Revisiting the sectoral Linder hypothesis: Aggregation bias or fixed costs?

• Author: Hendrik W. Kruse
• Published date: 01 September 2020
• DOI: http://doi.org/10.1111/roie.12482
• Date: 01 September 2020

1076

|

wileyonlinelibrary.com/journal/roie Rev Int Econ. 2020;28:1076–1112.

© 2020 John Wiley & Sons Ltd

Received: 18 April 2019

|

Revised: 25 October 2019

|

Accepted: 20 April 2020

DOI: 10.1111/roie.12482

ORIGINAL ARTICLE

Revisiting the sectoral Linder hypothesis:

Aggregation bias or fixed costs?

Hendrik W.Kruse1,2

1University of Göttingen, Göttingen,

Germany

2LEO and Labex Voltaire, CNRS,

University of Orléans, Orléans, France

Correspondence

Laboratoire d’Économie d’Orléans, Faculté

de Droit d’Économie et de Gestion, Rue de

Blois—BP 26739, 45067 Orléans Cedex 2,

France.

Email: hendrik.kruse@univ-orleans.fr

Abstract

This paper reassesses and revisits the Sectoral Linder

Hypothesis due to Hallak, which posits that similar tastes

for quality lead to more intensive trade between similar

countries at the sectoral level. First, the measure of demand

similarity used in this paper is based on the distribution of

income estimated from household surveys. The paper finds

that a similarity measure based on the income distribution

produces stronger results than the traditionally used meas-

ure based on GDP per capita. Moreover, the country/product

level extensive margin is taken into account. This is impor-

tant because similarity is likely to affect the fixed costs of

trade and the fixed costs of alternative means of servicing a

market (i.e., licensing and FDI). Fixed costs, in turn, affect

the number and average productivity of firms that engage in

bilateral trade and hence the overall volume of trade. This

paper employs the method by Helpman et al. to control for

the extensive margin. Heteroskedasticity is addressed using

a feasible generalized least squares (FGLS) approach. The

findings show that once controlling for the effect of simi-

larity on the extensive margin, the Linder hypothesis holds

at more aggregate levels. Other robustness checks suggest

that results are not confined to products that are vertically

differentiated.

JEL CLASSIFICATION

F14; D31

|

1077

KRUSE

1

|

INTRODUCTION

The Linder hypothesis has triggered a bulk of empirical literature in recent decades.1 In his most rad-

ical deviation from supply side explanations of international trade, Staffan Burenstam Linder (1961)

claimed that similarity of demand rather than differences in factor endowment determine the extent to

which countries could potentially benefit from bilateral trade. While early empirical tests were mixed,

recent evidence is increasingly vindicating Linder’s hypothesis (cf. e.g., Choi, 2002).

Many authors argue that demand shifts toward luxury goods as income increases.2 Others em-

phasize the role of quality differentiation.3 Notably, Hallak (2010) incorporates Linder’s theory in a

gravity model of sectoral trade flows. In any given sector, rich countries demand higher quality goods,

which gives them an advantage in their production. Of the money they devote to imports, most is spent

on imports of a suitable quality, resulting in similar countries trading more. Thus, quality differen-

tiation gives rise to sectoral Linder effects. One main empirical finding in Hallak (2010) is that the

hypothesis fails at more aggregate levels due to aggregation bias. A potential reason could be due to

intersectoral specialization according to factor proportions in the classical Heckscher–Ohlin sense (cf.

Eppinger & Felbermayr, 2015).4 Bernasconi (2013) moreover finds no evidence of aggregation bias,

neither at the intensive nor the extensive margin of trade.

In this paper, I revisit and reassess Hallak’s (2010) hypothesis. I examine the difference in perfor-

mance between the traditional Linder term—the similarity in per capita GDP—and an Overlap Index

as introduced by Bernasconi (2013) that allows for strict nonhomotheticity at both margins of trade.

The idea is that not only per capita GDP matters for preferences, but that the distribution of income

within the countries determines demand. In doing so, I largely follow Bernasconi (2013) with some

important exceptions. First, the Overlap Index—measuring the extent to which the income distribu-

tion of two countries overlap—is calculated using more detailed household survey data as in Choi,

Hummels, and Xiang (2009), instead of relying on income shares as in Bernasconi (2013). The com-

parison provided here allows us to see whether the more detailed measure of similarity in fact provides

value-added. Second, like Bernasconi (2013) I will distinguish between the extensive and intensive

margin but for each sector separately. The advantage of doing so is that it takes seriously the expec-

tation that the Linder effect is not present in every sector and may be of different size in each sector.

Thus, estimation with pooled data could mask a very heterogeneous distribution of effects. Third, I

deviate from Hallak (2010) and follow Bernasconi (2013) in allowing an influence of similarity at the

extensive margin; that is, in the selection equation. But instead of estimating the two margins of trade

separately, I apply the Helpman, Melitz, and Rubinstein (2008) methodology to additionally control

for the bias due to the omission of unobserved firm heterogeneity; that is, differences in the number

and productivity of firms engaged in trade.5

Moreover, a contribution of this paper is that it illuminates the role of the extensive margin and

selection in explaining aggregation effects. Selection occurs because of fixed costs (Hallak, 2010;

Helpman etal., 2008). On the one hand, income similarity may affect not only potential trade but also

fixed costs and thereby change the composition of traders. On the other hand, the costs of servicing

a market by other means matters for selection. Similarity reduces costs of servicing another market

by other means (e.g., FDI or licensing) (Fajgelbaum, Grossmann, & Helpman, 2015). To the extent

that this makes it relatively less profitable to export, results at the intensive margin will be downward

biased unless selection is controlled for. These fixed cost effects seem most plausible between sectors.

For instance, if a country has high profitability in a given sector it will find it easiest to service another

country’s market via FDI or licensing if that partner country has a workforce suitable for the sector

in question and an adequate infrastructure (Blyde & Molina, 2015). Since a similar association seems

much less plausible within a sector, a selection model could reduce aggregation effects.

1078

|

KRUSE

This issue is of some importance as aggregation effects would be biases only insofar as the quality

Linder hypothesis is concerned. If it holds for most sectors, we would not want to refute the sectoral

Linder hypothesis, even if it does not hold for aggregate data. Nevertheless, it would still be correct

to refute the “aggregate Linder hypothesis,” as Hallak does. This limits the relevance of Linder’s

reasoning.

Moreover, recent studies suggest that both within and between sectoral nonhomotheticity may

affect the distribution of the gains from trade. Using a hierarchy of needs model, in which expenditure

shifts from agricultural to manufactured goods and finally to services, Egger and Nigai (2018) provide

simulation evidence according to which the gains from trade liberalization are higher among the top

10% of workers and lower in the bottom 10% than homothetic preferences would suggest. Simulations

in a model with endogenous quality differentiation by Faber and Fally (2017) suggest that trade liber-

alization leads to a bigger increase in the variety of goods and decrease in the quality adjusted price

of products that the rich consume compared to the poor. Finally, Liu and Meissner (2019) stress that

nonhomothetic preferences play an important role in evaluating the market potential of a country.

In this study, I analyze a cross-section of countries for 2004. Heteroskedasticity is controlled for

using feasible generalized least squares (FGLS) (Martínez-Zarzoso, 2013). The findings show that the

effect of similarity is much stronger at the intensive margin. Also, allowing for strict nonhomotheticity

improves results at both margins. In fact, using the traditional Linder term the effect at the extensive

margin is on average negative, whereas under strict nonhomotheticity it is positive even though it is

significant in fewer cases than at the intensive margin. Furthermore, controlling for selection and in-

cluding the similarity measure in the selection equation makes the aggregation effect largely disappear

for both measures. The Linder hypothesis seems to hold at more aggregate levels.

The Linder hypothesis is usually applied only to differentiated goods. That is because differenti-

ated goods are most likely to show the relevant characteristics—that is, quality differentiation. While

this may seem as a limitation, the study of differentiated goods can yield important insights. If “what

you export matters”—as Hausmann, Hwang, and Rodrik (2007) assert in the title of their paper—

determinants of trade in differentiated products could teach us something about the difficulty of de-

veloping countries to gain market power in innovative markets. Nevertheless, in the sample employed

here I find that the hypothesis applies for other types of goods, as well. In accordance with other

robustness checks, this seems to suggest that quality is not the only important dimension of special-

ization that gives rise to Linder effects.

The remainder of this paper is organized as follows. Section 2 provides the motivating theoretical

remarks and introduces the concept of aggregation bias. In Section 3, the empirical implementation

is discussed. First, I describe how the Overlap Index is constructed. Then, briefly, the Helpman etal.

(2008) method is outlined. Finally, I discuss the treatment of heteroskedasticity. In Section 4, the data

sources are described and some descriptive statistics are presented. Section 5 presents the results,

including evidence concerning aggregation bias and robustness checks. Section 6 summarizes and

concludes.

2

|

MOTIVATION

In his classic essay, Linder (1961) hypothesized that trade is more intensive between similar countries.

He dismissed factor proportions as an explanation for trade flows mainly on the grounds that their

exploitation requires international entrepreneurial mobility (cf. ibid, pp. 92–92). The starting point of

his argument is that for an entrepreneur to begin production of a good, or an inventor to come up with

a new idea requires knowledge of the state of need. Arguably, an entrepreneur is most familiar with