WPS3763
OPENNESS CAN BE GOOD FOR GROWTH:
THE ROLE OF POLICY COMPLEMENTARITIES*
Roberto Chang Linda Kaltani Norman Loayza
Rutgers University American University World Bank
Abstract. This paper studies how the effect of trade openness on economic growth depends on
complementary reforms that help a country take advantage of international competition. This
issue is illustrated with a simple HarrisTodaro model where output gains after trade liberalization
depend on the degree of labor market flexibility. In that model, trade protection may ameliorate
the problem of underemployment (and underproduction) in sectors affected by labor market
distortions; hence trade liberalization unambiguously increases per capita income only when
labor markets are sufficiently flexible. We then present some panel evidence on how the growth
effect of openness depends on a variety of structural characteristics. For this purpose, we use a
nonlinear growth regression specification that interacts a proxy of trade openness with proxies of
educational investment, financial depth, inflation stabilization, public infrastructure, governance,
labormarket flexibility, ease of firm entry, and ease of firm exit. We find that the growth effects
of openness are positive and economically significant if certain complementary reforms are
undertaken.
Key Words: Openness, Growth, Economic Reform, Policy Complementarity
JEL Classification: E61, F13, F43, O40
World Bank Policy Research Working Paper 3763, November 2005
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the
exchange of ideas about development issues. An objective of the series is to get the findings out quickly,
even if the presentations are less than fully polished. The papers carry the names of the authors and should
be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely
those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors,
or the countries they represent. Policy Research Working Papers are available online at
http://econ.worldbank.org.
* We gratefully acknowledge the valuable support of the World Bank's Poverty Reduction and Economic
Management Network. We are also grateful to César Calderón, Rómulo Chumacero, Marco Cipriani,
Mieke Meurs, Romain Ranciere, Klaus SchmidtHebbel, Luis Servén, Vicente Tuesta, Roberto Zhaga, and
seminar participants at George Washington University, the World Bank, the Central Bank of Chile, the
Central Bank of Peru, the Tenth Conference of Dynamics, Growth, and International Trade, and the Paris
Meetings of the Latin American and Caribbean Economic Association for valuable comments and
suggestions. Finally, Chang thanks the hospitality and financial support of Princeton University.
OPENNESS CAN BE GOOD FOR GROWTH:
THE ROLE OF POLICY COMPLEMENTARITIES
1. INTRODUCTION
Ever since Ricardo's critique on the Corn Laws to the current debate on
globalization, few topics in economics have been more hotly contested as the importance
of international trade openness for economic development. The arguments in favor of
openness are well known and date back at least to Adam Smith's analysis of market
specialization. Openness promotes the efficient allocation of resources through
comparative advantage, allows the dissemination of knowledge and technological progress,
and encourages competition in domestic and international markets. Standard trade theory
captures the gains from openness as movements towards the production possibilities
frontier. To this incomelevel effect, recent theoretical models add a longrun growth
effect when the areas of specialization promoted by trade enjoy increasing returns to scale,
as illustrated in the endogenous growth models of Young (1991), Grossman and Helpman
(1991), Eicher (1993), and Lee (1993).
Opposing arguments are not too hard to build. If market or institutional
imperfections exist, openness can lead to subutilization of human and capital resources,
concentration in extractive economic activities, or specialization away from
technologically advanced, increasingreturn sectors. Grossman and Helpman (1991) and
Matsuyama (1992) provide theoretical models where a technologically backward country
specializes in a nondynamic sector as a result of openness, thus losing out on the benefits
of increasing returns. Underlying these models there is an imperfection in contracts or in
financial markets that makes people obey a myopic notion of comparative advantage.
Sachs and Warner (1995, 1999) develop a model where specialization in extractive,
naturalresource sectors prevents a country from the technological progress that eventually
leads to longrun growth. In this case, the underlying imperfection is an institutional
weakness that encourages naturalresource depletion for quick gains appropriated by
certain groups in society. Rodríguez and Rodrik (2000) review the theoretical arguments
as to why openness can be detrimental to developing countries; they do so in a secondbest
2
context, in which trade liberalization is the policy lever and the eventual culprit while
market and institutional weaknesses are accepted as immanent characteristics.
The theoretical ambiguity on the effects of openness is reflected in the available
empirical evidence. Some papers point to strongly positive growth effects by trade
liberalization. This is the case of Dollar (1992) and Sachs and Warner (1995), who run
crosscountry growth regressions on composite indices of the stance of trade policy; as
well as Edwards (1998), who prefers to base his positive evaluation by examining the
robustness of various individual indicators of trade liberalization in crosscountry growth
regressions. But others, most notably Harrison (1996) and Rodriguez and Rodrik (2000)
have cast doubt on the significance and robustness of the growth benefits of openness.
Their critique starts with the openness measures used in practice; for instance, some
purportedly openness indicators reflect general poor economic management (e.g., the black
market premium) or are primarily affected by geographic characteristics (e.g., the trade
volume). Other criticisms are based on econometric grounds, such as omittedvariable bias
(particularly due to the exclusion of institutional and geographyrelated variables) and joint
endogeneity bias (stemming from the effect of growth on certain policy regimes).
Recent empirical studies have addressed these criticisms by emphasizing the over
time variation in openness indicators and growth performance. (Harrison 1996 had already
noted that panel studies rendered a more positive evaluation of the growth effects of
openness than crosssectional studies.) Dollar and Kraay (2004) and Loayza, Fajnzylber,
and Calderón (2005) run growth regressions on panel data of large samples of countries.
Both papers use indicators fo openness based on trade volumes and control for their joint
endogeneity and correlation with countryspecific factors through GMM methods that
involve taking differences of data and instruments. This implies that, although they
continue to use crosscountry data, these papers favor withincountry changes as the main
source of relevant variation. Both papers conclude that opening the economy to
international trade brings about significant growth improvements. Wacziarg and Welch
(2003) arrive at a similar, though more nuanced, conclusion from a methodologically
different standpoint. Using an eventstudy methodology where an event is defined as a
year of substantial trade policy liberalization they find that liberalizing countries tend to
experience significantly higher volumes of trade, investment rates, and, most importantly,
3
growth rates. However, in an examination of 13 countrycase studies, Wacziarg and
Welch find noticeable heterogeneity in the growth response to trade liberalization.
Although their small sample does not allow for definite conclusions, it appears that the
growth response after liberalization is positively related to conditions of political stability.
This paper starts with the observation that although opening to trade is beneficial to
economic growth on average, the aftermath of trade liberalization varies considerably
across countries and depends on a variety of conditions related to the structure of the
economy and its institutions. A simple exercise may serve to convey this point. Figure 1
plots changes in growth rates of per capita GDP between the 1990s and 1980s versus
changes in the volume of trade (as a ratio to GDP) between those two decades for a
worldwide sample of 82 countries. Figure 1 has four panels; in each of them we separate
the country observations according to whether they belong to the top onethird (diamonds)
or bottom twothirds (circles) of a rank distribution given by, in turn, each of the following
criteria: a) secondary enrollment rates (a proxy for human capital investment); b) main
telephone lines per capita (a proxy for public infrastructure); c) a subjective index of the
quality of governance; and d) a de facto and de jure index of labor market flexibility.
(Appendix 2 gives details on variable definitions and sources.) Each criterion used for
ranking country observations is measured over the 1980s, the beginning period.
Dividing the country observations into top and bottom groups allows us to compare
the corresponding slopes for the relationship between changes in trade volume ratios and
changes in economic growth rates. In all panels, the OLS line described by the bottom
observations is basically flat, implying no relationship between trade opening and growth
improvement in the bottom groups. However, this changes for the top groups: for the top
observations, the slope of the OLS line is positive and steeper than that for the bottom
group.1 This clearly suggests that the empirical impact of trade opening on growth may
depend on the existence and degree of distortions in non trade areas. Of course, this is
quite a simple exercise and it does not control for other growth determinants (such as
initial per capita GDP), does not account for joint endogeneity, and does not use all
1 For the top observations according to educational investment, public infrastructure, and governance, the
slope of the OLS line is significantly positive at conventional levels. The top observations according to labor
market flexibility also describe a positive slope that is larger than that of the bottom group, but it is not
statistically significant. The more satisfactory methods later in the paper, however, indicate that the impact
of labor market reforms on the trade opening/growth relationship is in fact strongly significant.
4
information efficiently. But more careful econometric methods are used later in the paper
and confirm that the growth response to trade opening is heterogeneous, and not in random
ways but in relation to specific country conditions.
This paper studies how the eventual success of openness in terms of growth
performance depends on the economic and institutional characteristics that make a country
able to adjust to the new conditions imposed by international competition. This idea is
very general, but for concreteness our discussion starts with a simple theoretical example
where the gains in output after trade liberalization depend on the degree of labor market
flexibility. The example is a version of the well known HarrisTodaro model, and labor
market distortions are represented by a minimum wage that applies to the formal sector of
the economy. Trade restrictions are modeled as a tariff that also applies to formal sector
output. In the model, trade protection may serve to ameliorate the problem of
underemployment (and underproduction) in the sector affected by labor market distortions.
As a consequence, trade liberalization unambiguously increases per capita income only
when labor market distortions are sufficiently small. 2
Our model continues the examination of commercial policy in the presence of
labormarket distortions (Brecher 1974). But we regard it, more generally, as an example
in the tradition of the general theory of the second best (Lipsey and Lancaster 1956). From
this perspective, one should expect similar interactions between openness and
complementary reforms in other areas. For instance, in the influential work of Acemoglu
and Zilibotti (2001), openness (in the sense of unobstructed access to technological
progress) does not lead to productivity improvements in developing countries that fail to
improve their human capital (to adopt the new technologies) and to enforce intellectual
property rights (to encourage the development of technologies best suited to their skill
mix). Likewise, Banerjee and Newman (2004) have recently presented a model in which
lack of financial development and sluggish factor mobility make poor countries lose from
2In the model the potential gains from openness are given in terms of the level of output per capita. There is
no contradiction between this static treatment and our empirical emphasis on growth effects. This is so
because the time horizons used in current econometric studies do not allow discriminating longrun growth
effects from longlasting transitional level effects. Moreover, the finding of conditional convergence
suggests that growth impulses coming from improvements in growth determinants tend to decrease as per
capita GDP increases.
5
trade openness, as unproductive sectors are wiped out by foreign competition but the
capital and labor attached to them fail to divert to more efficient uses.
We then present some crosscountry empirical evidence on how the growth effect
of openness depends on a variety of structural characteristics, including some that may be
subject to reform. We build on the paneldata growth regressions presented in Dollar and
Kraay (2004) and Loayza, Fajnzylber, and Calderón (2005). As these papers do, we use a
GMM procedure that controls for endogeneity and unobserved countryspecific factors in
order to estimate the growth effect of openness, as well as those of other policy and non
policy variables. We, however, depart from those studies in that we interact the openness
measure with proxies of, respectively, educational investment, financial depth, inflation
stabilization, public infrastructure, governance, labormarket flexibility, and ease of firm
entry and exit. Our objective for using this nonlinear specification is to assess whether an
increase in openness may have a growth effect that depends on country characteristics that,
at least in principle, are subject to improvement through economic and institutional
reforms. We find that the growth effect of openness is positive and economically
significant if certain complementary reforms are undertaken. This quantitative assessment
may contribute towards identifying the specific reforms that are most needed to
complement a trade liberalization agenda.
The empirical growth literature offers some examples of nonlinear specifications
considering interaction effects. On the related topic of foreign direct investment,
Borensztein, De Gregorio and Lee (1998) and Alfaro, Chanda, KalemliOzcan, and Sayek
(forthcoming) find that the growth effect of FDI is significantly positive only when the
host country has, respectively, sufficiently high human capital and financial depth.
Specifically in the analysis of growth effects of trade openness, an important antecedent of
our work is the empirical study by Bolaky and Freund (2004). Using crosscountry
regressions in levels and changes of per capita GDP and controlling for simultaneity via
external instruments, they find that trade opening promotes economic growth only in
countries that are not excessively regulated. They argue that in highly regulated countries,
growth does not accompany trade openness because resources are prevented from flowing
to the most productive sectors and firms, and trade is likely to occur in goods where
comparative advantage is actually missing. Finally, Calderón, Loayza, and Schmidt
6
Hebbel (2004) interact in their panel growth regressions a measure of openness (volume of
trade / GDP) with linear and quadratic terms of GDP per capita, which they regard as
proxy for overall development. They find that the growth effect of trade opening is nearly
zero for low levels of per capita GDP, increases at a decreasing rate as income rises, and
reaches a maximum at high levels of income. Our strategy of interacting openness with
specific country characteristics is, to some extent, an attempt to decipher what lies behind
the dependence of the growth effect of openness on economic reform and development.
In section 2, we present a theoretical model to illustrate the ambiguous effect of
trade opening once labormarket rigidities are present. Section 3 is devoted to the
empirical analysis. There, we first introduce the sample and methodology, and then we
present the econometric results, illustrating them with straightforward simulations. In
section 4 we offer some concluding remarks.
2. AN ILLUSTRATIVE MODEL
The basic idea of our work is that economic reforms need to complement each
other to be effective. This general principle can be derived from several models, and
indeed one can see it as a straightforward implication of the theory of the second best. But
we believe that it will be useful to illustrate the principle in a concrete situation. This
section does just that, in the context of a simple open economy model in the spirit of Harris
and Todaro (1970).
The justly celebrated HarrisTodaro model focused on endogenous migration and
unemployment in the presence of labor market distortions. In our version below,
distortions in the labor market interact with tariffs or other distortions in international
trade. We show that, under certain conditions, a tariff reform reduces traderelated
distortions but exacerbates the labor market distortions. The implication is that the sign of
the impact of trade opening on productive efficiency depends on labor market conditions.
This observation provides the basis for the empirical work in later sections.
7
Production and Employment
Consider a static small open economy. There are two consumption goods, indexed
by i=1, 2, whose world prices are given in terms of a fixed numeraire. Both goods can be
produced at home with a simple Cobb Douglas technology:
(2.1) Yi = AiLi , i = 1,2
i
Labor is the only variable input in production. Home firms are owned by (a
measure one continuum of) identical entrepreneurs, which behave competitively in product
and factor markets. Profit maximization then implies that, in each productive sector i=1,2,
the value of the marginal product of labor will be equal to the wage in that sector:
(2.2) iAiPiLi i1=Wi
where 0 < i < 1, Pi is the home price of good i, and Wi is the wage prevailing in sector i.
The price of good i in domestic markets, Pi , may differ from its world price
(henceforth denoted by Pi*) because of trade policy. In particular, if there is a tariff on
imports of good i, Pi > Pi*. A "trade reform" is a reduction in the difference between Pi
and Pi*.
Also, as in the classic HarrisTodaro model, wages may be different in different
sectors, and there is a minimum wage in sector 1, which is assumed to exceed the wage in
sector 2:
(2.3) W1 = Wmin > W2
Note that a "labor market reform" would involve eliminating the minimum wage in sector
1.
There are L workers in this economy. Each one chooses whether to work in sector 1
or 2. Once the location decision has been made, workers cannot move from one sector to
the other. The critical aspect of the HarrisTodaro model is that, in equilibrium, the number
of workers that choose to locate in sector 1 will be too large for all of them to be
8
employed. Hence there will be a number, which we denote by U, of unemployed workers
in sector 1. Assuming that jobs in sector 1 are distributed randomly among workers located
in that sector, the probability that a worker in sector 1 is employed is L1 /( L1+U). As we
shall see, optimal location decisions by workers imply that the expected wage in the two
sectors must be the same:
(2.4) W2 = [L1/( L1+U)]Wmin
By definition
(2.5) L1 + L2 + U = L
Equations (2.1)(2.5) suffice to describe the production side of the economy. Given
the minimum wage Wmin and home prices P1 and P2, (2.1)(2.5) can be solved for Y1,Y2, L1,
L2, U, and W2; this is indeed the standard discussion of the HarrisTodaro model.
To see the implications of a trade reform, we will assume that P1 > P1* initially,
while P2 = P2*. That is, initially sector 1 is protected. A trade reform, therefore, involves
lowering P1 towards the world price P1*.
What is the effect of lowering P1? By (2.2) and (2.3), a lower P1 must reduce
employment in sector 1: since W1 is fixed at Wmin, a fall in P1 increases the real wage in
that sector, inducing firms to hire less workers. It is easy to show that L2 must then
increase.3
Decreasing marginal productivity of labor implies that W2 must fall. But then we
conclude, from (2.4), that L1/(L1+U) must fall or, in other words, that the rate of
unemployment in sector 1 must increase. The impact on U, the number of unemployed
workers, is ambiguous, and depends in particular on the elasticities of labor demand (the
is in (2.2)).
3Suppose L2 falls. Then, by (2.2), the wage in sector 2 must increase. By (2.4), the probability of
employment in sector 1 must then increase, so U must fall. But then L1, L2, and U would all fall,
contradicting (2.5).
9
It should also be intuitively obvious that a lower P1 increases distortions in the
labor market: this is because real wages in the two sectors move away from each other, and
hence the initial gap in marginal productivity of labor between the two sectors becomes
larger.
Of course, a trade reform may have beneficial effects as well. To characterize
those, and to add more precision to the analysis, it may be useful to complete the
description of this economy, in particular the demand side. To do this, we will make
specific assumptions about workers and entrepreneurs.
Demand
The typical worker consumes a Cobb Douglas aggregate of goods 1 and 2:
C = C1 C2 / (1 )1
1 
If his final income is I, the worker will choose C1 and C2 to maximize C subject to
the budget constraint
P1 C1 + P2 C2 = I
The solution is straightforward: let P denote the minimum cost of a unit of the
consumption aggregate:
P = P1 P2
1
Then
PC = I
and the worker will spend a fraction and (1) of his income in goods 1 and 2,
respectively.
10
All workers receive a transfer TW from the government. In addition, each worker
has one unit of time, and so his income will include his wage earned if he is employed. For
simplicity, assume that there is no disutility from labor.
Finally, each worker may choose to locate in sector 1 or in sector 2. Locating in
sector 2 implies that he will earn the wage W2 for sure. In contrast, if he chooses to locate
in sector 1, he will earn the wage Wmin only with probability L1/(L1+U). Assuming risk
neutrality, the worker will choose the location that maximizes the expected value of
income. An equilibrium in which there are workers in both sectors then requires each
worker to be indifferent between locating in sector 1 and sector 2. This is easily seen now
to involve that the expected wage in both sectors be the same (equation (2.4)).
For simplicity, assume that the typical entrepreneur consumes the same aggregate
C of goods 1 and 2 as the typical worker. He is assumed to receive a transfer TK from the
government, and all profits from production.
Finally, for concreteness let us assume that the government levies a tariff P1 P1*
on imports of good 1, and no tariff on imports of good 2. The government has no other
sources of revenue and transfers tariff revenues to workers and entrepreneurs. Then fiscal
balance requires:
(2.6) (P1  P1 )(C1 Y1) = LTW + TK
* a
where Ci denotes total domestic consumption of good i.
a
The efficiency losses associated with the tariff are now evident. Since each
domestic agent spends a fraction of his income in good 1, the same must be the case for
the aggregate, so
P1C1 = (P1Y1 + P2Y2 + LTW + TK )
a
Likewise, P2 C2 is a fraction (1) of aggregate income. And of course aggregate
a
expenditure in the two goods must equal aggregate income:
11
P1C1 + P2C2 = P1Y1 + P2Y2 + LTW + TK
a a
Using (2.6) in the last equation to eliminate LTW +TK and recalling that there is no
tariff on good 2, we obtain that:
(2.7) P1 C1 + P2 C2 = P1 Y1 + P2Y2
* a * a * *
In other words, the value of domestic consumption must equal the value of
production, both at world prices. Note that this relation must hold for any value of the
tariff.
This perspective helps to clarify the relationship between the analysis of this model
and the standard analysis of tariffs. The tariff causes a distortion in consumption, since
domestic agents face the after tariff relative price P1/P2 instead of the world price P1 /P2
* *
when making consumption decisions. This causes them to choose a consumption bundle
such that the social indifference curve is not tangent to the national budget constraint line.
Also, as in the standard case, the tariff causes a distortion on the production side, since it
increases the relative price of good 1, so domestic production of good 1 is inefficiently
large.
In our model, however, there is an additional distortion in the labor market, due to
the minimum wage in sector 1. This distortion pushes production of good 1 down; indeed,
in the absence of the tariff, production and employment in sector 1 would be inefficiently
low. A tariff in sector 1 reduces the distortion by increasing the price of good 1 and
inducing firms to expand hiring in that sector.
It follows that a trade reform (a reduction of the tariff to good 1) will generally
reduce consumption distortions but (assuming the minimum wage Wmin remains in place)
may increase production distortions. In this sense, the success of trade reform may depend
on a complementary labor market reform.
12
Complementary Reforms and Productive Efficiency
As highlighted in the previous subsection, the value of production at world prices is
an adequate summary of productive efficiency in this model. Denoting that value by Z, it
follows that
Z = P1 Y1 + P2Y2 = P1 A1L1 + A2L2
* * *
1 2
where we have assumed that the price of good 2, P2 , equals one.
*
Assuming again that a tariff may be imposed on good 1s imports, the impact of a
change in the tariff on productive efficiency is then given by
(2.8)
dZ *
= P1 1A1L1
* dL1 dL2 P1 dL1 dL2
Wmin
11 + A22L2 21 = +W2
dP1 dP1 dP1 P1 dP1 dP1
The last equality follows from (2.2) and (2.3).
Equation (2.8) is useful to understand the impact of a marginal change in the tariff
on Z. To understand its implications, assume that there are no initial trade or labor market
distortions. No tariffs imply that P1 = P1* and the absence of a minimum wage implies
that W1 = W2 = W, say. Therefore, dZ/dP1 = W[d(L1+L2)/dP1]. But d(L1+L2)/dP1 = 0,
since labor is fully employed if there is no minimum wage in sector 1. So dZ/ dP1 = 0: the
marginal impact of a tariff on productive efficiency is nil. This result is in line with
conventional theory.
Now suppose that initially there is no minimum wage, so that again W1 = W2 = W
and labor is fully employed, but that there is a positive tariff on good 1. The latter means
that P1 > P1* and the impact of a marginal tariff change is
13
dZ P1 dL1
* *
+ =W 1 < 0
dP1 =W P1 dP1 dL2
dP1 dL1 P1
dP1 P1
where the last equality follows from full labor employment. So, as expected, in the absence
of a minimum wage, if there is a positive tariff on good 1, a marginal tariff reduction will
increase productive efficiency.
If initially both a tariff and a minimum wage exist, the analysis is more involved.
Some algebra shows that then
(2.9)
dZ *
=Wmin
dP1 dL1 P1
dP1 P1 {(12)(L / L2) +2}1
By (2.2) and (2.3), dL1/dP1 > 0. Hence the impact of a marginal change in the
tariff depends on the quantity in square brackets, which captures the opposing effects on
trade distortions and labor market distortions. The term {(12)(L/L2)+2}ą is less than
one, so the term in brackets can be positive or negative. In other words, when distortions
exist in both trade and labor markets, a marginal reduction in tariffs (keeping the labor
market distortion fixed) can increase or reduce productive efficiency.
Equation (2.9) has, in fact, a straightforward interpretation. The term P1*/ P1 is a
measure of the tariff on good 1 imports: the larger the tariff, the smaller P1*/P1. On the
other hand, the distortionary impact of a minimum wage is given by L2/L: the smaller this
ratio, the larger the RHS of (2.9). Intuitively, the smaller the size of sector 2, the greater the
discrepancy between the marginal product of labor in sectors 1 and 2, and the more likely
it is that an increase in P1 will help increasing efficiency, by inducing more hiring in sector
1 , where the marginal product of labor is higher.
The conclusion is that a trade reform (here, a tariff reduction) may or may not
improve productive efficiency if other policy distortions remain. The outcome depends,
14
intuitively, on the relative importance of trade distortions and the other policyinduced
distortions.
The obvious but significant corollary is that trade liberalization will not, in general,
have an unambiguous effect on productive efficiency. In this model, in fact, trade
liberalization will reduce productive efficiency if the labor market distortion is
pronounced, but it will increase efficiency if the labor market distortion is mild. This
indicates the need to include a term for the interaction between trade opening and labor
market distortions in assessing the empirical connection between trade opening and
growth.
The discussion in this section has focused on the links between trade liberalization
and labor market reforms, but it should be apparent that the essence of the analysis can be
extended to analyze the complementarity between trade opening and other reforms.
Keeping this in mind, we now turn to an empirical evaluation of the complementarity of
trade reform and other reforms.
3. EMPIRICAL ANALYSIS
The objective of the empirical section is to examine how the growth effect of
openness may depend on a variety of country characteristics, including some that can be
changed by policy. For this purpose, we work with pooled crosscountry and timeseries
data, focusing on comparative information from withincountry changes. We start with a
linear growth regression specification and then extend it to account for interaction terms
between an openness measure and proxies for various country characteristics. These are
educational investment, financial depth, macroeconomic price stability, public
infrastructure, governance, labormarket flexibility, ease of firm entry, and ease of firm
exit. We build on the paneldata growth regression literature that uses a GMM procedure
to control for endogeneity and unobserved countryspecific factors, as presented for
example in Dollar and Kraay (2004) and Levine, Loayza, Beck (2000). Further details on
the methodology are given below.
15
3.1. Sample and Regression Specification
Our sample consists of an unbalanced panel dataset that comprises 82 countries.
For each of them, the dataset includes at most 8 observations, consisting of non
overlapping 5year averages spanning the 19602000 period. The sample includes 22
developed countries and 60 developing ones. Among the latter, 18 are from SubSaharan
Africa, 12 from Asia, 9 from the Middle East and North Africa, and 21 from Latin
America and the Caribbean. Appendix 1 provides the full list of countries in the sample.
The basic regression equation to be estimated is the following
yi,t  yi,t = 0 yi,t + r1'CVi,t + 2OPi,t + µt +i +i,t
1 1 (3.1)
where the subscripts i and t represent country and time period, respectively; y is the log of
GDP per capita, CV is a set of control variables, and OP represents trade openness; µt and
i denote unobserved time and countryspecific effects, respectively; and is the
regression residual.
As is standard in the literature, the dependent variable is the average rate of real per
capita GDP growth (i.e., the log difference of GDP per capita normalized by the length of
the period). The regression equation is dynamic in the sense that it includes the initial
level of per capita GDP as an explanatory variable. Our measure of trade openness is the
(structureadjusted) ratio of real exports and imports to real GDP. We select the set of
control variables considering both their importance as growth determinants per se and their
potential for affecting the growth response of trade opening. The control set includes
variables that vary both across countries and over time, as well as variables that vary only
across countries (that is, assumed constant over time). Among the former, we have the
average rate of secondary school enrollment to account for human capital investment, the
average ratio of private credit to GDP as a measure of financial depth, the average inflation
rate to account for macroeconomic price stability, and the average number of main
telephone lines per capita as proxy for public infrastructure. Among the variables that vary
only acrosscountries, we have a governance index from International Country Risk Guide
(Political Risk Services), labormarket and firmexit flexibility indices from Doing
Business (the World Bank), and a firmentry flexibility index from Doing Business (the
16
World Bank) and the Index of Economic Freedom (the Heritage Foundation). Appendix 2
provides full definitions and sources of all variables used in the paper, and Appendix 3
presents basic descriptive statistics for the data used in the regressions.
We then extend the regression specification by allowing the growth effect of
openness to vary with the country characteristics represented by the control set. We do this
by interacting the openness measure with each of the control variables in turn. The
regression equation with an interaction term is the following,
yi,t  yi,t = 0 yi,t + r1'CVi,t + 2OPi,t + 3cvi,t *OPi,t + µt +i +i,t
1 1 (3.2)
where cv represents one of the control variables in particular. We interact openness with
the control variables one at a time in order to both simplify the interpretation of the results
and not to overextend the parameter requirements on the data.
The interpretation of the coefficients on the timevarying variables and on their
interaction term with openness is straightforward. However, the interpretation of
coefficients related to the variables that are constant per country requires some
explanation. In linear regression models, they are well captured by the countryspecific
effect and, in general, would not be incorporated into the regression specification. In our
case, however, we need to include them in the regression in order to analyze their
interaction with openness. The coefficients on the constant variables themselves cannot be
identified unless we have additional information on their relationship with the country
specific effect or are willing to make assumptions about it. Nevertheless, in order to
complete the information set, we include them in the regression as explanatory variables on
their own whenever their respective interaction with openness is analyzed. In order to
avoid confusion, we do not report the estimated coefficients on the constant variables
themselves but only the coefficients on their respective interaction terms.
3.2. Estimation Methodology
The growth regression presented above poses some challenges for estimation. The
first is the presence of unobserved period and countryspecific effects. While the inclusion
of periodspecific dummy variables can account for the time effects, the common methods
17
of dealing with countryspecific effects (that is, withingroup or difference estimators) are
inappropriate given the dynamic nature of the regression. The second challenge is that
most explanatory variables are likely to be jointly endogenous with economic growth, so
we need to control for the biases resulting from simultaneous or reverse causation. The
following paragraphs outline the econometric methodology we use to control for country
specific effects and joint endogeneity in a dynamic model of panel data.
We use the generalized method of moments (GMM) estimators developed for
dynamic models of panel data that were introduced by HoltzEakin, Newey, and Rosen
(1988), Arellano and Bond (1991), and Arellano and Bover (1995). These estimators are
based, first, on differencing regressions or instruments to control for unobserved effects
and, second, on using previous observations of explanatory and laggeddependent
variables as instruments (which are called internal instruments).
After accounting for timespecific effects, we can rewrite equations 3.1 or 3.2 as
follows:
yi,t = yi,t + r' Xi,t +i +i,t
1 (3.3)
To eliminate the countryspecific effect, we take first differences of equation 3.3:
yi,t  yi,t1= yi,t
( 1  yi,t2) + (
' Xi,t  Xi,t1 ) + ( ) (3.4)
i,t i,t1
Note that by differencing we also eliminate the information provided by the variables that
are constant over time.
The use of instruments is required to deal with the likely endogeneity of the
explanatory variables and the problem that, by construction, the new error term, i,t i,t1,
is correlated with the lagged dependent variable, yi,t1 yi,t2. The instruments take
advantage of the panel nature of the data set in that they consist of previous observations of
the explanatory and laggeddependent variables. Given that it relies on past values as
instruments, this method only allows current and future values of the explanatory variables
to be affected by the error term. Therefore, while relaxing the common assumption of
18
strict exogeneity, our instrumentalvariable method does not allow the X variables to be
fully endogenous.
Under the assumptions that the error term, , is not serially correlated and that the
explanatory variables, X, are weakly exogenous (that is, the explanatory variables are
assumed to be uncorrelated with future realizations of the error term), the GMM dynamic
panel estimator uses the following moment conditions:
E yi,t i,t  i,t
[ s ( 1)] = 0 for s 2;t = 3,...,T (3.5)
E Xi,t i,t  i,t
[ s ( 1)] = 0 for s 2;t = 3,...,T (3.6)
for s 2 and t = 3,..., T. Although in theory the number of potential moment conditions is
large and growing with the number of time periods, T, when the sample size in the cross
sectional dimension is limited, it is recommended to use a restricted set of moment
conditions. In our case, we work only with the first acceptable lag as an instrument; that
is, for the regression in differences we use only the twicelagged level of the corresponding
variable.
The GMM estimator based on the conditions in 3.5 and 3.6 is known as the
difference estimator. Notwithstanding its advantages with respect to simpler panel data
estimators, the difference estimator has important statistical shortcomings. AlonsoBorrego
and Arellano (1999) and Blundell and Bond (1997) show that when the explanatory
variables are persistent over time, lagged levels of these variables are weak instruments for
the regression equation in differences. Instrument weakness influences the asymptotic and
smallsample performance of the difference estimator toward inefficient and biased
coefficient estimates, respectively.4
To reduce the potential biases and imprecision associated with the usual difference
estimator, we use a new estimator that combines the regression in differences and the
regression in levels into one system (developed in Arellano and Bover, 1995, and Blundell
and Bond, 1997). The instruments for the regression in differences are the same as above.
4An additional problem with the simple difference estimator involves measurement error: differencing may
exacerbate the bias stemming from errors in variables by decreasing the signaltonoise ratio (see Griliches
and Hausman, 1986).
19
For the regression in levels, however, the instruments are the lagged differences of the
corresponding variables. These are appropriate instruments under the following additional
assumption: although the levels of the righthandside variables may be correlated with the
countryspecific effect in equation 3.3, the differences of these variables are not. This
assumption results from the following stationarity property,
E[yi,t+p i] = E[yi,t+q i] and
E[Xi,t + pi] = E[Xi,t i] for all p and q (3.7)
+q
for all p and q. The additional moment conditions for the second part of the system (the
regression in levels) are:5
E[ yi,t  yi,t
( 1 2)( ) (3.8)
i +i,t ]= 0
E[ Xi,t  Xi,t
( 1 2 )( ) (3.9)
i+i,t ] = 0
Note that in the levels regression, the variables that are constant over time are not
eliminated; however, there are no available instruments for them based on either their own
lagged changes (since they are constant) or the lagged changes of the timevarying
variables (because if these changes are uncorrelated with the unobserved countryspecific
effect, they are also likely to be uncorrelated with the observed constant variables).
We thus use the moment conditions presented in equations 3.5, 3.6, 3.8, and 3.9
and employ a GMM procedure to generate consistent and efficient estimates of the
parameters of interest and their asymptotic variancecovariance (Arellano and Bond 1991;
Arellano and Bover 1995). However, given the limited size of our sample, in order to
reduce the risk of overfitting bias, in the regression in differences we use only the first
acceptable lag as an instrument. These are given by the following formulas:
^ = (X'Z1Z'X)1 X'Z1Z' y
^ ^ (3.10)
5Given that lagged levels are used as instruments in the differences specification, only the most recent
difference is used as an instrument in the levels specification. Using other lagged differences would result in
redundant moment conditions (see Arellano and Bover, 1995).
20
AVAR(^) = (X 'Z1Z' X )1
^ (3.11)
where is the vector of parameters of interest (, ); y is the dependent variable stacked
first in differences and then in levels; X is the explanatoryvariable matrix including the
lagged dependent variable (yt1, X) stacked first in differences and then in levels; Z is the
matrix of instruments derived from the moment conditions; and is a consistent estimate
^
of the variancecovariance matrix of the moment conditions.6
The consistency of the GMM estimators depends on whether lagged values of the
explanatory variables are valid instruments in the growth regression. We address this issue
by considering two specification tests suggested by Arellano and Bond (1991) and
Arellano and Bover (1995). The first is a Sargan test of overidentifying restrictions, which
tests the overall validity of the instruments by analyzing the sample analog of the moment
conditions used in the estimation process. Failure to reject the null hypothesis gives
support to the model.7
The second test examines the null hypothesis that the error term, i,t, is not serially
correlated. As in the case of the Sargan test, the model specification is supported when the
null hypothesis is not rejected. In the system specification, we test whether the differenced
error term (that is, the residual of the regression in differences) is secondorder serially
correlated. Firstorder serial correlation of the differenced error term is expected even if the
original error term (in levels) is uncorrelated, unless the latter follows a random walk.
Secondorder serial correlation of the differenced residual indicates that the original error
term is serially correlated and follows a moving average process of at least order one. This
would reject the appropriateness of the proposed instruments (and would call for higher
order lags to be used as instruments).
6 Arellano and Bond (1991) suggest the following twostep procedure to obtain consistent and efficient
GMM estimates. First, assume that the residuals, i,t, are independent and homoskedastic both across
countries and over time; this assumption corresponds to a specific weighting matrix that is used to produce
firststep coefficient estimates. Second, construct a consistent estimate of the variancecovariance matrix of
the moment conditions with the residuals obtained in the first step, and then use this matrix to reestimate the
parameters of interest (that is, secondstep estimates). Asymptotically, the secondstep estimates are superior
to the firststep ones insofar as efficiency is concerned.
7 There are cases where the Sargan test statistic cannot be computed given the near singularity of variance
covariance of the moment conditions. This arises when the crosssectional dimension is small relative to the
number of instruments. In those cases, of which we have a couple in our econometric results, we have to rely
only on the residual autocorrelation test.
21
3.3. Results
Regression results are presented in Tables 1 and 2. Table 1 shows the results of the
basic regression with no interaction terms (column 1) and the results of the regressions
where openness is interacted with timevarying variables (columns 25). These variables
represent areas where economic reform has been most active; they are human capital
investment, financial depth, macroeconomic price instability, and public infrastructure,
respectively. Table 2 shows the regression results where openness is interacted with time
invariant variables. They represent institutional and regulatory areas where reform often
called of second generation has been most sluggish. They are indices of governance,
labor market flexibility, firmentry flexibility, and firmexit flexibility. We treat them as
constant per country because their underlying institutional characteristics vary little over
time and, partly reflecting this, there is quite limited data on their time dimension.8
The basic regression (Table 1, Col. 1) shows results consistent with the previous
empirical literature. Initial GDP per capita carries a significantly negative coefficient,
commonly interpreted as evidence of conditional convergence. The proxies of human
capital investment, financial depth, and public infrastructure have positive and significant
coefficients, denoting their beneficial impact on economic growth. Inflation, on the other
hand, carries a negative coefficient, indicating the negative consequence of
macroeconomic price instability. Trade openness is also a significant explanatory variable;
as in other studies that rely on the crosscountry variation of withincountry changes, trade
openness is found to have a positive impact on economic growth. Since in this basic
specification only linear effects are allowed, the estimated openness impact on growth is an
average effect; below we attempt to uncover what is behind this average. The period shifts
indicate that international conditions for growth have deteriorated over time, resulting in
considerably poorer conditions in the 1980s and 1990s than in the previous decades.
Finally, both the Sargan and serialcorrelation tests indicate that the null hypothesis of
correct specification cannot be rejected, lending support to our estimation results. This is
8 The ICRG governance index is available since the mid 1980s and shows some time variation. Given that
we are forced to assume that its value was the same in the 1960s and 1970s as in the mid 1980s, we take the
conservative assumption that its growth effect cannot be estimated separately from that of the unobserved
fixed effect, as is the case with the other institutional variables that are completely constant over time.
22
the case for the exercises presented below, and we mention it only here in order to avoid
redundancy.
Table 1 also shows the regressions results that consider interaction effects between
openness and timevarying variables (Cols. 25). An interesting pattern of reform
complementarity emerges: the coefficient on the interaction between the trade volume ratio
and, in turn, the secondary enrollment rate, the private domestic credit ratio, and the
number of phone lines per capita is positive and significant. This indicates that the growth
effect of an increase in openness depends positively on the progress made in each of these
areas. That is, more openness results in a larger increase in economic growth when the
investment in human capital is stronger, financial markets are deeper, and public
infrastructure is more readily available. The shared explanation for these results is related
to the competitiveness of domestic firms in international markets: when domestic firms
find a better educated labor force and less costly credit and communications, they are able
to compete with foreign firms and expand their markets effectively. The interaction
between trade volumes and inflation is not significant, possibly reflecting the fact that for
most inflation values, relative price distortions are not severe.
Table 2 shows the growth regression results when openness is interacted with the
proxies of institutional and regulatory reform. Interestingly, as in the results related to
timevarying variables, we observe a pattern of complementarity between openness and
other reforms: the estimated coefficients on the interaction between the trade volume ratio
and, in turn, the proxies for governance, labormarket flexibility, and firmentry flexibility
are positive and statistically significant. The beneficial impact of an increase in trade
openness on economic growth is larger when society has a more efficient, accountable, and
honest government and where the rule of law is more respected. Likewise, the positive
growth effect of trade opening is stronger when flexible labor markets make it easier for
domestic firms to transform and adjust to changing environments, particularly those in
highly competitive foreign markets. Our results also point out the importance of
unrestricted firm renewal in order for trade opening to have a positive growth impact,
particularly regarding the firmentry margin. The interaction term between openness and
firmexit flexibility is, however, not significant; whether this reflects dataquality problems
or a more substantial difference with the opposite margin of firm dynamics is unclear.
23
The preceding discussion focuses on the interaction terms; however, in order to
ascertain whether the total impact of a change in openness leads to higher or lower growth,
we need to consider the coefficients on both the interaction term and the openness variable
itself. Since the total impact depends on the values of the variables with which openness is
interacted, it is not really informative to provide a single summary measure of the effect.
Instead, it may be best to show how the growth effect of a change in openness varies for
different levels of the other reform variables. We do so in Figure 2. Specifically, this
figure presents the total effect on economic growth of a onestandarddeviation change in
the openness measure for each value that a given complementary reform can take in the
sample. Since only linear interaction effects are considered, the growth effect of openness
can be represented as a linear function of each complementary reform. In addition to total
growth effects (based on the coefficient point estimates), the figure shows the
corresponding 90% confidence bands (constructed with the estimated coefficient standard
errors).9 Figure 2 has six panels, each corresponding to a reform variable whose
interaction with openness is statistically significant. For timevarying variables, the range
of values corresponding to the latest period (19962000) is found towards the higher values
of the full (allperiods) range; since for current policy analysis the latest values are the
most relevant, we highlight their range in the corresponding panel.
For all reform variables except the governance index, the total growth impact of
openness changes from negative to positive as progress occurs. Therefore, in principle, an
increase in openness could bring a reduction in economic growth if a given complementary
area is not sufficiently advanced. In practice, given the current state of reform progress
around the world, this concern is presently relevant for half the complementary areas under
consideration. For educational enrollment, financial development, and governance, our
9From our regression model, the growth effect of openness is given by,
Growth = (OPEN + INT REF) Openness
where OPEN and INT are, respectively, the estimated regression coefficients on openness and on the
interaction between openness and a given complementary reform variable (REF). Note that Openness is an
arbitrary constant (set to equal one sample standard deviation of the openness measure) and REF follows a
fixed set of values (and can thus be treated as a constant at any given point along its sample range). Then,
the confidence intervals can be constructed from the following expression for the variance of the growth
effect,
Var[Growth] = {Var(OPEN)+Var(INT) REF + Cov (OPEN , INT) REF} {Openness}2
2
where the variances and covariances of the estimated coefficients are obtained from our panel estimation
method.
24
results indicate that they would not cause growth to decline with increased openness given
that their current values in most countries exceed the corresponding threshold. However,
regarding infrastructure, labor market flexibility, and firm entry flexibility, there are many
countries that currently stand to lose from opening their markets. Focusing only on the
reform indicators used in the paper, we can derive the implication that the most urgent
reforms in order to make trade good for growth are related to infrastructure, labor markets,
and firm renewal. This is not to say, however, that countries will not benefit more from
trade openness if they improve their educational attainment, financial depth, and overall
governance.
So far we have considered only linear interactions between openness and other
variables one at a time. A richer specification would allow for both nonlinear interactions
and interactions with multiple variables at the same time. The problem with richer
specifications, however, is that the possibilities are almost endless. Our limited sample,
demanding econometric methodology, and the multicollinearity among growth
determinants prevent us from exploring a full set of interactions. However, some progress
can be made by expanding our specification to include the interaction between openness
and initial income to each of the regression specifications studied above and examining the
sign and significance of the coefficient on this interaction. This exercise links our results
to the literature discussed in the introduction, which finds that the growth effect of
openness depends on the countries' level of income. That literature argues that income
represents overall development and is thus strongly related to the social and economic
conditions that improve openness' beneficial impact. We find confirmation for this claim
in the fact that, in our sample, initial income has a 90% correlation with the first principal
component of our six complementary reforms, meaning that income represents well what
is common among these development variables. Therefore, we can analyze whether it is
the full package of reforms what makes openness better for growth by examining if, in the
expanded specification, the coefficient on the opennessinitial income interaction is
positive and significant in most cases. It is also of interest to check whether, after the
opennessinitial income interaction is included, the interaction of openness with certain
individual reforms remain significant. If so, the effect of those reforms on the openness
25
growth relation is independent to that of the other reforms as a whole. In that sense, they
would be "more complementary" to openness than the rest.
Table 3 presents the main results of the expanded regression specification. In the
benchmark (column 1), only the opennessinitial income interaction is considered; it
carries a significantly positive coefficient, as expected and predicted by previous literature.
In the remaining cases, the interactions between openness and, respectively, educational
investment, financial depth, public infrastructure, governance, labormarket flexibility, and
firmentry flexibility are added one at a time.
In five out of the six cases the opennessincome interaction remains significantly
positive, denoting that in the majority of cases the combination of all complementary
reform variables (represented by income) is a key determinant on the growth impact of
openness.10 Overall, therefore, our results underscore the benefits of a comprehensive path
to reform.
The inclusion of the opennessinitial income interaction term makes the
coefficients of the interaction between openness and human capital investment, financial
depth, public infrastructure, and governance lose their sign or become statistically
insignificant. As we have discussed, we interpret this result as suggesting that these
complementary areas must be addressed as a whole if they are to affect the openness
growth link. In contrast, the coefficients on the interaction of openness and, respectively,
labor market flexibility and firm entry flexibility remain significant at the five percent
level. Reforms of the labor market and firm entry conditions appear, therefore, to have a
significant effect on the impact of openness on growth, independently of other reforms.
4. CONCLUDING REMARKS
As discussed in the introduction, previous empirical evidence on the impact of
trade openness on economic growth has failed to reveal undisputed beneficial growth
effects from trade liberalization. Both the theory and the evidence reviewed in this paper,
however, indicate that such a failure should not have come as a surprise. It is not too hard
to find theoretical situations in which the removal of barriers to trade needs to be
10The exception occurs when the interaction between openness and firmentry flexibility is also present. In
that case, the coefficient of the opennessincome interaction term becomes insignificant, although its sign
remains positive.
26
accompanied by complementary reforms in nontrade areas if it is to improve productive
efficiency and growth. And our empirical work finds that such situations may not be
uncommon in practice.
One policy implication of our analysis is that the advisability of trade liberalization
may depend on the existence and degree of distortions in nontrade institutions, as well as
on the feasibility of removing those distortions. This underscores the need to reject a "one
size fits all" approach to trade opening in favor of packages that are tailored to the specific
circumstances of each country.
This being said, we believe that our findings provide fresh support to the view that
trade liberalization tends to enhance growth. Our empirical work does confirm that trade
opening results in faster growth on average, that is, when the interaction effects are omitted
from our regressions. And if the interaction terms are included, trade liberalization turns
out to still raise growth, except for countries in which complementary areas are strongly
distorted.
In addition, and conversely, our findings indicate that "second generation" reforms
have not only direct benefits but also indirect ones, in that they allow a country to take
fuller advantage of trade opening. This is a significant argument in the ongoing debate
about the gains from more comprehensive reform in developing countries.
27
References
Acemoglu, D. and F. Zilibotti (2001). "Productivity Differences." Quarterly Journal of
Economics 116(2): 563606.
Alfaro, L., A. Chanda, S. KalemliOzcan, and S. Sayek (2003). "FDI Spillovers, Financial
Markets and Economic Development." IMF Working Papers 03/186.
AlonsoBorrego, C. and M. Arellano (1999). "Symmetrically Normalized Instrumental
Variable Estimation Using Panel Data." Journal of Business and Economic Statistics
17(1): 3649.
Arellano, M., and S. Bond (1991). "Some Tests of Specification for Panel Data:
Montecarlo Evidence and an Application to Employment Equations." Review of Economic
Studies 58(2): 277297.
Arellano, M., and O. Bover (1995). "Another Look at the InstrumentalVariable
Estimation of ErrorComponents Models." Journal of Econometrics 68(1): 2952.
Banerjee, A. and A. Newman (2004). "Notes for Credit, Growth, and Trade Policy."
Mimeo. MIT.
Beck, T., A. DemirgucKunt, and R. Levine (2000). "A New Database on Financial
Development and Structure." The World Bank Economic Review 14(3): 597605.
Blundell, R. and S. Bond (1997). "Initial Conditions and Moment Restrictions in Dynamic
Panel Data Models". University College London Discussion Papers in Economics: 97/07,
July.
Bolaky, B. and C. Freund (2004). "Trade, Regulations, and Growth." World Bank Policy
Research Paper Series 3255.
Borensztein, E., J. De Gregorio, JW. Lee (1995). "How Does Foreign Direct Investment
Affect Economic Growth?" Journal of International Economics 45, 1: 11535.
Botero, J., S. Djankov, R. La Porta, F. LopezdeSilanes, and A. Shleifer (2004). "The
Regulation of Labor." Quarterly Journal of Economics 119: 13391382.
Brecher, R. (1974). "Optimal Commercial Policy for a Minimumwage Economy."
Journal of International Economics 4: 139149.
Calderon, C., N. Loayza, and K. SchmidtHebbel (2004). "External Conditions and
Growth Performance." Central Bank of Chile Working Paper N. 292.
Canning, D. (1998). "A Database of World Stocks of Infrastructure." The World Bank
Economic Review 12: 529547.
28
Djankov, S., R. La Porta, F. LopezdeSilanes and A. Shleifer (2002). "The Regulation of
Entry." Quarterly Journal of Economics 117: 137.
Djankov, S., O. Hart, T. Nenova, and A. Shleifer (2003). "Efficiency in Bankruptcy."
Working Paper, Department of Economics, Harvard University.
Dollar, D. (1992). "Outwardoriented Developing Economies Really D Grow More
Rapidly: Evidence from 95 LDCs, 197685." Economic Development and Cultural
Change, 52344.
Dollar, D. and A. Kraay (2004). "Trade, Growth, and Poverty." Economic Journal
114(493): F2249.
Easterly, W. and M. Sewadeh (2002). "Global Development Network Growth Database."
Washington: World Bank.
Edwards, S. (1998). "Openness, Productivity, and Growth: What Do We Really Know?"
Economic Journal 108 (447): 38398.
Eicher, T. (1993). "Trade and Converging Growth Rates in a Model with Endogenous
Human Capital and Technological Change: Dynamic Gains from Trade Reconsidered."
Unpublished paper. Columbia University.
Griliches, Z. and J. Hausman (1986). "Errors in Variables in Panel Data." Journal of
Econometrics 31(1): 93118.
Grossman, G. and E. Helpman (1991). Innovation and Growth in the Global Economy.
Cambridge, Mass.: MIT Press.
Harrison, A. (1996). "Openness and Growth: A TimeSeries, CrossCountry Analysis for
Developing Countries." Journal of Development Economics 48(2): 41947.
HoltzEakin, D. W. Newey, and H. S. Rosen (1998). "Estimating Vector Autoregressions
with Panel Data." Econometrica 56: 13711395
Lee, J. (1993). "International Trade, Distortions, and LongRun Economic Growth."
International Monetary Fund Staff Papers 40(2): 299328.
Levine, R., N. Loayza, and T. Beck (2000). "Financial Intermediation and Growth:
Causality and Causes." Journal of Monetary Economics 46, No.1: 3177.
Lipsey, R. and K. Lancaster (1956). "General Theory of Second Best." Review of
Economic Studies 24(1): 1132.
29
Loayza, N., P. Fajnzylber, and C. Calderon (2005). "Economic Growth in Latin America:
Stylized Facts, Explanations, Forecasts." in World Bank Latin American and Caribbean
Studies, Washington, D.C.: The World Bank.
Matsuyama, K. (1992). "Agricultural Productivity, Comparative Advantage, and Economic
Growth." Journal of Economic Theory 58(2): 317334.
O'Driscoll, G., E. Feulner, and M. A. O'Grady (2003). 2003 Index of Economic Freedom.
The Heritage Foundation and the Wall Street Journal.
Political Risk Services (2003). International Country Risk Guide. East Syracuse, N.Y.:
Political Risk Services.
Rodrik, D. and F. Rodríguez (2001). "Trade Policy and Economic Growth: A Skeptics
Guide to the CrossNational Evidence." in B. Bernanke and K. Rogoff (editors), NBER
Macroeconomics Annual 2000, Vol. 15, MIT Press: 261325.
Sachs, J. D. and A. Warner (1995). "Economic Reform and the Process of Global
Integration." Brooking Papers on Economic Activity, 1118.
Sachs, J. and A. Warner (1999). "The Big Push, Natural Resource Booms and Growth."
Journal of Development Economics 59(1): 4376.
Summers, R. and A. Heston (1991). "The Penn World Table (Mark 5): An Expanded Set
of International Comparisons, 19501988." Quarterly Journal of Economics 106(2): 327
68.
Wacziarg, R. and K. H. Welch (2003). "Trade Liberalization and Growth: New Evidence."
NBER Working Paper 10152, Cambridge, MA.
World Bank (2003). Doing Business. Washington, D.C.: The World Bank.
World Bank (2003). World Development Indicators. Washington, D.C.: The World Bank.
Young, A. (1991). "Learning by Doing and the Dynamic Effects of International Trade."
Quarterly Journal of Economics 106(2): 369405.
30
Table 1
Economic Growth and the Interaction between Openness and Other Economic Reforms
Crosscountry panel data consisting of nonoverlapping 5year averages spanning 19602000
Dependent variable: Growth rate of real GDP per capita
Estimation Method: GMMIV System Estimator (Arellano and Bover, 1995; Blundell and Bond, 1998)
Interaction of Openness with:
[1] [2] [3] [4] [5]
Benchmark: No Human Capital Public
Financial Depth Inflation
Interactions Investment Infrastructure
Control Variables:
Initial GDP per capita 3.1713 ** 3.2036 ** 3.2627 ** 3.2059 ** 3.3552 **
(in logs) 0.18 0.21 0.17 0.18 0.23
Human capital investment 1.1621 ** 0.8610 ** 1.2105 ** 1.1402 ** 1.2594 **
(secondary enrollment, in logs) 0.15 0.42 0.16 0.16 0.17
Financial depth 1.0272 ** 0.9421 ** 0.0262 1.0071 ** 0.9234 **
(private domestic credit/GDP, in logs) 0.11 0.09 0.21 0.11 0.07
Inflation 0.4580 ** 0.4350 ** 0.4895 ** 0.3243 0.4364 **
(deviation of inflation rate from 3%, in logs) 0.08 0.07 0.07 0.21 0.07
Public infrastructure 1.5764 ** 1.5904 ** 1.6053 ** 1.6050 ** 0.6423 **
(main telephone lines per capita, in logs) 0.13 0.16 0.14 0.14 0.19
Openness:
Trade Openness (TO) 1.1959 ** 2.0421 ** 0.2553 1.3497 ** 3.2821 **
(structureadjusted trade volume/GDP, in logs) 0.16 0.59 0.28 0.28 0.48
Interactions:
TO * Human capital investment 1.0031 **
0.18
TO * Financial depth 0.4629 **
0.08
TO * Inflation 0.0725
0.10
TO * Public infrastructure 0.4970 **
0.09
Period Shifts:
Intercept (base period: 196670) 26.6266 ** 33.8398 ** 30.5385 ** 26.8523 ** 24.3839 **
 7176 Period shift 0.2987 * 0.2371 0.2168 0.2698 0.2973 **
 7680 Period shift 1.1300 ** 1.1488 ** 1.0385 ** 1.1052 ** 1.1850 **
 8185 Period shift 3.3327 ** 3.3847 ** 3.2966 ** 3.3011 ** 3.4343 **
 8690 Period shift 2.9064 ** 3.0726 ** 2.9450 ** 2.8904 ** 3.1684 **
 9195 Period shift 3.6060 ** 3.8088 ** 3.6621 ** 3.6020 ** 3.9486 **
 9600 Period shift 4.3282 ** 4.6922 ** 4.4665 ** 4.3250 ** 4.8331 **
Countries / Observations 82/544 82/544 82/544 82/544 82/544
Specification Tests (pvalues)
 Sargan Test 0.42 0.42 0.37 0.39 0.47
 2nd. Order Correlation 0.15 0.14 0.15 0.14 0.14
Numbers below coefficients are the corresponding robust standard errors. * (**) denotes statistical significance at the 10 (5) percent level.
Source: Authors' calculations
31
Table 2
Economic Growth and the Interaction between Openness and Institutional/Regulatory Reforms1
Crosscountry panel data consisting of nonoverlapping 5year averages spanning 19602000
Dependent variable: Growth rate of real GDP per capita
Estimation Method: GMMIV system estimator for dynamic models with unobserved specific effects and endogenous regressors
Interaction of Openness with:
[1] [2] [3] [4]
Labor market Firm entry
Governance Firm exit flexibility
flexibility flexibility
Control Variables:
Initial GDP per capita 3.4019 ** 4.0229 ** 3.0202 ** 3.2063 **
(in logs) 0.33 0.24 0.21 0.18
Human capital investment 1.2845 ** 1.5146 ** 1.7603 ** 1.2424 **
(secondary enrollment, in logs) 0.16 0.16 0.16 0.11
Financial depth 0.9632 ** 1.2870 ** 0.9063 ** 1.3196 **
(private domestic credit/GDP, in logs) 0.12 0.12 0.12 0.12
Inflation 0.3830 ** 0.3513 ** 0.5266 ** 0.2848 **
(deviation of inflation rate from 3%, in logs) 0.08 0.08 0.08 0.07
Public infrastructure 1.5912 ** 1.6379 ** 1.4037 ** 1.0532 **
(main telephone lines per capita, in logs) 0.17 0.12 0.14 0.13
Openness:
Trade Openness (TO) 0.0802 3.7359 ** 3.5333 ** 1.6581 **
(structureadjusted trade volume/GDP, in logs) 0.33 0.64 0.69 0.27
Interactions:
TO * Governance 2.9617 **
(governance: index from ICRG, 0  1) 0.87
TO * Labor market flexibility 8.9986 **
(labor: index from DB, 0.210.80) 1.36
TO * Firm entry flexibility 7.4593 **
(entry: index from DB, 0.25  0.94) 1.31
TO * Firm exit flexibility 0.8598
(exit: index from DB, 0  1) 0.73
Period Shifts:
Intercept (base period: 196670) 30.1810 ** 39.9023 ** 34.5819 ** 20.0764 **
 7176 Period shift 0.2943 * 0.6062 ** 0.3485 * 0.6757 **
 7680 Period shift 1.1737 ** 1.5945 ** 1.2628 ** 1.5267 **
 8185 Period shift 3.4484 ** 3.7077 ** 3.6949 ** 3.5881 **
 8690 Period shift 3.1087 ** 3.3740 ** 3.3734 ** 2.9243 **
 9195 Period shift 3.9498 ** 3.9600 ** 4.0722 ** 3.5820 **
 9600 Period shift 4.6800 ** 4.4676 ** 4.8611 ** 3.8035 **
Countries / Observations 82/544 79/523 82/544 78/518
Specification Tests (pvalues)
 Sargan Test 0.37 n.a. 0.38 n.a.
 2nd. Order Correlation 0.12 0.28 0.13 0.25
Numbers below coefficients are the corresponding robust standard errors. * (**) denotes statistical significance at the 10 (5) percent level.
1 Our measures of institutional and regulatory reform do not vary, or vary little, over time. Their direct impact on growth cannot be separated
from that of the countryspecific effect; however, we include them as an additional control.
Source: Authors' calculations
32
Table 3
Economic Growth and the Interaction between Openness, Reforms, and Income1
Crosscountry panel data consisting of nonoverlapping 5year averages spanning 19602000
Dependent variable: Growth rate of real GDP per capita
Estimation Method: GMMIV System Estimator (Arellano and Bover, 1995; Blundell and Bond, 1998)
Interaction of Openness with:
[1] [2] [3] [4]
Human Capital Public
Benchmark Financial Depth
Investment Infrastructure
TO * Initial GDP per Capita 1.0067 ** 0.9237 ** 1.2174 ** 1.2644 **
0.21 0.27 0.30 0.41
TO * Human capital investment 0.1200
0.30
TO * Financial depth 0.2711 *
0.15
TO * Public infrastructure 0.1550
0.17
Countries / Observations 82/544 82/544 82/544 82/544
Interaction of Openness with:
[5] [6] [7]
Labor market Firm entry
Governance
flexibility flexibility
TO * Initial GDP per Capita 0.9340 ** 0.3532 ** 0.2452
0.36 0.16 0.35
TO * Governance 0.7291
(governance: index from ICRG, 0  1) 1.36
TO * Labor market flexibility 9.5158 **
(labor: index from DB, 0.210.80) 1.39
TO * Firm entry flexibility 6.3057 **
(entry: index from DB, 0.25  0.94) 2.04
Countries / Observations 82/544 79/523 82/544
Numbers below coefficients are the corresponding robust standard errors. * (**) denotes statistical significance at the 10 (5) percent
level.
1 Each column in the table is part of a larger regression which includes the same control variables as tables 1 and 2 but that
we do not present here for the sake of succinctness.
Source: Authors' calculations
33
1 1
34
8
0. up)o 8.0
gr
saera puo omtt p)
0.6 gr bo( 6.0
ility ou
omtt ar gr
exib
ourf erut nei Fl om
bo L
4 upo t
ni et
sre ucrtsar 0. 4.0
rka gr bot(r
ss Mr ss
mrofer nne omtt
nfI nnee nea
ope bo op bo Li
0.2 ni .20 ni
mot n:oireitr e e
anghc p) La:n hangc
C 0 ourg 0
8 6 4 2 0 2 4 6 8 iterior
bot 8 6 4 2 0 2 4 6 8
ngi C
nda nk upo ( opt
ing
Ra 3 gr
2 45 ear nk 1 71 up)o
optrof .2
980 gr
0. 57 0 55
037
0. 0. pot nLi Ra 0.
+ 0. +
6x
s0891 7x 3x
634.4 203x5.1 925
0. ouprg pot(r
0.4 988.0 = .4 =
y 0 y nea
= = opt Li
y y
and
s0991 6
0. 6.
htworg ni egnahc 0
htworg ni egnahc
eht
nee
wt 1
1
be
s
8.0
nnes up)o 8.0
p)
ope gr
ni ourg
s .60 omt
oup .60
gr
ngeahc bot(r upo tot mo
980s1 mott ce gr b(
ation 4.0 nea an omt
uss het bo Li 4.0 earn
ni bot Li
ss
ver el
P abi ucdE:n rnevo ss
nne
.20 ope G:n nne
ope
in .20 ni
GD var ge e
a m iterior
pitac orferfo C hanc iterior anghc
p) C
ing 0 g p)
8 6 4 2 0 2 4 6 8
el nk ourg 0
in 8 6 4 2 0 2 4 6 8 ou
per nk
Ra gr
of Ra
setar centir opt(
99
92 .20 5 upo
pe 674 gr .20 upo ( pot
0.
ht 0. 1048
x84 + pot earn 0516 gr
Li 0. earn
h 66 +
83 97x 1. opt Li
3. 06
owt g:n =
y 4.0 1. 7x
= 08x
y 4.
0
gr
ni nkiar 78 684
5. 0.
= =
y
1 s orf y
e .60
urg ngeah oni .60
htworg ni egnahc
htworg ni egnahc
ertir
Fi C C
Figure 2
Growth Effect of Trade Openness as a Function of Complementary Reforms*
A. Educational Enrollment B. Financial Depth
2.5 2
2
1.5
)
1.5
%)( %(
at ta
1 1
apic
0.5
per piacrep
P P 0.5
0 D
GD G
of 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 of
ht 0.5 th 0
ow ow 1.1 0.1 0.9 1.9 2.9 3.9 4.9
gr 1 gr
0.5
1.5
2 1
log of secondary school enrollments log of domestic credit to private sector/GDP
C. Telecommunications Infrastructure D. Governance
2.5 2.4
2.1
2
1.8
1.5
(%) (%)
a a 1.5
apitc 1 apitc 1.2
per per
P 0.5 P 0.9
D D
G G
0.6
of 0 of
ht 9 8 7 6 5 4 3 2 1 ht
0.3
grow 0.5 grow
0
0 0.2 0.4 0.6 0.8 1
1
0.3
1.5 0.6
log of per capita phones ICRG governance index
E: Labor Market Flexibility F: Firm Entry Flexibility
3 3
2 2
) )
%( (%
at
piacr 1 atipac 1
pe per
P P
D
G
0 0
of GDfo
ht 0.2 0.3 0.4 0.5 0.6 0.7 ht 0.25 0.35 0.45 0.55 0.65 0.75 0.85
wo
gr grow
1 1
2 2
labor market flexibility firm entry flexibility
*Notes:
1. The solid lines show the effect of a one standard deviation increase in the log of trade volume/GDP on the growth rate of GDP per capita.
2. The x axis represents the range of the reform area in the full sample. The thicker line on the x axis (when applicable) represents the range of the reform area in the period 19962000.
3. Dotted lines are 90% confidence bands.
35
Appendix 1: Sample of countries
I. Industrial Econom ies (22 countries)
Australia Germ any Norway
Austria Greece Portugal
Belgium Iceland Spain
Canada Ireland Sweden
Denm ark Italy Switzerland
Finland Japan United Kingdom
France Netherlands United States
New Zealand
II. Latin Am erica and the Caribbean (21 countries)
Argentina Ecuador Nicaragua
Bolivia El Salvador Panam a
Brazil Guatem ala Paraguay
Chile Haiti Peru
Colom bia Honduras Trinidad and Tobago
Costa Rica Jam aica Uruguay
Dom inican Republic Mexico Venezuela, RB
III. Asia (12 countries)
Bangladesh Korea, Rep. Philippines
China Malaysia Singapore
India Pakistan Sri Lanka
Indonesia Papua New Guinea Thailand
IV. Middle East and North Africa (9 countries)
Algeria Israel Syria, Arab Rep.
Egypt, Arab Rep. Jordan Tunisia
Iran, Islam ic Rep. Morocco Turkey
V. SubSaharan Africa (18 countries)
Burkina Faso Ghana Senegal
Botswana Kenya Sierra Leone
Cote d'Ivoire Madagascar South Africa
Congo, Rep. Malawi Togo
Congo, Dem ocratic Rep. Niger Zam bia
Gam bia, The Nigeria Zim babwe
36
Appendix 2: Definitions and Sources of Variables Used in Regression Analysis
Variable Definition and Construction Source
GDP per capita growth Log difference of real GDP per capita. Authors' construction using Summers and
Heston (1991) and The World Bank (2003).
Initial GDP per capita Initial value of ratio of total GDP to total population. GDP is Authors' construction using Summers and
in 1985 PPPadjusted US$. Heston (1991) and The World Bank (2003).
Education Ratio of total secondary enrollment, regardless of age, to the Easterly and Sewadeh (2002) and The World
population of the age group that officially corresponds to that Bank (2003).
level of education.
Public Infrastructure Telephone mainlines are telephone lines connecting a Canning (1998), International
customer's equipment to the public switched telephone Telecommunications Union.
network. Data are presented in per capita terms.
Governance First principal component of four indicators (prevalence of International Country Risk Guide (ICRG),
law and order, quality of bureaucracy, absence of corruption, Political Risk Services, 2003.
and accountability of public officials). www.icrgonline.com
Financial Depth Ratio of domestic credit claims on private sector to GDP Author's calculations using data from IFS, the
publications of the Central Bank and PWD.
The method of calculations is based on Beck,
DemirgucKunt and Levine (2000).
Trade Openness Residual of a regression of the log of the ratio of exports and Easterly and Sewadeh (2002) and The World
imports (in 1995 US$) to GDP (in 1995 US$), on the logs of Bank (2003).
area and population, and dummies for oil exporting and for
landlocked countries.
Inflation rate Deviation of annual % change in CPI from 3% Author's calculations with data from IFS.
Labor Market Flexibility Weighted average of three indicators: flexibility of hiring, Doing Business, The World Bank Group See
conditions of employement and flexibility of firing. The Botero, Djankov, La Porta, LopezdeSilanes
original index from Botero et al. has been rescaled to range and Shleifer, "The Regulation of Labor",
between 0 and 1 and in order for higher values to indicate Quarterly Journal of Economics, 119, 1339
more flexible labor markets. 1382, Nov. 2004.
The difficulty of hiring component measures i) whether term http://rru.worldbank.org/doingbusiness/explor
contracts can only be used for temporary taskes; ii) the etopics/hiringfiringwork
maximum duration of term contracts; and iii) the ratio of the
mandated minimum wage to the average valueadded per
working population.
The rigidity of hours component measures i) whether night
work is restricted; ii) whether weekend work is allowed; iii)
whether the workweek consists of fiveandahalf days or
more; iv) whether the workday can extend to 12 hours or
more; v) whether the annual paid vacation days are 21 days
or less.
The difficutly of firing component measures i) whether
redundancy is not grounds for dismissal; ii) whether the
employer needs to notify the labor union or the labor ministry
for firing one redundant worker; iii) whether the employer
needs to notify the labor union or labor ministry for group
dismissals iv) whether the employer needs approval from the
labor union or labor ministry for firing one redundant worker
v) whether the employer needs approval from the labor union
or the labor ministry for group dismissals; vi) whether the
law mandates training or replacement prior to dismissal vii)
whether the law mandates training or replacement prior to
dismissal vii) whether priority rules apply for dismissals; viii)
whether priority rules apply for reemployment.
37
Appendix 2 (continued): Definitions and Sources of Variables Used in Regression Analysis
Firm Entry Flexibility Composed of four indicators:
1) Entry procedures: The number of different procedures that Doing Business, The World Bank Group See
a startup has to comply with in order to obtain a legal status, Djankov, La Porta, LopezdeSilanes and
i.e. to start operating as a legal entity. The data cover (1) Shleifer, "The Regulation of Entry", Quarterly
procedures that are always required; (2) procedures that are
Journal of Economics, 117, 137, Feb. 2002.
generally required but that can be avoided in exceptional
http://rru.worldbank.org/doingbusiness
cases or for exceptional types of businesses.
2) Entry days: The average duration estimated necessary to
complete a procedure. The fastest procedure (independent of
cost) is chosen. It is assumed that the entrepreneur
completes the procedure in the most efficient way, ignoring
the time that the entrepreneur spends in information gathering.
3) Entry costs: Costs associated with startingup a business,
based on the texts of the Company Law, the Commercial
Code, or specific regulations. If there are conflicting sources
and the laws are not completely clear, the most authoritative
source is used. If the sources have the same rank the source
indicating the most costly procedure is used. In the absence
of express legal fee schedules, a governmental officer's
estimate is taken as an official source. If several sources
have different estimates, the median reported value is used.
In the absence of government officer's estimates, estimates of
incorporation lawyers are used instead. If these differ, the
median reported value is computed. In all cases, the cost
estimate excludes bribes.
4) Entry regulations: i) Very Low: existing regulations The Index of Economic Freedom, Heritage
straightforward and applied uniformly to all businesses; Foundation Based on: Economist Intelligence
regulations not much of a burden for business; corruption Unit, Country Commerce and Country Report,
nearly nonexistent. ii) Low: simple licensing procedures;
2001 and 2002, U.S. Department of State,
existing regulations relatively straightforward and applied
Country Commercial Guide 24 and Country
uniformly most of the time, but burdensome in some
instances; corruption possible but rare iii) Moderate: Reports on Economic Policy and Trade
complicated licensing procedure; regulations impose Practices. See O'Driscoll, G., E. Feulner, and
substantial burden on business; existing regulations may be M. A. O'Grady (2003).
applied haphazardly and in some instances are not even
published by the government; corruption may be present and
poses minor burden on businesses iv) High: governmentset
production quotas and some state planning; major barriers to
opening a business; complicated licensing process; very high
fees; bribes sometimes necessary; corruption present and
burdensome; regulations impose a great burden on business
v) Very High: Government impedes the creation of new
businesses; corruption rampant; regulations applied randomly
Firm Exit Flexibility Composed of three indicators: 1) A measure that documents Doing Business, The World Bank
the success in reaching the three goals of insolvency, as See Djankov, Simeon, Oliver Hart, Tatiana
stated in Hart (1999). It is calculated as the simple average Nenova, and Andrei Shleifer, "Efficiency in
of the cost of insolvency (rescaled from 0 to 100, where Bankruptcy", working paper, Department of
higher scores indicate less cost), time of insolvency (rescaled Economics, Harvard University, July 2003.
from 0 to 100, where higher scores indicate less time), the
observance of absolute priority of claims, and the efficient
outcome achieved. A score 100 on the index means perfect
efficiency. 2) The cost of the entire bankruptcy process,
including court costs, insolvency practitioners' costs, the cost
of independent assessors, lawyers, accountants, etc. In all
cases, the cost estimate excludes bribes.
38
Appendix 2 (continued): Definitions and Sources of Variables Used in Regression Analysis
The cost figures are averages of the estimates in a multiple
choice question, where the respondents choose among the
following options: 02 percent, 35 percent, 610 percent, 11
25 percent, 2650 percent, and more than 50 percent of the
insolvency estate value. 3) The degree to which the court
drives insolvency proceedings. It is an average of three
indicators: whether the court appoints and replaces the
insolvency administrator with no restrictions imposed by law,
whether the reports of the administrator are accessible only to
the court and not creditors, and whether the court decides on
the adoption of the rehabilitation plan. The index is scaled
from 0 to 100, where higher values indicate more court
involvement in the insolvency process.
Periodspecific Shifts Time dummy variables. Authors' construction.
39
ity
m it
Fir ex bilix 00.1
fle
40
rm yrt lity
ibi 00.1 .610
Fi en exfl
0 2 0
rketa yt 1.0 0.5 .40
ili
mr ibx
bo fle
La
ec .001 43.0 67.0 .630
an
rn
ve
Go
ssen 00.1 5 4 9 2
.00 0.2 0.0 .00
aderT enp
O
0 3
e ) .1 78 25 65 62
in on erp .01 0. 0. 0. 0.
ph
Ma es
le taipac gs 0
lo
te lin ni(

(in
n tear. 32
)] 00.1 13.0 0. 14.0 62.0 83.0 82.0
io nfi[
lat 3%
Inf gs
lo
2 0 3
26 24 92
um 0.1 0.1 4. 534.5 3 7 5 0
88 32 44 00
5. 0. 3. 1. 008.0 6
93
0. 000.1 citse 7
00 53
P
D ) 1. 0. 17.0 .00 56.0 52.0 85.0 64.0
m omd
axi e
at
M ivr t/Giderc gsol
n(i .s
ion
P at
4 3 4 6 3 9 0 0
10 00 00 58 61
92 24 08 34 35 53 00 25 y 1. 0. 0. 68.0 0 06.0 32.0 75.0 94.0 rves
um 0.1 6. 311.0 1. 1. 8. 0. 0. .20 0. .00 ob
ar nte .10
8
m
nii ndoce llm )sgol 51
ro n(i s
ons M S en ha
ity
6 2 7 1 9 2 9 0
essi 76 00 84 89 19 89 48 28
2. 1. 0. 0. 1. 1. 0. 0. 541.0 8
14
0. 212.0 rep ) 00
gs 1. .800 .730 59 71
0. .790 .250 .670 .600 bilix
s P lo 33.0 0. fle
it
s* Dev. D
Regr in( ex
ht onit d.
St antni Gla at m
iti
va In piac fir
owr er erm ileh
88 82 at 9 8 5 7 5 7 5 w
00 19 23
G obs
n 652.1 400.8 625.3 367.3 792.1 327.
3 056.2 538.0 Det .3 s
.40 655.0 of
.50 1. 0. .10 .20 0 0. 70.0 .20 .10 .20 .20
d
orf tear piac onti
544 ea an
th
cstisit s,eirt M ht per
wor P
D
G rvaesbo
G
un 3
Grow 52
co s
Sta 82 ) ene ha
evi s, gsol wt )s leb
log
be ria
age at ) ) ni( ]) ta at ) )s n(i ) ta
ipt er va
gs gs P
av lo gsol DP
n(i %3 piac snoi lo log D 3%] ity
escr
D doirep piacr in( et
pe erp at piacr G
y in(
pe in(t it/
at
P nte ilit at
P en tear. piacrep y bilix
ilit fle
D
3:
ar te piacr mll /Gtiderc ra.fn[i esnli ib m ib y
s
ro e es ytiilbix y D s
ilit Correl ilit
e piacr edrc nfi[ senli lity
oll e ibi
es rketa
ix ye
5 iara Gfo ib ib
e pe en citse gol on Gfo
e e pe estic log on
e P n(i lef mr
ph ennp yr exlf nre
P y m ni( exfl
ph exlf
ni iv atr D atr ennp exlftekra try bo
ht n leet D ar do
O ancn exlftekra it
ht e O it
la
a e nte atirav
ex nd no leet en ex
ppend Un) abl
owr Glaiti yradnoce mod
etavir iot er at e
n n
laf
A Dat (a Vari G In S P In Mai adrT ov
G mrobaL mriF rmiF Bi)b( lebai owr Glaiti
Var G In Seco Priv atiflnI Mai adrT ecnanrevo
G mrobaL rmiF rmiF heT
*