The core-periphery model with three regions and more

Abstract. We determine the properties of the core-periphery model with 3 regions and
compare our results with those of the standard 2-region model. The conditions for the
stability of dispersion and concentration are established. Like in the 2-region model,
dispersion and concentration can be simultaneously stable. We show that the 3-region
(resp. 2-region) model favours the concentration (resp. dispersion) of economic activity.
Furthermore, we provide some results for the n-region model. We show that the stability
of concentration of the 2-region model implies that of any model with an even number of
regions.

Keywords: new economic geography, core-periphery

JEL Classification Numbers: R12, R23

2
1 Introduction

The New Economic Geography literature has emerged from the long-existing need to
explain the spatial concentration of economic activity. The literature in the field provides
a general equilibrium framework addressing the emergence of economic agglomerations as
the result of a trade-off between increasing returns at the firm level and transportation
costs related to the shipment of goods.

In this paper, we consider a standard New Economic Geography model involving n regions
distributed along a circle. This model corresponds to the racetrack economy as studied
by Fujita et al. (1999) and can be viewed as the extension of the core-periphery model
of Krugman (1991) to the case of a spatial economy with n regions. Like in Krugman’s
original work, there are two sectors in the economy. While the agricultural sector employs
farmers and produces a single homogeneous good under constant returns to scale, the
manufacturing sector employs workers and produces differentiated goods which —unlike
the agricultural good— are costly to transport across regions.

In the case of the core-periphery model with 2 regions, the existence and uniqueness
of short-run equilibrium have been established by Mossay (2006). Also, the number
and stability of long-run equilibria have been determined by Robert-Nicoud (2005). If
transportation costs are low, all the industrial activity locates in one region (concentration
equilibrium). On the other hand, if transportation costs are high, the industrial activity
gets dispersed equally across regions (dispersion equilibrium).

As stressed by Fujita et al. (1999), a theoretical analysis of economic geography must get
beyond the 2-location framework. Interesting results regarding the size and number of
agglomerations in multi-location models have been obtained by Tabuchi et al. (2005) and
by Picard and Tabuchi (2010) in the context of quadratic preferences. However, except
for the work of Puga (1999), who restricted his analysis to the case of a finite number of
equidistant regions like in Tabuchi et al. (2005), no analytical result regarding the original
Krugman core-periphery model involving 3 regions or more has been derived so far. The
existing analysis of the multi-region core-periphery model relies on numerical simulations

3
exclusively, see Krugman (1993), Fujita et al. (1999), Brakman et al. (2001), and Ago et
al. (2006). While these simulations are very helpful in suggesting some possible outcomes
of the model, they provide little indication about whether some outcome is likely to remain
sustainable when the number of regions increases. The aim of this paper is to contribute
to fill this gap by providing analytical stability conditions and some insight regarding the
dependence of the original Krugman core-periphery model on the number of regions.

First we study the 3-region model. The numerical simulations in Fujita et al. (1999)
suggest that only two kinds of spatial equilibria can emerge: the dispersion configuration,
where the economic activity gets equally distributed across the 3 regions; and the con-
centration configuration, where the economic activity agglomerates in a single region. We
establish the conditions for the stability of the dispersion and concentration equilibria. As
expected and already suggested by the standard core-periphery model, high (resp. low)
transport costs favour the stability of the dispersion (resp. concentration) configuration.
We prove the existence of a region in the parameter space where the dispersion and con-
centration configurations are simultaneously stable. This result generalizes the overlap
interval determined in the case of the standard core-periphery model by Robert-Nicoud
(2005). By comparing the results of the 2- and 3-region models, we show that the 2-region
(resp. 3-region) model favours the dispersion (resp. concentration) of economic activity.

Second, we obtain further results regarding the n-region model. We provide a simple
sufficient condition for the stability of the concentration equilibrium, and show that the
stability of concentration of the 2-region model implies that of any model with an even
number of regions.

The main difficulties in studying the multi-region core-periphery model are twofold. Un-
like some other New Economic Geography models (e.g., Ottaviano et al. (2002)), the
original Krugman core-periphery model is not solvable due to high nonlinearities, so that
the short-run equilibrium of the model can only be determined by a set of implicit equa-
tions. Morever, as the number of regions of the model increases, the nature and complexity
of spatial interactions between regions increase dramatically unless regions are assumed to
be equidistant (e.g., Puga (1999) or Tabuchi et al. (2005)). When the economy involves

4
3 regions only, these regions are still equidistant which explains the reason for which we
are able to derive analytical results in that case. When the number of regions increases,
the stability of the concentration configuration is easier to obtain than that of the disper-
sion outcome. This is because the number of potential destabilizing perturbations to be
considered is much smaller in the concentration case than in the dispersion one.

In Section 2 we describe the n-region core-periphery model and provide some general
results regarding the steady states and the dynamics of the model. We derive the stability
analysis of the various spatial configurations emerging in the 3-region model in Section 3.
In Section 4 we compare our results with those of the standard core-periphery model. The
equilibria emerging in the n-region model are studied in Section 5. Section 6 concludes.

2 The model

2.1 Economic environment

We consider a spatial economy with a finite number of regions, i ∈ {1, 2, ..., n}. Regions
are evenly distributed along a circle meaning that successive regions are equidistant, see
the racetrack economy in Krugman (1993) and Fujita et al. (1999). There are two sectors
in the economy: the manufacturing sector, which exhibits increasing returns to scale,
and the agricultural sector, which has constant returns. Agents at location i enjoy a
Cobb-Douglas utility from the two types of goods:

μ
Ui = CM (i)CA1−μ (i) , (1)

where CA is the consumption of the agricultural good and CM is the consumption of the
manufactured aggregate, defined by
" n Z # σ−1
σ
X v(j)
σ−1
CM (i) = cz (j, i) σ dz , (2)
j=1 0

where v(j) is the density of manufactured varieties available at location j, cz (j, i) is the
consumption of variety z produced at j, and σ > 1 is the elasticity of substitution among

5
manufactured varieties. From utility maximization, μ is the share of manufactured goods
in expenditure.

There are two types of agents: workers and farmers. We normalize the total population
of workers to 1, and denote the number of workers in each region i by λi ∈ [0, 1], with
Pn
i=1 λi = 1. The number of farmers at any location i is constant and denoted by A.

Farming is an activity that takes place under constant returns to scale. The agricultural
output is:
QA (i) = A. (3)

Manufacturing variety z involves a fixed cost and a constant marginal cost. The number
of workers employed in location i to produce QM,z (i) units of variety z is:

Lz (i) = α + βQM,z (i). (4)

Transport costs only affect manufactured goods and take Samuelson’s iceberg form. More
precisely, when the amount Z of some variety is shipped from locations j to i, then the
amount X of that variety which is effectively available at location i is given by:

X Ti,j = Z, (5)

where Ti,j ≥ 1 denotes the transportation cost from location i to j.

There is a continuum of manufacturing firms. Each of them produces a single variety,

and faces a demand curve with a constant elasticity σ. This will be confirmed below, see
relation (13). The optimal pricing behaviour of any firm at location i is therefore to set
the price pz (i) of variety z at a fixed markup over marginal cost:
σ
pz (i) = βWi , (6)
σ−1
where Wi is the worker wage rate prevailing in region i.

Firms are free to enter into the manufacturing sector, so that their profits are driven to
zero. Consequently, their output is given by:
α
QM,z (i) = (σ − 1). (7)
β

6
Since all varieties are produced at the same scale, the density v(i) of manufactured goods
produced at each location is proportional to the density λi of workers at that location:
Z v(i)
λi = Lz (i)dz = ασv(i). (8)
0

Total income Y at location i is given by:

Yi = A + λi Wi , (9)

where the price of the agricultural good has been normalized to 1.

Workers are not interested in nominal wages but rather in utility levels. To consume at
location i one unit of variety z produced at location j, Ti,j units must be shipped. The
delivery price is, therefore, pz (j) Ti,j .

The price index of the manufactured aggregate for consumers at location i, denoted by Θi ,
is obtained by computing the minimum cost of purchasing one unit of the manufactured
aggregate CM (i):
" n Z #− σ−1
1
X v(j)
−(σ−1)
Θi = pz (j)−(σ−1) Ti,j dz . (10)
j=1 0

By using the pricing rule (6) and relation (8), Θ(i) may be rewritten as:
" n #− σ−1
1

βσ 1 X −(σ−1) −(σ−1)
Θi = (ασ) σ−1 λj Wj Ti,j
σ−1 j=1

The demand for variety z ∈ [0, v(j)] produced at j may be expressed for workers and
farmers located at i as:
−(σ−1)
cw
z (j, i) = μWi pz (j)
−σ
Ti,j Θσ−1
i ;
−(σ−1)
caz (j, i) = μpz (j)−σ Ti,j Θσ−1
i . (11)

The total demand for variety z produced at j is then obtained by summing up the demand
for that variety of all the consumers in the spatial economy:
Xn
QDM,z (j) = [λi cw a
z (j, i) + Aci (j, i)]
i=1
X
n
−(σ−1)
= μ[λi Wi + A]pz (j)−σ Ti,j Θσ−1
i . (12)
i=1

7
By using the total income expression (9), we get:
X
n
−(σ−1)
QD
M,z (j) = μYi pz (j)−σ Ti,j Θσ−1
i . (13)
i=1

The market-clearing price for variety z produced at j is obtained by equating the demand
QD
M,z (13) and the supply QM,z (7) of that variety:

" # σ1
μβ X
n
−(σ−1)
pz (j) = Yi Θσ−1
i Ti,j . (14)
α(σ − 1) i=1

Because of the optimal pricing rule (6), we get:

∙ ¸ σ1 "Xn
# σ1
σ−1 μβ −(σ−1)
Wj = Yi Θσ−1
i Ti,j .
βσ α(σ − 1) i=1

The manufacturing wage Wj is the wage prevailing at location j such that firms at j break
even.

The indirect utility Ui of a worker in location i is then obtained through expression (1):

μ
Ui = CM (Θi , Wi )CA1−μ (Θi , Wi )
= (μWi /Θi )μ [(1 − μ)Wi ]1−μ
= μμ (1 − μ)1−μ Θ−μ
i Wi . (15)

The adjustment dynamics postulates that workers migrate away from low-utility regions
toward high-utility regions, see Fujita et al. (1999)
·
λi = k(Ui − U)λi , (16)

where k denotes the adjustment speed and U denotes the average utility:
X
n
U= λi Ui .
i=1

In the short-run, each region i is described by the variables Yi , Θi , Wi , and Ui which denote
respectively the income level, the manufacturing price index, the nominal wage, and the

8
indirect utility level. Economic normalization leads to the following reduced system of
equations, see Fujita et al. (1999) or Mossay (2005):

Yi = (1 − μ)/n + μ λi Wi
" n #− σ−1
1
X
Θi = λj (Wj Ti,j )−(σ−1)
j=1
" n µ ¶σ−1 # σ1
X Θj
Wi = Yj
j=1
Ti,j
Ui = Θ−μ
i Wi
·
λi = (Ui − Ū)λi , i = 1, ..., n (17)

2.2 Equilibria and invariant

A simple symmetry argument establishes the existence of the dispersion and concentration
equilibria.

Lemma 1 The configurations of dispersion, ( n1 , n1 , ..., n1 ), and concentration, (1, 0, ..., 0)

and its permutations are equilibria.

Proof. This is obtained by direct substitution in the system of differential equations de-
scribing the dynamics (17).

Pn
The (n-1)-dimensional simplex is defined by {(λ1 , ..., λn ) ∈ Rn : i=1 λi = 1, λi ≥ 0,
i = 1, ..., n}. The boundary of the simplex corresponds to a distribution of workers that
leaves at least one of the regions empty; that is, on the boundary of the simplex, there
is some region i for which λi = 0. Because of the assumed dynamics (16), if a region is
initially deserted, then it will remain so over time unless there is some exogenous migration
to that region. This can be restated in the following Lemma.

Lemma 2 The boundary of the simplex is invariant for the dynamics.

Proof. See Appendix A.

9
3 The 3-region core-periphery model

In this Section, the spatial economy consists of 3 identical regions which are equally spaced
along a circle. The distance between any two regions is equal and the corresponding
transportation cost is denoted by T .

The existing literature has provided numerical simulations of this core-periphery model.
They suggest that only two kinds of outcomes can emerge: the dispersion configuration,
where the economic activity gets equally distributed across the 3 regions; and the concen-
tration configuration, where the economic activity agglomerates in a single region, see e.g.
Fujita et al. (1999). Our purpose is to support these numerical results, by providing an-
alytical results. We make clear the conditions under which dispersion and concentration
occur. In particular, we show that these two configurations can coexist in equilibrium
and determine the region in the parameter space for which this actually happens.

Lemma 3 The configurations ( 13 , 13 , 13 ), ( 12 , 12 , 0) and (0, 0, 1) are steady states of the

model.

Proof. Dispersion and concentration are equilibria by Lemma 1. The remaining result
is obtained by direct substitution in the system of differential equations describing the
dynamics (17).

Note that the dispersion equilibrium is fully symmetric. The remaining two equilibria
have partial symmetry: they are invariant by a reflection that swaps the first two regions
(coordinates). In other words, studying the stability of the above equilibria, provides the
stability of (1/2, 0, 1/2) and (0, 1/2, 1/2) from that of (1/2, 1/2, 0), and of (1, 0, 0) and
(0, 1, 0) from that of (0, 0, 1).

We provide conditions under which dispersion and concentration are stable. Our results
are obtained by studying the properties of the eigenvalues of the Jacobian matrix of
the dynamical system (17). We evaluate them at each of the above equilibria. As it is
usually assumed in the existing literature, we suppose that the “no-black-hole” condition,
μ < (σ − 1)/σ, holds, see Fujita et al. (1999).

10
Proposition 1 The dispersion configuration ( 13 , 13 , 13 ) is stable if and only if:
µ ¶ σ−1
1
σ − 1 + μ2 σ + 2μ(2σ − 1)
T > Td∗ = .
(1 − μ)[(1 − μ)σ − 1]

Proof. See Appendix B.

This result means that the dispersion configuration is stable for high values of the trans-
portation cost, as anticipated. Note that the “no-black-hole” condition guarantees that
the critical value, Td∗ , is positive. If the “no-black-hole” condition were to fail, then dis-
persion would be unstable regardless of the value of transportation cost T . This latter
scenario is not regarded as an interesting situation, see Fujita et al. (1999).

Proposition 2 The concentration configuration (0, 0, 1) is stable if and only if

£ ¤1
(1 + T σ−1 )(1 − μ) + (1 + 2μ)T 1−σ σ < 31/σ T μ .

Proof. See Appendix B.

A sufficient condition for the stability of concentration can be derived from the above
Proposition.

Corollary 1 The concentration configuration (0, 0, 1) is stable if

µ ¶ σ−1
1
1 + 2μ
T < Tc∗ = .
1−μ

Proof. See Appendix B.

This result means that the concentration configuration is stable for low values of the
transportation cost, as anticipated. It is important to stress that the above Corollary

11
provides a sufficient stability condition only, meaning that concentration is stable for a
wider range of parameter values, see Proposition 2.

We now address the possible coexistence of the above configurations.

Proposition 3 The concentration and dispersion configurations are simultaneously sta-

ble for an open subset in the parameter space (T , σ, μ).

Proof. See Appendix B.

This result proves the co-existence of concentration and dispersion, as illustrated so far by
numerical simulations, see Fujita et al. (1999, Chapter 6, Figure 6.3). The region in the
parameter space for which this co-existence of configurations actually occurs, lies between
the critical stability surfaces of the dispersion and concentration equilibria in the space
(T , σ, μ), see the representation of critical stability curves in the space (T , σ)
in Figure 1. This result contrasts with the finding by Ago et al. (2006) in the context of
a 3-region core-periphery model with asymmetric locations where such a co-existence of
equilibria is prevented due to the locational advantage of the central region. In contrast
to Ago et al. (2006), we provide analytical results regarding the stability analysis of
dispersion and concentration. This is possible because in our model regions are evenly
distributed along the circle (i.e., locations are symmetric).

12
 Figure 1: Critical stability curves in the space (T , σ) for μ =
0.6 (concentration: plain curve, dispersion: dashed curve).
Between the two curves, concentration and dispersion are
simultaneously stable.

The following result concerns the stability of the partial concentration configuration
(1/2, 1/2, 0).

Proposition 4 For μ > 2(σ−1)/(3σ), the partial concentration configuration (1/2, 1/2, 0)
is never stable.

Proof. See Appendix B.

This Proposition provides a sufficient condition for the instability of partial concentration.
This spatial distribution is likely to be unstable for a much larger set of parameters1 .

Following Fujita et al. (1999) and Robert-Nicoud (2005), we have determined the steady
states of the dynamic system (17) and studied their stability. An alternate strategy is to
study the stability of spatial equilibria like in Picard and Tabuchi (2010). The difference
1
Actually, numerical arguments suggest that the partial concentration configuration (1/2, 1/2, 0) is
never stable. Also, note that we have not been able to address the stability of the general
configuration ((1 − λ)/2, (1 − λ)/2, λ), 0 < λ < 1.

13
between the two approaches is that in a spatial equilibrium, workers do not have any
incentive to relocate to any other location given the equilibrium definition, while it is not
necessarily the case at a steady state. However, when studying stability, both approaches
are indeed equivalent; that is, there is a correspondence between stable steady states and
stable spatial equilibria.

4 Comparison between the equilibria of the 2- and

3-region models

In this Section, we compare the set of parameters for which concentration and dispersion
are simultaneously stable for the two models.

Lemma 4 The stability region of the dispersion equilibrium of the 2-region model con-
tains that of the 3-region model, meaning that the 2-region model favours dispersion.

Proof. See Appendix C.

In other words, dispersion in an economy with three regions implies dispersion in an

economy with only two regions (for the same parameter values). The reason for which
dispersion is less stable (e.g., the set of parameters (T , σ, μ) supporting a stable steady
state is smaller) in the 3-region economy than in the 2-region economy can be understood
in the following way. Consider workers and firms distributed uniformly across regions.
Now relocate some firm to another location. It must be the case that the relocated firm
has an incentive to remain in that location in the 3-region economy while it would not
necessarily have that incentive in the 2-region economy. This is because the relocated firm
can benefit from increasing returns in the new location while facing less local competition
with other firms in the 3-region model than in the 2-region model. This is because the
number of local competitors is 1/3 in the 3-region model while it is 1/2 in the 2-region
model.

14
Lemma 5 The stability region of the concentration equilibrium of the 3-region model
contains that of the 2-region model, meaning that the 3-region model favours con-
centration.

Proof. See Appendix C.

In other words, concentration in an economy with two regions implies concentration in

an economy with three regions. The reason for which concentration is less stable in the
2-region model than in the 3-region model is the following one. Consider workers and
firms concentrated in some location. Now relocate some firm to another location. It must
be the case that the relocated firm has an incentive to remain in the new location in the
2-region model while it would not necessarily have that incentive in the 3-region model.
This is because the relocated firm benefits from less competition in the new location and
supply locally half of the farmers in the 2-region model while it would only supply a third
of them in the 3-region model.

So far, when increasing the number of regions (from 2 to 3), we have been increasing the
size of the circumference. However, it is also interesting to consider an alternative scenario
where the perimeter of the circumference is kept fixed. To keep the transportation cost
of a full lap around the circle constant, say T , we should set the transportation cost
√ √
T of the 2-region economy to T , and that of the 3-region economy to 3 T . In this
alternate scenario, both Lemmas 4 and 5 still hold. This is because now the transportation
cost of the 3-region economy has been reduced with respect to the initial scenario. As
a consequence, the stability region of the dispersion equilibrium of the 3-region model
shrinks, while that of concentration expands, meaning that dispersion may no longer be
stable but concentration will surely remain stable.

We provide an additional interpretation of Lemma 5 in the case of an economy with

constant perimeter. In a concentration configuration, farmers are located closer to the
agglomeration of workers and firms in the 3-region model than what they would be in
the 2-region model. This fosters concentration in the 3-region model as compared to the

15
2-region model2 .

5 On the n-region model

The purpose of this Section is to provide an idea of how the stability of the dispersion and
concentration equilibria behaves for different values of n. In order to ease the comparison,
we consider n regions evenly distributed along a circumference with fixed perimeter as in
the racetrack economy studied by Fujita et al. (1999). The transportation cost along a
full lap around the circle is T = T n , where T remains the transportation cost between
two adjacent regions, and the transportation cost between regions i and j > i is Ti,j =
T d , where d = min{j − i, n − j + i}.

We provide the following characterization of the concentration equilibrium.

Proposition 5 In an economy with an even number of regions, if the condition (1 + μ)/2

T (1−σ−μσ)/2 + (1 − μ)/2 T (σ−1−μσ)/2 ≤ 1 holds, then the concentration equilibrium
(1, 0, . . . , 0) is stable.

Proof. See Appendix D.

Note that when n = 2, this sufficient condition turns out to be also necessary. This leads
to the following Corollary.

Corollary 2 If the concentration equilibrium is stable in the 2-region model, it will

remain stable in an economy with any even number of regions.

Our model constrasts with that of Picard and Tabuchi (2010) in many respects. First,
because of quadratic consumption preferences, the model of Picard and Tabuchi (2010) is
solvable; that is, the short-run equilibrium can be solved explicitly in terms of the spatial
2
This interpretation was suggested by a referee.

16
distribution of labor. In contrast, our model assumes a symmetric constant-elasticity-
of-substitution (CES) preference for differentiated varieties which makes the model not
solvable explicitly. This mainly explains the reason for which analytical results are difficult
to obtain for an arbitrary number of regions.

Second, unlike in Picard and Tabuchi, the set of regions is fixed so that workers do not
consider relocation to the hinterland as a migration possibility. This is likely to make
steady states more stable since the destabilizing perturbations to be considered have a
support constrained by the set of existing regions.

Despite of the differences between our model and that of Picard and Tabuchi, some of
our results are reminiscent of theirs (e.g., the possible co-existence of the symmetric
equilibrium with 3 regions and the full concentration equilibrium). However, while they
have obtained some stable configurations with 3 asymmetric regions, we have not been
able to identify any such stable asymmetric configuration in the context of the Krugman
setting.

6 Concluding remarks

We have provided results concerning the core-periphery model with more than two regions.
Our results regarding the 3-region model are analytical and complement those previously
obtained only by simulations in the existing literature (e.g., Fujita et al. (1999)). We
have compared the stable outcomes of the 2- and 3-region models and have established
that the 2-region model favours dispersion while the 3-region model favours concentration.
Furthermore, we have derived some results for the core-periphery model with more than
3 regions. The stability of concentration of the 2-region model implies that of any model
with an even number of regions.

17
7 References

Ago, T., Isono, I. and T. Tabuchi (2006), “Locational Disadvantage of the Hub”, Annals
of Regional Science 40 (4), 819-848.

Brakman, S., Garretsen, H. and C. van Marrewijk (2001), An Introduction to Geographical

Economics, Cambridge University Press.

Fujita, M., Krugman, P. and A. Venables (1999), The Spatial Economy: Cities, Regions
and International Trade, Cambridge, MA: MIT Press.

Krugman, P. (1991), “Increasing Returns and Economic Geography”, Journal of Political

Economy 99 (3), 483 — 499.

Krugman, P. (1993), “On the Number and Location of Cities”, European Economic Review
37, 293 — 298.

Mossay, P. (2005), “The Structure of NEG Models”, Chapter 5 in Contemporary Issues

in Urban and Regional Economics, L. Yee (ed.), Nova Publishers.

Mossay, P. (2006), “The core-periphery model: a note on the existence and uniqueness of
short-run equilibrium”, Journal of Urban Economics 59, 389 — 393.

Ottaviano, G., Tabuchi, T. and J.-F. Thisse (2002), "Agglomeration and Trade Revisited",
International Economic Review 43:2, 409-435.

Picard, P. and T. Tabuchi (2010), "Self-organized agglomerations and Transport costs",

Economic Theory, 42:3.

Puga, D. (1999), “The rise and fall of regional inequalities”, European Economic Review
43, 303 — 334.

Robert-Nicoud, F. (2005), “The structure of simple “new economic geography” models

(or, on identical twins)”, Journal of Economic Geography 5 (2), 201 — 234.

Tabuchi, T., Thisse, J.-F. and D. Zeng (2005), "On the Number and Size of Cities",
Journal of Economic Geography 5 (4), 423-448.

18
Appendix A

Proof of Lemma 2: The (n − 1)-dimensional simplex is such that

λ1 + λ2 + . . . + λn = 1,

and its boundary satisfies

λi = 0
λ1 + . . . + λi−1 + λi+1 + . . . + λn = 1.

Suppose that λi = 0. Then

λ̇1 = (U1 − Ū)λ1

..
.
λ̇i−1 = (Ui−1 − Ū )λi−1
λ̇i = 0
λ̇i+1 = (Ui+1 − Ū )λi+1
..
.
λ̇n = (Un − Ū)λn ,

and therefore, the boundary is invariant.

Appendix B

Proof of Proposition 1: First we determine the Jacobian matrix associated with the
dynamical system (17) in the following way. By inserting the income Yi and price index
Θi expressions into the wage equation, we get an equation determining wages W as an
implicit function of the worker distribution λ, that we denote k(W, λ) = 0. In a similar
way, the utilities Ui can be expressed in terms of W and λ; we write them as Ui (W, λ)
and define gi (W, λ) = [Ui (W, λ) − U (W, λ)]λi . By applying the implicit function theorem

19
to k(W, λ), we obtain ∂λ W = −[∂W k]−1 ∂λ k. Then the Jacobian matrix corresponds to
J = Dλ g = ∂W g ∂λ W + ∂λ g.

Second we evaluate J at the dispersion outcome (1/3, 1/3, 1/3) and evaluate its eigenval-
ues. We get one distinct eigenvalue

µ ¶ μ µ ¶
1 + 2T 1−σ σ−1 T σ − T T σ (1 − μ)[(1 − μ)σ − 1] − T (−1 + σ + μ2 σ − 2μ + 4μσ)
− · .
3 σ−1 T 2σ (1 − μ) + T 2 (1 + 2μ + 3σ) + T 1+σ (6σ − 2 − μ)
To ensure the stability of dispersion, the above eigenvalue should be negative,

T σ (1 − μ)[(1 − μ)σ − 1] − T (−1 + σ + μ2 σ − 2μ + 4μσ) > 0

⇔ T σ−1 (1 − μ)[(1 − μ)σ − 1] > −1 + σ + μ2 σ − 2μ + 4μσ

σ−1+μ2 σ+2μ(2σ−1)
⇔ T σ−1 > Td∗ = (1−μ)[(1−μ)σ−1]
,

where we used (μ − 1)[1 + (μ − 1)σ] > 0, given that the “no-black-hole” condition holds.

It remains to show that Td∗ > 1, otherwise dispersion would always be stable since T > 1.
We have
³ ´ 1
∗ σ−1+μ2 σ+2μ(2σ−1) σ−1
Td = (μ−1)(1+(μ−1)σ) >1

⇔ σ − 1 + μ2 σ + 2μ(2σ − 1) > (μ − 1)(1 + (μ − 1)σ)

⇔ σ − 1 + μ2 σ + 2μ(2σ − 1) > (μ − 1)(1 + (μ − 1)σ)

⇔ μ(6σ − 3) > 0,

which holds given that σ > 1 and μ ∈ (0, 1).

Proof of Proposition 2: We evaluate the Jacobian matrix as derived in the Proof of

Proposition 1 at the concentration outcome (1, 0, 0) and determine its eigenvalues. One
distinct eigenvalue is obtained
µ ¶1
−μ (1 + T σ−1 )(1 − μ) + (1 + 2μ)T 1−σ σ
−1 + T .
3
To ensure the stability of concentration, the above eigenvalue should be negative; that is,
£ ¤1
(1 + T σ−1 )(1 − μ) + (1 + 2μ)T 1−σ σ < 31/σ T μ . (18)

20
Proof of Corollary 1: Given that T > 1, we have T μ > 1 and 31/σ T μ > 31/σ . There-
fore, a sufficient condition for stability is that
£ ¤1
(1 + T σ−1 )(1 − μ) + (1 + 2μ)T 1−σ σ < 31/σ
⇔ (1 − μ)T σ−1 + (1 − μ) + (1 + 2μ)T 1−σ < 3
⇔ (1 − μ)T 2(σ−1) + (1 − μ)T σ−1 + 1 + 2μ < 3T σ−1
⇔ (1 − μ)T 2(σ−1) − (2 + μ)T σ−1 + 1 + 2μ < 0.

By replacing X = T σ−1 , we have

(1 − μ)X 2 − (2 + μ)X + 1 + 2μ < 0

⇔ X− < X < X+ ,

where X− and X+ are the roots of the polynomial in X, given by:

2 + μ ± 3μ
X± = .
2(1 − μ)

Since X− = 1, a sufficient condition for the stability of concentration is that

1 + 2μ
T σ−1 < Tc∗ = .
1−μ

Proof of Proposition 3: It turns out that we never have Td∗ < Tc∗ . This is because
the condition for stability of concentration is only sufficient and therefore, too strong.

We prove Proposition 3 by showing that for some values of T for which dispersion is
stable, the eigenvalue of the Jacobian matrix at the concentration equilibrium is negative.

By replacing T by ηTd∗ in the condition obtained in Proposition 2, where η > 1 is a real

number ensuring that we are considering values of T for which dispersion is stable, we
have
©£ ¤ ª1
1 + (ηTd∗ )σ−1 (1 − μ) + (1 + 2μ)(ηTd∗ )1−σ σ < 31/σ (ηTd∗ )μ ⇔
£ ¤
(ηTd∗ )σ−1 1 + (ηTd∗ )σ−1 (1 − μ) + 1 + 2μ < 3(ηTd∗ )μσ+σ−1 .

21
Both sides of the above inequality are continuous functions of parameters σ, μ and η. So,
if the inequality holds for a particular value of (σ, μ, η), it will also hold in an open set
containing that particular value.

Choose σ = 2. The no-black-hole condition then requires that μ < 1/2. Choose μ = 1/3
and η = 2. It is trivial to check that the inequality holds for these parameter values.

Proof of Proposition 4: We evaluate the Jacobian matrix as derived in the Proof

of Proposition 1 at the partial concentration outcome (1/2, 1/2, 0) and determine its
eigenvalues. The following two distinct eigenvalues are obtained
µ μ µ
¶ σ−1 ¶
1 + T 1−σ Tσ − T
·
2 σ−1

T (2 + 4μ − 7μσ − 2σ − 3σμ2 ) + T σ (1 − μ)[2(σ − 1) − 3μσ]

·
T 2 [−2 − 4μ − σ(1 − μ)] − T 2σ (1 − μ)(2 + σ) − 2T 1+σ [5σ − 2 + μ(σ − 1)]
and
µ ¶ σ−1
μ µ ¶ σ1
1 + T 1−σ − σ1 −μ T 2σ (1 − μ) + T 1+σ (1 − μ) + 2T 2 (2 + μ)
− +3 T .
2 T (T + T σ )

The first eigenvalue is negative if and only if the numerator of the large fraction is positive:

T (2 + 4μ − 7μσ − 2σ − 3σμ2 ) + T σ (1 − μ)[2(σ − 1) − 3μσ] > 0

⇔ 7μσ + 2σ + 3σμ2 − 2 − 4μ < T σ−1 (1 − μ)(2σ − 2 − 3μσ).

2 σ−1
For μ > 3 σ
(or equivalently 2σ −2−3μσ < 0), this is impossible, because the expression
on the left-hand side is positive while the expression on the right-hand side is negative.

Appendix C

Proof of Lemma 4: Denote the critical value of the 3-region model obtained in Propo-
∗
sition 1 by Td3 . We now turn to the 2-region model. Calculating the eigenvalue of the

22
Jacobian matrix at (1/2, 1/2), we find that dispersion is stable if and only if
³ 1 1
´−μ
σ−1 1−σ
− 2 (1 + T )
σ−1 (T σ − T ).
T σ (μ − 1)(1 + (μ − 1)σ) − T (1 + μ)(−1 + σ + μσ)
. < 0.
(σ − 1) (−T 2σ (μ − 1) + T 2 (1 + μ) + T 1+σ (4σ − 2))
³ 1 1
´−μ
Simplifying the above expression and taking into account that the expression − 2 σ−1 (1 + T σ−1 ) 1−σ
(T σ − T ) has a constant negative sign, we conclude that dispersion is stable if and only if

T σ−1 (μ − 1)(1 + (μ − 1)σ) − (1 + μ)(σ − 1 + μσ)

> 0.
(1 − μ)T 2 (σ − 1) + 2T σ−1 (2σ − 1) + 1 + μ
It is easy to see that the denominator is positive and therefore, the stability of dispersion
takes place when

0 < T σ−1 (μ − 1)(1 + (μ − 1)σ) − (1 + μ)(σ − 1 + μσ) ⇔

(1 + μ)(σ − 1 + μσ)
T σ−1 > ∗ σ−1
= (Td2 ) .
(μ − 1)(1 + (μ − 1)σ)

∗ ∗
We conclude the proof by showing that Td3 > Td2 . Because the denominators are equal,
we have

∗ ∗
Td3 > Td2 ⇔
σ − 1 + μ2 σ + 2μ(2σ − 1) > (1 + μ)(σ − 1 + μσ) ⇔
2μσ − μ > 0,

which is always the case.

Proof of Lemma 5: In the 2-region model, concentration is stable if and only if

µ ¶1/σ
−μ T σ−1 (1 − μ) + T 1−σ (1 + μ)
f2 (T ) = −1 + T < 0.
2
We have seen in Proposition 2 that concentration in the 3-region model is stable if and
only if
µ ¶1/σ
−μ (1 + T σ−1 )(1 − μ) + (1 + 2μ)T 1−σ
f3 (T ) = −1 + T < 0.
3

23
We conclude the proof by showing that f3 (T ) < f2 (T ). Given the above expressions, we
have to check that
µ ¶1/σ µ σ−1 ¶1/σ
−μ (1 + T σ−1 )(1 − μ) + (1 + 2μ)T 1−σ −μ T (1 − μ) + T 1−σ (1 + μ)
T < T ⇔
3 2
(1 + T σ−1 )(1 − μ) + (1 + 2μ)T 1−σ T σ−1 (1 − μ) + T 1−σ (1 + μ)
< ⇔
3 2
T 2(σ−1) − 2T σ−1 + 1 > 0 ⇔
(T σ−1 − 1)2 > 0,

which is always the case.

Appendix D

Proof of Proposition 5: We start by characterizing the concentration equilibrium in

region 1 (λ1 = 1).

From the reduced system of equations in Section 2.2, it is straightforward to show that
W1 = 1 and, then, to obtain the remaining variables.

The regional incomes are:

1−μ 1−μ
Y1 = + μ ; Yi6=1 = .
n n

The manufacturing price indexes are:

1 1
−(σ−1) − (σ−1)
θ1 = [W1 ] = 1 ; θi6=1 = [(W1 T1,i )−(σ−1) ]− (σ−1) = T1,i .

The utility of workers in each region is given by:

−μ
U1 = 1 ; Ui6=1 = T1,i Wi .

24
The nominal wages in each region are:
⎧
⎪
⎪ W1 = 1
⎪
⎪ ∙ ³ ´
⎪
⎪ σ−1 ³ ´σ−1 ³ ´σ−1 ¸1/σ
⎪
⎪ 1 σ−1 T1,3 T1,n
⎨ W2 = Y1 T2,1 + Y2 T1,2 + Y3 T2,3 + ... + Yn T2,n
⎪
⎪ ...
⎪
⎪ ∙ ³ ¸1/σ
⎪
⎪ ´σ−1 ³ ´σ−1
⎪
⎪ 1 T1,2 σ−1
⎩ Wn = Y1 T + Y2 Tn,2 + ... + Yn T1,n .
n,1

Is it possible that workers prefer to migrate from region 1 to another region? To address
that issue, consider region d + 1, which is d steps away from region 1. Without loss of
generality, let 1 ≤ d ≤ n/2. Wages are given by
µ ¶σ−1 µ ¶σ−1 µ ¶σ−1
σ 1 T1,2 σ−1 T1,n
Wd+1 = Y1 + Y2 + ... + Yd+1 T1,d+1 + ... + Yn
Td+1,1 Td+1,2 Td+1,n
1−σ 1−μ σ−1 1−μ X µ T1,j
¶σ−1
d d
= Y1 T n + T n + .
n n Td+1,j
j ∈{1,d+1}
/

In the above expression, each region’s specific term depends only on the difference between
the distances to region 1 and to region d + 1. By a careful evaluation3 , we have

X µ ¶σ−1 Xd µ ¶σ−1 n/2+1

X µ T (j−1)/n ¶σ−1
T1,j T (j−1)/n
= + +
Td+1,j j=2
T (d+1−j)/n j=d+2
T (j−d−1)/n
j ∈{1,d+1}
/

X µ T (n+1−j)/n ¶σ−1
d+n/2
X
n µ
T (n+1−j)/n
¶σ−1
+ +
T (j−d−1)/n T (n+d+1−j)/n
j=n/2+2 j=d+1+n/2
n/2+1 d+n/2
X
d
σ−1
X σ−1
X σ−1
X
n
1−σ
(2j−d−2) d (n−2j+d+2) d
= T n + T n + T n + T n

j=2 j=d+2 j=n/2+2 j=d+1+n/2

n/2+1
X
d
σ−1
X σ−1
X
n
1−σ
(2j−d−2) d d
= 2 T n + T n + T n

j=2 j=d+2 j=d+1+n/2

X
d
σ−1 n − 2d σ−1 n − 2d 1−σ
(2j−d−2) d d
= 2 T n + T n + T n .
j=2
2 2
3
Ignore the summation terms whenever the subscript is higher than the superscript.

25
Therefore, we get
σ 1−σ
d1 − μ σ−1 d (1 − μ)(n − 2d) ³ σ−1
d 1−σ
d
´
Wd+1 = Y1 T n + T n + T n +T n +
n 2n
2(1 − μ) X σ−1 (2j−d−2)
d
+ T n
n j=2

(1 − μ)(n − 2d + 2) ³ ´ 2(1 − μ) X
d
1−σ σ−1 1−σ σ−1
d d d (2j−d−2)
= μT n + T n +T n + T n .
2n n j=2

With more manipulation,

∙ ¸
σ 1−σ
d 1 − μ (1 − μ)(d − 1) ³ σ−1 d 1−σ
´
Wd+1 = μT n + − T n +T n d +
2 n
2(1 − μ) X σ−1 (2j−d−2)
d
+ T n
n j=2
∙ ¸ ∙ ¸
1−σ
d 1 + μ (1 − μ)(d − 1) σ−1
d 1 − μ (1 − μ)(d − 1)
= T n − +T n − +
2 n 2 n
2(1 − μ) X σ−1 (2j−d−2)
d
+ T n .
n j=2

Since T x + T −x is an increasing function of x, we can get rid of the summation and get
σ 1+μ 1−σ
d 1−μ σ−1
d
Wd+1 < T n + T n .
2 2

The welfare in region d + 1 is such that:

µ ¶ σ1
− μd 1+μ 1−σ−μσ
d 1−μ σ−1−μσ
d
Ud+1 = T n Wd+1 < T n + T n .
2 2

This means that workers won’t surely move to region d + 1 if:

1+μ 1−σ−μσ
d 1−μ σ−1−μσ
d
T n + T n ≤ 1.
2 2

By differentiation with respect to d, we obtain that if the above condition is satisfied for
d = n/2, then it is satisfied for any d, and concentration is stable. Hence, a sufficient
condition for the stability of concentration is:
1+μ 1−σ−μσ 1−μ σ−1−μσ
T 2 + T 2 ≤ 1.
2 2

26