WPS3925
Productivity Matters for Trade Policy:
Theory and Evidence
Baybars Karacaovali*
University of Maryland
Abstract
There is a growing literature that investigates the effect of trade liberalization on productivity. Nearly all
such studies assume that trade policy is determined independently of productivity, hence it is exogenous. I
show, both theoretically and empirically, that this assumption is not valid in general and that researchers
may be underestimating the positive effect of liberalization on productivity when they do not account for
the endogeneity bias. On the theory side, I demonstrate that under a standard political economy model of
trade protection, productivity directly influences tariffs. Moreover, this productivity-tariff relationship
partly determines the extent of liberalization across sectors even in the presence of a large exogenous
unilateral liberalization shock that affects all sectors. The link between productivity and tariffs is
maintained after I include in my political economy model a learning-by-doing motive of protection, which
also serves as the source of liberalization. On the empirical side, I examine total factor productivity (TFP)
estimates obtained at the firm level for Colombia between 1983 and 1998, and find that more productive
sectors receive more protection within this period. In estimating the effect of productivity on tariffs, I
control for the endogeneity of the two main right-hand-side variablesthe inverse import penetration to
import demand elasticity ratio and productivityby employing materials prices, the capital to output ratio, a
measure of scale economies, and the TFP of the upstream industries as robust instruments. I also account
for the large trade liberalization between 1990 and 1992, and find that the sectors with a higher productivity
gain are liberalized less. Finally, I use a system of equations to illustrate that the positive impact of
liberalization on productivity grows somewhat stronger when corrected for the endogeneity bias.
JEL Classification: D24, F13, F14.
Keywords: Productivity, trade liberalization, endogeneity, political economy of trade policy, learning-by-
doing.
World Bank Policy Research Working Paper 3925, May 2006
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.
*Department of Economics, University of Maryland, College Park, MD 20742, email: karacaov@econ.umd.edu. I
greatly acknowledge Nuno Limăo for his support throughout this project. I thank John Haltiwanger and especially
Marcela Eslava for generously sharing part of their data. I have benefited from the comments of Marco Arena, Thorsten
Beck, Roger Betancourt, Quy-Toan Do, Allan Drazen, Jonah Gelbach, Guliz Kalender, Harry Kelejian, Varun
Kshirsagar, Ramon López, John McLaren, Peter Murrell, John Shea, an anonymous referee for the World Bank Policy
Research Working Paper Series and the seminar participants at the World Bank Development Economics Research
Group, the University of Maryland at College Park and Fordham University.
1 Introduction
In general, how government policies affect economic outcomes is a crucial issue for both economists
and policymakers. In particular, it is important to know how government trade policies affect
productivity in the economy, hence eventually growth and development. This is a major issue
in development and international trade economics and the interest in this topic is not new. The
papers on this issue date back to the 1950s (e.g. Johnson 1955).1 We recently see an increased
interest in the topic through the ever-growing number of empirical studies testing the effect of
trade liberalization on productivity (e.g. Tybout and Westbrook 1995, Pavcnik 2002, Schor 2004).
Many developing countries (for example, Brazil, Colombia, Chile, India, Mexico, and Turkey)
have aggressively pursued trade liberalization in the late 1980s and the early 1990s, in part, to
boost productivity. So, does trade liberalization really increase productivity? Recent micro-level
empirical findings indicate that the answer is "Yes." However nearly all these studies fail to recognize
that trade policy might be endogenous with respect to productivity. And, even if they acknowledge
the existence of this endogeneity, most do not control for it. In this paper, I show, both theoretically
and empirically, that productivity directly affects trade policy. Thus, a concern for the endogeneity
bias is well-founded. Moreover, when we account for the bi-directional causality between trade
policy and productivity, the positive effect of trade reform on productivity may become stronger.
In Section 2, I provide a theoretical model of tariff policy determination for a small open
economy. I show that under a standard political economy setup of protection, the sectoral tariffs
depend positively on the industry production (size) and hence, on the sectoral productivity if these
sectors are organized and lobby for protection.2 The intuition for this result is that more productive
sectors have got more to gain from lobbying and the potential to generate more protection. More
specifically, the marginal benefit of a tariff, hence a higher domestic price is greater when it applies
to more units and more productive processes.
In this paper, I focus on the effect of productivity on trade policy due to political economy
motivations as I explain above. However, there are other plausible channels that could actually
lead to more (not less) protection for less productive sectors as I discuss in Section 2. My purpose
1A detailed historical review of the papers that are concerned with the effects of trade on industrial performance
appears in Pack (1988). He suggests that the early evidence is rather mixed.
2This is a modified result (as I explain later in the text) from the, now standard, political economy models such
as Grossman and Helpman (1994). The results from Grossman and Helpman (1994) have been widely tested and
confirmed. Also Ferreira and Facchini (2005), who find that more concentrated sectors receive more protection, share
a similar view with my paper in the sense that the industry characteristics matter for trade policy.
1
of using this standard model as the starting point is to create a tractable setup for the effect of
productivity on tariffs and then, having accounted for the economy-wide trade reform, to ultimately
obtain structural equations for econometric analysis.
It is often argued that trade reform may be used as an exogenous change and this uniform shift in
policy helps identify the effect of trade policy on productivity without worrying too much about its
endogeneity. In order to account for this argument, I model a unilateral trade liberalization shock
which is common across sectors. I find that under such a common exogenous shock, the reduction
in tariffs is lower for the sectors that experience a higher productivity gain (or a lower productivity
decline) as compared to the other sectors in the presence of political economy. Initially, I keep the
additional channel simple making sure that it is clearly the political economy consideration that
is driving the results. Next, I give more structure to the extra channel of protection and to the
way the liberalization shock manifests itself by allowing for an infant industry argument.3 This
argument led to the widespread use of import substitution policies in most developing countries
until the mid-1980s and its dismissal was the source of much unilateral trade liberalization since
then.
In my two period model, the government initially believes that there exists a learning-by-doing
(LBD) process and decides about the current tariffs by considering both the political economy
effects and the effect of these tariffs on future welfare through LBD. In the second period, the
government realizes that LBD does not actually exist and it is a false perception, and thus initi-
ates a trade reform. Given that political economy forces still determine tariffs, the extent of the
liberalization differs across sectors (due to political economy) despite this common shock across
sectors. The implication from the models I present is that, assuming cross-sectional differences in
trade policy to be independent of cross-sectional differences in productivity is incorrect.
I employ Colombian data for the empirical tests of my theory. Colombia has been used in
various studies (for example, Roberts 1996; Fernandes 2003; Melendez, Seim and Medina 2003)
given that it provides a great natural experiment environment. Colombia experienced a drastic
trade reform in the early 1990s and had a stable economy in this period without major crises.
Given the existence of other studies using the same country, I get the chance to test my predictions
with a comparable dataset. In Section 3, I discuss the trade policy in Colombia during the sample
period of 1983 through 1998, and describe how we can rule out a uniform change in tariffs across
3For example, the infant industry argument is mentioned as an important motive for protection in Grossman and
Horn (1988).
2
sectors.
In Section 4, I briefly review the empirical literature to the extent that it relates to this paper
and then, in Section 5 I make an empirical analysis based on the predictions from my theory using
Colombian data for the 1983-1998 period. I analyze the effect of productivity on trade protection
by following my theory closely. I confirm that the (4-digit ISIC4 level) sectoral tariffs are inversely
related to the import penetration ratio and import demand elasticity, whereas they are positively
related to total factor productivity. The results also indicate that the sectors with more productivity
gain (or less productivity decline), as compared to the other sectors, are liberalized less.
In the estimations, I account for the extra channels which I mention above and, more specifically,
I account for their elimination which leads to a big decline in tariffs. For this purpose, I allow for
a shift in the common terms across sectors over time. I tackle the potential endogeneity of the
right-hand-side variables, namely the inverse import penetration to import demand elasticity ratio
and productivity, by using instrumental variables. The instruments are the capital to output ratio,
materials prices, a measure of scale economies (value added/number of firms), and the TFP of the
upstream industries which I confirm to be valid based on a test of overidentifying restrictions and
also explain intuitively in Section 5.
My productivity estimates come from Eslava, Haltiwanger, Kugler and Kugler (2004) and they
are originally calculated at the firm-level which are then aggregated to the 4-digit ISIC level using
production shares of each firm. The dataset has the great advantage of using plant level input
and output prices. This enables obtaining good estimates with smaller bias as compared to the
majority of the studies which need to employ non-parametric estimations and sector level price
deflators due to lack of data.
By showing how trade policy depends on productivity both theoretically and empirically, I
provide solid evidence for the endogeneity of trade policy with respect to productivity. Finally, in
Section 6, I estimate a system of equations for illustrative purposes. In the system, I correct for
the endogeneity of tariffs and show that the positive effect of trade liberalization on productivity
can be underestimated when endogeneity bias is not accounted for.
4International Standard Industrial Classification, United Nations.
3
2 Theory
2.1 Basic Model
The output and factor markets are perfectly competitive. The numeraire good, i = 0, is pro-
duced with labor only, using a constant returns to scale process, whereas the non-numeraire goods,
i = 1,...n, are produced using labor and one sector-specific factor. The production function for the
non-numeraire goods is Xi(pi) = AiQi(pi),5 where Ai stands for the Hicks-neutral total factor pro-
ductivity (TFP). The international prices for all goods i = 0,...,N, denoted by pw, are normalized
i
to one. Furthermore, assuming a large enough aggregate supply of labor, the wage rate is also tied
at one--the marginal revenue productivity of labor in the numeraire sector.
The numeraire good is traded freely, hence its domestic price is equal to the world price of one.
The owners of each specific factor organize into lobbies, and ask for government protection in their
own sector only since they are assumed to constitute a negligible share of the total population.
The consumers cannot overcome the free-rider problem and are not organized as discussed in Olson
(1965). For simplification, export subsidies are not allowed and only tariffs are available for trade
protection.6 Maintaining the small country assumption and the world prices normalized to 1, the
domestic price of the remaining goods are given by pi = 1 + i, where i denotes both the specific
and the advalorem tariff rate.7 The government, then, sets its trade policy by maximizing the
following political support function
G L + PN (1)
i=1 µZ1+i
Di(i)di + Z0 1+i AiQi(i)di + iMi(i)¶
where L is the aggregate labor supply and income; Di(i) is the aggregate demand, Xi(i) =
AiQi(i) is the aggregate supply, and Mi(i) = Di(i)-AiQi(i) is the aggregate import demand
for good i. Thus, G is a weighted sum of the aggregate consumer and producer suplus, as well as
labor income and tariff revenue.8 The weight, 1, represents the relative importance given to
the producer surplus with respect to the rest of the social welfare.9
5In the estimations I give Qi(pi) more structure. Here, it just indicates part of the production function independent
of productivity.
6The trade with the rest of the world is balanced through movements of the numeraire good.
7Note that the domestic price for a small economy is related to the world price and tariff rates with pi =
(pw+specific tariff ) = pw(1+advalorem tariff ) but the assumption that pw = 1 makes the specific and advalorem
i i i
rates equal under this case.
8The tariff revenue, PN iMi(.), is fully rebated back to the public in a lump-sum manner.
i=1
9This setup can be easily interpreted as a reduced form of a model where lobbying is given micro-foundations
such as in Grossman and Helpman (1994). I prefer to take a shortcut here to keep the focus on the main subject
4
Given the additive separability of the government objective, we can obtain the optimal tariff
rate for sector i by maximizing equation (1) with respect to i. Consequently, the equilibrium
specific and advalorem tariff rates for sector i are implicitly defined as10
AiQi(i)/Mi(i)
i = ( - 1) (2)
i(i)
where i(.) stands for the elasticity of import demand.11 This expression is a standard one obtained
in various political economy models (Helpman 1997). Accordingly, the tariff rate for sector i is an
increasing function of the extra political economy weight provided to producers, whereas it is a
decreasing function of the import demand elasticity, i, and import penetration ratio, Mi/AiQi. A
tariff is a tax on imports, so just like a tax on a non-traded good, the deadweight loss created is
lower the more inelastic the (import) demand is. Thus, a smaller value of i allows for higher tariffs
to be applied. In addition, a relatively larger market for imports creates a greater price distortion
potential which should be avoided by the government. Finally, the marginal benefit of a tariff is
higher when it applies to more units and more productive processes.
Partially and implicitly differentiating equation (2) with respect to Ai, we obtain the following
relationships between tariff protection and productivity
i Qi(i)
(3a)
Ai = -( - 1)Mi0(i) > 0
di
= -Mi0(i) ( - 1)Qi(i) > 0 (3b)
dAi + ( - 1)AiQ0i(i)
I assume that Qi(i), Di(i) and hence, Mi(i) are linear for ease of exposition (i.e. Q00i (i) =
Di00(i) = 0).12 Although initially 1 is the only restriction on the value of , I further assume
it to be bounded above such that < 2 - Di0(.) based on the observations from the empirical
AiQ0i(.)
political economy literature. For example, Goldberg and Maggi (1999) estimate to be equal to
1.014 for the United States, whereas Karacaovali and Lim~ao (2005b) estimate it to be between
1.0025 and 1.0039 for the European Union. Although these estimates might be a bit small, they
matter: the trade policy and productivity linkage in the presence of unilateral liberalization.
10Again note that the specific (first term) and advalorem (second term) tariff rates are equivalent since the inter-
national prices are normalized to one. See the appendix (Section A.1) for the derivation of equation (2).
11Here, the import demand elasticity is defined as i -Mi0pw/Mi, so it differs from the standard definition
i
which is evaluated at the domestic price. I account for this in the empirical estimations as explained in the appendix
(Section A.2).
12I have the same assumption throughout the text. This is not a necessary condition for the inequality in equation
(3b) to hold or the other results to follow.
5
more than support my parameter restriction as a plausible one. If were so high to exceed
2 - Di0(.) 13, then more productivity would call for less protection. Under our political economy
AiQ0i(.)
setup, this would be counter-intuitive given that a high together with a high Ai only indicate a
stronger lobby and require more protection, not less.
The main result is that, based on a standard political economy model, we expect an organized
sector with higher productivity to receive more protection because it has got more to gain for a
marginal increase in the tariff rate, hence the domestic price level. Thus, this is a slight modification
of the size effect identified by the influential work of Grossman and Helpman (1994) which has
been tested and confirmed in various papers (like Gawande and Bandyopadhyay 2000, Goldberg
and Maggi 1999, Mitra et al. 2002 and so on). More importantly, this basic observation naturally
raises doubts about the assumption of exogeneity of trade policy with respect to productivity in
the earlier empirical literature.
Before I extend this basic setup by introducing common shocks, I should note that we could
potentially have a channel working in the opposite direction. That is, one can plausibly expect a
less productive sector to obtain more protection. One way to model this is by allowing the political
economy weight to differ across sectors based on certain sectoral characteristics as in Karacaovali
and Lim~ao (2005a). For example, a sector with a higher share of employment is likely to have a
higher weight given that it generates more political votes (Caves 1976). Furthermore, sectors with
lower wages may have a higher weight due to the government and society-wide sympathy with their
situation or simply due to the lower opportunity cost of lobbying by the low wage workers (Magee
et al. 1989). Typically, the low wage and labor intensive sectors are less productive so this could
potentially reverse the results. I exclude such concerns from my model because I want to focus
on the effect of productivity on tariffs in the presence of a big trade reform shock that affects all
sectors and accounting for further sector characteristics would only complicate this analysis. On
the other hand, I allow for such differences across sectors that could lead to different initial tariff
rates by considering fixed effects in my estimations (Section 5).
2.2 Trade Reform
Nearly all the papers examining the trade reform-productivity linkage involve a period of unilateral
liberalization which is usually considered to be an exogenous shock independent of productivity
13Note that Di0(.) < 0.
6
and common across sectors. Then, the exogeneity of the liberalization shock is used to defend the
argument that we should not be worried about the endogeneity of trade policy. Therefore, I provide
room for a unilateral liberalization motive common across sectors in order to create a similar setup
and analyze its effects. We would like to see whether such a common shock does indeed produce a
proportional, non-selective decline in tariffs.
In order to capture this common and exogenous shock argument, I simply augment the baseline
government objective function, G, with an additional term, () = PN i(i). This extra term
i=1
does not create a different economic structure, that is we still have consumers with quasilinear utility
functions, a constant returns to scale production with no spillovers and so on. The government
objective function can now be expressed as
G + PN i(i) (4)
i=1
where G is the same as in equation (1), i(.) is increasing in i (and concave), and > 0 is
a constant. () is meant to capture the government perceived benefit of using protective trade
policy and would call for protection even in the absence of lobbying. One can think of the perceived
benefit as a government view that favors import substitution or as an unquestioned historical legacy
of trade protectionism. Initially, I use this approach to be able to clearly show that the tariff changes
and levels depend on the productivity changes and levels even under the simplest setup of trade
reform. However, in the next sub-section, I put more structure on the way liberalization manifests
itself by modeling an infant industry argument, which is known to be a crucial protection motive
for developing countries.
The equilibrium tariff rate obtained by maximizing equation (4) is given by14
AiQi(i)/Mi(i) 0i(i)
i = ( - 1) + (5)
i(i) Mi(i)i(i)
The first part of the expression in equation (5) is essentially the same as equation (2). The
additional 0i term, on the other hand, captures the marginal perceived benefit of tariffs, again
weighted by imports and import demand elasticity. I assume that the 0i terms are identical across
sectors in order to get a uniform effect, that is 0i = 0j for i 6= j. As I show in the appendix
14The derivation is similar to the one for equation (2), so it is omitted.
7
(Section A.1), tariffs increase in the coefficient of perceived benefit
di
> 0 (6)
d
Many developing countries, including Colombia, have gone through significant unilateral trade
liberalization in the late 1980s through the early 1990s. As I mention above, it is often argued
that such liberalization episodes can be interpreted as an exogenous shock mostly uniform across
sectors. Thus, empirical researchers regress productivity on tariffs by exploiting this variation
over time maintaining an exogeneity assumption. Now, I model such a unilateral liberalization
shock as a dramatic decline in the parameter , say, all the way down to zero.15 Note that, this
shock is modeled to be common across sectors on purpose and it does not depend on any industry
characteristics. Yet, as we will see below, the political economy motives still affect tariffs so that
the reduction in tariffs for a sector depends on the change in its level of production or size and
hence, its productivity.
0i(it)
it+1 it+1|=0 - it|>0 = - ( - 1)Ait+1(Qit+1(it+1) + (7)
Mit0+1 it+1) ( -Mit0
1)AitQit(it)
+
(it) Mit0 (it)
Then, using the linearity of Mi,16 equation (7) can be re-expressed as
(Ait+1)Qit(it) + Ait+1(Qit+1) 0i(it)
it+1 = -( - 1) + (8)
Mi0 Mi0
The partial effect of Ait on it is the same as given in equation (3a). Now, we also obtain the
following relationships between tariffs and productivity in levels and changes
dit
dAit|0 = -Mit0 ( - 1)Qit(it) > 0 (9a)
(it) + ( - 1)AitQ0it(it) + 00i (it)
dit+1
dAit+1|it,Ait = -Mi0(. ( - 1)Qit+1(it+1) > 0 (9b)
) + ( - 1)Ait+1Q0it+1(it+1)
Equation (9a) is obtained by implicitly differentiating the first part of equation (5) (specific tar-
15This change could be due to a contingent loan from the IMF or a policy recommendation from the World
Bank which require certain stabilization and liberalization policies from our "small" country. Or it could be due
to a change in the paradigm having observed the success of other comparable liberalizing countries and a new
international consensus degrading import substitution type of policies. For example, Edwards (1997) analyzes the
role of the World Bank in its effect on trade liberalization reforms and acknowledges its contribution through research
and policy dialogue.
16I assume that the paramaters do not change over time, and combining with the earlier linearity assumption we
get Mit (.) = Mit(.) = Mi0(.).
0+1 0
8
iff) with respect to Ait whereas, equation (9b) is obtained by implicitly differentiating it+1 as
expressed in equation (8) with respect to Ait+1 for given initial levels of it and Ait.17
We see that a sector with higher productivity is expected to receive more protection and a sector
with a bigger increase (or a lower decline) in productivity is expected to have lower reduction in
tariffs despite an exogenous shock common across sectors. Thus, we have reasons to worry about
the endogeneity of tariffs with respect to total factor productivity. Accordingly, in the empirical
studies where the sector level productivity is regressed on the sector level tariffs that are assumed
to be exogenous, there will be a direct reverse causality problem. In the case of the firm level
productivity being regressed on the sector level tariffs, this problem will be smaller. However,
to the extent that the firm level productivities in a sector differ commonly from the firm level
productivities in the other sectors or the more correlated the firm level productivities with the
sector level productivities, the worse the endogeneity will be a problem. In the empirical section I
use productivity estimates obtained at the firm level that are then aggregated to the sector level
using production shares as weights to arrive at representative productivity values for each sector.
Finally, notice that the productivity-tariff linkage above is completely driven by the political
economy channel. If political economy is not a concern for the determination of tariffs for a given
sector, that is, if the extra political economy weight is null ( - 1 = 0), then productivity has no
effect on the tariffs. On the other hand, again because of political economy, the reduction in tariffs
varies across sectors based on the productivity differences regardless of the common shock.
2.3 Government Perceived Learning-by-Doing
In this sub-section, I provide more structure for the liberalization process and the government
perceived benefit of protection by introducing an infant industry argument. In developing countries,
learning-by-doing and infant industry arguments have been a major motivation for protection which
should be accounted for. Grossman and Helpman (1995) provide a comprehensive survey of the
literature on technology and trade and indicate that "some countries might wish to use trade or
industrial policies to alter their patterns of specialization... The short-run income loss for such a
country would be small, while the policy would generate a permanent boost to its productivity
growth..." (p. 1297). However, it should be noted that import substitution policies and infant
industry protection have been largely abandoned especially after the 1980s and some critics have
17Naturally, we also have it/Ait > 0 and it /Ait > 0.
+1 +1
9
indicated that the infants actually never seem to grow (see, e.g., Krueger and Tuncer 1982). In
this spirit, the trade liberalization episodes in the developing countries can be seen as a result of
the disillusion about the infant industry argument. That is, the governments go from strongly
believing in the argument to understanding that it does not work. I would like to examine the
effect of such a shift in the government beliefs on the structure of liberalization. Therefore, I model
a learning-by-doing (LBD) process which is merely a perception by the government.18 Although
there is no LBD, the government believes that there is some and thus sets its tariffs accordingly
until it realizes that this is a false perception and then embarks upon a trade reform.
More specifically, the government believes that more production today has a positive impact on
tomorrow's productivity and hence takes this relationship into account while determining its current
trade policy. On the other hand, the firms decide about their production by simply reacting to the
prices determined by the government trade policy and their decisions do not depend on any LBD
process. For simplicity, I assume that the government has a two-period policy setting horizon.19
This assumption is not only computationally convenient but also helps us partially capture the real
experience in Colombia20 which I study in the empirical section to follow.
In this setup, I aim to provide a plausible explanation for the way unilateral liberalization
is introduced. The liberalization shock is common across sectors as in the basic model but the
government's perceived benefit of protection now has a specific reasoning based on a LBD process.
I model the government objective in equation (1), but now the government has the following
belief about the form of the supply function
Xit(it) = Ait(it,it)Qit(it) (10)
Ait(.), like before, denotes the total factor productivity, it = (Xit-1,Xit-2,...) represents the
learning-by-doing process, and it stands for the determinants of TFP that are independent of LBD.
The government believes that (.) is an increasing function of the past production within the same
sector, that is 0(.) > 0. Note that the true supply function is actually Xit(it) = Ait(it)Qit(it).
Each period, the government sets tariffs considering their current effects on the weighted social
18I gratefully acknowledge Nuno Lim~ao for his suggestions here.
19One can think that the government sets trade policy quite infrequently such that tariffs are first determined when
the government believes that there exists a strong learning-by-doing process at play and later when this perception
is discarded because productivity gain is not observed or the process reaches its terminal point. Alternatively, this
might be a short lived government that expects to be in power for two periods only.
20This is a feature shared by many other developing countries such as Turkey, Brazil, and India that experienced
significant amounts of liberalization around the 1983-1985, 1991-1996, and 1990-1993 periods, respectively.
10
welfare as discussed in the previous section but now it additionally considers the perceived future
effects of current tariffs via learning-by-doing.
The equilibrium tariffs for period t can be obtained as
^t argmax[Gt + Et(Gt+1|t)] (11)
t
where the Gj terms are defined as in equation (1) but with time subscripts and Xit(.) takes the
form in equation (10). t denotes the information set of the government in period t, and < 1
is the time discount factor. At period t the government knows that TFP has some baseline value
Ait(.) = it and believes that the future expected TFP is given by Et(Ait+1(.)) = Et(it+1it+1) =
it+1(itQit).21
Solving backwards, we obtain the tariffs for period t+1. The realized values of these tariffs are
set after the government observes Ait+1 and finds out that there is actually no LBD. Therefore,
the actual period t + 1 tariff rate is equal in its form and value to the one in equation (2). On the
other hand, in order to determine the tariffs set in period t, the government needs to compute the
future expected welfare which depends on the expected period t + 1 tariffs. Given that there are
two periods, the expected tariffs for period t + 1 have the standard form similar to equation (2);
however, due to the LBD process, each of its components, hence itself is expected to depend on
period t tariffs
Et(it+1) = eit+1(it) = ( - 1) it+1(Xit(it))Qit+1(Xit(it))/Mit+1(Xit(it))
(12)
it+1(Xit(it))
As I show in Section A.1 in the appendix, the equilibrium tariff rate for period t is obtained
from equation (11) such that the tariffs now include the perceived learning-by-doing motive in
addition to the political economy channel.
itQit(it)/Mit(it) i/Mit(it)
it = ( - 1) + (13)
it(it) it(it)
The variable i stands for the LBD effect and it is defined as
i µ 0
(itQi(it))
(itQi(it))itQ0it¶Z0 1+eit+1
it+1(itQit)Qit+1dit+1 > 0 (14)
21Note that the government is not taking expected values over alternative values of it . Instead it expects it
+1 +1
to be equal to it with probability 1.
+1
11
Thus, i measures the government perceived growth in productivity due to the LBD process
multiplied by the responsiveness of the current supply to tariffs and weighted by the future producer
surplus. The tariff rate is increasing in the additional LBD term since the government considers
the positive effect of increased production through today's protection on tomorrow's welfare.
Let's consider the following functional form for the LBD process22
it+1 = (Xi(it)) = [itQi(it)]n, n < 1 (15)
Next, in order to see the effect of productivity on tariffs, we plug equation (15) in equation (13) and,
as I show in the appendix (Section A.1), obtain
dit
> 0 (16a)
dit+1
it
> 0 (16b)
dit
dit
> 0 (16c)
dit
Thus, assuming that it, the part of the government perceived productivity that is independent
of the LBD process, and it+1, its future expected value, are positively correlated with the actual
underlying determinants of TFP, the tariff rate increases in the current and expected future pro-
ductivity. This might be one of the reasons why we need to worry about using lagged tariff rates
as a way to get around the endogeneity problem while regressing productivity on tariffs. More im-
portantly, the positive effect of productivity on tariffs confirms the main result in my basic model
in this richer setup.
Next, I also confirm that the change in tariffs is positively related to the change in productivity
as in the previous section23
dit+1|it,it > 0 (17)
dit+1
We see that, introducing a government perceived LBD process adds a new channel of protection
and a structure for the onset of the unilateral trade liberalization. However, the productivity to
tariff linkages, as established in the basic political economy model, prevail.
22I assume (.) is concave. Otherwise, in the multi-period case, tariffs could be raised unboundedly which is quite
unrealistic.
23Apart from the supposedly temporary nature of protection due to the infant industry argument, I assume that
in period t+1 the government actually realizes that the LBD process is not working. See the appendix (Section A.1)
for the derivation.
12
As a final note, I acknowledge that in several empirical studies, declining industries such as
textiles, clothing, footwear, and steel receive more protection in industrialized nations. Therefore,
these studies naturally look at developed countries and especially the United States for evidence (see
for example Baldwin 1985, and Marvel and Ray 1983). On the theory side, two recent papers that
build up on Grossman and Helpman (1994) are worth noting: Baldwin (2002) uses a monopolistic
competition model with sunk entry costs and random Markov process demand shocks; whereas,
Tovar (2004) introduces loss aversion in preferences. One implication from these papers is that
less productive sectors could receive more protection. However, none of the models on declining
industries specifically account for unilateral liberalization or focus on establishing a link from
productivity to trade policy. The models I presented in this and previous sections intend to show
this link in the presence of unilateral liberalization which is taken to be an exogenous shock and
hence used as the working assumption to identify the effects of tariffs on productivity in the
earlier literature. Yet, the common denominator of my models and implications from these is that
productivity matters for trade policy. Thus, how productivity specifically affects tariffs needs to
be tested and documented empirically as I do in Section 5.
3 Trade Policy in Colombia
Colombia is a perfect example of a developing country that went through phases of heavy trade
protection prior to the mid-1980s and finally a dramatic unilateral trade liberalization in the early
1990s, as can be seen in Figure 1. Therefore, it is no surprise that Colombia has been used as the
case study of several papers to examine the impact of trade reform on productivity (for example,
Roberts 1996; Fernandes 2003; Melendez, Seim and Medina 2003).
The barriers were first lowered during the 1977-1981 period in response to an increase in the
coffee prices, increased foreign borrowing, and drug trafficking (Fernandes 2003). On the other
hand, the Latin American debt crisis and the worsening terms of trade led to an increase in
protection in the first half of the 1980s (Edwards 2001). President Virgilio Barco Vargas started
the initial movement towards a real trade reform after he took office in 1986. He was succeeded by
President Cesar Gaviria who completed the trade reform swiftly in two years (1991 and 1992).
As I mentioned above, nearly all studies in the trade reform and productivity literature neglect
the endogeneity of trade policy. Although some authors (e.g., Pavcnik 2002 for Chile; Ferreira
and Rossi 2003 for Brazil) acknowledge the potential for endogeneity, they argue that it may not
13
be such an issue in their studies given that the tariffs were reduced uniformly or proportionally
across sectors. This is not true at least for Colombia; the liberalization was not uniform. Edwards
(2001) notes that the trade liberalization reform of Colombia ("La Apertura") was "announced
during the presidential campaign [of Cesar Gaviria] as a `gradual' and `selective' process." As
can be observed from Figures 2 through 4, there is quite some variation in the tariff reductions
across sectors.24 Moreover, the Spearman's rank correlations of tariffs in Table 1 indicate that
the correlations reduced over time, implying a selective process as opposed to a uniform one in
liberalization. Otherwise, the ranking of sectors in terms of their protection rate would not change.
The average advalorem tariff rates in my sample of 4-digit ISIC industries declined from 43%
in 1983 to about 14% in 1992 and stayed around that rate in the following years (Table 2). The
dispersion of tariffs across sectors also declined (Table 2). If we just look at the standard deviations,
the decline appears to be markedly higher. However, we need to take into account the differences
in the magnitude of tariffs across periods and the coefficient of variation25 is a better measure
for that matter. The decline in the dispersion is notably lower when we use the coefficient of
variation. However, this outcome does not indicate that political economy is no more a factor in
tariffs after reform. The decrease in the dispersion is predicted by my models given the fact that
the liberalization occurs through the elimination of some extra channels other than the political
economy channel.
In my theoretical models and the main estimations, I use tariff rates as a measure of protection
for import-competing sectors. I also have access to effective rate of protection (ERP) data which I
use to augment my results with tariffs. The effective rates take into account the tariffs on inputs,
and they are based on the value added. They are considerably higher than the nominal rates but
show a similar pattern with the tariffs as can be observed from Figure 1.
Before I discuss my empirical methodology and the results, in the next section I briefly review
the existing empirical literature to the extent that it relates to this paper.
4 Empirical Literature Overview
Tybout (1991) briefly reviews the literature that contains implications for the linkages between
trade and productivity, and he indicates that the net effect of a liberalization episode is ambiguous.
24Note that the reductions are computed as percentages to account for the variation in the initial tariffs.
25Coefficient of variation is the mean divided by the standard deviation of a given group.
14
Therefore, a majority of the studies that appear in the last decade remain empirical and do not
test any particular theory.26,27
Nearly all researchers take a two-step approach, where they first estimate productivity usually
at the firm level (and some at the sector level), and then regress this productivity estimate on
trade policy measures such as import penetration or tariffs for a single country (e.g., Fernandes
2003, Schor 2004, and Tybout and Westbrook 1995). Another strand of the literature focuses on
the effect of imperfect competition in estimation and seeks to analyze the change in the price-cost
margins after liberalization (Harrison 1994, Krishna and Mitra 1998 and so on).28
Although nearly all the studies neglect the endogeneity issue, Fernandes (2003) is an exception.
She controls for endogeneity by using lagged tariff rates instead of current ones. She also consid-
ers the variables from Trefler's (1993) non-tariff barrier (NTB) equation as instruments for tariff
rates in a robustness check. However, using lagged tariff rates might not get around the endogene-
ity problem, because trade policy might differ across sectors due to persistent factors related to
productivity. For instance, productivity might be autocorrelated, or tariffs may be influenced by
anticipated changes in productivity as predicted by one of my theoretical results. What is more,
the validity of the instruments initially used by Trefler (1993) for a different study is debatable
since some of the instruments (like import penetration or regional concentration) could very well
be influenced by productivity, and hence be endogenous themselves.29 Muendler (2004) is another
exception in trying to control for the endogeneity of trade policy. He regresses the growth rate
of productivity on both tariffs and import penetration at the same time. He considers certain
components of the real exchange rate as instruments for the trade policy measures.30 However,
26The motivation for the micro-level liberalization impact studies is based on two basic conjectures. First, trade
liberalization may produce a productivity growth for the firm and the industry through economies of scale, improved
access to foreign technology, and the elimination of X-inefficiencies. Second, liberalization may reallocate resources
from the less efficient to the more efficient firms after the less efficient ones exit, hence provide a rise in the average
productivity.
27There are also ex-post theoretical studies that provide an explanation for some of the results found in the recent
empirical research. For example, in an influential paper, Melitz (2003) shows how industry productivity may grow
due to reallocation between firms after an exogenous trade reform shock.
28In this paper, I focus on the single-country micro studies and relate my results directly to these. However, there
is also a related group of empirical papers where authors analyze cross-country growth regressions (as summarized
in Harrison 1996) linking openness to output growth. Such studies miss the micro variation, which is crucial in
distinguishing among various channels of productivity changes, in the data. A recent criticism of these studies
appears in Rodriguez and Rodrik (2001) who are in turn criticized by Srinivasan and Bhagwati (2001).
29Fernandes (2003) acknowledges that these robustness results are not reliable, as some of her instruments are
clearly correlated with productivity.
30The measures are the nominal US dollar exchange rate, the average sector-specific European and US-Canadian
producer price indices, and the Brazilian consumer price index. Muendler (2004) recognizes that the domestic prices
can be correlated with the productivity of the firms so does not consider this as one of his baseline instruments but
rather keeps it as a component of the real exchange rate.
15
both the nominal and real exchange rates lack sectoral variation and cannot explain why tariff
rates differ across sectors.31
Harrison (1994) uses time dummies for capturing trade liberalization but these do not account
for the firm/industry level variation in policy. She also considers tariff changes and import penetra-
tion in her estimations by interacting the trade policy measures with the relevant mark-up variable.
These estimations invariably suffer from the same endogeneity problems I discussed above.
Pavcnik (2002) takes yet another approach and compares the productivity changes in the trad-
able versus non-tradable sectors around a trade liberalization period, finding that the import sectors
experienced a larger increase in productivity relative to the non-tradable sectors but the results
are inconclusive for export-oriented sectors.32 This methodology does not account for the sectoral
variations in trade policy as well. Furthermore, Tybout (1996) notes that firms usually self-select
their trade-orientation and if more productive firms are more likely to become an exporter, then
one must use caution in asserting a casual relationship from policy to performance. Pavcnik (2002)
also regresses productivity on tariffs and import penetration as a robustness check, and she does
not control for the endogeneity of trade policy.
Next, I would like to test my theory which predicts that tariffs depend on productivity, and
liberalization differs across sectors based on productivity changes despite a common exogenous
shock.
5 Estimation
5.1 Econometric Model
In the theory section I present two similar models of tariff policy where political economy is the
common determinant. Moreover, the government has some positive perception about using tariffs
which serves as the extra channel for protection. Both models imply protection even in the absence
of political economy. A large negative shock, which is common across sectors, appears through these
channels and serves as the source of trade liberalization.
In the estimations, I intend to capture the common features of protection and liberalization
31Note that, I show in my theoretical model and the corresponding estimations that tariffs directly depend on
import penetration as well as import penetration depends on tariffs in a systematic way so this might further create
multicollinearity problems in Muendler's (2004) estimations.
32Ozler and Yilmaz (2003) use the same approach to analyze the effect of trade liberalization on productivity in
Turkey.
16
implied by my models. According to the political economy channel, tariffs are inversely related
to import penetration (that is Imports/Domestic Production) and import demand elasticity. Re-
call that the production function is denoted as Xit = AitQit where Ait stands for total factor
productivity (TFP) and we have the following definition: Inverse Import Penetration/Import De-
mand Elasticity= AitQit/Mit. The additional source of tariff protection causes a major unilateral
it
liberalization when it vanishes. This occurs after the paradigm changes as discussed in Section 2.2,
or after learning-by-doing is realized to be a false perception as in Section 2.3. I model the addi-
tional channels of protection under both models with a combination of overall and sector specific
constants. Then, the trade reform that occurs due to the disappearance of such motives is a shift
in the intercept terms (constants) of the tariff determination rule. Given the parsimonious nature
of the models, the sector-specific effects also help to control for the other determinants of tariffs
that may not be considered already.
As illustrated in Figure 1, tariffs in Colombia declined drastically starting in 1990 and the
liberalization continued until 1992. Based on the theory, I first start out by assuming that liberal-
ization is a major, once and for all shift in tariffs and relax this assumption afterwards. I capture
the shift with a dummy variable, UNILIBt, that takes the value one for 1990 and onwards, and
zero otherwise. The econometric model can then be expressed as
log it = + 1 log(Qit/Mitit) + 2 log Ait + 3UNILIBt + µi4 + uit (18)
where it is the advalorem tariff rate for sector i = 1,...,N at period t = 1,...,T. Note that the
effect of Ait (TFP) on tariffs is taken in isolation with the use of logarithms. Qit/Mitit, together
with Ait, are measures of the main political economy channel.33 µi is a 1×(N-1) vector of industry
dummy variables and depicts sector-specific effects. UNILIBt serves as an intercept-shifter with
the interpretation I described above. According to the theory, we expect positive estimates of 1
and 2. On the other hand, the estimate of 3 should be naturally negative by definition (it is a
unilateral liberalization).34
In the theory section, the liberalization shock results in a one time permanent decline in tariffs
although it is not necessarily how a real reform progresses. By looking at Figure 1, we can distin-
guish three periods with plausibly three different intercept terms: 1983-1989 (pre-reform), 1990-1992
33Qit is not directly observable but it is estimated by dividing Xit by the estimate of Ait.
34, 1, 2, and 3 are scalars, whereas 4 is an (N - 1) × 1 vector of coefficients.
17
(reform), and 1993-1998 (post-reform). What is more, there exists considerable variation within
the pre-reform and reform periods.35 Therefore, I estimate two different versions of equation (18).
In the first version, I replace 3UNILIBt with 1REFt +2POSTREFt, where REFt is a dummy
variable which equals one for 1990-1992 and zero otherwise. Similarly, POSTREFt is equal to one
for 1993-1998 and zero otherwise.36 Both control for the shift in tariffs in their respective periods
relative to the constant term, . In the second version, I replace 3UNILIBt with t where t
is a 1 × (T - 1) vector of year dummies that capture the yearly common variation in tariffs and
further relaxes the assumption of a one time overall tariff reduction.37
Next, to eliminate the fixed effects, I use a first-differenced model based on equation (18)
logit = 1log(Qit/Mitit) + 2logAit + 3UNILIBt + vit (19)
where the error term is vit = uit and it is essentially autocorrelated so I correct for this autocor-
relation in my estimations. Given the definition of UNILIBt, UNILIBt becomes just a year
dummy for 1990. This may not be adequate to capture the action in the actual data. When we take
first differences, the differenced data for 1983, 1985, and 1988 automatically get dropped given the
gaps in the sample. Moreover, the trade liberalization took place gradually between 1990 and 1992.
I consider this downward trend in the reform years by employing a different version of equation
(19), where I replace 3UNILIBt with REFt. REFt is a dummy variable for 1990-1992 as
above, and is a scalar. I also estimate this modified version of equation (19) with a constant term
to account for the small tariff changes before and after the reform period where the interpretation
of REFt becomes the deviation from the constant term.
There are potentially endogeneity problems in the estimation. First, Qit/Mitit is endogenous
with respect to tariffs since it depends on domestic prices, hence on tariffs. Second, the previ-
ous empirical work documented that trade policy affects productivity which requires accounting
for a potential reverse causation. I use the following list of instruments to deal with the endo-
geneity issues: the capital to output ratio, materials prices, a measure of scale economies (value
added/number of firms), and the TFP of the upstream industries.38 Instruments should be cor-
35Tariffs actually increase between 1982 and 1984 and then start to decline in 1985. The sample, on the other
hand, only includes 1983, 1985, and 1988-1990 for the pre-reform era. Between 1983 and 1988 the trend for tariffs is
a gradual decline within the sample.
361 and 2 are scalars.
37
38 is a (T - 1) × 1 vector of coefficients.
The detailed variable definitions and sources are in the appendix (Section A.3).
18
related with the endogenous regressors and yet be orthogonal to the error term. I present the
formal tests of instrument validity in Section 5.4 but I would like to provide some intuition here.
Capital share is expected to be negatively related to the output/imports ratio (Qit/Mit) given that
Colombia is more likely to produce products with smaller capital content and import those rich
in capital based on a comparative advantage argument. The materials prices affect the domestic
output prices, hence Qit/Mitit but not the tariffs of a given sector i, conditional on Qit, Mit,
and it. Scale is positively correlated with productivity and it is an inherent characteristic of a
sector. The productivity of a sector is also expected to be affected from the embodied upstream
productivity which is likely to be independent of the sector's own tariffs.
Since the model is quite parsimonious, it is also prone to an omitted variable bias. The use
of the fixed industry effects in equation (18) and its different versions, and the first differencing
in equation (19) and its different versions should alleviate this potential problem along with the
Instrumental Variables (IV) estimation.
In the rest of this section I present and discuss the data, results, and robustness of my estima-
tions. In Section 6, I test how accounting for the endogeneity of tariffs with respect to productivity
affects the regressions that analyze the impact of trade reform on productivity.
5.2 Data
The base data for the estimations span 1982 through 1998 but given the lack of tariff and production
information for certain years the sample reduces to 1983, 1985, and 1988-1998. The tariff and
effective rate of protection (ERP) figures are obtained from DNP (National Planning Department)
of Colombia at the 8-digit product level,39 which are then aggregated to the 4-digit ISIC sectors by
using simple averages.40 The 4-digit ISIC level import data come from the COMTRADE dataset,
United Nations Statistics Division and the industry production data at the same level are available
through UNIDO's Industrial Statistics Database.
The productivity estimates, value added, input, and materials prices data are obtained from
Eslava, Haltiwanger, Kugler and Kugler (2004), where each variable (except the value added41) is
aggregated from the firm level to the 4-digit ISIC industry level with production shares used as
39The product classification code, called "Nabandina", is due to the Andean Community of Nations. I thank
Marcela Eslava at Universidad de Los Andes/CEDE, Colombia for generously sharing the data.
40I use simple averages to be consistent with the earlier literature. An alternative way would be to use the import
or production shares of each product as weights but these data do not exist for all sample years at this level of
disaggregation.
41The value added is used to compute a measure of scale economies where it is an unweighted total in each sector.
19
the weights. The main data source for Eslava et al. (2004) is the Colombian Annual Manufactur-
ers Survey (AMS) by DANE (National Statistical Institute). I discuss further details about the
productivity estimates below in Section 5.2.1.
The import demand elasticity measure is based on the structural estimates in Kee, Nicita and
Olarreaga (2004), which I combine with the GDP data from the World Development Indicators
(WDI), and import data from COMTRADE. The import demand elasticities are available only at
the 3-digit ISIC level.42
In order to obtain the TFP measure of the upstream industries, I employ the input-output
tables provided at the 3-digit ISIC level by Nicita and Olarreaga (2001), which were compiled from
version 4 of the Global Trade Analysis Project (GTAP) database. Excluding the inputs being used
from the own sector, the upstream measure is based on a combination of TFPs of the remaining
input sectors as weighted by their share of usage.
The variable definitions and sources are presented more in detail in the appendix (Section A.3).
In Table 4, I provide the summary statistics for all the variables I use in the estimations.
5.2.1 Productivity Estimates
The productivity estimates come from Eslava, Haltiwanger, Kugler and Kugler (2004). They
estimate total factor productivity (TFP) as the residual from the following production function for
each firm i = 1,...,N and period t = 1982,..., 1998
log Xit = b1 log Kit + b2 log Lit + b3 log Eit + b4 log Iit + log Ait (20)
where Kit, Lit, Eit, and Iit denote capital, labor (total employment hours), energy consumption,
and materials, respectively. An important concern in such an estimation is the simultaneity bias;
that is, productivity shocks may be correlated with the inputs. They correct for this bias by
considering a measure of downstream demand as an instrument for inputs along with regional
government expenditures and input prices. A great advantage of this dataset is that it involves
plant level input prices which have not been available to the other researchers in the field requiring
them to use non-parametric estimation techniques.43 Furthermore, the output measures commonly
42See the appendix for a discussion on how the import demand elasticity is computed (Section A.2).
43The methodology in these studies was developed by Olley and Pakes (1996), and advanced by Levinsohn and
Petrin (2003). They employ investment or intermediate inputs to control for the correlation between the input levels
and the unobserved firm-level productivity shocks.
20
used in the literature have usually been the firm revenue deflated by the the industry-level prices.
Thus, within-industry price differences (e.g. due to different markups) have been part of the output
and productivity estimates of such studies, potentially biasing their results.
5.3 Estimation Results
Before moving on to the results, let us first observe the simple correlations of total factor productiv-
ity and tariffs. The overall correlation coefficient for log Ait and log it in the whole sample of 920
observations is -0.222. This is significant at the 1% level and can be observed graphically in Figure
5 as well. In Table 3, I present the correlation matrix for all the combinations of log Ait and log it
across years. The two variables again have a relatively small negative correlation for the most part
which is insignificant for certain years such as 1992 through 1994. These relationships also appear
in Figure 6 which includes plots of TFP versus tariffs by year. In Table 3, it is interesting to note
that the two variables can be concurrently and also intertemporally correlated. This is one reason
why using lagged tariff rates may not get around the endogeneity problem of tariffs with respect to
productivity. Topalova (2004) notes a similar pattern in the Indian data for the 1997-2001 period
and excludes this period from her analysis due to her concern about endogeneity.
However, we cannot establish a causal relationship between tariff protection and productivity
with these crude observations alone. We need to control for the other important variables as
required by the theory and tackle the endogeneity issues. In this section, I show how productivity
influences tariffs after I control for the endogeneity of productivity. Later in Section 6, I estimate
a system of equations related to this setup and show that accounting for the effect of productivity
on tariffs may strengthen the positive impact of trade reform on productivity.
As noted in Section 5.1, the two right-hand-side variables--the inverse import penetration to
import demand elasticity ratio, and total factor produtivity--in the tariff regressions are portentially
endogenous. I use instrumental variables to address this problem. More specifically, I employ the
two-step efficient generalized method of moments (henceforth IV-GMM) estimator with either fixed
effects or first differences for my unbalanced panel. This methodology is more efficient than regular
instrumental variables in the presence of heteroskedasdicity of unknown form due to its use of
an optimal weighting matrix (Cragg 1983). A Pagan-Hall (1983) test confirms the presence of
heteroskedasticity in the data and further justifies the use of the IV-GMM methodology.
In Table 5, I present the main estimation results. In column 1, we have the estimates for equa-
21
tion (18). As predicted by theory, tariff rates depend positively on the inverse import penetration
to import demand elasticity ratio (Qit/Mitit) and positively on total factor productivity (Ait).
The coefficients for the two main variables, 1 and 2, are positive and statistically significant at
the 1% level. The unilateral liberalization variable, UNILIBt, takes out the common reduction
in the tariffs after 1990, and it is significant and negative as expected. In column 2, I provide the
estimates for the fist variant of equation (18). Here, we take into account the variation in the data
by dividing it into three periods as opposed to imposing a one time major decline in the tariffs.
The two intercept-shifters, REFt (period dummy for 1990-1992) and POSTREFt (period dummy
for 1993-1998), control for the common decline in tariffs across sectors relative to the 1983-1989
period and come out negative and significant. Thus, the results are in line with the ones in column
1. In column 3, we have the estimates for the second variant of equation (18) that allows for
further variation across time with the year effects and captures the gradual decline in tariffs. Both
1 and 2 are still positive and statistically significant at the 1% level in columns 2 and 3. The
year dummies in column 3 are jointly significant just like the industry fixed effects are in all three
equations.
A positive coefficient on Qit/Mitit, such that tariffs are inversely related to import penetration
and import demand elasticity is a result consistent with the previous findings in the empirical
political economy literature (such as Gawande and Bandyopadhyay 2000 for the U.S., Mitra et al.
2002 for Turkey, and Karacaovali and Lim~ao 2005a for the EU). A positive coefficient for Ait, that
is more productive sectors receive higher tariff protection, complements this result and confirms my
major theoretical prediction. This result is also important because none of the earlier researchers
separate the size effect into Ait and Qit. Moreover, I account for the exogenous unilateral liberal-
ization shock common across sectors in all specifications so there is no doubt that political economy
does matter for the sectoral variation in tariffs. Therefore, endogeneity of tariffs with respect to
productivity is a prevailing problem when researchers plainly regress tariffs on productivity.
In Table 6, I provide the estimates of equation (18) and its variants that measure the effect of
yearly productivity changes on tariff changes with the first differenced data. The methodology is
still IV-GMM and I employ the first differences of each instrument from Table 5. The sample now
reduces to 1988-1998 given the gaps in the data. In column 1, 1 and 2 have the expected signs
but are not significant. This result is not surprising given that UNILIBt fails to recognize the
gradual decline in tariffs and acts as a single year effect for 1990. I correct for this by estimating
22
equation (18) and replacing UNILIBt with a common term for the 1990-1992 period (REFt)
during which the liberalization took place step by step (column 2). In this version, 1 becomes
significant at the 10% level and 2 at the 5%. REFt has a negative and significant (at the 1%
level) coefficient capturing the common downward trend. In column 3, I further allow for a common
constant term on top of REFt recognizing the small changes in the other years, and both 1 and
2 become significant at the 5% level. These results indicate that due to political economy, the
extent of liberalization is smaller for the sectors with a smaller reduction or a higher increase in
their productivity as compared to similar sectors. Note that in all differenced results, 1 and 2
are statistically identical, which is predicted by the model. In levels, this may not occur because
of fixed effects.
Although the theoretical section involves protection through tariffs, I repeat the specifications
in Table 5 and Table 6 with the effective rates of protection (ERP) in order to see whether the
results hold with a different measure of protection. Effective rates are based on value added and
essentially take into account the effect of tariffs on the inputs as well. ERP data are provided by the
National Planning Department of Colombia (DNP) and I am limited by their computations since
I do not have the detailed data to calculate them myself. As can be observed from Figure 1 and
Table 4, the effective rates are higher than the regular tariff rates but otherwise display a similar
trend. I exclude the three sectors44 that exhibit negative ERP (in levels not logs), because it is
hard to argue that these sectors are indeed protected. In Table 7, I repeat the specifications from
Table 5 and the results appear to be totally consistent. The only difference is that the significance
levels for the main variables are lower, and the constant term becomes insignificant in columns 2
and 3. The same arguments apply to the figures in Table 8 which are the replicas of the estimates
from Table 6 with ERP. The results are again qualitatively similar but less significant.
In the next section, I provide specification tests and some sensitiviy analysis for the main
estimations I covered. Then, in Section 6, I discuss how accounting for the endogeneity of tariff
policy, as implied by my theoretical and empirical results, may improve the estimates of the effect
of trade reform on productivity.
44The excluded sectors are: a) ISIC 3122, manufacture of prepared animal feeds; b) ISIC 3512, manufacture of
fertilizers and pesticides; c) ISIC 3822, manufacture of agricultural machinery and equipment.
23
5.4 Robustness and Specification Tests
In Table 9, I examine the effect of past productivity on current tariffs to check whether policy
implementation occurs with a one period lag although it is not part of the model. I employ one
period lags of scale and upstream TFP as instruments for the lag of productivity, and hence repeat
the specifications in Table 5 with log Ait-1 instead of log Ait. I find that more productivity yesterday
calls for more protection today in all three specifications. However, precaution is required while
interpreting this result since it might be picking up the persistence in tariffs as well.
In tables 10 and 11, I present the biased ordinary least squares (OLS) results for comparison.
In Table 10, I provide the estimates of equation (18), and in Table 11 the estimates of equation
(19) with OLS using both tariffs and effective rates of protection. The OLS coefficients have the
same signs as the IV-GMM estimates but they are smaller. In addition to that, log(Qit/Mitit) and
log(Qit/Mitit) have significant coefficients in all specifications while the coefficients for logAit
and log Ait are insignificant in all except the one for log Ait in Table 11, column 2.
I confirm the endogeneity of log(Qit/Mitit) and log Ait econometrically through a Durbin-Wu-
Hausman endogeneity test, which further justifies the use of instrumental variables instead of OLS.
Furthermore, the Hansen-Sargan test of overidentifying restrictions indicate that our instruments
are valid, that is they are uncorrelated with the error term and correctly excluded from the es-
timated equations. The probability value for the null hypothesis that the instruments are valid,
range from 0.144 to 0.933 for the main specifications presented in Table 5. The Hansen-Sargan test
probability values are presented in the last row of each relevant table and they have high values,
as desired, for the estimations in first differences (Table 6) as well. The tests are not strong for the
ERP specifications, but given the endogeneity and good performance with tariffs, it is prudent to
keep these instruments and ensure comparability with the tariff results.
In Table 12, I report the first stage regressions for the main tariff specification in Table 5,
column 1, where we see that all the instruments are jointly significant and the regressions have
a high explanatory power. I also find that the results are not driven by any specific instrument
which I check by excluding each one at a time.45 In Table 13, we have the first stage regressions for
the main first-differenced specification (Table 6, column 1) which are not as strong as the ones in
Table 12 in terms of the explanatory power but all the instruments are still jointly significant. The
partial R-squared values based on Shea (1997) indicate that the instruments for log Ait explain a
45Including labor share as an additional instrument also does not change the results qualitatively but lowers the
probability value of the Hansen-Sargan test.
24
substantial fraction of its variation. The same is not true for log(Qit/Mitit) and the first-differenced
equations.
Figure 1 indicates that in Colombia the major trade liberalization era started in 1990 and
continued until 1992 where new persistently lower levels of tariffs were reached. Given the restric-
tiveness of UNILIBt by construction, I allowed it to take 1991 instead of 1990 as the cutoff point
as well and the results remain robust to this different cutoff value. Furthermore, when UNILIBt
is excluded, the coefficient magnitudes rise but the results carry through.
6 Endogeneity Bias and the Effect of Tariffs on Productivity
The theoretical and empirical results I presented in the earlier sections indicate that we should
be worried about the endogeneity of trade policy with respect to productivity. If the researchers
do not account for the endogeneity, their estimates of the trade policy effects on productivity will
be biased. However, it is hard to tell the direction and magnitude of the endogeneity bias unless
the system is very simple.46 Once we have other regressors in the system, the correlations among
them do not permit us to make any predictions about the bias. Therefore, I illustrate how the bias
might be working with a system of equations below.
In constructing this system, I partly rely on the setup of my estimations in the previous section.
On the other hand, I do not have a structural equation showing how productivity depends on tariffs
so I just try to keep it similar to the estimations in the earlier empirical literature. I model the
tariff and inverse import penetration equations as before and add a third equation for productivity
as follows
log Ait = 1 + 2 log it + (log Z1it)3 + µ1i4 + Z2t5 + 1it (21a)
log it = 1 + 2 log Ait + 3 log(Qit/Mitit) + µ2i4 + t5 + 2it (21b)
log(Qit/Mitit) = 1 + 2 log it + (log Z3it)3 + 3it (21c)
where 1it, 2it, and 3it are the error terms for sector i = 1,...,N at period t = 1,...,T. µ1i and
46Suppose that we have the following two equations that relate tariffs and productivity: (1) log Ait = a1+a2 log it+
w1it and (2) logit = b1 + b2 logAit + w2it where w1it and w2it are mean zero error terms with constant variances
2w1 and 2w2. Then assuming that the covariance between the two error terms is zero, i.e. cov(w1it,w2it) = 0, we
obtain cov(w1it,log it) = b2
have cov(w1it,log it) > 0. If we estimate a2 with OLS, ignoring equation (2), and get b2
1-b2a2 2w1. If the true values of a2 and b2 are such that a2 < 0 and b2 > 0, then we
a < 0 which is positively
correlated with a2, we would have an upward bias, hence underestimate a2.
25
µ2i are 1×(N -1) vectors of industry dummies, and t is a 1×(T -1) vector of year dummies. Z1it
and Z3it are 1×2 vectors of control variables at the 4-digit ISIC sector level, whereas Z2t is a 1×2
vector of economy-wide controls.47 Z1it includes the scale measure and upstream TFP, whereas
Z3it includes the capital to output ratio and materials prices. Note that these industry level control
variables are precisely the instruments I used in the instrumental variables estimations in Section
5 so they are expected to be exogenous. The other advantage of these controls is that I get a
consistent framework with the rest of my estimations. Z2t includes GDP growth and inflation to
control for the macro changes in the economy that might affect the productivity in all the sectors.
My estimates of equation (21a) are at the 4-digit industry level. This limitation precludes any
direct comparison between my estimates and the ones in the recent firm-level studies. However,
after I estimate the system with three-stage least squares (3SLS), I compare these results with the
simple OLS estimates of equation (21a) that ignore the endogeneity and get the chance to test
whether accounting for endogeneity improves the results within my dataset.
In Table 14, I provide the comparative results of estimating the whole system with 3SLS and
estimating equation (21a) with OLS only. As in line with the earlier literature, a negative OLS
estimate of the coefficient for tariffs in equation (21a), i.e. 2 < 0, indicates that productivity is
inversely related to tariffs. The 3SLS regression results not only confirm this finding but also show
b
that the positive effect of lower tariffs on productivity grows slightly stronger (by 1.5%) when I
account for the endogeneity of tariffs. The 3SLS results from equation (21b) are similar to the
findings I present in Table 5: more productive sectors receive higher protection, and tariffs are
inversely related to import penetration and import demand elasticity. In Table 15, I replicate
the estimations from Table 14 by using effective rates of protection instead of tariffs and find
identical results. However, this time the positive impact of liberalization on productivity is larger
by 17% when we estimate the whole system. These findings indicate that the trade policy effects
on productivity might be underestimated when endogeneity is not accounted for.
7 Concluding Remarks
I show, both theoretically and empirically, that we should be concerned about the endogeneity of
trade policy with respect to productivity. This has been neglected for the most part in the recent
47Note that 1, 2, 1, 2, 3, 1, and 2 are scalars. 3, 5, and 3 are 2×1, 4 and 4 are (N -1)×1 vectors,
and 5 is a (T - 1) × 1 vector.
26
empirical literature. Studies that investigate the effect of trade policy on productivity often argue
that the exogeneity of the trade liberalization shock helps to identify a linkage without worrying
too much about the endogeneity of tariffs or other forms of trade policy. I account for such an
argument in my theoretical and empirical models, and still obtain tariffs to be endogenous.
I employ a basic political economy of trade protection model and also introduce two different
channels of protection that lead to unilateral liberalization once they are removed. The extra
channels are meant to capture the perceived benefit of protection to the governments even in the
absence of political economy concerns. I first keep the liberalization channel very simple to ensure
that my results are not driven by any specific assumptions or complications and then give more
structure to it by modeling a learning-by-doing argument. The main result from my theoretical
models, simple yet compelling, is that despite an exogenous unilateral trade liberalization shock
that is common across sectors, we obtain a differentiated effect across sectoral protection based on
productivity. Based on my theory, I predict that more productive sectors receive more protection
and that the extent of liberalization is less for sectors that experience a higher productivity increase.
Next, I test and confirm these theoretical results using production, trade, and tariff data at
the 4-digit ISIC industry level for Colombia between 1983 and 1998. I keep all of my estimations
closely related to my theory and account for all the potential endogeneity problems by using relevant
instruments and methodologies.
Finally, I estimate a system of equations and show that by not accounting for the endogeneity
of trade policy with respect to productivity, we might underestimate the positive impact of trade
reform on productivity. Thus, correcting for the endogeneity bias does not overturn the results
in the early empirical literature but makes them somewhat stronger for Colombia which would be
interesting to test for different countries as well.
As a natural extension to this paper, it would be useful to carefully model the effect of tariffs
on productivity and obtain a more structural simultaneous equations model considering all the
interdependencies between tariffs, productivity, and their determinants.
27
A Appendix
A.1 Derivations
Equation (2)
We maximize equation (1) with respect to i to obtain the following first order condition
G
i = -Di(i) + AiQi(i) + Mi(i) + iMi0(i)
= ( - 1)AiQi(i) + iMi0(i) (22)
Equating to zero and solving for i, and then dividing both sides of this expression by pw = 1 and
i
using the following elasticity definition i -Mi0pw/Mi yields equation (2).
i
Equation (6)
In order to obtain equation (6), we implicitly differentiate the specific tariff version of i as
expressed in equation (5) with respect to and use the linearity assumption for Mi (so that
Mi00 = 0), the restriction 0 < < 2 - Di0/AiQ0i and concavity of i(i) (i.e. 0i > 0, 00i < 0)
di 0i
> 0 (23)
d = -Mi0(i) + ( - 1)AiQ0i(i) + 00i (i)
Equation (13)
The first order condition for a solution to equation (11) is
Et(Git + Git+1)
it it eit+1(it)Dit+1(. )
it = ( - 1)itZQ1+(eit ) + itMit0 - it
+1
+ it+1(itQit(it))Qit+1(it+1)dit+1 (24)
it 0
+ eit+1(it)Mit+1(. ) + eit+1(it)eit+1(it)Mit0
it it +1
which after a few steps of manipulation becomes
Et(Git + Git+1)
it it
it = ( -·1)0(QQ(it()it+ itMit0
it
+ (itQit(it))itQ0it
))
it Z0 1+eit
+1it+1(.)Qit+1(it+1)dit+1¸ (25)
+ eit+1(it) Ł(
it - 1)it+1(.)Qit+1(eit+1(it)) + eit+1(it)Mit0 +1 ¤
Using equation (12) to substitute in for eit+1(it) and employing the definition in equation (14),
28
the first order condition simplifies to
Et(Git + Git+1)
(26)
it = ( - 1)itQit(it) + itMit0 + i = 0
Dividing both sides of equation (26) by pw = 1 and using the same elasticity term i as described
i
above yields equation (13).
Equations (16a), (16b), and (16c)
Employing the functional form given in equation (15), the LBD term can now be expressed as
i = nnitQ0it(Qit(it))n-1 Z0 1+eit
+1
it+1Qit+1(it+1)dit+1 (27)
The relationships in equations (16a), (16b), and (16c) are then obtained by plugging equation
(27) in equation (26) and differentiating it in equation (26) with respect to it+1 (implicitly), it
(partially), and it (implicitly). For it+1 we get
dit nnitQ0it(Qit(it))n-1 1+eit+1Qit+1dit+1
dit+1|n<1 = -Mit0 > 0
+ n(n - 1)nit(Q0it)2(Qit)n-2 R01+eit
R0+1it+1Qit+1dit+1 + ( - 1)itQ0it (28)
Similarly for it we get the following two
it|n<1 n2nit-1Q0it(Qit)n-1
it = - R0
1+eit+1it+1Qit+1dit+1 + ( - 1)Qit > 0 (29)
Mit0
> 0
dit
dit|n<1 = -Mit0 + n(n - 1)nit(Q0it)2(QitR)n-2
n2nit-1Q0it(Qit)n-1 1+eit+1
0
R01+eit
it+1Qit+1dit+1 + ( - 1)Qit
+1 it+1Qit+1dit+1 + ( - 1)itQ0it (30)
Equation (17)
The actual tariff in period t +1 is similar to the one in equation (12) but now its terms are not
dependent on Xit, because I assume that the LBD process is realized to be a false perception:
it+1Qit+1(it+1)/Mit+1(it+1)
it+1 = ( - 1) (31)
it+1(it+1)
Now, by using equation (13) and equation (31), we can express the difference in the tariff rates
29
between the two periods as
1
it+1 = it+1|n=0 - it|n>0 = -Mi0( - 1)(it+1Qit+1 - itQit)
1 1+eit+1
+ Mi0nnitQ0it(Qit)n-1 Z0 it+1Qit+1dpit+1 (32)
Equation (17) is then obtained by implicitly differentiating equation (32) with respect to it+1.
dit+1
dit+1|it,it = -Mi0(. ( - 1)Qit+1(it+1) > 0 (33)
) + ( - 1)it+1Q0it+1(it+1)
30
A.2 Import Demand Elasticity
In the theory section, I define the import demand elasticity, i, as Mi0pw/Mi but traditionally and
i
in the empirical data it is evaluated at the domestic prices, not the world prices. I take this into
account in obtaining the elasticity adjusted inverse import penetration ratio, given the fact that
ouput value is evaluated at the domestic prices, whereas imports are evaluated at the world prices.
Therefore,
Xi/Mi Xi/Mi piXi/pwMi
= = i (34)
i Mi0pw/Mi
i Mi0pi/Mi
I use the structural estimates from Kee, Nicita and Olarreaga (2004) to compute the import
demand elasticities. Based on Kee et al. (2004), I obtain the import demand elasticity for sector i
as
ai
it = (35)
sit+ sit - 1
where sit is the negative of the imports to GDP ratio and ai is an estimated structural price
parameter from a GDP function.
31
A.3 Variable Definitions and Sources
Name Definition Source
it Advalorem tariff rate (%): Obtained at the 8-digit National Planning Department
product level ("Nabandina" code) and aggregated (DNP), Colombia
to the 4-digit ISIC level by simple averaging
it
eff Effective rate of protection (%): Obtained at the 8- National Planning Department
digit product level ("Nabandina" code) and (DNP), Colombia
aggregated to the 4-digit ISIC level by simple
averaging
Xit Output values in 1000 USD at the 4-digit ISIC UNIDO, Industrial Statistics
level Database
Mit Import values in 1000 USD at the 4-digit ISIC COMTRADE, United Nations
level Statistics Division
it Import demand elasticity at the 3-digit ISIC level: Structural estimates (Kee et al.
obtained by combining import and GDP data with 2004), GDP (World
estimated structural price parameters. Development Indicators,
World Bank), imports
(COMTRADE).
Ait Total factor productivity (TFP): Obtained at the Eslava, Haltiwanger, Kugler
firm level by estimating production function and Kugler (2004)
residuals with a 2SLS model. Aggregated from the
firm to the 4-digit ISIC level by using production
shares as weights.
Capital Capital stock series obtained at the firm level using Eslava, Haltiwanger, Kugler
Share fixed assets, gross investment, "observed" and Kugler (2004)
depreciation rates, and a gross capital formation
deflator. The ratio of capital stock to output is then
aggregated to the 4-digit ISIC level by using firms'
production shares as weights.
Materials Obtained at the firm level using Tornqvist indices Eslava, Haltiwanger, Kugler
Prices which are aggregated to the 4-digit ISIC level by and Kugler (2004)
using firms' production shares as weights.
Scale The ratio of total value added to the number of Eslava, Haltiwanger, Kugler
firms in a given 4-digit ISIC sector. and Kugler (2004)
Upstream Using the input-output tables at the 3-digit ISIC Input-output tables (Nicita and
TFP level, I exclude the inputs being used from the own Olarreaga 2001, originally
sector, and obtain the upstream measure based on a from Global Trade Analysis
combination of TFPs of the remaining input Project), TFP (Eslava et al.
sectors as weighted by their share of usage. 2004).
GDP The annual percentage change in the GDP World Development
Growth (constant 2000 US dollars) Indicators (WDI), World Bank
Inflation Annual percentage change in the GDP deflator World Development
Indicators (WDI), World Bank
32
References
[1] Baldwin, Richard E. (1985), The Political Economy of U.S. Import Policy, (Cambridge, MA:
MIT Press.)
[2] Baldwin, Robert E. (2002), "Entry and Asymmetric Lobbying: Why Governments Pick
Losers," National Bureau of Economic Research Working Paper No. 8756, February.
[3] Caves, R. E. (1976), "Economic Models of Political Choice: Canada's Tariff Structure," Cana-
dian Journal of Economics, 9(2), 278-300.
[4] Cragg, J. (1983), "More Efficient Estimation in the Presence of Heteroskedasticity of Unknown
Form," Econometrica, 51, 751-763.
[5] Edwards, S. (1997), "Trade Liberalization Reforms and the World Bank," American Economic
Review, Papers and Proceedings of the Hundred and Fourth Annual Meeting of the American
Economic Association, 87(2), 43-48.
[6] Edwards, S. (2001), The Economics and Politics of Transition to an Open Market Economy:
Colombia, (Paris: OECD.)
[7] Eslava, M., J. Haltiwanger, A. Kugler and M. Kugler (2004), "The Effects of Structural
Reforms on Productivity and Profitability Enhancing Reallocation: Evidence from Colombia,"
Journal of Development Economics, 75(2), 333-371.
[8] Fernandes, A. M. (2003), "Trade Policy, Trade Volumes and Plant-Level Productivity in
Colombian Manufacturing Industries" Policy Research Working Paper No. 3064, (Washington,
DC: World Bank.)
[9] Ferreira, P. C. and G. Facchini (2005), "Trade Liberalization and Industrial Concentration:
Evidence from Brazil," Quarterly Review of Economics and Finance, 45, 432-446.
[10] Ferreira, P. C. and J. L. Rossi (2003), "New Evidence from Brazil on Trade Liberalization and
Productivity Growth," International Economic Review, 44(4), 1383-1405.
[11] Gawande, K. and U. Bandyopadhyay (2000), "Is Protection for Sale? Evidence on the
Grossman-Helpman Theory of Endogenous Protection," Review of Economics and Statistics,
82(1), 139-152.
[12] Goldberg, P. K. and G. Maggi (1999), "Protection for Sale: An Empirical Investigation,"
American Economic Review, 89(5), 1135-1155.
[13] Grossman, G. M. and E. Helpman (1994), "Protection for Sale," American Economic Review,
84(4), 833-850.
[14] Grossman, G. M. and E. Helpman (1995), "Technology and Trade," in G. Grossman and K.
Rogoff (eds.), Handbook of International Economics, Vol. III, 1279-1337 (Amsterdam: North-
Holland.)
[15] Grossman, G. M. and H. Horn (1988), "Infant-Industry Protection Reconsidered: The Case
of Informational Barriers to Entry," Quarterly Journal of Economics, 103(4), 767-787.
33
[16] Harrison, A. (1994), "Productivity, Imperfect Competition and Trade Reform: Theory and
Evidence," Journal of International Economics, 36(1-2), 53-73.
[17] Harrison, A. (1996), "Openness and Growth: A Time-Series, Cross-Country Analysis for De-
veloping Countries" Journal of Development Economics, 48, 419-447.
[18] Helpman, E. (1997), "Politics and Trade Policy," in D. M. Kreps and K. F. Wallis (eds.), Ad-
vances in Economics and Econometrics: Theory and Applications, (New York, NY: Cambridge
University Press.)
[19] Johnson, H. G. (1955), "Economic Expansion and International Trade," Manchester School of
Economic and Social Studies, 23, 95-112.
[20] Karacaovali, B. and N. Lim~ao (2005a), "The Clash of Liberalizations: Preferential vs. Mul-
tilateral Trade Liberalization in the European Union," Policy Research Working Paper No.
3493, (Washington, DC: World Bank.)
[21] Karacaovali, B. and N. Lim~ao (2005b), "The Clash of Liberalizations: Preferential vs. Mul-
tilateral Trade Liberalization in the European Union," CEPR Discussion Paper No. 4973,
(London: Centre for Economic Policy Research.)
[22] Kee, H. L., A. Nicita and M. Olarreaga (2004), "Import Demand Elasticities and Trade Dis-
tortions," Policy Research Working Paper No. 3452, (Washington, DC: World Bank.)
[23] Krishna, P. and D. Mitra (1998), "Trade Liberalization, Market Discipline and Productivity
Growth: New Evidence from India," Journal of Development Economics, 56(2), 447-462.
[24] Krueger, A. O. and B. Tuncer (1982), "An Empirical Test of the Infant Industry Argument,"
American Economic Review, 72, 1142-1152.
[25] Levinsohn, J. and A. Petrin (2003), "Estimating Production Functions Using Inputs to Control
for Unobservables," Review of Economic Studies, 70, 317-341.
[26] Magee, S., W. Brock and L. Young (1989), Black Hole Tariffs and Endogenous Policy Theory:
Political Economy in General Equilibrium, (Cambridge: Cambridge University Press.)
[27] Marvel, H. P. and E. J. Ray (1983), "The Kennedy Round: Evidence on the Regulation of
International Trade in the United States," American Economic Review, 73(1), 190-197.
[28] Mitra, D., D. D. Thomakos and M. A. Ulubasoglu (2002), "Protection for Sale in a Developing
Country: Democracy vs. Dictatorship," Review of Economics and Statistics, 84(3), 497-508.
[29] Melendez, M., K. Seim and P. Medina (2003), "Productivity Dynamics of the Colombian
Manufacturing Sector," Documento CEDE 2003-23, Universidad de los Andes.
[30] Melitz, M. J. (2003), "The Impact of Trade on Intra-Industry Reallocations and Aggregate
Industry Productivity," Econometrica, 71(6), 1695-1725.
[31] Muendler, M. (2004), "Trade, Technology, and Productivity: A Study of Brazilian Manufac-
turers, 1986-1998," mimeo, University of California, San Diego.
[32] Nicita, A. and M. Olarreaga (2001), "Trade and Production, 1976-99," Policy Research Work-
ing Paper No. 2701, (Washington, DC: World Bank.)
34
[33] Olley, S. and A. Pakes (1996), "The Dynamics of Productivity in the Telecommunications
Equipment Industry," Econometrica, 64(6), 1263-1297.
[34] Olson, M. (1965), The Logic of Collective Action: Public Goods and the Theory of Groups,
(Cambridge, MA: Harvard University Press.)
[35] Ozler S. and K. Yilmaz (2003), "Does Trade Liberalization Improve Productivity? Plant Level
Evidence from Turkish Manufacturing Industry," in Regional Development: Reality or Myth?
Selected Papers from the Ninth ERF Annual Conference, (Cairo: Economic Research Forum.)
[36] Pack, H. (1988), "Industrialization and Trade," in H. Chenery and T.N. Srinivasan (eds.)
Handbook of Development Economics, Volume I, 333-380 (Amsterdam: North-Holland.)
[37] Pagan, A. R. and D. Hall (1983), "Diagnostic Tests as Residual Analysis," Econometric Re-
views, 2(2), 159-218.
[38] Pavcnik, N. (2002), "Trade liberalization, Exit, and Productivity Improvements: Evidence
from Chilean Plants," Review of Economic Studies, 69(1), 245-76.
[39] Roberts, M. J. (1996), "Colombia, 1977-85: Producer Turnover, Margins, and Trade Expo-
sure," in M. J. Roberts and J. R. Tybout (eds.), Industrial Evolution in Developing Countries:
Micro Patterns of Turnover, Productivity, and Market Structure, (New York, NY: Oxford Uni-
versity Press), 227-258.
[40] Rodriguez, F. and D. Rodrik (2001), "Trade Policy and Economic Growth: A Skeptic's Guide
to the Cross-National Evidence," in B. Bernanke and K. Rogoff (eds.), NBER Macroeconomics
Annual 2000, (Cambridge, MA: MIT Press.)
[41] Schor, A. (2004), "Heterogeneous Productivity Response to Tariff Reduction. Evidence from
Brazilian Manufacturing Firms," Journal of Development Economics, 75, 373-396.
[42] Shea, J. (1997), "Instrument Relevance in Multivariate Linear Models: A Simple Measure,"
Review of Economics and Statistics, 49(2), 348-352.
[43] Srinivasan, T.N. and J. Bhagwati (2001), "Outward Orientation and Development: Are Re-
visionists Right?" in D. Lal and R. Snape (eds.), Trade, Development and Political Economy:
Esays in Honour of Anne O. Krueger, (Basingstoke, UK: Palgrave.)
[44] Topalova, P. (2004), "Trade Liberalization and Firm Productivity: The Case of India," IMF
Working Paper WP/04/28, (Washington, DC: International Monetary Fund.)
[45] Tovar, P. (2004), "The Effects of Loss Aversion on Trade Policy and the Anti-Trade Bias
Puzzle," mimeo, University of Maryland.
[46] Trefler, D. (1993), "Trade Liberalization and the Theory of Endogenous Protection: An Econo-
metric Study of U.S. Import Policy," Journal of Political Economy, 101(1), 138-160.
[47] Tybout, J. (1991), "Linking Trade and Productivity: New Research Directions," The World
Bank Economic Review, 6(2), 189-211.
[48] Tybout, J. (1996), "Heterogeneity and Productivity Growth: Assessing the Evidence" in M.
J. Roberts and J. R. Tybout (eds.), Industrial Evolution in Developing Countries: Micro
Patterns of Turnover, Productivity, and Market Structure, (New York, NY: Oxford University
Press), 43-72.
35
[49] Tybout, J. and M. Westbrook (1995), "Trade Liberalization and the Dimensions of Efficiency
Change in Mexican Manufacturing Industries," Journal of International Economics, 39(1-2),
53-78.
36
FIGURE 1. Average Tariffs and Effective Rates of Protection in Colombia 1983-1998
0.9
0.8 ERP (%)
0.7
0.6
0.5
Tariff (%)
0.4
0.3
0.2
0.1
0
1983 1985 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998
Year
Source: DNP and author's own calculations.
FIGURE 2. Histogram of the Percentage Decline in Tariffs between 1983 and 1995 at the 4-
digit ISIC Level
4
.0
3
.0
yti
ns 2
.0
De
1
.0
0
0 20 40 60
Tariff Reduction (%)
Note: Tariff reduction is calculated as [log(1+i1983)-log(1+ i1995)]*100
Source: DNP and author's own calculations.
37
FIGURE 3. Histogram of the Percentage Decline in Tariffs between 1988 and 1995 at the 4-
digit ISIC Level
6
.0
4
.0
ytisn
De
2
.0
0
0 10 20 30 40
Tariff Reduction (%)
Note: Tariff reduction is calculated as [log(1+i1988)-log(1+ i1995)]*100
Source: DNP and author's own calculations.
FIGURE 4. Histogram of the Percentage Decline in Tariffs between 1983-1989 average and
1992-1998 average at the 4-digit ISIC Level
5
.0
4
.0
3
ytisn .0
De 2
.0
1
.0
0
0 10 20 30 40
Tariff Reduction (%)
Note: Tariff reduction is calculated as [log(1+avgi1983-1988)-log(1+avgi1992-1998)]*100
Source: DNP and author's own calculations.
38
FIGURE 5. Tariffs and Productivity: Whole Sample
0
-1
)ffi
arT -2
g(
Lo
-3
-4
0 1 2 3 4
Total Factor Productivity
FIGURE 6. Tariffs and Productivity: By Year
1983 1985 1988 1989
0
-1
-2
-3
-4
1990 1991 1992 1993
)ffi 0
-1
arT -2
g( -3
Lo
-4
1994 1995 1997 1998
0
-1
-2
-3
-4
0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4
Total Factor Productivity
Graphs by year
39
TABLE 1. Spearman's Rank Correlation Matrix for Tariffs over Time
1983 1985 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997
1985 0.928 1
1988 0.861 0.953 1
1989 0.862 0.954 0.997 1
1990 0.806 0.925 0.952 0.954 1
1991 0.733 0.823 0.830 0.847 0.893 1
1992 0.688 0.764 0.771 0.768 0.795 0.866 1
1993 0.700 0.761 0.773 0.770 0.779 0.861 0.993 1
1994 0.688 0.763 0.777 0.778 0.797 0.876 0.997 0.991 1
1995 0.720 0.794 0.801 0.807 0.828 0.898 0.986 0.978 0.989 1
1996 0.727 0.847 0.874 0.882 0.897 0.857 0.885 0.874 0.887 0.922 1
1997 0.732 0.852 0.882 0.882 0.907 0.859 0.887 0.876 0.889 0.925 0.999 1
1998 0.724 0.845 0.874 0.882 0.899 0.856 0.888 0.877 0.889 0.924 0.998 0.999
Note: t stands for the average 4-digit ISIC level tariff in year t.
40
TABLE 2. Summary Statistics for Tariffs over Time
Observations Mean Standard Coefficient of Minimum Maximum
Deviation Variation
1983 78 0.427 0.221 0.516 0.09 1.15
1985 78 0.377 0.148 0.393 0.059 0.70
1988 78 0.347 0.155 0.448 0.07 0.70
1989 75 0.344 0.155 0.451 0.07 0.70
1990 76 0.297 0.115 0.386 0.07 0.50
1991 74 0.211 0.093 0.442 0.016 0.35
1992 68 0.134 0.045 0.334 0.05 0.20
1993 65 0.135 0.046 0.343 0.05 0.20
1994 63 0.136 0.045 0.333 0.05 0.20
1995 65 0.136 0.046 0.334 0.043 0.20
1996 67 0.139 0.046 0.333 0.048 0.20
1997 66 0.140 0.046 0.332 0.048 0.20
1998 67 0.140 0.045 0.323 0.048 0.20
Note: t stands for the average 4-digit ISIC level tariff in year t.
41
TABLE 3. Correlation Matrix for Tariffs and Productivity over Time
log1983 log1985 log1988 log1989 log1990 log1991 log1992 log1993 log1994 log1995 log1996 log1997 log1998
logA1983 -0.187 -0.153 -0.064 -0.071 -0.049 0.003 -0.100 -0.117 -0.073 -0.088 -0.007 -0.014 0.007
logA1985 -0.233** -0.179 -0.108 -0.116 -0.104 -0.076 -0.085 -0.097 -0.076 -0.108 0.012 0.008 0.028
logA1988 -0.220* -0.237** -0.269** -0.211* -0.249** -0.198* -0.144 -0.151 -0.158 -0.226* -0.241** -0.244** -0.239*
logA1989 -0.208* -0.217* -0.197* -0.199* -0.233** -0.206* -0.141 -0.125 -0.129 -0.196 -0.219* -0.223* -0.220*
logA1990 -0.223* -0.241** -0.235** -0.229** -0.260** -0.253** -0.138 -0.135 -0.146 -0.198 -0.233* -0.231* -0.233*
logA1991 -0.177 -0.195* -0.238** -0.224* -0.234** -0.268** -0.127 -0.124 -0.132 -0.203 -0.205 -0.204 -0.208*
logA1992 -0.145 -0.156 -0.158 -0.162 -0.163 -0.251** -0.039 -0.054 -0.074 -0.199 -0.113 -0.112 -0.115
logA1993 -0.170 -0.193 -0.200 -0.204 -0.229* -0.265** -0.161 -0.148 -0.163 -0.201 -0.257** -0.258** -0.260**
logA1994 -0.195 -0.251** -0.263** -0.262** -0.308** -0.362*** -0.185 -0.161 -0.175 -0.218* -0.275** -0.271** -0.277**
logA1995 -0.179 -0.225* -0.239* -0.238* -0.269** -0.346*** -0.154 -0.132 -0.147 -0.227* -0.225* -0.225* -0.227*
logA1996 -0.180 -0.217* -0.245** -0.245** -0.257** -0.297** -0.141 -0.125 -0.144 -0.213* -0.241** -0.244** -0.247**
logA1997 -0.167 -0.194 -0.226* -0.231* -0.241* -0.294** -0.146 -0.126 -0.143 -0.220* -0.225* -0.232* -0.234*
logA1998 -0.206* -0.229* -0.272** -0.271** -0.294** -0.337*** -0.223* -0.205 -0.219* -0.265** -0.302** -0.307** -0.311**
Note: logt stands for the natural logarithm of the average 4-digit ISIC level tariff in year t, and logAt stands for the natural logarithm of the
average 4-digit ISIC level total factor productivity in year t
42
TABLE 4. Summary Statistics for All the Variables in the Estimations
Observations Mean Standard Minimum Maximum
Deviation
logit 920 -1.646 0.641 -4.107 0.140
logit 840 -0.093 0.228 -1.971 1.583
logit eff 902 -1.128 0.884 -4.294 1.556
logit eff 821 -0.113 0.526 -12.613 2.377
log(Qit/Mitit) 920 0.248 2.376 -6.139 11.470
log(Qit/Mitit) 840 -0.098 0.641 -6.025 3.201
logAit 920 1.508 0.590 0.091 4.097
logAit 840 0.031 0.270 -1.951 2.154
logCapital Share 920 -1.765 0.777 -5.549 0.598
logCapital Share 840 0.027 0.324 -3.714 1.771
logMaterials Prices 920 -0.067 0.267 -1.488 0.929
logMaterials Prices 840 -0.018 0.134 -0.777 1.338
logScale 920 11.726 1.489 5.595 16.264
logScale 830 0.050 0.528 -5.107 3.488
logUpstream TFP 920 1.523 0.140 1.206 2.096
logUpstream TFP 840 0.017 0.067 -0.205 0.490
GDP growth 920 3.445 1.621 0.570 6.042
Inflation 920 24.037 7.158 14.773 45.357
Notes: The tariff data are not available for 1982, 1986, and 1987 so we start out with 1310 4-digit ISIC
tariff lines. When we take into account the missing output figures (not present for the whole year of 1984),
the sample reduces to 1004 observations. Finally, considering the other missing observations on the right-
hand-side, the sample further declines to around 920 for the main estimations.
43
TABLE 5. The Effect of Productivity on Tariffs
(1) (2) (3)
log(Qit/Mitit) 0.390*** 0.606*** 0.864***
(1>0) (0.073) (0.142) (0.173)
logAit 0.271*** 0.386*** 0.551***
(2>0) (0.066) (0.097) (0.116)
UNILIBt -0.533***
(3<0) (0.065)
REFt -0.492***
(1<0) (0.035)
POSTREFt -0.306**
(2<0) (0.154)
Constant -1.536*** -1.884*** -2.186***
(0.190) (0.269) (0.296)
Year Effects No No Yes
Observations 920 920 920
Chi2-test p-val for all µi=0 a 0.000 0.000 0.000
Chi2-test p-val for all t=0 b n/a n/a 0.000
Hansen's J p-val c 0.144 0.180 0.933
Notes: (1) Standard errors are in parentheses. (2) *, **, and *** indicate significance at a 10%, 5%, and 1%
level, respectively. (3) The dependent variable is the natural logarithm of the advalorem tariff rate (logit).
(4) The predicted signs for the coefficients of the regressors are indicated in parentheses below them.
(5) All the regressions include 4-digit ISIC industry dummies as regressors but are not reported. (6) List of
the instruments (all in logs): Capital share, materials prices (deviated from the producer price index),
measure of scale economies (value added/number of firms), and the TFP of upstream sectors.
a"Chi2-test p-val for all µi=0" provides the probability value for the Chi-squared test of H0: All µi (industry
fixed effects) are jointly insignificant.
b "Chi2-test p-val for all t =0" provides the probability value for the Chi-squared test of H0: All t (year
effects) are jointly insignificant.
c "Hansen's J p-val" provides the probability value for the Hansen-Sargan test of overidentifying
restrictions for H0: Excluded instruments are uncorrelated with the error term and correctly excluded from
the estimated equation.
44
TABLE 6. The Effect of Productivity Differences on Tariff Differences
(1) (2) (3)
log(Qit/Mitit) 0.190 0.514* 0.449**
(1>0) (0.726) (0.290) (0.213)
logAit 0.231 0.519** 0.476**
(2>0) (0.658) (0.263) (0.209)
UNILIBt -0.120***
(3<0) (0.031)
REFt -0.185*** -0.231***
( <0) (0.069) (0.045)
Constant 0.031*
(0.016)
Observations 676 676 676
Hansen's J p-val a 0.572 0.929 0.958
Notes: (1) Standard errors are in parentheses. (2) *, **, and *** indicate significance at a 10%, 5%, and 1%
level, respectively. (3) The dependent variable is the one year change in the natural logarithm of the
advalorem tariff rate (logit). (4) The predicted signs for the coefficients of the regressors are indicated in
parentheses below them. (5) List of the instruments (all in logs): First differences of capital share, materials
prices (deviated from the producer price index), measure of scale economies (value added/number of firms),
and the TFP of upstream sectors. (6) All estimations allow for arbitrary intra-industry correlation over time.
a "Hansen's J p-val" provides the probability value for the Hansen-Sargan test of overidentifying
restrictions for H0: Excluded instruments are uncorrelated with the error term and correctly excluded from
the estimated equation.
45
TABLE 7. The Effect of Productivity on the Effective Rates of Protection
(1) (2) (3)
log(Qit/Mitit) 0.333*** 0.270* 0.303**
(1>0) (0.084) (0.142) (0.139)
logAit 0.261*** 0.190** 0.198**
(2>0) (0.070) (0.091) (0.096)
UNILIBt -0.598***
(3<0) (0.079)
REFt -0.500***
(1<0) (0.045)
POSTREFt -0.715***
(2<0) (0.165)
Year Effects No No Yes
Constant -0.415** -0.207 -0.066
(>0) (0.198) (0.363) (0.342)
Observations 887 887 887
Chi2-test p-val for all µi=0 a 0.000 0.000 0.000
Chi2-test p-val for all t=0 b n/a n/a 0.000
Hansen's J p-val c 0.004 0.001 0.000
Notes: (1) Standard errors are in parentheses. (2) *, **, and *** indicate significance at a 10%, 5%, and 1%
level, respectively. (3) The dependent variable is the natural logarithm of the effective rate of protection
(logit ). (4) The predicted signs for the coefficients of the regressors are indicated in parentheses below
eff
them. (5) All the regressions include 4-digit ISIC industry dummies as regressors but are not reported. (6)
List of the instruments (all in logs): Capital share, materials prices (deviated from the producer price index),
measure of scale economies (value added/number of firms), and the TFP of upstream sectors.
a"Chi2-test p-val for all µi=0" provides the probability value for the Chi-squared test of H0: All µi (industry
fixed effects) are jointly insignificant.
b "Chi2-test p-val for all t =0" provides the probability value for the Chi-squared test of H0: All t (year
effects) are jointly insignificant.
c "Hansen's J p-val" provides the probability value for the Hansen-Sargan test of overidentifying
restrictions for H0: Excluded instruments are uncorrelated with the error term and correctly excluded from
the estimated equation.
46
TABLE 8. The Effect of Productivity Differences on the Effective Rate of Protection
Differences
(1) (2) (3)
log(Qit/Mitit) 0.337 0.553 0.519*
(1>0) (0.838) (0.337) (0.269)
logAit 0.297 0.504* 0.490*
(2>0) (0.758) (0.294) (0.251)
UNILIBt -0.105***
(3<0) (0.040)
REFt -0.182** -0.223***
(<0) (0.089) (0.065)
Constant 0.032
(0.021)
Observations 652 652 652
Hansen's J p-val a 0.482 0.695 0.765
Notes: (1) Standard errors are in parentheses. (2) *, **, and *** indicate significance at a 10%, 5%, and 1%
level, respectively. (3) The dependent variable is the one year change in the natural logarithm of the
effective rate of protection (logit ). (4) The predicted signs for the coefficients of the regressors are
eff
indicated in parentheses below them. (5) List of the instruments (all in logs): First differences of capital
share, materials prices (deviated from the producer price index), measure of scale economies (value
added/number of firms), and the TFP of upstream sectors. (6) All estimations allow for arbitrary intra-
industry correlation over time.
a "Hansen's J p-val" provides the probability value for the Hansen-Sargan test of overidentifying
restrictions for H0: Excluded instruments are uncorrelated with the error term and correctly excluded from
the estimated equation.
47
TABLE 9. The Effect of Past Productivity on Current Tariffs
(1) (2) (3)
log(Qit/Mitit) 0.586*** 0.518*** 0.351***
(1>0) (0.116) (0.118) (0.101)
logAit-1 0.186** 0.168** 0.205***
(2>0) (0.088) (0.080) (0.067)
UNILIBt -0.361***
(3<0) (0.104)
REFt -0.494***
(1<0) (0.044)
POSTREFt -0.381***
(2<0) (0.139)
Constant -2.207*** -2.058*** -1.610***
(0.289) (0.293) (0.239)
Year Effects No No Yes
Observations 895 895 895
Chi2-test p-val for all µi=0 a 0.000 0.000 0.000
Chi2-test p-val for all t=0 b n/a n/a 0.000
Hansen's J p-val c 0.406 0.061 0.056
Notes: (1) Standard errors are in parentheses. (2) *, **, and *** indicate significance at a 10%, 5%, and 1%
level, respectively. (3) The dependent variable is the natural logarithm of the advalorem tariff rate (logit).
(4) The predicted signs for the coefficients of the regressors are indicated in parentheses below them.
(5) All the regressions include 4-digit ISIC industry dummies as regressors but are not reported. (6) List of
the instruments (all in logs): Capital share, materials prices (deviated from the producer price index), one
period lag of the measure of scale economies (value added/number of firms), and one period lag of the TFP
of upstream sectors.
a"Chi2-test p-val for all µi=0" provides the probability value for the Chi-squared test of H0: All µi (industry
fixed effects) are jointly insignificant.
b "Chi2-test p-val for all t =0" provides the probability value for the Chi-squared test of H0: All t (year
effects) are jointly insignificant.
c"Hansen's J p-val" provides the probability value for the Hansen-Sargan test of overidentifying
restrictions for H0: Excluded instruments are uncorrelated with the error term and correctly excluded from
the estimated equation.
48
TABLE 10. The Effect of Productivity on Tariffs and Effective Rates of Protection: OLS
Results
(1) (2)
logit logit eff
log(Qit/Mitit) 0.125*** 0.134***
(1>0) (0.015) (0.019)
logAit 0.046 0.071
(2>0) (0.032) (0.046)
UNILIBt -0.731*** -0.741***
(3<0) (0.024) (0.031)
Constant 3.418*** 0.199**
(>0) (0.078) (0.092)
Observations 920 887
R2 0.823 0.842
Wald test p-val a 0.000 0.000
Notes: (1) Standard errors are in parentheses. (2) *, **, and *** indicate significance at a 10%, 5%, and 1%
level, respectively. (3) logit is the natural logarithm of the advalorem tariff rate, and logit is the natural
eff
logarithm of the effective rate of protection. (4) The predicted signs for the coefficients of the regressors
are indicated in parentheses below them. (5) The estimates include 4-digit ISIC industry dummies as
regressors but are not reported.
aWald test p-val provides the probability value for the F-test of H0: The regressors are jointly insignificant.
49
TABLE 11. The Effect of Productivity Differences on Tariff and Effective Rate of
Protection Differences: OLS Results
(1) (2)
logit logit eff
log(Qit/Mitit) 0.076*** 0.084***
(1>0) (0.022) (0.031)
logAit 0.079* 0.071
(2>0) (0.046) (0.066)
UNILIBt -0.128*** -0.113***
(3<0) (0.017) (0.031)
Observations 676 652
R2 0.056 0.042
Wald test p-val a 0.000 0.021
Notes: (1) Standard errors are in parentheses. (2) *, **, and *** indicate significance at a 10%, 5%, and 1%
level, respectively. (3) logit is the one year change in the natural logarithm of the advalorem tariff rate,
and logit is the one year change in the natural logarithm of the effective rate of protection. (4) The
eff
predicted signs for the coefficients of the regressors are indicated in parentheses below them. (5) All
estimations allow for arbitrary intra-industry correlation over time.
aWald test p-val provides the probability value for the F-test of H0: The regressors are jointly insignificant.
50
TABLE 12. First Stage Regressions: Table 5 Column 1 Specification
log(Qit/Mitit) logAit
UNILIBt -0.686*** 0.097***
(0.0672) (0.020)
logCapital Share -0.109 -0.334***
(0.069) (0.020)
logMaterials Prices 0.432*** 0.008
(0.159) (0.047)
logScale -0.139*** 0.187***
(0.0458) (0.013)
logUpstream TFP -0.901*** 0.435***
(0.303) (0.088)
Constant 4.531*** -2.164***
(0.703) (0.205)
Observations 920 920
R2 0.897 0.858
Adjusted R2 0.887 0.844
Shea's partial R2 0.033 0.434
F statistic 86.67 59.96
Wald test p-val a 0.000 0.000
F statistic for excluded instruments 7.02 160.72
Wald test p-val excluded instruments b 0.000 0.000
Notes: (1) Standard errors are in parentheses. (2) *, **, and *** indicate significance at a 10%, 5%, and 1%
level, respectively. (3) These first stage regressions refer to the main tariff specification in column 1 of
Table 5. (4) The dependent variables are indicated in the header row of each column.
a Wald test p-val provides the probability value for the F-test of H0: The instruments are jointly
insignificant.
bWald test p-val provides the probability value for the F-test of H0: The excluded instruments are jointly
insignificant.
51
TABLE 13. First Stage Regressions: Table 6 Column 1 Specification
log(Qit/Mitit) logAit
UNILIBt -0.064 0.053**
(0.075) (0.026)
logCapital Share 0.379*** -0.405***
(0.085) (0.030)
logMaterials Prices 0.017 -0.061
(0.183) (0.064)
logScale -0.077* 0.103***
(0.045) (0.016)
logUpstream TFP 0.345 0.004
(0.377) (0.132)
Observations 676 676
R2 0.041 0.289
Adjusted R2 0.034 0.283
Shea's partial R2 0.002 0.014
F statistic 85.09 58.11
Wald test p-val a 0.000 0.000
F statistic for excluded instruments 6.80 155.58
Wald test p-val excluded instruments b 0.000 0.000
Notes: (1) Standard errors are in parentheses. (2) *, **, and *** indicate significance at a 10%, 5%, and 1%
level, respectively. (3) These first stage regressions refer to the main first-differenced tariff specification in
column 1 of Table 6. (4) The dependent variables are indicated in the header row of each column.
a Wald test p-val provides the probability value for the F-test of H0: The instruments are jointly
insignificant.
bWald test p-val provides the probability value for the F-test of H0: The excluded instruments are jointly
insignificant.
52
TABLE 14. Tariffs and Productivity: A System of Equations
OLS 3SLS
logAit logAit logit log(Qit/Mitit)
logit -0.066*** -0.067*** 1.869***
(0.022) (0.021) (0.107)
log(Qit/Mitit) 0.362***
(0.096)
logAit 0.198***
(0.057)
logCapital Share 1.038***
(0.082)
logMaterials Prices 1.041***
(0.241)
logScale 0.246*** 0.245***
(0.029) (0.014)
logUpstream TFP 0.260*** 0.257***
(0.092) (0.096)
Constant -1.842*** -1.828*** -1.612*** 5.227***
(0.321) (0.219) (0.229) (0.239)
Observations 920 920 920 920
R2 0.812 0.812 0.770 0.338
Wald test p-val 0.000 0.000 0.000 0.000
Notes: (1) Standard errors are in parentheses. (2) *, **, and *** indicate significance at a 10%, 5%, and 1%
level, respectively. (3) The dependent variables are indicated in the header row of each column. (4) The
logAit equations include industry dummies, GDP growth, and inflation as controls which are not reported
here. The logit equation includes both industry and year dummies which are not reported.
aWald test p-val provides the probability value for the Chi-squared-test of H0: The regressors are jointly
insignificant.
53
TABLE 15. Effective Rates of Protection and Productivity: A System of Equations
OLS 3SLS
logAit logAit logit eff log(Qit/Mitit)
logit eff -0.059*** -0.069*** 1.513***
(0.022) (0.020) (0.080)
log(Qit/Mitit) 0.468***
(0.127)
logAit 0.320***
(0.074)
logCapital Share 1.074***
(0.082)
logMaterials Prices 0.966***
(0.239)
logScale 0.247*** 0.247***
(0.029) (0.014)
logUpstream TFP 0.265*** 0.247**
(0.092) (0.098)
Constant -1.767*** -1.729*** -0.545* 3.906***
(0.332) (0.229) (0.302) (0.183)
Observations 887 887 887 887
R2 0.809 0.809 0.756 0.363
Wald test p-val 0.000 0.000 0.000 0.000
Notes: (1) Standard errors are in parentheses. (2) *, **, and *** indicate significance at a 10%, 5%, and 1%
level, respectively. (3) The dependent variables are indicated in the header row of each column. (4) The
logAit equations include industry dummies, GDP growth, and inflation as controls which are not reported
here. The logit equation includes both industry and year dummies which are not reported.
eff
aWald test p-val provides the probability value for the Chi-squared-test of H0: The regressors are jointly
insignificant.
54