Labour Economics 15 (2008) 1416 1434

www.elsevier.com/locate/econbase

Migration and the wage and unemployment gaps

between urban and non-urban sectors: A dynamic
general equilibrium reinterpretation of the
HarrisTodaro equilibrium
Chul-In Lee
53, 3-ga, Myongnyun-dong, Chongno-gu, School of Economics,
SungKyunKwan Univ., Seoul, 110-745, Republic of Korea
University of Michigan, United States
Received 2 November 2006; received in revised form 3 December 2007; accepted 23 January 2008
Available online 8 February 2008

Abstract
This paper offers a dynamic general equilibrium reinterpretation of the static partial migration equilibrium
by Harris and Todaro [Harris, J., Todaro, M., 1970. Migration, unemployment and development; a two-sector
analysis. American Economic Review 60, 126142], under (i) flexible urban and rural wages and (ii) free
mobility of workers and free entry of firms. The proposed model accounts for the set of stylized facts in
developing countries: rural to urban migration and higher urban wages and unemployment.
The model allows us to view the wage gap as a compensating differential for the negative amenities
associated with job destruction and subsequent costly search on the consumption side, which can also be
seen as a match-specific premium based on a sectoral productivity differential on the production side. Our
model predicts the comovements among urban and non-urban wages and migration flows to the urban
sector, an empirical regularity observed over the urbanization process of developing economies. Finally, we
also conduct a welfare analysis.
2008 Elsevier B.V. All rights reserved.
JEL Classification Codes: J31; J61; J64
Keywords: Migration; Urban vs. non-urban wage and unemployment gaps; Matching; Equilibrium unemployment;
Compensating wage differential; Welfare

Tel.: +82 2 760 0421.

E-mail address: leeci@skku.edu.
0927-5371/$ - see front matter 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.labeco.2008.01.003

C.-I. Lee / Labour Economics 15 (2008) 14161434

1417

1. Introduction
Three salient stylized facts that characterize the economic development process are: (i) workers
migrate from the rural to the urban sector, (ii) wages in the urban sector exceed those in the rural sector,
and (iii) urban unemployment exceeds that in the rural sector.1 While these phenomena are common
across developing economies, it has been difficult to obtain a cogent theoretical explanation. The early
seminal papers by Harris and Todaro (1970) (henceforth, HT) and Todaro (1969) took some important
strides toward providing an explanation. In its simplest form, HT's model postulates that migration
proceeds in response to differences in urban and rural expected wages, and the HT equilibrium holds
when the expected wages are equalized.2 Nevertheless, a caveat to their approach is that migration is
considered in a partial equilibrium setting with urban wages given exogenously. Moreover, the
equilibrium is defined based on ex ante expected value calculations. But given the differences in jobfinding and -destruction rates in the usual job-matching process entailing costly and time-consuming
search, such ex ante HT migration decisions are not supported by the ex post decisions in a dynamic
general equilibrium setting.3 This point therefore implies that migratory flows in the traditional HT
models are not an ongoing equilibrium phenomenon.
While many important studies have extended or improved the HT migration equilibrium (see the
literature survey in footnote 4),4 what separates this study from previous studies lies in the following
three points. First, this paper builds a dynamic general equilibrium model of migratory flows that is
consistent with the set of stylized facts. The proposed model provides a coherent framework that permits:
(i) both urban and rural wages to be fully flexible, (ii) both labor mobility and firm entry to be free across
1
A substantial wage gap between urban and non-urban sectors has been widely observed throughout the world. Unskilled full-time
nominal city wages are about 41% higher than farm wages in contemporary developing countries; a similar gap of about 51% can also
be found among late nineteenth-century industrializers (Hatton and Williamson, 1992). In Lucas's (1985) study of migration in
Botswana, he found that unadjusted urban earnings are 68% greater than rural earnings for males, and this wage gap generally
becomes smaller when schooling and experience are controlled for. Overall, he shows evidence consistent with that of Todaro and
HT. The wage gap still remains in the modern developed economy with a much smaller gap, e.g., around 30% currently in the U.S.
2
The resulting wage gap in the HT model has long been invoked to explain the migration flows and urban
unemployment of developing countries: typical developing countries exhibit large and sustained migration flows from
rural areas to cities, while urban unemployment rates are high. In their model, the wage gap arising from institutional
reasons (e.g., minimum wage laws) generates high urban unemployment. Unemployment can thus be interpreted as a key
pricing instrument allocating scarce resources in an economy in a partial static equilibrium context.
3
The ex ante migration decision in HT is based on the expected sectoral wage equalization, but once an urban job searcher from the rural
sector finds an urban job ex post, his or her next-period set of choices becomes very different from that of a comparable searcher who ends
up being unemployed in the dynamic setting. Looked at differently, the payoffs to workers are dependent on past job search outcomes. So
as we will see later, the expected sectoral wage equalization does not hold in the dynamic general equilibrium context.
4
Many studies have also attempted to extend or improve the original HT migration equilibrium by endogenizing urban wages,
which are assumed to be fixed at a level higher than non-urban wages. The first type of these models uses the efficiency wage
approach under the assumption that the urban sector gives efficiency wages for incentive reasons (Stiglitz, 1974, 1976; Moene,
1988; Zenou and Smith, 1995; Brueckner and Zenou, 1999; Brueckner and Kim, 2001). The second approach endogenizes urban
wages through a bargaining between a trade union and firms (Calvo, 1978; Hazari and Sgro, 1987; Quibria, 1988). The third
approach applies search-matching models with migration (Coulson et al., 2001; Ortega, 2000; Laing et al., 2005; Sato, 2004).
Earlier papers have introduced search frictions into an HT model with only one side of the market (the workers) (e.g., Fields, 1975,
1989; Banerjee, 1984; Mohtadi, 1989; Vishwanath, 1991). One study also explicitly takes into account the elasticity of urban
demand for labor (Stark et al., 1991). While the first two approaches are certainly appealing, this paper belongs to the third group,
since it views migration as a forward-looking behavior that involves unemployment due to search frictions that cause externalities.
Among a few studies related to ours, Coulson et al. (2001) build a search equilibrium model for a city with central and suburban
labor markets and explain the nature of a spatial mismatch problem. Ortega (2000) proposes a search model with two-country
international migration and shows that multiple equilibria arise. Laing et al. (2005) develop a search model for China to deal with
the effects of the household registration rule, taking into account the specificities of the Chinese labor market and institutions.
Finally, Sato (2004) proposes a static search-matching model and deals with welfare-improving policies in the ruralurban setting.

1418

C.-I. Lee / Labour Economics 15 (2008) 14161434

sectors and (iii) unmatched workers and firms to conduct costly and time-consuming search, such that it
still keeps the original motivation of HT as closely as possible. In such an environment, it is possible to
address questions concerning migration flows and the evolution of relative wages and relative
unemployment rates as development actively takes place, not to mention a more general welfare
analysis. In our model, migration occurs freely, even within a period, in response to changes in the
present value of income arising from sectoral wage changes or job-finding, job-arrival, or job-destruction
rates. In a general dynamic search-matching context, we can find the exact conditions for the HT
equilibrium. As we will see later, for the model to explain the wage and unemployment gaps across
sectors as an equilibrium phenomenon, the urban sector should exhibit the properties of (i) greater
heterogeneity and specificity of jobs, and (ii) higher productivity, compared with the non-urban sector.
Second, this paper presents a more structural interpretation of the wage gap. Rather than appealing
to an exogenous wage gap between the urban and rural sectors that is large enough to compensate
migrants for the risks of being unemployed (as in HT),5 the gap in this paper is endogenous and
consistent with the traditional labor economics literature (e.g., Roback, 1982). It can be interpreted as a
compensating differential for the negative amenities associated with job destruction and subsequent
costly search in a fully general equilibrium setting on the consumption side. On the production side, it
can also be seen as a match-specific premium based on a sectoral productivity differential. Given that
the urban wage is exogenously set above the rural wage in the HT model, there is only one pricing
instrument left that can allocate labor: the rate of urban unemployment. Thus, HT do not really tell us
that a higher urban wage is a compensating differential for unemployment risk; instead, unemployment
risk is a pricing scheme for scarce resources under conditions of congestion.6
Third, the proposed model predicts the comovements among urban and non-urban wages and migration
flows to the urban sector, an empirical regularity observed over the urbanization process of developing
economies. Along with a more structural interpretation of the wage gap, we examine the changes in rural
and urban wages and migration flows over the urbanization process, which previous studies did not explore.
Our view on the urban sector accords with that of Fujita and Thisse (1996), Behrman (1999), and Sato
(2004): compared with non-urban jobs, urban jobs are more heterogeneous and specific. A natural
implication of this point is that the urban sector faces a more costly and time-consuming job-matching
process under imperfect information.7 We argue that these differences in matching technology between
5
The urban wages in the HT models are assumed to be high and institutionally fixed due to minimum wage legislation or the
power of labor unions (e.g., Calvo, 1978; Quibria, 1988). This assumption is inevitable in a partial equilibrium model focusing on
the incipient urbanization stage. But mobility of workers can affect wages, and thus endogenizing both urban and non-urban wages
is necessary to understand the labor market equilibrium over time. Empirically, neither minimum wage laws nor unions are very
important forces in developing countries. In addition, the nature of urban unemployment needs to be considered more explicitly
instead of assuming that a certain level of unemployment is required to establish equilibrium without any structural reasons.
6
The author is grateful to an anonymous referee for suggesting this insightful interpretation of the HT equilibrium.
7
Our model uses the matching function that gives the outcome of firms' and workers' investment of resources in the trading
process of labor services in the urban sector (i.e., matched workerfirm pairs) as a function of the inputs (vacant jobs and job
searchers). With this modeling device, we can capture the implications of the costly trading process without the need to make the
heterogeneities and other features that give rise to them explicit. This modeling spirit can be found in macroeconomic aggregate
functions such as the production function and the demand for money function. In these functions, we do not make a physical
technology and a transaction technology combined with a portfolio choice explicit, respectively. See Pissarides (2000) for the
motivation of using the matching function. In light of this, we cannot eliminate the possibility that urban workers are more likely to
be high-skilled workers, on average, compared with non-urban workers. However, considering the mass migration in a developing
country that has just begun to reallocate the workforce from a large agricultural sector to a small, modern manufacturing sector, we
may view differences in skill level among rural workers as insubstantial, although heterogeneities may be crucial. Then the
remaining issue is information problems: firms are not sure whether workers can fit their needs; at the same time, workers do not
know what kinds of vacancies are available and where, etc. This is a sort of two-sided uncertainty, which takes resources to make
matching between firms and workers possible. In the non-urban labor market, however, this issue can be said to be less crucial.

C.-I. Lee / Labour Economics 15 (2008) 14161434

1419

the sectors, when combined with a productivity differential, permit a steady-state equilibrium
in which the wage gap is accounted for by a match-specific premium based on a sectoral
productivity differential and the unemployment gap by urban job-search frictions (e.g., a timeand resource-consuming process). Our two-sector (urban and non-urban) job-matching model
is a natural integration of the original HT model and the classic one-sector macroeconomic
search-matching models of Diamond (1984) and Mortensen and Pissarides (1994). As in most
matching models, unemployment occurs in the costly and time-consuming trade of labor
services when workers and firms are heterogeneous and information about them is imperfect.
As a result, we not only endogenize variables such as the number of displaced workers,
vacancies, and job searchers, and the unemployment rate, sectoral wages, and the non-urban
population, but we also allow for search externalities arising from congestion in job search and
worker recruitment.
The remainder of the paper consists of the following sections. Section 2 describes the basic
structure of our model. Section 3 presents the steady-state equilibrium, key results, and an
interpretation of them. We show that our model can account for widely observed comovement
among sectoral wages and the urbanization rate. Then we conduct a welfare analysis to
characterize the efficiency of the equilibrium with some new policy implications. A short
summary and conclusions are given in Section 4.
2. Basic structure of the model
2.1. A dynamic location choice model with free mobility and entry
2.1.1. Individuals' decisions
The labor force with its size normalized at unity chooses its location in order to maximize the
present value of its future income stream under the economic environment to be explained below.
To discuss its location choice and the conditions for inter-regional labor market equilibrium, we
first define three different value functions.
The present value of urban employment at t, W(t), can be expressed using the Bellman
equation as follows. To simplify notations and save space, we henceforth denote the current
period's variable x(t) simply as x and the next period's variable x(t + 1) by x in the Bellman
equations.
1 rW w f1  k  max W V; X V k  X Vg;

where w is the flow urban wage received at the end of period t (henceforth, all flow variables are
measured at the end of period), r is the discount factor, X = max[U,R], and U and R are the
values of urban unemployment and non-urban employment, respectively, to be defined below.
Eq. (1) implies that (1 + r)W equals a flow wage rate w plus the value of the state-contingent
choices in the next period: a weighted average of (i) the maximum value among W, R and U
conditional on the current employment continuing with probability 1 and (ii) the
greater value between R and U conditional on the current employment discontinuing with
probability .
Similarly, we can construct the value function for an urban unemployed worker searching for
an urban job, U(t): it is equal to the flow unemployment income b(t) plus the weighted average
of (i) the maximum value conditional on a job offer arriving with probability q() that is

1420

C.-I. Lee / Labour Economics 15 (2008) 14161434

derived from the CRS matching technology8 and (ii) the maximum value conditional on a job
offer not arriving with probability 1 q():
1 rU b fh VqhV max W V; X V 1  h Vqh VX Vg:

For the sake of simplicity, we will normalize b at zero. Meanwhile, we assume without loss of
generality that rural workers must move to the city if they want to find an urban job. This
assumption accords well with the original intuition of HT; for the case where rural workers can
make an on-the-job search, see the earlier version of this paper (Lee, 2002).9 Taking account of
this, the value function R(t) for non-urban workers is defined using the flow non-urban wage wR
(t) and the value of X(t + 1):
1 rR wR X V:

2.1.2. Firms' decisions

Both urban and non-urban firms produce an identical good with its price normalized at unity.
Their entry is regionally unrestricted. Let J(t) be the present value of the expected profit stream
from a filled urban job. Similarly, V(t) is the present value of the expected profit stream from a
vacant job. Then, an urban firm with a vacancy filled faces the following value function:
1 rJ p  w f1  k  max fJ V; V Vg kV Vg;

where p(t) is the flow output when the match occurs at time t, p(t) w(t) is the flow profit
conditional on the match, and the last term is the value of the firm's state-contingent choices in the
next period: a weighted average of (i) a greater value between J and V conditional on continuous
operation with probability 1 and (ii) the value V conditional on shutting down with
probability . Without loss of generality, we assume that urban productivity p is exogenously
given.
Similarly, an urban firm with a vacancy faces the following value function:
1 rV c fqhV  max J V; V V 1  qh V  V Vg

where c is the flow opportunity cost of a vacant job and the firm can hire a worker with the
probability q() as explained in footnote 8. As will be shown later, perfect competition drives the
steady-state value of a vacancy, V(t), down to zero.

8
The CRS matching technology is: m(sL,vL = m(s,v)L), where v is the aggregate level of vacancy; s is the aggregate
search intensity of urban unemployed workers: s = eu1; e is the search intensity of urban unemployed workers with size
u1 (henceforth, e is normalized at unity); and L is the urban workforce. From this, we can define a measure of job-market
tightness by the ratio of the number of vacancies to that of searchers, (t) v(t) / u1(t). The probability that a firm with a
vacancy will find a worker (i.e., the worker-arrival rate) therefore becomes q((t)) m(s(t)) / v(t), 1). By the properties of
the matching technology, q() b 0 and the elasticity of q()is a number between 0 and 1. Meanwhile, urban
unemployed workers can obtain an urban job with probability (t)q(q(t)). It is often called the unemployment hazard rate
or the job-arrival rate. See Pissarides (2000) for details.
9
An earlier version of this paper (Lee, 2002) allows non-urban workers to search for urban jobs while staying in the
non-urban sector. It uses a high search efficiency parameter for urban workers and a low efficiency one for rural workers.
Since the main results do not change, we show the simplest model here.

C.-I. Lee / Labour Economics 15 (2008) 14161434

1421

Meanwhile, wages for urban matched jobs are derived from the generalized Nash bargaining
solution: wi(t) maximizes the weighted product of firm i's and its workers' net return from a
productive job match. In this bargaining the threat points for the worker and the firm are the
opportunity values, X(t) = max[U(t),R(t)] and V(t), respectively. Since the threat point for
workers should be the second highest wage (the opportunity wage) at which urban firms can
attract workers, we thus use the greater one between U(t) and R(t) as a worker's threat point.
Therefore the wage rate for this job satisfies:
n
o
wi t arg max Wi t  X t g Ji t  V t 1g

where is the parameter for bargaining power with 0 1. This equation suggests that there
may be a premium for matched workers and firms because they are engaged in a bilateral
monopoly situation to share the benefit of greater productivity (e.g., arising from agglomeration
in the urban sector).
Next, the aggregate non-urban production is described by AR(t) = F(u2(t),z(t)), where u2(t) is
the non-urban population and z(t) is the size of land. To simplify the discussion, we will use the
CRS agricultural production function and inelastic supply of land as follows:
F u2 t ; zt u2 t b z1b :

If we normalize land size at unity, the aggregate production function is f(u2(t)) F(u2(t),1) = u2(t);
the corresponding wage and rent functions are given, respectively, by:
wR t bu2 t b1 ;

8  1

rR t 1  bu2 t b :

8  2

and

In this formulation, migration to the urban sector reduces the non-urban population, leading to
an increase in non-urban labor productivity (i.e., see Eq. (8-1)). Based on these results, we can
express the value of a representative non-urban firm, JR(t), as follows:
JR t

l
X
f u2 t  wR t  u2 t  rR t
t

t0

j 1 r s

where r0 0:

s0

In equilibrium, non-urban firms will receive a zero profit due to competition combined with the
CRS production technology. Also, the value of an urban firm with a vacancy, V(t), equals zero
due to inter- and intra-sectoral competition.
3. The equilibrium and key results
Now, we can define the economy's steady state based on the framework described earlier.

1422

C.-I. Lee / Labour Economics 15 (2008) 14161434

3.1. Steady-state equilibrium

The following system of equations can describe the steady-state HT equilibrium under flexible
wages and free mobility and entry:
kl1 h  qhu1

10  1

rV c qh J  V

10  2

rJ p  w  k J  V

10  3

rW w k R  W

10  4

rU h  qhW  R

10  5

rR wR u2

10  6

W  R g J W  V  R

10  7

R U bW

10  8

where 1 = l1 + u1 + u2 (l1 is the urban employed population, u1 is the urban unemployed

population, l1 + u1 is the urban population or the urbanization rate, and u1 / (1 u2) is the urban
unemployment rate), = v(u1), and V = 0 and b is normalized at zero.
Eq. (10-1) is the so-called Beveridge equation modified in the context of developing economies
with two sectors. Steady-state equilibrium values of l1, u1, and a measure of market tightness
imply that the number of workers who enter urban unemployment is equal to that of workers who
leave urban unemployment in the steady state, so the steady-state urban unemployment rate is
constant. Eqs. (10-2)(10-6) represent steady-state value functions derived from Eqs. (1)(5)
combined with Eq. (10-8), which deserves some discussion. Eq. (10-8) is the inter-regional labor
market equilibrium condition, and it can also be interpreted as an analogue to HT's migration
equilibrium in a dynamic setting: staying in the non-urban sector is equal in value to migrating to the
urban sector as an unemployed worker searching for a job, i.e., R = U. This condition also ensures the
model's empirical relevance: workers are allocated in urban and non-urban areas with non-zero
urban unemployment. Given that the value of employed workers exceeds that of unemployed
workers (W(t) N U(t)), the only possibility of explaining the coexistence of urban unemployment and
non-urban employment arises when W(t) N R(t) = U(t).10 Using this condition, we can express the
complex location choices of firms and workers simply by Eqs. (10-2)(10-6), which are based on the
steady-state values of , w, wR and u2. Eq. (10-7) shows the result of wage bargaining. Finally, since
In the case of W N R N U or W N U N R, urban unemployed workers move to the non-urban sector, or non-urban workers
move to the urban sector to become unemployed workers, respectively, until W(t) N R(t) = U(t) is established.
10

C.-I. Lee / Labour Economics 15 (2008) 14161434

1423

any firm is free to open a job vacancy and can enter any sector, the profit from an additional vacancy
is set at zero: V = 0, a zero-profit condition for an additional firm entry.
To illustrate the equilibrium, we begin our discussion by expressing the equations for U, R,
and W as functions of endogenous variables w, u2 and :
rW

wr hqh
;
r k hqh

rU

whqh
;
r k hqh

rR wR u2 :

11

Substituting these values (W, R, and U) and the value of J from Eq. (10-3) into the wagebargaining equation (10-7), we obtain a reduced-form version of the wage-bargaining equation:
w

gpr k hqh
:
r k ghqh

12

Eq. (12) contains the general intuition that w increases with the following partial changes11:
increases in the worker's bargaining power , the job-finding rate q(), and the urban
productivity p; and decreases in the job-destruction rate and the interest rate r.
Next, by using V = 0 and equating the J value from Eq. (10-2) with the J value from Eq. (10-3),
we obtain the standard job creation equation:
pw

r kc
:
q h

13

This equation tells us that the marginal product of urban labor, p, is equal to the wage cost plus the
expected capitalized value of the firm's opportunity cost of a vacant job, (r + )c / q(). Solving
Eqs. (12) and (13) along with the modified Beveridge curve (10-1) for , w, and l1, we obtain the
following results:


1 p1  g r k

h
;
14
g
c
q h
wp

c r k
;
q h

15

l1 1 

pqhu1 g  1 cru1 u1 gk
:
cgk

16

To determine the steady-state values of u1, u2, and v, we use the R = U condition. Substituting the
expressions for R and U from Eq. (11) into R = U yields:
whqh wR r k hqh:

17

From Eq. (17), we can derive a direct expression for the wage gap:
w=wR

r k hqh
rk
1
N1:
hqh
hqh

17

Here, the word partial is used in the sense of the partial derivative of a certain variable, holding the rest of the
variables constant.
11

1424

C.-I. Lee / Labour Economics 15 (2008) 14161434

From this, we can see that the wage gap depends positively on the interest rate and jobdestruction rate, and negatively on the urban job-finding rate q(). Some notable implications
are as follows. First, the wage gap is partly due to r + , which arises because the urban labor
market is characterized by job destruction () and subsequent costly (time-consuming) search (r),
factors that are absent in the non-urban labor market. Second, the wage gap also depends on the
urban job-finding rate (q()); if it is hard to find an urban job, the wage gap rises. Third, if we
solve the equilibrium condition (10-1), l1 = q()u1,for q()and
substitute it into Eq. (17),

then the wage gap can be expressed as w=wR 1 kr 1 ul11 . This expression tells us that the
wage gap is a function of the interest rate, the job-destruction rate, and the ratio of the urban
unemployed to employed population, a measure of the severity of unemployment. We can call all
these elements of the wage gap search frictions for future reference. The overall discussion here
allows us to interpret the wage gap as the compensating wage differential for the negative
amenities arising from job destruction and subsequent costly search as shown above.
Next, substituting Eq. (8-1) wR = u2 1 into Eq. (17) gives the solution for u2 as follows:

1
br k hqh 1b
u2
:
whqh

18

Eq. (18) provides useful results consistent with the empirical evidence: the non-urban population
declines (or urbanization advances) for the following partial changes: urban wage w goes up,
the urban job-finding rate q()goes up, the urban job-destruction rate goes down, or the
interest rate r goes down.12
To present the existence of equilibrium, we determine the value of first. is not solved
in an analytically simple form due to the non-linearity of the matching probability function,
but given q() b 0 in the CRS matching technology, Brouwer's fixed-point theorem can be
used to prove the existence of from Eq. (14) and the uniqueness can also be shown by
h 

1
g

p1g
c

 qrk
h

i

h
b1 for all . q can be determined immediately (i.e., q =q()) and
=dh r gqkqV
h
2

|{z}


then w is obtained from Eq. (15). Next, with the steady-state values of , q, and w already
determined, we can solve for u1 and u2 using the Beveridge Eqs. (10-1) and (16). Other variables are
recursively determined, and we can verify the existence and uniqueness of equilibrium. The main
findings can be summarized as follows.
3.2. The nature of the wage gap
Proposition 1. Our model accounts for the aforementioned set of stylized facts, if and only if the urban
sector exhibits higher productivity and greater search frictions, compared with the non-urban sector.
Proof. Eq. (17) and the subsequent discussions already described the nature of the wage gap on
the consumption side that arises from the so-called search frictions, i.e., the sources of the
unemployment gap.

12

The partial effect of an increase in is uncertain:

Au2
Ab

u2

!
1
Log u2

.
2
1  bb 1  b
|{z}
?

C.-I. Lee / Labour Economics 15 (2008) 14161434

1425

Next, we examine the role of the sectoral productivity differential in the wage gap on the
production side. Proposition 1 says that the gap is also based on the following implicit condition that urban average labor productivity p(t) is greater than its comparable non-urban
counterpart bf u2u2tt : pt N bf u2u2tt wR t . To see this, we substitute w from Eq. (15) into
Eq. (17) to obtain:
wR
hqh
:

p  cr k=qh r k hqh

17

After inverting the above equation, we rearrange it as follows to show that the urban-to-non-urban
labor productivity ratio is greater than unity, i.e., p(t) N wR(t):


p
r k hqh cr k=qh
r k cr k=qh

N1:

1
wR
hqh
wR
hqh
wR
19
|

{z

}
|{z}
|{z}
w=wR


This exercise shows that behind the wage differential, there exists the sectoral productivity
differential. Rearranging the first line of Eq. (19), we can express the wage differential as a
function of the labor productivity differential, p / wR:


w
r k hqh
p
cr k=qh


:
19
wR
hqh
wR
wR
|{z}

Since the last term in the second line of Eq. (19) is negative, the wage differential should be
smaller than the labor productivity differential, implying that the productivity differential is the
underlying source of the wage gap.
Last, migration from the rural to urban sector is motivated by the wage gap, which is again
based on the productivity gap. It continues until the equilibrium gap is established.

Proposition 1 stresses that the sectoral productivity differential combined with search frictions
in the urban sector is the underlying source of migration and the wage and unemployment gaps of
developing economies in a fully general dynamic search equilibrium context. It provides answers
to why the wage gap as a compensating differential for unemployment and costly search can also
be seen as a match-specific premium based on a sectoral productivity differential. Moreover, this
proposition can account for or is consistent with the aforementioned three stylized facts that
characterize the economic development process13:
Lemma 1. The current period incomes, w(t), wR(t), and b(t), can be ordered as follows:
wt NwR t Nbt :

20

13
The initial sectoral productivity differential condition is affected by migration because migration raises non-urban
productivity. However, this endogenous change in the productivity differential does not mean that equilibrium arises at
any value of p. For instance, if urban productivity p is low, many workers move to the rural sector until they bring the
rural marginal product of labor below p, ensuring that the rural wage is below the urban wage. However, this is not an
empirically interesting case because we usually observe migration from the rural to urban sector in most developing
countries. Meanwhile, in their study of the spatial mismatch hypothesis, Coulson et al. (2001) use an equilibrium
condition that sectoral productivity differential does not exceed a specified threshold.

1426

C.-I. Lee / Labour Economics 15 (2008) 14161434

Proof. The inequality w(t) N wR(t) holds according to Proposition 1: compensating differentials
always lead to w(t) N wR(t). From the Nash bargaining theory, it is obvious that w(t) N b(t). Now,
what is left is whether the condition wR(t) N b(t) really holds. To show this directly, we reexpress
Eq. (10-5) with the benefit level b un-normalized as follows:
rU b h  qhW  R:

10  5

Substituting Eq. (10-5) and rR =wR(u2) (Eq. (10-6)) into the inter-regional equilibrium condition U =R,
we can derive the equation: b + q()(W R)=wR(u2), which is rearranged to give an inequality:
wR  b h  qh W  R N0:
|{z} |{z}

10  5

Given that the job-finding rate, q(), is positive and WR is also positive in equilibrium (i.e., having
an urban job is better than having a non-urban job), we find that wR(t)N b(t) in equilibrium. Putting the
results together, our model can account for the empirical regularity of w(t)N wR(t) Nb(t).

Looking at this regularity, HT postulate that migration proceeds in response to differences in urban and
rural expected incomes. Given that the urban wage is exogenously set above the rural wage, in the original
HT model there is only one pricing instrument left that can allocate labor: the rate of urban unemployment.
Thus, HT do not really tell us that a higher urban wage is a compensating differential for unemployment risk.
Instead, unemployment risk is a pricing scheme for scarce resources under conditions of congestion.14
In contrast, our paper allows us to view the regional wage gap in developing economies as a
compensating differential for a negative amenity (e.g., job destruction and costly search) in a fully
general equilibrium setting. This interpretation is consistent with the traditional labor economics
literature (e.g., Roback, 1982). With this interpretation, our model can account for the same phenomenon
in a more structural framework with less restrictive assumptions.
3.3. Comovement among sectoral wages and the urbanization rate
While many previous studies have looked at the nature of the wage gap per se, they pay little attention
to the changes in rural and urban wages and migration flows over the urbanization process. Our model
has several structural parameters, and the changes in the parameters can generate the changes in wages
and migration flows. The relevance of the model can be supported if the changes in these key endogenous
variables are consistent with the empirical regularity that urban and non-urban wages and urbanization
comove in developing economies.15
14
Sato (2004) also noted that unemployment causes a wage differential. This paper highlights the wage gap as a compensating wage
differential for job destruction and subsequent costly and time-consuming search with job-finding probabilities on the consumption
side and as a match-specific premium based on a sectoral productivity differential on the production side in a dynamic general
equilibrium setting.
15
To give partial evidence of the comovement, I have examined the Korean data during the rapid urbanization period for many
useful properties to examine the model: (i) unlike other developing countries, Korea provides rich data about sectoral wages and
unemployment since 1963; (ii) this matching model does not require institutions such as minimum wage laws and unions (as in HT
(1970) or Calvo (1978)), so we need data about a country that was not influenced by these institutions. In light of this, the Korean
data can provide useful information because it was not until recently that institutions such as minimum wages or unions began to
affect the economy; and (iii) more important, Korea has experienced an unusually rapid urbanization process, so its data on wages
and unemployment offer another advantage in examining the implications of the model. We found some results consistent with the
model: (i) despite the absence of minimum wages and unions, the wage and unemployment gaps still appear during the early
development period; (ii) our data reveal empirical evidence lending support to Proposition 2. In particular, the data show the
comovement among urban and rural wages and the urbanization rate. More detailed results are available upon request.

C.-I. Lee / Labour Economics 15 (2008) 14161434

1427

Proposition 2. The urbanization rate and urban and non-urban wages comove for each of the
changes in {p, r, }.
Proof. See Appendix A.

This result also offers an opportunity to view labor migration in a more structural setting.
Apparently, the wage gap is a reason for migration, but our model studies why and how urban and
non-urban wages change, and it predicts that migration occurs as an adjustment toward
equilibrium in response to unanticipated changes in exogenous parameters.
For instance, a primary driving force for urban migration may be the urban productivity
shock, which raises p(t). Our model predicts a series of changes: urban wage goes up, job
vacancies rise, migration to the urban sector occurs, and the non-urban wage increases; these
changes are observed until a new steady-state equilibrium is reached. Other typical changes
involved in the economic development process would be gradual decreases in the interest rate r
and the job-destruction rate . These changes also generate effects qualitatively similar to the
increase in p.
3.4. Welfare analysis
Discussions of social optimality focus on the internalization of search externalities within the
urban sector and the between-sector allocation of resources. In our model, there are two-sided
search externalities: a worker's job search affects other workers' search due to a congestion
externality and, conversely, a firm's search for the right worker creates a similar kind of
congestion for other firms.
To address the optimality of the competitive equilibrium, we propose the utilitarian social
planner's problem, which is a standard dynamic programming problem, to model an economy's
optimal production in a multi-period setting.16 In the following Bellman equation for the value of
net production, we express current total net production as the sum of urban output y and nonurban output u2 minus the recruitment cost to the firm, cv = cu1:

i
1 h
y ub2  chu1 V yV; u V1
V y; u1 max
21
h;u2
1r
subject to
(i) yV hqhu1 p 1  ky
(ii) uV1 k  1  u1  u2 1  hqh  u1 :
The first constraint (i) is the dynamics of the urban output, suggesting that the next period's
urban output y is equal to the output of the newly operating firms with their vacancies filled q
()u1p plus the output of the surviving firms (1 )y. Constraint (ii) is the dynamics of urban
unemployment, which includes job destructions and job matches. As the right-hand side of the
Bellman equation is a contraction map for all r N 0, a unique solution exists for the value function
V(y,u1).
16

Given that individuals maximize their wealth, which comes from production in the general equilibrium setting, the
utilitarian social optimum is achieved when we maximize the overall production.

1428

C.-I. Lee / Labour Economics 15 (2008) 14161434

The first-order conditions are given by:

pVy V  Vu V1

dhqh
 c 0;
dh

22  1

bub1
 kVu V1 0:
2

22  2

And the envelope conditions are:

1 rVy 1 1  kVy V;

23  1

1 rVu1 ch Vu V1 k 1  hqh Vy Vhqhp:

23  2

1
. Similarly,
In the perfect-foresight steady-state equilibrium where Vy = Vy, we obtain Vy rk
chphq

h
=

rk

. And then, substituting these equations into the

using Vu1 = Vu1, we obtain Vu1
rkhqh
first-order conditions (22-1) and (22-2), we finally obtain:

pdhqh=dh
;
r k hqh  hdhqh=dh

24  1

and


1
1b
br k hqh
u2
:
kch phqh=r k

24  2

The other two variables y and u1 are described by the two constraints in Eq. (21), and we thus
have four variables and four equations. To attain the socially optimal allocation of resources
across regions, we solve for {, u2, y, and u1} in accordance with these four equations.
3.4.1. Welfare properties of the competitive equilibrium
Next, we compare the socially optimal solution with the decentralized competitive equilibrium
described in the earlier section. To do that, we should compare Eq. (24-1) with Eq. (14), and Eq.
(24-2) with Eq. (18), respectively. (Note that the other two equations, i.e., the two constraints in
(21), are identical.) Obviously, they are all different from each other, so we need a combination of
two policy instruments {t, s} for the competitive equilibrium to replicate the socially optimal
values of {, u2}; positive values of t and s represent a tax and a subsidy, respectively.
The following is the case in which we implement a tax t on the output of the urban firm p and a
subsidy s to non-urban wages. To facilitate comparison, we show below the equations for the socially
optimal solution on the left-hand side and those for the competitive equilibrium on the right-hand side.
c

pg
r k 1  ghqh

vs: c

1  t p1  g
:
r k ghqh

where dln[q() / dln()], i.e., the elasticity of the job-finding rate with respect to .

u2

1
1b
br k hqh
kch phqh=r k


vs:

u2

1
1b
1 sbr k hqh
:
r kch phqh=r k

C.-I. Lee / Labour Economics 15 (2008) 14161434

1429

By comparing these two sets of socially optimal and competitive equilibrium conditions, we
find that they are not identical, but the competitive equilibrium discussed earlier happens to be
equal to the optimal solution if the following condition holds:



g
r k ghqh

t 1
;
25
1  g r k 1  ghqh
and
s r=k:
To make sense of these equations, we look at the first equation.
(i) If = 1 , i.e., the usual Hosios (1990) condition established in the one-sector setting, then
t = 0, i.e., we do not need any further intervention in the urban market. However, this is not the
end of the story when we consider the inter-sectoral allocation of labor. In the competitive
equilibrium, non-urban workers make their migration decisions knowing the probabilistic statecontingent possibilities of being matched or unmatched to an urban firm. But to search for an
urban job, they have to move to the urban sector as unemployed workers; unemployed job
searchers cannot produce at all until next period, which, from the perspective of the social planner,
is seen as a waste of scarce resources. For this reason, the social planner can improve overall
efficiency by relocating some of the unmatched urban workers to the non-urban sector. Even
when the search and recruitment externalities in the urban sector cancel out with the usual Hosios
condition = 1 , the competitive equilibrium involves inefficiently many unemployed workers
and inefficiently many firms with job vacancies as a result.
To correct for the problem of excessive migration to the urban sector, we are better off
subsidizing the non-urban wage at the rate of s = r / . The intuition for this form of s is simple
and clear: as rises, we can afford to give a smaller subsidy because migration becomes less
attractive (note that U goes down). The r in the subsidy formula is to correct for the
underutilization of resources due to the time-consuming urban job-matching process if time
cost r rises, then underutilization of scarce resources becomes more costly, so we are better off
limiting migration by offering a greater subsidy.
(ii) If b 1 , then too many potential firms have incentives to post vacancies; from the
worker's perspective, a worker's chance of finding a job is (relatively) not high. In this case, we
are better off limiting both the entry of firms and the migration of workers to the urban sector by
taxing the output of the urban sector while subsidizing the wages of the non-urban sector by
s = r / as above.
(iii) If N 1 , then the opposite of case (ii) applies. In this case, we are better off subsidizing
both the output of the urban sector and the wages of the non-urban sector. These results are
summarized by Proposition 3.
Proposition 3. A tax-subsidy scheme can improve social welfare by correcting for the withinurban externalities in job matching and the between-sector distortions.
3.4.2. Discussion of the results
The intuition behind these results is that the competitive search equilibrium is likely to involve
the between-sector distortion, i.e., excessive migration to the urban sector, which generates a
time- and resource-consuming job-matching process, i.e., inefficient use of scarce resources. To
correct for this distortion, we need to limit excessive migration by subsidizing the non-urban

1430

C.-I. Lee / Labour Economics 15 (2008) 14161434

sector. In addition, we need to correct the within-sector distortion by taxing or subsidizing the
output of the urban sector accordingly. The common feature of the optimal policies in the early
HT models is that because a high urban wage causes too high urban unemployment, it is
necessary to implement policies that attract firms to the urban sector in order to increase urban
employment opportunities and reallocate urban job searchers to the rural sector.17 Albeit not
perfectly, this paper lends some partial support to this prediction in a dynamic general equilibrium
setting: the rural population in the competitive equilibrium tends to be lower than the optimal
level, so we need to subsidize rural labor; however, we do not necessarily need to subsidize urban
labor because the urban wage is flexible in our model in contrast to HT models.18
Of course, it should be noted that our model considers the search externalities in job matching
only, excluding other externalities such as agglomeration or congestion in production. Perhaps
future research that includes agglomeration or congestion externalities would offer new insight
into migration and economic growth along with more appropriate policies.
4. Summary and conclusions
Migration from the rural to the urban sector, combined with higher wages and unemployment
in cities than in rural areas, has been a central issue in developing economies. We attempted to
provide a dynamic general equilibrium reinterpretation of the static partial migration equilibrium
by Harris and Todaro (HT) (1970), under (i) flexible urban and rural wages and (ii) free mobility
of workers and free entry of firms. Our model highlights the exact conditions for the dynamic HT
equilibrium to hold in a search-matching context. To explain the observed migration pattern and
the wage and unemployment gaps across sectors as an equilibrium phenomenon, we showed that
the urban sector should exhibit the properties of (i) greater heterogeneity and specificity of jobs
and (ii) higher productivity compared with the non-urban sector.
The model allows us to view the wage gap as a compensating differential for the negative
amenities associated with job destruction and subsequent costly search on the consumption side,
which can also be seen as a match-specific premium based on a sectoral productivity differential on
the production side. The proposed model predicts the comovements among urban and non-urban
wages and migration flows to the urban sector, an empirical regularity observed over the
urbanization process of developing economies. Finally, welfare analysis suggests some interesting
policy implications.

17
Harris and Todaro (1970) argue that no single policy measure can attain the socially optimal resource allocation, and
they propose a policy combination: a wage subsidy policy in the urban sector and a labor-mobility restriction policy.
Later, Bhagwati and Srinivasan (1974) show that a uniform wage subsidy without migration restriction can yield the
optimal, first best solution. A subsequent study by Basu (1980) relaxes the condition for the Bhagwati and Srinivasan
result to hold.
18
In a study related to this paper, Sato (2004) shows in a static setting that the Hosios (1990) condition is sufficient
enough only to generate the optimality of the search equilibrium. As discussed in the introduction, this study takes a
dynamic setting in which non-urban workers have to move to the urban sector as unemployed workers to conduct a timeconsuming search for an urban job; while they search as unemployed urban workers, they cannot produce at all, an
inefficient use of scarce resources. In response to this overmigration, too many firms open vacancies and pay the real cost
as a result. These features cannot be considered in static frameworks, since time is a valuable resource only in a dynamic
setting. Further, the social welfare function in Sato is based on workers' utility only, while this paper considers a broader
measure of production, because workers' wages plus profits arise from production, and the utilitarian social welfare is
maximized when production is maximized in our model. For these reasons, the Hosios (1990) condition is not sufficient
enough, and we thus need to correct for inefficient allocation of resources across regional markets.

C.-I. Lee / Labour Economics 15 (2008) 14161434

1431

While the sectoral productivity differential across sectors is a crucial element of our model and
results, this paper did not mention what drives it and instead assumed sufficiently high urban
productivity. Perhaps the differential arises from the agglomeration effect in the urban sector, given
that (i) according to the new economic geography, close spatial proximity involves pecuniary
externalities it reduces the costs of trading both intermediate and final goods, and/or (ii) according
to the recent economic growth literature, urban agglomeration promotes human capital accumulation
through active learning. Incorporating these features into the present model seems to be a promising
avenue for future research.
Acknowledgments
The author thanks Charles Brown, Yongsung Chang, Hyungjun Kim, Bonggeun Kim,
participants at numerous seminars, two anonymous referees and co-editor Fabien Postel-Vinay
for useful comments and constructive suggestions. This study was supported by the 2007
SungKyunKwan University faculty research fund.
Appendix A. Proof of Proposition 2
Proof. For the sake of exposition, we will find the derivatives of w, wR, and 1 u2 (the urban
population) with respect to the parameters of our model {p, r, }. If the endogenous variables {w,
wR, and 1 u2} change in the same direction in response to the changes in {p, r, }, comovement
can be established. First, we obtain the following results from Eq. (14):
dh 1  g

=AN0;
dp
gc

where

Au1 

r kqVh
gqh2

N0

A  1  1

dh
1

=Ab0;
dr
gqh

A  1  2

dh
1

=Ab0:
dk
gqh

A  1  3

By manipulating Eqs. (14) and (15), we can derive w = (p + c). Combining this with the above
differentiation results, we obtain the following results:
0
1
B
dw
dh C
B
C
gB1 c 
C N0;
@
dp
dp A
|{z}

A  2  1

dw
dh
gc
b0;
dr
dr
|{z}


A  2  2

dw
dh
gc
b0:
dk
dk
|{z}

A  2  3

1432

C.-I. Lee / Labour Economics 15 (2008) 14161434

Next, we obtain the following set of partial derivatives from the rural population equation (18):
Au2
0;
Ap

A  3  1



Au2
1
b
b
u

N0;
1  b 2 whqh
Ar


Au2
1
b
ub2

N0;
1b
whqh
Ak


Au2
1
br k hqh
ub2

b0;
1b
w2 hqh
Aw
0
1

z}|{
Bbwr k dq=d C 19
Au2
1
B
C
ub2 B

Cb0:
A
1b @
Ah
whqh2

A  3  2
A  3  3
A  3  4

A  3  5

Then, we can determine the net effects of parameter changes on the non-urban population using
the chain rule:
du2
Au2 dh Au2 dw

b0

dp
Ah dp |{z}
Aw dp
|{z}
|{z}
|{z}
 
du2
Au2 Au2 dh Au2 dw

N0

dr |{z}
dr
dr
Ar |{z}
Ah |{z}
Aw |{z}
|{z}
   
du2
Au2 dh Au2 dw

N0
dk |{z}
dk
dk
Ah |{z}
Aw |{z}
|{z}
   

d 1  u2
N0;
dp

A  4  1

d 1  u2
b0;
dr

A  4  2

d 1  u2
b0:
dk

A  4  3

Similarly, we can find the net effects of parameter changes on the non-urban wage:
dwR AwR du2

N0; wR bub1
2
dp
Au2 dp
|{z} |{z}
 
dwR AwR du2

b0;
dr
Au2 |{z}
dr
|{z}

dwR AwR du2

b0:
dk
Au2 |{z}
dk
|{z}


A  5  1

A  5  2

A  5  3

dq() / d = q()(1 + q,) N 0 is derived because the elasticity of q() is bounded between 1 and 0 in the CRS
matching technology: 1 q, 0.
19

C.-I. Lee / Labour Economics 15 (2008) 14161434

1433

Using these results, we can obtain the following. For each of the changes in {p, r, }, we see that
the derivatives of w, wR and 1 u2 take the same signs, implying that the w, wR and 1 u2
comove.20

References
Banerjee, B., 1984. Information flow, expectations and job search. Journal of Development Economics 15, 239257.
Basu, K., 1980. Optimal policies in dual economies. Quarterly Journal of Economics 100, 10671071.
Behrman, J.R., 1999. Labor markets in developing countries. In: Ashenfelter, O., Card, D. (Eds.), Handbook of Labor
Economics, vol. 3. Elsevier, Amsterdam, pp. 28592939.
Bhagwati, J., Srinivasan, T., 1974. On reanalyzing the HarrisTodaro model: policy rankings in the case of sector-specific
sticky wages. American Economics Review 64, 502508.
Brueckner, J., Kim, H., 2001. Land markets in the HarrisTodaro model: a new factor equilibrating ruralurban migration.
Journal of Regional Science 41, 507520.
Brueckner, J., Zenou, Y., 1999. HarrisTodaro models with a land market. Regional Science and Urban Economics 29,
317339.
Calvo, G., 1978. Urban unemployment and wage determination in LDC's: trade unions in the HarrisTodaro model.
International Economic Review 19, 6581.
Coulson, N.E., Laing, D., Wang, P., 2001. Spatial mismatch in search equilibrium. Journal of Labor Economics 19, 949972.
Diamond, P., 1984. A Search-equilibrium Approach to the Microfoundations of Macroeconomics. MIT Press, Cambridge.
Fields, G., 1975. Ruralurban migration, urban unemployment and underemployment, and job-search activity in LDCs.
Journal of Development Economics 2, 165187.
Fields, G., 1989. On-the-job search in a labor market model: ex ante choices and ex post outcomes. Journal of
Development Economics 30, 159178.
Fujita, M., Thisse, J.-F., 1996. Economics of agglomeration. Journal of the Japanese and International Economies 10,
339378.
Harris, J., Todaro, M., 1970. Migration, unemployment and development: a two-sector analysis. American Economic
Review 60, 126142.
Hatton, T., Williamson, J., 1992. What explains wage gaps between farm and city? Exploring the Todaro model with
American evidence, 18901941. Economic Development and Cultural Change 40, 267294.
Hazari, B.R., Sgro, P.M., 1987. Disguised urban employment and welfare in a general equilibrium model with segmented
labour markets. Journal of Regional Science 27, 461475.
Hosios, A.J., 1990. On the efficiency of matching and related models of search and unemployment. Review of Economic
Studies 57, 279298.
Laing, D., Park, C., Wang, P., 2005. A modified HarrisTodaro model of ruralurban migration for China. In: Kwan, F.,
Yu, E. (Eds.), Critical Issues in China's Growth and Development. Ashgate, London, pp. 245264.
Lee C.I. A theory of the long-run gaps in wages and unemployment between urban and non-urban sectors. Mimeo,
University of Michigan, 2002.
Lucas, R., 1985. Migration amongst the Botswana. Economic Journal 95, 358382.
Moene, K.O., 1988. A reformulation of the HarrisTodaro mechanism with endogenous wages. Economics Letters 27,
387390.
Mohtadi, H., 1989. Migration and job search in a dualistic economy: a TodaroStigler synthesis. Economics Letters 29,
373378.
Mortensen, D., Pissarides, C., 1994. Job creation and job destruction in the theory of unemployment. Review of Economic
Studies 61, 397415.
Ortega, J., 2000. Pareto-improving immigration in an economy with equilibrium unemployment. Economic Journal 110,
92112.
Pissarides, C., 2000. Equilibrium Unemployment Theory, 2nd ed. MIT Press, Cambridge.
Quibria, M., 1988. Migration, trade unions, and the informal sector: a note on Calvo. International Economic Review 29,
557563.
20

Although not reported in the text, we have examined the comovement in response to other exogenous parameters. (i)
First, for the change in c, we can find the comovement if ,c b 1. (ii) Second, for the change in , it is uncertain how wR
and 1 u2 respond to that, so we cannot say that the comovement always happens. Finally, (iii) for the change in , wR
and 1 u2 are found to comove but w does not necessarily comove with wR and 1 u2.

1434

C.-I. Lee / Labour Economics 15 (2008) 14161434

Roback, J., 1982. Wages, rents, and the quality of life. Journal of Political Economy 90, 12571278.
Sato, Y., 2004. Migration, frictional unemployment, and welfare-improving labor policies. Journal of Regional Science 44,
773793.
Stark, O., Gupta, M., Levhari, D., 1991. Equilibrium urban unemployment in developing countries: is migration the
culprit? Economics Letters 37, 477482.
Stiglitz, J., 1974. Alternative theories of wage determination and unemployment in LDCs: the labor turnover model.
Quarterly Journal of Economics 88, 194227.
Stiglitz, J., 1976. The efficiency wage hypothesis, surplus labor and the distribution of income in LDCs. Oxford Economic
Papers 28, 185207.
Todaro, M., 1969. A model of labor migration and urban unemployment in less developed countries. American Economic
Review 59, 138148.
Vishwanath, T., 1991. Information flow, job search, and migration. Journal of Development Economics 36, 313335.
Zenou, Y., Smith, T.E., 1995. Efficiency wages, involuntary unemployment and urban spatial structure. Regional Science
and Urban Economics 25, 547573.