SSRN Id2087930

Discussion Paper
No. 2012-25 | May 22, 2012 | http://www.economics-ejournal.org/economics/discussionpapers/2012-25
Openness to International Trade and Economic

Growth: A Cross-Country Empirical Investigation
Bülent Ulaşan
Central Bank of the Republic of Turkey
Abstract In this paper, we revisit the empirical evidence on the relationship between
trade openness and long-run economic growth over the sample period 1960-2000. In
contrast to previous studies focusing mainly on the period 1970-1990, this paper
reassesses the openness-growth nexus over a much longer sample period, enabling us
to better account both trade policy stance and long-run growth dynamics. We carry out
our empirical investigation by employing various openness measures suggested in the
literature rather than relying on a few proxy variables. We also construct three
additional composite trade policy indexes directly measuring trade policy stance. Our
findings indicate that many openness variables are positively and significantly
correlated with long-run economic growth. However, in some cases, this result is
driven by the presence of a few outlying countries. Adding to the fragility of the
openness-growth association, the significance of openness variables disappears once
other growth determinants, such as institutions, population heterogeneity, geography
and macroeconomic stability are accounted for.
JEL F13, F43, O47.

Keywords Economic Growth; trade openness; outliers; weighted least squares
Correspondence Bülent Ulaşan, Research and Monetary Policy Department, Central

Bank of the Republic of Turkey, Istiklal Cad. No:10, Ulus, 06100, Ankara, Turkey; e-
mail: [email protected]
The views expressed in this paper are solely the author's and do not reflect those of the
Central Bank of the Republic of Turkey. The author is grateful to Wyn Morgan, Peter
Wright and Mehmet Eris for their helpful comments and suggestions, but all remaining
errors are the author's own.
© Author(s) 2012. Licensed under a Creative Commons License - Attribution-NonCommercial 2.0 Germany
1 Introduction
Does openness to international trade boost economic growth in the long run?
Although this is one of the oldest questions in economics, the existing theory
does not provide a decisive answer.1 Therefore, the openness-growth nexus
is basically an empirical question and has been extensively investigated by
empirical cross-country work dating back to 1970s. This issue especially
attracted renewed interest since the early 1990s, with almost all studies find-
ing a strong and statistically significant positive relationship between trade
openness and economic growth.2
However, the cross-country growth literature is still far from settled since
the findings of this literature have been subject to an important criticism
in terms of robustness. In particular, Edwards (1993) and Rodrik and
Rodrı́guez (2000) argue that the strong results in favour of openness may
arise from model misspecification and/or openness measures may be acting
as a proxy for other macroeconomic policies or other important factors such
as institutions and geography. In a nutshell, it is fair to say that the cross-
country studies suffer from lack of robust and convincing evidence on the
openness-growth connection and this issue is still highly controversial.
An outstanding but generally neglected feature of the empirical literature
is that the substantial part of existing studies solely focuses on the period
of 1970-1990. Although data availability is an important reason for this,
investigating growth and openness link over the 1970-1990 period is trouble-
some: The first problem is that a time period of 20 years is rather limited to
fully reflect long-run growth dynamics. Secondly, and more importantly, the
sample period from 1970 to 1990 is inappropriate as most of the developing
countries followed protectionist trade polices not only during the 1970s but
1
The traditional Ricardian-Hecksher-Ohlin trade theory points out that openness to
international trade brings only a one-time increase in output, yet does not suggest any
certain implications for long-run growth. The neoclassical growth model concludes that
the long-run growth rate of per capita output is determined by the exogenous technological
progress. Only the newer endogenous growth theories pay attention to implications of trade
openness on growth in the long run since openness facilitates the transmission of technology
by providing communication with foreign counterparts, directs domestic resources towards
more research intensive sectors and increases market size (see Rivera-Batiz and Romer
(1991) and Grossman and Helpman (1991, Chapters 6 and 9)). However, these models do
not necessarily predict that openness leads to economic growth in all circumstances and
for all countries. In other words, whether openness causes economic growth in the long
run depends on country specific conditions.
2
Examples include Dollar (1992), Edwards (1992, 1998), Lee (1993), Sachs and Warner
(1995), Harrison (1996), Vamvakidis (1999), Frankel and Romer (1999), Greenaway et al.
(2002), Yanikkaya (2003), Lee et al. (2004), Aksoy and Salinas (2006), Foster (2008),
Kneller et al. (2008), Wacziarg and Welch (2008), Chang et al. (2009), Kim (2011).
2
also during 1960s and it is more likely that their trade policy measures did
not change substantially over the period 1960-1980. However, most of them
experienced relatively higher growth performance during the 1960s. There-
fore the empirical evidence based on the sample period 1970-1990 is highly
likely to be biased since it does not include the growth information of 1960s.
In this paper, we revisit the empirical evidence on the relationship be-
tween trade openness and long-run economic growth within the sample period
1960-2000. We do so by extending the augmented neoclassical growth model
developed by Mankiw, Romer and Weil (1992) with a measure of openness.
We carry out our empirical investigation by employing various openness mea-
sures suggested in the literature rather than relying on a few proxy variables
since the empirical studies fail to provide satisfactory openness measures and
their findings are very sensitive to employed openness variables as pointed
out by many authors.3 In doing so, we classify openness measures under
four broad categories: trade volumes, direct trade policy measures, deviation
measures, and subjective indexes.
Among these openness measures, trade volumes (conventionally expressed
as the ratio of exports plus imports to GDP) is the most problematic one, at
least conceptually since we define openness as removing or reducing policy
barriers to international trade rather than trade intensity. It is obvious that
a country’s trade volume is affected not only by trade policy but also by other
factors such as country size, distance to trade partners, transportation costs,
world demand and so on. This implies that direct trade policy measures such
as tariffs, non-tariff barriers are ideal measures to capture a country’s degree
of openness to trade. The main problem with these measures is, however, that
they do not pick up differences in trade policy barriers across countries and
hence testing openness-growth connection by using a single policy variable
may be misleading. That’s why we need reasonable weights for aggregation
of different policy instruments if we want to investigate openness growth link
by employing a composite indicator directly addressing and encompassing
every aspect of trade policy.
We attempt to construct such a measure in this paper. We compute three
composite trade policy indexes consisting of weighted averages of tariff rates,
non-tariff barriers and black market premium. Weights are estimated using
the models in which both nominal and real trade volumes as a share of GDP
are regressed on the initial level of income, country size and trade policy
instruments.
The contributions of this paper are twofold: First, in contrast to previous
studies, mainly focusing on the period 1970-1990, this paper analyses the
3
See, for instance, Pritchett (1996) and Rodrik and Rodrı́guez (2000) amongst others.
3
openness-growth link over a much longer time period. In other words the
sample period is sufficiently long in order to account for both trade policy
stance and growth dynamics in the long run. Second, we employ a myriad
of openness measures suggested in the literature. Providing a wider picture,
this enables us to better evaluate both existing openness variables and the
openness-growth connection. We construct three additional composite open-
ness variables arguably better capturing trade policy stance. We also check
whether our findings are driven by outlying countries and carry out a sen-
sitivity analysis, accounting for other important growth theories in order to
check the robustness of our basic findings.
The cross-country empirical investigation in this paper indicates that
many openness variables are positively and significantly correlated with long-
run economic growth. However, in some cases, this result depends on the
presence of a few outlying countries. More importantly, the significance of
openness variables disappears once other growth determinants, such as eco-
nomic institutions, population heterogeneity, geography and macroeconomic
stability related to government consumption are accounted for.
The structure of this paper is as follows. Section 2 provides the ba-
sic framework for the empirical cross-country investigation of the openness-
growth connection. Section 3 presents OLS estimates based on the cross-
country data over the period 1960-2000. Section 4 examines how robust our
findings are to alternative model specifications. Finally, Section 5 summarises
our results and concludes.
2 Economic Growth and Openness to Inter-

national Trade: Baseline Model
We adopt the empirical framework of the augmented neo-classical growth
model suggested by Mankiw et al. (1992) for investigation openness-growth
link. In other words, we extend this model by adding a proxy variable for
openness as follows:
log yi (t) − log yi (0) = γ0 + γ1 log yi (0) + γ2 log(ni + g + δ)

+ γ3 log si,K + γ4 log si,H + γ4 OPi + εi (1)
where, yi and (ni + g + δ) denote the level of GDP per worker and the
sum of rates of population growth, technological progress and depreciation
in country i, respectively. Similarly, the terms si,K and si,H represent the
4
rates of accumulation of both physical and human capital for country i,
respectively. Finally, the term OP indicates country i’s degree of openness.
Following MRW, we assume that the sum of rates of depreciation and
technological progress is constant and equal to 0.05 across countries. We
measured si,K by the ratio of real investment to real GDP and si,H by the
secondary school gross enrolment rate.4 Data are compiled from standard
sources: GDP per capita and investment rates are taken from the Penn World
Tables Version 6.1 (Heston, Summers and Aten, 2002); population, labour
force and gross secondary school enrolment rates come from the World Bank
World Development Indicators (2002, 2006). Using labour force as the total
population between ages 15 and 64, GDP per capita is converted to GDP per
worker. All of these variables are averaged over the period 1960-2000 except
the initial level of income. The variables and their sources are detailed in the
Appendix.
In summary, we estimate the following cross-country growth regression in
this paper:
log yi,2000 − log yi,1960 = γ0 + γ1 log yi,1960 + γ2 log(ni + g + δ)

+ γ3 log(Investment ratei )
(2)
+ γ4 log(School enrolment ratei )
+ γ5 Opennessi + εi
Once we specify the cross-country growth regression expressed in equa-

tion (2), the problem of simultaneity arises immediately: Does openness leads
to economic growth or vice versa? A country may abandon protectionist or
inward-looking trade polices after specialising in sectors with the dynamic
comparative advantage in terms of rapid productivity growth and large tech-
nological spillovers. In this case, causality occurs from economic growth to
trade openness. The openness variable in equation (2) is, therefore, likely
to be correlated with the error term, εi . This means that the OLS esti-
mate of γ5 will be biased and the direction of causality between growth and
openness will remain unclear. However, overcoming this problem is not easy
because finding a valid instrumental variable (IV) is almost impossible in the
cross-country growth works. The reason for this is that growth theories are
4
Using the school enrolment rate as a proxy for the saving rate of human capital is
problematic in the literature and leads researchers to employ average years of schooling
as more reliable variables for human capital (see, for instance Bils and Klenow (2000),
amongst others). We, however, employ the secondary school enrolment rate in order to
follow the theoretical framework suggested by Mankiw et al. (1992) more strictly as years
of schooling are a stock rather than a flow variable for human capital. In addition, school
enrolment rates are available for a larger sample of countries.
5
openended, that is they are complementary and hence there is a wide range of
different explanations for growth such as economic and political institutions,
trade openness, geography, culture and so on. Therefore, finding an instru-
ment which is not a direct growth determinant and/or correlated with other
omitted growth determinants is extremely difficult.5 More importantly, if IV
is not valid, the coefficient estimate will be again biased and in this case the
OLS estimate would be more preferable as argued by Durlauf et al. (2005).
It may be, therefore, possible to conclude that the cross-country growth
regressions can never reveal the direction of causality. Despite this fact,
the cross-country works still provide useful information between growth and
a variable of interest. Even if we can not establish the casuality between
growth and openness, a statistically significant partial association can be used
to reject alternative hypotheses which fail to provide statistically significant
correlation and one can provide plausible causal statement (Mankiw (1995),
Wacziarg (2002)). For instance, if we conclude that OLS estimate of γ5
is positive and statistically significant and fail to conclude the statistically
significant negative association between openness and growth, then it is not
reasonable to reach an inference such that openness is harmful for economic
growth. Similarly, Warner (2003) argues that it is very difficult to attribute
a positive coefficient estimate on openness variable to reverse casuality from
growth to trade polices. The reason is that there are no specific cases in
which countries opened to international trade, grew slowly and then closed
again because of poor economic performance.
3 Empirical Results
In this section we present and discuss the findings of our empirical inves-
tigation under four categories of openness measures: Before evaluating the
regression results, we want to emphasise two points about the regressions.
First, in each regression we check the normality assumption applying median
5
Mankiw (1995, p.303) points out “[W]hen looking for instruments, it is easy to fall prey
to temptation.” Durlauf, Johnson and Temple (2005, p.) argue that “[t]he belief that it is
easy to identify valid instrumental variables in the growth context is deeply mistaken.” In
some studies the lagged values of endogenous variables are used as instruments. However,
lagged values of endogenous variables do not guarantee that they are directly uncorrelated
with growth and hence they are proper instruments. The reason is that many growth
variables such as measures of educational attainment affect growth with a substantial
delay. More importantly, even if they are valid instruments, whether the instrumental
estimate shows the effect of endogenous variable or of lagged value of that variable on
economic growth remain unanswered. According to Mankiw (1995) the answer is generally
neither.
6
and inter quartile range comparison suggested by Hamilton (1992) which is
originally based on Hoaglin, Iglewicz and Tukey (1986) on regression residu-
als and conclude that residuals are normally distributed. Therefore, we may
assume that actual errors are normally distributed (at least approximately).
Second, in each regression we also check the constant error variance as-
sumption by employing the Breusch-Pagan test for heteroscedasticity and
carry out regression analysis employing t-statistics based on the usual stan-
dard errors unless we reject the homoscedasticity assumption. We report t-
statistics based on the heteroscedasticity consistent (White-robust) standard
errors only for the regressions in which the assumption of homoscedastic error
variance is rejected.6
3.1 Trade Volumes

We start our estimations with the ratio of trade volume to GDP. We obtain
two measures for this variable: one is from the World Bank and the other is
from Penn World Tables (Version 6.1). One advantage of the World Bank
measure is that the data are published in terms of exports and imports.
Thus, this allows us to investigate the export-growth connection and import-
growth connection separately. On the other hand the trade ratio of the Penn
World Tables is published only as a sum of exports and imports at current
prices. This data have the advantage that they are available for a larger
number of countries. This is a very common proxy in the cross-country
growth literature and some times referred to as “current openness”. The
simple correlation between these two trade ratios is almost one (Table 1).
6
The common practice in cross-country growth literature for dealing heteroscedastic-
ity is reporting regression results with the heteroscedasticity consistent (White-robust)
standard errors since they work well regardless of heteroscedasticity in the actual errors.
However, these errors are consistent but not unbiased. More clearly they are justified
only asymptotically. In small samples, heteroscedasticity consistent standard errors may
have distributions that are not close to those of usual standard errors which means that
they may be larger or smaller than the usual ones. As pointed out by Wooldridge (2003)
heteroscedasticity consistent standard errors are generally found to be larger than the
usual standard errors. This can affect the subsequent statistical inference such that one
can conclude that a variable is statistically insignificant according to t-test based on the
heteroscedasticity consistent standard errors even if that variable is significant (at least
marginally) in the case of usual t-test. Therefore, there is no reason to use heteroscedas-
ticity consistent standard errors as long as the homoscedastic error variance assumption
holds and the errors are normally distributed.
7
Table 1: Economic Growth and Trade Volumes: Pairwise Correlations
1960-2000 averages
Average GDP per Exports Imports Trade Current Real Trade

growth worker ratio ratio Ratio open. open. ratio with
in 1960 OECD
8
Average growth 1.000
GDP per worker in 1960 0.137 1.000
Exports ratio of WB 0.359 0.231 1.000
Imports ratio of WB 0.288 0.004 0.840 1.000
Trade ratio of WB 0.335 0.120 0.958 0.960 1.000
Current openness of PW 0.344 0.152 0.958 0.962 0.996 1.000
Real openness by A&C 0.426 0.381 0.846 0.721 0.812 0.868 1.000
Trade ratio with OECD 0.280 0.184 0.851 0.850 0.876 0.909 0.815 1.000
Trade ratio with NonOECD 0.276 0.036 0.672 0.661 0.687 0.880 0.759 0.687
Table 2 provides the estimation results. Columns 1 and 2 show the regres-
sion results using the ratio of exports and the ratio of imports, respectively.
Column 3 includes the trade ratio as a sum of the ratio of exports and the
ratio of imports. In each regression the coefficient of the openness variable is
to be found positive but not significant. Therefore, a significant association
between growth and openness is not established using the World Bank data.
However, we find that coefficient estimate of openness is positive and highly
significant when employing current openness of the Penn World Tables in
column 4. The regression results indicate a 10 percent increase in trade ra-
tio would raise the growth by 2.73 percent over the 1960-2000 period. This
is equivalent to 0.07 percent increase in annual growth rate over the same
period (2.73/40=0.07). In column 5 we estimate our baseline model with
the variable of real openness from the Penn World Tables. This variable is
suggested by Alcalá and Ciccone (2004) and defined as the ratio of exports
plus imports relative to GDP in constant prices. Alcalá and Ciccone (2004)
argue that real openness is a better measure of openness compared to current
openness in the presence of trade-driven productivity. As can be seen, we
conclude positive and strongly significant coefficient estimate. Furthermore,
the estimated coefficient of real openness is higher compared to that of cur-
rent openness and implies that a 10 percent increase would raise the growth
rate 3.99 percentage points over the period 1960-2000 (which is equal to 0.10
percent increase in annual growth rate for the same period).
We estimate our benchmark model adding trade ratio with OECD and
trade ratio with non-OECD countries in columns 6 and 7. The rationale
is straightforward as Yanikkaya (2003) points out. In the light of recent
endogenous growth theories countries, particularly the developing ones, can
benefit more from trade with technologically advanced countries in order to
stimulate growth. If this argument is true then one would expect a higher
coefficient estimate of trade ratio with the OECD with respect to that of
the trade ratio with non-OECD countries. However, the regression results
in columns 6 and 7 show that the coefficient of the trade ratio with OECD
is not only less than the coefficient of trade ratio with non-OECD but also
is statistically insignificant. These results confirm the findings of Yanikkaya
(2003) and imply that technology spillover effects of international trade on
economic growth are not very important compared to the effects of compar-
ative advantage and scale economies.
In summary, the regression results in Table 2 show a positive association
between economic growth and international trade and confirm the findings
of previous work (Vamwakidis (2002), Dollar and Kraay (2003), Yanikkaya
(2003), Alcalá and Ciccone (2004) are a few examples). However, it is puz-
zling that we could not find any statistically significant relation between
9
Table 2: Economic Growth and Trade Volumes: OLS Estimates†
(1) (2) (3) (4) (5) (6) (7)

log GDP per worker 1960 -0.471 -0.456 -0.462 -0.429 -0.464 -0.454 -0.434
(6.46) (6.16) (6.31) (6.46) (6.97) (5.81) (5.66)
log(ni + g + δ) -1.108 -1.063 -1.080 -1.100 -1.021 -1.153 -1.286
(2.66) (2.52) (2.58) (2.93) (2.77) (2.97) (3.38)
log of Investment rate 0.410 0.400 0.403 0.360 0.338 0.507 0.468
(3.80) (3.66) (3.72) (3.98) (3.74) (4.68) (4.38)
log of School enrolment 0.436 0.453 0.445 0.430 0.438 0.391 0.390
(4.28) (4.46) (4.38) (5.11) (5.28) (3.89) (4.00)
Exports ratio of WB 0.388 - - - - - -
(1.58)
Imports ratio of WB - 0.316 - - - - -
(1.37)
Trade ratio of WB - - 0.185 - - - -
(1.51)
Current Openness of PW - - - 0.273 - - -
(2.85)
Real Openness by A&C - - - - 0.399 - -
(3.29)
Trade ratio with OECD - - - - - 0.358 -
(1.74)
Trade ratio with NonOECD - - - - - - 0.787
(2.80)
Constant 2.730 2.734 2.725 2.244 2.731 2.603 2.018
(2.23) (2.22) (2.22) (2.08) (2.60) (2.20) (1.71)
Number of observations 93 93 93 105 105 89 89
p-value for heteroscedasticitya 0.95 0.94 0.95 0.64 0.81 0.43 0.69
Adjusted R2 0.60 0.60 0.60 0.63 0.64 0.64 0.66
Note: t-statistics are in parenthesis. In the regressions where the heteroscedasticity test is
failed to pass at 15 % level t-statistics based on heteroscedastic-consistent (White-robust)
standard errors are reported.
†
Dependent variable is the log difference of real GDP per worker between 1960 and 2000.
a
Breusch-Pagan test for heteroscedasticity in which the null refers to the homoscedastic
errors.
10
trade ratios of the World Bank and growth despite the very high correlations
between these measures and current openness.7 The reason is that regres-
sions including trade ratios of World Bank have smaller samples than those
including current openness and real openness. More clearly, some countries
are dropped from the regressions when we use the World Bank data and
hence this implies that the positive and statistically significant relation be-
tween growth and trade ratios of Penn World data may be driven by outlying
countries.
When we look at the data of current openness, we highlight that Singa-
pore has the highest trade ratio with a value of 323 percent. This country
not only has the highest trade ratio but also records the highest growth rate
over the 1960-2000 period. However, Singapore is missing in the trade ra-
tios of World Bank since data for this country are available only over the
1974-2000 period. When we add Singapore to the regressions based on the
trade ratios of the World Bank using the average values over the 1974-2000
period instead of the 1960-2000 period, we find a positive and statistically
significant relationship between growth and all three trade ratios of World
Bank.8 This finding clearly indicates that Singapore is a highly influential
country in the cross-country regressions in Table 2.
In order to check possible outliers we apply the method suggested by Hadi
(1992) on each data set subject to cross-country growth regressions in Table 2.
We identify four countries namely Singapore, Hong Kong, Luxembourg and
Tanzania as outliers in the regressions including current openness and real
openness. Except Tanzania, these countries are clearly outliers in current
openness and/or real openness (See figures 1(a) and 1(b) for a graphical
inspection).9 Their outstanding characteristics are that they have the highest
trade ratios with an average value of 244 percent according to the current
openness and experience very high growth performances over the sample
period. In addition, we also identify Hong Kong and Luxembourg as outliers
for the data set including trade ratios of the World Bank. A sample of
countries based on the trade ratios with the OECD and non-OECD concludes
that only Singapore for the data set including trade ratio with the OECD and
five countries namely Congo Democratic Republic, Hong Kong, Singapore,
Malaysia and Hungary for the data set including trade ratio with non-OECD
as outlying observations.
7
For a comparison see Table 1. As can be easily seen, the correlation between current
openness and real openness is smaller than the correlations between current openness and
trade ratios of the World Bank.
8
Regression results are available from the author on request.
9
Tanzania has the smallest school enrolment rate in our sample and hence it is an
outlier in the saving rate for human capital.
11
Singapore
1
Hong Kong
Luxembourg
Growth Rate (Unexplained Part)
−1 −.5 0
−1.5 .5
−1 0 1 2 3
Current Openness
coef = .27289843, se = .0956058, t = 2.85
(a) Current Openness: Exports plus Imports as a ratio of GDP in

current prices
Singapore
1
Hong Kong
Luxembourg
Growth Rate (Unexplained Part)
−1 −.5 0
−1.5 .5
−.5 0 .5 1 1.5 2
Real Openness
coef = .39876586, se = .12103841, t = 3.29
(b) Real Openness: Exports plus Imports as a ratio of GDP in PPP
Figure 1: Partial Associations between Growth Rate and Trade Shares
When we drop all outlying countries from our regressions, we find that
all openness variables except trade ratios with the OECD and non-OECD
12
are negative but statistically insignificant (Table 3). Therefore, our findings
such that both current openness and real openness are significantly and pos-
itively correlated with growth may not be robust. As shown in Table 3 after
dropping outliers we conclude very similar findings to our first result for the
trade ratio with the OECD such that this variable is still positive but not
significant. Contrary to our first findings, we find a significantly negative
coefficient estimate of the trade ratio with non-OECD which implies that
trading with non-OECD countries, and hence technologically less advanced
countries, is not beneficial for economic growth.
However, omitting outliers from the sample may not be a good solution
for cross-country growth analysis as Temple (2000) points out. Instead of
dropping these countries an alternative and better way is to carry out the
analysis by employing weighted least squares (WLS).10 Therefore, we esti-
mate the cross-country growth regressions in Table 2 employing Iteratively
Reweighted Least Squares (IRLS).11 However, the most important shortcom-
ing of IRLS is that it is not robust to high leverage data points. This implies
that estimation results will be more robust if one can identify and drop the
observations with the high leverage value and then estimate the sample of
remaining observations by IRLS. Hence, before applying IRLS, we firstly
highlight high leverage countries in each data set. To do so, following Huber
(1981) we determine the countries as risky whose leverage values are greater
than 0.2. In the second step, we delete these risky countries from each sample
and then estimate the cross-country growth regressions by IRLS.
Table 4 provides IRLS regression results. The first four columns of the
table shows that both trade ratios of the World Bank and current openness
are significantly and positively associated with economic growth. However,
we could not conclude the same thing for the real openness. As seen in column
5 of Table 4, the coefficient estimate of real openness is positive but not
statistically significant. In columns of 6 and 7, we report regressions including
the trade ratios with OECD and non-OECD, respectively and conclude that
both are positive. Yet, trade ratios with OECD is not found to be statistically
significant.
In summary, we conclude a positive and significant association between
10
Indeed, omitting some observations from OLS regression is another kind of WLS
estimation such that we assign each observation a weight of 1 if it is included in the
regression or a weight of 0 if omitted. In contrast to these extreme weights, WLS gives
each observation a weight between 1 and 0 such that outlying cases are down-weighted
gradually.
11
IRLS is based on iterative computation of case weights obtained from the residu-
als. Weight functions for the observations are first Huber weights and second the Tukey
bisquare weights.See Hamilton (1992) for more information about the IRLS.
13
Table 3: Economic Growth and Trade Volumes: OLS Estimates without
Outlying Countries†
(1) (2) (3) (4) (5) (6) (7)

(6.83) (6.67) (6.76) (6.81) (6.64) (5.97) (7.72)
log(ni + g + δ) -1.170 -1.175 -1.174 -1.159 -1.167 -1.269 -1.429
(2.86) (2.84) (2.86) (3.16) (3.17) (3.30) (4.20)
(3.62) (3.62) (3.63) (3.70) (3.66) (4.30) (3.53)
(4.71) (4.76) (4.75) (5.00) (4.95) (4.15) (5.77)
Exports ratio of WB -0.073 - - - - - -
(0.24)
Imports ratio of WB - -0.044 - - - - -
(0.17)
Trade ratio of WB - - -0.031 - - - -
(0.21)
Current Openness of PW - - - -0.071 - - -
(0.49)
Real Openness by A&C - - - - -0.024 - -
(0.12)
Trade ratio with OECD - - - - - 0.132
(0.58)
Trade ratio with NonOECD - - - - - - -1.299
(2.09)
Constant 2.862 2.855 2.860 2.406 2.318 2.327 2.467
(2.36) (2.35) (2.36) (2.25) (2.19) (2.00) (2.34)
p-value for heteroscedasticitya 0.58 0.60 0.59 0.76 0.84 0.74 0.47
Adjusted R2 0.61 0.61 0.61 0.62 0.62 0.63 0.69
†
a
errors.
14
Table 4: Economic Growth and Trade Volumes: IRLS Estimations†
(1) (2) (3) (4) (5) (6) (7)

(7.99) (7.31) (7.60) (7.79) (7.97) (7.64) (7.24)
log(ni + g + δ) -1.157 -1.115 -1.124 -1.302 -1.215 -1.280 -1.307
(3.28) (3.07) (3.14) (4.05) (3.77) (3.87) (3.95)
(4.33) (3.98) (4.08) (3.90) (3.57) (4.34) (4.55)
(5.13) (5.32) (5.26) (5.53) (5.65) (5.30) (5.07)
Exports ratio of WB 0.564 - - - - - -
(2.72)
Imports ratio of WB - 0.385 - - - - -
(1.93)
Trade ratio of WB - - 0.247 - - - -
(2.36)
Current Openness of PW - - - 0.231 - - -
(2.27)
Real Openness by A&C - - - 0.209 - -
(1.29)
Trade ratio with OECD - - - - - 0.279 -
(1.42)
Trade ratio with NonOECD - - - - - - 0.771
(2.07)
Constant 2.741 2.668 2.693 1.805 2.234 2.592 2.423
(2.64) (2.51) (2.56) (1.95) (2.41) (2.59) (2.35)
Adjusted R2 0.69 0.67 0.68 0.69 0.67 0.70 0.70
Note: t-statistics are in parentheses.

†
trade volume and economic growth and except real openness, confirm the
findings of previous literature. However, the ratio of trade volume to GDP
is not a good proxy to measure trade openness.12 First, some large countries
may appear closed economy by this measure. For instance, according to the
current openness the United States and Japan are the fourth and seventh
12
The ratio of trade volume in GDP can be defined as follows:
X +M
openness =
A+X −M
where X is exports, M is imports and A is domestic absorbtion (sum of consumption and
investment). Assume that the world economy consists of two identical countries. In this
setting the country running trade deficit will be more open than the other country.
15
most closed economies in our sample. Second, the volume of international
trade is affected not only by trade policies but also by other factors such as
transportation costs, world demand, geography, natural resource dependence
and so on. Hence, it is likely that a positive association between trade to
GDP ratio and growth rate implies the impact of international trade rather
than the effect of trade polices on economic growth.
3.2 Direct Trade Policy Measures

In the second step we investigate the openness-growth connection by em-
ploying direct trade policy measures namely tariff rates, non-tariff barriers
on imports and black market premium for exchange rate. It is obvious that
the first two measures directly affect a country’s trade volume and reducing
or removing them clearly indicates a more open trade regime. Yet, the use
of the black market premium as a measure of trade barriers is debatable in
the literature. Authors such as Sachs and Warner (1995) and Warner (2003)
argue that a high black market premium causes increases in the price of im-
ports relative to domestic prices and thus plays the same role with tariff and
non-tariff barriers. Others, such as Rodrik and Rodrı́guez (2000) claim that
a high black market premium generally shows macroeconomic imbalances
rather than trade policy barriers.
Undoubtedly, directly addressing trade policy both tariff rates and non-
tariff barriers are ideal measures in order to capture a country’s degree of
openness to international trade. Yet, unfortunately we do not have good
data for these measures across countries and over time. The most common
data for these measures in the literature come from the data set by Barro
and Lee (1994). In this data set, tariff rate and non-tariff coverage ratios
are weighted averages by the product shares in a country’s overall imports.
However, both variables cover the only imports of intermediate inputs and
capital goods over the 1983-1985 period. In addition, they have a problem of
downward bias since both measures are weighted by their own-import value.
Finally, it is not likely that all non-tariff barriers can be measured and they
accurately reflect the intensity of non-tariff barriers.
Employing these two measures as an openness variable, we estimate our
baseline model. The estimation results are reported in Table 5. In columns
1 and 2 of Table 5, we only include tariff rate and non-tariff barriers, respec-
tively. Both measures enter the regressions with negative but insignificant
coefficient estimates. In column 3, we allow both tariff rate and non-tariff
barriers together. As can be seen, tariff rate and non-tariff barriers are nei-
ther individually nor jointly significant. In the fourth column, we replace
tariff rate with the collected import duties as a ratio of imports over the
16
1970-1998 period and find a positive but insignificant coefficient estimate. It
is well-known fact that the ratio of collective import duties in a country’s
overall imports is a problematic measure in order to reflect a country’s tar-
iff structure due to the fact that a country with very high tariff rates may
appear open by this measure. In column 5, we include unweighted average
tariff rate over the 1990-2000 period that is provided by Wacziarg and Welch
(2008). The difference between this measure and tariff rate of Barro and Lee
(1994) is that the former is simply averages of ad valorem tariff rates across
commodities subjected to imports. The estimated coefficient of unweighed
tariff rates is negative but again statistically insignificant.
We include average black market premium over the 1960-2000 period in
column 7 and find that the black market premium is negatively and signif-
icantly associated with economic growth. In columns 8 and 9, we replace
average black market premium with two dummy variables, respectively. The
first dummy variable takes the value of 1 if the average black market premium
exceeds 20 percent in the 1960s or the 1970s or the 1980s or the 1990s while
the second one is equal to 1 if the average black market premium is higher
than 20 percent over the 1960-2000 period. Our aim in constructing for these
dummies is to check the relation between growth and a larger level of black
market premium. Following Sachs and Warner (1995), we assume 20 percent
as a threshold level. As shown in the table, in each case the dummy variables
are negatively correlated with growth and strongly significant. In column 10,
we include tariff rate, non-tariff barriers and average black market premium
jointly. The result is essentially same. Both tariff rates and non-tariff mea-
sures are not significant but black market premium is. In the last column,
the regression is exactly the same as the regression in column 10, the only
difference is that we replace average black market premium with the dummy
variable for black market premium. As seen, the result is unchanged.
17
Table 5: Economic Growth and Direct Trade Measures: OLS Estimates†
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
log GDP per worker 1960 -0.485 -0.476 -0.497 -0.446 -0.479 -0.484 -0.499 -0.506 -0.493 -0.505
(5.82) (5.74) (5.78) (5.29) (6.14) (7.30) (7.24) (7.59) (6.88) (7.46)
log(ni + g + δ) -1.291 -1.274 -1.229 -1.109 -1.061 -1.092 -0.996 -1.043 -1.269 -1.194
(2.85) (2.77) (2.65) (2.60) (2.80) (3.09) (2.74) (2.98) (3.54) (3.39)
log of Investment rate 0.428 0.433 0.420 0.443 0.400 0.327 0.323 0.329 0.319 0.336
(4.20) (4.24) (4.07) (4.02) (4.14) (3.53) (3.42) (3.61) (2.49) (2.77)
log of School enrolment 0.423 0.426 0.440 0.445 0.448 0.475 0.495 0.483 0.441 0.436
(4.08) (4.12) (4.21) (4.14) (4.79) (5.57) (5.63) (5.71) (4.85) (5.06)
owtia -0.327 - -0.290 - - - - - -0.309 -0.096
(1.12) (0.93) (1.07) (0.33)
owqib - -0.116 -0.050 - - - - - 0.088 0.019
(0.60) (0.24) (0.48) (0.10)
18
Import Dutiesc - - - 0.377 - - - - - -
(0.44)
uwtid - - - - -0.477 - - - - -
(0.90)
log (1+BMP)e - - - - - -0.233 - - -0.224 -
(2.84) (2.26)
BMP dummyf - - - - - - -0.199 - - -
(2.32)
BMP dummyg - - - - - - - -0.263 - -0.275
(3.22) (3.01)
Constant 2.556 2.495 2.821 2.654 3.109 2.937 3.357 3.301 2.523 2.874
(1.96) (1.91) (2.08) (1.88) (2.56) (2.78) (3.00) (3.11) (2.14) (2.51)
Number of observations 87 85 85 93 101 101 101 101 83 83
p-value for heteroscedasticityh 0.82 0.70 0.83 0.59 0.81 0.30 0.47 0.39 0.12 0.11
F -valuei - - 0.61 - - - - - 2.09 3.51
Continued on Next Page. . .
Table 5 – Continued
2
Adjusted R 0.60 0.60 0.60 0.56 0.61 0.64 0.64 0.65 0.62 0.63
Note: t-statistics are in parenthesis. In the regressions where the heteroscedasticity test is failed to pass at 15 %
level t-statistics based on heteroscedastic-consistent (White-robust) standard errors are reported.
†
a
Own-import weighted tariff rates on intermediate inputs and capital goods.
b
Own-import weighted non-tariff frequency on intermediate inputs and capital goods.
c
Collected import duties as ratio of imports over 1970-1998 period.
d
Unweighted average tariff rate over the 1990-99 period.
e
Logarithm of one plus average value of black market premium over the 1960-1999 period.
f
Dummy variable is equal to 1 if the average black market premium exceeds 20 % in either the 1960s or the
19
1970s or the 1980s or the 1990s.
g
Dummy variable is equal to 1 if the average black market premium exceeds 20 % over the 1960-2000 period.
h
Breusch-Pagan test for heteroscedasticity in which the null refers to the homoscedastic errors.
i
Test for joint significance of openness variables.
As a result, our empirical investigation between growth and direct trade
policy measures indicates that the significant association with economic growth
is established only for the black market premium. However, Warner (2003)
points out that the time period should be 1970-1990 for testing the impact
of trade protection through tariff rates. The reason is that the majority of
developing countries have liberalised their trade regime during the late 1980s
and early 1990s. In other words, the large cross-country variation in tariff
rates in the earlier period was eliminated after the 1980s. In addition, he
indicates that India is a clear outlying observation. Therefore, according to
Warner (2003), one can find a negative and significant correlation between
growth and tariff rate if he omits India from the sample and estimates the
growth regression over the 1970-1990 period.
In order to test Warner’s claim we estimate our baseline model over the
period 1970-1990 for the same sample without India. The regression result
is given in the first column of Table 6 and shows a negative but statistically
insignificant coefficient estimate of the tariff rate. In addition to India, we
identify three more countries namely Burkina Faso, Guyana, and Tanzania
as outliers applying the Hadi methodology on the data set over the 1970-1990
period. In the second column of Table 6, we drop these countries as well as
India from the regression and conclude that tariff rate is again negative but
not significant. Therefore, our findings indicate that Warner’s claim is not
valid over the 1970-1990 period.
However, it may be more reasonable to investigate growth-tariff connec-
tion over the period 1960-2000. As we argue in the introduction, most of the
developing countries followed protectionist trade polices not only during the
1970s but also during 1960s and experienced relatively higher growth perfor-
mance in the 1960s. This means that the time period suggested by Warner
(2003) might be biased since it does not include the growth information of
1960s. Therefore, in column 3, we estimate our baseline growth model with
the tariff rate whilst dropping India over the 1960-2000 period and conclude
that coefficient of tariff rate is negative and significant at the 11 percent sig-
nificance level. In addition if we also omit Tanzania, the other outlier over
the sample period, from the regression in column 4, we find that tariff rate
is negative and significant at 9 percent significance level. It is possible to
conclude that tariff rate is negatively associated with economic growth over
the 1960-2000 period at the marginally significant level once we take into
account outlying countries.
In summary, our cross-country empirical investigation indicates that among
the direct trade policy measures only tariff rate and black market premium
are negatively correlated with economic growth. However, we conclude that
this correlation is marginally significant in the tariff case while strongly sig-
20
Table 6: Economic Growth and Tariffs: OLS Estimates under Different
Time Periods and without Outliers†
1970-1990 1960-2000
(1) (2) (3) (4)
log of Initial GDP per worker -0.289 -0.357 -0.488 -0.494
(4.64) (6.33) (5.87) (5.95)
log(ni + g + δ) -0.494 -0.818 -1.151 -1.209
(1.57) (2.69) (2.47) (2.59)
log of Investment rate 0.302 0.364 0.424 0.493
(4.57) (5.40) (4.17) (4.30)
log of School enrolment 0.193 0.174 0.420 0.355
(2.92) (2.51) (4.07) (3.10)
owtia -0.360 -0.444 -0.725 -0.762
(1.12) (1.48) (1.64) (1.73)
Constant 2.280 2.140 3.006 2.990
(2.45) (2.46) (2.22) (2.21)
Number of observations 86 83 86 85
p- value for heteroscedasticityb 0.36 0.21 0.86 0.80
Adjusted R2 0.41 0.54 0.60 0.61
†
a
b
errors.
nificant for the black market premium.
3.2.1 Black Market Premium: A Proxy for Trade Policy or Macroe-

conomic Imbalances?
As mentioned above, whether the significant association between black mar-
ket premium and economic growth reflects the relationship between growth
and restrictive trade policies or the connection between growth and other
poor macroeconomic polices rather than trade polices is debatable. Now, we
consider this point further.
Our data for black market premium come from Global Development Net-
work (2005). Unfortunately, we lack data on the black market premium
for several countries in the 1960s. In addition, for many countries data are
missing during the mid 1990s. In order to increase the number of observa-
tions, we take the averages of black market premium for the 18 countries
21
over the 1970-2000 period instead of 1960-2000 period. Of course, this ap-
proach implicitly assumes that for these countries the black market premium
in the 1970s reflects the black market premium in the 1960s. Indeed, this
assumption is not very realistic since we do not observe a certain pattern on
the black market premium during the 1960s and the 1970s for the countries
whose data are available in both decades. Among the 103 countries we iden-
tify, 43 of them experience a higher level of black market premium in the
1970s compared to the 1960s. Most of these countries are located in Africa
and Latin America. On the other hand only 39 countries mostly located in
the Middle East, North Africa, East Europe, Asia and Pacific have a lower
level of black market premium in the 1970s with respect to the 1960s. 21
developed countries have zero black market premium in both decades.
Table 7: Black Market Premium (%): Summary Statistics
Number of Mean Standard Minimum Maximum

Observations value Deviation value value
The Largest Sample
BMP 1960-2000 121 259.701 2114.377 -0.442 23235.250
BMP in 1960s 103 48.685 228.780 -0.090 2276.400
BMP in 1970s 121 36.319 72.469 -6.960 451.790
BMP in 1980s 121 129.106 600.604 -1.410 6406.610
BMP in 1990s 121 1205.328 12627.680 -0.350 138935.900
The 2nd Largest Sample
BMP 1960-2000 103 297.5641 2291.072 -0.442 23235.250
BMP in 1960s 103 48.68505 228.7795 -0.090 2276.400
BMP in 1970s 103 39.39447 77.2462 -1.290 451.790
BMP in 1980s 103 139.4033 648.8455 -1.410 6406.610
BMP in 1990s 103 1411.573 13686.08 -0.350 138935.900
The Regression Sample
BMP 1960-2000 101 66.520 199.778 -0.442 1796.679
BMP in 1960s 90 47.521 241.983 -0.090 2276.400
BMP in 1970s 101 32.399 64.711 -6.960 451.790
BMP in 1980s 101 137.674 654.984 -1.410 6406.610
BMP in 1990s 101 30.501 124.572 -0.350 1199.310
The Smallest Sample
BMP 1960-2000 90 67.668 209.826 -0.442 1796.679
BMP in 1960s 90 47.521 241.983 -0.090 2276.400
BMP in 1970s 90 33.895 67.323 -0.870 451.790
BMP in 1980s 90 142.192 692.166 -1.410 6406.610
BMP in 1990s 90 32.662 131.741 -0.350 1199.310
Table 7 provides summary statistics of black market premium for each

decade. As can be seen, for the first and largest sample in Table 7, both the
22
mean and standard deviation of black market premium is higher in the 1960s
compared to the 1970s. In order to provide a better comparison we also report
the summary statistics of black market premium for 103 countries whose data
are available in each decade (the second largest sample). Again both the
mean and standard deviation of the black market premium is higher in the
1960s compared to the 1970s. Therefore, it is likely that we underestimate
average black market premium over the 1960-2000 period by using the 1970-
2000 averages for the countries whose data are missing during the 1960s.
However, since we conclude that a negative and highly significant association
between black market premium and economic growth over the 1960-2000
period, this bias in the data makes our result stronger.
The other important point is that the mean of the black market premium
is substantially higher in the 1990s compared to the other decades in the
largest sample. At the first sight, this might be thought to be surprising be-
cause most of the developing countries have liberalised their capital accounts
since the late 1980s and one would expect very low black market premium for
these countries during the 1990s. However, this is mainly a result of a small
number of countries with the extreme values of black market premium in this
decade such as Iran, Iraq, Afghanistan, Liberia, Syria and Libya. As shown
in Table 7, not only is the mean value of black market premium substantial
but also its standard deviation is high during the 1990s. Of course, from
our point summary statistics of regression sample are of great concern rather
than those of the largest samples. When we consider only the regression
sample, both mean and standard deviation of black market premium in the
1980s are considerably higher than other decades. In addition, the statis-
tics of black market premium based on the smallest sample consisting of the
countries whose data are available in each decade support this fact. This im-
plies that the negative and statistically significant association between black
market premium and economic growth over the 1960-2000 period may be as
a result of both the high level and variation of the black market premium
during the 1980s.
Therefore, in Table 8 we estimate our baseline model with the averages of
black market premium in each decade. In column 1, we allow average black
market premium in each decade to vary continuously and conclude that none
of them are statistically significant despite a negative sign. In addition, they
are jointly insignificant. However, the t-statistics of black market premium
in the 1980s is relatively higher. In columns 2-5, we insert average black
market premium in each decade separately and find that the only average
black market premium in the 1980s is negatively and significantly correlated
with growth. Therefore, it is possible to conclude that the significant and
negative correlation between black market premium and economic growth
23
Table 8: Economic Growth and Black Market Premium: OLS Estimates†
(1) (2) (3) (4) (5)

log GDP per worker 1960 -0.514 -0.507 -0.483 -0.456 -0.475
(7.14) (6.39) (8.00) (6.99) (6.92)
log(ni + g + δ) -1.032 -1.180 -1.222 -1.008 -1.070
(2.94) (3.05) (3.64) (2.85) (2.91)
log of Investment rate 0.294 0.324 0.341 0.336 0.362
(2.69) (3.17) (3.14) (3.68) (3.84)
log of School enrolment 0.511 0.506 0.452 0.453 0.470
(5.93) (5.11) (6.05) (5.36) (5.34)
log (1+BMP) in 1960s -0.113 -0.132 - - -
(0.76) (1.16)
log (1+BMP) in 1970s -0.068 - -0.244 - -
(0.26) (1.16)
log (1+BMP) in 1980s -0.130 - - -0.186 -
(1.35) (3.07)
log(1+BMP) in 19990s -0.034 - - - -0.181
(0.28) (1.52)
Constant 3.318 2.855 2.560 2.925 2.941
(2.84) (2.53) (2.50) (2.79) (2.64)
Number of observations 90 90 101 101 101
p-value for heteroscedasticitya 0.08 0.53 0.11 0.21 0.91
F -valueb 1.81 - - - -
Adjusted R2 0.62 0.61 0.63 0.65 0.62
†
a
errors.
b
Test for joint significance of log of average black market premium in the decades.
over the 1960-2000 period mainly depends on the high level and high varia-
tion in the black market premium during the 1980s in which many developing
countries launched the liberalisation programs after the debt crises in the late
1970s and the early 1980s. Hence, it is more likely that negative and signif-
icant connection between black market premium and economic growth over
the period 1960-2000 reflects the adverse relation between macroeconomic
24
Table 9: Economic Growth and Black Market Premium Dummy: OLS
Estimates†
(1) (2) (3) (4) (5)

(6.24) (6.43) (7.63) (8.46) (7.96)
log(ni + g + δ) -1.117 -1.139 -1.181 -1.001 -1.266
(3.64) (2.97) (3.39) (3.18) (3.67)
(2.70) (3.10) (3.75) (3.21) (4.25)
(5.32) (5.11) (5.64) (5.97) (5.36)
BMP dummy in 1960sa -0.026 -0.137 - - -
(0.21) (1.26)
BMP dummy in 1970sa -0.150 - -0.261 - -
(1.34) (3.17)
BMP dummy in 1980sa -0.103 - - -0.275 -
(0.85) (3.38)
BMP dummy in 1990sa -0.173 - - - -0.385
(1.01) (3.65)
Constant 3.555 3.007 3.022 3.138 3.052
(3.11) (2.62) (2.88) (3.06) (2.96)
p-value for heteroscedasticityb 0.09 0.58 0.18 0.12 0.50
F -valuec 3.09 - - - -
Adjusted R2 0.65 0.61 0.65 0.66 0.66
†
a
Dummy variable is equal to 1 if the average black market premium exceeds 20 %.
b
errors.
c
Test for joint significance of the black market premium dummies.
imbalances and growth. In Table 9 we exactly replicate the cross country

growth regressions in Table 8. The only difference is that in Table 9 we use
a dummy variable which takes the value of 1 if the average value of black
market premium exceeds 20 percent. As can be seen, the dummy variables
25
are continuously and separately negative and significant, except the dummy
in the 1960s. This implies that a higher level of black market premium is
particularly harmful for economic growth.
In conclusion, our findings about the relationship between economic growth
and direct trade policy measures are not in favour of more liberal trade poli-
cies. We find evidence for the adverse growth-tariff connection, but the
tariff rate is only marginally significant. More to the point this finding im-
plies that imposition of tariffs on intermediate inputs and capital goods is
harmful for growth rather than the negative impact of all kind of tariffs on
economic growth as our tariff variable includes only imports of intermediates
and capital goods. On the other hand our finding concerning the statistically
significant association between black market premium and economic growth
may indicate the negative relation between growth and macroeconomic im-
balances rather than the trade restrictive effect of black market premium.
3.3 Deviation Measures

Cross-country growth regressions including only single trade policy measures
have low-power for testing openness-growth connection. The reason is that
each trade policy measure takes into account only one aspect of trade policy
and hence tells only part of story (Leamer (1988)). Therefore, we need a
single general measure encompassing all aspects of trade policy for testing
the protectionism in general. As noted by Warner (2003, p.4) “The right
regression for testing the impact of protectionism would seem to entail some
aggregation of the policy instruments.”
One way in order to overcome this problem is employing deviation mea-
sures. These measures basically show the difference between predicted and
actual trade and hence can be used as an indicator of the overall level of
trade protectionism. Deviation measures are essentially based on the fol-
lowing rationale. According to the traditional small country model, inter-
national trade is determined by the factor endowments, international prices,
technology, preferences, natural trade barriers and trade policy instruments
(Leamer (1988)). Therefore, differences among the countries in the level of
trade can be considered as trade policy barriers if the countries are substan-
tially identical in terms of factor endowment, technology, preferences, and
natural barriers or their effects are controlled for. This implies that if one
constructs a model which sufficiently explains trade flows across countries,
then residuals obtained from that model can be considered as an overall mea-
sure of trade barriers subject to only trade policy. Of course this approach
implicitly assumes that trade policy barriers are the only important omitted
variables and they are uncorrelated with the other determinants of trade.
26
In the literature many deviation measures as an indicator of openness to
international trade have been suggested. In this respect, outward orientation
index by Syrquin and Chenery (1989) and predicted trade shares by Frankel
and Romer (1999) are the most well-known measures. In addition to these,
we also use estimated residuals from a very simple model as:
((X + M )/GDPi ) = α0 + α1 log(Initial per worker GDPi )

+ α2 log(Areai ) + α3 log(Average labour f orcei ) + εi
(3)
The model in (3) includes the dependent variable which is the exports plus
imports as a share of GDP and hence already takes into account the size of
country. We employ the real GDP per worker as a proxy for factor endow-
ments of country. In order to avoid the possible endogeneity problem we use
the 1960 value of per worker GDP. The other explanatory variables are the
land area and average labour force. Both variables are expressed in loga-
rithms and represent the country size. As pointed out by Frankel and Romer
(1999) country size is an important determinant of international trade due
to the fact that there are more opportunities for within country trade in the
larger countries.
The specification in (3) is of course very simple in many aspects. First,
the dependent variable is the average total trade as a ratio of GDP rather
than the sum of bilateral trades across countries. Obviously estimating total
trade as a sum of bilateral trades by employing a model including some grav-
ity variables such as distance between two countries, common border dummy
as well as other determinants would be better. However, unfortunately we
lack data on bilateral trades across countries over the period 1960-2000. Sec-
ond, this specification assumes that the only important omitted variables
are trade policy barriers. A better specification therefore would be to in-
clude trade policy barriers such as tariffs and non-tariff barriers on imports
(we will consider this point later). Finally we assume that preferences and
technology are constant among countries. Even though these are important
shortcomings for the model in (3), we believe that the estimated residu-
als from this model can be used as a more reliable indicator for openness
compared to simple actual trade ratios. Obviously, a large value of residual
implies that the country is more open to international trade once the initial
factor endowments and country’s size are controlled.
Employing the current openness as a dependent variable we estimate this
model by OLS over the 1960-2000 period. Column 1 of Table 10 presents the
resulting OLS estimate. In column 2, we also include a dummy variable for
27
Table 10: Exports plus Imports as a Share of GDP: OLS Estimates
Dependent variable† Current Openness Real Openness

(1) (2) (3) (4)
log GDP per worker 1960 0.061 0.050 0.150 0.150
(2.08) (1.59) (5.23) (4.70)
log Area -0.110 -0.106 -0.094 -0.094
(2.57) (2.40) (2.62) (2.54)
log Average Labour Force -0.049 -0.054 0.003 0.003
(1.52) (1.56) (0.11) (0.09)
Landlocked country dummya - -0.072 - -0.005
(0.91) (0.09)
Constant 2.218 2.354 0.222 0.231
(5.73) (5.59) (0.72) (0.67)
p-value for heteroscedasticityb 0.00 0.00 0.00 0.00
Adjusted R2 0.45 0.45 0.40 0.40
Note: t-statistics based on heteroscedastic-consistent (White-robust) standard errors are

in parenthesis.
†
1960-2000 averages.
a
The landlocked countries in Europe are dropped.
b
errors.
landlocked countries in order to control for natural trade barriers.13 However,

inserting the landlocked country dummy neither changes the basic results
nor improves the fit of the model. Therefore, we prefer to employ residuals
which come from the first regression as an openness variable and label it as
RESID Current Openness. In columns 3 and 4, we replicate the regressions
in the first two columns. The only difference is that in columns 3 and 4
we use the real openness as the dependent variable. As can be seen, the
regression results in columns 3 and 4 are different from those in columns 1
and 2 in some aspects: First, when the dependent variable is real openness
the coefficient estimate of initial income is considerably larger and highly
significant. Second, average labour force is not only significant but also very
close to zero. Third, the R2 values of the regressions in columns 3 and 4 is
13
While constructing this dummy variable, we omit the landlocked countries in Europe
such as Austria, Luxembourg, Switzerland and Hungary since becoming landlocked for
these countries may not create an important disadvantage.
28
smaller and hence it is possible to conclude that these regressions are less
precise. On the other hand, as in the case of current openness, including
a landlocked country dummy does not improve the goodness of fit of the
regression in column 3. Therefore, we again use the estimated residuals in
column 3 as an openness measure and label it as RESID Real Openness.
Table 11 reports the estimation results by employing deviation measures
in the framework of our baseline cross-country growth model. In column 1 we
include the outward orientation index by Syrquin and Chenery (1989) and
conclude that its coefficient estimate is negative but insignificant. However,
it is more plausible to estimate this index over the period 1960-1985 since
the index covers 1965-1980 period. Therefore, in column 2 we setup our
benchmark model over the 1960-1985 period and estimate it with the outward
orientation index. Now, coefficient estimate of index is positive but again
insignificant. In column 3 and 4 we include RESID Current Openness and
RESID Real Openness and conclude that both variables are positively and
significantly associated with growth. Notice that this association is stronger
for the RESID Real Openness, with the RESID Current Openness being only
marginally significant.
The regression in column 5 includes the Frankel-Romer predicted trade
shares from a gravity model. Since this variable is predicted for only 1985 the
cross-country growth regression covers 1960-1985 period. Estimation results
show that the Frankel-Romer predicted trade share is positive and statis-
tically significant. Recall that this variable is predicted trade shares from
a gravity model based on the geographical characteristics of the countries.
That is why this result can be only seen as an indication of a positive impact
of international trade on economic growth if one assumes that geography
influences growth only through international trade. In column 6, we esti-
mate our benchmark model with the Frankel-Romer predicted trade shares
over the 1960-2000 period. For this regression we employ the predicted trade
shares provided by Dollar and Kraay (2003). This variable is different from
the original Frankel-Romer predicted trade shares in two aspects: First it is
based on the data on bilateral trade in 1995; Second, it expresses the bilateral
trade in US dollar as a fraction of GDP at purchasing power parity (PPP)
US dollar rather than current prices. Therefore, this variable is identical to
real openness once the geographical characteristics of the countries are con-
trolled for. The regression results implies that the predicted trade share has
a positive and significant effect on economic growth. As seen, this effect is
substantially stronger than the effect found in the previous regression.
29
Table 11: Economic Growth and Deviation Measures: OLS Estimates†
(1) (2) (3) (4) (5) (6)

log GDP per worker 1960 -0.480 -0.331 -0.415 -0.418 -0.340 -0.406
(6.39) (5.87) (6.09) (6.16) (6.71) (6.22)
log(ni + g + δ) -1.199 -0.339 -1.198 -1.118 -0.191 -0.817
(3.00) (1.26) (3.03) (2.91) (0.74) (2.27)
log of Investment rate 0.415 0.296 0.359 0.349 0.302 0.397
(4.42) (5.16) (3.80) (3.66) (5.41) (4.52)
log of School enrolment 0.471 0.266 0.429 0.446 0.279 0.405
(5.11) (4.75) (4.99) (5.22) (5.29) (4.82)
Outward orientation by S&C -0.036 0.082 - - - -
(0.42) (1.29)
RESID Current Openness - - 0.268 - - -
(1.91)
RESID Real Openness - - - 0.323 - -
30
(2.07)
log F&R predicted trade in 1985 - - - - 0.091 -
(2.39)
log F&R predicted trade in 1995 - - - - - 0.209
(3.35)
Constant 2.745 3.283 2.038 2.268 3.544 3.672
(2.39) (4.05) (1.76) (2.04) (4.65) (3.43)
Number of observations 103 103 105 105 107 107
p-value for heteroscedasticitya 0.73 0.35 0.69 0.71 0.21 0.93
Adjusted R2 0.61 0.53 0.61 0.62 0.57 0.65
Note: t-statistics are in parenthesis. In the regressions where the heteroscedasticity test is failed to pass at 15 % level t-statistics based
on heteroscedastic-consistent (White-robust) standard errors are reported.
†
a
3.4 Subjective Measures
In the fourth and last step, we consider some subjective measures for open-
ness. These measures are in some sense similar to the deviation measures
such that both try to capture all aspects of trade policy. The main difference
is that they are partly or completely based on the subjective judgment.
We start with the real exchange rate distortion index suggested by Dollar
(1992) as an openness variable. Our data on this variable come from Global
Development Network (2005) and cover the 1970-2000 period. This measure
compares the domestic prices of tradable goods across countries. Assuming
that the law of one price always holds, a higher level of distortion index
indicates a more distorted trade regime. Our estimation results reported
in column 1 of Table 12 show that the distortion index is significantly and
negatively associated with growth. However, the coefficient estimate of the
index is very small (-0.003!) which implies that the distortion index does not
have any influence on growth. In column 2, we include the variability index
which is simply a coefficient of variation of the real exchange rate distortion
index over the 1970-2000 period and conclude that the variability is negative
and significant with a substantially larger coefficient estimate. However, in a
difference to Dollar (1992), we find an insignificant coefficient estimate of the
distortion index when both distortion and variability indexes are included in
the regression together. As shown in column 3, the regression result shows
that the distortion index is not only insignificant but also is very close to
zero (-0.001). It is, therefore, more reasonable to attribute these findings to
the importance of real exchange rate stability rather than the liberal trade
polices for better growth performance.
In column 4, we include the dummy variable for openness suggested by
Sachs and Warner (1995, SW henceforth). The SW dummy variable is a sin-
gle openness measure covering all major kinds of trade restrictions, namely
non-tariff barriers, average tariff rate, black market premium for exchange
rate, a socialist country, state monopoly on major exports. However, differ-
ently from SW, we extend the openness dummy over the period 1960-2000
rather than 1970-1989 period. This means that we consider only 26 coun-
tries as always open during the 1960-2000 period while SW define 33 open
countries between 1970 and 1989. More clearly we define the countries Tai-
wan, Jordan, Ireland, South Korea, Indonesia, Japan and Australia as closed
over the 1960-2000 period since these countries opened their trade regimes
during the 1960s according to the SW criteria. Notice that in our regressions
Germany and Taiwan are always omitted because of missing data on these
countries over the sample period. In spite of this difference, we conclude that
the SW dummy variable is statistically significant and positive. Regression
31
result in column 4 implies that in the long run GDP per worker in an open
economy would have 2.6 times that in a closed economy once the other de-
terminants are controlled.14 15 In addition, in column 5 we employ the SW
dummy over the period 1970-2000 period which is exactly identical to the
original SW dummy variable. Now the coefficient estimate of the dummy
is larger and indicates that GDP per worker in an open economy would be
equal to 3.2 times the GDP per worker in a closed economy in the long run.
The SW dummy variable is, however, heavily criticized by Harrison and
Hanson (1999) and Rodrik and Rodrı́guez (2000). The most important crit-
icism is that the strength of the dummy is mainly the result of the criteria
related to state monopolies on exports and the black market premium. Ac-
cording to Rodrik and Rodrı́guez (2000), the export monopolies component of
the SW dummy acts like a sub-Saharan Africa dummy while the black market
premium component reflects poor macroeconomic conditions and imbalances
rather than restrictive trade polices. That is why, in column 6 we insert a
sub-Saharan Africa dummy to the regression. The regression result indicates
that the coefficient estimate of the SW dummy is now higher and highly sig-
nificant. In column 7, we substitute the sub-Saharan African dummy with a
composite regional dummy for both sub-Saharan Africa and Latin America.
Now, the coefficient estimate of SW dummy is relatively smaller, but still sta-
tistically significant. In column 8, we introduce the black market premium
dummy which takes the value of 1 if the average black market premium ex-
ceeds 20 % in any of the 1960s, 1970s, 1980s or 1990s as well as the composite
regional dummy variable. The regression result shows that the coefficient es-
timate of SW dummy is not only smaller but also marginally significant at
the 7 % level. In column 9, we substitute the black market premium dummy
with logarithm of average black market premium and conclude the same re-
14
The effect of openness dummy on the long run income level can be calculated as
exp(−γ5 /γ1 ) where γ5 is the coefficient of openness dummy and γ1 is the coefficient of the
initial level of income. According to the regression result in Column 4, γ5 = 0.463 and
γ1 = −0.483, hence the long run level of GDP per worker in an open country would be
2.6 = exp(−0.463/ − 0.483) times the GDP per worker in a closed economy.
15
Moreover, we test the absolute convergence hypothesis for open economies over the
1960-2000 period. In order to facilitate comparison with SW, we employ annual growth
rate and initial income according to GDP per capita rather than GDP per worker and
conclude the following cross-country growth regression for 24 countries which are always
open during the 1960-2000 period (robust t-statistics are in parentheses).
[log yi,2000 − log yi,1960 ]/40 = 15.349 − 1.397 log yi,1960 R̄2 = 0.68
(6.14) (5.00)
where yi is the real GDP per capita. As can be seen, the coefficient estimate of initial
income is very close to that estimated by SW (They report the coefficient estimate of
initial GDP per capita as minus 1.368 in column 3 of Table 11 p. 48).
32
sult. Therefore, it is possible to conclude that the SW dummy is sensitive to
the black market premium for exchange rate, but not to dummy variable for
sub-Saharan Africa.
We employ the fraction of open years according to SW liberalisation dates
over the 1960-2000 period in column 10.16 This variable is more reasonable
with respect to the SW dummy since SW liberalisation dates are based on
the intensive survey of the country cases as pointed out by Wacziarg (2001)
and Wacziarg and Welch (2008). The regression result indicates that the
fraction of open years is strongly and significantly correlated with economic
growth over the period 1960-2000.
16
For the period 1990-2000, we employ the liberalisation dates provided by Wacziarg
and Welch (2008) updating the SW dummy and liberalsation status. In their systematic
review, Wacziarg and Welch (2008) disagree with SW on the liberalization status or dates
in the case of several countries. Some countries such as Panama and Cape Verde which
were not included in SW are classified in the study by Wacziarg and Welch (2008). There
are five countries namely, Ivory Coast, the Dominican Republic, Mauritania, Niger and
Trinidad and Tobago for which Wacziarg and Welch (2008) disagree with SW assignment
of liberalization dates and four countries which remains closed as of 2001 according to
Wacziarg and Welch (2008) while SW classifies them as open in the early 1990s. These
countries are Belarus, Croatia, Estonia, and India. In this study, we follow the Wacziarg
and Welch (2008) for the disagreement cases. See Sachs and Warner (1995) and Wacziarg
and Welch (2008) and appendices therein for more information about SW liberalisation
dates.
33
Table 12: Economic Growth and Subjective Measures: OLS Estimates†
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
log GDP per worker 1960 -0.498 -0.520 -0.515 -0.483 -0.478 -0.500 -0.454 -0.469 -0.472 -0.496
(6.91) (7.63) (7.42) (7.10) (7.44) (7.37) (6.72) (6.57) (7.81) (7.43)
log(ni + g + δ) -1.245 -1.042 -1.041 -0.315 -0.310 -0.294 -0.258 -0.483 -0.506 -0.524
(3.29) (2.83) (2.81) (0.79) (0.84) (0.75) (0.66) (1.21) (1.39) (1.44)
log of Investment rate 0.536 0.526 0.524 0.335 0.316 0.313 0.302 0.278 0.261 0.317
(5.49) (5.65) (5.58) (3.71) (3.65) (3.48) (3.38) (2.98) (2.36) (3.57)
log of School enrolment 0.421 0.466 0.460 0.471 0.426 0.383 0.409 0.424 0.421 0.397
(4.36) (5.05) (4.92) (5.65) (5.35) (4.04) (4.76) (4.70) (4.95) (4.84)
Distortion Index by Dollar (1992)a -0.003 - -0.001 - - - - - - -
(2.47) (0.46)
Variability Index by Dollar (1992)b - -0.468 -0.428 - - - - - - -
(3.81) (2.83)
34
Openness Dummy (1960-2000)c - - - 0.463 - 0.510 0.397 0.269 0.244 -
(3.69) (4.03) (3.15) (1.86) (1.89)
Openness Dummy (1970-2000)c - - - - 0.549 - - - - -
(4.88)
Sub-saharan Africa - - - - - -0.242 - - - -
(1.85)
Latin America and Africad - - - - - - -0.234 -0.210 -0.234 -
(2.30) (2.06) (2.44)
BMP dummye - - - - - - - -0.072 - -
(0.73)
log (1+BMP)f - - - - - - - - -0.195 -
(2.42)
Fraction of open yearsg - - - - - - - - - 0.645
(4.31)
Constant 3.240 3.828 3.831 4.839 4.675 4.971 4.765 4.302 4.269 4.097
Continued on Next Page. . .
Table 12 – Continued
(2.94) (3.57) (3.56) (4.18) (4.43) (4.34) (4.21) (3.66) (3.51) (3.90)
Number of observations 87 87 87 102 102 102 102 98 98 102
p-value for heteroscedasticityh 0.68 0.82 0.72 0.77 0.37 0.68 0.55 0.26 0.15 0.91
Adjusted R2 0.67 0.70 0.70 0.65 0.68 0.66 0.67 0.65 0.67 0.67
Note: t-statistics are in parenthesis. In the regressions where the heteroscedasticity test is failed to pass at 15 % level t-statistics based
on heteroscedastic-consistent (White-robust) standard errors are reported.
†
a
Real exchange rate distortion index, 1970-2000 averages.
b
Coefficient variation of real exchange rate distortion index over the 1970-2000 period.
c
Dummy variable for open countries according to the Sachs and Warner (1995) criteria.
d
Composite regional dummy variable for Latin America and Sub-saharan Africa.
e
Dummy variable is equal to 1 if the average black market premium exceeds 20 % in either the1960s or the 1970s or the 1980s or
35
the 1990s.
f
g
Fraction of open years according to librealization dates in Sachs and Warner (1995) and Wacziarg and Welch (2008)
h
Table 13: International Trade and Trade Policy Indexes: OLS Estimates
Dependent variable† Current Openness Real Openness

(1) (2) (3) (4)
log GDP per worker 1960 0.019 0.041 0.091 0.100
(0.47) (1.25) (2.61) (3.29)
log Area -0.124 -0.072 -0.103 -0.055
(2.24) (1.89) (2.28) (1.87)
log Average Labour Force -0.021 -0.055 0.034 0.003
(0.46) (1.46) (0.98) (0.09)
owtia -0.140 -0.063 -0.236 -0.193
(0.72) (0.42) (1.19) (1.20)
owqib -0.237 -0.162 -0.196 -0.115
(2.06) (1.75) (2.01) (1.49)
log (1+BMP)c - 0.016 - -0.038
(0.36) (1.11)
Constant 2.369 2.014 0.403 0.205
(4.10) (4.71) (0.83) (0.52)
p-value for heteroscedasticityd 0.00 0.00 0.00 0.00
F -valuee 2.19 1.11 2.60 1.76
Adjusted R2 0.43 0.43 0.40 0.40

in parenthesis.
†
1960-2000 averages.
a
b
Own-import weighted non-tariff frequency on intermediate inputs and capital goods.
c
d
errors.
e
Test for joint significance of owti, owqi and log(1+BMP)
Finally, we construct a simple subjective openness measure. Recall that

our simple model for international trade expressed in equation (3). As men-
tioned above, one of the weaknesses of this model is the assumption that
the only important omitted variable is trade policy barriers. Now, we relax
this assumption and introduce three trade policy instruments namely tariffs,
non-tariff barriers and the black market premium. Our aim is to obtain ap-
proximate weights for these trade policy instruments in order to construct a
composite trade policy indicator. Estimation results are given in Table 13.
36
Regression results are not very precise compared to our previous estimations
in Table 10. First, introducing trade policy instruments does not improve
the goodness of fit. Second, except for non-tariff barriers, all trade policy
instruments are found to be statistically insignificant. One reason for the less
precise results is that our data on tariffs and non-tariff barriers are not very
satisfactory. Multicollinearity among the policy instruments may be another
reason. However, in spite of the lack of precision, our results indicate that
all trade policy instruments have the expected sign. The only exception is
the regression in column 2 in which the sign of the black market premium
is positive. Hence, except this regression, the coefficient estimates of trade
policy instruments can be used as approximate weights. In the light of the
regressions in Table 13, we define the following three trade policy indicators;
T rade policy 1 = −0.14(owti) − 0.24(owqi)

T rade policy 2 = −0.24(owti) − 0.20(owqi)
T rade policy 3 = −0.19(owti) − 0.12(owqi) − 0.04 log(1 + BM P )
where BMP is the average black market premium and owti and owqi denote
the own-import weighted tariff rates and non-tariff frequency on intermediate
inputs and capital goods, respectively. Notice that the higher level of trade
policy index implies a more open country since weights are negative num-
bers. Thus, one would expect a positive coefficient estimate of the indexes
if openness is positively correlated with economic growth. Employing these
indexes we estimate our baseline model and conclude them all of them have
the positive but insignificant coefficient estimates (Table 14).
4 Sensitivity Analysis
In this section, we investigate the sensitivity of our findings on the relation-
ship between openness and economic growth in the previous section. As
acknowledged by many authors,17 most of the studies in the empirical cross-
country growth literature include a small set of explanatory variables. How-
ever, the main problem in these studies is that their results are very sensitive
to changes in the list of explanatory variables. The empirical literature on
openness and growth is particularly subject to this problem because many
studies in this literature employ simple growth models and it is likely that
17
See, for instance, Levine and Renelt (1992), Mankiw (1995), Sala-i-Martin (1997),
Temple (2000), Brock and Durlauf (2001).
37
Table 14: Economic Growth and Composite Trade Policy Measures: OLS
Estimates†
(1) (2) (3)

log GDP per worker 1960 -0.484 -0.489 -0.489
(5.78) (5.81) (5.97)
log(ni + g + δ) -1.237 -1.221 -1.298
(2.68) (2.65) (2.87)
log of Investment rate 0.431 0.428 0.363
(4.24) (4.22) (3.65)
log of School enrolment 0.432 0.436 0.439
(4.17) (4.20) (4.35)
Trade policy 1 0.574 - -
(0.86)
Trade policy 2 - 0.643 -
(0.99)
Trade policy 3 - - 1.188
(1.40)
Constant 2.682 2.783 2.464
(2.01) (2.06) (1.83)
Number of observations 85 85 83
p-value for heteroscedasticitya 0.73 0.77 0.37
Adjusted R2 0.60 0.60 0.60
†
a
errors.
many of the results of these works arise from model misspecification (Ro-
drik and Rodrı́guez (2000)). Therefore, the estimating framework requires a
reasonably comprehensive set of explanatory variables. For this purpose, we
redefine our baseline growth model expressed in equation (2) as:

(4)
+ γ5 Opennessi + γ6 Zi + εi
38
where Z is a vector of other explanatory variables. We determine Z as
follows: First, we include two variables related to macroeconomic policy,
namely inflation rate and government consumption expenditures. Inclusion
of these variables is particularly important since an important criticism on
the openness-growth literature is that openness measures are proxy for other
macroeconomic policies rather than trade policy per se. Second, we employ
two variables in order to consider the effect of institutions and geography.
For this, following Hall and Jones (1999), we measure institutional quality
by using a composite index based on the data set of International Country
Risk Guide (ICRG) published by a private international consulting company
Political Risk Services. This index consists of equally weighting an average of
four ICRG components for the years 1984-2000: i) investment profile as a av-
erage of three subcomponents namely, contract viability, profits repatriation
and payment delays; ii) law and order; iii) corruption; and iv) bureaucratic
quality. We use the share of population in geographical tropics in order to
capture the effect of geography (Sachs (2001)). Finally, we include ethnolin-
guistic fragmentation index (ELF). This index shows the probability that two
randomly selected persons of a given country do not belong to the same eth-
nolinguistic group and has become a standard variable in the cross-country
growth literature since the important studies by Mauro (1995) and Easterly
and Levine (1997).
Before carrying out our sensitivity analysis, we estimate an augmented
neoclassical growth model with the only Z variables. Table 15 reports the
estimation results. As seen in the first column, all Z variables are found to be
statistically significant with the anticipated signs. Moreover, this regression
has substantially high explanatory power such that it explains 76 percent of
variation in growth of per worker GDP over the period 1960-2000. However,
applying the Hadi method we identify six countries (namely Brazil, Bolivia,
Argentina, Peru, Nicaragua and Congo Democratic Republic) as outliers. In
the second column we present the regression results without these six outly-
ing countries and conclude that, except for the inflation rate, all Z variables
are statistically significant. In column 3, instead of dropping these coun-
tries, we estimate the same model by IRLS. Notice that in this regression
we exclude two risky countries, Congo Democratic Republic and Nicaragua,
because of their high leverage values. We again conclude that all Z vari-
ables are statistically significant with the expected signs, except the inflation
rate. Therefore, in column 4 we omit the inflation rate and estimate the
augmented neoclassical growth model with the remaining control variables.
The regression results show that government consumption, institutions, hav-
ing population living in tropics and ELF are significantly associated with
growth. This regression does not include any outliers and hence can be con-
39
Table 15: Other Determinants of Economic Growth†
(1) (2) (3) (4)

log GDP per worker 1960 -0.617 -0.624 -0.620 -0.650
(8.54) (7.25) (9.01) (8.69)
log(ni + g + δ) -0.208 -0.233 -0.246 -0.281
(0.57) (0.55) (0.71) (0.73)
(3.46) (2.84) (4.11) (2.01)
(3.93) (4.26) (3.59) (4.52)
Inflation Rate -0.099 -0.245 -0.005 -
(3.71) (0.68) (0.09)
Government Consumption/GDP -1.855 -1.761 -1.917 -1.725
(2.17) (1.71) (2.36) (1.81)
Institutional Quality Index 0.219 0.215 0.238 0.255
(3.54) (2.44) (4.03) (3.42)
Population in Tropics -0.244 -0.272 -0.308 -0.279
(2.36) (2.90) (3.11) (3.22)
Ethnolinguistic Fragmentation -0.340 -0.321 -0.253 -0.301
(2.45) (1.97) (1.90) (1.85)
Constant 6.013 5.972 5.852 5.706
(5.54) (4.87) (5.69) (5.03)
p-value for heteroscedasticitya 0.17 0.06 - 0.05
Adjusted R2 0.76 0.71 0.75 0.71
†
a
errors.
sidered as a reasonable model for sensitivity analysis. In this regard, we set
40
up the following cross-country growth regression:

+ γ5 (Government Consumptioni ) (5)
+ γ6 (Institutional Qualityi )
+ γ7 (P opulation in T ropicsi )
+ γ8 ELFi + γ9 Opennessi + εi
We carry out our sensitivity analysis under four categorizations: trade

volumes, direct trade policy measures, deviation measures and subjective
indexes. However, we focus only on openness variables found to be significant
in the previous section.
We start our sensitivity analysis with trade volumes, with Table 16 re-
porting our estimation results. All openness variables are found to be positive
but not statistically significant. Table 17 then presents the results of sen-
sitivity analysis for direct trade policy measures. As seen in column 1-3,
the tariff rate and non-tariff barriers are neither individually nor jointly sig-
nificant, albeit they have anticipated negative signs. In columns 4 and 5,
we include the average black market premium and a dummy variable for the
black market premium, respectively. Both variables enter the regression with
negative but insignificant coefficient estimates. This may be attributed as
another evidence in favour of black market market premium as a proxy for
macroeconomic imbalances rather than trade policy.
In table 18, we estimate deviation measures for openness. In the first and
second columns, we include adjusted trade ratios for current openness and
real openness respectively, and conclude that both variables are insignificant.
In column 3, we check the sensitivity of the Frankel-Romer predicted trade
shares and find that the coefficient estimate on this variable is positive but
statistically insignificant (Note that this regression is run over the 1960-1985
period because the Frankel-Romer predicted trade share is calculated for only
1985). In column 4, we enter the Frankel-Romer predicted trade shares at
PPP in 1995. We again find that this variable is not statistically significant
at conventional levels.
Finally, sensitivity analysis on subjective openness indexes is provided in
Table 19. In column 1, we include the fraction of open years according to SW
liberalization dates and conclude that this variable is significant at the 10
percent significance level. In columns 2-4 we include three composite trade
policy indexes with only the third index being statistically significant at the
41
Table 16: Economic Growth and Trade Volumes: Sensitivity Analysis†
(1) (2) (3) (4) (5)

(8.39) (8.28) (8.35) (8.52) (8.67)
log(ni + g + δ) -0.036 -0.020 -0.028 -0.264 -0.266
(0.09) (0.05) (0.07) (0.67) (0.68)
(1.70) (1.69) (1.70) (1.98) (2.02)
(4.63) (4.70) (4.67) (4.46) (4.47)
Government Consumption /GDP -1.820 -1.976 -1.901 -2.009 -1.860
(1.82) (1.95) (1.89) (1.93) (1.85)
Institutional Quality Index 0.246 0.249 0.247 0.249 0.244
(3.17) (3.22) (3.20) (3.30) (3.10)
Population in Tropics -0.326 -0.327 -0.327 -0.308 -0.290
(3.62) (3.55) (3.60) (3.64) (3.31)
Ethnolinguistic Fragmentation -0.344 -0.331 -0.338 -0.297 -0.297
(1.89) (1.81) (1.85) (1.80) (1.81)
Exports ratio of WB 0.256 - - - -
(1.18)
Imports ratio of WB - 0.295 - - -
(1.40)
Trade ratio of WB - - 0.140 - -
(1.31)
Current Openness of PW - - - 0.100 -
(0.92)
Real Openness by A&C - - - - 0.094
(0.71)
Constant 6.256 6.214 6.237 5.713 5.797
(5.18) (5.15) (5.17) (4.93) (5.06)
Adjusted R2 0.70 0.70 0.70 0.71 0.71

in parenthesis.
†
a
errors.
42
Table 17: Economic Growth and Direct Trade Policy Measures: Sensitivity
Analysis†
(1) (2) (3) (4) (5)

(8.89) (7.68) (8.61) (8.93) (8.85)
log(ni + g + δ) -0.428 -0.561 -0.460 -0.293 -0.230
(1.14) (1.34) (1.14) (0.76) (0.57)
(2.05) (1.80) (1.80) (1.95) (1.96)
(3.04) (3.17) (3.18) (4.73) (4.96)
Government Consumption /GDP -3.174 -2.441 -2.896 -1.559 -1.646
(2.66) (2.27) (2.39) (1.62) (1.66)
Institutional Quality Index 0.226 0.221 0.217 0.229 0.219
(3.03) (2.66) (2.77) (2.73) (2.54)
Population in Tropics -0.324 -0.286 -0.317 -0.287 -0.285
(3.26) (3.03) (3.21) (3.23) (3.22)
Ethnolinguistic Fragmentation -0.194 -0.256 -0.174 -0.271 -0.314
(1.10) (1.45) (0.99) (1.71) (1.92)
owti -0.493 -0.524
(1.95) (2.09)
owqi 0.028 0.122
(0.19) (0.80)
log (1+BMP) -0.057
(0.47)
BMP dummy -0.079
(0.82)
Constant 5.622 4.753 5.422 5.823 6.105
(4.72) (3.71) (4.26) (5.05) (5.00)
F -valueb 0.09
Adjusted R2 0.71 0.69 0.71 0.71 0.71

in parenthesis. The variables owti and owqi refers to own-import weighted tariff rates
and own-import weighted non-tariff frequency on intermediate inputs and capital gods,
respectively. The term log (1+BMP) denotes the logarithm of one plus average value of
black market premium over the 1960-2000. The variable BMP dummy is a dummy variable
which is equal to 1 if the average black market premium exceed 20 % in either the 1960s
or the 1970s or the 1980s or the 1990s.
† 43
a
errors.
b
Test for joint significance of owti and owqi.
Table 18: Economic Growth and Deviation Measures: Sensitivity Analysis†
(1) (2) (3) (4)

(8.77) (8.76) (6.29) (7.59)
log(ni + g + δ) -0.300 -0.289 0.361 -0.220
(0.77) (0.74) (1.35) (0.55)
(1.88) (1.92) (2.82) (2.07)
(4.42) (4.50) (2.96) (4.20)
Government Consumption /GDP -1.803 -1.756 -0.427 -1.936
(1.71) (1.76) (0.56) (1.86)
(3.43) (3.39) (2.99) (3.35)
(3.31) (3.19) (2.76) (3.14)
Ehnolingustic Fragmentation -0.315 -0.307 -0.086 -0.282
(1.80) (1.83) (0.66) (1.70)
RESID Current Openness 0.057 - - -
(0.27)
RESID Real Openness - 0.040 - -
(0.20)
log F&R predicted trade in 1985 - - 0.024 -
(0.62)
log F&R predicted trade in 1995 - - - 0.055
(0.90)
Constant 5.633 5.674 5.416 5.941
(4.96) (4.93) (6.14) (5.06)
p-value for heteroscedasticitya 0.04 0.04 0.18 0.04
Adjusted R2 0.71 0.71 0.59 0.71
†
a
errors.
44
9 percent level.
5 Summary and Concluding Remarks

This paper revisits the empirical evidence on the relationship between trade
openness and economic growth over the sample period of 1960-2000. De-
parting from previous cross-country studies focusing mainly on 1970-1990
period, we reassess the openness-growth link over a much longer time period,
enabling us to better account for both trade policy stance and growth dy-
namics in the long run. In doing so, we experiment with a large number of
openness measures in the framework of the augmented neo-classical growth
model developed by Mankiw et al. (1992). We briefly summarise our main
findings as follows:
We conclude that ratio of trade volume to GDP is positively and signifi-
cantly associated with economic growth. However, their significance in OLS
estimate is mainly driven by a few outlying countries. When we repeat the
analysis using weighted least squares, giving relatively lower weight to out-
lying countries, we find a significant and positive correlation between growth
and trade ratios at nominal prices. We could not conclude the same result
for the real openness which measures the ratio of trade volume to GDP in
constant prices.
Our findings relating to direct trade policy measures indicate only weak
evidence for the adverse growth-tariff connection such that the tariff rate is
only marginally significant within the period 1960-2000. We could not find
any statistically significant relation between non-tariff barriers and economic
growth while the black market premium is negatively and significantly asso-
ciated with growth. However, our empirical investigation on the link between
black market premium and economic growth indicates the negative relation-
ship between growth and macroeconomic imbalances rather than the trade
restrictive effect of black market premium.
We find that adjusted trade ratios and the Frankel-Romer predicted trade
shares are significantly and positively correlated with growth. Our findings,
however, show an insignificant association between growth and outward ori-
entation index by Syrquin and Chenery (1989).
Both the Sachs-Warner openness dummy and the fraction of open years
on the basis of Sachs and Warner (1995) and Wacziarg and Welsh (2008)
liberalization dates are positively and significantly correlated with economic
growth over the period 1960-2000. We could not find any significant rela-
tion between growth and our composite trade policy indexes consisting of
weighted averages of tariff rates, non-tariff barriers and the black market
45
Table 19: Economic Growth and Subjective Measures: Sensitivity Analysis†
(1) (2) (3) (4)

(8.98) (7.71) (7.90) (8.20)
log(ni + g + δ) -0.138 -0.502 -0.458 -0.432
(0.30) (1.21) (1.12) (1.08)
(2.01) (1.91) (1.98) (1.98)
(3.98) (3.17) (3.17) (3.26)
Government Consumption /GDP -1.519 -2.577 -2.707 -2.751
(1.52) (2.38) (2.45) (2.48)
(2.51) (2.69) (2.72) (2.68)
(3.20) (3.13) (3.19) (3.31)
Ethnolinguistic Fragmentation -0.299 -0.246 -0.230 -0.200
(1.85) (1.37) (1.28) (1.11)
Fraction of Open Years 0.278 - - -
(1.66)
Trade policy 1 - 0.257 - -
(0.53)
Trade policy 2 - - 0.498 -
(1.19)
Trade policy 3 - - - 0.967
(1.71)
Constant 6.003 5.026 5.258 5.451
(4.55) (3.91) (4.15) (4.45)
p-value for heteroscedasticitya 0.06 0.00 0.00 0.00
Adjusted R2 0.72 0.69 0.70 0.70

in parenthesis.
†
a
errors.
46
premium.
We also conclude that existing openness variables are very sensitive to the
inclusion of other growth determinants into the baseline growth model. To
show this, we expand our baseline model by adding government consumption,
economic institutions, geography, and ethnolinguistic fragmentation. Once
we consider this model, our findings indicate that openness measures become
insignificant whilst other variables remain significant with the expected signs.
In sum, in contrast to many previous cross-country growth studies, this
paper does not support the proposition that openness has a direct robust rela-
tionship with economic growth in the long run. In light of data evidence here,
one may conclude that trade openness does not matter for economic growth.
However, it may be more reasonable to conclude that without building better
institutions, maintaining conflict management along ethnolinguistic dimen-
sion, and following sound and stable fiscal policies, openness to international
trade will not guarantee economic growth. Economic reforms in these areas
should take priority over the policies enhancing trade openness.
47
References
Aksoy, A. and Salinas, G. (2006). Growth before and after trade liberal-
ization, Policy Research Working Paper 4062, World Bank, Washington,
DC.
Alcalá, F. and Ciccone, A. (2004). Trade and productivity, Quarterly Journal

of Economics 119(2): 613–646.
Barro, R. J. and Lee, J.-W. (1994). Data set for a panel of 138 countries,
Technical report. Downloadable at http://www.nber.org/pub/barro.lee/.
Bils, M. and Klenow, P. J. (2000). Does schooling cause growth?, American

Economic Review 90(5): 1160–1183.
Brock, W. A. and Durlauf, S. N. (2001). Growth empirics and reality, World

Bank Economic Review 15(2): 229–272.
Center for International Development at Harvard Univer-

sity (n.d.). Research Datasets, Boston, MA. Web page:
http://www.cid.harvard.edu/ciddata/ciddata.html.
Chang, R., Kaltani, L. and Loayza, N. V. (2009). Openness can be good

for growth: The role of policy complementarities, Journal of Development
Economics 90(1): 33–49.
Dollar, D. (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.
Dollar, D. and Kraay, A. (2003). Institutions, trade, and growth, Journal of

Monetary Economics 50(1): 133–162.
Durlauf, S. N., Johnson, P. A. and Temple, J. R. (2005). Growth economet-

rics, in P. Aghion and S. N. Durlauf (eds), Handbook of Economic Growth,
Vol. 1A, North-Holland, Amsterdam, chapter 8, pp. 555–677.
Easterly, W. and Levine, R. (1997). Africa’s growth tragedy: Policies and

ethnic divisions, Quarterly Journal of Economics 112(4): 1203–1250.
Edwards, S. (1992). Trade orientation, distortions and growth in developing

countries, Journal of Development Economics 39(1): 31–57.
Edwards, S. (1993). Openness, trade liberalization, and growth in developing

countries, Journal of Economic Literature 31(3): 1358–1393.
48
Edwards, S. (1998). Openness, productivity and growth: What do we really
know?, Economic Journal 108(447): 383–398.
Foster, N. (2008). The impact of trade liberalisation on economic growth:

Evidence from a quantile regression analysis, Kyklos 61(4): 543–567.
Frankel, J. A. and Romer, D. (1999). Does trade cause growth?, American

Global Development Network (2005). Growth Database, New York University

Development Research Institute, New York.
Greenaway, D., Morgan, W. and Wright, P. (2002). Trade liberalisation

and growth in developing countries, Journal of Development Economics
67(1): 229–244.
Grossman, G. M. and Helpman, E. (1991). Innovation and Growth in the

Global Economy, MIT press, Cambridge, USA and London, UK.
Hadi, A. S. (1992). A new measure of overall potential influence in linear

regression, Computational Statistics and Data Analysis 14: 1–27.
Hall, R. E. and Jones, C. I. (1999). Why do some countries produce so much

more output per worker than others?, Quarterly Journal of Economics
114(1): 83–116.
Hamilton, L. C. (1992). Regressions with Graphics: A Second Course in

Applied Statistics, Duxbury Press, Belmont, CA.
Harrison, A. (1996). Openness and growth: A time-series, cross-country

analysis for developimg countries, Journal of Development Economics
48(2): 419–447.
Harrison, A. and Hanson, G. (1999). Who gains from trade reform? some
remaining puzzles, Journal of Development Economics 59(1): 125–154.
Heston, A., Summers, R. and Aten, B. (2002). Penn World Table Version
6.1, Center for International Comparisons at the University of Pennsylva-
nia (CICUP). Available at http://pwt.econ.upenn.edu/.
Hoaglin, D. C., Iglewicz, B. and Tukey, J. W. (1986). Performance of some

resistant rules for outlier labeling, Journal of the American Statistical As-
sociation 81(396): 991–999.
Huber, P. J. (1981). Robust Statistics, Wiley, New York.
49
Kim, D.-H. (2011). Trade, growth and income, Journal of International
Trade & Economic Development 20(5): 677–709.
Kneller, R., Morgan, C. W. and Kanchanahatakij, S. (2008). Trade liberali-

sation and economic growth, World Economy 31(6): 701–719.
Leamer, E. E. (1988). Measures of opennes, in R. E. Baldwin (ed.), Trade

Policy Issues and Emprical Analysis, University of Chicago, Chicago and
London, pp. 147–204.
Lee, H. Y., Ricci, L. A. and Rigobon, R. (2004). Once again, is openness

good for growth?, Journal of Development Economics 75(2).
Lee, J.-W. (1993). International trade, distortions, and long-run economic

growth, IMF Staff Papers 40(2): 299–328.
Levine, R. and Renelt, D. (1992). A sensitivity analysis of cross-country

growth regressions, American Economic Review 82(4): 942–963.
Mankiw, G. N. (1995). The growth of nations, Brookings Papers on Economic

Activity 1995(1): 275–310.
Mankiw, G. N., Romer, D. and Weil, D. N. (1992). A contribution to the

empirics of economic growth, Quarterly Journal of Economics 107(2): 407–
437.
Mauro, P. (1995). Corruption and growth, Quarterly Journal of Economics

110(3): 681–712.
Political Risk Services (n.d.). International Country Risk Guide (ICRG),

Political Risk Services Group Data Base: 1984-2002. Web page:
http://www.prsgroup.com/ICRG.aspx.
Pritchett, L. (1996). Measuring outward orientation in LDCs: Can it be

done?, Journal of Development Economics 49(2): 307–335.
Rivera-Batiz, L. A. and Romer, P. M. (1991). International trade with en-

dogenous technological change, European Economic Review 35(4): 971–
1001.
Rodrik, D. and Rodrı́guez, F. (2000). Trade policy and economic growth: A

skeptic’s guide to the cross-national evidence, in B. Bernanke and K. Rogoff
(eds), NBER Macroeconomics Annual, Vol. 15, MIT Press, Cambridge,
USA, pp. 261–325.
50
Sachs, J. D. (2001). Tropical underdevelopment, Working Paper 8119,
National Bureau of Economic Research, Cambridge, MA. Available at
http://www.nber.org/papers/w8119.
Sachs, J. D. and Warner, A. M. (1995). Economic reform and the process of

global integration, Brookings Papers on Economic Activity 1995(1): 1–118.
Sala-i-Martin, X. X. (1997). I just run two million regressions, American

Syrquin, M. and Chenery, H. (1989). Three decades of industrialization,

World Bank Economic Review 3(2): 145–181.
Temple, J. (2000). Growth regressons and what the textbooks don’t tell you,
Bulletin of Economic Research 52(3): 181–205.
Vamvakidis, A. (1999). Regional trade agreements or broad liberalization:

Which path leads to faster growth?, IMF Staff Papers 46(1): 42–68.
Vamwakidis, A. (2002). How robust is the evidence growth-openness connec-

tion? historical evidence, Journal of Economic Growth 7: 57–80.
Wacziarg, R. (2001). Measuring the dynamic gains from trade, World Bank
Wacziarg, R. (2002). Review of Easterly’s “The Elusive Quest for Growth”,

Journal of Economic Literature 40(3).
Wacziarg, R. and Welch, K. H. (2008). Trade liberalization and growth: New

evidence, World Bank Economic Review 22(2): 187–231.
Warner, A. M. (2003). Once more into the breach: Economic growth and
integration, Working Papers 34, Center for Global Development. available
at http://ideas.repec.org/p/cgd/wpaper/34.html.
Wooldridge, J. M. (2003). Introductory Econometrics: A Modern Approach,

South-Western, Mason, Ohio.
World Bank (2002, 2006). World Development Indicators (CD-ROM data),

Washington, DC.
Yanikkaya, H. (2003). Trade openness and economic growth: A cross-country

empirical investigation, Journal of Development Economics 72(1): 57–89.
51
6 Appendix: Descriptions and Sources of Vari-
ables used in Cross-Country Growth Re-
gression Analysis
6.A Augmented Neo-classical Growth Model
Real GDP per capita (RGDPCH) : 1996 international prices, chain se-
ries. Source: Global Development Network Growth Database (2005)
which rely on Penn World Tables Version 6.1 (Heston, Summers and
Aten (2002)).
Population (TP) : Total population is based on the de facto definition

of population, which counts all residents regardless of legal status or
citizenship. Source: The World Bank World Development Indicators
(2002, 2006).
Labour force (LF) : Labour force or economically active population de-

fined as the total population between ages 15 and 64. Source: The
World Bank World Development Indicators (2002, 2006)
Share of labour force (SLF) : Share of labour force in total population.

The exact calculation is LF/T P .
Real GDP per worker (PWGDP) : 1996 international prices, chain se-
ries. The exact calculation is P W GDP = RGDP CH ∗ (1/SLF ).
Growth : Average growth rate of real GDP per worker over the 1960-2000
period. The exact calculation is log(P W GDP 2000/P W GDP 1960),
where PWGDP1960 and PWGDP2000 is the real GDP per worker in
1960 and 2000, respectively.
Initial income (PWGDP1960) : Real GDP per worker in 1960.
Population growth (n) : Average rate of population growth between 1960

and 2000. The exact calculation is (1/40) ∗ log(T P 2000/T P 1960),
where TP1960 and TP2000 are total population in 1960 and 2000,
respectively.
(g+δ) : Sum of exogenous rates of technological process and depreciation

over the 1960-2000 period and assumed to be equal to 0.05.
(n+g+δ) : Sum of rates of population growth, technical process and depre-

ciation over the 1960-2000 period.
52
Investment rate (INV) : Average of Investment share in GDP at con-
stant prices over the 1960-2000 period. Source: Penn World Tables
Version 6.1 (Heston, Summers and Aten (2002)) and the World Bank
World Development Indicators (2002, 2006).
School enrolment rate (SCH) : Average gross rate of secondary school

enrolment over the 1960-2000 period. Source: The World Bank World
Development Indicators (2002, 2006).
6.B Trade Policy

Imports share from the World Bank (MGDP WB) : Average share
of imports of goods and services in GDP over the 1960-2000 period.
Source: The World Bank World Development Indicators (2002, 2006).
Exports share from the World Bank (XGDP WB) : Average share
of exports of goods and services in GDP over the 1960-2000 period.
Trade ratio from World Bank (XMGDP WB) : Average share of ex-
ports plus imports of goods and services in GDP over the 1960-2000 pe-
riod. The exact calculation is XM GDP W B = XGDP W B+M GDP W B.
Current Openness of Penn World (COPEN) : Average share of ex-

ports plus imports of goods and services in GDP in current prices over
the 1960-2000 period. Source: Penn World Tables Version 6.1 (Hes-
ton, Summers and Aten (2002)).
Real Openness of Penn World (ROPEN) : Average share of imports

plus exports in US dollar in GDP in PPP US dollar over the 1960-2000
period. The Exact calculation is ROP EN = (P GDP/100) ∗ COP EN ,
where P GDP is the Penn World variable of price level of GDP, unit:
US dollar=100 in current prices, and COP EN is the current open-
ness variable of Penn World. Source: Penn World Tables Version 6.1
(Heston, Summers and Aten (2002)).
Trade ratio with OECD (XM OECD) : Trade with OECD members
over the 1960-1998 period (Exports plus Imports as a ratio to GDP).
Source: Global Development Network Growth Database (2005).
Trade ratio with NonOECD (XM NonOECD) : Trade with Non-OECD

countries over the 1960-1998 period (Exports plus Imports as a ratio
53
to GDP). Source: Global Development Network Growth Database
(2005).
Tariff rate (OWTI) : Own-import weighted tariff rates on intermediate

inputs and capital goods over the 1983-1985 period. Source: Barro
and Lee (1994),
Non-tariff Barriers (OWQI) : Own-import weighted non-tariff frequency

on intermediate inputs and capital goods over the 1983-1985 period.
Source: Barro and Lee (1994).
Import Duties (M DUTY) : Collected import duties as a fraction of im-

ports over the 1970-1998 period. Source: World Bank World Devel-
opment Indicators (2002).
Tariff rate (UWATR) : Unweighted average tariff rates on all commodi-

ties over the period 1990-1999 period. Source: Wacziarg and Welch
(2008)
Black Market Premium (BMP) Average of black market premium over

the period 1960-1999. The black market premium is calculated as
(P arallel Exchange rate/Of f icial Exchange rate) − 1. Source:
Global Development Network Growth Database (2005).
log (1+BMP) : Logarithm of one plus average of black market premium

over the period 1960-1999.
Black Market Premium Dummy 1 (BMP Dummy1) : The dummy

variable is equal to 1 if the average black market premium exceeds 20
percent in the 1960s or the 1970s or the 1980s or the 1990s.
Black Market Premium Dummy 2 (BMP Dummy2) : The dummy

variable is equal to 1 if the average black market premium exceeds 20
percent over the 1960-2000 period.
Outward Orientation Index by Syrquin and Chenery (SCOUT) : Dummy

variable for outward orientation based on Syrquin and Chenery (1989).
Source: Levine and Renelt (1992)
log F&R predicted trade 1985 (F&R85) : Logarithm of the Frankel-

Romer predicted trade shares from gravity model in 1985. Source:
Frankel and Romer (1999).
54
log F&R predicted trade 1995 (F&R95) : Logarithm of the Frankel-
Romer predicted trade shares from gravity model as a fraction of GDP
in PPP in 1995. Source: Dollar and Kraay (2003).
Exchange rate distortion index (RERD) : The real exchange rate dis-
tortion index over the period 1970-2000. Source: Dollar (1992) and
Global Development Network Growth Database (2005).
Exchange rate variability index (RERV) : Coefficient of variation of

the real exchange rate distortion index over the 1970-2000 period.
Sachs-Warner openness dummy (SW-Dummy) : The Sachs-Warner

openness dummy over the period 1960-2000. Source: Sachs and Warner
(1995) and Wacziarg and Welch (2008).
Sachs-Warner Open Years( SW-Years) : Fraction of open years on the

basis of Sachs-Warner and Wacziarg-Welch liberalisation dates over the
1960-2000 period. Source: Sachs and Warner (1995) and Wacziarg and
Welch (2008).
6.C Other Growth Determinants

Inflation Rate (INFLATION) : Average inflation rate based on con-
sumer price index over the 1960-2000 period. Source: The World
Bank World Development Indicators (2002, 2006) based on Interna-
tional Monetary Fund, International Financial Statistics and data files.
Government Consumption (GOVCONS) : The ratio of general gov-

ernment final consumption expenditure in GDP. Source: The World
Bank World Development Indicators (2002, 2006) based on World Bank
national accounts data, and OECD National Accounts data files.
Economic Institutions (ICRG Measure) A measure of economic in-

stitutions based on four International Country Risk Guide (ICRG)
components of Political Risk Services Group: 1) Investment Profile
as a average of three subcomponents namely, contract viability, prof-
its repatriation and payment delays; 2) law and order; 3) corruption;
4) bureaucratic quality. The index is calculated as the equal weight-
ing average of these four categories for the years 1984-2000 (annual
observations are calculated as the averages of the monthly indexes).
The higher points indicates the greater institutional quality. Source:
Political Risk Services (n.d.)
55
Ethno-linguistic fragmentation index (ELF) : The index shows the
probability that two randomly selected persons do not belong to the
same ethnolinguistic group in a given country and ranges between 0
and 1. The lower value of the index implies the more homogenous
population. Source: Easterly and Levine (1997)
Population in Tropics (TROPICPOP) : The fraction of population liv-

ing within the geographical tropics. Source: Center for International
Development at Harvard University (n.d.)
56
Table 20: Summary Statistics of Openness Measures
] of Obs. Mean Std. Dev. Min. Max.

Trade Volumes
MGDP WB 107 0.3377 0.1887 0.0723 1.0051
XGDP WB 107 0.2958 0.1848 0.0656 1.0270
XMGDP WB 107 0.6335 0.3583 0.1453 2.0320
COPEN 114 0.6432 0.4165 0.1477 3.2260
ROPEN 114 0.3734 0.3526 0.0436 2.4601
XM OECD 106 0.3912 0.2627 0.0655 1.7622
XM NonOECD 106 0.1409 0.1636 0.0156 1.2300
Direct Trade
Policy Measures
OWTI 104 0.1688 0.1630 0 1.319
OWQI 102 0.1858 0.2372 0 0.888
M DUTY 117 0.1229 0.0888 0 0.4645416
UWATR 121 0.1496 0.0932 0.003 0.547
log (1+BMP) 121 0.3776 0.6716 -0.004 5.453
BMP Dummy1 121 0.5455 0.5000 0 1
BMP Dummy2 121 0.4463 0.4992 0 1
Deviation Measures
SCOUT 119 0.3445 0.4772 0 1
F&R85 148 2.9882 0.7905 0.833 5.639
F&R95 195 -2.8315 0.6409 -4.4070 -0.6378
RESID COPEN 111 0.0000 0.3071 -0.5700 1.9100
RESID ROPEN 111 0.0000 0.2706 -0.5495 1.5548
Subjective Measures
RERD 94 114.6070 39.2534 59.2693 358.0221
RERV 94 0.2917 0.3220 0 2.7937
SW-Dummy 114 0.2281 0.4214 0 1
SW-Years 114 0.4322 0.3832 0 1
POLICY1 102 -0.0684 0.0695 -0.3978 -0.0002
POLICY2 102 -0.0780 0.0729 -0.4942 -0.0002
POLICY3 92 -0.0724 0.0617 -0.3666 0.0001
57
Please note:
You are most sincerely encouraged to participate in the open assessment of this
discussion paper. You can do so by either recommending the paper or by posting your
comments.
Please go to:
http://www.economics-ejournal.org/economics/discussionpapers/2012-25
The Editor
© Author(s) 2012. Licensed under a Creative Commons License - Attribution-NonCommercial 2.0 Germany

SSRN Id2087930

Uploaded by

Copyright:

Available Formats

SSRN Id2087930

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

SSRN Id2087930

Uploaded by

Copyright:

Available Formats

Discussion Paper

No. 2012-25 | May 22, 2012 | http://www.economics-ejournal.org/economics/discussionpapers/2012-25

Openness to International Trade and Economic

JEL F13, F43, O47.

Correspondence Bülent Ulaşan, Research and Monetary Policy Department, Central

2 Economic Growth and Openness to Inter-

log yi (t) − log yi (0) = γ0 + γ1 log yi (0) + γ2 log(ni + g + δ)

log yi,2000 − log yi,1960 = γ0 + γ1 log yi,1960 + γ2 log(ni + g + δ)

Once we specify the cross-country growth regression expressed in equa-

3.1 Trade Volumes

Average GDP per Exports Imports Trade Current Real Trade

(1) (2) (3) (4) (5) (6) (7)

(a) Current Openness: Exports plus Imports as a ratio of GDP in

(b) Real Openness: Exports plus Imports as a ratio of GDP in PPP

Figure 1: Partial Associations between Growth Rate and Trade Shares

(1) (2) (3) (4) (5) (6) (7)

(1) (2) (3) (4) (5) (6) (7)

Note: t-statistics are in parentheses.

3.2 Direct Trade Policy Measures

nificant for the black market premium.

3.2.1 Black Market Premium: A Proxy for Trade Policy or Macroe-

Table 7: Black Market Premium (%): Summary Statistics

Number of Mean Standard Minimum Maximum

Table 7 provides summary statistics of black market premium for each

(1) (2) (3) (4) (5)

(1) (2) (3) (4) (5)

imbalances and growth. In Table 9 we exactly replicate the cross country

3.3 Deviation Measures

((X + M )/GDPi ) = α0 + α1 log(Initial per worker GDPi )

Dependent variable† Current Openness Real Openness

Note: t-statistics based on heteroscedastic-consistent (White-robust) standard errors are

landlocked countries in order to control for natural trade barriers.13 However,

(1) (2) (3) (4) (5) (6)

Dependent variable† Current Openness Real Openness

Note: t-statistics based on heteroscedastic-consistent (White-robust) standard errors are

Finally, we construct a simple subjective openness measure. Recall that

T rade policy 1 = −0.14(owti) − 0.24(owqi)

(1) (2) (3)

log yi,2000 − log yi,1960 = γ0 + γ1 log yi,1960 + γ2 log(ni + g + δ)

(1) (2) (3) (4)

sidered as a reasonable model for sensitivity analysis. In this regard, we set

log yi,2000 − log yi,1960 = γ0 + γ1 log yi,1960 + γ2 log(ni + g + δ)

We carry out our sensitivity analysis under four categorizations: trade

(1) (2) (3) (4) (5)

Note: t-statistics based on heteroscedastic-consistent (White-robust) standard errors are

(1) (2) (3) (4) (5)

Note: t-statistics based on heteroscedastic-consistent (White-robust) standard errors are

(1) (2) (3) (4)

5 Summary and Concluding Remarks

(1) (2) (3) (4)

Note: t-statistics based on heteroscedastic-consistent (White-robust) standard errors are

Alcalá, F. and Ciccone, A. (2004). Trade and productivity, Quarterly Journal

Bils, M. and Klenow, P. J. (2000). Does schooling cause growth?, American

Brock, W. A. and Durlauf, S. N. (2001). Growth empirics and reality, World

Center for International Development at Harvard Univer-

Chang, R., Kaltani, L. and Loayza, N. V. (2009). Openness can be good

Dollar, D. (1992). Outward-oriented developing economies really do grow

Dollar, D. and Kraay, A. (2003). Institutions, trade, and growth, Journal of

Durlauf, S. N., Johnson, P. A. and Temple, J. R. (2005). Growth economet-