2. Theory

Ornelas (Reference Ornelas2005a) develops a political–economy oligopolistic model in which external tariffs fall with a free trade area, and there is overall trade creation. The model is based on Krishna (Reference Krishna1998), with the important modification that external tariffs are endogenous.

A homogenous good is produced by firms from three countries. Countries X and Y are potential PTA partners and country Z represents the rest of the world. There are n_i firms in country i, which engage in Cournot behavior under segmented markets. The quantity sold by a firm from country i in j's market is $q_j^i $ and the equilibrium price of the good in country j is P _j. The marginal cost of production, c, is constant, and the demand is linear: Q _i =   ∝ _i − P _i, where ∝_i > c. Government i can impose a specific tariff on imports from j, $t_i^j $.Footnote ¹¹ Each government maximizes a welfare function that is given by consumer surplus, tariff revenue, and profits, which receive an extra weight of b _i ≥ 0. Thus, for country X:

(1)$$W_x = Q_x^2 /2 + ( {n_yt_x^y q_x^y + n_zt_x^z q_x^z } ) + ( {1 + b_x} ) n_x\mathop \sum \limits_{\,j = x, y, z} ( {P_j-c} ) q_j^x $$

The marginal effect of a change in $t_x^z $ on W _x is:

$$\eqalign{\displaystyle{{dW_x} \over {dt_x^z }} = & \,\,( {n_zq_x^z + n_yq_x^y + n_xq_x^x } ) \left({-\displaystyle{{dP_x} \over {dt_x^z }}} \right) + \left[{n_yt_x^y \displaystyle{{dq_x^y } \over {dt_x^z }} + n_z\left({q_x^z + t_x^z \displaystyle{{dq_x^z } \over {dt_x^z }}} \right)} \right] \cr \,\,\, + & \,( {1 + b_x} ) n_x\left[{q_x^x \displaystyle{{dP_x} \over {dt_x^z }} + ( {P_x-c} ) \displaystyle{{dq_x^x } \over {dt_x^z }}} \right]} $$

or:

(2)$$\eqalign{\displaystyle{{dW_x} \over {dt_x^z }} = & \,\left[{n_zq_x^z \left({1-\displaystyle{{dP_x} \over {dt_x^z }}} \right)-n_yq_x^y \displaystyle{{dP_x} \over {dt_x^z }}} \right] + \left[{n_yt_x^y \displaystyle{{dq_x^y } \over {dt_x^z }} + n_zt_x^z \displaystyle{{dq_x^z } \over {dt_x^z }}} \right]\cr + &\,\left[{( {1 + b_x} ) ( {P_x-c} ) n_x\displaystyle{{dq_x^x } \over {dt_x^z }} + b_xn_xq_x^x \displaystyle{{dP_x} \over {dt_x^z }}} \right]} $$

Now consider a PTA between X and Y. The PTA implies that $t_x^y = t_y^x = 0$, consistent with Article XXIV of the GATT.Footnote ¹² All derivatives on the right-hand side of (2) are independent of $t_x^y $.Footnote ¹³ Therefore, they are unaffected by the agreement's requirement that $t_y^x = 0$. Equilibrium prices and quantities will of course be affected by $t_x^y $. It can be shown that $dt_x^{z\ast } /dt_x^y > 0$; that is, a reduction in the tariff that X applies against imports from Y leads to a reduction in the tariff that X imposes against imports from Z.Footnote ¹⁴ This complementarity between the tariffs that a country applies against imports from different trading partners can be explained by means of three effects, corresponding to the three terms in square brackets in equation (2). Letting a tilde denote a variable in the presence of the PTA, to present the effects more clearly, we compare (2) with the effects of a change in the external tariff, $t_x^z $, on $\widetilde{{W_x}}$ (the government's payoff under the PTA), which is given by:

(2')$$\eqalign{\displaystyle{{d\widetilde{{W_x}}} \over {dt_x^z }} = &\left[{n_z\widetilde{{q_x^z }}\left({1-\displaystyle{{dP_x} \over {dt_x^z }}} \right)-n_y\widetilde{{q_x^y }}\displaystyle{{dP_x} \over {dt_x^z }}} \right] + \left[{n_y\widetilde{{t_x^y }}\displaystyle{{dq_x^y } \over {dt_x^z }} + n_zt_x^z \displaystyle{{dq_x^z } \over {dt_x^z }}} \right]\cr & \quad + \left[{( {1 + b_x} ) ( {\widetilde{{P_x}}-c} ) n_x\displaystyle{{dq_x^x } \over {dt_x^z }} + b_xn_x\widetilde{{q_x^x }}\displaystyle{{dP_x} \over {dt_x^z }}} \right]}$$

As already mentioned, the derivatives on the right-hand side of (2) are independent of $t_x^y $. The three effects are hence the following. The first is that, as the tariff against Y decreases, imports $( {n_zq_x^z + n_yq_x^y } ) $ increase via higher imports from the partner, $n_yq_x^y \;$(while imports from Z, $n_zq_x^z , \;$ fall), and thus the gain from increasing the external tariff (against Z) decreases. The reason is that, because imports from Y rise, the terms of trade deterioration with Y from increasing the external tariff becomes more prominent.Footnote ¹⁵ This effect is given by the first term in brackets in (2′). More precisely, equation (2) is the marginal effect of an increase in the external tariff on W _x, and we analyze how equation (2) varies when $t_x^y $ varies (because the PTA will bring about a change in the preferential tariff, $t_x^y $). Therefore, via the first term brackets in (2’), we compare the change in the terms of trade due to the change in imports, where the increase in imports is given by the fact that the term $n_yq_x^y $ in (2′) increases when $t_x^y $ falls.Footnote ¹⁶ This is the ‘terms of trade effect’ in Ornelas (Reference Ornelas2005a).

The second effect is that, because the tariff against Y is lower (or 0), the increase in tariff revenue resulting from the increase in imports from the partner from increasing the external tariff $( {n_yt_x^y dq_x^y /dt_x^z } ) $ falls (or goes away when $t_x^y = 0$). Note that, when comparing this term, it varies only due to the change in $t_x^y $, since $dq_x^y /dt_x^z $ does not change.Footnote ¹⁷ In other words, a lower level of $t_x^y $ with the PTA will make the increase in imports from the partner brought about by an increase in the external tariff fall. This is the ‘tariff revenue effect’ in Ornelas (Reference Ornelas2005a).

The third effect is that, as the tariff against Y falls, an increase in the external tariff brings about a smaller increase in profit. This has two components. One is that the markup of domestic firms in X, (P _x − c), decreases, and hence the increase in market share for the domestic firms due to increasing the external tariff generates less profit. This is the ‘strategic effect’. The other component is that, when the tariff against Y decreases, sales by domestic firms in X, $( n_xq_x^x ) $, decrease, and thus the increase in the domestic price in X from increasing the external tariff generates less profit (or less rents transferred from local consumers to local producers). This is the ‘distributive effect’.

Those three effects make $dW_x/dt_x^z $ necessarily fall due to the PTA. Because at the optimal level of $t_x^{z\ast } $, $dW_x/dt_x^z = 0$, the new value of $[ {dW_x/dt_x^z } ] ^{post\;PTA}$ when evaluated at $t_x^{z\ast } $ is negative, which implies that the equilibrium external tariff decreases with the PTA. Furthermore, the third effect in (2′) is scaled by b _x, the political–economy weight. Thus, the more politically biased are the PTA governments, the more their external tariffs will fall with the PTA.Footnote ¹⁸

As a final note, on the adequacy of this type of model for Latin American and other developing countries’ manufacturing industries, see Ornelas (Reference Ornelas2007). Also, Erosa and Hidalgo-Cabrillana (Reference Erosa and Hidalgo-Cabrillana2008) show that under weak contracts, capital market imperfections arise, which create rents for entrepreneurs and thus, as argued by Ornelas (Reference Ornelas2007), the use of a model where there are rents is especially applicable to developing countries, where contracting environments can be relatively weak.

3. Econometric Model and Predictions

The theoretical predictions are for how changes in preferential tariffs due to a PTA lead to changes in external tariffs via various mechanisms. Therefore, in order to examine how changes in the import protection level that country j applies on its partners on good i under a PTA impact the import protection level that the same country applies on the same good toward countries that do not belong to the PTA, I estimate the following equation:

(3)$$\eqalign{& \Delta {\rm \tau }_{ijt} = {\propto} _j + {\propto} _I + {\propto} _t + \beta ( {\Delta {\rm Pre}{\rm f}_{ijt-1}} ) + \gamma \Delta {\rm M}\_{\rm PT}{\rm A}_{ijt} + \delta ( {\Delta {\rm Pre}{\rm f}_{ijt-1}} ) {^\ast} \Delta {\rm M}\_{\rm PT}{\rm A}_{ijt} \cr & \quad + \theta \Delta {\rm \Pi }_{ijt} + \emptyset ( {\Delta {\rm Pre}{\rm f}_{ijt-1}} ) {^\ast} \Delta {\rm \Pi }_{ijt} + \sigma bhi_j {^\ast \Delta} {\rm \Pi }_{ijt} + \rho ( {\Delta {\rm Pre}{\rm f}_{ijt-1}} ) {^\ast} bhi_j {^\ast} \Delta {\rm \Pi }_{ijt} + \epsilon _{ijt}} $$

where Δτ_ijt is the change in the import protection level that country j applies on good i imported from PTA non-members between years t and t – 1, while ΔPref_ijt−1 represents the lagged change in the bilateral level of import protection that country j applies good i imported from its PTA partners. To explain the inclusion of all the right-hand-side variables, I now refer to the three effects from the theoretical model.

The first effect is that, if the preferential tariff decreases, imports increase via higher imports from the partner, and hence the terms of trade loss with the partner from increasing the external tariff is higher; therefore, the external tariff will decrease. I capture this effect by including the change in preferential imports interacted with the (lagged) change in preferential import protection $( {( {\Delta {\rm Pre}{\rm f}_{ijt-1}} ) {^\ast} \Delta {\rm M}\_{\rm PT}{\rm A}_{ijt}} ) $.Footnote ¹⁹ Because a larger preferential (or external) tariff cut will be measured in this paper as a more negative ΔPref_ijt−1 (or Δτ_ijt), according to the theory, the coefficient of this interaction, $( {\Delta {\rm Pre}{\rm f}_{ijt-1}} ) {^\ast} \Delta {\rm M}\_{\rm PT}{\rm A}_{ijt}$, should be positive.

The second effect is that, when the preferential tariff is lower, the increase in tariff revenue derived from the increase in preferential imports generated from increasing the external tariff falls, and so the external tariff will fall. This means that the coefficient of the change in the preferential import protection (ΔPref_ijt−1) should be positive. As captured by the third term in square brackets in equation (2’), this effect is due only to the change in $t_x^y $ (recall that the derivatives do not change).

The third effect is given by the fact that, as the preferential tariff falls, the markup and sales of domestic firms diminish, and therefore the increase in both the domestic firm's market share and the domestic price from raising the external tariff creates less profit. This effect is expected to be stronger, or more likely to be relevant, when the government is more politically motivated. Therefore, I capture the effect by including the change in profits interacted with the change in preferential import protection ((ΔPref_ijt−1)*ΔΠ_ijt), as well as the same interaction multiplied by a dummy for a high b _j (denoted bhi _j), which is equal to 1 when the political–economy weight, b _j, belongs to the top tercile in the sample. The sum of the coefficients of both profit interactions (scaled and not scaled by b _j) is expected to be negative, since the effect should be present or larger when political–economy forces are stronger.Footnote ²⁰ The political–economy parameters, b _j, are obtained using Gawande et al. (Reference Gawande, Krishna and Olarreaga2009)'s estimates of the weight that governments give to social welfare relative to private interests, as explained in section 5. I also include each, the change in imports and the change in profits alone, as controls. There could be some unobservable industry characteristics that impact the changes in external import protection. An example would be political–economy factors that can affect the size of import protection changes (e.g. lobbying for import protection can take place at the industry level). I thus add an industry fixed effect, ∝_I.Footnote ²¹ I include a country fixed effect, ∝_j, as well, to control for country characteristics, such as economic or political factors, that could impact changes in import protection. A year fixed effect, ∝_t, captures the impact of broad macroeconomic shocks affecting the Latin American countries. And ε _ijt is the error term.Footnote ²²

An important difference with the majority of the literature, which has focused on MFN and preferential tariffs only, is that I also include in my measures of import protection the temporary trade barriers of antidumping and safeguards that country j applies.Footnote ²³ This means, firstly, that τ_ijt in (3) will be defined as the sum of the (applied) ad valorem MFN tariff plus the ad valorem temporary trade barrier that country j imposes on imports of good i from countries that do not belong to its PTAs; and, secondly, that Pref_ijt−1 in (3) will be defined as the sum of the (applied) preferential tariff plus the ad valorem temporary trade barrier that country j applies on imports of good i from countries that are members of its PTAs. In section 5, I explain the procedure I use to sum tariffs and temporary trade barriers.

As in related literature, there are some issues potentially affecting the econometric estimation of the link between changes in import protection against PTA members and outsiders. One is that I use a one-year lag of ΔPref_ijt−1 because it may be considered predetermined with respect to the MFN tariff (or TTB), and that could generate a lag in its impact (countries negotiate reductions in preferential tariffs in each trade agreement, and they are implemented according to a specified liberalization schedule).Footnote ²⁴ Using the lagged variable also helps lower simultaneity bias.

I set the preferential tariff equal to the MFN tariff in the year before a country offers its first preference for a given product, so that the effect of the first preferential tariff cut is measured. When a country does not offer a preference in years t and t – 1 in an industry, the change in the preferential tariff is set to zero. Moreover, when the MFN tariff is zero, the preferential tariff will also have to be zero, leading to a potential bias in the estimates. Therefore, as it is also standard in this literature, I exclude observations with a zero MFN tariff.

Additionally, given that a country may belong to various PTAs at the same time, I define the preferential tariff that country j imposes in industry i in year t as the minimum of the preferential tariffs that country j imposes in industry i in year t toward all of its different PTA partners, following Estevadeordal et al. (Reference Estevadeordal, Freund and Ornelas2008) and others.Footnote ²⁵

There is also potential endogeneity. First, it could be that trade liberalization across industries in the Latin American countries may have occurred in a similar manner multilaterally and under PTAs; for example, some products may be easier to liberalize. Second, there may be reverse causation from external import protection to preferential liberalization. Actually, changes in preferential tariffs are predetermined to changes in MFN tariffs (and to temporary trade barriers), given that tariff concessions are negotiated in the PTA and their decreases over time occur following a detailed schedule (also established under the PTA). That notwithstanding, it could be that some changes in MFN tariffs were expected when negotiating the preferences and, if so, they could have influenced those preference levels. Therefore, I use instrumental variables to deal with these issues. To instrument for a country's preferential liberalization, I resort to the preferential liberalization of the two countries in the sample most correlated with the country's own preferential liberalization.Footnote ²⁶ Such correlations are high, and those instruments are valid provided that the aspects that influence the preferential tariffs of a country's partners are not the same affecting the country's own MFN tariffs or TTBs (Estevadeordal et al., Reference Estevadeordal, Freund and Ornelas2008). Moreover, Tovar (Reference Tovar2019) discusses the trade reforms in Latin America in more detail and argues that there exist differences in the patterns of trade liberalization among those countries. It also posits that the use of TTBs varies across those countries along important dimensions as well, which are economically meaningful for my analysis (including variation in the use of TTBs across years, the countries targeted, and the products affected by those TTBs).

Endogeneity may also affect the change in preferential imports and the change in profits. For each country and each of those variables, I use as instruments the corresponding variables from the two countries in the sample for which those variables are most correlated with the country's own variable and which are never its PTA partners during the sample period. I use non-PTA partners because the change in preferential imports and the change in profits of a country's PTA partners are affected by market conditions in each of those countries, which in turn could be affected by the preferential liberalization and the change in the aforementioned variables of the country itself. For the change in profits, I also obtained and used data from other Latin American countries for which value-added data are available (Bolivia and Honduras). I included those countries to make the selection of the countries with most correlated data to be used as instruments for the change in profits, in order to improve the correlation between the endogenous variable and its instrument.Footnote ²⁷ I use a test to evaluate the endogeneity of the variables and also a test of overidentifying restrictions to confirm the validity of the instruments.

4. Latin America's Trade Liberalization Reforms and Temporary Trade Barriers

Subsequent to the implementation of the import substitution policies, which were characterized by protectionism, the 1980s was the period when most unilateral trade reforms were initiated in Latin America. Most of them were implemented from the mid or late 1980s until the early 1990s (Chile being an exception, with an early and abrupt process started at the end of 1973 by a military government (Sáez, Reference Sáez, Michael Finger and Nogués2006)). They involved decreases in tariffs, the number of tariff levels, a reduction or elimination of quantitative restrictions on imports and other non-tariff barriers, and institutional reforms. Furthermore, the 1990s was a decade in which a number of important preferential trade agreements, including NAFTA and Mercosur, were negotiated and implemented.Footnote ²⁸

With the exception of Argentina, whose antidumping (AD) legislation was enacted in 1972 and allowed the use of safeguards (SGs) as well (Moore, Reference Moore and Bown2011), in the majority of Latin American countries, antidumping legislations were introduced in the 1980s or early 1990s, overlapping with the trade reforms. The first AD investigations were typically initiated in the late 1980s or early 1990s (i.e., also during the trade liberalization phases). The sector with the largest AD use in Latin America is metals, followed by chemicals, plastics/rubbers, textiles, and machinery/electrical. These sectors are also major users of AD worldwide. The country most frequently targeted by Latin America's AD investigations is China. Other frequent targets are the European Union, the United States, and Brazil. Subsequently, in the 1990s, the legislations were reformed and also brought into consistency with GATT/WTO rules. Legislations on safeguards were usually established in the 1990s, and the first SG investigations were started later in the 1990s or, in a few cases, in the 2000s. As in other countries, AD is by far the most used TTB in Latin America and the use of SGs is much smaller. The legislation in some countries allows the freedom to apply a lesser duty (below the dumping margin), and some include a national interest clause that permits them to deny AD measures even in the presence of dumping and injury. Some preferential trade agreements have rules for the use of TTBs.Footnote ²⁹

5. Data

The data I employ for the estimations are for four-digit ISIC manufacturing industries in eight countries in Latin America (Argentina, Chile, Colombia, Ecuador, Mexico, Peru, Uruguay, and Venezuela) from 1990 to 1999. I obtained data on value added and wage bill to construct the profit variable. The data on value added and wage bill are missing for Paraguay for that period, as well as for many industries in Brazil. Therefore, I do not include those two countries and the countries used are a subset of the ten countries in Estevadeordal et al.'s (Reference Estevadeordal, Freund and Ornelas2008) dataset.

Data on MFN tariffs were obtained from the World Integrated Trade Solution (WITS). Data on preferential tariffs derive from the tariff schedules of the different preferential trade agreements, which contain the timeline for tariff cuts across products, years, and countries. Simple averages of the preferential tariffs are used to aggregate the data at the four-digit ISIC level.Footnote ³⁰

I measure changes in import protection (in Δτ_ijt and ΔPref_ijt−1 in equation (3)) by including not only changes in applied MFN and preferential tariffs, but also the ad valorem TTBs – AD and SG policy measures – that a country imposes against the rest of the world and its PTA partners. To construct the ad valorem measure of a TTB, I use data at the exporter–product level. Some AD policies were imposed as specific duties; however, I have the data on the final dumping margin as well, which are measured in ad valorem terms.Footnote ³¹ The dumping margin is sometimes reported for each exporting firm in an investigated country but, in some cases, it is only reported as a range of values of the import restrictions facing the exporters of the good in a targeted country. Therefore, I compute two alternative variables: (i) AD_min, defined as the average of the minimum AD margins, where the average is calculated across all foreign exporting countries that are being subject to the given country's AD measure in that good; and (ii) AD_max, analogously defined as the average of the maximum AD margins across all foreign exporting countries subject to the country's AD measure in the good.Footnote ³² For robustness purposes, I will use and present results with both measures. An important difference relative to tariffs is that AD duties may be imposed against some exporting countries only. To adjust for this when calculating the final import protection measure, I use the sum of the applied import tariff and the AD margin weighted by the affected countries’ share in total imports of the good by the imposing country.Footnote ³³

I also perform the estimations by measuring the changes in import protection as the sum of tariffs plus AD and SGs, using data on the safeguard duties imposed by the Latin American countries. The data on AD and SG import restrictions at the product level are obtained from the Temporary Trade Barriers Database (Bown, Reference Bown2015), and originate from government sources from each of the sample countries.

Data on bilateral imports by sector and year used to calculate imports from the PTA partners are from COMTRADE, obtained via WITS. Profit is calculated as value added or, alternatively, as value added minus payroll, with data from the World Bank's Trade and Production database (Nicita and Olarreaga, Reference Nicita and Olarreaga2007). These data are at the three-digit ISIC level, and because they are not available for all country years, I use the long change in profits, set as the change from 1990 to 1999.Footnote ³⁴ Given that variation in the profit variables is at the three-digit ISIC level, I adjust the standard errors for clustering at the country-ISIC3 level in all regressions. I account for clustering at the country-sector level because variables such as preferential and external liberalization are expected to be correlated for products within an industry in a given country (for example, lobbying for protection usually takes place at the industry level). Finally, I use the political–economy parameters (b _j) derived from estimates of the weight that governments attach to social welfare relative to private interests from Gawande et al. (Reference Gawande, Krishna and Olarreaga2009). Using the Grossman and Helpman (1994) model as the theoretical basis, Gawande et al. (Reference Gawande, Krishna and Olarreaga2009) provide estimates of the welfare-mindedness of governments (known as parameter a in the literature on the political–economy of trade protection) for 54 countries for the period 1988–2000, including the countries used in this paper. I calculate b _j as the inverse of each country's parameter a from Gawande et al. (Reference Gawande, Krishna and Olarreaga2009), and then I define bhi _j as a dummy equal to 1 when b _j belongs to the top tercile in the sample.

6. Estimation Results

6.1 Baseline Estimates and Robustness

I begin by presenting baseline panel estimates of the econometric model. Table 1 reports summary statistics for the variables used in the estimations. Table 2 shows the results from the estimation of equation (3). Beginning with column 1, the dependent variable there is the change only in the MFN tariff applied on a good against PTA non-members. The main explanatory variable is the lagged change only in the preferential tariff applied on a good against PTA partners. The (lagged) change in the preferential tariff has a positive coefficient, statistically significant at the 1 percent level. This is consistent with the ‘tariff revenue’ effect from the theory, which is that, when the preferential tariff falls, the increase in tariff revenue arising from the increase in imports from the partner from increasing the MFN tariff decreases, and thus the MFN tariff will decrease.

Table 1. Summary statistics

Table 2. The effect of preferential liberalization on import protection against non-members

Notes: Robust standard errors in parentheses with *, **, and *** indicating statistically significant at 10%, 5%, and 1% levels, respectively. Constant included but not reported. 1/ Columns 4–5 exclude AD measures that are DPU (duty if the price falls under a given level). 2/ Variable is scaled by 100,000. Standard errors are adjusted for clustering at the country-ISIC3 level in all regressions.

The coefficient of the change in the preferential tariff interacted with the change in preferential imports is positive, in accordance with the model's prediction (the ‘terms of trade’ effect), but it is not statistically significant.

Finally, the sum of the coefficients of both change in profit interactions with the change in the preferential tariff (scaled and not scaled by the political–economy parameter, b _j) is negative and statistically significant at the 10 percent level (as shown at the bottom of the table), which means that, when political motives are sufficiently powerful, the results are aligned with the prediction from the model that, as the preferential tariff falls, both the markup of and the sales by domestic firms decline, and thus increasing the external tariff generates less profit (the ‘strategic and distributive’ effect). Hence, the external tariff will be lower. Note that the coefficient of the change in the preferential tariff interacted with the change in profits by itself is not significant, which means the effect is not present when the political–economy parameter is low, but it is present when political–economy motives are stronger. This is also consistent with the theory's prediction that the effect increases with the strength of the political–economy forces.

Overall, these initial results support most of the predictions from Ornelas’ (Reference Ornelas2005a) political–economy model on the effects of PTAs on trade liberalization toward non-member countries.

In columns 2 and 3, the dependent variable and the main explanatory variable are redefined more comprehensively to include not only tariffs, but also the antidumping import policy restrictions that the country imposed on imports originating from the rest of the world (in Δτ_ijt), and on imports coming from PTA partners (in ΔPref_ijt−1). In column 2, I present results using the average of the minimum AD margins and in column 3, I use the average of the maximum AD margins (calculated as explained in the previous section). I exclude from the sample some outlier observations in which the dependent variable (the change in the MFN tariff plus the AD duty) was greater than 100 percent or lower than −100 percent.Footnote ³⁵ The results are very similar to those from column 1, which only use tariffs. The coefficients of the three variables of interest have the signs predicted by the theory, and the coefficient of the change in the preferential tariff and the sum of the coefficients of both change in profit interactions with the change in the preferential tariff are statistically significant (the latter now at the 1 percent level in column 3).

However, some AD measures are imposed as a duty if the price falls below a given level (DPU) and, in those instances, firms have an incentive to increase their price and ‘enjoy’ the corresponding rents, instead of facing the duty. In such event, unlike an import tariff or AD duty, there would not be any tariff revenue collected. Since that would affect the first and second theoretical effects from the model (as they are related to tariff revenue), in columns 4 and 5 I exclude AD measures imposed as DPU. The results are again similar. The coefficient of the change in bilateral import protection toward PTA partners alone and the sum of coefficients of the change in profit interactions (in absolute value) are a bit lower when AD is included in the import protection measures (in columns 2–5 relative to column 1), which suggests that the building block effect of preferential trade liberalization falls slightly when more comprehensive measures of import protection toward PTA members and non-members are considered. Other studies have used the first coefficient to establish whether there is a building block or stumbling block effect of preferential liberalization on trade liberalization against non-member countries (e.g. Estevadeordal et al., Reference Estevadeordal, Freund and Ornelas2008), although most studies have only used tariffs.Footnote ³⁶

To account for the use of safeguards, in columns 6 and 7 I redefine the import protection variables to include both AD and SG measures imposed against other PTA members and against non-PTA members.Footnote ³⁷ The estimated results are very similar to the previous ones, providing support for most of the theory (i.e., for all mechanisms except the terms of trade effect).

Turning to the magnitudes of the effects, with the estimates from column 1, a one standard deviation increase in the preferential tariff reduction leads to a decrease of 0.60 percentage points in the MFN tariff, which is sizeable given that median MFN tariff in the sample is 14.7 percentage points. Furthermore, a one standard deviation increase in the sum of the change in profits interactions lowers the MFN tariff by 0.87 percentage points. Adding the two effects, they entail a decrease of 10.0 percent in the MFN tariff when political–economy forces are strong, which is substantial (in the case of low political–economy forces, only the first effect would be present, i.e., a decrease of 4.1 percent in the MFN tariff). If we also consider TTBs, with the coefficients from column 7, a one standard deviation increase in preferential liberalization leads to a reduction in protection (MFN tariff plus AD and SGs) against non-member countries by 0.56 percentage points. Meanwhile, the median import protection against non-members in the sample is 15.0 percentage points. Moreover, a one standard deviation increase in the sum of the change in profits interactions decreases protection against non-members by 0.79 percentage points. The sum of the two effects represents a reduction of 9.0 percent in the external import protection. Thus, the building block effect becomes a bit lower when TTBs are taken into account, and the results are not only statistically but also economically significant.

In Table 2, profit was measured as value added. Alternatively, I measure profit as value added minus payroll, which may concord better with the theoretical model (for this reason, I keep this measure in all the estimations in subsequent tables as well). Columns 1–3 of Table 3 replicate the specifications from columns 1–3 of Table 2 with the difference in the profit measure only. There is again support for the ‘tariff revenue’ effect, since the coefficient of the change in the preferential tariff is always positive and highly statistically significant. The coefficient of the preferential imports interaction is again positive, as expected, but not statistically significant; therefore, I do not find evidence of the ‘terms of trade’ effect. The sum of coefficients of the profit interactions is always negative, as predicted by the theory, and it is statistically significant at the 1 percent level in column 3, when using the average of the maximum AD margins. This provides some support for the presence of the ‘strategic/distributive’ effect. In columns 4 and 5, I include not only AD but also SG measures (in addition to tariffs), and the results are similar to those with AD only (columns 2–3).

Table 3. The effect of preferential liberalization on import protection against non-members—robustness to profit measure

Notes: Robust standard errors in parentheses with *, **, and *** indicating statistically significant at 10%, 5%, and 1% levels, respectively. Constant included but not reported. 1/ Variable is scaled by 100,000. Standard errors are adjusted for clustering at the country-ISIC3 level in all regressions.

As a sensitivity analysis, instead of using the change in preferential imports, I use the change in total imports. This allows me to also incorporate the decrease in the imports originating from non-members of the PTA that may take place. In Table 4, I replicate the specifications from Table 3 but changing the import measure as explained. The results support all the predictions from the model. The change in the preferential tariff and the change in imports interaction have positive and statistically significant coefficients, and the sum of coefficients of the profit interactions is always negative, and it is significant in most regressions.

Table 4. The effect of preferential liberalization on import protection against non-members—robustness to import measure

Notes: Robust standard errors in parentheses with *, **, and *** indicating statistically significant at 10%, 5%, and 1% levels, respectively. Constant included but not reported. 1/ Variable is scaled by 100,000. Standard errors are adjusted for clustering at the country-ISIC3 level in all regressions.

Finally, I tried dividing the sample into two periods: 1990–1994 and 1995–1999. The second period begins with the start of the two customs unions in the sample, Mercosur and the Andean Community, which are major PTAs in Latin America and went from being FTAs to becoming CUs in 1995. The results from replicating the specifications from Table 3 indicate that the building block result in period 1 (1990–1994) is always present, both using only tariffs and adding TTBs, similarly to the results I had found for the whole period (in Table 3).Footnote ³⁸ Also, the main mechanism that drives this building block effect in period 1 is the ‘tariff revenue’ effect. In period 2 (1995–1999), there is a building block effect (of around the same magnitude as in period 1) only once TTBs are added to tariffs, which is driven by the ‘strategic and distributive’ effect (using tariffs only, no effect is found).Footnote ³⁹

These results provide interesting evidence on the dynamics of the effects; in particular, of how different theoretical channels are the main drivers in different periods in Latin America. It is also interesting that the ‘strategic and distributive’ effect plays a role in the second period, which is the CU period. Since the literature has shown that free-riding issues in lobbying activities can worsen in CUs relative to FTAs (e.g., Richardson, Reference Richardson1994; Panagariya and Findlay, Reference Panagariya, Findlay, Feenstra, Grossman and Irwin1996), it is possible that a weakening in lobbying may have contributed to larger reductions in external import protection in response to preferential liberalization via the decrease-in-profits channel during the CU period.Footnote ⁴⁰

To sum up, the results provide empirical support for most of the theoretical predictions from Ornelas’ (Reference Ornelas2005a) model, according to which preferential trade liberalization will lead to liberalization against countries that do not belong to the preferential trade agreements. More precisely, there is strong support for two of the three theoretical mechanisms (the tariff revenue and the strategic/distributive effects), and only weak support for the terms of trade effect (since the corresponding estimated coefficient is only statistically significant in Table 4). The model incorporates political–economy motives for protection, which turn out to be empirically important, and the results hold not only using tariffs but also using more expansive import protection measures that include AD and SG policies imposed against PTA members as well as outsiders.

6.2 Instrumental Variables Estimates

Since there is possible endogeneity, in this section I address it via the use of instrumental variables. To instrument for the change in preferential liberalization, I use the change in the preferential liberalization of the two countries in the sample most correlated with the country's own preferential liberalization. I use an analogous procedure to instrument for the change preferential imports and the change in profits, with the difference that I now use countries that are not PTA partners at any time during the sample period.

Table 5 presents the results of the estimation using instrumental variables (IV). Column 1 repeats the specification from column 1 of Table 3, but using an IV-GMM estimation procedure, in which I instrument for the change in the preferential tariff, the change in preferential imports, and the change in profits. The coefficients of all the variables of interest have their predicted signs, but only the (lagged) change in the preferential tariff is statistically significant. The Hansen-J statistic (shown at the bottom of the table) confirms the validity of the instruments and their correct exclusion from the equation.Footnote ⁴¹ In addition, in Table A1 in the Appendix, I provide the results from the first-stage estimations corresponding to column 1.Footnote ⁴² However, I test for the endogeneity of the variables with the Durbin and Wu–Haussman tests, and the results indicate that we cannot reject the null hypothesis of exogeneity of the change in profits. Hence, column 2 replicates the specification from column 1 but considering the change in profits as econometrically exogenous, and instrumenting for the remaining variables. Again, all the coefficients of the relevant variables have signs in accordance with the theory, and the coefficient of the (lagged) change in the preferential tariff and the sum of the coefficients of the change in profits interaction are statistically significant, consistent with the results from the previous section also.

Table 5. IV estimates of the effect of preferential liberalization on import protection against non-members

Notes: Robust standard errors in parentheses with *, **, and *** indicating statistically significant at 10%, 5%, and 1% levels, respectively. Constant included but not reported. 1/ Variable is scaled by 100,000. Standard errors are adjusted for clustering at the country-ISIC3 level in all regressions.

The results from column 2 imply that a one standard deviation increase in the preferential tariff reduction generates a decrease of 0.94 percentage points in the MFN tariff, which is again sizeable since the median MFN tariff in the sample is 14.7 percentage points. Furthermore, a one standard deviation increase in the sum of the change in profits interactions lowers the MFN tariff by 2.47 percentage points. Adding the effects, they represent an overall decrease in the MFN tariff of 23.2 percent, a considerable effect.

To account for the use of temporary trade barriers and examine how this might impact the results, in columns 3 and 4 of Table 5, I redefine the variables to incorporate antidumping measures (column 3 uses the average of the minimum of the AD margins and column 4 uses the average of the maximum). The results for the preferential tariff change and the change in profits are qualitatively similar to those from column 2, and the change in profits has a larger effect on external protection once AD measures are included (as shown by the sum of coefficients at the bottom of the table), which could mean that external protection that includes AD is more sensitive to changes in profit (when political forces are strong). Moreover, the effect tends to be stronger, considering both size and significance, with AD_max. Since AD_max uses the average across targeted exporting countries of the maximum AD margins (i.e., of the highest exporting firm-specific dumping margins), using AD_max means that there will be a larger change in external protection due to a given change in profits (relative to AD_min), thus leading to a larger magnitude for the estimated coefficient. However, the change in preferential imports interaction is negative and statistically significant, which goes against the theoretical prediction and implies that there is not support for the terms-of-trade effect. Lastly, in columns 5 and 6 of Table 5, I incorporate not only tariffs and AD but also safeguard measures. The results are similar to those from columns 3 and 4.

Does the inclusion of TTBs strengthen or weaken the building block effect? Considering the results from column 6, a one standard deviation increase in preferential liberalization generates a reduction in external protection (MFN tariff plus AD and SGs) of 0.62 percentage points. Meanwhile, the median external import protection in the sample is 15.0 percentage points. Moreover, a one standard deviation increase in the sum of the change in profits interactions decreases protection against non-members by 4.78 percentage points. On the other hand, a one standard deviation increase in the change in preferential imports interaction now increases protection against non-members by 2.33 percentage points. The sum of the three effects represents a net reduction of 20.5 percent in the external import protection. Thus, and as found also in the non-IV results in the previous section, the building block effect becomes a bit lower when TTBs are taken into account, and the results are not only statistically but also economically significant.

Comparing these results with those of Tovar (Reference Tovar2019), I obtain that the estimates of this model, which incorporates political–economy forces, lead to a substantially larger building block effect of preferential liberalization in Latin America.Footnote ⁴³

As an additional robustness test, I define the dummy for a high b _j as being equal to 1 when the political–economy weight belongs to the top two terciles in the sample (instead of the top tercile). Columns 1–4 from Table 6 replicate the specifications from columns 3–6 of Table 5, with the difference in the definition of the dummy for the political–economy parameter. The results are very similar to those of Table 5.

Table 6. IV Estimates of the effect of preferential liberalization on import protection against non-members: Additional Robustness Tests

Notes: Robust standard errors in parentheses with *, **, and *** indicating statistically significant at 10%, 5%, and 1% levels, respectively. Constant included but not reported. 1/ Variable is scaled by 100,000. Standard errors are adjusted for clustering at the country-ISIC3 level in all regressions.

Thus far, the industry fixed effects have been defined at the three-digit ISIC level, which is the level of variation of the change in profit variables. In column 5 of Table 6, I re-run the tariffs-only specification from column 2 of Table 5 but using four-digit ISIC level fixed effects. The results remain quantitively very similar, but now the change in preferential imports interaction becomes significant at the 10 percent level (again with the predicted sign). In columns 6 and 7 of Table 6, I repeat the estimations from columns 4 and 6 of Table 5 and the results remain qualitatively and quantitatively similar.Footnote ⁴⁴

Overall, there is strong support for two of the three predicted mechanisms from Ornelas’ (Reference Ornelas2005a) model – in particular, for the tariff revenue effect and for the strategic/distributive effect – as well as evidence of a building block effect of preferential liberalization for the Latin American countries in the 1990s, which is also a prediction from the model. I do not find evidence supporting the terms-of-trade effect. Theoretically, the magnitude of the terms-of-trade effect depends on the increase in preferential imports. Therefore, one possible reason why there is not support for this theoretical channel in the sample could be that Latin America has a low level of intraregional trade.Footnote ⁴⁵ Bown et al. (Reference Bown, Lederman, Pienknagura and Robertson2017) show that the Latin American countries have, on average, very similar patterns of revealed comparative advantage. They also have very similar trade structures, import demands, and export baskets (and especially so with the countries that are nearby), which does not necessarily occur in other regions. Bown et al. (Reference Bown, Lederman, Pienknagura and Robertson2017) argue that the similarity in comparative advantage is one of the constraints to deeper regional integration in the region. It would therefore be interesting to test whether the terms-of-trade effect is present in regions with higher levels of preferential trade.