Journal of Urban Economics 63 (2008) 306–324

www.elsevier.com/locate/jue

New economic geography with heterogeneous

preferences: An explanation of segregation
Dao-Zhi Zeng a,b,∗
a Graduate School of Management, Kagawa University, Kagawa, 760-8523, Japan
b Center for Research of Private Economy, Zhejiang University, Zhejiang 310027, China

Received 23 March 2006; revised 20 October 2006

Available online 12 March 2007

Abstract
The Tiebout hypothesis (residential choice depends solely on local public goods) is extensively applied
to explain geographic segregation, and the related literature finds that residents are segregated according
to their heterogeneous preferences for public goods. This paper further examines the heterogeneous prefer-
ences for private goods in a spatial economy without public goods. Specifically, we employ a new economic
geography framework in which the heterogeneous preferences of mobile workers on manufactured goods
are incorporated. The rigorous general equilibrium analysis conducted here shows that the increasing-
returns technology and monopolistic competition form a mechanism endogenously leading to persistent
residential segregation. There is an evolving path with decreasing transport costs in which the two types of
mobile workers are segregated, while two industries evolve from dispersion to agglomeration.
© 2007 Elsevier Inc. All rights reserved.

JEL classification: F12; R12

Keywords: Segregation; Heterogeneous preferences; Economic geography; Agglomeration

1. Introduction

Spatial segregation is evident in a lot of cities in the United States (see [6,7,19]). While the
traditional meaning of segregation in the USA is the separation of Blacks and Whites, the United

* Fax: +81 87 832 1905.

E-mail address: zeng@gsm.kagawa-u.ac.jp.

0094-1190/$ – see front matter © 2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.jue.2007.02.004
 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324 307

States Census 2000 reveals that this issue has been complicated by new forms of Latino and
Asian separateness and mixing.
In economics, spatial segregation is widely interpreted by the Tiebout theory [27] on local
public goods and its counterpart in the theory of clubs (see the original paper of Buchanan [4]
and Scotchmer’s survey papers [20,21]). The literature examines Tiebout’s conjecture that effi-
cient provision of public goods is realized in an economy where people “vote with their feet.”
It turns out that the answer rests on a number of assumptions. For example, when there is no
benefit in having people of different skills or cultures together in one community, a group of
individuals with heterogeneous preferences would be better off forming a separate club for each
homogeneous preference group (see Berglas [2], Brueckner [3], Conley and Wooders [5]).
While local policies clearly matter for residential choice, a paper of Rhode and Strumpf [19]
reports that long-run trends in geographic segregation are inconsistent with the simple Tiebout
theory. Instead, employment and other factors are more important. In other words, non-Tiebout
incentives are perhaps the driving forces in residential decisions.
To find such a non-Tiebout incentive, this paper examines a spatial economy without public
goods. Specifically, we focus on people’s heterogeneous preferences for private goods only. Con-
sumers in the real world typically hold heterogeneous preferences for private goods. With respect
to cars, the Thais like pickup trucks while the Indonesians like minivans. This paper shows that
the preference heterogeneity is another reason for the persistent segregation. The importance of
heterogeneous preferences is already noted by Scott [22], who finds that while external barriers
(e.g. housing discrimination based on race) are still viewed as the main cause of black–white
segregation, preference plays a larger role among Asians. In fact, there are districts, frequently
referred to as “China Town,” which are populated primarily by Asians in many large US cities.
Numerous varieties of goods preferred by Asian people agglomerate there. Preference plays a
more important role for segregation in an ethnically homogeneous country, such as Japan. For
example, the people in Kitakata City of Fukushima Prefecture have a preference for Ramen,
a kind of Chinese noodles, while people in the Kagawa Prefecture prefer Udon, a Japanese noo-
dle.
Since there is a wide variety of private goods in the real world, the differences in preferences
deserve more attention in segregation research. The findings in this research will fill this theo-
retical gap by showing how heterogeneous preferences lead to the segregation of consumers and
industries. The results may clarify why the leading automaker Toyota selected Indonesia as its
global production base for minivans and Thailand for pickup trucks,1 and how the difference
in preferences between Kitakata and Kagawa people in Japan has resulted in the production of
the well-known Kitakata–Ramen and Sanuki–Udon (Sanuki being another name given to Ka-
gawa).
It was not easy to formulate a general-equilibrium model for the examination of heteroge-
neous preferences for various private goods, but the techniques devised in the framework of the
new economic geography (NEG) have made it possible. NEG is concerned with how agglom-
eration is derived from monopolistic competition, increasing-returns technology, and transport
costs. Extensive reviews on this subject can be found in the book of Fujita et al. [9] and the
survey paper of Fujita and Mori [10]. The rapid development in recent years makes this research
area a mainstream of trade theory and regional science. Unfortunately, due to model tractability,

1 See Hakim’s article in http://www.thejakartapost.com/yearender/bus06.asp and Runckel’s article in http://www.

business-in-asia.com/auto_article2.html.
308 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324

there are only a few papers considering heterogeneous residents in the NEG literature so far. Fo-
cusing on non-market factors, Tabuchi and Thisse [24] and Murata [15] find that heterogeneous
preferences for residential location induce a strong dispersion force. Another paper of Glazer
et al. [11] places the heterogeneous preferences of commodities into their framework, but the
transport costs are assumed to be sufficiently high to make consumers buy goods locally and
the workers are immobile. Therefore, the segregation and industrial evolution pattern due to the
heterogeneity of the population remains unclear.
By reconstructing the footloose entrepreneur model of Ottaviano [16] and Forslid and Otta-
viano [8], this paper incorporates the heterogeneity of mobile workers into the NEG framework.
Specifically, there are two symmetric regions with the same number of unskilled workers who
are immobile across regions and work in the traditional sector. In contrast, there are two types
of mobile workers who work in the manufacturing sector. There are also two industries in the
manufacturing sector. Different types of workers have the same productivity in the manufactur-
ing production, but they have different preferences for two types of industrial goods. Therefore,
a firm of a particular type prefers a location with more consumers of the same type. However,
since all workers consume both types of industries goods, the two industries are unlikely to ag-
glomerate even if the workers are separated when the transport costs are high.
A prominent equilibrium is the distribution of full dispersion, in which two types of mobile
workers and two industries equally disperse in two regions. However, our rigorous general-
equilibrium analysis shows that this equilibrium is unstable. Intuitively, when a worker migrates,
the destination market becomes bigger, which increases the demand and encourages firms to
move (the so-called “backward linkage” in the literature). However, if a firm follows the moving
worker, then the competition of firms leads to a reduction in the local price index, which encour-
ages the moving worker but discourages the moving firm. In the original core-periphery model,
the move of homogeneous workers automatically implies the move of homogeneous firms. The
two opposite forces are synthesized as the result of agglomeration (respectively, dispersion) when
transport costs are small (respectively, large). In contrast, workers and firms are heterogeneous
in our framework. The move of a worker does not automatically imply the move of firms of one
type. On the contrary, two types of firms follow the moving worker in proportion to the new mar-
ket demand, which encourages further move of the same type workers (the so-called “forward
linkage”). As a result, the increasing-returns technology and monopolistic competition form a
mechanism endogenously leading to persistent residential segregation, and a consumer lives in a
region in which other consumers share the same preferences. This is true even if the difference
between the two types of workers is small.
Previewing the equilibrium, there is an evolving path with decreasing transport costs in which
two types of workers are segregated in two regions and two types of industry separate gradually.
Two industries do not separate for large transport costs because of the local demands. However,
the “home market effect” occurs in the sense that each industry concentrates in the region with a
higher demand for its products but the distribution ratio is higher than the ratio of local demands.
The paper is constructed as follows. The basic model is established in Section 2. In Sec-
tion 3, we show that there is no stable interior equilibrium in which two types of mobile workers
disperse. Then we examine all possible corner equilibria and see how industries evolve with de-
creasing transport costs. Although the full agglomeration is also possible for sufficiently small
transport costs, it is unlikely to occur because the evolving path of segregation equilibria is con-
tinuous. Finally, Section 4 concludes the paper by indicating that the preference heterogeneity is
a fundamental reason but not a unique one for segregation.
 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324 309

2. Brief setup

We extend the footloose entrepreneur model of Ottaviano [16] and Forslid and Ottaviano
[8] to the case of two kinds of population and two industries. We make this choice because
this framework retains most features of the first core-periphery model of Krugman [13] and is
analytical solvable.
Specifically, we consider the economy of two symmetric regions 1 and 2 with the same num-
ber L of unskilled workers who may work in the traditional sector. There are two industries x
and y in the manufacturing sector. Two kinds of skilled workers are mobile across these two re-
gions. The population of each type is the same H . For simplicity, all skilled workers are involved
in manufacturing production and their productivity in any industry is supposed to be the same.2
Call the unskilled workers type 0, and two kinds of mobile workers types x and y. Those workers
have heterogeneous preferences. Specifically, the preference of the representative consumer of
type k ∈ {0, x, y} in region i ∈ {1, 2} is captured by the following utility functions:
  μ  1−μ
Ui0 = Xi0 Yi0 2 A0i ,
 x μη  x μ(1−η)  x 1−μ y  y μ(1−η)  y μη  y 1−μ
Ui = X i
x
Yi Ai , Ui = X i Yi Ai ,

where Aki is the consumption of the traditional good in region i of a type k consumer, and
   σ    σ
σ −1 σ −1
σ −1 ky σ −1
Xik = dikx (s) σ ds , Yik = di (s) σ ds , k ∈ {0, x, y}
s∈N x s∈N y

represent the consumption of two types of manufactured goods. In the above expressions, N x
(respectively, N y ) is the mass of variety in industry x (respectively, industry y), and dikx (s) (re-
ky
spectively, di (s)) is the consumption of a typical good in industry x (respectively, industry y)
by a consumer of type k ∈ {0, x, y}. Parameter μ represents the relative importance of the man-
ufactured goods and the traditional good, while η represents the relative consumption of two
industries inside the manufacturing sector. A type k ∈ {x, y} mobile worker enjoys an absolute
advantage over another type of mobile workers in the consumption of the type k industry goods.
Finally, the parameter σ (>1) is both the elasticity of demand of any variety and the elasticity of
substitution between any two varieties within an industry. Note that the unskilled workers do not
differentiate the production of two industries, which keeps the symmetric role of the immobile
factor as an NEG model. In addition, since the productivity is the same for both types of mobile
workers, a manufacturing firm is related to a particular type of worker only through its market.
The two industries, x and y, are modeled in a symmetric way. Their only difference is the
preference of the mobile workers for them. We assume
1
<η<1 (1)
2
to reflect the fact that type k mobile workers prefer type k industry goods more than another type.

2 Some authors consider the case of heterogeneous productivity, e.g., Amiti and Pissarides [1] and Zeng [29]. Particu-
larly, Zeng [29] has shown that, in the case of multiple types of skilled workers with different skills, the workers of the
same type are likely to agglomerate in a region when transport costs decrease. Therefore, this paper focuses on the case
of homogeneous productivity.
310 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324

Let nkj be the set of all varieties in industry k ∈ {x, y} produced in region j ∈ {1, 2}. Given
the price pjki (s) of variety s ∈ nkj consumed in region i, the CES demand djki1 k2 (s) of variety s
consumed by workers of type k1 in location i is calculated as
pjxi (s)−σ μ 0 pj i (s)−σ μ 0
y
0y
dj0xi (s) = I , dj i (s) = y I , (2)
(Pix )1−σ 2 i (Pi )1−σ 2 i
pjxi (s)−σ pj i (s)−σ
y
xy
djxxi (s) = μηIix , dj i (s) = y μ(1 − η)Iix , (3)
(Pix )1−σ (Pi )1−σ
pjxi (s)−σ pj i (s)−σ
y
yx y yy y
dj i (s) = μ(1 − η)Ii , dj i (s) = y μηIi , (4)
(Pix )1−σ (Pi )1−σ
where
    1
1−σ
Pix = piix (s)1−σ ds + pjxi (s)1−σ ds ,
s∈nxi s∈nxj
    1
1−σ
y y y
Pi = pii (s)1−σ ds + pj i (s)1−σ ds
y y
s∈ni s∈nj

are the price indices and Iik is the local income of type k in region i determined by

I1k = w1 λk H, I2k = w2 (1 − λk )H, for k = x, y;

I10 = w10 L, I20 = w20 L, for k = 0.
In the above expressions, wi and wi0 are the wages of skilled and unskilled workers in region i,
respectively. Since both types of skilled workers are assumed to have the same productivity, their
wages are the same.
Let
yk
djki (s) = dj0ki (s) + djxki (s) + dj i (s) (5)
be the total demand of variety s in industry k produced in region j and consumed by all con-
sumers in region i.
Firms in industries x and y are monopolistically competitive and employ any type of skilled
workers and unskilled workers under increasing returns to scale. Specifically, in order to produce
q(s) units of variety s, a firm incurs a fixed input requirement of one unit of any type of skilled
worker and a marginal input requirement of βq(s) units of unskilled workers, so the total cost is
TCi (s) = wi + wi0 βq(s).
Note that the firms of two industries have no productivity difference. Let λx and λy be the popu-
lation shares of type x and y workers in region 1, respectively. For convenience, we also denote
y y
the numbers of firms in two regions by nx1 , nx2 , n1 , and n2 . Then we have
y y

nx1 + n1 = (λx + λy )H, nx2 + n2 = (1 − λx ) + (1 − λy ) H,

y y
nx1 + nx2 = N x , n1 + n2 = N y .
 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324 311

Therefore, N x + N y = 2H .
Without loss of generality, we suppose that one unskilled worker produces one unit of tradi-
tional good and take the traditional good as the numéraire. Then,
piA = wi0 = 1
holds in any region i.
Assume Samuelson’s iceberg form of interregional transportation costs: τ > 1 units of the
manufactured good must be shipped for one unit to reach the other region. A typical manufactur-
ing firm located in region i determines prices by maximizing profit

Πix (s) = piix (s)diix (s) + pij

x
(s)dijx (s) − β diix (s) + τ dijx (s) − wi ,
y y y y y y y

Πi (s) = pii (s)dii (s) + pij (s)dij (s) − β dii (s) + τ dij (s) − wi .
According to (5), the first-order condition of maximization gives
βσ τβσ y βσ y τβσ
piix (s) = , x
pij (s) = , pii (s) = , pij (s) = .
σ −1 σ −1 σ −1 σ −1
Therefore, the price index is
βσ  x  1 y βσ  y y 1
Pix = ni + φnxj 1−σ , Pi = ni + φnj 1−σ , (6)
σ −1 σ −1
where φ = τ 1−σ ∈ (0, 1].
Due to the free-entry, the profit of any firm is zero. Therefore, the wage wi of skilled workers
in region i is determined by

β[diix (s) + τ dijx (s)]
wi = piix (s)diix (s) + pij
x
(s)dijx (s) − β diix (s) + τ dijx (s) = (7)
σ −1
for working in firms of industry x and
y y
y y y y y y
β[dii (s) + τ dij (s)]
wi = pii (s)dii (s) + pij (s)dij (s) − β dii (s) + τ dij (s) = (8)
σ −1
for working in firms of industry y.
Given the population shares λx and λy in region 1, the expenditure Eik in region i for industry
k goods is calculated as
μ
E1x = L + w1 λx H μη + w1 λy H μ(1 − η),
2
y μ
E1 = L + w1 λx H μ(1 − η) + w1 λy H μη,
2 (9)
μ
E2 = L + w2 (1 − λx )H μη + w2 (1 − λy )H μ(1 − η),
x
2
y μ
E2 = L + w2 (1 − λx )H μ(1 − η) + w2 (1 − λy )H μη.
2
y y
Let xi = diix (s) + τ dijx (s) and yi = dii (s) + τ dij (s) be the amounts of production of any variety
s of industries x and y in region i (evidently, xi and yi do not depend on the name s due to the
symmetry across varieties). In the case of nonempty production, (2)–(4) give
312 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324

x )−σ
(p11 x )−σ
(p12
x1 = d11
x
+ τ d12
x
= E x
+ τ Ex
(P1x )1−σ 1 (P2x )1−σ 2
 
σ −1 E1x φE2x
= + , (10)
βσ nx1 + φnx2 φnx1 + nx2
 
σ −1 E2x φE1x
x2 = + x , (11)
βσ nx2 + φnx1 n1 + φnx2
 y y 
σ −1 E1 φE2
y1 = y y + y y ,
βσ n1 + φn2 φn1 + n2
 y y 
σ −1 E2 φE1
y2 = y y + y y . (12)
βσ n2 + φn1 n1 + φn2
The above equations show that the demands depend on the distribution of workers, firms, and
wages of workers. Although they appear to be complicated, they make it possible to conduct a
general equilibrium analysis.

3. Equilibrium analysis

In order to study the spatial equilibrium and its stability, following established traditions in
economic geography, we assume that markets for goods adjust instantaneously, while interre-
gional migration of firms and workers is relatively slow, implying that wages adjust much faster
than the labor share. The dynamics of labor shares is simply as follows
⎧ x
⎪
⎪
dλ w1 w2
⎨ dt = V (λ , λ ) ≡ (P x )μη (P y )μ(1−η) − (P x )μη (P y )μ(1−η) ,
x x y
⎪
1 1 2 2
(13)
⎪
⎪ dλ y w 1 w 2
⎪
⎩ = V (λ , λ ) ≡ x μ(1−η) y μη − x μ(1−η) y μη .
y x y
dt (P1 ) (P1 ) (P2 ) (P2 )
The stability of interior and corner equilibria in such dynamics is extensively discussed
in [26,28].

3.1. Absence of stable interior equilibrium

Similar to the case of homogeneous population, there are at most three interior equilibria.
However, the different conclusion is that none of them is stable.
This can be shown as follows. Note that, in an interior equilibrium, we have λx , λy ∈ (0, 1).
According to (13), the real wages working in any industry and in any region must be the same at
the equilibrium
w1 w2
y = y , (14)
(P1x )μη (P1 )μ(1−η) (P2x )μη (P2 )μ(1−η)
w1 w2
x μ(1−η) y μη = x μ(1−η) y .
(P1 ) (P1 ) (P2 ) (P2 )μη
From (6), the above relation implies that the ratio of two industries in two regions are the same:
y y
nx1 /n1 = nx2 /n2 = N x /N y . Meanwhile, non-empty production of two industries in both regions
requires wi = βxi /(σ − 1) = βyi /(σ − 1) for i = 1, 2 from (7) and (8). Then we have
 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324 313

y
(1 − φ 2 )E2x (1 − φ 2 )E2
σ (w2 − φw1 ) = = ,
(1 − λ + φλ)N x (φλ + 1 − λ)N y
y
(1 − φ 2 )E1x (1 − φ 2 )E1
σ (w1 − φw2 ) = = ,
[λ + φ(1 − λ)]N x [λ + φ(1 − λ)]N y
y y y
where λ = nx1 /N x = n1 /N y . The above equalities imply that E1 = E1x N y /N x and E2 =
E2x N y /N x . On the other hand, according to (9), the total expenditures of x and y in two re-
y y
gions with respect to the local wages are balanced: E1x /w1 + E2x /w2 = E1 /w1 + E2 /w2 . This
is because the two industries are symmetric. The balance equality can be rewritten as
  
Ny w1 x
1− x E1 +
x
E = 0.
N w2 2
Since E1x , E2x and w1 /w2 are all positive, it follows that N x = N y (= H ), and then, nx1 =
y y y
n1 = λH , nx2 = n2 = (1 − λ)H . Furthermore E1x = E1 implies the same population shares
λ = λ (= λ).
x y

On the other hand, wi = βxi /(σ − 1) gives the expressions for w1 and w2 as follows
μ L 2σ φλ + [σ − μ + (σ + μ)φ 2 ](1 − λ)
w1 = ,
σ − μ 2H φσ [λ2 + (1 − λ)2 ] + [σ − μ + (σ + μ)φ 2 ]λ(1 − λ)
μ L 2σ φ + λ(1 − φ)[σ − μ − (σ + μ)φ]
w2 = ,
σ − μ 2H φσ [λ2 + (1 − λ)2 ] + [σ − μ + (σ + μ)φ 2 ]λ(1 − λ)
and (14) says that λ is a solutions of
  μ
λ + φ(1 − λ) 1−σ 2σ φ + (1 − λ)(1 − φ)[σ (1 − φ) − μ(1 + φ)]
= . (15)
1 − λ + φλ 2σ φ + λ(1 − φ)[σ (1 − φ) − μ(1 + φ)]
We have obtained the same wage expressions for the homogeneous population model (see [8,
(16)]). Ottaviano [16, Appendix A] and Forslid and Ottaviano [8, p. 236] have shown that there
are at most three roots in (15). Among them, symmetric equilibrium λ = 1/2 is stable for small
φ and two asymmetric equilibria are always unstable. Since the stability conditions of (13) in our
heterogeneous population model are more stringent than that of homogeneous population model,
two asymmetric equilibria remain to be unstable here. Surprisingly, Appendix A shows that the
symmetric equilibrium is also unstable when workers are heterogeneous. Therefore, we conclude
as follows.

Proposition 1. There is no stable interior equilibrium, in which two types of mobile workers
disperse.

Intuitively, the instability results from the fact that heterogeneous preferences of workers im-
ply weaker competition effect among firms. In the framework of homogeneous workers, the
move of a worker is linked with the move of a firm. Although the moving worker increases the
destination market, the moving firm increases the competition which decreases the local prices.
Therefore, the net benefit is not definitely positive, particularly when transport costs are large.
However, when the workers and industries are heterogeneous, the move of a worker is not linked
with the move of a firm in the same type industry. On the contrary, firms in both industries may
follow the moving worker in proportion to the new market demand. This decreases the compet-
itive effects, and does not discourage the moving firms to a great degree. At the same time, the
moving worker is encouraged by the lower price indices and he/she is followed by more workers
of the same type. This finally causes the instability of the interior equilibrium.
314 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324

3.2. Industry distribution

Appendix B further rejects the possibility of a stable asymmetric equilibrium in which only
one type of skilled worker disperses. Therefore, only two kinds of population distributions are
possible:

(i) Two types of mobile worker segregate in two regions, and

(ii) both types of worker agglomerate in the same region.

They are examined in the following two parts, respectively.

3.2.1. Segregating workers

Without loss of generality, let the type x workers be in region 1 and the type y workers in
y y
region 2. In notation, λx = 1, λy = 0. Then, nx2 = N x − nx1 , n1 = H − nx1 and n2 = H − N x + nx1
hold.
The two industries are not necessarily separated because all workers consume all goods of
both industries. If both industries disperse in two regions, then (7) and (8) give
βx1 βy1 βx2 βy2
w1 = = , w2 = = .
σ −1 σ −1 σ −1 σ −1
The only solution to the above equations is
μL
w1 = w 2 = , (16)
(σ − μ)H
σ (1 − φ) + (2η − 1)μ(1 + φ)
nx1 = H, (17)
2σ (1 − φ)
y σ (1 − φ) + [2(1 − η) − 1]μ(1 + φ)
n1 = H,
2σ (1 − φ)
N x = N y = H.
Two industries disperse iff nx1 is in (H /2, H ), which is equivalent to
N σ − H (2η − 1)μ σ − μ(2η − 1)
φ< = . (18)
H (2η − 1)μ + N σ σ + μ(2η − 1)
The skilled workers have no incentive to migrate because their real wages are higher
w1∗ w2∗
y > y ,
(P1x )μη (P1 )μ(1−η) (P2x )μη (P2 )μ(1−η)
w1∗ w2∗
y < y .
(P1x )μ(1−η) (P1 )μη (P2x )μ(1−η) (P2 )μη
To check the stability of this equilibrium, we further investigate whether a firm benefits from
switching its production between x and y industries or moving between two regions. Since
∂x1 ∂y1 ∂x2 ∂y2
< 0, y < 0, > 0, y > 0,
∂nx1 ∂n1 ∂nx1 ∂n2
hold at the equilibrium iff (18) is satisfied, no firms switch or move, and the equilibrium is stable
under (18).
 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324 315

Note that (17) increases with η. Therefore, for more heterogeneous preferences, the industries
agglomerate more intensively. Meanwhile, (17) increases with φ, which means that the industries
agglomerate more for smaller transport costs.
Krugman [12] uses a framework of heterogeneous preferences framework to examine trade
theory. His model also involves two groups of consumers but each consumer is assumed to con-
sume only one type of industrial goods. Furthermore, the distribution of consumers is given
exogenously. Krugman finds the so-called “home market effect,” in which each industry con-
centrates in the country with a higher demand for its products. In contrast, our model assumes
that each consumer consumes both types of industrial goods to various degrees and the con-
sumers segregate endogenously. The increasing-returns technology employed here allows us to
reproduce the home market effect. In fact, the ratio of demands of x industry in regions 1 and 2
is
μηw1 H + μ2 w10 L σ + (2η − 1)μ
= ,
μ(1 − η)w1 H + μ 0
2 w1 L
σ − (2η − 1)μ
where w1 is given in (16). The reader can check that the ratio above is smaller than nx∗ x∗
1 /n2 ,
where n1 is given in (17) and n2 = N − n1 . In other words, it implies that region 1 is a net
x∗ x∗ x∗

exporter of goods in industry x.

Next, we turn to the examination of the equilibrium of separating industries. This full separa-
tion equilibrium is described as
y
λx = 1, λy = 0, nx1 = n2 = N x = N y = H.
Solving the equations of w1 = βx1 /(σ − 1), w2 = βy2 /(σ − 1), (10) and (12), we obtain
μL
w1 = w 2 = .
(σ − μ)H
The price indices in two regions are calculated as follows
βσ 1 y βσ 1
P1x = H 1−σ , P1 = (φH ) 1−σ ,
σ −1 σ −1
y βσ 1 βσ 1
P2 = H 1−σ , P2x = (φH ) 1−σ .
σ −1 σ −1
Then, with respect to the real wages,
w1 w2 w1 w2
x μη y μ(1−η) > x μη y μ(1−η) , x μ(1−η) y μη < x μ(1−η) y
(P1 ) (P1 ) (P2 ) (P2 ) (P1 ) (P1 ) (P2 ) (P2 )μη
hold at the equilibrium for all φ ∈ (0, 1) from (1). Therefore, none of the skilled workers has the
incentive to migrate.
To check the stability of this equilibrium, we investigate whether firms benefit from switching
their production between industries and/or moving between two regions. For example, if an x-
firm moves from region 1 to region 2, then its profit becomes
βx2 Lμ(1 − φ)[N σ (1 − φ) − H (2η − 1)μ(1 + φ)]
− w2 = .
σ −1 2σ Nφ(σ N − μH )
To be stable, the above expression should be non-positive, which is equivalent to
N σ − H (2η − 1)μ σ − μ(2η − 1)
φ = . (19)
H (2η − 1)μ + N σ σ + μ(2η − 1)
316 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324

Other kinds switching and moving are also prevented under (19). Therefore, the separating dis-
tribution is a stable equilibrium when τ is small enough. Note that the RHS of (19) is positive
since σ > 1 > μ.
Three remarks are given here.
First, (19) is the complement of (18). Therefore, we obtain a continuous path with respect
to decreasing τ . There is no drastic change of distribution patterns in the case of heterogeneous
preferences.
Second, the RHS of (19) is related to η (actually, it decreases with η). Therefore, for more
heterogeneous preferences, the full separation distribution is more likely. In other words, the less
heterogeneous the workers are, the less likely industries are to agglomerate. Particularly, when
η = 1/2, the RHS is 1, and the above expression is always false so that the industries do not
separate. This shows that, when the workers are homogeneous, the distribution of two separating
industries is not a stable equilibrium. This is consistent with the results in the research of multiple
industries in Tabuchi and Thisse [25] and Zeng [29].
Third, the whole manufacturing sector appears dispersing in the full separation distribution.
However, it is different from the full dispersing distribution because each type of industry is
agglomerated. This verifies that the difference between the re-dispersion (when τ is small) and
the dispersion (when τ ) is large, as shown in [29].
The above results are summarized as follows.

Proposition 2. (i) Separating workers and dispersing industries is stable if τ is large enough to
satisfy (18). (ii) Full separation is stable if τ is small enough so that (18) fails.

3.2.2. Full agglomeration

Although the dispersing equilibrium in the standard core-periphery model is replaced by the
two equilibria of Proposition 2, the full agglomeration is again a possible distribution in this new
model. Without loss of generality, we examine the distribution of agglomeration in region 1:
y∗
λx∗ = λy∗ = 1, nx∗ 1 = n1 = N = N = N . After solving equations w1 = βx1 /(σ − 1) =
x y

βy1 /(σ − 1), (10) and (11), we obtain

 
∗ μL ∗ μL 1 + φ 2 φμ
w1 = , w2 = + .
(σ − μ)H σH 2φ σ −μ
To prevent the migration of type x workers from region 1 to region 2, it should hold that
w1∗ w2∗
x y − x y  0,
(P1 )μη (P1 )μ(1−η) (P2 )μη (P2 )μ(1−η)
or, equivalently,
μ σ +μ 2 σ −1−μ
1− + φ − 2φ σ −1  0. (20)
σ σ
It is easy to find that (20) is sufficient and necessary to prevent all possible migration and ensures
the stability of this corner equilibrium.
It is noteworthy that the LHS in the above expression is the same as (25) of Forslid and Otta-
viano [8], which is not related to η. This implies that the distribution of agglomerated industries
and residents does not depend on the preference heterogeneity. It can be shown that the LHS of
(20) increases with φ if σ  1 + μ. In this case, (20) is always true and the economy always
 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324 317

Fig. 1. Evolving paths. FA = full agglomeration, FS = full separation.

agglomerates in a region. Therefore, Forslid and Ottaviano [8] impose the no-black-hole condi-
tion3 σ > 1 + μ. Under the no-black-hole condition, (20) is true when φ is large enough (or, τ is
small enough). For example, when σ = 5, μ = 0.3, L = 10, H = 4, (20) is true iff φ > 0.762.
To summarize, we obtain two evolving paths. One involves the segregation of workers for
all τ , while the other involves full agglomeration for small τ . See Fig. 1.
Under the no-black hole condition, there are two possible stable equilibria for small τ : full
separation of both workers and industries (FS) and full agglomeration (FA). For a comparison,
Appendix C provides a utility calculation for all workers in the FS and FA equilibria. The results
are as follows

Unskilled workers Core of FA>FS>periphery of FA

Skilled workers FA>FS

If the social welfare function is the total sum of all the utilities of individuals, then it can be
shown that the FA equilibrium is better than the FS equilibrium.
Although the full agglomeration is better than the full separation except for the unskilled
workers in the periphery region, the latter is more likely for two reasons. First, since the evolving
path of Proposition 2 is continuous, the FS is chosen by history [14] without exogenous force
from policy makers. Second, more importantly, the full agglomeration turns out to be unstable
if the agglomeration diseconomy due to intra-city congestion is included [23], and/or the trans-
port costs of the traditional goods are positive [17], and/or the non-market factors of residential
behavior are included [15,24]. In contrast, the full separation equilibrium remains stable even if
those dispersion forces are included. This is because the numbers of residents in two regions are
the same in the full separation equilibrium.

4. Conclusion

Quigley [18] reviewed two fields, “urban agglomeration” and “neighborhood effects,” as the
prominent examples representing the renaissance in regional research. The former refers to
the NEG literature, and the latter refers to the research related to segregation. This paper is a

3 Note that our previous result of segregation does not depend on the no-black-hole condition.
318 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324

combination of both fields and provides an explanation of segregation by extending the foot-
loose entrepreneur model of [8] to incorporate the heterogeneity of workers. We show that the
increasing-returns technology and monopolistic competition form a mechanism endogenously
leading to persistent residential segregation. Specifically, there is a continuous path in which two
types of mobile workers are segregated and two industries evolve from dispersion to agglom-
eration separately with decreasing transport costs. Although the full agglomeration pattern is
possibly stable for small transport costs offers higher utilities for all except the unskilled work-
ers in the periphery region than the full separation equilibrium, the full agglomeration pattern is
unlikely to appear without exogenous force from a policy maker.
Despite the importance of heterogeneous preferences, research on it is almost limited to the
literature of the Tiebout theory, in which preferences with respect to local public goods are stud-
ied. The heterogeneous preference for private goods is hardly discussed in economic literature.
Therefore, this paper fills the theoretical gap.
Needless to say, preference heterogeneity for private goods is not a unique non-Tiebout in-
centive of segregation. For example, the heterogeneity of productivity is also one of the reasons.
By use of a multiple-industry model, in which the production in each industry requires a specific
skill owned by a specific type of worker, Zeng [29] shows that at most one industry disperses and
other industries agglomerate in a region for sufficiently small transport costs. Since the whole
manufacturing sector is dispersing, skill segregation is observed. Segregation in the real world is
much more complicated, and it is fair to say that more factors will need to be examined to fully
clarify segregation.

Acknowledgments

The author is indebted to the editor (Jan Brueckner), two anonymous referees, and to A. Kajii,
Y. Kanemoto, T. Mori, Y. Murata, R. Nakamura, K. Sasaki, Y. Sato, T. Tabuchi, H. Takatsuka and
X. Zhu for helpful suggestions and discussions. Further acknowledgment is due for the financial
support received from the Japanese Ministry of Education, Culture, Sports, Science, and Tech-
nology (Grand-in-Aid for Science Research 18530179), from Kagawa University (Exploratory
Research), and from Zhejiang University (CRPE Joint Research).

Appendix A. The instability of the interior equilibrium

Given λx and λy , it holds that

y y
nx2 = N x − nx1 , n1 = (λx + λy )H − nx1 , n2 = nx1 + (2 − λx − λy )H − N x ,

which imply that

y y
∂nx2 ∂N x ∂nx1 ∂n1 ∂nx1 ∂n2 ∂nx1 ∂N x
= − , = H − , = − H − .
∂λx ∂λx ∂λx ∂λx ∂λx ∂λx ∂λx ∂λx
Because w1 , w2 , nx1 , N x are obtained from equations
(σ − 1)w1 (σ − 1)w2
= x1 = y1 , = x2 = y2 ,
β β
we have the following equations for ∂w1 /∂λx , ∂w2 /∂λx , ∂nx1 /∂λx and ∂N x /∂λx :
 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324 319

 
∂w1 1 ∂w1 x

σ x = x H μ ηλ + (1 − η)λ + w1 H μη y
∂λ (n1 + φnx2 )2 ∂λx
 
 x  ∂nx1 ∂N x
× n1 + φn2 − E1 (1 − φ) x + φ x
x x
∂λ ∂λ
 
φ ∂w2

+ H μ (1 − λ )η + (1 − λ )(1 − η) − w2 H μη
x y
(φnx1 + nx2 )2 ∂λx
 
  ∂nx ∂N x
× φnx1 + nx2 − E2x (φ − 1) 1x + ,
∂λ ∂λx
 
∂w1 1 ∂w1

σ x = y y H μ (1 − η)λ + ηλ + w1 H μ(1 − η)
x y
∂λ (n1 + φn2 )2 ∂λx
   
 y y y ∂nx ∂N x
× n1 + φn2 − E1 (1 − φ) H − 1x − φ x
∂λ ∂λ

φ ∂w2

+ y y 2 H μ (1 − λx )(1 − η) + (1 − λy )η
(φn1 + n2 ) ∂λ x
    
 y y y ∂nx1 ∂N x
− w2 H μ(1 − η) φn1 + n2 − E2 (φ − 1) H − x − ,
∂λ ∂λx
 
∂w2 1 ∂w2

σ x = x H μ (1 − λ x
)η + (1 − λ y
)(1 − η) − w 2 H μη
∂λ (n2 + φnx1 )2 ∂λx
 
  ∂nx ∂N x
× nx2 + φnx1 − E2x (φ − 1) 1x +
∂λ ∂λx
 
φ ∂w1 x

+ x H μ λ η + λ y
(1 − η) + w 1 H μη
(n1 + φnx2 )2 ∂λx
 
  ∂nx ∂N x
× nx1 + φnx2 − E1x (1 − φ) 1x + φ x ,
∂λ ∂λ

∂w2 1 ∂w2

σ x = y y 2 H μ (1 − λx )(1 − η) + (1 − λy )η
∂λ (n2 + φn1 ) ∂λ x
   x  
 y y y ∂n1 ∂N x
− w2 H μ(1 − η) n2 + φn1 − E2 (1 − φ) −H −
∂λx ∂λx
 
φ ∂w1 x

+ y y H μ λ (1 − η) + λ η + w1 H μ(1 − η)
y
(n1 + φn2 )2 ∂λx
   
 y y y ∂nx ∂N x
× n1 + φn2 − E1 (1 − φ) H − 1x − φ x .
∂λ ∂λ
Solving the above four equations, we obtain
 σ φ(3−4λ) 
∂w1 Lμ(1 − φ) (σ +μ)φ−(σ −μ) − (1 − λ) (1 − φ)
2
=  2 ,
∂λx σφ
4H (σ − μ) (σ +μ)φ−(σ + λ(λ − 1)(1 − φ)
−μ)
 σ φ(1−4λ) 
∂w2 Lμ(1 − φ) (σ +μ)φ−(σ −μ) + λ2 (1 − φ)
=  2 ,
∂λx 4H (σ − μ) σφ
+ λ(λ − 1)(1 − φ)
(σ +μ)φ−(σ −μ)
320 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324

∂nx1 H [σ (1 − φ) + (2η − 1)μ(1 + φ)]

=
∂λ x 2σ (1 − φ)
 2μφ 
H (2η − 1)(2λ − 1)μφ λ(1 + φ) − (σ +μ)φ−(σ −μ)
+   ,
(σ 2 −μ2 )φ
2σ (1 − φ 2 ) [(σ +μ)φ−(σ −μ)]2
+ λ(1 − λ)
∂N x H (2η − 1)(2λ − 1)μφ[(σ − μ)2 − φ 2 (σ + μ)2 ]
=
∂λx 2σ (1 + φ){(σ 2 − μ2 )φ + λ(1 − λ)[(σ + μ)φ − (σ − μ)]2 }
at equilibrium λx = λy = λ, where λ is a solution of (15).
Similarly, we can obtain the derivatives with respect to λy . Their values at the symmetric
equilibrium λ = 1/2 are summarized as follows
∂w1 Lμ(1 − φ)[(σ + μ)φ − (σ − μ)] ∂w1 ∂w2 ∂w2 ∂w1
= = y, = y =− x ,
∂λx H (σ − μ)(1 + φ)[(σ + μ)φ + σ − μ] ∂λ ∂λx ∂λ ∂λ
y y
∂nx1 H [σ (1 − φ) + (2η − 1)μ(1 + φ)] ∂n1 ∂nx2 ∂n2 ∂nx1
= = y, = = − ,
∂λ x 2σ (1 − φ) ∂λ ∂λx ∂λy ∂λx
y y
∂nx1 ∂n1 ∂nx1 ∂nx2 ∂n2 ∂nx1 ∂N x
= = H − , = = − H, = 0.
∂λy ∂λx ∂λx ∂λy ∂λx ∂λx ∂λx
On the other hand, price indices are
βσ  x  1 y βσ  y y 1
Pix = ni + φnxj 1−σ , Pi = ni + φnj 1−σ , i, j = 1, 2, j = i.
σ −1 σ −1
Then,
y y  
∂P1x 2(1 − φ)P1x ∂nx1 ∂P1 2(1 − φ)P1 ∂nx1
= , = H −
∂λ x (1 − σ )(1 + φ)H ∂λx ∂λx (1 − σ )(1 + φ)H ∂λx
hold at the equilibrium, and furthermore,
 
∂V x ∂ w1 w2
= −
∂λx ∂λx (P1x )μη (P1y )μ(1−η) (P2x )μη (P2y )μ(1−η)
 1 2(1−φ)μw1 ∂n1
x ∂nx1

2 ∂w
∂λx + (σ −1)(1+φ)H η ∂λx + (1 − η)(H − ∂λx )
= y
(P1x )μη (P1 )μ(1−η)
2Lμ(P1x )−μη (P1 )μ(η−1)
y
=− δ xx ,
H (σ − μ)(σ − 1)σ (1 + φ)[(σ + μ)φ + σ − μ]
where

δ xx = (σ − 1)σ 2 (1 − φ)2 − μ(1 − φ 2 ) σ (2σ − 1) − μ2 (2η − 1)2

− 4μ2 σ φ + η(η − 1)(1 + φ)2

= φ 2 (μ + σ ) (σ + μ − 1)σ − (2η − 1)2 μ2

− 2σ φ σ (σ − 1) + 2μ2 (2η2 − 2η + 1)

+ (σ − μ) (σ − μ − 1)σ − (2η − 1)2 μ2 . (A.1)

Similarly,
 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324 321

 
∂V x ∂ w1 w2
= y −
∂λy ∂λ (P1x )μη (P1y )μ(1−η) (P2x )μη (P2y )μ(1−η)
∂V x 2 2L(2η − 1)2 μ3
= − x μη y μ(1−η) ,
∂λ x (P1 ) (P1 ) H (σ − μ)(σ − 1)σ
   
∂V y ∂ w1 w2 ∂V x P1x μ(2η−1)
= − = ,
∂λx ∂λx (P1x )μ(1−η) (P1y )μη (P2x )μ(1−η) (P2y )μη ∂λy P1y
   
∂V y ∂ w1 w2 ∂V x P1x μ(2η−1)
= − = .
∂λy ∂λy (P1x )μ(1−η) (P1y )μη (P2x )μ(1−η) (P2y )μη ∂λx P1y
Therefore, the symmetrical equilibrium is stable if and only if
δ xx > 0, (A.2)
 ∂V x ∂V x 
 ∂λx  16L2 (2η − 1)2 μ4 (1 − φ)[μ2 + (σ − 1)σ + μ(2σ − 1)]
 ∂λy =
 ∂V y ∂V y  H 2 (σ − μ)2 (σ − 1)2 σ (1 + φ)[(σ + μ)φ + σ − μ]
∂λx ∂λy
 
μ + μ2 − 2μσ + (σ − 1)σ 1
× φ− 2 > 0. (A.3)
μ + (σ − 1)σ + μ(2σ − 1) (P1 P1y )μ
x

However, inequalities (A.2) and (A.3) are inconsistent for any φ ∈ [0, 1]. In fact, let
μ + μ2 − 2μσ + (σ − 1)σ
φ0 = ,
μ2 + (σ − 1)σ + μ(2σ − 1)
which is smaller than 1. Then (A.3) is true if and only if φ ∈ (φ0 , 1]. On the other hand, from
(A.1) we know that δ xx is negative at φ = 1 and φ0 :
δ xx |φ=1 = −4(2η − 1)2 μ2 σ < 0,
4μ2 (1 − 2η)2 (σ − 1)(σ − μ)[μ2 + σ (σ − 1)]
δ xx |φ=φ0 = − < 0.
(σ + μ)(σ + μ − 1)2
Since the coefficient of φ 2 in (A.1) is positive, δ xx < 0 holds for all φ ∈ [φ0 , 1].

Appendix B. No stable equilibrium in which only one type of skilled workers disperses

Without loss of generality, we consider a population distribution (λx , 0), in which λx H ∈

(0, H ) workers of type x live in region 1 while all the remaining workers live in region 2. The
equilibrium conditions of the migration dynamics are
w1 w2
x μη y μ(1−η) = x y , (B.1)
(P1 ) (P1 ) (P2 ) (P2 )μ(1−η)
μη
w1 w2
x μ(1−η) y μη  x μ(1−η) y .
(P1 ) (P1 ) (P2 ) (P2 )μη
The expenditures are specified as follows
μ y μ
E1x = L + w1 λx H μη, E1 = L + w1 λx H μ(1 − η),
2 2
μ
E2 = L + w2 (1 − λ )H μ(1 − η) + w2 H μη,
x x
2
y μ
E2 = L + w2 (1 − λx )H μ(1 − η) + w2 H μη.
2
322 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324

Let γ x = nx1 /nx2 . Then

y
n1 (1 + γ x )λx H − γ x N x
γ =
y
y = . (B.2)
n2 (1 + γ x )(2 − λx )H − N x
We have the following two possible cases.
Case 1. Both industries locate in two regions.
In this case, wi = βxi /(σ − 1) = βyi /(σ − 1) hold for i = 1, 2 from (7) and (8). They are
written as
E1x φE2x
σ w1 = x + ,
[γ + φ(1 − γ x )]N x [(1 − γ x ) + φγ x ]N x
E2x φE1x
σ w2 = + x ,
[(1 − γ ) + φγ ]N
x x x [γ + φ(1 − γ x )]N x
y y (B.3)
E1 φE2
σ w1 = y + ,
[γ + φ(1 − γ y )](2H − N x ) [(1 − γ y ) + φγ y ](2H − N x )
y y
E2 φE1
σ w2 = + .
[(1 − γ y ) + φγ y ](2H − N x ) [γ y + φ(1 − γ y )](2H − N x )
Then we have six equations (B.1), (B.2) and (B.3) for five variables γ x , γ y , N x , w 1 and w 2 .
Furthermore, for λx ∈ (0, 1), the six equations are independently. Therefore, there is no solution
satisfying all equations and such an equilibrium does not exist.
Case 2. Only industry x locates in two regions.
y
In this case, we have nx1 = λx H , nx2 = N x − λx H , ny = 0 and n2 = 2H − N x = N y . We have
the following four independent equations:
βx1 βx2 βy2
w1 = , w2 = = , and (B.1),
σ −1 σ −1 σ −1
but three variables N x , w1 and w2 only. So there is no solution again.

Appendix C. The calculation of welfare

C.1. Full agglomeration

In the distribution of full agglomeration in region 1, we have λ1 = λ2 = 1, nx1 = nx2 = H .

The prices of industrial goods are
y βσ y τβσ
x
p11 (s) = p11 (s) = , x
p12 (s) = p12 (s) = ,
σ −1 σ −1
and the price indices are
y βσ 1 y βσ 1
P1x = P1 = H 1−σ , P2x = P2 = (φH ) 1−σ .
σ −1 σ −1
The skilled workers are all in region 1 with wage w1 = μL/(σ − μ)H . The consumption is given
as
μσ −1 1 μσ −1 1
X10 = Y10 = H σ −1 , X20 = Y20 = H σ −1 ,
2 βσ 2 τβσ
y μ2 ηL σ − 1 1 y μ2 (1 − η)L σ − 1 1
X1x = Y1 = H σ −1 , X1 = Y1x = H σ −1 ,
(σ − μ)H βσ (σ − μ)H βσ
 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324 323

y μ(1 − μ)L
A01 = A02 = (1 − μ), Ax1 = A1 = .
(σ − μ)H
Then the utility levels are
 μ  μ
μσ −1 1 μσ −1 1
U10 = H σ −1 (1 − μ)1−μ , U20 = H σ −1 (1 − μ)1−μ τ −μ ,
2 βσ 2 βσ
 1−μ  μ
y μ(1 − μ)L μ L σ −1
2 1
U1 = U1 =
x
H σ −1 ημη (1 − η)μ(1−η) .
(σ − μ)H (σ − μ)H βσ
The social welfare is the total sum of the utilities
 μ
σ −1 1
WFA = 2μ (1 − μ)
μ 1−μ
L H σ −1
βσ
 
−1−μ −μ μ
× 2 (1 + τ ) + η (1 − η)
μη μ(1−η)
.
σ −μ

C.2. Full separation

Consider the distribution that two types of skilled workers and two types of industries are
y
separated in regions 1 and 2: λ1 = 1, λ2 = 0, nx1 = N x = n2 = N y = H .
The prices of industrial goods are
y βσ y τβσ
x
p11 (s) = p22 (s) = , x
p12 (s) = p21 (s) = ,
σ −1 σ −1
and the price indices are
y βσ 1 y βσ 1
P1x = P2 = H 1−σ , P1 = P2x = (φH ) 1−σ .
σ −1 σ −1
The wages are given by w1 = w2 = μL/(σ − μ)H , and the consumption is
μσ −1 1 μσ −1 1
X10 = Y20 = H σ −1 , X20 = Y10 = H σ −1 ,
2 βσ 2 τβσ
y μ ηL σ − 1
2 1 y μ2 (1 − η)L σ − 1 1
X1x = Y2 = H σ −1 , X2 = Y1x = H σ −1 ,
(σ − μ)H βσ (σ − μ)H τβσ
y μ(1 − μ)L
A01 = A02 = (1 − μ), Ax1 = A2 = .
(σ − μ)H
Then the utility levels are
 μ
μσ −1 1 μ
U1 = U 2 =
0 0
H σ −1 (1 − μ)1−μ τ − 2 ,
2 βσ
   μ
μ(1 − μ)L 1−μ μ2 L σ − 1 1
H σ −1 ημη (1 − η)μ(1−η) τ −μ(1−η) .
y
U1x = U2 =
(σ − μ)H (σ − μ)H βσ
The social welfare is the total sum of the utilities
 μ
σ −1 1
WFS = 2μμ (1 − μ)1−μ L H σ −1
βσ
 
μ μ
× 2−μ τ − 2 + ημη (1 − η)μ(1−η) τ −μ(1−η) .
σ −μ
324 D.-Z. Zeng / Journal of Urban Economics 63 (2008) 306–324

References

[1] M. Amiti, C.A. Pissarides, Trade and industrial location with heterogeneous labor, Journal of International Eco-
nomics 67 (2005) 392–412.
[2] E. Berglas, Distribution of tastes and skills and the provision of local public goods, Journal of Public Economics 6
(1976) 409–423.
[3] J.K. Brueckner, Tastes, skills, and local public goods, Journal of Urban Economics 35 (1994) 201–220.
[4] J.M. Buchanan, An economic theory of clubs, Economica 32 (1965) (1994) 1–14.
[5] J.P. Conley, M.H. Wooders, Tiebout economies with differential genetic types and endogenously chosen crowding
characteristics, Journal of Economic Theory 98 (2001) 261–294.
[6] D.M. Cutler, E.L. Glaeser, J.L. Vigdor, The rise and decline of the American ghetto, Journal of Political Econ-
omy 107 (1999) 455–506.
[7] C.J. Dawkins, Tiebout choice and residential segregation by race in US metropolitan areas, 1980–2000, Regional
Science and Urban Economics 35 (2005) 734–755.
[8] R. Forslid, G.I.P. Ottaviano, An analytically solvable core–periphery model, Journal of Economic Geography 3
(2003) 229–240.
[9] M. Fujita, P.R. Krugman, A.J. Venables, The Spatial Economy: Cities, Regions and International Trade, MIT Press,
Cambridge, MA, 1999.
[10] M. Fujita, T. Mori, Frontiers of the new economic geography, Papers in Regional Science 84 (2005) 377–405.
[11] A. Glazer, M. Gradstein, P. Ranjan, Consumption variety and urban agglomeration, Regional Science and Urban
Economics 33 (2003) 653–661.
[12] P. Krugman, Scale economics, product differentiation, and the pattern of trade, American Economic Review 70
(1980) 950–959.
[13] P.R. Krugman, Increasing returns and economic geography, Journal of Political Economy 99 (1991) 483–499.
[14] P.R. Krugman, History vs. expectations, Quarterly Journal of Economics 106 (1991) 651–667.
[15] Y. Murata, Product diversity, taste heterogeneity, and geographic distribution of economic activities: Market vs.
non-market interactions, Journal of Urban Economics 53 (2003) 126–144.
[16] G.I.P. Ottaviano, Monopolistic competition, trade, and endogenous spatial fluctuations, Regional Science and Urban
Economics 31 (2001) 51–77.
[17] P. Picard, D.-Z. Zeng, Agricultural sector and industrial agglomeration, Journal of Development Economics 77
(2005) 75–106.
[18] J.M. Quigley, The renaissance in regional research, The Annals of Regional Science 35 (2001) 167–178.
[19] P.W. Rhode, K.S. Strumpf, Assessing the importance of Tiebout sorting: Local heterogeneity from 1850 to 1990,
American Economic Review 93 (2003) 1648–1677.
[20] S. Scotchmer, Public goods and the invisible hand, in: J. Quigley, E. Smolensky (Eds.), Modern Public Finance,
Harvard Univ. Press, Cambridge, MA, 1994, pp. 93–119.
[21] S. Scotchmer, Local public goods and clubs, in: A.J. Auerbach, M. Feldstein (Eds.), Handbook of Public Economics,
vol. 4, Elsevier Science B.V., Amsterdam, 2002, pp. 1997–2042.
[22] J. Scott, Rethinking segregation beyond black and white, New York Times, July 29, 2001.
[23] T. Tabuchi, Agglomeration and dispersion: A synthesis of Alonso and Krugman, Journal of Urban Economics 44
(1998) 333–351.
[24] T. Tabuchi, J.-F. Thisse, Taste heterogeneity, labor mobility and economic geography, Journal of Development
Economics 69 (2002) 155–177.
[25] T. Tabuchi, J.-F. Thisse, Regional specialization, urban hierarchy, and commuting costs, International Economic
Review 47 (2006) 1295–1317.
[26] T. Tabuchi, D.-Z. Zeng, Stability of spatial equilibrium, Journal of Regional Science 44 (2004) 641–660.
[27] C. Tiebout, A pure theory of local expenditures, Journal of Political Economy 64 (1956) 416–424.
[28] D.-Z. Zeng, Equilibrium stability for a migration model, Regional Science and Urban Economics 32 (2002) 123–
138.
[29] D.-Z. Zeng, Re-dispersion is different from dispersion: Spatial economy of multiple industries, The Annals of
Regional Science 40 (2006) 229–247.