The Economic Journal, 108 (March), 383±398. # Royal Economic Society 1998.

Published by Blackwell
Publishers, 108 Cowley Road, Oxford OX4 1JF, UK and 350 Main Street, Malden, MA 02148, USA.

OPENNESS, PRODUCTIVITY AND GROWTH: WHAT DO

WE REALLY KNOW?

Sebastian Edwards
Comparative data for 93 countries are used to analyse the robustness of the relationship
between openness and total factor productivity growth. Nine indexes of trade policy are used
to investigate whether the evidence supports the view that total factor productivity growth is
faster in more open economies. The results are robust to the use of openness indicator,
estimation technique, time period and functional form, and suggest that more open countries
experienced faster productivity growth. Although the use of instrumental variables help
dealing with endogeneity, issues related to causality remain somewhat open, and require time
series analyses to be adequately addressed.

Old controversies die slowly. For over a century social analysts have debated
the connection between trade policy and economic performance. While
according to liberal economists freer trade results in faster growth, some
analysts have argued that protectionism may help economic performance. This
controversy continues today, even as the world is experiencing an unprece-
dented period of trade liberalisation, and in spite of numerous empirical
studies that claim to have found a positive effect of openness on growth. The
most prominent trade liberalisation sceptics include Krugman (1994) and
Rodrik (1995), who have argued that the effect of openness on growth is, at
best, very tenuous, and at worst, doubtful. Two issues have been at the core of
these controversies: ®rst, until recently theoretical models had been unable to
link trade policy to faster equilibrium growth. And second, the empirical
literature on the subject has been affected by serious data problems.1
During the last decade, however, the `new' theories of growth pioneered by
Romer (1986) and Lucas (1988) have provided persuasive intellectual support
for the proposition that openness affects growth positively. Romer (1992),
Grossman and Helpman (1991) and Barro and Sala-i-Martin (1995), among
others, have argued that countries that are more open to the rest of the world
have a greater ability to absorb technological advances generated in leading
nations. Barro and Sala-i-Martin (1995, Ch. 8), for example, consider a two-
countries world (one advanced and one developing), differentiated inputs,
and no capital mobility. Innovation takes place in the advanced (or leading)
nation, while the poorer (or follower) country con®nes itself to imitating the
new techniques. The equilibrium rate of growth in the poorer country
depends on the cost of imitation, and on its initial stock of knowledge. If the
costs of imitation are lower than the cost of innovation, the poorer country will

 I thank Andrew Warner for providing me time series of the Sachs-Warner openness indicator. I am
indebted to Alejandro Jara for excellent assistance, and to seminar participants at a seminar at the
Banco Central de Chile for helpful comments. I am grateful to Jose De Gregorio and two referees for
helpful suggestions.
1
See the surveys by Edwards (1993) and Rodrik (1995). In this paper the emphasis is on policy-
induced openness.

[ 383 ]
384 THE ECONOMIC JOURNAL [MARCH
grow faster than the advanced one, and there will be a tendency towards
convergence. In this type of model it is natural to link the cost of imitation to
the degree of openness: more open countries have a greater ability to capture
new ideas being developed in the rest of the world (Obstfeld and Rogoff,
1996).2
In spite of these theoretical advances, the empirical literature has continued
to be affected by some serious limitations. In particular, researchers have been
unable to generate satisfactory indexes of trade policy orientation. According
to Rodrik (1995), in most studies on openness and growth `the trade-regime
indicator used is typically measured very badly', and `openness in the sense of
lack of trade restrictions is often confused with macroeconomic aspects of the
policy regime (p. 2941).' The lack of high quality comparative data on total
factor productivity (TFP) has complicated things further, imparing the analysis
of the connection between openness productivity growth.
The complex nature of commercial policy ± international trade can be
affected by tariffs, quotas, licenses, prohibition, and exchange controls, among
others ± suggests that attempts to construct a single indicator of trade
orientation may be futile, and will tend to generate disagreements and
controversies. This means that in order for research on the relationship
between trade policy and growth to be persuasive, its results have to be robust
to the way in which (policy-induced) openness is measured.
In this paper I use a new comparitive data set for 93 countries to analyse the
robustness of the relationship between openness and TFP growth. I use nine
alternative indexes of trade policy to investigate whether the evidence supports
the view that, with other things given, TFP growth is faster in more open
economies. The paper is organised as follows: In Section 1 I review alternative
methodoligical approaches used to measure openness. In Section 2 I report
new results on the relationship between TFP growth and openness obtained
with the new data set. In Section 3 I analyse the robustness of these results.
Section 4 contains the conclusions.

1. On the Measurement of Trade Orientation and Openness

For a long time economists have tried to provide comparative measures of
openness. This has proven to be controversial and elusive. This is illustrated by
the fact that while for some South Korea has been an open and outward-
orientated economy (Greenaway and Nam, 1988), for others it is a prime
example of a semi-closed economy with a high degree of government interven-
tion (Wade, 1994).
Many early cross country comparative studies used ready-available data on
the external sector as proxies for openness. These have included trade
dependency ratios and the rate of growth exports (Balassa, 1982). The main
limitation of these indicators is that they are not necessarily related to policy ±

2
Grossman and Helpman (1991), however, present a model where openness affects growth
negatively.

# Royal Economic Society 1998

1998] OPENNESS, PRODUCTIVITY AND GROWTH 385
a country can distort trade heavily, and still have a high trade dependency ratio
± and that they are largely endogenous.
Some authors have tried to avoid these problems by using (partial) informa-
tion on policy to classify countries according to the degree to which trade is
distorted. The Heritage Foundation index of trade policy ± which classi®es
countries into ®ve categories according to the level of tariffs and other
(perceived) distortions ± is a good recent example of this approach (Johnson
and Sheehy, 1996).3 Michaely et al. (1991) constructed a `subjective index' of
trade liberalisation for their cross country regressions. The World Bank's 1987
World Development Report constructed an `outward orientation' index for 41
countries at two points in time (see also Greenaway and Nam, 1988). The
classi®cation of Korea as a `strongly outward oriented' country both in 1963±
73 and 1973±85 periods elicited, however, some controversy, since it has been
argued that during the earlier years the Korean trade regime was considerably
more restrictive than in the latter period (Edwards, 1993). Sachs and Warner
(1995) have used a series of trade-related indicators ± tariffs, quotas coverage,
black market premia, social organisation and the existence of export market-
ing boards ± to construct a composite openness index. Although this indicator
is an improvement over previous attempts, it provides only a binary classi®ca-
tion ± a country is either open or closed. As a result, countries with different
degrees of trade intervention are equally classi®ed as open. Also, many of the
underlying data used by Sachs and Warner to construct their index are only
available at one point in time.
Some analysts have tried to use observed values of variables associated with
trade restrictiveness as indicators of openness. The most popular among these
are tariff averages, average coverage of quantitative restrictions (QRs), and
collected tariff ratios (CTR), de®ned as the ratio of tariff revenues to imports.
In addition, some authors have argued that the black market premium for
foreign exchange is a good proxy for the overall degree of external sector
distortions (Levine and Renelt, 1992). An advantage of these indicators is that
they are drawn from observed data, and that they allow for intermediate
situations where countries are neither totally open nor totally closed. Pritchett
(1991) analysed the relationship between some of these indicators and
concluded that they are correlated among themselves somewhat weakly.
Pritchett and Sethi (1994) found that `collected tariff rates (CTR)' under-
estimated true protection. Using samples of over 3,000 observations for
Jamaica, Kenya and Pakistan, they found out that the relationship between
of®cial tariff rates and CTRs is not linear. However, they found out that in all
cases but one the ordering of of®cial tariff rates and CTRS was maintained.
Anderson and Neary (1994) argue that if tariffs are the only form of
protection, a unique openness index can be de®ned as the welfare-equivalent
`average tariff'. This indicator is based on the concept of `average tariff'
developed by Corden (1966). The change in this index resulting from a trade
reform is equal to the weighted average of change in domestic prices.

3
Krueger (1978) is an early attempt to use dummy variables to classify trade regimes.

# Royal Economic Society 1998

386 THE ECONOMIC JOURNAL [MARCH
Anderson (1994) calculated the Anderson-Neary indicator for a group of 23
countries and found that the weighted average tariff tended to underestimate
the `true' degree of trade restrictions. The extent of underestimation is directly
proportional to the degree of dispersion in the protective structure. Although
the development of this new index is an important step in the evaluation of
trade policy, the data requirements for its calculation are quite formidable.4
Serious measurement dif®culty arise when there are both tariffs and QRs.
Some authors have faced this problem by using regression-based analyses to
construct openness indexes. Leamer (1988), for example, used an empirical
Hecksher-Ohlin model with nine factors to estimate net trade ¯ows and trade
intensity ratios for 183 commodities at the three digit SITC level for 53
countries. He then took the differences between predicted and actual trade
intensity ratios as indicators of trade barriers. Wolf (1993) extended Leamer's
approach, using a larger set of factors of production, a more disaggregated set
of commodities and three alternative base years (1975, 1980 and 1985). He
constructed trade orientation indexes as `the distance between actual trade
and the trade predicted by the ``true'' model in the absence of distortions'
(page 60). As Leamer (1988) himself has recognised, however, the usefulness
of these type of indexes is not all that clear.
To sum up, the discussion in this section has chronicled the quest for
®nding an appropriate comparative openness index. In spite of signi®cant
efforts and ingenuity, there has not been too much progress in this area. The
vast majority of indexes continue to be subject to limitations. Most empirical
studies on the relationship between trade and economic performance have
relied on one ± or at most two ± of these indexes and have, thus, left
themselves open to criticism by reform-sceptics. In my view, the dif®culties in
de®ning satisfactory summary indexes suggest that researchers should move
away from this area, and should instead concentrate on determining whether
econometric results are robust to alternative indexes. The empirical analysis
presented in the rest of this paper is an attempt to move in this direction.

2. Openness and Total Factor Productivity Growth: New Results

In this section I report results on the relationship between trade policy and
productivity growth for a cross-section of countries. I begin by sketching the
mechanics of TFP growth in an open economy. I then turn to a description of
the data, and I ®nally present the regression results.

2.1. The Simple Mechanics of TFP Growth and Openness

Consider a standard aggregate production function where GDP depends on
physical capital (K ), labour measured in ef®ciency units (L) and the stock of
knowledge, or total factor productivity (B): Y t ˆ B t f (K t , L t ). Growth at any

4
The basic data required to calculate these equivalent tariff measures for 30 countries have recently
made available in electronic form. See Feenstra et al. (1997).

# Royal Economic Society 1998

1998] OPENNESS, PRODUCTIVITY AND GROWTH 387
moment in time will depend on the rate of change of these three factors.5
Assume that there are two sources of TFP growth: a domestic source ±
associated with innovation ± and an international one, related to the rate at
which the country is able to absorb (or imitate) technological progress
originated in the leading nations. The rate of domestic innovation is assumed
to depend on the level of human capital (education); imitation, on the other
hand, depends on a `catch-up' term. That is, countries with a lower initial stock
of knowledge will tend to imitate faster than those with a higher initial stock of
TFP. The rate of growth of B can be written as:
_
B=B ˆ ä ‡ è(W ÿ B)=B, (1)
where W is the world stock of knowledge assumed to grow at a rate of g ; ä is
the domestic rate of innovation assumed to depend on human capital, and not
to exceed g (i.e. g > ä). If the country in question is `the' leading innovating
nation, ä ˆ g , and W ˆ B. Finally, è is the speed at which the country closes
the `knowledge gap', and is assumed to depend on national policies, including
trade policy.6 In particular, in the spirit of many new models of growth, it is
assumed that more open countries have a greater ability to absorb ideas from
the rest of the world and, thus, have a higher è. The country's stock of
knowledge will converge to [è=(è ‡ g ÿ ä)]W , and there will be an equili-
brium gap between the country's and the world's level of TFP. This model
implies that countries with a more open economy will have a higher steady
state stock of knowledge and, with other things given, higher GDP. In steady
state equilibrium all countries' TFP will grow at g, as long as è . ä. An
important property of this simple model is that countries that liberalise trade
will experience transitional productivity growth that exceeeds that of countries
that maintain their trade distortions. To sum up, the model sketched here
suggests that TFP growth will be positively affected by the level of human
capital and openness, and negatively affected by the initial stock of know-
ledge.7

2.2. Data
Total factor productivity growth
Ten-years averages of TFP growth were constructed for 93 advanced and
developing countries ± see Appendix for the list. The starting point was the
estimation of a random effect growth equation using panel data for 1960±90.
The raw data on total physical capital, human capital and real GDP were taken

5
In equilibrium, of course, there will only be per capita growth if there are incentives to accumulate
more physical capital (K ), or knowledge (B). A suf®cient condition to generate steady state growth is
that the equilibrium marginal productivity of capital exceeds the rate of time preference.
6
This framework was ®rst proposed by Nelson and Phelps (1966) in the context of human capital
accumulation. See Edwards (1989, 1992) for an application to the case of trade policy.
7
Barro and Sala-i-Martin (1995), Grossman and Helpman (1991) and Obstfeld and Rogoff (1996)
develop detailed micro models of innovation and imitation with similar implications.

# Royal Economic Society 1998

388 THE ECONOMIC JOURNAL [MARCH
from Neruh and Dhareshwar (1993). The following equation was estimated
using 3,555 observations panel data set:
g jt ˆ á d log K jt ‡ â d log L jt ‡ ë ‡ æ j ‡ E t ‡ î jt , (2)
where æ, E, and î are a country speci®c, time speci®c and common i.i.d error
terms, and ë is a common ®xed effect term. á ‡ â were restricted to add up to
one. The estimated factor shares were then used to construct yearly estimates
of TFP growth. Finally, these were averaged to construct decade-long estimates
of TFP growth. A 1960±90 average was also constructed. In addition to (2), a
growth equation that included human capital as an additional regressor was
estimated. These estimates were used to construct a second set of TFP growth
series, called TFPH. These two TFP growth variables behave very similarly ± a
regression of average TFP on average TFPH for the complete 1960±90 period
yielded a slope coef®cient of 0.93 with a standard error of 0.04.
Two batteries of consistency checks suggest that new estimates of TFP growth
are `reasonable'. First, correlation coef®cients between these new indexes and
those calculated by Neruh and Dhareshwar (1994) and Fischer (1993) were
high (between 0.77 and 0.91). And second, the orders of magnitude of the
new TFP growth estimates for a number of randomly selected countries
correspond approximately to what has been calculated by other authors.
Cook's (1977) distance statistic, however, indicated that Iraq was a gross
outlier, with an estimated average rate of TFP growth of ÿ13% in the 1980s.
After removing Iraq from the sample, the number of observations was 92
countries. Table 1 contains summary statistics for the new estimates of average
TFP growth for 1960±90 and the 1980s. As may be seen, during the 1980s
there was a productivity slowdown; moreover, the variability of TFP growth
across countries (measured by the standard deviation) increased greatly during
this decade.8

Table 1
New TFP Growth Estimates: Summary Statistics
TFP TFPH TFP TFPH
1960±90 1960±90 1980±90 1980±90

First quartile 0.004 0.004 ÿ0.007 ÿ0.008

Median 0.011 0.010 0.008 0.007
Third quartile 0.018 0.018 0.016 0.015
Average 0.009 0.009 0.003 0.003
Standard deviation 0.013 0.012 0.026 0.022

Source: See text.

8
These new TFP growth estimates can provide some additional light on the recent debate on the
role of productivity improvements in the East Asian `miracle' countries. See Harberger (1996),
Krugman (1994) and Young (1995). The new estimates suggest that while Taiwan's TFP growth has
been very high, that of Korea and Singapore has been more mediocre.

# Royal Economic Society 1998

1998] OPENNESS, PRODUCTIVITY AND GROWTH 389
Openness
In this paper I use nine alternative openness indexes to analyse the connection
between trade policy and productivity growth during 1980±90. While three of
these indexes measure openness proper, the other six measure the extent of
trade policy-induced distortions. The following openness indexes were used (a
higher value denotes a lower degree of policy intervention in international
trade): 1. Sachs and Warner Openness Index (OPEN): This binary index takes a
value of 1 if the country is considered to be open in that particular year, and
zero if it is closed. I averaged it by decades to generate summary indicators. 2.
World Development Report Outward Orientation Index (WDR): This indicator classi-
®es countries in four categories, according to their perceived degree of open-
ness. In most of the regressions I use the index for 1973±85. 3. Leamer's
Openness Index (LEAMER): An openness index estimated by Leamer (1988) as
the average residuals from disaggregated trade ¯ows regressions was used.
In addition to these three indicators of openness, the following six indexes
of trade distortions were used ± in these cases a higher value denotes a
greater departure from free trade: 4. Average Black Market Premium (BLACK):
The average value of the black market premium in the foreign exchange rate
market during the 1980s was used as a proxy for the overall extent of
distortions in the external sector. The raw data are from Barro and Lee
(1994). 5. Average Import Tariff on Manufacturing (TARIFF): The average import
tariff for 1982 reported by UNCTAD and provided by Barro and Lee (1994)
was used. 6. Average Coverage of Non Tariff Barriers (QR): This variable is also
taken from the Barro and Lee (1994) data set. Its source is UNCTAD. 7. The
Heritage Foundation Index of Distortions in International Trade (HERITAGE): This
index takes values of 1 to 5 and tries to measure the extent to which
government policy distorts trade. 8. Collected Trade Taxes Ratio (CTR): This
variable was constructed as the average for 1980±85 of the ratio of total
revenues on taxes on international trade (imports plus exports) to total trade.
The index was calculated from raw data obtained from the IMF. 9. Wolf's
Index of Import Distortions (WOLF): Wolf's (1993) regression-based index of
imports distortions for 1985.
Table 2 contains summary statistics for these nine indexes. As may be seen,
their coverage varies signi®cantly, ranging from 87 countries in the case of
BLACK to only 38 countries for WDR. Spearman rank coef®cients suggest that,
in spite of some important differences, these indicators tend to tell a somewhat
similar story. The rank coef®cients have, in all cases, the expected sign and
range (in absolute value) from 1.0 to 0.25. Fig. 1 plots four of these trade
distortions indexes against average TFP growth for the 1980s.9 In all cases
there is evidence of a negative relationship: countries with a higher level of
trade distortions have had lower TFP growth. The regression analysis reported
below deals with this issue in greater detail.

9
These four indexes were selected to be displayed because they have `continuous' values. All other
indexes ± with the exception of Leamer's openness indicator ± are discrete.

# Royal Economic Society 1998

390 THE ECONOMIC JOURNAL [MARCH
Table 2
Alternative Trade Orientation Indexes: Summary Statistics
Index Obs Mean Std. Dev. Min Max Period
OPEN 63 0.36 0.44 0 1 1980s
WDR 38 1.97 0.88 1 4 1973±82
LEAMER 49 ÿ0.32 0.63 ÿ2.8 0.22 1982
BLACK 87 0.63 1.68 0 14.00 1980s
TARIFF 80 0.18 0.18 0.012 1.32 1982
QR 78 0.21 0.25 0 0.88 1982
HERITAGE 69 3.75 1.26 1 5 late 1980s
CTR 58 0.03 0.03 0.0003 0.16 1980±85
WOLF 62 38.10 21.43 1 75 1985

Source: see text.

Fig. 1. TFP Growth and Alternative Measures of Trade Distortions

# Royal Economic Society 1998
1998] OPENNESS, PRODUCTIVITY AND GROWTH 391
2.3. Basic Results
The basic regression results reported in this section follow the model sketched
above and focus on three possible determinants of TFP growth: (1) The log of
initial GDP per capita (GDP 65). This variable captures the existence of TFP
convergence, and its coef®cient is expected to be negative. (2) Initial level of
human capital. This variable, measured as the mean number of years of
education in 1965, captures the view that countries with a more developed
educational system have a greater ability to innovate and absorb new ideas. (3)
Openness. I use the nine alternative openness indexes for the 1980s discussed
above to analyse whether, when controlling for other factors, countries with a
lower degree of external distortions have tended to have faster productivity
growth.10 The results reported here were obtained using TFP as the dependent
variable; when TFPH was used, however, the main thrust of the results was
maintained. The analysis concentrated on 1980s averages since many of the
openness indicators are only available for that period. In Section 4 I analyse
the potential role of other regressors, the existence of outliers, and alternative
functional forms and time periods.
Tables 3 and 4 contain the basic results from TFP growth regressions using
weighted and instrumental weighted least squares (see the notes to the tables
for details on the estimation). As may be seen, in every regression the coef®-
cient of initial GDP per capita is negative, and in 16 out of the 18 cases it is
signi®cant at conventional levels. These results indicate that after controlling
for the initial level of human capital and the degree of openness, TFP exhibits
(conditional) convergence (see also Edwards (1995)). The very small value of
the estimated coef®cients suggest, however, that this convergence process is
extremely slow. When human capital and openness are excluded, the coef®-
cient of GDP 65 becomes positive, indicating that the hypothesis of absolute
TFP convergence is rejected.
A second important result in Tables 3 and 4 is that in every one of the 18
regressions the coef®cient of initial human capital is positive, as expected, and
signi®cant at conventional levels. When alternative measures of human capital
were used ± including measures of the quality of education, such as the
teacher=students ratio ±, the thrust of the results was maintained.
Third, and more important for the purposes for this study, in all but one of
the 18 equations the estimated coef®cient of the openness indicator has the
expected sign, and in the vast majority of the cases (13 out of 17) it was
signi®cant.11 In my view these results are quite remarkable, suggesting with
tremendous consistency that there is a signi®cantly positive relationship be-
tween openness and productivity growth. The fact that this relationship is

10
A dif®culty faced in this analysis is that for many of the (potential) determinants of TFP ±
including the alternative openness indexes ± data availability is limited; for many regressors the country
coverage is smaller than the original 92-country sample. After examining the different series, and in
contrast to Hall and Jones (1996), I decided against interpolating the missing observations, since there
are almost no bases for doing it. The sample size, thus, is limited and dictated by the availability of data.
11
Notice that although the coef®cient of WOLF in eq. 9 has the `wrong' sign, it is not signi®cant.
# Royal Economic Society 1998
392 THE ECONOMIC JOURNAL [MARCH
Table 3
TFP Growth Regressions (Weighted Least Squares)
Eq. No. Openness Trade
Measure GDP 65 HUMAN 65 Orientation R2 N
1. OPEN ÿ0.011 0.005 0.94E-2 0.24 51
(ÿ2.41) (3.27) (2.12)
2. WDR ÿ0.013 0.004 0.75E-2 0.45 32
(ÿ2.53) (2.17) (3.57)
3. LEAMER ÿ0.005 0.003 0.41E-2 0.23 44
(ÿ0.90) (1.94) (1.03)
4. BLACK ÿ0.008 0.003 ÿ0.022 0.28 75
(ÿ2.43) (2.53) (ÿ3.59)
5. TARIFF ÿ0.010 0.003 ÿ0.045 0.24 67
(2.69) (2.99) (ÿ2.77)
6. QR ÿ0.008 0.004 ÿ0.005 0.16 66
(ÿ2.06) (3.19) (ÿ0.54)
7. HERITAGE ÿ0.007 0.002 ÿ0.58E-2 0.42 58
(ÿ2.81) (2.58) (ÿ4.56)
8. CTR ÿ0.017 0.004 ÿ0.484 0.34 45
(ÿ3.24) (3.34) (ÿ3.04)
9. WOLF ÿ0.009 0.004 0.35E-4 0.14 53
(ÿ1.91) (2.83) (0.27)
 Each row corresponds to a TFP growth regression using a different openness indicator. The indicator
being used is identi®ed in column 2 (openness measure), and its estimated coef®cients appears in
column 5 (trade orientation). All the regressions were estimated using weighted least squares. GDP per
capita in PPP dollars in 1985 was used as a weight. These equations were estimated with a constant. Its
estimated value, however, is not reported due to space restrictions. The numbers in parentheses are t-
statistics.

found for eight very different measures of openness, covering (slightly) differ-
ent samples, and under alternative estimation techniques cannot be dismissed
lightly. The computation of standardised beta coef®cients, however, indicates
that in spite of having the expected sign and being signi®cant, openness is
relatively less important than initial GDP and human capital in explaining
cross country differences in TFP growth. In six of the nine regressions in Table
3 the trade orientation index had the lowest standardised beta, and only in two
(equations 2 and 7) it had the highest.
Although all nine of the indicators used in this study are proxies for
openness, not all of them capture the same aspect of trade policy. This is the
case, for example, of TARIFF, QR and BLACK, which cover very different angles
of commercial policy. This means that information may be lost by introducing
them separately in the regressions. Ef®ciency could be gained if some of the
nine openness indicators are combined into a `grand' composite index. I
explore this avenue by using a principle components approach. In order to
maintain a reasonable size sample I restrict the exercise to those indicators
with at least 60 countries: OPEN, BLACK, TARIFF, QR and WOLF. The ®rst
principle component was called COM, and explained more that 60% of the
variance of these ®ve indicators. Since a higher value of this hybrid trade
orientation index re¯ects a higher degree of trade intervention and distor-
# Royal Economic Society 1998
1998] OPENNESS, PRODUCTIVITY AND GROWTH 393
Table 4
TFP Growth Regressions (Instrumental Weighted Least Squares)
Eq. No. Openness Trade
Measure GDP 65 HUMAN 65 Orientation R2 N
10. OPEN ÿ0.011 0.005 0.89E-2 0.24 49
(ÿ2.28) (3.187) (2.84)
11. WDR ÿ0.011 0.003 0.013 0.41 30
(ÿ2.04) (2.35) (3.36)
12. LEAMER ÿ0.010 0.002 0.20E-2 0.22 33
(ÿ1.36) (2.24) (1.95)
13. BLACK ÿ0.008 0.003 ÿ0.019 0.27 71
(ÿ2.32) (2.23) (ÿ1.95)
14. TARIFF ÿ0.014 0.003 ÿ0.106 0.10 64
(ÿ2.07) (1.97) (ÿ2.95)
15. QR ÿ0.010 0.004 ÿ0.004 0.12 63
(ÿ2.00) (2.29) (ÿ1.42)
16. HERITAGE ÿ0.011 0.001 ÿ0.011 0.36 56
(ÿ3.23) (1.97) (ÿ3.74)
17. CTR ÿ0.040 0.005 ÿ1.67 0.24 42
(ÿ2.42) (2.28) (ÿ2.15)
18. WOLF ÿ0.009 0.003 ÿ0.26E-4 0.04 51
(ÿ1.87) (2.49) (ÿ0.72)
 Each row corresponds to a TFP growth regression using a different openness indicator. The indicator
being used is identi®ed in column 2 (openness measure), and its estimated coef®cients appears in
column 5 (trade orientation). All the regressions were estimated using instrumental weighted least
squares. Instruments were selected using the usual criteria. Additionally, an effort was made to have as
many observations as possible in each regression. The following instruments were used (although not
all of them in every equation): TFP growth in the 1970s, OPEN in the 1970s, imports=GDP ratio in the
1970s, exports=GDP ratio in the 1970s, the average black market premium in the 1970s, the Heritage
Foundation index of property rights protection, and the change in the terms of trade. These equations
were estimated with a constant. Its estimated value, however, is not reported due to space restrictions.
The numbers in parentheses are t-statistics.

tions, its signs expected to be negative.12 The weighted least squares estimate
of the TFP using this composite trade orientation index yielded the following
result (t-statistics in parentheses):
TFPGROWTH ˆ 0:08 ÿ 0:013 GDP 65 ‡ 0:005 HUMAN 65 ÿ 0:07 COM
(2:0) (ÿ2:3) (2:7) (ÿ2:8)
N ˆ 60; R2 ˆ 0:32:
This con®rms the ®ndings reported above suggesting that, with other things
given, more open countries will tend to experience faster productivity growth
than more protectionist countries.

12
The ®rst principle component is computed as: COM ˆ ÿ0:469OPEN ‡ 0:320BLACK ‡
0:494TARIFF ‡ 0:553QR ‡ 0:354WOLF . The openness indicators have been standardised to having
zero mean and unitary variance. A well known limitation of this approach is that the coef®cients used to
construct the principle components have no economic meaning. In this case, however, they have the
expected signs.

# Royal Economic Society 1998

394 THE ECONOMIC JOURNAL [MARCH
3. Omitted Variables, Outliers and Nonlinearities
3.1. Omitted Variables
An important question is whether the openness indicators used in this paper
indeed re¯ect the effects of trade policy on productivity growth, or whether
they are capturing the possible effect of omitted variables. In this section I
explore this issue by considering the role of institutions, politics and macro-
economic stability as (potential) determinants of TFP growth.
Institutions: North (1990), among others, has argued that institutions are
important determinants of entrepreneurship and growth. Institutions that
lower transaction costs allow individuals to devote more time to productive
pursuits, instead of protecting their interests from the rapacity of others.
Property rights protection is, perhaps, the most important among these
institutions. I explore its role by including a comparative index of property
rights protection calculated by Johnson and Sheehy (1995) in the regressions.
This index (PROPERTY ) captures the extent to which the legislation and the
judiciary guarantee and protect property rights; the index can take values from
1 to 5, with higher numbers re¯ecting a more tenuous degree of protection.
Political Instability: Societies subject to a greater degree of political upheaval are
more volatile, and tend to discourage investment in innovation and productiv-
ity enhancements (Barro, 1996). I used a series of proxies to investigate the
extent to which the political system affects TFP growth. The results reported in
this section were obtained when the average number of politically motivated
attacks during the 1980s was included. This variable was called POLINST.
Macroeconomic Instability: A number of authors have argued that greater macro-
economic instability ± and in particular, higher in¯ation rates ± tends to affect
economic performance negatively (Fischer, 1993). The theoretical work on the
subject, however, has not speci®cally focused on TFP growth. This means that,
in a way, the question of how in¯ation affects productivity growth is still
somewhat open. In this paper I investigate this issue by introducing an estimate
of in¯ation tax revenues as a percentage of GDP into the regression analysis.
The following equation is representative of the results obtained when these
additional variables were introduced into the regression analysis.13
TFPGROWTH ˆ 0:05 ÿ 0:007 GDP 65 ‡ 0:002 HUMAN 65 ‡ 0:08 OPEN
(1:9) (ÿ1:9) (1:7) (2:4)
ÿ 0:006 PROPERTY ÿ 0:012 POLINST ÿ 0:001 INFTAX
(ÿ2:0) (ÿ1:5) (ÿ0:4)
N ˆ 39; R2 ˆ 0:56:
The results are quite interesting. First, the coef®cients of the original variables
were not affected by the inclusion of the new regressors. In particular, these

13
The complete set of regressions is not reported due to space considerations. Overall, however,
they continued to support the idea that more open economies experience a faster rate of productivity
growth. The complete set is available on request.

# Royal Economic Society 1998

1998] OPENNESS, PRODUCTIVITY AND GROWTH 395
results continue to endorse the idea that there is a positive relationship
between openness and productivity growth, Second, the protection of property
rights plays an important role in explaining cross country differences in TFP
growth. The standardised beta for PROPERTY is 0.4, higher than that for the
openness variables. Third, the in¯ation tax does not appear to have affected
TFP growth signi®cantly.

3.2. Outliers
I used two alternative methods for analysing whether the results reported
above were being driven by outliers. Both Cook's (1977) distance statistics and
the dbetas index for the marginal effect of individual observations suggest that
Trinidad and Tobago may be (somewhat) of an outlier. However, when this
observation is excluded the results are not affected in any fundamental way.

3.3. Nonlinearities and Alternative Time Periods

An interesting question is whether the relationship between TFP growth and
openness exhibits some nonlinearities. In order to explore this possibility I
added a quadratic term to a subset of the regressions in Table 3. I restricted
the analysis to those indicators which allow for continuous values. The results
were mixed and suggested that for some indicators (LEAMER, BLACK and
TARIFF ) these is some evidence of a nonlinear effect (the estimated quadratic
coef®cient was positive). When the principle components-based overall indica-
tor was considered, however, its squared term was nonsigni®cant.
Finally, in order to analyse whether the results reported in this paper were
period-speci®c, I estimated TFP growth regressions for the complete 1960±90
period. A limitation with this analysis, however, is that most of the openness
measures are available for 1980±90, or a subperiod of this decade. Sachs and
Warner's (1995) index is the only measure available for the complete period.
The estimation of a TFP growth equation for 1960±90 yielded the following
result:
TFPGROWTH ˆ 0:01 ÿ 0:003 GDP 65 ‡ 0:001 HUMAN 65 ‡ 0:07 OPEN
(0:9) (ÿ1:4) (2:1) (2:6)
N ˆ 51; R2 ˆ 0:20:
The positive relationship between openness and TFP growth seems to hold for
longer periods of time; it is not the consequence of the selection of a speci®c
time period for the analysis.

4. Conclusions
Empirical work on trade and openness has often looked suspicious. This is
mainly because of serious measurement and endogeneity problems (Rodrik,
1995). In this paper I have investigated this issue using a new data set and
# Royal Economic Society 1998
396 THE ECONOMIC JOURNAL [MARCH
making a special effort to analyse the robustness of the ®ndings. The regres-
sions reported here are robust to the use of openness indicator, estimation
technique, time period and functional form, and suggest that more open
countries have indeed experienced faster productivity growth. Although the
use of instrumental variables goes a long way towards dealing with endogene-
ity, issues related to causality are still somewhat open, and will require time
series analyses to be adequately addressed. Although I do believe that these
results are forceful and persuasive, I am convinced that further understanding
on the economics of innovation and productivity growth ± and its connection
to openness ± will require additional detailed microeconomic work at the
country level.
University of California, Los Angeles
and
National Bureau of Economic Research

Appendix: List of Countries

Algeria Honduras Norway
Angola Iceland Pakistan
Argentina India Panama
Australia Indonesia Paraguay
Austria Iran Peru
Bangladesh Iraq Phillippines
Belgium Ireland Portugal
Bolivia Israel Rwanda
Brazil Italy Senegal
Cameroon Jamaica Sierra leone
Canada Japan Singapore
Chile Jordan South Africa
China Kenya Spain
Colombia Korea Sri Lanka
Costa Rica Kuwait Sudan
Cote d'Ivoire Libya Sweden
Cyprus Luxembourg Switzerland
Denmark Madagascar Taiwan
Dominican Republic Malaysia Tanzania
Ecuador Malawi Thailand
Egypt Mali Trinidad
El Salvador Malta Tunisia
Ethiopia Mauritius Turkey
Finland Mexico UK
France Morocco USA
Germany Mozambique Uganda
Ghana Myanmar Uruguay
Greece Netherlands Venezuela
Guatemala New Zealand Zaire
Guyana Nicaragua Zambia
Haiti Nigeria Zimababwe

# Royal Economic Society 1998

1998] OPENNESS, PRODUCTIVITY AND GROWTH 397
References
Anderson, James E. (1994), `Trade restrictiveness benchmarks', The World Bank.
Anderson, James E. and Neary, Peter (1994), `Measuring the restrictiveness of trade policy', World Bank
Economic Review, vol. 8(1), pp. 151±69.
Balassa, Bela (1982), Development Strategies in Semi-Industrial Countries, Oxford: Oxford University Press.
Barro, Robert J. (1996), `Determinants of economic growth: a cross-country empirical study', National
Bureau of Economic Research, Working Paper no. 5698.
Barro, Robert J. and Lee, Jong-Wha (1994), `Sources of economic growth'. Carnegie-Rochester Conference
Series on Public Policy.
Barro, Robert J. and Sala-i-Martin, Xavier (1995), Economic Growth, New York; McGraw-Hill.
Cook, R. D. (1977), `Detection of in¯uential observations in linear regressions', Technometrics, vol. 19,
pp. 15±8.
Corden, W. Max (1966), `The effective protective rate, the uniform tariff equivalent and the average
tariff.' Economic Record, vol. 42, pp. 200±16.
Edwards, Sebastian (1989), `Openness, trade liberalization and economic performance in developing
countries', National Bureau of Economic Research Working Paper no. 2908.
Edwards, Sebastian (1992), `Trade orientation, distortions and growth in developing countries', Journal
of Development Economics, vol. 39(1), pp. 31±58.
Edwards, Sebastian (1993), `Openness, trade liberalization and growth in developing countries', Journal
of Economic Literature, vol. 31(3), pp. 1358±93.
Edwards, Sebastian (1995), `Trade policy, exchange rates and growth', in (R. Dornbusch and S.
Edwards, eds.), Reform, Recovery and Growth: Latin America and the Middle East, Chicago: University of
Chicago Press.
Feenstra, Robert C,. Lipsey, Robert E. and Bowen, Harry P. (1997), `World trade ¯ows, 1970±92, with
production and tariff data', National Bureau of Economic Research Working Paper no. 5910.
Fischer, Stanley (1993), `The role of macroeconomic factors in growth', Journal of Monetary Economics,
vol. 32, no. 3, pp. 485±512.
Greenaway, David and Nam, Chong H. (1988), `Industrialisation and macroeconomic performance in
developing countries under alternative trade strategies', Kyklos, vol. 41, no. 3, pp. 419±35.
Grossman, Gene and Helpman, Elhanan (1991), Innovation and Growth in the Global Economy, Cam-
bridge: MIT Press.
Johnson, Bryan and Sheehy, Thomas (eds.) (1996), 1996 Index of Economic Freedom, Washington: The
Heritage Foundation.
Hall, Robert E. and Jones, Charles I. (1996), `The productivity of nations', National Bureau of
Economic Research Working Paper no. 5812.
Harberger, Arnold C. (1996) `Re¯ections on economic growth in Asia and the Paci®c', Journal of Asian
Economics, vol. 7, no. 3, pp. 365±92.
Krueger, Anne (1978), Foreign Trade Regimes and Economic Development, Cambridge: Ballinger.
Krugman, Paul (1994), `The myth of Asia's miracle', Foreign Affairs, November=December, pp. 62±78.
Leamer, Edward (1988), `Measures of openness' in (R. Baldwin, ed.) Trade Policy and Empirical Analysis,
Chicago: Univeristy of Chicago Press.
Levine, Ross and Renelt, David (1992), `A sensitivity analysis of cross-country growth regressions',
American Economic Review, vol. 82, no. 4, pp. 942±63.
Lucas, Robert E. (1988), `On the mechanics of economic development', Journal of Monetary Economics,
vol. 22, no. 1, pp. 3±42.
Michaely, Michael, Papageorgiou, Demetris and Choksi, Armeane (eds) (1991), Liberalizing Foreign
Trade, Oxford: Blackwell.
Nelson, Richard and Phelps, Edmund S. (1966), `Investment in humans, technological defusion and
economic growth', American Economic Review, vol. 56, no. 2, pp. 69±75.
Neruh, Vikram and Dhareshwar, Asok (1993), `A new database on human physical capital stock:
sources, methodology and results', Revista Analisis Economico, vol. 8, no. 1, pp. 37±59.
Neruh, Vikram and Dhareshwar, Asok (1994), `New estimates of total factor productivity growth for
developing and industrial countries', Policy Research Working Paper 1313, The World Bank.
North, Douglass (1990), Institutions, Institutional Change and Economic Performance, Cambridge and New
York: Cambridge University Press.
Obstfeld, Maurice and Rogoff, Kenneth (1996), Foundations of International Macroeconomics, Cambridge:
MIT Press.
Pritchett, Lant (1991), `Measuring outward orientation in developing countries: can it be done?',
Working Paper, The World Bank.
Pritchett, Lant and Sethi, Geeta (1994), `Tariff rates, tariff revenue and tariff reform: some new facts',
World Bank Economic Review, vol. 8(1), pp. 1±16.
Rodrik, Dani (1995), `Trade policy and industrial policy reform' in (Jere Behrman and T. N. Srinivasan,
eds) Handbook of Development Economics, vol. 3B, Amsterdam: North Holland.

# Royal Economic Society 1998

398 THE ECONOMIC JOURNAL [ M A R C H 1998]
Romer, Paul M. (1986), `Increasing returns and long run growth', Journal of Political Economy, vol. 94,
no. 5, pp. 1002±37.
Romer, Paul M. (1992), `Two strategies for economic development: using ideas and producing ideas',
World Bank Annual Conference on Economic Development, Washington, DC, The World Bank.
Sachs, Jeffrey and Warner, Andrew (1995), `Economic reform and the process of global integration',
Brookings Papers on Economic Activity, vol. 1, pp. 1±118.
Wade, Robert (1994), `Is the miracle study right?' Working Paper, Washington, DC, Overseas Develop-
ment Council.
Wolf, Holger (1993), `Trade orientation: measurement and consequences', Estudios de EconomõÂa,
vol. 20(2), pp. 52±72.
World Bank (1987), World Development Report 1987, Oxford: Oxford University Press.
Young, Alwyn (1995), `The tyranny of numbers: confronting the statistical realities of the East Asian
growth experience', Quarterly Journal of Economics, vol. 109, no. 3, pp. 641±80.

# Royal Economic Society 1998