, s) is

such that

W 1 (x, u)

= υ(w) − ψe+

+βE y,ẑ [W 0 (h(u, ẑ)) + (1 − δ)(1 − s(y, ẑ))W (x ∗ (h(u, ẑ)), h(u, ẑ))|e] .

(2)

In the current period, the worker consumes w units of output and exerts effort e. At the

separation stage of next period, the worker keeps his job with probability (1−δ)(1−s(y, ẑ)),

in which case his continuation lifetime utility is W 0 (h(u, ẑ)) + W (x ∗ (h(u, ẑ)), h(u, ẑ)). With

probability 1 − (1 − δ)(1 − s(y, ẑ)), the worker loses his job and his continuation lifetime

utility is W 0 (h(u, ẑ)).

The difference W (x, u) between W 1 (x, u) and W 0 (u) represents the gains from trade to

a worker employed under the contract x. From (1) and (2), it follows that W (x, u) is such

9

that

W (x, u)

= υ(w) − υ(b) − ce+

+βE y,ẑ {[(1 − δ)(1 − s(y, ẑ)) − λ(θ(ẑ, u), u)]W (x ∗ (h(u, ẑ)), h(u, ẑ))|e} .

(3)

We find it useful to denote as V (u) the gains from trade for a worker employed under the

equilibrium contract x ∗ (u), i.e. V (u) = W (x(u), u). We refer to V (u) as the equilibrium

gains from trade accruing to the worker.

The present value of profits F (x, u) for a firm that employs a worker under the contract

x = (e, w, s) is such that

F (x, u) = E y [y|e] + βE y,ẑ [(1 − δ)(1 − s(y, ẑ))F (x ∗ (h(u, ẑ)), h(u, ẑ))|e] (4)

In the current period, the firm enjoys a profit equal to the expected output of the worker

net of the wage. At the separation stage of next period, the firm retains the worker with

probability (1 − δ)(1 − s(y, ẑ)), in which case the firm’s continuation present value of profits

is F (x ∗ (h(u, ẑ)), h(u, ẑ)). With probability 1 − (1 − δ)(1 − s(y, ẑ)), the firm loses the worker,

in which case the firm’s continuation present value of profits is zero. We find it useful to

denote as J(u) the present value of profits for a firm that employs a worker at the equilibrium

contract x ∗ (u), i.e. J(u) = F (x ∗ (u), u). We refer to J(u) as the equilibrium gains from trade

accruing to the firm.

The equilibrium contract x ∗ (u) is the Axiomatic Nash Solution to the bargaining problem

between the firm and the worker. That is, x ∗ (u) is such that

subject to the logical constraints

max W (x, u)F (x, u), (5)

x=(e,w,s)

e ∈ {0, 1} and s(y, ẑ) ∈ [0, 1],

and the worker’s incentive compatibility constraints

ψ ≤ β(p h (1) − p h (0))Eẑ [(1 − δ)(s(y l , ẑ) − s(y h , ẑ))V (h(u, ẑ))] , if e = 1,

ψ ≥ β(p h (1) − p h (0))Eẑ [(1 − δ)(s(y l , ẑ) − s(y h , ẑ))V (h(u, ẑ))] , if e = 0.

In words, the equilibrium contract x ∗ (u) maximizes the product between the gains from

trade accruing to the worker, W (x, u), and the gains from trade accruing to the firm, F (x, u),

among all contracts x that satisfy the worker’s incentive compatibility constraints. The first

incentive compatibility constraint states that, if the contract specifies e = 1, the cost to the

10

worker from exerting effort must be smaller than the benefit. The second constraint states

that, if the contract specifies e = 0, the cost to the worker from exerting effort must be

greater than the benefit. The cost of effort is ψ. The benefit of effort is given by the effect of

effort on the probability that the realization of output is high, p h (1) − p h (0), times the effect

of a high realization of output on the probability of keeping the job, (1−δ)(s(y l , ẑ)−s(y h , ẑ)),

times the value of the job to the worker, βV (h(u, ẑ)).

The equilibrium market tightness θ(u, ẑ) must be consistent with the firm’s incentives

to create vacancies. The cost to the firm from creating an additional vacancy is k. The

benefit to the firm from creating an additional vacancy is given by the job-filling probability,

η(θ(u, ẑ), u), times the value to the firm of filling a vacancy, J(h(u, ẑ)). The market tightness

is consistent with the firm’s incentives to create vacancies if k = η(θ(u, ẑ), u)J(h(u, ẑ)) when

θ(u, ẑ) > 0, and if k ≥ η(θ(u, ẑ), u)J(h(u, ẑ)) when θ(u, ẑ) = 0. Overall, the market tightness

is consistent with the firm’s incentives to create vacancies iff

k ≥ η(θ(u, ẑ), u)J(h(u, ẑ)) and θ(u, ẑ) ≥ 0, (6)

where the two inequalities hold with complementary slackness.

The equilibrium law of motion for unemployment, h(u, ẑ), must be consistent with the

equilibrium firing probability s ∗ (y, ẑ, u) and with the job-finding probability λ(θ(u, ẑ), u).

Specifically, h(u, ẑ) must be such that

h(u, ẑ) = u − uµ(J(h(u, ẑ), u) + (1 − u)E y [δ + (1 − δ)s ∗ (y, ẑ, u)] , (7)

where

µ(J, u) = λ η −1 (min{k/J, 1}, u) , u ,

and η −1 (min{k/J, 1}, u) is the labor market tightness that solves (6). The first term on

the right-hand side of (7) is unemployment at the beginning of the bargaining stage in the

current period. The second term in (7) is the measure of unemployed workers who become

employed during the matching stage of next period, which is given by unemployment u

times the probability that an unemployed worker becomes employed µ(J(h(u, ẑ), u). The

last term in (7) is the measure of employed workers who become unemployed during the

separation stage of next period. The sum of the three terms on the right-hand side of (7) is

the unemployment at the beginning of the bargaining stage in the next period.

We are now in the position to define a recursive equilibrium for our model economy.

Definition 1: A Recursive Equilibrium is a tuple (W, F, V, J, x ∗ , h) such that: (i) The gains

from trade accruing to the worker, W (x, u), and to the firm, F (x, u), satisfy (3) and (4) and

11

V (u) = W (x ∗ (u), u), J(u) = F (x ∗ (u), u); (ii) The employment contract x ∗ (u) satisfies (5);

(iii) The law of motion h(u, ẑ) satisfies (7).

Over the next three sections, we will characterize the properties of the recursive equilibrium.

We are going to carry out the analysis under the maintained assumptions that the

equilibrium gains from trade are strictly positive, i.e. J(u) > 0 and V (u) > 0, and that the

equilibrium contract requires the worker to exert effort, i.e. e ∗ (u) = 1. The first assumption

guarantees that firms and workers trade in the labor market, and the second assumption

guarantees that firms and workers find it optimal to solve the moral hazard problem. 6

3 Optimal Contract

In this section, we characterize the properties of the Axiomatic Nash Solution to the bargaining

problem between the firm and the worker. That is, we characterize the properties

of the employment contract that maximizes the product of the gains from trade accruing to

the worker and the gains from trade accruing to the firm subject to the worker’s incentive

compatibility constraint. 7

We refer to such contract as the optimal employment contract.

Our key finding is that the worker is fired if and only if the realization of output is low and

the realization of the state of the world is such that the cost to the worker from losing the

job relative to the cost to the firm from losing the worker is suffi ciently high.

We carry out the characterization of the optimal employment contract in four lemmas,

all proved in Appendix A. In order to lighten up the notation, and without risk of confusion,

we will drop the dependence of the gains from trade to the worker, W , and to the firm, F , as

well as the dependence of the optimal contract, x ∗ , on unemployment in the current period,

u. We will also drop the dependence of the continuation gains from trade to the worker

and to the firm on unemployment in the current period and write V (h(u, ẑ)) as V (ẑ) and

J(h(u, ẑ)) as J(ẑ).

Lemma 1: Any optimal contract x ∗ is such that the worker’s incentive compatibility holds

with equality. That is,

ψ = β(p h (1) − p h (0))Eẑ [(1 − δ)(s ∗ (y l , ẑ) − s ∗ (y h , ẑ))V (ẑ)] . (8)

6 It is straightforward to verify that the first assumption is satisfied as long as b is suffi ciently low relative

to p h (1)y h + p l (1)y l , and that the second assumption is satisfied as long as y l is suffi ciently low relative to

y h .

7 As mentioned at the end of Section 2, we will carry out the analysis under the maintained assumptions

that the gains from trade are strictly positive and that it is optimal to recommend the worker to exert effort.

12

To understand Lemma 1 consider a contract x such that the worker’s incentive compatibility

constraint is lax. Clearly, this contract prescribes that the worker is fired with some

positive probability after a low realization of output, i.e. s(y l , ẑ). If we lower s(y l , ẑ) by some

small amount, the worker’s incentive compatibility constraint is satisfied. Moreover, if the

lower s(y l , ẑ), the survival probability of the match increases. Since the continuation value

of the match is strictly positive for both the worker and the firm, an increase in the survival

probability of the match raises the gains from trade accruing to the worker, W , the gains

from trade accruing to the firm, F , and the Nash product W F . Therefore, the contract x

cannot be optimal.

Lemma 2: Any optimal contract x ∗ is such that, if the realization of output is high, the

worker is fired with probability 0. That is, for all ẑ ∈ [0, 1],

s ∗ (y h , ẑ) = 0. (9)

To understand Lemma 2 consider a contract x such that the worker is fired with positive

probability when the realization of output is high, i.e. s(y h , ẑ) > 0. If we lower the firing

probability s(y h , ẑ), the incentive compatibility constraint of the worker is relaxed. Moreover,

if we lower the firing probability s(y h , ẑ), the survival probability of the match increases. In

turn, the increase in the survival probability of the match raises the gains from trade accruing

to the worker, W , the gains from trade accruing to the firm, F , and the Nash product W F .

Thus, the contract x cannot be optimal.

Lemma 3: Let φ(ẑ) ≡ V (ẑ)/J(ẑ). Any optimal contract x ∗ is such that, if the realization

of output is low, the worker is fired with probability 1 if φ(ẑ) > φ ∗ , and the worker is fired

with probability 0 if φ(ẑ) < φ ∗ . That is, for all ẑ ∈ [0, 1],

s ∗ (y l , ẑ) =

1, if φ(ẑ) > φ ∗ ,

0, if φ(ẑ) < φ ∗ .

(10)

Lemma 3 is one of the main results of the paper. It states that any optimal contract x ∗ is

such that, if the realization of output is low, the worker is fired with probability 1 in states

of the world ẑ in which the continuation gains from trade to the worker, V (ẑ), relative to

the continuation gains from trade to the firm, J(ẑ), are above some cutoff, and the worker

is fired with probability 0 in states of the world in which the ratio V (ẑ)/J(ẑ) is below the

cutoff. There is a simple intuition behind this result. Firing is costly– as it destroys a

valuable relationship– but also necessary– as it is the only tool to provide the worker with

an incentive to exert effort. However, only the value of the destroyed relationship that

would have accrued to the worker serves the purpose of providing incentives. The value of

13

the destroyed relationship that would have accrued to the firm is “collateral damage.”The

optimal contract minimizes the collateral damage by concentrating firing in states of the

world in which the value of the relationship to the worker would have been highest relative

to the value of the relationship to the firm. In other words, the optimal contract minimizes

the collateral damage by concentrating firing in states of the world in which the cost to the

worker from losing the job, V (ẑ), is highest relative to the cost to the firm from losing the

worker, J(ẑ). Notice that this property of the optimal contract follows immediately from the

linearity of the problem with respect to the firing probability, and it does not depend on the

fact that the optimal contract maximizes the product of the gains from trade rather than

the gains from trade to the firm taking subject to delivering a given level of gains from trade

to the worker. In this sense, the property is rather general to contractual environments in

which firing takes place along the equilibrium path.

In any optimal contract x ∗ , the firing cutoff φ ∗ is such that

Z

Z

ψ = β(p h (1) − p h (0))(1 − δ) V (ẑ)dẑ + s ∗ (y l , ẑ)V (ẑ)dẑ , (11)

φ(ẑ)>φ ∗ φ(ẑ)=φ ∗

The above equation is the worker’s incentive compatibility constraint (8) written in light

of the fact that s ∗ (y h , ẑ) is given by (9) and s ∗ (y l , ẑ) is given by (10). Figure 1 plots the

right-hand side of (11), which is the worker’s benefit from exerting effort, as a function of the

firing cutoff φ ∗ . On any interval [φ 0 , φ 1 ] where the distribution of the random variable φ(ẑ)

has positive density, the right-hand side of (11) is strictly decreasing in φ ∗ . On any interval

[φ 0 , φ 1 ] where the distribution of φ(ẑ) has no density, the right-hand side of (11) is constant.

At any value φ where the distribution of φ(ẑ) has a mass point, the right-hand side of (11)

can take on an interval of values, as the firing probability s ∗ (y l , ẑ) for ẑ such that φ(ẑ) = φ

varies between 0 and 1. Overall, the right-hand side of (11) is a weakly decreasing function

of the firing cutoff φ ∗ .

In any optimal contract x ∗ , the firing cutoff φ ∗ is such that the right-hand side of (11) is

equal to the worker’s cost ψ from exerting effort. There are three cases to consider. First,

consider the case in which the right-hand side of (11) equals ψ at a point where the righthand

side of (11) is strictly decreasing in φ ∗ . In this case, the equilibrium firing cutoff is

uniquely pinned down. Moreover, since the right-hand side of (11) is strictly decreasing in

φ ∗ , the random variable φ(ẑ) has no mass point at the equilibrium firing cutoff. Hence, in

this case, the firm either fires the worker with probability 0 or with probability 1. This is the

case of ψ 1 in Figure 1. Second, consider the case in which the right-hand side of (11) equals

ψ at a point where the right-hand side of (11) can take on a range of values. In this case, the

14

equilibrium firing cutoff is uniquely pinned down. However, the random variable φ(ẑ) has a

mass point at the equilibrium firing cutoff. Hence, in this case, for any realization of φ(ẑ)

equal to the equilibrium firing cutoff, the firm fires the worker with the probability s ∗ (y l , ẑ)

that satisfies (11). This is the case of ψ 2 in Figure 1. Finally, consider the case in which there

is an interval of values of φ(ẑ) such that the right-hand side of (11) equals ψ. In this case,

the equilibrium firing cutoff can take on any value in the interval. However, the choice of

the cutoff is immaterial, as the probability that the random variable φ(ẑ) falls in the interval

is zero. This is the case of ψ 3 in Figure 1. In any of the three cases, the equilibrium firing

cutoff is effectively unique and so is the firing probability for any realization of the random

variable φ(ẑ).

Lemma 4: Any optimal contract x ∗ is such that the wage w ∗ satisfies

W (x ∗ )

F (x ∗ ) = υ′ (w ∗ )

. (12)

1

Lemma 4 states that any optimal contract x ∗ prescribes a wage such that the ratio of

the marginal utility of consumption to the worker to the marginal utility of consumption to

the firm, υ ′ (w ∗ )/1, is equal to the ratio of the equilibrium gains from trade accruing to the

worker to the equilibrium gains from trade accruing to the firm, W (x ∗ )/F (x ∗ ) = V/J. This

15

is the standard optimality condition for the wage in the Axiomatic Nash Solution. Lemma

4 is important, as it tells us that the relative gains from trade accruing to the worker are

higher in states of the world in which the worker’s wage is lower. Hence, it follows from

Lemma 3 that the optimal contract is such that the firm fires the worker if and only if the

realization of output is low and the realization of the state of the world is such that the

worker’s wage next period would have been suffi ciently low.

We are now in the position to summarize the characterization of the optimal contract.

Theorem 1: (Contracts) Any optimal contract x ∗ is such that: (i) the worker is paid the

wage w ∗ given by (12); (ii) If the realization of output is high, the worker is fired with

probability s ∗ (y h , ẑ) given by (9); (iii) If the realization of output is low, the worker is fired

with probability s ∗ (y l , ẑ) given by (10), where the φ ∗ is uniquely pinned down by (11).

4 Properties of Equilibrium

In this section, we characterize the properties of the equilibrium. First, we characterize

the role of the sunspot within a period. We show that there exists a Perfect Coordination

Equilibrium in which all firms fire their non-performing workers with probability 1

for some realization of the sunspot and with probability 0 for the other realizations of the

sunspot. In this equilibrium, firms use the sunspot to randomize over keeping or firing their

non-performing workers in a perfectly correlated fashion. There is also a No Coordination

Equilibrium in which firms fire workers with the same probability independently of the realization

of the sunspot. In this equilibrium, firms randomize over firing or keeping their

non-performing workers independently from each other. However, we show that the No Coordination

Equilibrium only exists because of the firms randomize simultaneously and have

to rely on an inherently meaningless signal to coordinate. Indeed, we show that, in a version

of the model where firms randomize sequentially, the unique equilibrium is the one with

perfect coordination. Finally, we establish the existence and characterize the properties of

the recursive equilibrium of the economy.

4.1 Stage Equilibrium

In any Recursive Equilibrium, the probability s ∗ (y l , ẑ) with which firms fire non-performing

workers and the worker’s relative gains from trade φ(ẑ) must simultaneously satisfy two conditions.

For any ẑ, the firing probability s ∗ (y l , ẑ) must be part of the optimal employment

contract given the worker’s relative gains from trade φ(ẑ) and the probability distribution

16

of the worker’s relative gains from trade across realizations of the sunspot. Moreover, for

any ẑ, the worker’s relative gains from trade φ(ẑ) must be those implied by the evolution

of unemployment, given the firing probability s ∗ (y l , ẑ). Formally, in any equilibrium, the

functions φ(ẑ) and s ∗ (y l , ẑ) must be a fixed-point of the mapping we just described. Borrowing

language from game theory, we refer to such a fixed-point as the stage equilibrium,

as it describes the key outcomes of the economy within one period.

We first characterize the effect of the firm’s firing probability s ∗ (y l , ẑ) on the worker’s

relative gains from trade φ(ẑ). Given that unemployment at the beginning of the period is u

and that the firing probability at the separation stage is s(ẑ) = s ∗ (y l , ẑ), the law of motion

(7) implies that unemployment at the bargaining stage is û(s(ẑ)) such that

û(s(ẑ)) = u − µ(J(û(s(ẑ))), u) + (1 − u)(δ + (1 − δ)p l (1)s(ẑ)). (13)

We conjecture that the gains from trade accruing to the firm are a strictly increasing function

of unemployment, i.e. J(û(s(ẑ))) is strictly increasing in û(s(ẑ)). Under this conjecture,

there is a unique û(s(ẑ)) that satisfies (13) and û(s(ẑ)) is strictly increasing in the firing

probability s(ẑ).

Given that unemployment at the bargaining stage is û(s(ẑ)), the worker’s wage is w ∗ (û(s(ẑ))).

Then, it follows from the optimality condition (12) that the worker’s relative gains from trade

are such that

φ(ẑ) = V (û(s(ẑ)))

J(û(s(ẑ))) = υ′ (w ∗ (û(s(ẑ))))

. (14)

1

We conjecture that the wage is a strictly decreasing function of unemployment, i.e. w ∗ (û(s(ẑ)))

is strictly decreasing in û(s(ẑ)). Under this conjecture, the worker’s relative gains from trade

φ(ẑ) are strictly increasing in the unemployment û(s(ẑ)) and, since û(s(ẑ)) is strictly increasing

in s(ẑ), they are also strictly increasing in the firing probability s(ẑ). The solid red line

in Figure 2 illustrates the effect of the firing probability on the worker’s relative gains from

trade.

The conjectures that the worker’s wage is decreasing in unemployment and the firm’s

gains from trade are increasing in unemployment are natural and will be verified in Section

4.3. Intuitively, when the matching function has decreasing returns to scale, an increase in

unemployment tends to lower the job-finding probability of unemployed workers. In turn, a

decline in the job-finding probability lowers the value of unemployment. Since the value of

unemployment is the worker’s outside option in bargaining, the equilibrium wage falls and

the worker’s relative gains from trade increase.

17

Next, we characterize the effect of the worker’s relative gains from trade φ(ẑ) on the

probability s(ẑ) = s ∗ (y l , ẑ) with which firms fire their non-performing workers. From the

optimality condition (10), it follows that the firing probability s(ẑ) is such that

8

< 0, if φ(ẑ) < φ ∗ ,

s ∗ (y l , ẑ) = ∈ [0, 1], if φ(ẑ) = φ ∗ ,

:

1, if φ(ẑ) > φ ∗ ,

where φ ∗ is implicitly defined by the worker’s incentive compatibility constraint (11). As

explained in the previous section, the higher are the worker’s relative gains from trade in

ẑ, the stronger is the firm’s incentive to fire its non-performing workers in that state of the

world. The dashed green line in Figure 2 illustrates the effect of the worker’s relative gains

from trade φ(ẑ) on the firing probability s(ẑ).

For any realization ẑ of the sunspot, the firing probability must be optimal given the

worker’s gains from trade (i.e.

(15)

we must be on the dashed green line) and the worker’s

gains from trade must be consistent with the firing probability (i.e.

we must be on the

solid red line). As it is clear from Figure 2, for any realization of ẑ, only three outcomes

are possible: points A, B and C. The first outcome, point A, is such that the firm’s firing

probability s(ẑ) is zero, and the worker’s relative gains from trade φ(ẑ) are smaller than

18

φ ∗ . The second outcome, point B, is such that the firm’s firing probability s(ẑ) is greater

than zero and smaller than one, and the worker’s relative gains from trade φ(ẑ) are equal to

φ ∗ . The third outcome, point C, is such that the firm’s firing probability s(ẑ) is one, and

the worker’s relative gains from trade φ(ẑ) are greater than φ ∗ . The coexistence of multiple

outcomes is a consequence of the fact that firms have a desire to coordinate the outcome

of the randomization over firing or keeping their workers. If other firms are more likely to

fire their workers in one state of the world than in another, an individual firm wants to do

the same, because, in the state of the world where other firms are more likely to fire their

workers, unemployment is higher and so are the worker’s relative gains from trade.

Let Z 0 denote the realizations of the sunspot for which firms fire non-performing workers

with probability 0, and let π 0 denote the measure of Z 0 . Let Z 1 denote the realizations of the

sunspot for which firms fire non-performing workers with a probability s ∗ (y l , ẑ) = s 1 ∈ (0, 1),

and let π 1 denote the measure of Z 1 . Similarly, let Z 2 denote the realizations of the sunspot

for which firms fire non-performing workers with probability 1, and let π 2 denote the measure

of Z 2 .

Depending on the value of π 1 , we can identify three different types of stage equilibria.

If π 1 = 1, we have a No Coordination Equilibrium. In this equilibrium, firms fire the nonperforming

workers with probability s ∗ (y l , ẑ) = s 1 ∈ (0, 1) for all ẑ ∈ [0, 1]. Basically,

firms ignore the sunspot and randomize over firing or keeping their non-performing workers

independently from each other. In a No Coordination Equilibrium, the worker’s incentive

compatibility constraint (11) becomes

ψ = β(p h (1) − p h (0))(1 − δ)s 1 V (û(s 1 )). (16)

When we solve the constraint with respect to s 1 , we find that the constant probability with

which firms fire their non-performing workers is

s 1 =

ψ

β(p h (1) − p h (0))(1 − δ)V (û(s 1 )) . (17)

If π 1 = 0, we have a Perfect Coordination Equilibrium. In this equilibrium, firms fire their

non-performing workers with probability 0 for the realizations of the sunspot ẑ ∈ Z 0 , and

with probability 1 for the other realization of the sunspot. Basically, firms use the sunspot to

randomize over firing or keeping their non-performing workers in a perfectly correlated fashion.

In a Perfect Coordination Equilibrium, the worker’s incentive compatibility constraint

(11) becomes

ψ = β(p h (1) − p h (0))(1 − δ)π 2 V (û(1)). (18)

19

When we solve the constraint with respect to π 2 , we find that the probability with which

firms coordinate on firing all of their non-performing workers is given by

π 2 =

ψ

β(p h (1) − p h (0))(1 − δ)V (û(1)) . (19)

If π 1 ∈ (0, 1), we have a Partial Coordination Equilibrium. In this equilibrium, firms

fire workers with probability s ∗ (y l , ẑ) = s 1 ∈ (0, 1) for all ẑ ∈ Z 1 . Hence, when ẑ ∈ Z 1 ,

firms randomize over firing or keeping their non-performing workers independently from

each other. However, if ẑ /∈ Z 1 , firms fire their workers with probability 0 if ẑ ∈ Z 0 and

with probability 1 if ẑ ∈ Z 2 . Hence, when ẑ /∈ Z 1 , firms use the sunspot to randomize over

firing or keeping their non-performing workers in a correlated fashion. Overall, a Partial

Correlation Equilibrium is a combination of a No Coordination and a Perfect Coordination

Equilibrium. In a Partial Coordination Equilibrium, the worker’s incentive compatibility

constraint (11) becomes

ψ = β(p h (1) − p h (0))(1 − δ) [π 1 s 1 V (û(s 1 )) + π 2 V (û(1))] . (20)

When we solve the constraint with respect to s, we find that the constant probability with

which firms fire their non-performing workers for ẑ ∈ Z 1 is

s 1 = ψ − β(p h(1) − p h (0))(1 − δ)π 2 V (û(1))

. (21)

β(p h (1) − p h (0))(1 − δ)π 1 V (û(s 1 ))

The above results are summarized in Theorem 2.

Theorem 2: (Stage Equilibrium). Three 1-Period Equilibria exist: (i) No Coordination

Equilibrium where s ∗ (y l , ẑ) = s 1 for all ẑ ∈ Z 1 , where Z 1 has probability measure π 1 = 1

and s 1 is given by (17); (ii) Perfect Coordination Equilibrium where s ∗ (y l , ẑ) = 0 for all

ẑ ∈ Z 0 and s ∗ (y l , ẑ) = 1 for all ẑ ∈ Z 2 , where Z 0 has probability measure 1 − π 2 and

Z 2 has probability measure π 2 and π 2 ∈ (0, 1) is given by (19); (iii) Partial Coordination

Equilibrium where s ∗ (y l , ẑ) = 0 for all ẑ ∈ Z 0 , s ∗ (y l , ẑ) = s 1 for all ẑ ∈ Z 1 , and s ∗ (y l , ẑ) = 1

for all ẑ ∈ Z 2 , where Z 1 has probability measure π 0 ∈ (0, 1) and s 1 is given by (21).

The Perfect Coordination Equilibrium exists because firms have an incentive to coordinate

the outcome of the randomization over firing and keeping their non-performing workers

in order to minimize the “collateral damage”involved in providing workers with incentives.

Moreover, firms are able to coordinate the outcome of the randomization because of the

sunspot. The No Coordination Equilibrium and the Partial Coordination Equilibrium exist

because the sunspot is inherently meaningless and, hence, there always exist an equilibrium

20

in which it is ignored. However, if firms do not need to rely on an inherently meaningless

signal to coordinate, they will always be able to do so and the No Coordination and Partial

Coordination Equilibria will disappear. Indeed, in the next subsection, we consider a version

of the stage game in which firms fire sequentially and show that its unique equilibrium is the

one with perfect coordination.

In a Perfect Coordination Equilibrium the economy experiences aggregate fluctuations:

for some realizations of the sunspot, all of the firms fire their non-performing workers and

unemployment is high; for other realizations of the sunspot, all of the firms keep their nonperforming

workers and unemployment is low. First, notice that these aggregate fluctuations

are endogenous. Indeed, they do not originate from exogenous shocks to fundamentals, or

from exogenous shocks to the selection of the equilibrium played by the market participants.

They are caused by the fact that, in the unique robust equilibrium, different firms randomize

over firing or not firing in a correlated fashion. Second, notice that these aggregate

fluctuations are stochastic. Indeed, the economy does not follow a more or less complicated

deterministic path (as in the earlier literature on endogenous cycles), but a genuinely stochastic

process. We believe that our model is the first example of a theory of endogenous

and stochastic fluctuations. In fact, we do not know of any other model where the stochastic

process for aggregate fluctuations is an equilibrium outcome as it is in our model (see

equation (19)).

4.2 Equilibrium Refinement

As we discussed above, the existence of No Coordination and Partial Coordination Equilibria

is an artifact of the simplifying assumption that all firms randomize over firing or keeping

non-performing workers simultaneously. When firms randomize simultaneously, they need

to rely on the sunspot to correlate the outcome of their randomization. However, since the

sunspot is inherently meaningless, there is always an equilibrium in which the sunspot is

ignored and firms cannot correlate the outcome of their randomization. In this subsection,

we consider a version of the environment in which firms randomize over firing or keeping

their non-performing workers sequentially. We show that, firms moving later can always

condition their randomization on the outcome of firms moving earlier and, hence, the unique

equilibrium is the one with perfect coordination.

Here is a formal description of the modified environment. Let 1 − u denote the measure

of employed workers at the bargaining stage of the current period. The measure of employed

workers is equally divided into a large number NK firms, each employing one worker of

21

“measure”(1−u)/NK. Firms are clustered into a large number K of groups, each comprising

a large number N of firms. Firms and workers bargain over the terms of the one-period

employment contract knowing the group to which they belong. At the separation stage of

next period, firm-worker pairs in different groups break up sequentially. First, the firmworker

pairs in group 1 simultaneously decide to separate or not. Second, after observing

the outcomes of group 1, the firm-worker pairs in group 2 simultaneously decide to separate

or not. Third, after observing the outcomes of groups 1 and 2, the firm-worker pairs in group

3 simultaneously decide to separate or not. The process continues until the firm-worker pairs

in group K simultaneously decide to separate or not, after having observed the outcomes of

groups 1 through K − 1. Naturally, in this version of the model, we do not need the sunspot.

Let T i denote the measure of workers separating from firms in groups 1 through i. We

assume that each firm in group i takes as given the probability distribution of T i conditional

on T i−1 , which we denote as P i (T i |T i−1 ). The assumption means that each firm views itself as

small compared to its group. The assumption is reasonable when N is large. We also assume

that P i (T i |T i−1 ) is increasing, in the sense of first-order stochastic dominance, in T i−1 . The

assumption means that firms in a group view themselves as small compared to the whole

economy. The assumption is reasonable when K is large. We also approximate the worker’s

gains from trade, V (û), and the firm’s gains from trade, J(û), with linear functions. The

approximation implies that the expectation of the worker’s relative gains from trade over next

period’s unemployment, E[V (û)]/E[J(û)], is equal to the worker’s relative gains from trade

evaluated at the expectation of next period’s unemployment, V (E[û])]/J(E[û]) = φ(E[û]).

We can now characterize the optimal contract between a worker and a firm in group

i = 2, 3, . . . K. The contract can condition the firing probability s i (y, T i−1 ) on the realization

of the worker’s output y and on the measure T i−1 of workers separating from firms in groups

1 through i − 1. As in Section 3, it is easy to show that the optimal contract is such that: (i)

the worker’s incentive compatibility constraint holds with equality; (ii) if the realization of

output is high, the worker is fired with probability 0, i.e. s i (y h , T i−1 ) = 0 for all T i−1 ; (iii) if

the realization of output is low, the worker is fired with probability 0 if the worker’s relative

gains from trade are below a cutoff φ ∗ i , and with probability 1 if they are above the cutoff, i.e.

s i (y l , T i−1 ) = 0 if φ(E[û|T i−1 ]) < φ ∗ i , and s i (y l , T i−1 ) = 1 if φ(E[û|T i−1 ]) > φ ∗ i . The optimal

contract between a worker and a firm in group 1 can only condition the firing probability

on the realization of the worker’s output y. In this case, the optimal contract is such that

s 1 (y h ) = 0 and s 1 (y l ) = s 1 , where s 1 is such that the worker’s incentive compatibility

constraint holds with equality.

22

The following lemma shows that the probability s i (y l , T i−1 ) with which firms in group

s i (y l , T i−1 ) fire their non-performing workers has a threshold property with respect to T i−1 ,

the measure of workers separating from firms in groups 1 through T i−1 . This property of

equilibrium is intuitive, as a higher T i−1 leads to a higher expectation for unemployment

and, in turn, to a higher expectation for the worker’s relative gains from trade.

Lemma 5: For i = 2, . . . K, the firing probability s i (y l , T i−1 ) equals 0 for all T i−1 < T ∗

i−1,

and it equals 1 for all T i−1 > T ∗

i−1.

Proof : In Appendix B.

We can now compute the equilibrium probability distribution of the measure t i of workers

separating from firms in group i = 1, 2, . . . K, conditional on the measure T i−1 of workers

separating from firms in groups 1 through i − 1. Any firm-worker pair in group 1 separates

with probability τ 1 = δ + (1 − δ)p l (1)s 1 . Since N is large, we can use the Central Limit

Theorem to approximate the measure t 1 of workers separating from firms in group 1 with a

Normal distribution with mean E[t 1 ] = τ 1 (1 − u)/K and variance V ar[t 1 ] = τ 1 (1 − τ 1 )[(1 −

u)/K] 2 /N. Conditional on T i−1 , a firm-worker pair in group i = 2, 3, . . . .K separates with

probability τ i (T i−1 ) = δ + (1 − δ)p l (1)s i (y l , T i−1 ). Since N is large, we can approximate

the measure t i of workers separating from firms in group i with a Normal distribution with

mean E[t i |T i−1 ] = τ i (T i−1 )(1 − u)/K and variance V ar[t i |T i−1 ] = τ i (T i−1 )(1 − τ 1 (T i−1 ))[(1 −

u)/K] 2 /N. We find it convenient to define t l = δ(1 − u)/K, and t h = [δ + (1 − δ)p l (1)] (1 −

u)/K.

The following lemma shows that, for N → ∞, there exists an equilibrium in which the

firms in groups 2 through K either all fire or all keep their non-performing workers depending

on the measure of workers separating from firms in group 1. Let us give some intuition for

this result in the case of K = 3. If the firms in group 1 happen to break up with more

than T1

∗ workers, firms in group 2 fire their non-performing workers with probability 1. If

the firms in group 1 happen to break up with less than T1

∗ workers, firms fire in group

2 fire their non-performing workers with probability 0. Since the variance of t 1 and t 2 is

vanishing as N → ∞, T 2 = E[t 1 ] + t h with probability 1 − P 1 (T ∗ 1 ) and T 2 = E[t 1 ] + t l

with probability P 1 (T ∗ 1 ). Suppose that T ∗ 2 = E[t 1 ] + (t h + t l )/2. Then, firms in group 3 fire

their non-performing workers with probability 1 if T 2 = E[t 1 ] + t h and with probability 0

if T 2 = E[t 1 ] + t l . Since the variance of t 3 is vanishing as N → ∞, T 3 = E[t 1 ] + 2t h with

probability 1 − P 1 (T ∗ 1 ) and T 3 = E[t 1 ] + 2t l with probability P 1 (T ∗ 1 ). Overall, firms in group

2 fire their non-performing workers with probability 1 − P 1 (T ∗ 1 ) and, in that case, expect

total separations T 3 = E[t 1 ] + 2t h . Firms in group 3 fire their non-performing workers with

23

probability 1 − P 1 (T1 ∗ ) and, in that case, they expect total separations T 3 = E[t 1 ] + 2t h .

Therefore, if T1

∗ satisfies the incentive compatibility constraint of workers in group 2, then

satisfies the incentive compatibility constraint of workers in group 3. This confirms the

T ∗ 2

existence of the desired equilibrium for K = 3. Clearly, for K → ∞, this equilibrium

converges to the Perfect Coordination Equilibrium.

Lemma 6: For N → ∞, there is an equilibrium in which firms in groups 2 through K

fire their non-performing workers with probability 0 if T 1 < T1 ∗ , and with probability 1 if

T 1 > T1 ∗ , where T1

∗ is such that

ψ = β(1 − δ)(p h (1) − p h (0))(1 − P 1 (T1 ∗ ))V (E[û|T K = E[t 1 ] + (K − 1)t h ]). (22)

Proof : In Appendix B.

Next, we rule out the existence of other equilibria. To this aim, notice that, in any

equilibrium, firms in group 2 fire their non-performing workers with probability 0 if T 1 < T1 ∗ ,

and they fire them with probability 1 if T 1 > T1 ∗ , where T1 ∗ is such that the worker’s incentive

compatibility constraint is satisfied, i.e.

ψ = β(1 − δ)(p h (1) − p h (0))(1 − P 1 (T1 ∗ ))V (E[û|T 1 > T1 ∗ ]) (23)

For N → ∞, T 2 is approximately equal to E[t 1 ] + t l with probability P 1 (T1 ∗ ) and, it is

approximately equal to E[t 1 ] + t h with probability 1 − P 1 (T1 ∗ ).

Now, suppose that the threshold T2 ∗ is such that P 2 (T2 ∗ ) > P 1 (T1 ∗ ). Then, conditional on

any T 2 approximately equal to E[t 1 ] + t l , firms in group 3 fire their non-performing workers

with probability 0. Conditional on T 2 being approximately equal to E[t 1 ]+t h , firms in group 3

do not fire their non-performing workers with probability (P 2 (T2 ∗ )−P 1 (T1 ∗ ))/(1−P 1 (T1 ∗ )) and

they do with probability (1 − P 2 (T2 ∗ ))/(1 − P 1 (T1 ∗ )). The incentive compatibility constraint

for workers employed by firms in group 3 is thus given by

ψ = β(1 − δ)(p h (1) − p h (0))(1 − P 2 (T2 ∗ ))V (E[û|T 2 > T2 ∗ ]) (24)

However, the incentive compatibility constraints (23) and (24) cannot hold simultaneously

and, hence, there cannot be an equilibrium in which P 2 (T2 ∗ ) > P 1 (T1 ∗ ). To see why this is

the case, notice that E[û|T 1 > T1 ∗ ] is equal to E[û|T 2 > T2 ∗ ] + E[û|T 2 < T2 ∗ , T 1 > T1 ∗ ] and

1 − P 1 (T1 ∗ ) > 1 − P 2 (T2 ∗ ). Therefore, the right-hand side of (23) is strictly greater than the

right-hand side of (24). Following a similar argument, we can rule also out equilibria in which

P 2 (T2 ∗ ) < P 1 (T1 ∗ ). Hence, in any equilibrium, P 2 (T2 ∗ ) = P 1 (T1 ∗ ), which implies that firms in

24

group 3 fire their non-performing workers if and only if firms in group 2 do. Repeating the

above argument for i = 4, 5, . . . .K, we can show that, in any equilibrium, firms in group i

fire their non-performing workers if and only if firms in group i − 1 do. Therefore, the only

equilibrium of the modified environment is the one described in Lemma 6.

We have thus completed the proof of the following theorem.

Theorem 3: (Refinement) For K → ∞ and N → ∞, the unique equilibrium of the environment

with K groups of N firms firing sequentially is the Perfect Coordination Equilibrium.

The analysis of the environment where firms fire sequentially sheds additional light on

the nature of aggregate fluctuations in our model. The firms in the first group find it optimal

to randomize on whether to fire or keep their non-performing workers. If the first group of

firms fire enough workers, then all the other firms in the economy find it optimal to fire their

non-performing workers. Otherwise, all the other firms in the economy find it optimal to

keep their non-performing workers. That is, the equilibrium is such that the firing decision

of the first group of firms leads to a “firing cascade”. In contrast to the models of herding of

Banerjee (1992) and Bikhchandani et al. (1992), the cascades in our model do not take place

because the actions of the first group of firms contain information about the realization of

an exogenous aggregate shock, but because the actions of the first group of firms affect the

incentive of subsequent firms from taking the same action. Hence, cascades in our model

start from the realization of idiosyncratic shocks, absent any aggregate uncertainty. In this

sense, aggregate fluctuations in our model have a granular origin, as in Jovanovic (1987) and

Gabaix (2011). However, unlike Jovanovic (1987) and Gabaix (2011), idiosyncratic shocks

in our model propagate because of strategic interactions between firms rather than because

of the input-output structure of the economy.

4.3 Recursive Equilibrium

In the previous subsections, we established the existence of a Perfect Coordination Equilibrium

of the stage game under the conjecture that unemployment is increasing and the wage

is decreasing in unemployment. We also argued that the Perfect Coordination Equilibrium

is the only equilibrium of the stage game that is robust to a natural perturbation of the environment.

In this subsection, we show that, given a Perfect Coordination Equilibrium of the

stage game, there exists a Recursive Equilibrium such that the unemployment is increasing

in firing and the wage is decreasing in unemployment.

25

The existence proof is an application of Schauder’s fixed point theorem. The proof is

lengthy and relegated in Appendix C. Here we outline the structure of the proof. Denote

as Ω the set of bounded functions (V + , J + ), V + : [0, 1] −→ R and J + : [0, 1] → R, such

that for all u 0 and u 1 , 0 ≤ u 0 ≤ u 1 ≤ 1, V + (u 1 ) − V + (u 0 ) is greater than D V+

(u 1 − u 0 )

and smaller than D V+ (u 1 − u 0 ), and J + (u 1 ) − J + (u 0 ) is greater than D J+

(u 1 − u 0 ) and

smaller than D J+ (u 1 − u 0 ), where D V+ > D V+

> 0, D J+ > 0 ≥ D J+

. That is, Ω is the

set of functions (V + , J + ) that are bounded and Lipschitz continuous, with Lipschitz bounds

respectively given by D V+

and D V+ , and D J+

and D J+ . 8 Also, we denote as µ u and µ u the

upper and the lower bound of the partial derivative of the job-finding probability µ(J, u)

with respect to u, and as µ J and µ J the upper and the lower bound of the partial derivative

of µ(J, u) with respect to J.

Take an arbitrary pair of functions V + (u) and J + (u) from the set Ω. We let V + be the

worker’s expected gains from trade at the end of the production stage, and we let J + be the

firm’s expected gains from trade at the end of the production stage. First, given (V + , J + ),

we use the fact that W (w, u) = υ(w) − υ(b) − ψ + V + (u) and F (w, u) = E[y] − w + J + (u)

and condition (12) for the optimal contract to compute the equilibrium wage function w(u).

We prove that w(u) is strictly decreasing in u. Intuitively, given the choice for the bounds

D V+

and D J+ , an increase in unemployment leads to a larger increase in W (w, u) than in

υ ′ (w)F (w, u) for any given wage w. For this reason, the wage must fall to make sure that

the worker’s relative gains from trade, W (w, u)/F (w, u), are equal to the worker’s relative

marginal utility of consumption υ ′ (w).

Second, given the functions V + (u) and J + (u) and the wage function w(u), we use the

fact that V (u) = W (w(u), u) and J(u) = F (w(u), u) to compute the equilibrium gains from

trade accruing to the worker and to the firm. We prove that J(u) is strictly increasing in

u. Intuitively, given the choice for the bound D J+

, an increase in unemployment leads to a

decline in the wage that more than compensates the largest possible decline in J + (u). Similarly,

we prove that V (u) is strictly increasing in u. Intuitively, V (u) is equal to J(u)υ ′ (w(u))

and both J(u) and υ ′ (w(u)) are strictly increasing in u.

Third, given the function J(u), we use (7) to compute the law of motion for unemployment

h(u, ẑ). We prove that, as long as (1 − δ)p h (1) − µ(J, u) > 0, next period’s unemployment

h(u, ẑ) is strictly increasing in current period’s unemployment. Moreover, we prove that

next period’s unemployment h(u, ẑ) is strictly greater for the realization of the sunspot for

which firms coordinate on firing their non-performing workers, i.e. for ẑ ∈ Z 2 , than for

8 The reader can find the expression for the Lipschitz in Appendix B.

26

the realizations for which firms coordinate on keeping their non-performing workers, i.e. for

ẑ ∈ Z 0 .

Finally, given the functions V (u), J(u) and h(u, ẑ), we compute updates FV + (u) and

FJ + (u) for the worker’s and firm’s expected gains from trade at the end of the production

process. More precisely, we compute FV + (u) and FJ + (u) as

FV + (u) = βEẑ {[(1 − δ)(1 − p l (1)s(y l , ẑ)) − µ(J(h(u, ẑ)), u)]V (h(u, ẑ))},

FJ + (u) = βEẑ [(1 − δ)(1 − p l (1)s(y l , ẑ))J(h(u, ẑ))],

(25)

where the firing probability s(y l , ẑ) = 1 for ẑ ∈ Z 2 and s(y l , ẑ) = 0 for ẑ ∈ Z 0 , while the

probability that ẑ ∈ Z 2 is given by π 2 in (19) and the probability that ẑ ∈ Z 2 is π 0 = 1 − π 2 .

We prove that, as long as µ u − µ J (1 − δ + µ u ) > 0, FV + (u) is bounded and such that

FV + (u 1 ) − FV + (u 0 ) is greater than D V+

(u 1 − u 0 ) and smaller than D V+ (u 1 − u 0 ) for all

u 0 , u 1 with 0 ≤ u 0 ≤ u 1 ≤ 1. Similarly, we prove that FJ + (u) is bounded and such that

FJ + (u 1 ) − FJ + (u 0 ) is greater than D J+

(u 1 − u 0 ) and smaller than D J+ (u 1 − u 0 ) for all u 0 , u 1

with 0 ≤ u 0 ≤ u 1 ≤ 1.

The above observations imply that the operator F is a self-map, in the sense that it maps

pairs of functions in the set Ω into pairs of functions that also belong to the set Ω. The

set Ω is a non-empty, bounded, closed convex subset of the space of bounded continuous

functions with the sup norm (see Lemma A.1 in Menzio and Shi, 2010). We also establish

that the operator F is continuous. Finally, we establish that the family of functions F(Ω) is

equicontinuous. Equicontinuity, which is typically rather diffi cult to establish, here follows

immediately because the functions FV + and FJ + have the same Lipschitz bounds for all

(V + , J + ) ∈ Ω.

The properties of the operator F are the conditions of Schauder’s fixed point theorem

(see Theorem 17.4 in Stokey, Lucas and Prescott 1989). Thus, there exists a pair of functions

(V+, ∗ J+) ∗ ∈ Ω such that F(V+, ∗ J+) ∗ = (V+, ∗ J+). ∗ Given the functions V+ ∗ and J+, ∗ we construct

the associated wage function, w ∗ , the gains from trade to the worker and to the firm, V ∗ and

J ∗ , and the law of motion for unemployment, h ∗ . These objects, together with an optimal

contract x ∗ (u) that prescribes the wage w ∗ (u) and the firing probabilities s ∗ (y h , ẑ, u) = 0,

s ∗ (y h , ẑ, u) = 1 for ẑ ∈ Z 2 and s ∗ (y h , ẑ, u) = 0 for ẑ ∈ Z 0 constitute a Recursive Equilibrium

in which, for all u, the equilibrium stage game is such that firms perfectly coordinate on

firing or keeping their non-performing workers.

This completes the proof of the following theorem.

27

Theorem 4: (Recursive Equilibrium) Assume (1 − δ)p h (1) − µ(u, J) > 0 and µ u − µ J (1 −

δ + µ u ) > 0. For all (β, ψ) such that β ∈ (0, β ∗ ) and ψ ∈ (0, ψ ∗ ), where β ∗ > 0 and ψ ∗ > 0,

there is a Recursive Equilibrium in which the stage equilibrium is a Perfect Coordination

Equilibrium.

5 **Agency** **Business** **Cycles**

In a Perfect Coordination Equilibrium, the economy experiences aggregate fluctuations that

are endogenous and stochastic. We refer to these aggregate fluctuations as **Agency** **Business**

**Cycles** (ABC). In this section, we calibrate the model to the US economy to assess the

magnitude of ABC. We find that ABC can generate large fluctuations in unemployment,

in the rate at which employed workers become unemployed (EU rate) and in the rate at

which unemployed workers become employed (the UE rate). We then test three distinctive

features of ABC. First, in ABC, a recession starts with an increase in the EU rate which

drives up unemployment and, because of decreasing returns to scale in matching, lowers the

UE rate. This causal chain implies that the EU rate leads the unemployment rate and the

UE rate. We find the same pattern of leads and lags in the US data. Second, in ABC, the

probability of a recession is endogenous and, in particular, it becomes higher the lower is

unemployment. We find that, in the US economy, the probability of a recession depends

negatively on unemployment. Third, in ABC, a recession is a period when the value of time

in the market relative to the value of time at home is abnormally high. We find preliminary

evidence that this is also the case for the US economy.

5.1 Calibration

We start by calibrating the primitives of the model. Preferences are described by the worker’s

periodical utility function, υ(c) − ψe, and by the discount factor, β. Market production is

described by the realizations of output, y h and y l , by the probability that output is high

given the worker’s effort, p h (1) and p h (0), and by the exogenous job destruction probability

δ. Home production is described by the output of an unemployed worker, b. The search and

matching process is described by the vacancy cost, k, and the matching function, M(u, v).

We specialize the utility function for consumption to be of the form υ(c) = c 1−σ /(1 − σ),

where σ is the coeffi cient of relative risk aversion. We specialize the matching function to be

of the form M(u, v) = A(u)m(u, v), where m(u, v) = uv(u ξ + v ξ ) −1/ξ is a constant returns

to scale matching function with an elasticity of substitution ξ, and A(u) = exp(−ρu) is a

matching effi ciency function with a semi-elasticity with respect to unemployment of −ρ.

28

We calibrate the parameters of the model to match some key statistics of the US labor

market between 1951 and 2014, such as the average unemployment rate, the average UE

rate, and the average EU rate. We measure the US unemployment rate as the CPS civilian

unemployment rate. We measure the UE and the EU rates using the civilian unemployment

and short-term unemployment rates from the CPS, following the same methodology as in

Shimer (2005). We measure labor productivity as output per worker in the non-farm sector.

We calibrate the basic parameters of the model as is now standard in the literature.

We choose the model period to be one month. We set the discount factor, β, so that the

annual real interest rate, (1/β) 1/12 − 1, is 5 percent. We choose the vacancy cost, k, and the

exogenous job destruction probability, δ, so that the model matches the average UE and EU

rates in the US economy (respectively, 44% and 2.6%). We normalize the average value of

market production, p h (1)y h + (1 − p h (1))y l , to 1. We choose the value of home production, b,

to be 70% of the average value of market production, which Hall and Milgrom (2010) argue

is a reasonable estimate for the US economy.

We calibrate the parameters of the model that determine the extent of the agency problem

as follows. The probability that the realization of output is y h given that the worker exerts

effort, p h (1), affects the number of non-performing workers and, hence, the magnitude of

firing bursts. The disutility of effort, ψ, affects the frequency at which firms need to fire nonperforming

workers, hence, the frequency of firing bursts. Therefore, we choose y h so that

the model generates the same standard deviation in the cyclical component of the EU rate

as in the US economy (9.85%). We choose ψ so that, on average, firms coordinate on firing

their non-performing workers once every 50 months. The parameters y h and y l and p h (0)

cannot be uniquely pinned down. Given that average output is 1, the realizations of output

y h and y l do not affect the equilibrium, as long as it is optimal for firms to require effort

from their workers. Similarly, the probability p h (0) only affects the equilibrium through the

ratio ψ/(p h (1) − p h (0)). Therefore, we choose some arbitrary values for y h , y l and p h (0) such

that firms find it optimal to require effort.

Finally, we need to choose values for the parameters in the utility and the matching

functions. We set the coeffi cient σ of relative risk aversion in the utility function υ to 1.

This is a standard value from micro-estimates of risk aversion. We set the elasticity ξ of

substitution between unemployment and vacancy in the matching function m to 1.24. This

is the value estimated by Menzio and Shi (2011). 9 We tentatively set the parameter ρ in the

9 Correctly estimating a matching function requires taking into account the fact that unemployed and

employed workers all search for vacancies to some degree. Using a model of search off and on the job,

29

matching effi ciency function A to 6. 10

Table 1: Calibrated Parameters

Description

Value

y h high output 1.03

y l low output .000

p h (1) probability of y h given e = 1 .967

p h (0) probability of y h given e = 0 .467

b UI benefit/value of leisure .700

ψ disutility of effort .007

δ exogenous job destruction .025

k vacancy cost .257

ξ elasticity of sub. btw u and v 1.24

ρ semi-elasticity of A wrt u 6.00

5.2 Magnitude and Properties of ABC

Table 1 reports the calibrated value of the parameters of the model. Given these values, we

simulate the model and create monthly time-series for the unemployment rate, the UE rate,

the EU rate and other labor market variables. For each variable, we construct quarterly

time-series by taking 3-month averages. We then compute the cyclical component of each

variable as the percentage deviation of its quarterly value from a Hodrick-Prescott trend

constructed using a smoothing parameter of 10 5 . We use the same procedure to construct

the cyclical component of labor market variables in the US data. Figure D1 in Appendix

D presents a sample of the time-series generated by the model for the cyclical component

of the unemployment rate, the UE rate and the EU rate. Figure D2 presents the cyclical

component of the unemployment rate, the UE rate and the EU rate in the US economy over

the period 1990-2014.

Table 2 reports some statistics about the unemployment rate, the UE rate, the EU rate

and the labor productivity generated by the model and about the same variables in the data.

Menzio and Shi (2011) estimate the elasticity of substitution between searching workers and vacant jobs in

the matching function to be 1.24.

10 When we calibrate the model using a higher value for ρ, we find that the model generates larger fluctuations

in the UE rate and, hence, larger fluctuations in the unemployment rate. The qualitative predictions of

the model, though, remain unchanged. We refer the reader to Footnote xx for a justification of our baseline

choice of rho.

30

First, Table 2 shows that ABC can account for a significant fraction of the volatility of the

US labor market. Specifically, the standard deviation of unemployment in the model is 55%

of what we observe in the data. Similarly, the standard deviation of the UE rate in the

model is 33% of its empirical counterpart, and the standard deviation of the EU rate in the

model is the same as in the data. When we recalibrate the model using a higher value of ρ,

the model generates larger fluctuations in the UE rate and, hence, larger fluctuations in the

unemployment rate. 11 The reader should keep in mind that the model is calibrated under

the identifying assumption that the volatility in the EU rate observed in the data is entirely

explained by our theory. 12 Table 2: **Agency** **Business** **Cycles**

u rate UE rate EU rate APL

Model std 9.34 4.09 9.11 0

cor. wrt u 1 -.98 .32 -

Data: 1951-2014 std 16.9 12.9 9.7 1.98

cor wrt u 1 -.94 .80 -.37

Data: 1984-2014 std 17.3 13.8 6.91 1.38

cor wrt u 1 -.96 .70 .09

Second, Table 2 shows that ABC feature the same pattern of comovement between the

unemployment rate, the UE rate and the EU rate as in the data. In particular, in the model

as in the data, the unemployment rate and the UE rate are negatively correlated, while

the unemployment rate and the EU rate are positively correlated. However, in the model

unemployment and vacancies are mildly positively correlated, while in the data these two

variables are almost perfectly negatively correlated. This discrepancy between model and

data is an artifact of the simplifying and counterfactual assumption that workers search the

labor market only when they are unemployed. Indeed, under the assumption of off-the-job

search, an increase in unemployment causes an increase in the number of workers searching

11 We are not arguing that ABC are the only source of aggegate fluctuations in the labor market and,

hence, that it should explain all of the empirical volatility of unemployment, UE and EU rates. Indeed, we

believe that ABC can either create additional fluctuations relative to those caused by fundamental shocks,

or that correlated firings might amplify fundamental shocks. However, as we want to isolate the effect of

ABC, in this paper we abstract from all fundamental shocks.

12 Nonetheless, the finding that ABC can create large fluctuations in the unemployment, the UE and the

EU rates is important. Indeed, as shown by Shimer (2005), the textbook search-theoretic model of the labor

market with productivity shocks explains less than 10% of the empirical volatility of unemployment. Hall

(2005), Menzio (2005), Hagedorn and Manovskii (2008), Kennan (2010), Menzio and Shi (2011) develop

search-theoretic models in which productivity shocks generate larger fluctuations in unemployment. However,

these models counterfactually predict a perfect negative correlation between unemployment and labor

productivity.

31

the labor market which, in turn, gives firms an incentive to create more vacancies. Under the

more realistic assumption of on and off-the-job search, an increase in unemployment does

not cause an increase in the number of workers searching the labor market and, hence, does

not give firms a clear incentive to create more vacancies.

Finally, Table 2 shows that the model generates relatively large fluctuations in unemployment

that are uncorrelated with fluctuations in labor productivity. This is an important

feature of the model.

The empirical correlation between unemployment and labor

productivity– which was significantly negative for the period 1951-1984– has become basically

zero for the period 1984-2014.

Therefore, over the period 1951-1984, fluctuations

in labor productivity may have driven the cyclical movements of the US labor market. In

contrast, over the period 1984-2014, fluctuations in labor productivity seem an unlikely

driver of cycles in the US labor market. Our theory provides an explanation for the recent

lack of comovement between labor productivity and unemployment by identifying a novel,

non-technological source of aggregate fluctuations. 13

Overall, Table 2 shows that **Agency** **Business** **Cycles** can be large.

Now, we turn to

examine some of the distinctive features of ABC. According to ABC, a recession starts with

an increase in the EU rate which drives up unemployment and, because of decreasing returns

to scale in matching, lowers the UE rate. This causal chain can be seen in Figure 3, where we

display the correlation between unemployment in quarter t and other labor market variables

in quarter t + x, with x going from −5 to +5. The red solid line is the correlation between

the EU rate in quarter t + x and the unemployment rate in quarter t. The green solid line is

the correlation between the UE rate in quarter t+x and the unemployment rate in quarter t.

The black solid line is the correlation between the unemployment rate in quarter t+x and in

quarter t. It is immediate to see that the EU rate leads the unemployment rate by a quarter–

in the sense that the absolute value of the correlation between unemployment in quarter t

and the EU rate in quarter t+x is highest for x = −1– while the UE rate is contemporaneous

with unemployment—in the sense that the correlation between unemployment in quarter t

and the UE rate in quarter t + x is highest for x = 0. Moreover, the correlation between

current unemployment and the future EU rate dies offmuch more rapidly than the correlation

between current unemployment and the future UE rate.

The US labor market displays

exactly the same pattern of leads and lags, as can be seen from the dashed lines in Figure 3.

13 Gali and van Rens (2014) document in detail the decline in the negative correlation between unemployment

and labor productivity. Gali and van Rens (2014), Kaplan and Menzio (2015) and Beaudry, Galizia and

Portier (2015) advance theories, alternative to ours, in which labor productivity does not correlate negatively

with unemployment.

32

Figure 3: Leads and Lags

Correlation with u at 0

1

0.8

0.6

0.4

0.2

0

-0.2

-0.4

-0.6

-0.8

-1

-5 -4 -3 -2 -1 0 1 2 3 4 5

u-model EU-model UE-model

u-data EU-data UE-data

The main difference between the correlation functions in the model and in the data is that,

in the model, all correlations die off more quickly than in the data. This shortcoming of the

model is due to the fact that the driving force behind ABC (i.e. the coordinated firing of

non-performing workers) has no persistence. 14

Fujita and Ramey (2009) were the first to point out that the EU rate leads the unemployment

rate, while the UE rate is contemporaneous with the unemployment rate. Our theory

explains this pattern as a causal link from the EU rate to the unemployment rate and from

the unemployment rate to the UE rate. 15 Also, Fujita and Ramey (2009) used a VAR model

to show that, once one takes into account the negative correlation between the current EU

rate and the future UE rate, fluctuations in the EU rate can account for approximately 60%

of the overall volatility of unemployment. Our theory can explain this finding. Indeed, as

shown in Table 2, our theory implies that the fluctuations to the EU rate account for 60%

of the volatility of unemployment. 16

14 We believe we could create persistence in firings by assuming that firm and workers observe output in a

staggered fashion– let’s say half in odd periods and half in even periods.

15 Ahn and Hamilton (2015) estimate a statistical model of flows in and out of unemployment in which

workers differ by their job-finding probability. They find that the increase in the EU rate at the onset of

the Great Recession contained a disproportionate fraction of low job-finding probability workers and, hence,

caused the subsequent decline in the UE rate. They also find that the low job-finding probability workers

are typically those fired from their job.

16 Since ρ determines the effect of shocks to the EU rate on the UE rate and, hence, on the unemployment

33

In ABC, the probability of a recession is endogenous and given by equation (19). The

lower is unemployment in (19), the lower is the cost to the worker from losing his job and,

hence, the higher is the probability with which firms need to fire their non-performing workers

in order to give them an incentive to perform. Thus, the lower is unemployment, the higher

is the probability that a recession starts. In order to find out whether this feature of our

theory is borne out in the data, we take the time-series for the unemployment rate and define

the start of a recession as a quarter in which the unemployment rate turns from decreasing

to increasing and keeps growing for at least two consecutive quarters. 17 We then estimate a

probit model for the probability of the start of a recession as a function of the unemployment

rate. The estimated coeffi cient on the unemployment rate is −.21, with a standard deviation

of 11%. The estimated coeffi cient implies that the probability of a recession increases from

8% to 11% as the unemployment falls from 5.5% to 4.5%. Clearly, it is not possible to

precisely estimate the effect of unemployment on the probability that a recession starts

because recessions are relatively rare events. In order to gather more observations, we take

the time-series of unemployment for Australia, Canada, Italy, Japan, France and the UK. 18

After taking out the average unemployment from the time-series of each country, we merge

these data to those for the US and re-estimate the probit model. The estimated coeffi cient

on the unemployment rate is −.20 with a standard deviation of 4%.

We now turn to testing what is perhaps the most distinctive feature of ABC. In our theory,

a recession is a period when the value of time in the market relative to its value at home is

abnormally high. In contrast, in the Real **Business** Cycle theory of Kydland and Prescott

(1982), in Mortensen and Pissarides (1994) or in any other theory where business cycles are

driven by either exogenous or endogenous shocks to the value of production, a recession is a

period when the value of time in the market relative to its value at home is abnormally low.

To paint a picture, in RBC, a recession is a day when it is raining in the marketplace and,

for that reason, workers find it optimal to stay at home. In ABC, a recession is a day when

the TV set is broken at home and, for that reason, firms find it optimal to get rid of their

non-performing workers. It is natural to wonder whether, empirically, the relative value of

labor in the market is pro or countercyclical. In order to address this question, we construct

rate, our choice for the value of rho is such that the model is consistent with the VAR findings of Fujita and

Ramey (2009).

17 The estimates are robust to alternative ways to define the start of a recession.

18 These are countries for which the OECD provides suffi ciently long time-series for unemployment. We

drop Germany from the sample because of the large movements in unemployment related to the unification.

When estimated for each country separately, the coeffi cient on unemployment in the probit model is always

negative. In Appendix D, we report the estimated relationships between unemployment and the probability

of a recession for each country separately and for the pool of countries.

34

some rudimentary empirical measures of the net value of employment to a worker.

We measure the value of employment to a worker, W 1,t , and the value of unemployment

to a worker, W 0,t , as

W 1,t = w t + β

h EU

t+1W 0,t+1 + (1 − h EU

t+1)W 1,t+1 ,

W 0,t = b t + β (26)

h UE

t+1W 1,t+1 + (1 − h UE

t+1)W 0,t+1 ,

where w t is a measure of the real wage in month t, b t is a measure of unemployment benefit/value

of leisure in month t, h EU

t+1 is a measure of the EU rate in month t + 1, h UE

t+1 is a

measure of the UE rate in month t + 1, and W 1,t+1 and W 0,t+1 are respectively the value of

employment and unemployment in month t + 1. We measure the net value of employment

to a worker, V t , in month t as the difference between W 1,t and W 0,t .

We measure w t using the time-series for the hourly wage that have been constructed by

Haefke, Sonntag and van Rens (2013). We consider two alternative time-series: the average

hourly wage in the cross-section of all employed workers, and the average hourly wage in the

cross-section of newly hired workers after controlling for the composition of new hires. The

first time-series may be more appropriate when we want to interpret V t as the cost of losing a

job to a worker, the second-time series may be more appropriate when we want to interpret

V t as the benefit of finding a job to a worker. As in Shimer (2005), we measure h UE

t and h EU

t

using, respectively, the values for the EU rate and UE rates implied by the time-series for

unemployment and short-term unemployment. In order to make the time-series for w t , h UE

t

and h EU

t stationary, we construct their Hodrick-Prescott trend using a smoothing parameter

of 10 5 . We then take the difference between the value of each variable and its trend and add

this difference to the time-series average for that variable. Since b t is not directly observable,

we tentatively set it to be equal to 70% of the average of w t .

We are now in the position to construct time-series for the value of employment to a

worker, W 1,t , and the value of unemployment to a worker, W 0,t , over the period going from

January 1979 to December 2014. We compute the values for W 1 and W 0 in December 2014

by assuming that, from January 2015 onwards, w, h UE , and h EU are equal to their historical

averages. Given the values for W 1 and W 0 in December 2014, we compute the values for W 1

and W 0 from November 2014 back to January 1979 by using equation (26) and the timeseries

for w, b, h UE , and h EU . The reader should notice that the values of W 1 and W 0 thus

computed differ from their theoretical counterpart because they are constructed using the

realizations of future w, h UE , and h EU , rather than the expectation of these variables.

Figure 4 presents the result of our calculations. Figure 4 displays the time-series for the

net value of employment to a worker, V t , computed using the average wage of all employed

35

Figure 4: Net Value of Employment

18

16

14

12

10

8

6

4

2

0

1979 1984 1989 1994 1999 2004 2009 2014

V-new V-all u

workers (solid black line) and the average wage of newly hired workers (dashed black line).

Figure 4 also displays the time-series for the detrended unemployment rate (solid grey line).

The figure clearly shows that V t is countercyclical, oin the sense that V t moves together with

the unemployment rate. This is the case whether we measure V t using the average wage

of all employed workers– in which the correlation between V t and unemployment is 80%–

or whether we measure V t using the average wage of newly hired workers– in which case

the correlation between V t and unemployment is 71%. Mechanically, V t is countercyclical

because, when unemployment increases, the decline in the value of being unemployed caused

by the large decline in the UE rate is larger than the decline in the value of being employed

cause by the small decline in the wage and by the short-lived increase in the EU rate. The

countercyclicality of V t is a very robust finding. The measure of V t becomes even more

countercyclical if b t is assumed to be constant fraction of w t rather than a constant. The

correlation between V t and u t remains practically unchanged if we do not filter the data.

The finding that V t is countercyclical means that recessions are times when unemployed

workers find it especially valuable to find a job, and when employed workers find it especially

costly to lose a job. The finding is in stark contrast with the view of recessions as “days of

rain in the marketplace” advanced by Kydland and Prescott (1982) or by Mortensen and

Pissarides (1994). In contrast, the finding is supportive of the view of recessions as “days

of no TV at home” advanced by our theory. Notice that our finding that V t should not

36

e entirely surprising. In fact, using a different, more sophisticated approach and richer

data, Davis and von Wachter (2011) show that the lifetime earning cost of losing a job is

much higher in recessions than in expansions. Even if one is skeptical about our theory

of recessions, the finding that V t is countercyclical represents a serious challenge for most

existing theories of business cycles. 19

6 Conclusions

This paper proposed a new theory of business cycles. At a very abstract level, our theory

states that business cycles emerge because different agents in the economy find it optimal to

randomize over some individual decision in a perfectly correlated fashion. More concretely,

business cycles emerge because firms need to randomize over firing or keeping workers who

have performed poorly in the past, in order to give them an ex-ante incentive to perform.

Moreover, firms find it optimal to correlate the randomization outcomes, as doing so allows

them to load up the firing probability on states of the world in which it is costlier for

workers to become unemployed and, hence, it allows them to economize on agency costs.

In the unique robust equilibrium, firms use a sunspot to perfectly correlate the outcome

of their individual randomizations. In this equilibrium, the economy experiences aggregate

fluctuations that are endogenous– in the sense that they are not caused by exogenous shocks

to fundamentals or by exogenous shocks to the selection of equilibrium, but they are an

inherent feature of the unique equilibrium– and are stochastic– in the sense that they do

not follow a deterministic path, but a genuinely stochastic one. We believe that this may be

the first theory of endogenous and stochastic business cycles.

References

[1] Ahn, H., and J. Hamilton. 2015. “Heterogeneity and Unemployment Dynamics.”Manuscript,

University of California at San Diego.

[2] Angeletos, M., and J. La’O. 2013. “Sentiments.”Econometrica, 81: 739-780.

19 An alternative explanation is that business cycles are driven by productivity shocks and wages are rigid

(see, e.g., Hall 2005, Menzio 2005, Gertler and Trigari 2009, Menzio and Moen 2010 and Kennan 2010 for

search theoretic models of the labor market with rigid wages). In fact, if wages are rigid, productivity shocks

may generate, just like in our theory, a positive correlation between unemployment and V t . However, if

wages are rigid, productivity shocks should generate a strong negative correlation between unemployment

and the value of a worker to a firm J t , while this correlation is positive in our theory. Applying the same

technique used to compute the time-series for V t , we can compute a time-series for J t . Using average labor

productivity as our measure of output per worker, we find that J t is positively correlated with u t .

37

[3] Banerjee, A. 1992. “A Simple Model of Herd Behavior.” Quarterly Journal of Economics,

107: 797-817.

[4] Beaudry, P., D. Galizia, and F. Portier. 2015 “Reviving the Limit Cycle View of Macroeconomic

Fluctuations,”NBER Working Paper 21241.

[5] Beaudry, P., and F. Portier. 2004. “Exploring Pigou’s Theory of **Cycles**.” Journal of

Monetary Economics, 51: 1183-1216.

[6] Benhabib, J., and R. Farmer. 1994. “Indeterminacy and Increasing Returns.” Journal

of Economic Theory 63: 19—41.

[7] Bikhchandani, S., D. Hirshleifer, and I. Welch. 1992. “A Theory of Fads, Fashion,

Custom and Cultural Change as Information Cascades.”Journal of Political Economy

100: 992-1026.

[8] Bloom, N. 2009. “The Impact of Uncertainty Shocks.”Econometrica, 77: 623-685.

[9] Boldrin, M., and L. Montrucchio. 1986. “On the Indeterminacy of Capital Accumulation

Paths.”Journal of Economic Theory, 40: 26-39.

[10] Boldrin, M., and M. Woodford. 1990. “Equilibrium Models Displaying Endogenous

Fluctuations and Chaos.”Journal of Monetary Economics, 25: 189-222.

[11] Burdett, K., S. Shi, and R. Wright. 2001. “Pricing and Matching with Frictions.”Journal

of Political Economy 109: 1060-85.

[12] Clementi, G., and H. Hopenhayn. 2006. “A Theory of Financing Constraints and Firm

Dynamics.”Quarterly Journal of Economics 121: 229-65.

[13] Cooper, R., and A. John. 1988. “Coordinating Coordination Failures in Keynesian Models.”Quarterly

Journal of Economics, 103: 441—63.

[14] Davis, S., and T. von Wachter. 2011. “Recessions and the Costs of Job Loss.”Brookings

Papers on Economic Activity.

[15] Diamond, P. 1982. “Aggregate Demand Management in Search Equilibrium.” Journal

of Political Economy 90: 881—94.

[16] Diamond, P., and D. Fudenberg. 1989. “Rational Expectations **Business** **Cycles** in Search

Equilibrium.”Journal of Political Economy 97: 606—19.

38

[17] Fujita, S., and G. Ramey. 2009. “The Cyclicality of Separation and Job Finding Rates.”

International Economic Review 50: 415—30.

[18] Gabaix, X. 2011. “The Granular Origins of Aggregate Fluctuations.”Econometrica 79:

733-772.

[19] Galí, J., and T. van Rens. 2014. “The Vanishing Procyclicality of Labor Productivity.”

Manuscript, University of Warwick.

[20] Gertler, M., and A. Trigari. 2009. “Unemployment Dynamics with Staggered Nash Wage

Bargaining.”Journal of Political Economy, 117: 38-86.

[21] Haefke, C., M. Sonntag, and T. van Rens, 2013. “Wage Rigidity and Job Creation.”

Journal of Monetary Economics.

[22] Hagedorn, M., and I. Manovskii. 2008. “The Cyclical Behavior of Unemployment and

Vacancies Revisited.”American Economic Review 98: 1692—706.

[23] Hall, R., 2005. “Employment Fluctuations with Equilibrium Wage Stickiness.”American

Economic Review 95: 50-65.

[24] Hall, R., and P. Milgrom. 2008. “The Limited Influence of Unemployment of the Wage

Bargain.”American Economic Review 98: 1653—74.

[25] Heller, W. 1986. “Coordination Failure Under Complete Markets with Applications to

Effective Demand.” Equilibrium analysis: Essays in Honor of Kenneth J. Arrow, 2:

155—175, 1986.

[26] Jaimovich, N., and S. Rebelo. 2009. “Can News About the Future Drive **Business** **Cycles**?”American

Economic Review 99: 1097-1118.

[27] Jovanovic, B. 1987. “Micro Shocks and Aggregate Risk.” Quarterly Journal of Economics

102: 395-409.

[28] Kaplan, G., and G. Menzio. 2015. “Shopping Externalities and Self-Fulfilling Unemployment

Fluctuations.”Journal of Political Economy, Forthcoming.

[29] Kennan, J. 2010. “Private Information, Wage Bargaining and Employment Fluctuations.”Review

of Economic Studies 77: 633-64.

[30] Kydland, F., and E. Prescott. 1982. “Time to Build and Aggregate Fluctuations.”

Econometrica 50: 1345-70.

39

[31] Menzio, G. 2005. “High Frequency Wage Rigidity.”Manuscript, Northwestern University.

[32] Menzio, G., and E. Moen. 2010. “Worker Replacement.” Journal of Monetary Economics,

57: 623-36.

[33] Menzio, G., and S. Shi. 2010. “Block Recursive Equilibria for Stochastic Models of

Search on the Job.”Journal of Economic Theory 145: 1453-94.

[34] Menzio, G., and S. Shi. 2011. “Effi cient Search on the Job and the **Business** Cycle.”

Journal of Political Economy 119: 468—510.

[35] Mortensen, D. 1999. “Equilibrium Unemployment Dynamics.”International Economic

Review, 40: 889—914.

[36] Mortensen, D., and C. Pissarides. 1994. “Job Creation and Job Destruction in the

Theory of Unemployment.”Review of Economic Studies 61: 397—415.

[37] Petrongolo, B., and C. Pissarides. 2001. “Looking into the Black Box: A Survey of the

Matching Function.”Journal of Economic Literature 39: 390-431.

[38] Pissarides, C. 1985. “Short-Run Equilibrium Dynamics of Unemployment, Vacancies

and Real Wages.”American Economic Review 75: 676—90.

[39] Shimer, R. 2005. “The Cyclical Behavior of Unemployment and Vacancies.”American

Economic Review 95: 25—49.

[40] Stokey, N., R. Lucas, and E. Prescott, “Recursive Methods in Economic Dynamics.”

Harvard University Press, Cambridge, Massachusetts, 1989.

40

A Proof of Lemmas 1-4

A.1 Proof of Lemma 1

Appendix

Let ρ ≥ 0 denote the Lagrange multiplier on the worker’s incentive compatibility constraint,

let ν(y, ẑ) ≥ 0 denote the multiplier on the constraint 1 − s(y, ẑ) ≥ 0 and let ν(y, ẑ) denote

the multiplier on the constraint s(y, ẑ) ≥ 0.

The first order condition with respect to the firing probability s(y l , ẑ) is given by

(1 − δ) [F (x)V (ẑ) + W (x)J(ẑ)] = ρβ(1 − δ)(p h (1) − p h (0))V (ẑ) + ν(y l , ẑ) − ν(y l , ẑ), (A1)

together with the complementary slackness conditions ν(y l , ẑ)·(1−s(y l , ẑ)) = 0 and ν(y l , ẑ)·

s(y l , ẑ) = 0. The left-hand side of (A1) is the marginal cost of increasing s(y l , ẑ). This cost

is given by the decline in the product of the worker’s and firm’s gains from trade caused

by a marginal increase in the firing probability s(y l , ẑ). The right-hand side of (A1) is the

marginal benefit of increasing s(y l , ẑ). This benefit is given by the value of relaxing the

worker’s incentive compatibility and the s(y l , ẑ) ≥ 0 constraints net of the cost of tightening

the s(y l , ẑ) ≤ 1 constraint by marginally increasing the firing probability s(y l , ẑ).

by

Similarly, the first order condition with respect to the firing probability s(y h , ẑ) is given

(1 − δ) [F (x)V (ẑ) + W (x)J(ẑ) + ρβ(p h (1) − p h (0))V (ẑ)] = ν(y h , ẑ) − ν(y h , ẑ), (A2)

together with the complementary slackness conditions ν(y h , ẑ)·(1−s(y h , ẑ)) = 0 and ν(y h , ẑ)·

s(y h , ẑ) = 0. The left-hand side of (A2) represents the marginal cost of increasing s(y h , ẑ).

The rigth-hand side of (A2) represents the marginal benefit of increasing s(y h , ẑ). Notice

that increasing the firing probability s(y h , ẑ) tightens the worker’s incentive compatibility

constraint and, hence, the term in ρ is now on the left-hand side of (A2).

Suppose ρ = 0. First, notice that the left-hand side of (A1) is strictly positive as V (ẑ) > 0,

J(ẑ) > 0 by assumption, and W (x) > 0, F (x) > 0 at the optimum x ∗ . The right-hand side

of (A1) is strictly positive only if ν(y l , ẑ) > 0. Hence, if ρ = 0, the only solution to the

first order condition with respect to the firing probability s(y l , ẑ) is 0. Next, notice that

the left-hand side of (A2) is strictly positive and the right-hand side is strictly positive

only if ν(y h , ẑ) > 0. Hence, if ρ = 0, the only solution to the first order condition with

respect to the firing probability s(y h , ẑ) is 0. However, if s(y l , ẑ) = s(y h , ẑ) = 0, the worker’s

41

incentive compatibility constraint is violated. Therefore, ρ > 0 and the worker’s incentive

compatibility constraint holds with equality.

A.2 Proof of Lemma 2

The first order condition with respect to s(y h , ẑ) is given by (A2) together with the complementary

slackness conditions ν(y h , ẑ)(1 − s(y h , ẑ)) = 0 and ν(y h , ẑ)s(y h , ẑ) = 0. The

left-hand side of (A2) is strictly positive. The right-hand side of (A2) is strictly positive

only if ν(y h , ẑ) > 0. Therefore, the first order condition is satisfied only if ν(y h , ẑ) > 0 and,

hence, only if s(y h , ẑ) = 0.

A.3 Proof of Lemma 3

Using the definition of φ(ẑ), we can rewrite the first order condition with respect to the firing

probability s(y l , ẑ) as

(1 − δ)V (ẑ) [F (x) + W (x)/φ(ẑ) − ρβ(p h (1) − p h (0))] = ν(y l , ẑ) − ν(y l , ẑ), (A3)

together with ν(y l , ẑ) · (1 − s(y l , ẑ)) = 0 and ν(y l , ẑ) · s(y l , ẑ) = 0. The left-hand side of

(A3) is strictly decreasing in φ(ẑ). The right-hand side of (A3) is strictly positive if ν(y l , ẑ)

is strictly positive and it is strictly negative if ν(y l , ẑ) is strictly positive. Therefore, there

exists a φ ∗ such that if φ(ẑ) > φ ∗ , the left-hand side is strictly negative and the solution to

(A3) requires ν(y l , ẑ) > 0. In this case, the solution to the first order condition for s(y l , ẑ)

is 1. If φ(ẑ) < φ ∗ , the left-hand side is strictly positive and the solution to (A3) requires

ν(y l , ẑ) > 0. In this case, the solution to the first order condition for s(y l , ẑ) is 0.

A.4 Proof of Lemma 4

The first order condition with respect to the wage w is given by

F (x)υ ′ (w) − W (x) = 0.

(A4)

The left-hand side of (A4) is the increase in the product of the worker’s and firm’s gains

caused by a marginal increase in the worker’s wage w. A marginal increase in w, increases the

worker’s gains from trade by υ ′ (w) and decreases the firm’s gains from trade by 1. Therefore,

a marginal increase in w, increases the product of the worker’s and firm’s gains from trade

by F (x)υ ′ (w) − W (x). The first order condition for w states that the effect of a marginal

increase in w is zero.

42

B Proof of Lemmas 5-6

B.1 Proof of Lemma 5

We first consider a firm-worker pair in group K. Since P K (T K |T K−1 ) is strictly increasing in

T K−1 and û is strictly increasing in T K , E[û|T K−1 ] is strictly increasing in T K−1 . In turn, since

E[û|T K−1 ] is strictly increasing in T K−1 and φ(û) is strictly increasing in û, φ(E[û|T K−1 ]) is

strictly increasing in T K−1 . It then follows from property (iii) of the optimal contract that

there exists a T ∗ K−1 such that s K(y l , T K−1 ) = 0 for all T K−1 < T ∗ K−1 , and s K(y l , T K−1 ) = 1

for all T K−1 > T ∗ K−1 . This establishes that the firing probability s K(y l , T K−1 ) has the desired

threshold property. The threshold property implies that the measure t K of workers separating

from firms in group K is increasing in the measure T K−1 of workers separating from firms in

groups 1 through K − 1.

Next, consider a firm-worker pair in group K − 1. Since P K−1 (T K−1 |T K−2 ) is strictly

increasing in T K−2 and t K is increasing in T K−1 , it follows that E[û|T K−2 ] and, in turn,

φ(E[û|T K−2 ]) are strictly increasing in T K−2 . Then property (iii) of the optimal contract

implies that there exists a T ∗ K−2 such that s K−1(y l , T K−2 ) = 0 for all T K−2 < T ∗ K−2 , and

s K−1 (y l , T K−2 ) = 1 for all T K−2 > TK−2 ∗ . This establishes that the firing probability

s K−1 (y l , T K−2 ) has the desired threshold property. The threshold property implies that the

measure t K−1 of workers separating from firms in group K − 1 is increasing in the measure

T K−2 of workers separating from firms in groups 1 through K − 2. By repeating the above

argument for firm-worker pairs in groups K − i, with i = 2, 3, . . . K − 1, we can establish

that the firing probability s K−i (y l , T K−i−1 ) has the threshold property and that the measure

t K−i is increasing in T K−i−1 .

B.2 Proof of Lemma 6

Let T1 ∗ be given as in (22) and let Ti

∗ = E[t 1 ]+(i−1)(t h −t l )/2 for i = 2, 3, . . . K −1. Given

the thresholds {T1 ∗ , T2 ∗ , ...TK−1 ∗ }, we can compute the unconditional probability distribution

of the random variable T i . For N → ∞, T 2 is approximately equal to E[t 1 ]+t l if T 1 < T1 ∗ , and

it is approximately equal to E[t 1 ]+t h if T 1 > T ∗ 1 . Since E[t 1 ]+t l < T ∗ 2 and E[t 1 ]+t h > T ∗ 2 , T 3

is approximately equal to E[t 1 ] + 2t l if T 1 < T ∗ 1 , and T 3 is approximately equal to E[t 1 ] + 2t h

if T 1 > T ∗ 1 . Similarly, for i = 4, 5, . . . K, T i is approximately equal to E[t 1 ] + (i − 1)t l if

T 1 < T ∗ 1 , and it is approximately equal to E[t 1 ] + (i − 1)t h if T 1 > T ∗ 1 . Hence, if T 1 < T ∗ 1 ,

firms in groups 2 through K fire their non-performing workers with probability 0 and the

total measure of workers separating from firms is T K = E[t 1 ] + (K − 1)t l . If T 1 > T ∗ 1 , firms

43

in groups 2 through K fire their non-performing workers with probability 1 and the total

measure of workers separating from firms is T k = E[t 1 ] + (K − 1)t h .

The above observations imply that benefit from exerting effort for a worker employed at

a firm in group i = 2, 3, . . . K is given by

β(1 − δ)(p h (1) − p h (0))(1 − P 1 (T1 ∗ ))V (E[û|T K = E[t 1 ] + (K − 1)t h ]). (B1)

The benefit in (B1) is equal to the cost ψ of exerting effort given the choice of T1 ∗ . Thus, the

incentive compatibility for workers employed by firms in groups i = 2, 3, . . . K is satisfied.

The incentive compatibility constraint for workers employed by firms in group 1 is satisfied

given the choice of s 1 .

C Proof of Theorem 4

The existence proof is based on an application of Schauder’s fixed point theorem. In particular,

we are going to take arbitrary value functions V + (u) and J + (u) denoting, respectively,

the worker’s gains from trade at the end of the production stage and the firm’s gains from

trade at the end of the production stage. Then, we are going to use the conditions for

a Recursive Equilibrium with perfect coordination to construct a mapping F that returns

updates for V + (u) and J + (u). Using Schauder’s fixed point theorem, we are going to prove

that the mapping F admits a fixed point and we are going to show that this fixed point is

a Recursive Equilibrium with perfect coordination in the stage game.

The functions V + (u) and J + (u) are chosen from a set Ω of Lipschitz continuous functions

with fixed Lipschitz bounds. These property of the set Ω is critical to establish that the

operator F satisfies the conditions for Schauder’s fixed point theorem (namely, that the

family of functions F(⊗) is equicontinuous). Formally, let Ω denote the set of bounded and

continuous functions ω(u, i) = iV + (u) + (1 − i)J + (u), with ω : [0, 1] × {0, 1} → R, and such

that: (i) for all u 0 , u 1 ∈ [0, 1] with u 0 < u 1 , V + (u 1 )−V + (u 0 ) is greater than D V+

(u 1 −u 0 ) and

smaller than D V+ (u 1 − u 0 ); (ii) for all u 0 , u 1 ∈ [0, 1] with u 0 < u 1 , J + (u 1 ) − J + (u 0 ) is greater

than D J+

(u 1 − u 0 ) and smaller than D J+ (u 1 − u 0 ). In other words, Ω is the set of bounded

and Lipschitz continuous functions ω = (V + , J + ) with fixed Lipschitz bounds D V+

, D V+ ,

D J+

and D J+ . We choose the Lipschitz bounds to satisfy D V+ > D V+

> 0, D J+ > 0 ≥ D J+

,

and D V+

> υ ′ (D J+ − D J+

).

Starting from V + (u) and J + (u) in Ω, we use the equilibrium conditions to construct the

equilibrium wage, the equilibrium gains from trade accruing to the worker and the firm, the

44

equilibrium law of motion for unemployment and an update for V + (u) and J + (u). This

process implicitely defines the operator F. In order to make sure that F maps functions in

Ω into functions in Ω (which is a condition of Schauder’s fixed point theorem), we need to

verify that all equilibrium objects are Lipschitz continuous with fixed Lipschitz bounds. In

order to make sure that F is continuous (which is also a condition of Schauder’s fixed point

theorem), we need to verify that all equilibrium objects are continuous. To carry out these

tasks, we need a few more pieces of notation. In particular, we use V , V , J and J to denote

lower and upper bounds on the worker’s and firm’s gains from trade constructed as in (??).

Also, we use µ u and µ u to denote the minimum and the maximum of the (absolute value) of

the partial derivative of the job-finding probability µ(J, u) with respect to i = u. That is,

µ u denotes min |∂µ(J, u)/∂u| for (J, u) ∈ [J, J] × [0, 1], and µ u denotes max |∂µ(J, u)/∂u| for

(J, u) ∈ [J, J]×[0, 1]. Similarly, we use µ J and µ J to denote the minimum and the maximum

on the (absolute value) of the partial derivative of µ(J, u) with respect to J.

C.1 Wage

Take an arbitrary pair of value functions ω = (V + , J + ) ∈ Ω. For any u, the equilibrium

wage function w(u) takes a value w such that

υ(w) − υ(b) − ψ + V + (u) = υ ′ (w) E[y|e = 1] − w + J + (u) .

(C1)

For any u, there is a unique wage that satisfies (C1). In fact, the right-hand side of (C1) is

strictly increasing in w, as υ(w) − υ(b) − ψ + V + (u) is strictly increasing in w. The left-hand

side of (C1) strictly decreasing in w, as υ ′ (w) and E[y|e = 1] − w + J + (u) are both strictly

decreasing in w. Therefore, there exists a unique wage w that solves (C1) for any u.

The next lemma proves that the equilibrium wage function w(u) is Lipschitz continuous

in u with fixed Lipschitz bounds. Moreover, the lemma proves that w(u) is strictly decreasing

in u.

Lemma C1: For all u 0 , u 1 ∈ [0, 1] with u 0 < u 1 , the wage function w(u) is such that

D w (u 1 − u 0 ) ≤ w(u 0 ) − w(u 1 ) ≤ D w (u 1 − u 0 ),

(C2)

where the bounds D w and D w are defined as

D w = D V +

/υ ′ − D J+

2 + Jυ ′′ /υ ′ , D w = D V+ /υ ′ − D J+

. (C3)

Proof : To alleviate notation, let w 0 denote w(u 0 ) and w 1 denote w(u 1 ). First, we establish

45

that w 0 > w 1 . To this aim, notice that

υ ′ (w 0 ) [E[y|e = 1] − w 0 + J + (u 1 )]

≤ υ ′ (w 0 )E[y|e = 1] − w 0 + J + (u 0 )] + υ ′ D J+ (u 1 − u 0 ).

(C4)

Also, notice that

υ(w 0 ) − υ(b) − c + V + (u 1 )

≥ υ(w 0 ) − υ(b) − c + V + (u 0 ) + D V + (u 1 − u 0 )

= υ ′ (w 0 )E[y|e = 1] − w 0 + J + (u 0 ) + D V + (u 1 − u 0 )

> υ ′ (w 0 )E[y|e = 1] − w 0 + J + (u 0 ) + υ ′ D J+ (u 1 − u 0 ),

(C5)

where the third line follows from (C1) and the last line follows from D V+

> υ ′ D J+ . Taken

together (C4) and (C5) imply that υ(w 0 ) − υ(b) − c + V + (u 1 ) is strictly greater than

υ ′ (w 0 ) [E[y|e = 1] − w 0 + J + (u 1 )]. Therefore, the left-hand side of (C1) is strictly smaller

than the right-hand side of (C1) when evaluated at w = w 0 and u = u 1 . Since the left-hand

of (C1) is strictly increasing in w and the left-hand side is strictly decreasing in w, w 1 is

strictly smaller than w 0 .

Next, we derive lower and upper bounds on w 0 − w 1 . From (C1), it follows that

υ(w 0 ) − υ(w 1 ) + V + (u 0 ) − V + (u 1 )

= υ ′ (w 0 ) [E[y|e = 1] − w 0 + J + (u 0 )] − υ ′ (w 1 ) [E[y|e = 1] − w 0 + J + (u 0 )]

+ υ ′ (w 1 ) [E[y|e = 1] − w 0 + J + (u 0 )] − υ ′ (w 1 ) [E[y|e = 1] − w 1 + J + (u 1 )]

(C6)

The above equation can be rewritten as

=

−

w 0 − w 1

V + (u 1 ) − V + (u 0 )

− [J + (u

υ ′ 1 ) − J + (u 0 )]

(w 1 )

υ(w0 ) − υ(w 1 ) υ ′ (w 1 ) − υ ′ (w 0 )

−

(E[y|e = 1] − w

υ ′ (w 1 )

υ ′ 0 + J + (u 0 )) .

(w 1 )

(C7)

The first term in square brackets on the right-hand side of (C6) is greater than (D V + /υ ′ )(u 1 −

u 0 ) and smaller than (D V + /υ ′ )(u 1 − u 0 ). The second term in square brackets is greater than

D J+ (u 1 − u 0 ) and smaller than D J+ (u 1 − u 0 ). The third term in square brackets is greater

than zero and smaller than w 1 − w 0 . The last term on the right-hand side of (C6) is greater

than zero and smaller than υ ′′ D w J/υ ′ (w 1 −w 0 ). From the above observations, it follows that

w 0 − w 1 ≤

DV +

υ ′

− D J+

(u 1 − u 0 ). (C8)

46

Similarly, we have

w 0 − w 1 ≥

−1

2 + υ′′

υ D DV +

wJ

′ υ ′ − D J+ (u 1 − u 0 ). (C9)

The inequalities (C8) and (C9) represent the desired bounds on w 0 − w 1 .

The next lemma proves that the equilibrium wage function w is continuous with respect

to the value functions V + and J + . Specifically, consider ω 0 = (V +

0 , J + 0 ) and ω 1 = (V +

1 , J + 1 )

with ω 0 , ω 1 ∈ Ω. Denote as w i the wage function computed using ω i in (C1) for i = {0, 1}.

If the distance between ω 0 and ω 1 goes to 0, so does the distance between w 0 and w 1 .

Lemma C2: For any κ > 0 and any ω 0 , ω 1 in Ω such that ||ω 0 - ω 1 || < κ, we have

||w 0 − w 1 || < α w κ, α w = 1 + 1/υ ′ . (C10)

Proof : Take an arbitrary u ∈ [0, 1]. To alleviate notation, let w 0 denote w 0 (u) and w 1 denote

w 1 (u). From (C10), it follows that

υ(w 0 ) − υ(b) − c + V +

0 (u) − υ ′ (w 0 ) E[y|e = 1] − w 0 + J + 0 (u) = 0,

υ(w 1 ) − υ(b) − c + V +

1 (u) − υ ′ (w 1 ) E[y|e = 1] − w 1 + J + 1 (u) = 0.

Subtracting the second equation from the first, we obtain

V +

0 (u) − V +

1 (u) + υ(w 0 ) − υ(w 1 )

+ υ ′ (w 0 ) J + 1 (u) − J + 0 (u) + w 0 − w 1

+ [υ ′ (w 1 ) − υ ′ (w 0 )] E[y|e = 1] − w 1 + J + 1 (u) = 0.

(C11)

(C12)

Suppose without loss in generality that w 0 ≥ w 1 . We can rewrite the above equation as

w 0 − w 1 = J 0 + (u) − J 1 + (u) V

+

+ 1 (u) − V 0 +

(u) υ(w1 ) − υ(w 0 )

+

υ ′ (w 0 )

υ ′ (w 0 )

+ E[y|e = 1] − w 1 + J 1 + (u)

(C13)

υ ′ (w 0 ) − υ ′ (w 1 )

.

υ ′ (w 0 )

The first term in square brackets on the right-hand side of (C13) is strictly smaller than κ.

The second term in square brackets is strictly smaller than κ/υ ′ . The third term in square

brackets is strictly negative. The last term is strictly negative. Hence, we have

0 ≤ w 0 − w 1 < (1 + 1/υ ′ )κ. (C14)

Since the above inequality holds for any u ∈ [0, 1], we conclude that ‖w 0 − w 1 ‖ < α w κ where

α w = (1 + 1/υ ′ ).

47

C.2 Gains from trade to worker and firm

Given V + , J + and w, the equilibrium gains from trade accruing to the firm and to the

worker, J and V , are respectively given by

J(u) = E[y|e = 1] − w(u) + J + (u),

V (u) = υ(w(u)) − υ(b) − ψ + V + (u)

(C15)

The next lemma proves that the equilibrium gains from trade J(u) and V (u) are Lipschitz

continuous in u with fixed Lipschitz bounds. Moreover, the lemma proves that J(u) and V (u)

are strictly increasing in u.

Lemma C3: For all u 0 , u 1 ∈ [0, 1] with u 0 < u 1 , the equilibrium gains from trade accruing

to the firm, J(u), and to the worker, V (u), are such that

D J (u 1 − u 0 ) ≤ J(u 1 ) − J(u 0 ) ≤ D J (u 1 − u 0 ),

D V (u 1 − u 0 ) ≤ V (u 1 ) − V (u 0 ) ≤ D V (u 1 − u 0 ),

(C16)

where the bounds are defined as

D J = D w + D J + > 0, D J = D w + D J +,

D V = υ ′ (D w + D J +) > 0, D V = υ ′ (D w + D J +) + υ ′′ D w J.

(C17)

Proof : To simplify notation, let w 0 denote w(u 0 ) and w 1 denote w(u 1 ). Then J(u 1 ) − J(u 0 )

is given by

J(u 1 ) − J(u 0 ) = w 0 − w 1 + J + (u 1 ) − J + (u 0 ).

From the above expression, it follows that

J(u 1 ) − J(u 0 ) ≥ (D w + D J+ )(u 1 − u 0 ),

J(u 1 ) − J(u 0 ) ≤ (D w + D J+ )(u 1 − u 0 ).

(C18)

The difference V (u 1 ) − V (u 0 ) is given by

V (u 1 ) − V (u 0 ) = υ ′ (w 1 )J(u 1 ) − υ ′ (w 0 )J(u 0 )

= [υ ′ (w 1 )J(u 1 ) − υ ′ (w 1 )J(u 0 )] + [υ ′ (w 1 )J(u 0 ) − υ ′ (w 0 )J(u 0 )]

From the above expression, it follows that

V (u 1 ) − V (u 0 ) ≥ υ ′ (D w + D J+ )(u 1 − u 0 ),

V (u 1 ) − V (u 0 ) ≤ υ ′ (D w + D J+ ) + υ ′′ D w J (u 1 − u 0 ).

(C19)

Lemma 7 follows directly from the inequalities in (C18) and (C19).

48

The next lemma proves that the firm’s and worker’s gains from trade are continuous with

respect to V + and J + . Specifically, consider ω 0 = (V +

0 , J + 0 ) and ω 1 = (V +

1 , J + 1 ) with ω 0 ,

ω 1 in Ω. For i ∈ {0, 1}, denote as w i the equilibrium wage computed using ω i , and as J i

and V i the equilibrium gains from trade computed using ω i and w i in (C15). If the distance

between ω 0 and ω 1 goes to 0, so does the distance between J 0 and J 1 and between V 0 and

V 1 .

Lemma C4: For any κ > 0 and any ω 0 , ω 1 in Ω such that ||ω 0 - ω 1 || < κ, we have

||J 0 − J 1 || < α J κ, α J = 1 + α w ,

||V 0 − V 1 || < α V κ, α V = 1 + υ ′ α w . . (C20)

Proof : Take an arbitrary u ∈ [0, 1]. The difference J 0 (u) − J 1 (u) is such that

|J 0 (u) − J 1 (u)| ≤ |w 0 (u) − w 1 (u)| + J

+

0 (u) − J + 1 (u)

< (α w + 1)κ.

(C21)

The difference V 0 (u) − V 1 (u) is such that

|V 0 (u) − V 1 (u)| ≤ |υ(w 0 (u)) − υ(w 1 (u))| + V

+

0 (u) − V +

1 (u)

< (υ ′ α w + 1)κ.

(C22)

Since the inequalities (C21) and (C22) hold for all u ∈ [0, 1], we conclude that ‖J 0 − J 1 ‖ <

α J ′κ and ‖V 0 − V 1 ‖ < α V ′κ, where α J = α w + 1 and α V = υ ′ α w + 1.

C.3 Law of motion for unemployment

Given J, the law of motion for unemployment h(u, ẑ), is such that– for any current period’s

unemployment u and any realization of the sunspot ẑ– next period’s unemployment takes

on a value û such that

û = u − uµ(J(û), u) + (1 − u)[δ + (1 − δ)p l (1)s(y l , ẑ)], (28)

Since we are trying to construct a Recursive Equilibrium in which there is perfect coordination

in the stage game, we set s(y l , ẑ) = 0 if ẑ ∈ Z 0 and s(y l , ẑ) = 1 if ẑ ∈ Z 2 .

Next period’s unemployment û is uniquely determined by (28). In fact, the left-hand

side of (28) equals zero for û = 0, it is strictly increasing in û and it equals one for û = 1.

The right-hand side of (28) is strictly positive for û = 0 and it is strictly decreasing in û,

as the worker’s job finding probability µ is strictly increasing in the firm’s gains from trade

J, and J is strictly increasing in û. Therefore, there exists one and only one û that satisfies

49

(28). Next period’s unemployment û is strictly increasing in u. In fact, the left-hand side

of (28) is independent of u. The right-hand side of (28) is strictly increasing in u, as its

derivative with respect to u is greater than (1 − δ)p h (1) − µ, which we assume to be strictly

positive. Therefore, the û that solves (28) is strictly increasing in u. Moreover, next period’s

unemployment û is strictly higher if ẑ ∈ Z 2 than if ẑ ∈ Z 0 . To see this, it is suffi cient to

notice that the left-hand side of (28) is independent of s(y l , ẑ), while the right hand side is

strictly increasing in it.

The next lemma proves that h(u, ẑ) is Lipschitz continuous in u with fixed Lipschitz

bounds.

Lemma C5: For all u 0 , u 1 ∈ [0, 1] with u 0 < u 1 , h(u, ẑ) is such that

D h (u 1 − u 0 ) < h(u 1 , ẑ) − h(u 0 , ẑ) ≤ D h (u 1 − u 0 ),

(C24)

where the bounds D h and D h are defined as

D h = 0, D h = 1 − δ + µ u . (C25)

Proof : Let û 1 denote the solution to (28) for u = u 1 and let û 0 denote the solution to (28)

for u = u 0 , i.e.

û 1 = u 1 − u 1 µ(J(û 1 ), u 1 ) + (1 − u 1 )[δ + (1 − δ)p l (1)s(y l , ẑ ′ )],

û 0 = u 0 − u 0 µ(J(û 0 ), u 0 ) + (1 − u 0 )[δ + (1 − δ)p l (1)s(y l , ẑ ′ )].

Subtracting the second equation from the first one, we obtain

û 1 − û 0 = (u 1 − u 0 )(1 − δ)[1 − p l (1)s(y l , ẑ ′ )]

+u 0 [µ (J(û 0 ), u 0 ) − µ (J(û 1 ), u 0 )]

+u 0 [µ (J(û 1 ), u 0 ) − µ (J(û 1 ), u 1 )]

+u 0 µ (J(û 1 ), u 1 ) − u 1 µ (J(û 1 ), u 1 ) .

(C26)

The first line on the right-hand side of (C26) is positive. Since µ(J, u) is increasing in J,

J is increasing in u and, as established in the main text, û 1 > û 0 , the second line on the

right-hand side of (C26) is negative. Since µ(J, u) is decreasing in u and u 1 > u 0 , the third

line on the right-hand side of (C26) is positive. The fourth line on the right-hand side of

(C26) is obviously negative. Hence, an upper bound on û 1 − û 0 is given by

û 1 − û 0 ≤ (u 1 − u 0 )(1 − δ)[1 − p l (1)s(y l , ẑ ′ )] + u 0 [µ (J(û 1 ), u 0 ) − µ (J(û 1 ), u 1 )]

≤ (u 1 − u 0 ) [1 − δ + µ u ] .

(C27)

50

Combining the above inequality with û 1 < û 0 , we obtain

D g (u 1 − u 0 ) < û 1 − û 0 ≤ D g (u 1 − u 0 ) ,

(C28)

where

D g = 0, D g = 1 − δ + µ u .

Next, we prove that h(u, ẑ) is continuous with respect to V + and J + .

Formally, we

consider ω 0 = (V +

0 , J + 0 ) and ω 1 = (V +

1 , J + 1 ) with ω 0 , ω 1 in Ω. For i = {0, 1}, we denote

as J i the firm’s equilibrium gains from trade computed using ω i in (C15), and as h i the

equilibrium law of motion for unemployment computed using J i in (28). Then, we show

that, if the distance between ω 0 and ω 1 goes to 0, so does the distance between h 0 and h 1 .

Lemma C6: For any κ > 0 and any ω 0 , ω 1 in Ω such that ||ω 0 - ω 1 || < κ, we have

||h 0 − h 1 || < α h κ, α h = µ J α J . (C29)

Proof : Take an arbitrary u ∈ [0, 1] and ẑ ∈ {Z 0 , Z 2 }. Let û 0 denote h 0 (u, ẑ) and û 1 denote

h 1 (u, ẑ). From (28), it follows that

û 0 = u − uµ(J 0 (û 0 ), u) + (1 − u)[δ + (1 − δ)p l (1)s(y l , ẑ)],

û 1 = u − uµ(J 1 (û 1 ), u) + (1 − u)[δ + (1 − δ)p l (1)s(y l , ẑ)].

Without loss in generality suppose that û 0 ≥ û 1 . In this case,

û 0 − û 1 = u {[µ(J 1 (û 1 ), u) − µ(J 0 (û 1 ), u)] + [µ(J 0 (û 1 ), u) − µ(J 0 (û 0 ), u)]} .

(C30)

The term µ(J 0 (û 1 ), u) − µ(J 0 (û 0 ), u) on the right-hand side of (C30) is negative as µ(J, u) is

increasing in J, J 0 is increasing in u and û 0 ≥ û 1 . Therefore, we have

û 0 − û 1 ≤ u [µ(J 1 (û 1 ), u) − µ(J 0 (û 1 ), u)] < µ J α J κ.

(C31)

Since the above inequality holds for any u ∈ [0, 1] and any ẑ ′ ∈ {B, G}, we conclude that

‖h 0 − h 1 ‖ < α h κ, where α h = µ J α J .

C.4 Updated value functions

Given J, V and h, we can construct an update, V +′ , for the worker’s gains from trade at

the end of the production stage as

V +′ (u) = βEẑ {[(1 − δ)(1 − p l (1)s(y l , ẑ)) − µ(J(h(u, ẑ)), u)]V (h(u, ẑ))}

(C32)

51

Similarly, we can construct an update, J +′ , for the firm’s gains from trade at the end of the

production stage as

J +′ (u) = βEẑ [(1 − δ)(1 − p l (1)s(y l , ẑ))J(h(u, ẑ))]

(C33)

Since we are looking for a Recursive Equilibrium with perfect coordination in the stage game,

we set s(y l , ẑ) = 0 for ẑ ∈ Z 0 and s(y l , ẑ) = 1 for ẑ ∈ Z 2 . Also, we set the probability π 0

that ẑ ∈ Z 0 and the probability π 2 that ẑ ∈ Z 2 so that π 0 = 1 − π 2 and

π 2 =

ψ

β(p h (1) − p h (0))(1 − δ)V (h(u, Z 2 )) .

(C34)

In the next lemma, we prove that V +′ (u) and J +′ (u) are Lipschitz continuous in u with

fixed Lipschitz bounds.

Lemma C7: For all u 0 , u 1 ∈ [0, 1] with u 0 < u 1 , V +′ and J +′ are such that

D ′ V +(u 1 − u 0 ) < V +′ (u 1 ) − V +′ (u 0 ) ≤ D ′ V +(u 1 − u 0 ),

D ′ J+(u 1 − u 0 ) < J +′ (u 1 ) − J +′ (u 0 ) ≤ D ′ J+(u 1 − u 0 ).

where the bounds D ′ V + and D ′ V + are defined as

(C35)

D ′ V + = βV (µ u

− µ J D J D h ) −

D ′ V + = β V µ u + (1 − δ)D V D h

,

2ψD V D h V

(p h (1) − p h (0))(1 − δ)V 2 ,

(C36)

and the bounds D ′ J+ and D ′ J+ are defined as

D ′ 2cD V D h J

J+ = −

(p h (1) − p h (0))(1 − δ)V 2 ,

(C37)

D ′ J+ = β(1 − δ)D J D h

Proof : Take arbitrary u 0 , u 1 ∈ [0, 1] with u 1 > u 0 . To simplify notation, let û j,0 denote

h(u 0 , ẑ) for ẑ ∈ Z j and let û j,1 denote h(u 1 , ẑ) for ẑ ∈ Z j . Similarly, let π j,0 denote the

probability that the realization of the sunspot is ẑ ∈ Z j for u 0 , and let π j,1 denote the

probability that the realization of the sunspot is ẑ ∈ Z j for u 1 . Using this notation and

(C32), we can write V +′ (u 1 ) − V +′ (u 0 ) as

V +′ (u 1 ) − V +′ (u 0 )

= β P j {(π j,1 − π j,0 ) [(1 − δ)(1 − p l (1)s(y l , ẑ)) − µ(J(û j,1 , u 1 )]V (û j,1 )

+ π j,0 [(1 − δ)(1 − p l (1)s(y l , ẑ)) − µ(J(û 2,1 , u 1 )] [V (û j,1 ) − V (û j,0 )] (C38)

+ π j,0 V (û j,0 )[µ(J(û 2,0 , u 0 ) − µ(J(û 2,0 , u 1 )].

+ π j,0 V (û j,0 )[µ(J(û 2,0 , u 1 ) − µ(J(û 2,1 , u 1 )]}.

52

The first term on the right-hand side of (C38) is negative. In absolute value, this term is

smaller than cD V D h V (u 1 − u 0 )/[β(p h (1) − p h (0))(1 − δ)V 2 ]. The second term on the righthand

side of (C38) is geater than 0 and smaller than π j,0 (1 − δ)D V D h (u 1 − u 0 ). The third

term on the right-hand side of (C38) is positive, greater than π j,0 V µ u (u 1 − u 0 ) and smaller

than π j,0 V µ u (u 1 −u 0 ). The last term on the right-hand side of (C38) is negative. In absolute

value, this term is smaller than π j,0 V µ J D J D h (u 1 − u 0 ). Overall, we have

V +′ (u 1 ) − V +′ (u 0 ) ≥

2cD V D h V

βV (µ u

− µ J D J D h ) −

(p h (1) − p h (0))(1 − δ)V 2 (u 1 − u 0 ), (C39)

and

V +′ (u 1 ) − V +′ (u 0 ) ≤ β V µ u + (1 − δ)D V D h

(u1 − u 0 ).

(C40)

Using (C33), we can write J +′ (u 1 ) − J +′ (u 0 ) as

J +′ (u 1 ) − J +′ (u 0 )

= β P j {(π j,1 − π j,0 ) [(1 − δ)(1 − p l (1)s(y l , ẑ))]J(û j,1 )

+ π j,0 [(1 − δ)(1 − p l (1)s(y l , ẑ))] [J(û j,1 ) − J(û j,0 )] .

(C41)

The first term on the right-hand side of (C41) is negative. In absolute value, this term is

smaller than ψD V D h J(u 1 − u 0 )/[β(p h (1) − p h (0))(1 − δ)V 2 ]. The second term on the righthand

side of (C41) is geater than zero and smaller than π j,0 (1 − δ)D J D h (u 1 − u 0 ). Overall,

we have

J +′ (u 1 ) − J +′ 2ψD V D h J

(u 0 ) ≥ −

(p h (1) − p h (0))(1 − δ)V 2 (u 1 − u 0 ), (C42)

and

J +′ (u 1 ) − J +′ (u 0 ) ≤ β(1 − δ)D J D h (u 1 − u 0 ).

(C43)

Lemma C7 follows directly from the inequalities in (C39)-(C40) and (C42)-(C43).

The next lemma shows that, under some parametric conditions, there exists a fixed point

for the Lipschitz bounds on the firm’s and worker’s gains from trade at the end of the

production stage.

Lemma C8: Assume µ u − µ J (1 − δ + µ u ) > 0. Then there exist β ∗ > 0 and ψ ∗ > 0 such

that, if β ∈ (0, β ∗ ) and ψ ∈ (0, ψ ∗ ), there are Lipschitz bounds D V + , D V + , D J+ , D J+ such

that: (i) D ′ V + = D V + , D ′ V + = D V + , D ′ J+ = D J+ , and D ′ J+ = D J+ ; (ii) D V+ > D V+

> 0,

D J+ > 0 ≥ D J+

and D V+

> υ ′ (D J+ − D J+

).

Proof : Set D ′ V + and D ′ V + in (C36) equal to D V + and D V + , and D ′ J+ and D ′ J+ in (C37) equal

to D J+ and D J+ . Then solve for D V + , D V + , D J+ and D J+ . It is immediate to verify that,

53

since µ u − µ J (1 − δ + µ u ) > 0, D V+ > D V+

> 0, D J+ > 0 ≥ D J+

and D V+

> υ ′ (D J+ − D J+

)

for β and ψ small enough.

In the last lemma, we prove that V +′ and J +′ are continuous with respect to V + and J + .

Specifically, consider ω 0 = (V 0 + , J 0 + ) and ω 1 = (V 1 + , J 1 + ) with ω 0 , ω 1 ∈ Ω. For i ∈ {0, 1},

denote as V +′

i and J +′

i the continuation value functions computed using V +

i , J + i and h i in

(C32) and (C33). If the distance between ω 0 and ω 1 goes to 0, so does the distance between

V +

0 and V +

1 and between J + 0 and J + 1 .

Lemma C9: For any κ > 0 and any ω 0 , ω 1 ∈ Ω such that ||ω 0 - ω 1 || < κ, we have

||V 0 +′ − V 1 +′ || < α V ′

+

κ and ||J 0 +′ − J 1 +′ || < α J ′

+

κ, where

2ψV

α V ′

+

=

(p h (1) − p h (0))(1 − δ)V 2 + 1

α V + D V α h + V µJ α J + D J α h ,

(C44)

2ψJ α V + D V α h

α J ′

+

=

(p h (1) − p h (0))(1 − δ)V 2 + α

J + D J α h .

Proof : Take an arbitrary u ∈ [0, 1]. To simplify notation, let û j,0 denote h 0 (u, ẑ) for ẑ ∈ Z j ,

and let û j,1 denote h 1 (u, ẑ) for ẑ ∈ Z j . Similarly, let π j,0 denote the probability that the

realization of the sunspot is ẑ ∈ Z j defined by using V 0 and h 0 in (C34), and let π j,1 denote

the probability that the realization of the sunspot is ẑ ∈ Z j defined by using V 1 and h 1 in

(C34). Using this notation and (C32), we can write V +′

0 (u) − V +′

1 (u) as

V +′

0 (u) − V +′

1 (u)

= β P j {(π j,0 − π j,1 ) [(1 − δ)(1 − p l (1)s(y l , ẑ)) − µ(J 0 (û j,0 , u)]V 0 (û j,0 )

+ π j,1 [(1 − δ)(1 − p l (1)s(y l , ẑ)) − µ(J 0 (û j,0 , u)] [V 0 (û j,0 ) − V 1 (û j,1 )]

+ π j,1 V 1 (û j,1 )[µ(J 1 (û j,1 , u) − µ(J 0 (û j,0 , u)]}.

(C45)

From (C45), it follows that

V

+′

0 (u) − V 1 + (u) ≤

2cV

(p h (1) − p h (0))(1 − δ)V 2 α V + D V α h

κ + αV + D V α h

κ + V µJ α J + D J α h

κ,

(C46)

where the first term on the right-hand side of (C46) is an upper bound on the absolute value

on the first line on the right-hand side of (C45), the second term on the right-hand side of

(C46) is an upper bound on the absolute value on the second line on the right-hand side of

(C45), and the last term on the right-hand side of (C46) is an upper bound on the absolute

value on the last line on the right-hand side of (C45).

54

Using (C33) we can write J +′

0 (u) − J +′

1 (u) as

J +′

0 (u) − J +′

1 (u)

= β P j {(π j,0 − π j,1 ) (1 − δ)(1 − p l (1)s(y l , ẑ))J 0 (û j,0 )

+ π j,1 (1 − δ)(1 − p l (1)s(y l , ẑ)) [J 0 (û j,0 ) − J 1 (û j,1 )] .

(C47)

From (C47), it follows that

J

+′

0 (u) − J 1 +′ (u) ≤

2ψJ

(C48)

(p h (1) − p h (0))(1 − δ)V 2 α V + D V α h κ + αJ + D J α h κ,

where the first term on the right-hand side of (C48) is an upper bound on the absolute value

on the first line on the right-hand side of (C47), and the second term on the right-hand side

of (C48) is an upper bound on the absolute value on the second line on the right-hand side

of (C47). Since the inequalities (C46) and (C48) hold for any u ∈ [0, 1], we conclude that

V

+′

0 − V 1

+′ < αV + κ and J

+′

0 − J +′ < αJ+ κ.

1

C.5 Existence

In the previous subsections, we have taken a pair of continuation gains from trade V + and

J + and, using the conditions for a perfect coordination equilibrium, we have constructed an

updated pair of continuation gains from trade V +′ and J +′ . We denote as F the operator

that takes ω(u, i) = (1 − i)V + (u) + iJ + (u) and returns ω ′ (u, i) = (1 − i)V +′ (u) + iJ +′ (u).

The operator F has three key properties. First, the operator F maps functions that

belong to the set Ω into functions that also belong to the set Ω. In fact, for any ω =

(V + , J + ) ∈ Ω, ω ′ = (V +′ , J +′ ) is bounded and continuous and, as established in Lemma

C8, it is such that: (i) for all u 0 , u 1 ∈ [0, 1] with u 0 < u 1 , the difference V +′ (u 1 ) − V +′ (u 0 )

is greater than D V+

(u 1 − u 0 ) and smaller than D V+ (u 1 − u 0 ); (ii) for all u 0 , u 1 ∈ [0, 1]

with u 0 < u 1 , the difference J +′ (u 1 ) − J +′ (u 0 ) is greater than D J+

(u 1 − u 0 ) and smaller

than D J+ (u 1 − u 0 ). Second, the operator F is continuous, as established in Lemma C9.

Third, the family of functions F(Ω) is equicontinuous. To see that this is the case, let

||.|| E denote the standard norm on the Euclidean space [0, 1] × {0, 1}. For any ɛ > 0, let

κ ɛ = min{(max{D V + , D J+ , |D J+ |}) −1 , 1}. Then, for all (u 0 , i 0 ), (u 1 , i 1 ) ∈ [0, 1] × {0, 1} such

that ||(u 0 , i 0 ) − (u 1 , i 1 )|| E < κ ɛ , we have

|(Fω)(u 0 , i 0 ) − (Fω)(u 1 , i 1 )| < ɛ, for all ω ∈ Ω. (29)

55

Since F : Ω −→ Ω, F is continuous and F(Ω) is equicontinuous, the operator F satisfies the

conditions of Schauder’s fixed point theorem. Therefore there exists a ω ∗ = (V +∗ , J +∗ ) ∈ F

such that Fω ∗ = ω ∗ .

Now, we compute the equilibrium wage function w ∗ using V +∗ and J +∗ . We compute

the equilibrium gains from trade accruing to the worker, V ∗ , and to the firm, J ∗ , using w ∗ ,

V +∗ and J +∗ . We compute the equilibrium law of motion for unemployment h ∗ using J ∗ .

Finally, we construct the equilibrium employment contract x ∗ (u) as the effort e ∗ = 1, the

wage function w ∗ and the firing probabilities s ∗ (y h , ẑ) = 0, s ∗ (y l , ẑ) = 0 for ẑ ∈ Z 0 and

s ∗ (y h , ẑ) = 1 for ẑ ∈ Z 2 . Since the wage function w ∗ is strictly decreasing in u and h ∗ (u, Z 2 )

is strictly greater than h ∗ (u, Z 0 ), these objects constitute a Recursive Equilibrium in which

there is perfect coordination at the stage game.

D

Additional Figures

Figure D1: Model Labor Market

50

40

30

20

10

0

-10

-20

u UE EU

56

Figure D2: US Labor Market

60

50

40

30

20

10

0

-10

-20

-30

-40

-50

1992 1995 1998 2001 2004 2007 2010 2013

u UE EU

Figure D3: Recession Probability

probability recession

0.45

0.4

0.35

0.3

0.25

0.2

0.15

0.1

0.05

USA

POOL

AUS

CAN

FR

IT

JAP

UK

0

4 5 6 7 8 9

unemployment

57