The Relationship Between International

Trade and Economic Growth in Mexico

HENDRIK VAN DEN BERG

ABSTRACT
The available empirical evidence on the relationship between international trade and
economic growth in Mexico is not conclusive. This article first identifies the contribu-
tions and shortcomings of previous empirical work. Then several new econometric
approaches are pursued using data for 1960-1991. Modem time-series methods and
improved data are used to replicate previous Granger causality tests and single-equation
regressions. Then, a simultaneous equation time-series regression model is used to con-
firm hypothesis that trade and growth are directly related. A direct test of the relation-
ship between trade and total factor productivity is also carried out using a simultaneous-
equations model. A positive relationship is again confirmed. The Lucas critique
reminds us, however, that we cannot use econometric results, based on data from the
past, to justify Mexico’s new “outward-oriented” economic policies. Nevertheless, the
results appear supportive when we consider that the sample period includes two
decades of import substitution policies; the econometric results are likely to understate
the actual strength of the trade-growth relationship under Mexico’s current open trade
regime.

1. INTRODUCTION
The collapse of economic growth in Mexico after the 1982 debt crisis led the Mexican
authorities to completely overhaul their economic policies. Import-substituting industrial-
ization (ISI), the dominant policy in Mexico since the 1930s was replaced by trade liber-
alization and, more recently, the North American Free Trade Agreement. The current
government continues to view free trade as a key component of Mexico’s overall strategy
to put its economy back on a steady growth path.
There are many sound theoretical reasons why trade liberalization will help to increase
economic growth in Mexico. In addition to the traditional arguments in favor of free trade,
Aspe (1993) and Lustig (1992) describe the important role that trade li~ralization plays
within Mexico’s overall economic reform program. But while the theoretical literature is
often quite convincing, the available econometric evidence on the relationship between
trade and growth in Mexico may not appear to justify the radical policy shift. The hypoth-
esis that, in Mexico, trade and economic growth are positively related appears to enjoy only
“mixed’ empirical support, and the statistical methods of many of the available studies are
open to serious criticism.
This paper offers a detailed examination of the econometric evidence on the relationship
between trade and growth in Mexico. The methods and results of previous studies are crit-

Hendrik van den Berg l Associate F’rofessor, Department of Economics, University of Nebr~ka-Lincoln,
Lincoln, NE 68588-0489; e-mail: hvandenb~unlinfo.uul.edu.

North American Journal of Economics & Finance 8(l): 1-2 I Copyright 0 1997 by JAI Press Inc.
ISSN 1062-9408 All rights of reproduction in any form reserved
2 VANDENBERG

ically evaluated. To correct for some of the potential sources of errors in previous work,
other econometric approaches are pursued. Specifically, this paper applies modem time-
series methods, new specifications of simultaneous equations models, better data, and a
more direct test of the effect of trade on total factor productivity covering the period 1960-
1991. The improved methods generate consistent results in favor of the hypothesis of a
positive relationship between trade and growth in Mexico.

II. A CRITIQUE OF THE AVAILABLE EVIDENCE

Empirical studies of the relationship between foreign trade and economic growth have
nearly always specified linear econometric models of the form

GGDP = a0 + alGCAP + a2GLAB + ajTRADE, (1)

where GGDP, GCAP, and GLAB are the growth rates of real gross domestic product, cap-
ital stock, and labor force, respectively, and TRADE is a measure of international trade.
Model (1) resembles the “sources of growth” equation,

GY = GTFP + aGK + (1 -a)GL, (2)

derived from a neoclassical production function, in which GY, GK, GL, and GTFP are the
growth rates of total output, capital, labor, and total factor productivity, respectively, and a
and (1 - a) are the relative income shares of capital and labor.’
In practice, the ratio of investment to output, Z/Y, has commonly been used in place of
GCAP. Population growth, GPOP, replaced GLAB when detailed labor force data were not
available. The growth of real exports, GREX, has been the preferred proxy for TRADE.
Equation (1) then becomes

GRGDP = a0 + al(UYj + a,GPOP + a,GREX. (3)

This specification suffers some measurement problems, but because the data are readily
available, it has been widely applied.2 Among the many authors who have applied models
similar to Equation (3) are, in alphabetical order, Dollar (1992) Edwards (1988), Feder
(1982), Greenaway and Nam (1988), Greenaway and Sapsford (1994), Kavoussi (1984),
Moschos (1989), Ram (1985, 1987), Salvatore and Hatcher (199 l), Sheehey (1992) Tyler
(1981), and Van den Berg and Schmidt (1994).

A. Trade and Productivity

The TRADE variable in Equation (l), usually GREX, is positioned to “explain” some
portion of the constant GTFP in Equation (3).3 A specification such as (1) or (3) is thus in
accordance with those who suggest that the influence of international trade shows up as an
increase in total factor productivity.
Trade can improve total factor productivity in a number of ways. According to Dom-
busch (1993):

The channels through which trade liberalization could bring benefits are broadly these:
improved resource allocation in line with social marginal costs and benefits; access to
/~tematio~a/Trade and Economic Growth in Mexico 3

better technologies, inputs and intermediate goods; an economy better able to take
advantage of economies of scale and scope; greater domestic competition; availability
of favorable growth externalities, like the transfer of know-how; and a shake-up of
industry that may create a Schumpeterian environment especially conducive to
growth.4

Krueger (1980) furthermore points out that, in contrast to free trade, import substitution
policies and their inherent need for heavy governmental involvement in economic deci-
sions can lead to unproductive rent seeking, costly paperwork, bureaucratic bottlenecks,
and other waste. And, especially relevant in the case of Mexico, export earnings may allow
a country to borrow external capital at more favorable terms and without danger of debt
servicing difficulties and sudden discontinuities in capital flows (See, e.g., Collins and
Park 1989; Dollar 1992; World Bank 1993).
Equation (1) is also compatible with the growing literature on endogenous growth.
Grossman and Helpman (1994), Edwards (1992), and Ruffin (1993) develop endogenous
growth models in which open economies absorb new technology at a faster rate than closed
economies. Because the availability of knowledge in the rest of the world lowers the cost of
creating new knowledge, policies that expand trade can also increase the international flow
of knowledge and, hence, economic growth. Increases in technology and know-how are
captured as total factor productivity growth; hence, specification (1) provides a test of the
hypothesis that trade increases growth by expanding knowledge transfers.”

8. Measurement Error
Equation (3) may not offer a very precise test of the relationship between trade and total
factor productivity, however. If I/Y and GPOP do not accurately represent the growth of
capital and labor, then the constant will misrepresent TFP growth.
In the case of GPOP, ignoring differences in labor productivity in a cross-section study
(or changes in labor productivity in time-series studies) can lead to a biased estimate of the
coefficient of the TRADE variable, ax. For example, if both exports and education grow
more rapidly in high-growth economies, but GLAB is not adjusted for human capital, then
standard cross-section statistical techniques will erroneously give GREX credit for
changes in output that are actually due to increases in labor productivity.
Other variables in (3) are also imprecise. I/Y ignores depreciation. More serious, per-
haps, is the likelihood that export growth does not accurately represent TRADE. The suc-
cess of the “Asian Tigers” and their “export-led growth” has understandably focused
attention on the role of exports, but theorists suggest that imports also matter for economic
growth. For example, the well-known “twin-gap” literature focuses on the shortage of for-
eign exchange to acquire needed imports. More recently, outward orientation seems to
have helped countries maintain access to external financing and thereby avoid the foreign
debt crises that devastated so many Latin American economies in 1982.6 If both imports
and exports matter, the popular model (3) is misspecified.

C. Cross-Section Regression Results

Most empirical studies of trade and growth have used the cross-section approach.’ For
these studies, a significantly positive relationship between trade and growth is revealed for
nearly all types of country groupings and all periods of time. Since Mexico was included in
the samples of most cross-section studies, it is tempting to interpret the positive cross-sec-
4 VANDENBERG

tion results as providing empirical support for Mexico’s shift to free trade policies. But
cross-section results do not reflect the relationship between trade and growth in any indi-
vidual country.* Indeed, most authors of cross-section studies have avoided making infer-
ences about specific countries in their samples.’
Estimates of models such as (1) also implicitly assume that production functions are
identical across countries. If production functions differ, which is likely, then the relation-
ship between trade and growth is not properly estimated with model (1). In addition, cross-
section studies are especially likely to suffer from omitted variable bias because there are
many unmeasured factors (cultural, political, environmental, etc.), important to economic
growth, that vary across countries.
Cross-section regressions will also generate inconsistent results if the variables used are
average values of nonstationary time series. lo The variables in cross-section data sets are
normally period averages rather than single observations; such averages reduce the
“noise” of cyclical variations or one-time exogenous shocks. But if a variable used in the
cross-section studies is nonstationary, as is often the case, the calculated average value
does not represent the true mean of the series. The mean of a nonstationary time series
with drift is, of course, always changing through time, and the series has an infinite vari-
ance when the source of the nonstationarity is a stochastic trend (unit root). Such a series
technically has no mean, and thus it is unclear what the regression results tell us. Thus, not
only do results from cross-section analyses not describe the behavior of any particular
country, but cross-sections of structurally diverse countries will not even produce consis-
tent “average” estimates.

D. Time-Series Analyses
The shortcomings of cross-section studies have led to the use of time-series data for indi-
vidual countries. Time-series analyses may be better able to capture the relationship among
trade and growth for a specific country, and simple causality tests can easily be conducted.
More recently, researchers have learned to apply modern time-series methods that can deal
with nonstationary variables. Table 1 lists time-series studies on Mexico.

TABLE 1. Summary of Previous Time-Series Studies Relating Trade and

Economic Growth in Mexico

Study Sample Method Variables Results

Jung and 37 countries Granger GRGDP and GREX Exports do not “cause”
Marshall( 1985) Causal. growth in Mexico
Chow (1987) 8 NIC’s Sims Mfg. output and GREX Uni-directional causal. in
1960-1980 Causal. Mexico
Ram (1987) 88 countries OLS Production function GREX significant in 38 of
1960-1982 GREX, WY,GLAB, GOVT 88, no individual results
Dodaro ( 1993) 87 countries Granger GRGDP and GREX Exports do not “cause”
1967-l 986 Causal. growth in Mexico
Salvatore and 26 countries OLS Production function GREX not significant in
Hatcher (1991) 1963-1985 GREX, I, GINDUSTRY Mexico
Van den Berg and 17 countries OLS, Production function GREX significant in
Schmidt (1994) 1960-1987 Er. Con: GREX, I/Y, GLAB Mexico
infernafional Trade and Economic Groin in Mexico 5

Most early time-series studies applied causality tests. Jung and Marshall (1985) used
1950-1980 data to perform Granger causality tests for 37 developing economies, and
Dodaro (1993) performed similar Granger causality tests using 1967-1986 data for 87
countries. In both studies, export growth was found to “cause” output growth in very few
cases, as Table 1 reports. Mexico was never among the significant cases.
Chow (1987), on the other hand, performed Sims’ causality tests on export growth and
industrial output for eight Newly Industrialized Countries (NICs) over 1960- 1980, includ-
ing Mexico. Causality was found to run in both directions for six of the eight countries;
Mexico was the only country for which causality was uni-directional from export growth to
indust~~ output growth. These causality tests have been criticized, however, for not taking
into consideration the possibility that the time series may have been nonstationary. Giles,
Giles, and McCann (1992) and Bahmani-Oskooee and Alse (1993), for example, per-
formed causality studies of trade and growth, meticulously conducting unit root and coin-
tegration tests where appropriate. But, their studies did not include Mexico.
The stationarity issue is not the only shortcoming of past causality studies. The general
validity of causality tests has been widely disputed. First of all, the simple formulation of
Granger causality regressions lacks a theoretical foundation. Furthermore, the relationship
between two variables, in this case trade and growth, is examined in isolation; omitted vari-
able bias is likely. Learner (1985) also points out that the term “precedence” would more
accurately describe the Granger methodology, and precedence no more impfies true causal-
ity than does contemporaneous correlation. ”
Richer relationships among variables, such as those suggested by the production-func-
tion specification of Equations ( 1) and (3), have been tested by applying regression analy-
sis to the time-series data. For example, Salvatore and Hatcher (1991) used time-series data
and a specification similar to Equation (3). They first ran pooled regressions for 1963-1973
and 1973-1985, and Mexico is included in a subgroup of “moderately inward oriented”
countries. The coefficient of export growth is significantly positive in both periods for
Mexico’s subgroup. But when the authors applied only Mexican time-series data for the
entire 1963-1985 period, the coefficient for the export variable was not significant.
Salvatore and Hatcher did not test their time series data for unit roots. The study by
Van den Berg and Schmidt (1994) used specification (3) for 17 Latin American countries,
and they performed two unit root tests for each country’s 1960-1987 time-series of
GRGDP, Ify, GLAB, and GREX. Where called for, they also tested for cointegrating
relationships. Then, in strict accordance with the unit root and cointegration test results,
they specified variations on Equation (3). They confirm a positive relationship between
export growth and economic growth for the majority of the 17 countries, including Mex-
ico.12 Van den Berg (1996) performed new time-series regressions for six Latin Ameri-
can countries, including Mexico, using data through. 1990 and addressing some of the
measurement problems mentioned above; he again confirmed a significant relationship
between exports and growth in Mexico.

E. The Potential for Simultaneity Bias

The time-series regressions appear to solve some of the estimation problems associated
with cross-sections studies. Another problem remains, however. Several authors who
used models similar to Equation (l), including Balassa (1978), Tyler (1981), Feder
(1982), and Ram (1985), acknowledged the possibility that a positive relationship
between exports and GDP might be “predetermined” because exports are a component of
 VANDENBERG

GDP. However, each of these authors justifies maintaining GREX and GRGDP in their
single-~uat~on models.
The justifications for ignoring the possibility of simultaneity bias are several. First of
all, Equation (3) uses the grawth rates of exports and GDP, not levels, in a production
relation. Thus, there is no strict national income accounting identity present in the specifi-
cation. Also, Equation (3) contains the growth rates of the factors of production as
explanatory variables, and hence export growth helps to explain only that portion of total
GDP growth, including exports, not accounted for by the growth of the factors of produc-
tion. This is precisely the productivity effect that we seek to measure, that is, the contri-
bution to GDP growth of the reallocation of resources and, perhaps, the reduction in
unemployment brought about by expansion of the export sector. Finally, there is no a pri-
ori reason why the coefficient “3 in Equation (I) must be positive. For example, depen-
dency theory suggests that export activity in ~~nderdevelop~d” economies causes
praductivity to stagnate. If true, regression analysis would most likely generate a negative
value for a3.
Despite these arguments, some researchers altered their econometric specifications to
avoid simultaneity bias. Michaely (1977) used the ratio of exports to GDP, X/Y, rather than
the growth of exports, claiming that this would avoid bias. Michaely’s results are still very
positive. Sheehey (1992) similarly uses several other measures of TRADE in model (l),
and he finds that “the positive effects of shifting more resources into exports were confined
to a limited number of more industrialized developing countries in the 196Os, a period of
strong growth for world trade.“‘” Mexico was in this latter cross-section group.14
The sjmultaneity problem is, of course, best addressed using a simultaneous-equations
model that explicitly specifies the suspected bi-directional relationships among vari-
ables.15 This was the approach taken by Sprout and Weaver (l993), who specified (3) as
the first equation in a three equation model estimated for 72 developing countries using a
cross-section of average values for 1970-1984. The second and third equations explained
GCAP and GREX. Sprout and Weaver confirmed a strong relationship between export
growth and GDP growth, although they also found a weak direct relationship between
growth and exports elsewhere in their simultaneous equation model,
Sprout and Weaver further split their sample into “small primary product exporters,”
“small nonprimary product exporters,” and “large LDC’s.” Latin American countries were
dispersed among the three groups. For the latter two groups, coefficients for GREX were
significant and similar in magnitude to those found in single-equation cross-section stud-
ies, but for the first group export growth was not significantly reiated lo GNP growth.
Thus, in this case, even cross-section analysis did not un~fo~~y support the hypothesis of
a trade-growth relationship. Export growth apparently mattered in some countries but not
in others.
Esfahani (1991) estimated a different three equation cross-section model, adding import
growth to his first equation, a production function equation that was atherwise similar to
(1). His second and third equat,ions explained exports and imports, respectively. Compared
to the results of single equation cross-section regressions, the coefficient of the export
growth variable became less significant in his model, and imports appeared to be more
important to economic growth than exports. But both the Esfahani and Sprout and Weaver
studies only applied cross-section data; therefore, their resufts are subject to the criticisms
discussed above and of limited relevance to any specific country such as Mexico.
International Trade and Economic Growth in Mexico 7

F. Summa~ of the Available Econometric Evidence

In general, most cross-section studies point to a strong positive relationship between
trade and growth, but the time-series results are less conclusive. These general results carry
over to Mexico. Causality tests offer little support for a positive relationship between trade
and growth in Mexico. None of the causality tests for Mexico control for unit root behav-
ior. A time-series study of Mexico using regression equations based on the production
function, such as Equation (3), while also adjusting for unit root behavior (Van den Berg
and Schmidt 1994) found a significant positive relationship between trade and growth in
Mexico. But measurement errors and simultaneity were not addressed.

G. Extending the Analysis

In the remainder of this paper, time-series analysis is used to uncover the relationship
between trade and growth in Mexico. The analysis builds on the studies reviewed above.
First of all, unit root tests are applied to all variables. Both causality tests and time-series
regression models are then adjusted to explicitly deal with variables found to have unit
roots.
Secondly, the results of previous studies may have been seriously biased by the use of
inaccurate measures of capital and labor in specifications such as Equation (3). Hofman’s
(1992) time-series of capital stocks for Mexico, plus an adjustment of labor force data for
education, age, and gender, provides a specification closer to the ideal production-function
model ( 1).
Instead of simple two-variable Granger causality tests, a more elaborate VAR model is
used to test causality from trade to growth. All variables suggested by the production func-
tion model are included in the VAR.
Next, the issue of simultaneity bias is examined for Mexico. A simultaneous equations
model, based on the cross-section models of Sprout and Weaver (1993) and Esfahani
(199 l), is tested using Mexican time-series data.
Finally, the relationship between trade and total factor productivity is directly examined
using a two-step method that first estimates TFP growth and then regresses the estimated
growth of TFP on the growth of trade. This two-step method also permits the application of
a VAR causality tests as well as single- and simultaneous-equations estimation models.

Ill. TESTING FOR STATIONARITY

Time-series regressions may generate spurious results if the series are not stationary. l6 In
the case of nonstationa~ty, conventional statistical tests such as OLS are biased toward
finding a significant relationship among variables in levels when in fact none existsI
The standard method for detecting nonstationary behavior in a time series is to test for
the presence of a unit root. Testing can be extended to incorporate the prospect of a deter-
ministic trend as well as the stochastic type of trend represented by a unit root. In this
paper, two unit root tests for stationarity are used. The Augmented Dickey-Fuller (ADF)
test takes the existence of a unit root as the null hypothesis.18 A criticism of the ADF test
is that it is not very powerful and, therefore, biased toward accepting the presence of a
unit root. Kwiatkowski, Phillips, Schmidt, and Shin (1992) have proposed a test (hence-
forth referred to as the KPSS test) that takes the absence of a unit root as the nu11.19 By
conducting tests under both types of null hypotheses, the analysis will not be predis~sed
toward acceptance or rejection of unit roots. Table 2 reports the results for both the ADF
8 VANDENBERG

TABLE 2. Unit Root Tests, Mexico, 1960-1991

Augmented Dickey-Fuller Test’ KPSS Tes?
Level 1st Difl Level 1st DifJ:
GRGDP -3.71(a)* -6.05(a)* .082(b) .064(b)
GLAB -1.85 -6.17* .553* ,138
GCAP - 1.39(a) -4.06(a)* .189(b)* .063(b)
GREX -3.35* -5.64* .108 ,083
FDUY -3.76* -6.70* ,182 ,135
GRIM -5.02* -5.00* ,154 ,067
GPCY -3.54(a)* -6.28(a)* ,068 ,062
PCY -1.41(a) -3.43(a)* .181(b)* .080(b)
TPGROWTH -3.72* -4.61* ,294 ,119
FORCAP -2.12 -4.10* .418* ,141
Discrepancy between ADF and KPSS:
GTFF’ -3.64* -7.23* .445* ,098
RER -1.71 -5.10* ,170 ,080
Notes: ’ Augmented Dickey-Fuller (1979) tp umt root test wth constant and no trend. u&s\ otherwise noted. A lag from 0 to 3
was selected according to the Akalke Informanon Critermn.
* Kwiatkowski et.al. (1992) unit root tests with constant and no trend (H,: no unit root).
(a) Augmented Dickey-Fuller (1979) t, unit root test with constant and trend. Lag from 0 to 3 was selected in each case
according to the Akalke Information Criterion.
(b) Kwiatkowski &al. (1992) tests with constant and trend.
* Significant at the 90% level using critval values from Table 8.5.2 in Fuller (1976). p. 373, or in the case of a trend.
Table VI, Dickey and Fuller (1979). p. 1063. In case of ADF, * implies no unit mot; for KPSS test. * implies unit root.

and KPSS unit root tests for all the variables that appear in the models estimated in this
paper. Note that the results of the ADF and KPSS tests are mostly compatible. The tests
conflict for only two variables, the growth of total factor productivity and the real
exchange rate.
The unit root test results have several implications. First of all, some of the previous stud-
ies may have generated misleading results. For example, all the studies that specified Equa-
tions (1) or (3) apparently were, in the case of Mexico, regressing acombination of stationary
and nonstationary variables2’ In the analysis that follows, we will pay careful attention to
the various combinations of stationary and nonstationary variables included in the models.

IV. IMPROVED MEASURES OF CAPITAL AND LABOR

Most empirical studies employ a specification such as (3) rather than (1) because accurate
measures of capital and labor stocks were not readily available. But in the case of Mexico,
variables that fit the spirit of Equation (1) either exist or can be derived, and it is not nec-
essary to use I/Y and GPOP as explanatory variables.
The capital stock series by Hofman (1992) provides us with a time series of Mexican
capital stock, GCAP, through 1989. The series is extended through 1991 by using available
data on investment and assuming a depreciation rate.*’
A series of annual quality-adjusted labor force figures is generated by adjusting the
available figures on the economically-active population for age and education. Specifi-
~nferna~~ona/ Trade and Economic broad in Mexico 9

tally, labor force figures are augmented according to the estimated average number of
years of education of men and women. They are reduced for the number of young workers,
who are assumed to be half as productive as older (age 25 and up) workers. The Appendix
describes in detail how the series GLAB was derived.
As discussed earlier, imports may also matter for economic growth. Imports can improve
productivity by reducing shortages, providing better inputs, and, in the case of capital
goods, incorporating new technologies. Through its maze of import restrictions, Mexico
effectively rationed imports during much of the 1960-1991 period. If imports indeed mat-
ter, the growth of real imports, GRIM, should be included in the analysis as welt as GREX.
To reduce the potential for measurement error, the basic specification for the time-series
estimates that follow below will be

CRGDP = a0 + a,GCAP + a@LAB + a3GREX + ahGRIM. (4)

Unit root test results for the series GRGDP, GREX, GRIM, GCAP, and GLAB are given
in Table 2. According to both the ADF and KPSS tests, GCAP and GLAB exhibit unit root
behavior, while the remaining series appear to be stationary.

V. NEW TIME-SERIES ANALYSIS OF TRADE AND GROWTH

Previous causality tests relating trade and growth in Mexico reported contradictor results.
Regression analysis using time-series data was more favorable to the hypothesis of a posi-
tive relationship between trade and growth. Here those tests are updated for 1960-1991
while also correcting for unit root behavior, measurement error, and, later, simultaneity.

A. A Simple Granger Causality Test

The simple two-variable Granger causality test regresses the rate of economic growth,
GRGDP, on the lagged values of GREX. In order to test the possibility of reverse causality,
GREX is regressed on lagged values of GRGDP. Table 3, panel I, presents the results from
simultaneously estimating both types of Granger equations. Trade Grcznger causes eco-
nomic growth in Mexico over 1960-1991, but GRGDP does not Granger cause GREX.
These results refute the lack of causality found for Mexico by Jung and Marshall and
Dodaro, and they confirm the uni-directional causality reported for Mexico by Chow.

B. Vector Autoregression
If there are good reasons to believe that the relationship between trade and growth is
more accurately captured by the production-function specification, as in Equation (4), in
which factor supplies and productivity also exercise their influence on output, then the
results of simple two-variable Granger tests must be questioned. It would be useful to
examine “causality” within an extended vector autoregression (VAR) model containing all
the variables in Equation (4).
Table 3, part 2, shows the results of estimating a VAR relating GRGDP, GCAP, GLAB,
GREX, and GRIM. Since unit roots were confirmed for GCAP and GLAB, the VAR can
be consistently estimated either by differencing the two unit root series and running a spec-
ification that consists of a mixture of differenced and levels variables or differencing all
five variables. In the latter case we can be sure to capture the short-run relationship
between the variables; it is not clear what the estimates of the “mixed’ model would tell us.
10 VANDENBERG

TABLE 3. Trade and Growth-Causality Tests’ Mexico: 1960-1991

1. Granger tests, 3 lags”:

GREX -_j GRGDP3 16.2326** GRGDP -+ GREX 0.0205
2. VAR, only unit root variables differenced, 3 lags:
GCAPDIF + GRGDP a.9094** GRGDP -+ GCAPDIF 1.0403
GLABDIF + GRGDP 11.7322*** GRGDP --+ GLABDIF 3.7972
GREX + GRGDP 0.6960 GRGDP -+ GREX 3.1775
GRIM --f GRGDP 4.0016 GRGDP -+ GRIM 1.7934
3. VAR, all differenced, 2 lags:
GCAPDIF -+ GRGDPDIF 5.8678* GRGDPDIF -+ GCAPDIF 2.1747
GLABDIF -+ GRGDPDIF 6.1468** GRGDPDIF -+ GLABDIF 0.7067
GREXDIF + GRGDPDIF 1.8529 GRGDPDIF -+ GREXDIF 2.8003
GRIMDIF -_;, GRGDPDIF 0.0723 GRGDPDKF -+ GRIMDIF 2.9587
i-f&~: I. Test statistics are Wald x2 statistics.
2. The number of lags was chosen according to the Akaike Information Criterion (AK).
3. Variables are described in the Data Appendix.
*** 99% significance level.
** 95% significance level.
* 90% significance level.

Only GCAPDIF and GLABDIF

cuu~e GRGDPDIF. Unlike the results from the Granger
causality test, GREXDIF does not cause GRGDPDIF. GRIMDIF is not significant either.
There appears to be no reverse causality for any of the four causal variables.

C. Single-Equation Time Series Regression Analysis

While the VAR model is superior to the simple two-variable Granger model, if the pro-
duction function model is really the preferred theoretical formulation, then Equation (4)
should be estimated directly using standard regression methods.22 But in a time-series
model, consistent estimation procedures depend on the combinations of stationary and
nonstationa~ variables in the model. When nonstationary variables in a regression model
have unit roots, a frequently used procedure is to difference the variables. Consistent esti-
mation can then proceed only if the first differences are stationary, provided that the series
are I( 1) and the nonstationary variables are not cointegrated. Of course, the coefficients of
first differences reveal only the short-run relationship among the variables.2” A discussion
these points is provided by Engle and Granger (1987) and Mehra (1991).
Results from applying time series data for Mexico over 1960-1991 to specification (4)
are reported in Table 4, column 1. The results from estimating the equation without GRlM
are reported for comparative purposes in column 2. A time variable was added to capture a
possible secular time trend. The results support the hypothesis that trade is positively
related to export growth in Mexico. The earlier results by Van den Berg and Schmidt
(1994) do not appear to have been biased by measurement errors or omitted variables.

0. Simultaneous Equations Regression

The positive results from estimating the single equation regression model still may
have been biased by simultaneity between GREX and GGDP or GRIM and GGDP. as
 11

TABLE 4. Model of GDP Growth Rates Wing Specifcation (4)

and Improved Estimates of GCAP and GLAB Mexico:
Annual Observations for 1960-l 991

(lI.gXf** (12.50)*” (14.31)“*

GCAP I.536 1.565 1.642
(3.14)** (C-).36)** (4.59)**
GLAB -0.114 -0.110 -0.076
(-0.40) (-0.40) (-0.37)
GREX 0.062 0.062 0.085
(3.01)** (3.38)X” (5.27)**
GRIM 0.002 -0.007
(0.07) - (-0.32)
Time -0.002 -0.002 -0.002
(-4.72F (-5.81)** t-6.32)*”

was discussed in Section II. To examine the extent of simultaneity in single equation
models (1) and (4), a simultaneous equations model is specified that incorporates
aspects of both the Esfahani (1991) and Sprout and Weaver (1993) models described
earlier.24
The model is specified as follows:

GGDP = q-j 4” a 1 GCAP + a*GLAB f n3GREX + ~4GR~~ + a+

GCAP = b. + hl GPCY + b2PCY + b3GREX + ~~FURCAP
GREX = co + c i GGDP + c2RER + c~T~GR~WTH + r4t (71
GRIM = do + dlGGDP + d2RER + d3PCY + d4FORCAP + d5t

In addition to those variables introduced previously, GPCY is per capita GDP growth,
PCY is just per capita GDP, FORCAP is a measure of foreign capital inflows, RER is the
real exchange rate, and TPGROWTH is the growth rate of real GDP of the United States,
Mexico’s principal trade partner (See Appendix for details),
Table 5 presents the complete results of estimating model (7) using three-stage least
squares. All variables with unit roots, as signalled by the ADF and KPSS tests, were first
differenced.‘s Since the dependent variable and at1 but one of the right hand side variables
in the second equation were found to have unit roots, all variables in the second equation
were differenced. Notice that the cmfficient for GREX in the first equation of the model is
highly significant and positive. In fact, the value of the CREX coefficient is larger and
more significant than in the single equation regressions given in Table 4. (The simulta-
 12 VANDENBERG

TABLE 5. Mexico: Simultaneous Equation Model of Trade and Output 3SLS

Estimation Using Annual Observations for 19fiO-1991

GRGDP = 0.080 + 1642GCAP - 0.076GLAB + 0.085GREX - 0.007GRIM - 0.002t

(14.31) (4.59) (-0.37) (5.27) (-0.32) (-6.32)
GCAP =-0.006+O.O16Gpcy +O.O004PCy - 0.003GREX +0.226FORCAP
(-0.30)(0.50) (9.12) (-0.37) (5.94)
GREX = 0.006 + 3.017GRGDP - 0.106m - 0.200TPGROWTH + 0.01 It
(0.27) (3.16) (-1.69) (-0.20) (3.39)
GRIM = -0.231 +6.091GRGDP - 0.067= - 0.004m +5622FORCAP + 0.015t
(-1.14) (2.13) (-1.30) (-0.85) (7.41) (3.72)

Nofe.~ Values in parentheses are r-values.

Bold. underlined variables are first differences.

neous-equations regression results are also given in column 3 of Table 4 for easy compar-
ison to the single-equation results derived earlier). Simultaneity bias, if it is indeed present,
seems to cause single-equation estimates to understate, not overstate, the relationship
between trade and growth.
Summarizing our results so far, time series evidence suggests that the relationship
between trade and growth has been positive in Mexico over the period 1960-1991. The
results are robust across both single-equation and simultaneous-equations regressions.

VI. THE RELATIONSHIP BETWEEN TRADE

AND FACTOR PRODUCTIVI~
It was noted earlier, in Section II.A., that specification (1) provides for an indirect test of
the relationship between trade and total factor productivity growth. A more direct altema-
tive test is proposed here. It consists of a two-step procedure that first derives estimates of
TFP growth and then applies both causality tests or regression analysis to measure the rela-
tionship between the estimated TFP and export growth. Total factor productivity is calcu-
lated according to

GTFP = GRGDP - aGCAP - (1 - a)GUB, (8)

which is just a restatement of the “sources of growth” Equation (2). The GCAP and GLAB
variables are again the more accurate estimates described in the previous section. The
availability of these improved measures permits us to estimate Mexican total factor pro-
ductivity for the period 1960- 199 1 (see also the Appendix).

A. VAR Causality Study

The shortcomings of simple two-variable Granger causality tests have been noted earlier.
If causality tests are to be conducted at all, it is better to use a VAR model that incorporates
all the relevant explanatory variables, although even such a more elaborate model still
lacks a solid theoretical foundation.
The growing literature on endogenous growth theory suggests a great number of possible
explanations for productivity growth, such as research and development expenditures, edu-
International Trade and Economic Growth in Mexico 13

TABLE 6. Trade and Total Productivity Causality Tests from the VAR
Mexico: 1960-l 991

II. SpeciJied According to KPSS Tests (All DifSerenced, 3 Lags):

GREXDIF + GTFPDIF 25.122** GTFPDIF + GREXDIF 2.925
GRIMDIF + GTFPDIF 7.663* GTFPDIF + GRIMDIF 6.790*
FDIDIF _j GTFPDIF 42.310** GTFPDIF -+ FDIDIF 21.565**
GCAPDIF + GTFPDIF 43.320** GTFPDIF + GCAPDIF 22.901**
Variables are described in the Data Appendix.
Test statistics are Wald x2 statistics.
The number of lags was determined according to the Akaike Information Criterion (AK).
** 99% significance level.
* 90% significance level.

cation, learning-by-doing, infrastructure investment, and government policies to promote

research and new product development, in addition to international trade.26 Investment has
often been distinguished as an important medium for introducing new technology, innova-
tions, and improvements in productive efficiency. Foreign direct investment (FDI) has
been singled out for its role in facilitating international transfers of technology.27 To cap-
ture these influences, the growth of the capital stock, GCAP, and the ratio of FDI to GDP,
denoted as FDI/Y, are included in the VAR along with GREX, GRIM, and GTFP.
The ADF tests reveal no unit roots for GTFP, GREX, GRIM, or FDIA’. Both the ADF
and KPSS tests confirm unit roots for the GCAP series, and the KPSS test confirms a unit
root for the GTFP series as well. While it would be statistically consistent to run a “mixed”
VAR model, with some variables differenced and others in levels, there is no obvious the-
oretical justification for mixing levels and changes. Thus, the model is run using all first
differences, which means that only the short run relationships among the variables will be
revealed. The estimation results are given in Table 6. GREXDIF and GRIMDIF both cause
GTFPDIF.

B. Trade and Productivity in a Single-Equation Regression Model

A theory-based regression model is preferable to the somewhat “mechanical” VAR
model. A regression model also captures the contemporaneous relationships among the
variables. The following regression model is therefore estimated for Mexico using time-
series data for 1960- 199 1:

GTFP = a0 + a,GREX + a2GRIM + a3GCAP + a,FDUY, (9)

As pointed out in the above section, the ADF unit root tests in Table 2 confirm that all
variables except GCAP are stationary, and the KPSS tests confirm unit roots for both
GCAP and GTFP, the dependent variable. Consistent estimates can be generated if all the
variables in Equation (9) are differenced, but again only the short run relationship between
the explanatory variables and GTFP would be revealed. Table 7, part 1, presents the results
of running model (9) entirely in first differences. Only the coefficient for GRIM is statisti-
cally significant.
The results from the single-equation model (9) of trade and factor productivity conflict
with the single and simultaneous-equations regression results from Section V. Those
14 VANDENBERG

TABLE 7. Models of Total Factor Productivity Growth Specified According

to Unit Root Tests Mexico: Annual Observations for 1960-1991
I. Linear Regression, all variables differenced’
GTFP = -0.001 - 0.004GREX + 0.072GRIM + 0.077GCAP + O.l94FDI/Y
(-0.36) (-0.12) (3.77)** (0.25) (0.14)
2. Simultaneous-Equations Model, all variables differenced
GTFP = -0.00 1 + 0.037GREX + 0.07 1GRIM + 0.542GCAP + 0.467FDYY
(-0.09) (1.64)* (4.45)** (1.50)* (0.40)
GCAP = -0.005 + 0.199GPCY + O.OOOlPCY - 0.023GREX + 0.746FDIJY
(-3.27)** (4.73)** (4.99)** (-3.39)** (2.30)**
GREX = 0.0001 + 2.458GTFP + 0.016RER + 0.937T~ROWTH
(0.001) (2.19)** (0.18) (0.87)
GRIM = 0.059 + 7.707GGDP - 0.145RER - O.OOlPCY - 10.56FDIN
(1.41)C (7.25)** (- 1.30) (-1.81)” (-1.26)
FDI/Y = 0.0001 + 0.252GTFP + 0.253GCAP - 0.276GRGDP
(0.86) (2.91)** (3.59)** (-3.19)“”

Nores: Figures in parentheses are t-statistics.

* Slgnlflcant at 95% level.
** Significant at 99% level.
’ Allowance for serial correlation in model estimation

results confirmed a positive relationship between trade, represented by both exports and
imports, and the growth of real GDP.

C. Trade and Productivity in a Simultaneous-Equations Regression Model

Further econometric analysis may be able to clear up the conflicting results. As before,
we should be wary of testing a complex relationship, such as that between trade and pro-
ductivity, using a single-equation model. There is likely to be some simultaneity among the
variables included in Equation (9).2x These potential simultaneities and interrelationships
can be captured in a five equation model that uses Equation (9) as its first equation and adds
four more equations to “explain” the right-hand side variables in (9):

GTFP = a0 + ai GCAP + a2FDUY + a3GREX + a&RIM

GCAP = b, i- b,GPCY + b2PCY + b,GREX + b4FDIff
GREX = co i- qGTFP -t c,RER + c~TPGR~WT~ (10)
GRIM = do + dlGGDP + dzRER + d3PCY + d4FDI/Y
FDIff = e. + e I GTFP + qGCAP + e3GRGDP

Given the number of variables for which the ADF and KPSS tests confirmed unit roots,
the model (10) is estimated entirely in first differences in order to avoid a confusing mix-
ture of levels and differences. Results are reported in Table 7, part 2. Comparing the
results in part 2 for Equation (9) when it is estimated as part of the simultaneous-equa-
tions model (10) to the single-equation results in part 1, note that now both GREX and
GRIM are statistic~ly significant at the 90% level or better. Thus, the conflict appears to
be solved.
/nternational Trade and Econurnic Growth in Mexico 15

The relationships between exports and imports on the one hand and the growth in output
and productivity on the other are thus both confirmed using simultaneous-equations time-
series regression models, Since these models most thoroughly address the methodological
shortcomings of previous statistical models, it is quite conceivable that the “mixed” out-
comes of previous studies were the result of methodological shortcomings, not the lack of
a significant relationship between trade and growth in Mexico.

VW. SUMMARY AND CONCLUSIONS

Numerous econometric studies have tested the relationship between inte~ation~ trade and
economic growth. The first part of this paper focused on the studies that generated results
relevant to Mexico. While a vast number of cross-section studies overwhelmingly confirm
that growth is positively related to trade, country-specific causality tests and time-series
regressions for Mexico do not present clear and consistent results.
This study extends the previous empirical studies in an attempt to establish firmer empir-
ical support for Mexico’s decisive shift away from import substitution and toward free
trade. The time-series approach is used exclusively, and recent advances in time-series
analysis are applied throughout. Improved data series for capital and labor are used to fit
regressions based on an underlying production function. Regression analysis is performed
using both singIe-equation models and simultaneous-equation models. The results are
clear: over the period 1960-1991 trade and real output growth were positively related in
Mexico.
The improved measures of capital and labor conveniently permit the calculation of a
time-series of Mexican total factor productivity (TFP). These series then permit direct tests
of the hypothesis that total factor productivity serves as the channel through which trade
stimulates growth. A VAR model confirms that exports or imports cause total factor pro-
ductivity growth, but a single equation regression model confirms only that imports are
related to TFP growth. The problem appears to be related to the complex relationships
among the variables, which a single equation model cannot accurately capture. When a
simultaneous-equations model is estimated, both exports and imports significantly explain
the growth of total factor productivity in Mexico.
Before jumping to the conclusion that the positive statistical results support Mexico’s
recent pro-trade policies, the limits of the econometric analysis presented in this article
must be recognized. First of all, there has been a general shift in the composition of Mexi-
can exports over the past three decades. Petroleum exports came to dominate in the late
197Os, and the rapid growth of the “maquiladora” industries along Mexico’s border with
the United States in the 1980’s greatly increased the labor component of the value added of
Mexican manufacturing exports. According to the literature on “enclave development,” the
petroleum industry, and perhaps the maquiladora plants, may not generate the same “spill-
over” effects on technology and efficiency throughout Mexico’s economy as trade in other
types of goods.29 Hence, the relationship between trade and growth is likely to have varied
considerably over the full 1960-1991 period.
There is yet another reason why the relationships between trade and output or trade and
productivity are likely to have changed over time: policy shifts. The 1960- 1991 period cov-
ers three different policy regimes-the “stabilizing development” period from the mid-
1950s through 1970, the “populist” years from 1970 through 1982, and the adjustment, or
“reform”, period starting after 1982.30 The first two periods are characterized by protec-
tionism, government regulation, and high levels of government investment relative to pri-
vate investment. During the latter period, trade policy was liberalized, government
investment was cut back, and most government-owned firms were privatized. The statisti-
cal evidence may correctly describe the “average” relationship for 1960- 1991, but to base
policy on these results would be misleading.
Empirical results based on data from the import substitution period of Mexican devel-
opment cannot reveal the present links between trade, output, and pr~uctivity in Mex-
ico. This, of course, is the well-known Lucas (1976) critique of econometric policy
analysis: in the case of a regime change, statistical analysis based on historical data can-
not provide reliable info~ation about the relationships between variables after the
regime change. Xt is likely that the relationship between trade and growth is now, under
Mexico’s current open trade policies, probably even stronger than the statistical results
reveal. But, because the empirical results cannot confirm the present or future relation-
ship between trade and growth, the role of economic theorists remains most important.
While the econometric evidence is certainly supportive of Mexico’s recent shift in trade
policy, Mexico’s economic policy must clearly also be based on forward-looking theo-
retical reasoning.

APPENDIX

Data and Suurces

The International Monetary Fund’s intermztional Financial Statistics (IFS), CD-RUM
version, March, 1995, was the source for GGDP (series 99b.p), GPCY and PCY (99b.p and
99z), GREX and GRIM (series 70 and 71, given in U.S. dollars and divided by the U.S.
GDP deflator), FORCAP (exports minus imports plus factor payments; 98c minus 9Oc plus
90e), RER (series ae and GDP deflators), and TPGROWTH (US. real GDP growth, fram
series 99b.p).
GCAP for 1960-1989 is from Hofman (1992), 1990 and 1991 are extrapolated using
investment (series 93e from the IFS) and a ten-year moving average rate of depreciation
implied by Hofman’s GCAP estimates and published investment data from the IFS.
GLAB is the “economically active populations’ from CELADE, ~e~~~r~~~~~ ~~~~er~~,
Vol. 18 and Vol. 25. adjusted according to

LADJ= (ae”)MO + .5(aeM)MY + (aPf)FO + .5(aef)FY. (Al)

The a’s represent the rise in productivity per year of education (set at 1.1, in line with data
in World Bank (199 1), em and ef are the average years of education for men and women,
respectively (from data in UNESCO, Statistical Yearbook, Paris: UNESCO, various
issues), and MO, MY, FO, and FY are older (over 25) and younger (15-24) males and
females. Young workers are assumed half as productive as older workers.
Alpha is based on national income data in: Economic Commission for Latin America
and the Caribbean, Statisficat Ye~r~u~k~~r Latin America and the Carihttean (Santiago:
ECLA), various issues.
GTFP was calculated using Equation (8). GTFP averaged a mere 0.07% per year over
196091, although it varied considerably. It averaged 2,27% over 1960-69, -0.34% over
1970-79, and a dismal -1.73% after 1980.
Acknowledgments: James Schmidt provided frequent and valuable advice during the writing of
this paper.

NOTES
1. For example, assume output, Y, is a function of the stock of capital, K, and labor, .&,according
to the function Y = @K-L“-. Expressing Yin logarithms and differentiating yields Equation (2).
2. The coefficient al in Equation (3) is the marginal product of capital, not the shnre of capital as
in specification ( 1).
3. This has been pointed out by, among others, Kavoussi (1984).
4. Dornbusch fl993:87).
5. Edwards (1992) specifies a model very similar to Equation (1) except that, in addition to a mea-
sure of trade restrictiveness, which serves as a negative measure of TRADE, he also adds a variable
to proxy the technology “gap” between the developing economy and developed economies in order
to capture the “catch-up’” effect.
6. See Eaton and Cersovitz (1980). Khan and Knight (1988), and Esfahani (1991).
7. See, for example, Dollar (1992), Edwards (1988), Feder (1982). Greenaway and Nam (1988)*
Kavoussi (1984), Moschos (1989), Ram (1985, 1987), and Tyler (198 I ). See aiso the excellent sur-
vey by Edwards (1993).
8. The Lucas (1976) critique further warns us against making policy inferences based on econo-
metric results. This issue wili be dealt with again at the end of the article, after the more technical
econometric probtems have been addressed.
9. Balassa (1978) was an exception, however, and be compared his “average” cross-section results
to specific country experiences. He showed that during 1966-1973 the increase in real Mexican GNP
would have been 8% greater had its exports grown at the rate of the “average” country instead of at
Mexico’s actual rate.
10. Van den Berg and Schmidt (1994).
11. VAR models, even if they include a large number of variables, still lack a theoretical founda-
tion. But even the modest step of using larger VAR models tq test causality has not yet been under-
taken for Mexico.
12. For Mexico, Van den Berg and Schmidt found different patterns of stationary and nonstation-
ary variables, depending on which of two alternative unit root tests were used. In one case, the
regression (3) was estimated with the export growth and GDP growth variables in levels, and the
coefficient for export growth was positive and significant. In the second case, three of the variables,
including growth of exports and GDP, were found to be non$tationa~ and cointegrated. An error
correction model was used in place of Equation (3), but again the coefficient for export growth was
also positive and significant. Their results were thus robust to the particular unit root test used.
13. Sheehey (1992:733).
14. Causality studies have also been called on to shed light on the simultaneity issue. Chow
(1987) finds that causality is often bi-directional, and Jung and Marshall (1985) find that causality
runs from growth to trade more often than from trade to growth This may imply that the relationship
between exports and growth is a complex one that simply cannot be captured in a single-equation
model. La1 and Rajapatirana (1987:195), however, claim that “the results showing that output
growth causes export growth are not inconsistent with the export-growth link found by the more
conventional studies,” The reversed ‘%ausality” is interpreted by them as a further confirmation that
trade is positively related to growth because it causes shifts in resources in accordance with compar-
ative advantage.
15. A simultaneous equations model is attractive for other reasons as well. Levine and Renelt
(1992) point out that the effects of trade and investment are often difficult to disentangle in empirical
studies. Taking a thearetical perspective, Fagerberg (1994) suggests that the various sources of
18 VANDENBERG

growth as given in a model such as Equation (4) may be interrelated, especially investment and total
factor productivity.
16. Two types of trends, deterministic and stochastic, can cause nonstationarity in economic time-
series. A time-series is said to have a dere~inistic trend if it can be represented as a function of
time. Such a series can easily be rendered stationary by standard detrending procedures and, as a
result, is occasionally referred to as a trend stationary series. On the other hand, a time-series has a
srochastic trend if it contains an unpredictable random component that is permanently incorporated.
A series that possesses a stochastic trend but whose first differences are stationary is said to be inte-
grated of order one, denoted as I( I).
17. Known as the spurious regressions problem, it was popularized and studied extensively by
C.W.J. Granger and Paul Newbold (1974).
18. See Dickey and Fuller (1979).
19. The KPSS test is based on the regression

Y[ = a, + P,+ e1’

where t is a time trend, e, is a random error, and cx, follows a random walk

with uI being a random error having variance 62U. The null hypothesis of a deterministic trend in
the variable requires that 02U equal zero, in which case CI,becomes a constant, say CLu.If the null is
rejected, then we conclude that there is a unit root with drift in the variable.
20. The simple two-variable causality studies by Jung and Marshall (1985) Dodaro (1993), and
Chow (1987) used only the stationary GREX and GRGDP series and were not biased.
21. Hofman’s (1992) 1950-1989 estimates are generated using the perpetual inventory method
and annual investment figures. Figures for 1990 and 1991 are generated using investment data and
the depreciation rate implied by a ten year moving average of Hofman’s data.
22. The lack of causality between trade and growth found in the VAR estimates may indicate that
the relationship is contemporaneous rather than lagged, further strengthening the case for regression
analysis.
23. If the nonstationary variables are cointegrated, differencing will actually cause estimates to be
inconsistent. Nonstationary I( 1) variables are said to be cointegrated if there is a linear combination
of them that is stationary. In such a case, there is a stable relationship between the variables that can
be called a long-run equilibrium path. Estimation of the model in levels in this case will provide con-
sistent estimates, but only the long-run relationship will be revealed. These points are not relevant to
the analysis here, however, because our regression combines both stationary and nonstationary I( 1)
variables which cannot be cointegrated.
24. Both import and export growth variables are included in the production function equation, as
in the Esfahani model. GCAP is explicitly determined, as in the Sprout and Weaver model. And, the
reverse influence of GGDP on exports and imports is tested, as in both the Sprout and Weaver and
Esfahani models.
25. RER is differenced even though only the ADF test signals a unit root,
26. Evaluations of endogenous growth theory’s accomplishments and shortcomings are discussed
in Grossman and Helpman (1994), Pack (1994), and Romer (1994).
27. See Cardoso and Dombusch (1989).
28. See Fagerberg (1994).
29. See for example, Myrdal (1956), Caimcross (1962), or Best (1968).
30. For a description of these policy regimes, see C. Bazdresch and S. Levy (1991) or Lustig
(1992).
International Trade and Economic Growth in Mexico 19

REFERENCES
Aspe, Pedro. 1993. Economic Transformation the Mexican Way. Cambridge, MA: MIT Press.
Balassa, Bela. 1978. “Exports and Economic Growth: Further Evidence.” Journal of Development
Economics 5: 181-189.
Bahmani-Oskooee, Mohsen, and Janardhanan Alse. 1993. “Export Growth and Economic Growth:
An Application of Cointegration and Error-Correction Modeling.” The Journal of Developing
Areas 27: 53.5542.
Bazdresch, Carlos, and Santiago Levy. 1991. “Populism and Economic Policy in Mexico, 1970-
1982.” Pp. 223-262 in The Macroeconomics of Populism in Latin America, edited by R. Dorn-
busch and S. Edwards. Chicago: University of Chicago Press.
Best, Lloyd. 1968. “The Mechanism of Plantation Type Economies: Outline of a Model of Pure Plan-
tation Economy.” Social and Economic Studies 17: 283-326.
Caimcross, A.K. 1962. Factors in Economic Development. London: George Allen & Unwin.
Cardoso, Eliana, and Rudiger Dombusch. 1989. “Foreign Private Capital Flows.” Chapter26 in
Handbook of Development Economics, vol. 2, edited by Hollis B. Chenery and T.N. Srinivasan.
Amsterdam: North-Holland.
Chow, Peter C.Y. 1987. “Causality Between Export Growth and Industrial Development.” Journal of
Development Economics 26: 55-63.
Collins, Susan M., and Won-Am Park. 1989. “External Debt And Macroeconomic Performance in
South Korea.” Chapter 6 in Developing Country Debt and the World Economy, edited by Jeffrey
D. Sachs. Chicago: The University of Chicago Press.
Dickey, David A., and Wayne A. Fuller. 1979. “Distributions of the Estimators for Autoregressive
Time Series with a Unit Root.” Journal of the American Statistical Association 74: 427-43 1.
Dodaro, Santo. 1993. “Exports and Growth: A Reconsideration of Causality.” Journal of Developing
Areas 27: 227-244.
Dollar, David. 1992. “Outward-Oriented Developing Economies Really Do Grow More Rapidly:
Evidence from 95 LDCs, 1976-1985.” Economic Development and Cultural Change 40(3): 523-
544.
Dombusch, Rudiger. 1993. Stabilization, Debt, and Reform. Englewood Cliffs, NJ: Prentice Hall.
Eaton, J., and M. Gersovitz. 1980. “LDC Participation in International Financial Markets.” Journal
of Development Economics 5: 3-21.
Edwards, Sebastian. 1988. Exchange Rate Misalignment in Developing Countries. Baltimore: The
Johns Hopkins University Press.
Edwards, Sebastian. 1992. “Trade Orientation, Distortions and Growth in Developing Countries.”
Journal of Development Economics 39( 1): 3 l-57.
Edwards, Sebastian. 1993. “Openness, Trade Liberalization, and Growth in Developing Countries.”
Journal of Economic Literature 31(3): 1358-1393.
Engle, R.F., and C.W. Granger. 1987. “Cointegration and Error Correction: Representation, Estima-
tion and Testing.” Econometrica 55: 25 1-276.
Esfahani, Hadi S. 1991. “Exports, Imports, and Economic Growth in Semi-Industrialized Countries.”
Journal of Development Economics 35: 93-116.
Fagerberg, Jan. 1994. “Technology and International Differences in growth Rates.” The Journal of
Economic Literature 32: 1147-l 175.
Feder, Gershon. 1982. “On Exports and Economic Growth.” Journal of Development Economics 12:
59-73.
Fuller, W.A. 1976. Introduction to Statistical Time Series. New York: Wiley.
Giles, E.A., Judith A. Giles, and Ewen McCann. 1992. “Causality, Unit Roots and Export-Led
Growth: The New Zealand Experience.” The Journal of International Trade & Economic Devel-
opment l(2): 195-218.
Granger, C.W.J., and Paul Newbold. 1974. “Spurious Regressions in Econometrics.” Journal of
Econometrics 2: 11 1- 120.
20 VANDENBERG

Greenaway, David, and Chong Hyun Nam. 1988. “Industrialization and Macroeconomic Perfor-
mance in Developing Countries under Alternative Trade Strategies.” Kyklos 41: 419-435.
Greenaway, David, and David Sapsford. 1994. “What Does Liberalization Do for Exports and
Growth?’ Weltwirfschufrliches Archiv 129(2): 153-173.
Grossman, Gene M., and Elhanan Helpman. 1994. “Endogenous Innovation in the Theory of
Growth.” Journal of Economic Perspectives 8( 1): 23-44.
Hofman, Andre A. 1992. “Capital Accumulation in Latin America: A Six Country Comparison for
1950-89.” Review of Income and Wealth 38(4): 365401.
Jung, Woo S., and Peyton J. Marshall. 1985. “Exports, Growth and Causality in Developing Coun-
tries.” Journal of Development Economics 18: 1-12.
Kavoussi, Rostam M. 1984. “Export Expansion and Economic Growth, Further Empirical Evi-
dence.” Journal of Development Economics 14: 241-250.
Khan, M.S., and M.D. Knight. 1988. “Import Compression and Export Performance in Developing
Countries.” Review of Economics and Statistics 70: 315-321.
Krueger, Anne. 1980. “Trade Policy as an Input to Development.” American Economic Review
70(2): 288-292.
Kwiatkowski, Denis, Peter C.B. Phillips, Peter Schmidt, and Yongcheol Shin. 1992. “Testing the
Null Hypothesis of Stationarity against the Alternative of a Unit Root.” Journal of Econometrics
54: 159-178.
Lal, Deepak and Sarath Rajapatirana. 1987. “Foreign Trade Regimes and Economic Growth in
Developing Countries.” World Bank Research Observer 2(2): 189-217.
Learner, Edward E. 1985. “Vector Autoregressions for Causal Inference?’ Carnegie Rochester
Series on Public Policy 22. Amsterdam: North Holland.
Levine, Ross, and David Renelt. 1992. “A Sensitivity Analysis of Cross-Country Growth Regres-
sions.” American Economic Review 82: 942-963.
Lucas, Robert E. Jr. 1976. “Econometric Policy Evaluation: A Critique.” Carnegie Rochester Con-
ference on Public Policy 1: 19-46.
Lustig, Nora. 1992. Mexico: The Remaking of an Economy. Washington, DC: The Brookings Institu-
tion.
Mehra, Yash P. 1991. “An Error Correction Model of U.S. M2 Demand.” Federal Reserve Bank of
Richmond’s Economic Review May/June: 3-12.
Michaely, Michael. 1977. “Exports and Growth.” Journal of Development Economics 4: 49-53.
Moschos, Demetrios. 1989. “Export Expansion, Growth and the Level of Economic Development,
An Empirical Analysis.” Journal of Development Economics 30: 93-102.
Myrdal, Gunnar. 1956. Development and Underdevelopment. National Bank of Egypt Fiftieth Anni-
versary Commemoration Lectures. Cairo: National Bank of Egypt.
Pack, Howard. 1994. “Endogenous Growth Theory: Intellectual Appeal and Empirical Shortcom-
ings.” Journal of Economic Perspectives 8( 1): 55-72.
Ram, Rati. 1985. “Exports and Economic Growth: Some Additional Evidence.” Economic Develop-
ment and Cultural Change 33: 415-25.
Ram, Rati. 1987. “Exports and Growth in Developing Countries: Evidence from Time- Series and
Cross-Section Data.” Economic Development and Cultural Change 36(l): 51-63.
Romer, Paul M. 1994. “The Origins of Endogenous Growth.” Journal of Economic Perspectives
8( 1): 3-22.
Ruffin, Roy J. 1993, August. “Endogenous Growth and International Trade.” Research Paper No.
9332, Federal Reserve Bank of Dallas.
Salvatore, Dominick, and Thomas Hatcher. 1991. “Inward Oriented and Outward Oriented Trade
Strategies.” The Journal of Development Studies 27(3): 7-25.
Sheehey, Edmund J. 1992. “Exports and Growth: Additional Evidence.” The Journal of Development
Studies 28(4): 730-734.
Sprout, Ronald V.A., and James H. Weaver. 1993. “Exports and Economic Growth in a Simultaneous
Equations Model.” Journal of Developing Areas 27: 289-306.
Tyler, William G. 198 1. “Growth and Export Expansion in Developing Countries.” Journal of Devel-
opment Economics 9: 12 1- 130.
Van den Berg, Hendrik, and James R. Schmidt. 1994. “Foreign Trade and Economic Growth: Time
Series Evidence from Latin America.” The Journal of International Trade & Economic Develop-
ment 3(3): 249-268.
Van den Berg, Hendrik. 1996. “Libre Comercio y Crecimiento: La Evidencia EconomCtrico para
America Latina.” Comercio Exterior 46(5): 364-373.
World Bank. 1993. The East Asian Miracle. New York: Oxford University Press.