Is the wage equation spatial enough? Evidence from a novel regional trade dataset

• Date: 01 August 2018
• Author: Aurélien Fichet de Clairfontaine,Christoph Hammer
• Published date: 01 August 2018
• DOI: http://doi.org/10.1111/roie.12315

SPECIAL ISSUE PAPER

Is the wage equation spatial enough? Evidence from

a novel regional trade dataset

Aur

elien Fichet de Clairfontaine

|

Christoph Hammer

Vienna University of Economics and

Business (WU), Vienna, Austria

Correspondence

Christoph Hammer, Department of

Economics, Vienna University of

Economics and Business (WU),

Welthandelsplatz 1, 1020 Vienna,

Austria.

Email: chhammer@wu.ac.at

Abstract

This study focuses on the market accessibility of European

regions and its relationship to income per capita, summar-

ized in the new economic geography (NEG) “wage

equation”. In a first step, we make use of a novel dataset of

bilateral trade flows for 254 European nomenclature of terri-

torial units for statistics (NUTS-2) regions (for 26 European

countries excluding Bulgaria and Romania) in order to esti-

mate trade costs and ultimately construct a regional measure

of access to markets. In a second step, we test the hypothesis

that access to domestic as well as to foreign markets

increases income per capita. We find that, in spite of its spa-

tial formulation, the wage equation is not able to capture

local spatial patterns of the distribution of European regional

income per capita.

1

|

INTRODUCTION

Ever since the seminal work of Krugman (1991), models of “new economic geography”(NEG) have

attempted to shed light on the uneven economic distribution using a general equilibrium framework

with a focus on geography represented by transport or trade costs. They aim to explain agglomeration

patterns by allowing mobile production factors to move across space to regions with the highest

rewards. While historical and institutional factors as well as the physical geography of a region can

help explain agglomeration, NEG models focus on forces that reflect the behavior of optimizing

(mobile) economic agents.

One central element of NEG models that is derived from optimizing agents is the so-called “wage

equation.”It states that the “maximum wage that each firm in a specific region can afford to pay is a

function of trade-cost-weighted market and supply capacities”(Redding & Venables, 2004, p. 58).

This means that NEG models imply a spatial structure in which remunerations of factors are higher in

regions that have better access to markets or a higher “real market potential.”

This relation was first introduced by Harris (1954), who used geographical distance to weigh

income from all other regions. Real market potential has some advantages over the basic concept of

Harris. It is derived from microeconomic theory and considers competition among firms in export

610

|

V

C2017 JohnWiley & Sons Ltd wileyonlinelibrary.com/journal/roie Rev IntEcon. 2018;2 6:610–633.

DOI: 10.1111/roie.12315

markets. Additionally, it is constructed with trade costs that are more broadly defined than as a function

of distance only.

The empirical literature can be divided into three main approaches that to some extent also differ in

their research questions.

1

The first strand was pioneered by Redding and Venables (2004). They use a

two-step approach to estimate the wage equation. First, in order to construct a measure for market

access they estimate a gravity-type trade equation to obtain estimates of trade costs. Trade costs repre-

sent every form of friction in shipping goods, services, people or ideas over space. In the most general

case one can think of the sum of transport costs, information and time costs, institutional and cultural

barriers (such as tariffs, product standards, and language).

2

In a second step, the wage equation is esti-

mated with the constructed measure for market access. They find robust evidence for the role that mar-

ket access plays for explaining income differences between 101 countries.

A second line of research usually uses more disaggregated data to estimate the wage equation.

Hanson (2005) studies the wage equation derived from Helpman’s extension of Krugman’s core–

periphery model, for U.S. counties between 1970 and 1990. In contrast to the above-mentioned two-

step approach, he estimates the wage equation in a single step using nonlinear least squares. The

advantage of this approach is the possibility to identify most theoretical parameters. Among the draw-

backs are the sensitivity of nonlinear least squares to starting values and demanding data requirements

to simultaneously model all the equilibrium conditions.

3

The third approach builds on the second and is a combination of estimation and simulation. It tries

to bring the estimated coefficients back into the theoretical model. Brakman, Garretsen, and Schramm

(2006) and Bosker, Brakman, Garretsen, and Schramm (2010) argue that, in order to interpret the

results of a NEG wage equation, one should study the implications for economic agglomeration within

the NEG model from which the equation was motivated. For them, the wage-equation estimates are a

prior for model simulations.

We follow the two-step approach put forward by Redding and Venables (2004) but in contrast to

earlier studies, our analysis is based on a novel cross-regional trade dataset for 254 European NUTS-2

regions for 2010. To the best of our knowledge there is no study that follows the two-step approach

with such detailed data for the overall European Union. One advantage of this is the ability to stay

close to the theoretical model and to use Krugmans real market potential.

4

Since the internal economic

structure of countries is usually biased towards large urban areas, regional heterogeneity would be lost

in a national analysis. Compared with other studies with a regional focus that also use two steps (see,

e.g., Head & Mayer, 2004; Breinlich, 2006), we are able to estimate trade costs for each of 254

2

region

pairs with a fully specified gravity equation. Earlier papers either had to work with national data and

impose homogeneous trade costs within a country or modeled trade costs as a function of distance

only.

An additional new element is the way in which we considerspatial correlation when we estimate the

regional wageequation in the second step. Comparedwith earlier formulations (seeFingleton & Fischer,

2010; Fingleton, 2010; or Bruna, Lopez-Rodriguez,& Faí~

na, 2016) we argue that a NEG wage equation

with real market potential should not be represented by a spatial autoregressive model. Instead we use

spatial filtering techniques (Tiefelsdorf and Griffith, 2007) in orderto control for spatial dependence.

Our results are consistent with earlier findings on the ability of real market potential, and more

especially its foreign component, to explain regional distribution of income per capita among European

regions. While the effect of domestic market potential, that is, a region’s home market, is negligible,

we show that foreign market potential is a strong determinant of income per capita even when correct-

ing for regional spatial dependences. Furthermore, an analysis of remaining spatial autocorrelation sup-

ports recent findings in the literature, at the time of writing, that the wage equation cannot capture local

patterns but rather accounts for a global spatial trend.

FICHET DE CLAIRFONTAINE AND HAMMER

|

611