DETERMINANTS OF INTRA‐INDUSTRY TRADE: A SENSITIVITY ANALYSIS*
| Date | 01 August 1996 |
| Author | Johan Torstensson |
| Published date | 01 August 1996 |
| DOI | http://doi.org/10.1111/j.1468-0084.1996.mp58003005.x |
OXFORD BULLETIN OF ECONOMICS AND STATISTICS, 58,3(1996)
0305-9049
DETERMINANTS OF INTRA-INDUSTRY
TRADE: A SENSITIVITY ANALYSIS*
Johan Torstensson
I. INTRODUCTION
Since the 1950's, there has been a growing emphasis on empirical applica-
tions of theoretical models of international trade, but various empirical
studies have produced conflicting results. Tests of trade models seem to
be afflicted with a number of econometric problems. Often, because
several conflicting theoretical models have been used to explain the trade
flows that are examined, it is unclear which variables should be included
in the empirical model. In addition, proxy-variables are often used to
capture the theoretical variables. Yet empirical testing of trade models
has not explicitly treated the robustness of the estimated coefficient to
changes in the set of control variables and to errors in variables (EIV).
Tests of intra-industry trade (lIT) seem, in particular, to be prone to
these type of problems (see Torstensson, 1994a).
ITT has received increasing attention during the last decades. Econo-
metric studies have tried to identify determinants of the country and
industry pattern of lIT, but some criticism has been directed towards
these studies. In particular, Learner (1994, section 4.5) argues that inter-
pretation of tests of lIT is difficult. One reason for Learner's criticism is
that the coefficients of determination are so low that (p. 87) 'the signs of
the estimated coefficients are not resistant to measurement-error adjust-
ments' and another reason is that (p. 87) 'it is difficult to interpret a
partial correlation which controls for a haphazardly selected group of
other variables'.
This paper examines the robustness of econometric estimates of the
industry pattern of lIT by examining determinants of lIT in Sweden 1983
and 1989. In so doing, the aim is to contribute to an understanding of the
sensitivity of determinants of lIT in general and acquire some more
*A1 anonymous referee has provided very useful and competent comments. Earlier
versions of this paper have been presented at the SPES-Workshops on 'Trade Specialisation
and Market Structure in the European Community After 1992' in Antwerp, October 1993
and in Athens, April 1994. I thank workshop participants for their comments. I have also
received comments and other help from Karolina Ekholm, Pär Hansson, Lars Lundberg and
Rasha Torstensson. The Tore Browaldh foundation, Crafoordska stiftelsen and the Swedish
Council for Research in the Humanities and Social Sciences (HSFR) have provided financial
support.
507
© Blackwell Publishers 1996. Published by Blackwell Publishers, 108 Cowley Road, Oxford 0X4 1JF,
UK & 238 Main Street, Cambridge, MA 02142, USA.
508 BULLETIN
general insights into the usefulness of applying extreme bounds-analysis
(EBA) and methods of dealing with EIV in the field of international
trade.1 Sweden should be an appropriate case study because there are an
unusually large number of publicized results on determinants of ITT in
Sweden.2 Note, though, that this is strictly an empirical paper. Although
an attempt to base theoretical hypotheses more firmly in theory would be
welcome, it is outside the scope of this study to undertake such a task.
Instead, the aim here is simply to examine how robust are the results
arrived at in the large number of previously published papers on lIT.
This paper addresses three specific questions. First, we examine in
Section III whether estimated coefficients in regression analysis are
robust to changes in the set of selected control variables.3 In so doing, we
apply EBA as proposed by e.g. Leamer (1983) and Learner and Leonard
(1983). It turns out that only one variable is robust in both the 1983 and
1989 sample. Second, in Section IV we analyse by reverse regressions as
proposed by Klepper and Learner (1984) the effects of EIV that are
orthogonal to the true unobserved variables. It is found that if only
variables that are significant in more than 95 percent of the EBA-analysis
are included in the econometric model, the estimated coefficients are
bounded. However, when more variables are included in the model, the
estimated coefficients are unbounded unless restrictions are put on the
increase in R2 that would occur if the measurement errors were removed.
Finally, in Section V we use the method of Krasker and Pratt (1986) in
which independence of the error terms is not required and we estimate
the required correlation coefficient between the proxy-variable and the
'true' variable to ensure that they necessarily are of the same sign. We
found that the required correlation coefficient is much higher than could
otherwise be expected from the relative crude proxy-variables we employ
to measure product differentiation.
II. EARLIER EMPIRICAL STUDIES OF lIT
A large number of studies have investigated the industry pattern of ITT
econometrically (for a comprehensive survey, see the now classic study by
'In his comprehensive study of net trade, Learner (1984) applies EBA-analysis and reverse
regressions to deal with measurement errors. Maskus (1991) and Torstensson (1995a) use
reverse regression to examine measurement errors in studies of net trade flows.
2Lundberg, 1982; Lundberg and Gavelin, 1983; Lundberg, 1988; Hansson and
Lundberg, 1989; Hansson, 1991; Lundberg, 1992.
restrict our analysis to one country (Sweden) and two years (1983 and 1989). Culem
and Lundberg (1986) find that the pattern of ITT seems to be fairly stable over time, the
simple correlation coefficient being in the range of 0.6-0.8. However, the industry pattern of
ITT seems less stable across countries. In trade with developed countries the correlation
coefficients is seldom above 0.4.
4EBA-analysis has been applied to growth studies by Levine and Renelt (1992) and later
by Torstensson (1994b) and to estimation of money demand by Cooley and LeRoy (1981).
The approach has been criticized by McAleer, Pagan and Volker (1986), but defended by
Cooley and LeRoy (1986) and Learner (1986).
© B!ackwel! Publishers 1996.
To continue reading
Request your trial