Market access for sale: Latin America's lobbying for US tariff preferences Hiau Looi Kee Marcelo Olarreaga Peri Silva§ March 2003 Abstract This paper assesses the foreing lobbying forces behind the tariff preferences that the United States grants to Latin American coun- tries. The basic framework is the one developed by Grossman and Helpman (1994) that is extended to explain the relationship between foreign lobbying and tariff preferences. Results suggest that returns to Latin American exporters lobbying for tariff preferences in the United States are above 50 percent. The reason for these large returns is due to the relatively low weight given to social welfare in the US government's objective function relative to Latin American exporters' lobbying contributions for tariff preferences. JEL classification numbers: F10, F11, F13 Keywords: Trade, Political Economy, Latin America, United States. We are grateful to Olivier Cadot, Caroline Freund, Kishore Gawande, Bernard Hoek- man, Daniel Lederman and Caglar Ozden for helpful discussions. The views expressed here are those of the authors and do not necessarily reflect those of the institutions to which they are affiliated. Development Research Group, The World Bank, Washington, DC 20433, USA; Tel. (202)473-4155; Fax: (202)522-1159; e-mail: hlkee@worldbank.org Development Research Group, The World Bank, Washington, DC 20433, USA, and CEPR, London, UK; Tel. (202)458-8021; Fax: (202)522-1159; e-mail: molarreaga@worldbank.org §Department of Economics, University of Illinois at Urbana-Champaign, Champaign, Illinois; Tel. (217)333-0120; Fax: (217)244-6678; e-mail: pasilva@students.uiuc.edu 1 Introduction The purpose of this paper is to assess the importance of lobbying by foreign exporters in determining the extent of tariff preferences granted by an im- porting country. We focus on tariff preferences granted by the United States (US) to Latin American countries, for at least two reasons. First, the US is the only country where foreign lobbying expenditures are publicly avail- able. Second, the extent of tariff preferences granted to Latin America by the US is on paper quite important (Andean Act, CBI, CBTPA, GSP, Nafta, Puerto Rico-CBI, and more recently Chile-United States). In practice, it also accounts for a large share of Latin American exports to the United States, around 50 percent.1 An important and growing empirical literature has been exploring the importance of lobbying in determining trade policy in different countries. Most of the recent empirical and analytical literature is based on the work of Grossman and Helpman (1994), which provided a flexible framework to analyze issues of lobbying and endogenous policy determination.2 So far most of the literature has focused on the United States mainly for data issues (see Gawande and Krishna, 2002).3. Empirical estimates suggests that 1Around 82 percent of preferential exports entered under the NAFTA regime (Mexico only) in the year 2000; CBTPA countries followed with 6 percent; GSP accounted for 4 percent; CBI for 3 percent and the Andean act regime for 2 percent of Latin America preferential exports to the United States. Other special import regimes, such as the Civil Aircraft, Pharmaceuticals and Dyes accounted for the rest of non-program-claimed imports of the US from Latin America. 2For a recent review of this literature see Grossman and Helpman (2001). 3Studies focused on developing countries include Cadot, de Melo and Olarreaga (2003), Cadot, Grether and Olarreaga (2003), Gawande, Sanguinetti and Bohara, (2002) and Mitra, Thomakus and Ulubasoglu (2002). there is a non trivial role for domestic lobbying in the determination of the US trade policy (see Gawande, 1997, Gawande and Bandhopadhyay, 2000 and Goldberg and Maggi, 1999). More recently, Gawande, Krishna and Robbins (2002) focused on the role of foreign lobbying in determining US Most Favoured Nation (MFN) tar- iffs.4 They found strong evidence that foreign lobbying tends to decrease MFN tariffs in the US. In what follows below we try to assess the role of foreign contributions by Latin American countries in explaining tariff pref- erences granted by the US to these countries. Indeed, in the case of small countries, such as those in Latin America which represent individually less than 1 percent of world trade, contributions would probably rather target preferential access rather than MFN tariff reductions that would benefit all other exporting countries. Although the empirical literature has found significant support in the data for domestic and foreign lobbying being an important determinant of trade policy in the United States, it still faces an important puzzle: most estimates of the weight granted to social welfare relative to contributions in the US government's objective function are extremely high: generally above 100. Such high values suggest that focusing on a non-political economy objective function for the United States would be a quite good approximation. There are several reasons why estimates for the weight granted to social welfare are so high (see Gawande and Krishna, 2002 for an exhaustive discussion of these 4Most of the empirical literature in the US actually mainly explains non-tariff barriers. This is generally justified by the fact that tariffs are subject to multilateral tariff negoti- ations and therefore the political economy of trading partners would also determine their level. Note that for any analysis after the Tokyo round this may also apply to non-tariff barriers depending on their nature. 2 estimates). The one explored in this paper is that contributions functions may not necessarily be continuous. Indeed, in the case of tariff preferences granted by the US, these are unlikely to be continuous as the US government either grants full tariff preference (zero tariff) or no preferences at all under most of its preferential schemes.5 The paper addresses two main questions: Does lobbying by Latin Amer- ica's producers in the US explains the pattern of tariff preferences each coun- try enjoys? And if yes, what is the return on $1 dollar of lobbying in the United States by Latin American exporters? Results suggests that political contributions by Latin American exporters to the US government can indeed help explain the variation in tariff prefer- ences across products and countries. Moreover, the returns to foreign lobby- ing seem to be relatively high, around 50 percent. Finally, contrary to the empirical literature for the United States described above we found very low values for the estimates of the weight granted to social welfare in the govern- ment's objective function (around two times the weight granted to foreign contributions), which underscores the importance of foreign (and domestic) lobbying in determining US trade policies. The paper is organized as follows. Section 2 discusses the pattern of foreign contributions by Latin American producers in the United States, as well as the structure of tariff preferences granted by the United States. Section 3 presents the analytical setup. Section 4 describes the empirical strategy and section 5 presents the results. Section 6 concludes. 5The exception being the phase-out periods, of Nafta for example. 3 2 Foreign lobbying and tariff preferences This section discusses the data sources and variable construction and provides a description of the pattern of Latin American lobbying and tariff preferences by country and sector. 2.1 Foreign Lobbying The data set on foreign lobbying used in this paper was provided by the US Department of Justice. Foreign lobbying activities have to be reported fol- lowing the legislation known as the Foreign Agent Registration Act (FARA) from 1938. The US Department of Justice organizes annually a report on foreign lobbying activities which is sent from the US Attorney General to the US Congress. The FARA annual reports contain the name and address of foreign agents, the name of the principals, the purpose of the agency and the amount of money in return for the agent's activities. First we separated the contribution data related to trade on agricultural and industrial goods from those involving trade in services and also the con- tributions from political purposes6. Then, each contribution related to trade on goods was mapped into 3-digit ISIC industries. This mapping process was repeated for each entry found on the FARA report for 33 countries in Latin America. We used the FARA reports for the years of 1997, 1998, 1999 and 2000. Trade related FARA contributions by Latin American exporters reach 60 million dollars during this period. Finally, we calculated the aver- age contribution over this four year period by 3-digit ISIC industry and by 6See appendix for more information on the foreign lobbying data. 4 country. One characteristic of the contribution is the high concentration by sector and country. The ISIC 3 digit industries with largest contributions are agri- culture and livestock production (ISIC 111), crude petroleum and natural gas production (ISIC 220) and manufacture of transport equipment (ISIC 384) with, respectively, 62, 26 and 8 percent of total contributions. Thus, this three industries account for 96 percent of total Latin America's con- tributions. Concentration is also very high when one looks to the share of different countries in some of these industries. In agriculture and livestock production, Colombia accounted for 98 percent of the total7, in crude petro- leum and natural gas production, Venezuela accounted for 99 percent of the total, and, in the case of transport equipment, Brazil was responsible for 98 percent of the total. We can also split the countries in three groups according to its geograph- ical localization. In this case each country can belong to South America or Caribe and Central America or NAFTA (used for Mexico). South Amer- ican countries are the largest contributors with 95 percent of total contri- butions. Mexico shows more diversified contributions. The Mexican sector with largest contribution is non-metallic mineral products (ISIC 230) with 34 percent of total contributions. For instance, in South America the sector with the largest contributions is agriculture and livestock production (ISIC 111), which accounts for 64 percent of total South American contributions. 7Colombian contributions on agriculture and livestock production were 100 percent given by that Nation's coffee producer association. 5 2.2 Tariff Preferences Tariff preferences are defined as the share of the Most-Favoured-Nation tar- iff that is waived to Latin American exporters under different preferential schemes. The data source is the excellent customs data set provided by the United States International trade Commission (USITC).8 It provides in- formation on the value of imports and duties paid under different import regimes from each country at the tariff line level (Harmonized System (HS) 8-digit). This allows us to calculate actual tariff preferences (i.e., those ac- tually granted at customs and not `on paper'). The advantage is that actual preferences capture the effects of non-trade barriers (agriculture and textile & clothing quotas, rules or origin, standards, etc...) on the value of the preference granted `on paper'. To calculate the tariff preferences at the 3 digit level of the ISIC clas- sification we proceeded as follows. We obtained data on duties and import values for each product exported by Latin American countries at the 8-digit level of the HS classification for the period 1997-2000. The average across these four year period was calculated by country and by HS 8 digit line. For the products exported by Latin American countries, we also collected data on duties and import values from the world, but that enter under the MFN regime. The tariff rate of each of these 8-digit tariff line product was then calculated for US imports from Latin American (both MFN and preferen- tial) and from the world under the MFN regime by simply dividing duties collected by the value of imports. This provided us with two tariffs: the (potentially) preferential tariff on each Latin American country, denoted tF, 8See dataweb.usitc.org. 6 and the MFN tariff, denoted t, at 8 digit HS level. Because the contribution data is only available at the 3-digit ISIC level, we needed to filter our tariff preferences from the 8-digit HS level to the 3-digit ISIC level. The tariff data was aggregated to the 3-digit ISIC classification using HS-8 digit exports of each Latin American country to the US within a 3-digit ISIC classification as weights (both for the Latin American tariff and for the MFN tariffs).9 Then, we define the US tariff preferences to Latin America as follows, = 1 - tF if tF t, t 0 otherwise. where it is clear that is censored between 0 and 1. Using (1) we notice that the closer is from 1 the larger is the tariff preference that exporters from a Latin American country in a respective 3-digit ISIC receive from the United States. The total number of US tariff preferences across countries and products used in this work is 1087. About 22 and 14 percent of the US trade preferences calculated correspond to values 0 and 1, respectively.10 Aggregating all data by country or 3-digit ISIC industry helps to identify which Latin American countries or products receive the highest and lowest US trade preferences. The countries which received the largest US trade preferences from 1997 to 2000 were Suriname, Bahamas and Trinidad and 9We used the same weights in order to avoid aggregation bias when moving from the tariff line level to the 3-digit ISIC level. 10About 61 percent of the values of which are zero are negative and were censored as described by equation (1). One reason for the existence of negative preferences in our data is the use of anti-dumping duties by the United States. Another reason, for agriculture products is the presence of in- and out-of-quota tariff rates. 7 Tobago with values for preferences of 0.97, 0.96 and 0.92, respectively. Al- though Mexico belongs to the North American Free Trade Area (NAFTA), seven Latin American countries received higher preferences than Mexico. One reason for the unexpected lower US trade preference to Mexico vis-a-vis other countries not members of NAFTA is that rules of origin in that trade bloc are very stiff, obligating many Mexican exports within NAFTA to use the US's MFN regime as shown by Estevardeordal et al. (2002). If the data aggregation is done by 3-digit ISIC industry then the sectors which received the highest US trade preferences from 1997 to 2000 are hunting and trapping (ISIC 113), other mining (ISIC 290) and manufacture of paper and paper products (ISIC 341) with values for preferences ranging from 0.98 to 0.99. Agriculture and livestock production (ISIC 111), manufacture of footwear (ISIC 324) and petroleum refineries products (ISIC 353) receive the lowest levels of US trade preferences with values of 0, 0.19 and 0.20, respectively. In particular, agriculture and livestock production is the only sector that received negative preferences from the US. Again, as mentioned above this is explained by the existence of in- and out-of-quota tariffs on agriculture products and antidumping duties. Note that tariff preferences granted by the United States at the tariff line level are usually full preferences, i.e., = 1. In our data set is a continuous variable for several reasons. First, the data is aggregated at the 3-digit ISIC level. Second, some of the preferential agreements have phase-out periods for preferential tariff elimination. Finally, even if there is full preferences granted on paper, some exports may still enter under the MFN regime, because they do not satisfy rules of origin. 8 3 Analytical setup Consider an economy in which consumers maximize a quasi-linear utility function m U = c0 + u(ck). k=1 Good zero is the numeraire. Given this functional form, there is no income or substitution effect on demand. The supply side is a specific-factor model where primary inputs into production are sector-specific capital and mobile labor. Production of good zero uses labor only under CRS, which fixes the wage rate. Thus, there is no general-equilibrium effect on the supply side either. Owners of sector-specific capital have an incentive to get politically organized and lobby for trade policies so as to raise capital's return. Resident owners of specific capital have mass zero in the population and consequently do not consider their consumption bundle or share of tariff revenue when lobbying the home government. The political game is as follows. Politically organized owners of specific capital, whether nationals or foreigners, lobby the domestic government for trade policies that are advantageous to them. For domestic import-competing producers, this means asking for tariffs on imports, whereas for foreign pro- ducers exporting into the domestic market, it means asking for tariff pref- erences. In order to simplify the setup, we will assume that ROW imports (imports from non-preferred countries) are sufficiently large to absorb the increase in preferential imports that would result from full preferences. This `large market' assumption ensures that there is no political rivalry between domestic and preferential-partner lobbies who in effect try to influence two 9 distinct and independent policy instruments (the MFN tariff and the rate of preference respectively). Lobbies move first, domestic ones by offering contributions conditioned on the MFN tariff in their sector of activity, and foreign (`preferred') ones by simultaneously offering contributions conditioned on the rate of preference. Then the government picks an m × 2 matrix of MFN tariffs and preference rates. Given the absence of general-equilibrium effects and the `large market' assumption, the game is a collection of independent principal-agent problems. The form of these principal-agent relationships differs, however, because the government's action is a continuous variable in the case of MFN tariffs and a binary one in the case of preferences. In the logic of common-agency models a la Bernheim-Whinston (1986b), domestic lobbies face the government with truthful contribution schedules, i.e. functions of the MFN tariff whose derivative is equal to the derivative of their own profit function. Foreign lobbies, by contrast, face the govern- ment with pairs of transfers corresponding to the two possible values of the preference rate (zero and full). Alternatively, one may think of the game between the government and foreign lobbies as a standard auction in which the latter buy indivisible preferences. This game is considerably simpler than a menu auction a la Bernheim-Whinston (1986a). Sticking to the principal- agent interpretation, each foreign lobby offers the smallest transfer inducing the government to grant preferences, which means, in the absence of hidden action, that the lobby keeps the entire protection rent. government, ignoring the simultaneous game between other (foreign and domestic) lobbies by virtue of the model's independence properties. Let 10 i = 1, ..., n denote trading partners (i.e. countries) and let k = 1, ..., m denote tariff lines (i.e., products). Let tk be the home country's MFN tariff on good k and tik {0, tk} be the preferential tariff applied on good k originating from preferential partner i. As indicated by the notation, preferences are either full (tik = 0) or nil (tik = tk). Interior values of tik will be obtained by aggregation in the empirical part. Define also ik = 1 - tik/tk as the `rate of preference' (what foreign lobbies are interested in). Let Ck(ik) i be the contribution schedule offered by foreign lobby k from country i to the home government, and Ck(tk) that offered by domestic import-competing lobby k. The home government's objective function is n m m V = Ci(ik) + Ck(tk) + aW(1, ..., n, t) i=1 k=1 k=1 where i = (i1, ..., im) is the vector of tariff preferences granted to partner i and t = (t1,...,tm) is the vector of MFN tariffs. Lobby (i, k)'s objective function net of contribution is vk = ik(ik) - Ci(ik). i Let V (0) be the value of the government's objective function when ik = 0 and V0 its value when lobby (i, k) does not contribute, i.e. when Ci = 0. The latter is the government's reservation value. In equilibrium, the two will necessarily be equal, but this equality is a property of the equilibrium, not a 11 part of the game's definition. The equilibrium of the game between foreign lobby (i, k) and the domestic government satisfies (ik,Ck) arg max(V + vk), i i (1) V (ik) V (0), (2) V (ik) V0. (3) Expression (1) ensures that the deal is jointly optimal for both parties, a standard requirement of incentive contracts. Inequality (2) is an incentive constraint whereby the government finds it profitable to choose the lobby's preferred action (grant positive preference). Inequality (3) is a participation constraint whereby accepting the deal is at least as good as leaving it. Trans- fers from lobby to government being costly, both inequalites are binding in equilibrium; thus, Ck(0) = 0, and Ck(ik) is (implicitly) determined, after i i elimination of independent terms and rearrangement, by Ck(ik) = -a W(1, ..., n,t) - W(01,...,0n, t) = -aW. i Because foreign exporters are sufficiently small, their entire export sup- ply at existing home prices cannot be larger than the home country import demand.11 This ensures that home prices remain unchanged after granting tariff preferences to foreign exporters. This implies that there are no changes in consumer surplus or producer surplus associated with the tariff prefer- 11This assumption was checked in the application to Latin American preferences in the US market. Indeed, at the six digit of the HS (most disaggregated level at which we can compare data) exports of each LAC country to the world are smaller than US imports from the world. 12 ences.12 Thus, the change in the importing country welfare is simply driven by the change in tariff revenue and the financial contribution received by the foreign lobby of exporters of good k in country i. Thus, equation (3) becomes: a Ck = -aW = -aTR - aCk Ck = -1 i i i TR (4) + a where TR is tariff revenue. Note that Ck enters the home country welfare i function, as these contributions represent additional income to the home country. Tariff revenue is defined as: TR = tkmk - (tk - tik)xik (5) where mk are total home imports of good k, xik are exports of good k by country i, which is lobbying for preferences. Using (5), (4) becomes, a Ck = i (6) 1 + a (tk - tik)xik The return to foreign lobbying by exporters from country i for tariff prefer- ences in good k is then given by the ratio of the increase in exports associated with the tariff preference over the foreign contribution (minus 1): 1 + a 1 rk = i (tk - tik)xik (7) a/(1 + a)(tk - tik)xik - 1 = a - 1 = a 12And as suggested above no incentives for domestic producers to counter-lobby the tariff preference. Note that this also assumes that the whole rent from the tariff preferences is captured by the exporters and not the importers. 13 Thus, the inverse of the weight given to social welfare in the home country government's objective function yields the return to foreign lobbying for tariff preferences. The higher is the weight granted to social welfare in the govern- ment's objective function and the lower are the returns to foreign lobbying. Solving (6) for the tariff preference ik: i ik = 1 + a Ck (8) a tkxik A stochastic version of (8) will allow us to estimate the parameter a in the US government's objective function, and then induce the returns to Latin American exporters lobbying for tariff preferences. Note that the coefficient (1+a)/a needs to be larger than one by definition (unless a < 0). A discussion of the empirical methodology can be found in the next section. Recent estimates of a for the United States (Gawande and Bandyopad- hyay, 2000 and Goldberg and Maggi, 1999) suggest figures above 100. This would imply that the return to foreign lobbying is very small: below 1 per- cent. But this seems at odds with the figures discussed in the previous sections. On average for the period 1997-2000, Latin American trade related contributions under FARA amounted to US$ 15 millions, whereas the value of preferences, calculated as txf, was around US$ 4 billions. This suggests a much higher return (around 26000 percent) and a value for a close to zero. Two questions may be raised. First, one may ask why not stop here if the objective is to obtain the return to foreign lobbying in Latin America. What's wrong with the 26000 percent number calculation?. The problem is that these are average numbers and contributions may not necessarily reflect tariff pref- erences, but other trade-related issues. So the 26000 percent number may 14 be an underestimate of returns to preference lobbying by Latin American exporters to the US. On the other hand, official FARA contributions may seriously under-estimate total (including non-official) contributions. If this was the case, returns to foreign lobbying can actually be much smaller. Per- haps, more importantly, tariff preferences are not exclusively due to foreign lobbbying contributions. Other political and economic factors entered also into consideration. In order to check, whether the tariff preferences reflect foreign lobbying, as suggested above, we will estimate equation (8). As long as the under-reporting or other factors influencing the granting of tariff pref- erences is consistent (i.e. proportional) within countries and industries, and econometric approach would allow us to estimate consistently the parameter a and therefore the returns to foreign lobbying r, using equation (8) and introducing country (i) and industry (k) dummies. The second question that may be raised is: why not using the first order condition of the government's problem to estimate a, as in previous work for the United States (Gawande and Bandyopadhyay, 2000, and Goldberg and Maggi, 1999). Two reasons for this. First by using the government's `no regret' condition, we do not need to assume that contribution functions (and welfare) are differentiable to esti- mate the parameter a. Grossman and Helpman (1994) aruged that there are good reasons to assume that contribution functions are indeed differentiable; otherwise the lobby may end up given up on a profitable exchange of contri- butions and tariffs (see footnote 8, page 841).13 However, when considering 13In their paper, they also suggest the use of truthful contribution functions, which are differentiable and have the additional interesting property of being coalition proof. Note that as disccused above, given our setup there are no incentives for foreign lobbies to form 15 lobbying for tariff preferences in the US market, contributions are unlikely to be continous. The reason is that most tariff preferences granted by the United States are either full preference or no preference, i.e., at the tariff line level only takes the value 0 or 1. Thus, given that contribution functions for tariff preferences are not differentiable, it is not possible to work with the first order condition of the government's problem. To our knowledge, all empirical studies of lobbying and trade policy in the US have used the assumption of truthful contribution function when estimating the first order condition of the government's problem (to our knowledge, none of the existing studies has yet focused on tariff preferences, though). If this assumption does not hold one could obtain bias estimates. As discussed in section 5.2, thruful contribution may to lead to upward bias estimates of a. Second, to estimate a using the first order condition of the government's problem, one would need to obtain estimates of the elasticity of export supply of each Latin American country to the United States (and not to the world) by product for the late 1990s. These are not available, to our knowledge, and would represent a tedious exercise, which would probably raise additional problems. For these two reasons we prefer to work with the non-regret condition rather than the first order condition of the government's problem in order to obtain estimates of the weight granted by the US government to social welfare, a, and the returns to Latin American lobbying in the United States, r. coalitions. 16 4 Empirical Strategy Our empirical strategy consist of estimating a stochastic version of equation (8). Because the endogenous variable, , is bound at 0 (no preference) and 1 (full preference) a tobit estimator is necessary. When aggregating tariff and trade data from the 8 digit HS level to the 3 digit ISIC level one may introduce some heteroscedasticity due to group aggregation if one believes that the equation to be estimated is determined at the 8 digit HS level and that the number of 8 digit HS tariff lines in each 3 digit ISIC industry is not the same. In order to correct for this potential for heteroscedasticity we follow a parametric correction suggested by Dickens (1990). It consist in estimating (8) and then running the error term against the inverse of the number of 8 digit HS tariff lines in each 3 digit ISIC industry. The constant of such a regression provides an estimate of the variance of the industry level component of the error term, whereas the coefficient in front of the inverse of the number of lines provides and estimate of the variance of the tariff line level component of the error term. To obtain asymptotically efficient estimates one re-weighs each observation at the industry level using these variance estimates.14 An important problem with the estimation of equation (8) is that the righ-hand-side is endogenous. Indeed, the ratio of contributions to Latin American exports is obviously endogenous to the preference margin. Previ- ous empirical studies of the political economy of tariffs in the US, such as Goldberg and Maggi (1999), Gawande and Bandyopadhyay (2000) use factor 14We alternatively provide estimates of the standard error using boot-strapping. STILL TO COME. 17 shares and traditional exogenous political economy variables to control for endogeneity problems in a protection equation similar to the one shown in 8. We follow a similar approach using factor shares and their products to control for endogeneity problems in the estimation of (8).15 However, other political economy variables used in the mentioned studies like concentration ratios and unionization indexes are not available for most Latin American economies and their estimation are beyond the scope of this work. We, therefore, proceeded using a two-stage tobit. In the first stage we run the ratio of contributions to Latin American exports (C/(txF)) on the instruments described above. In the second stage, we use the fitted value of the above regression instead of the ratio itself as an explanatory variable.16 Finally, there could be systematic differences in the tariff preferences granted by the United States across countries or products. For example, Mexico is the only member of Nafta; to benefit from CBI preferences you need to be a Caribbean country. Also, the tariff lines included in the GSP preferences granted by the US to each Latin American country is almost exogenously given. Thus, in order to control for any systematic differences across countries or products in tariff preferences granted by the United States to Latin American countries we included country and industry time dummies. 15See appendix on the description of the instruments used. 16Results reported in the next section are robust to the two-stage tobit methodology suggested by Gawande and Bandyopadhyay (2000), which consisted of entering the error term of the first stage regression as an additional variable in the second stage to correct for the endogeneity of C/(txF). We also provide standard error estimates using boot-strapping to avoid any problems due to inefficient estimates linked to the two-stage procedure. 18 5 Results Table 1 provides results for the estimation of a stochastic version of (8) for our sample of LAC countries. The first column provides a simple estimation of (8) without constant, country or product dummies and without any correction for endogeneity or heteroscedasticity. The second column provides estimates with the endogeneity correction and the third column adds a constant to the endogeneity correction. The fourth column further corrects for the poten- tial heteroscedasticity introduced by aggregating the data at the industry level when preferences are determined at the tariff line level. The paramet- ric correction we introduced follows Dickens (1990). The fifth column adds country and industry dummies and corrects both for heteroscedasticity and endogeneity of our right-hand-side variable. The sixth column is the same as in the fifth column, but standard error are calculated using bootstrapping. Estimates of (1 + a)/a oscillate between 0.6 and 4.8. But these two extremes are obtained before a constant is introduced into the regression. Once the constant is introduced estimates of (1 + a)/a are relatively robust. They osciallate between 1.17 and 1.48. This in turn implies that estimates of a oscillate between 2 and 6; and that returns to foreign lobbying by Latin American exporters in the US provides returns of 17 to 48 percent.17 Our prefered estimates are those in the last two columns as the control for any systematic country or product variation in preference (as well as for endogeneity and heteroscedasticity). The point estimate for a is around 2.08 17Note that using the estimated standard errors for a and r non-linearly, we can reject the hypothesis that a > 14 (and therefore r < 8 percent) in all regressions with 95 percent confidence. 19 and therefore r = 48 percent. Table 2 provides the same estimates as in Table 1 but the sample excludes Mexico. The reason is that Mexico is a member of Nafta and may therefore be subject to a very different regime than the other Latin American countries that our country and product dummies may not capture. Moreover, Mex- ico accounts for more than 50 percent of total preferential exports of Latin American into the US.18 Results are very similar to the ones reported in Ta- ble 1. If we go by our prefered estimates in the last two columns of Table 1, it seems that returns of foreign lobbying are lower for the non-Mexico Latin American countries. Indeed, returns to foreign lobbying contributions are es- timated around 32 percent instead of 48 percent when we include Mexico.19 Finally, note that lobbying only explains a very small part of the tariff preference variation. Without the inclusion of country and product dummies, the pseudo-R2 are well below 1 percent. Many other determinants of tariff preferences are captured by the product and country dummies, as well as the error term. Nevertheless, the returns to foreign lobbying seem to be relatively high. 5.1 Is the estimate for a too low? HERE WE EXPLAIN WHY IF CONTRIBUTIONS FUNCTION ARE NOT CONTINOUS OR DIFFERENTIABLE WORKING WITH THE GOVERN- MENT FOC CAN BIAS THE ESTIMATES OF 'a' UPWARDS. WORKING 18Almost all enter under Nafta. 19The estimate for a is around 3.1. Note that this suggests that monetary contributions from different countries are valued differently by the US government. 20 WITH THE NON-REGRET CONDITION ALLOWS US TO AVOID THIS 6 Concluding remarks Almost 50 percent of US imports from Latin America enter under a prefer- ential tariff regime. This paper explores the importance of lobbying in the US by Latin American exporters in explaining the extent of tariff preferences granted by country and product. Empirical results suggest that lobbying by Latin American exporters is indeed a significant determinant of tariff preferences, although not a very important one. Other country and product characteristics seem to explain a much larger share of the variation in tariff preferences granted by the US government to Latin American exporters. However, it pays Latin American exporters to lobby for tariff preferences in the US. Returns to foreign lobbying are estimated to be around 50 percent in the case of Latin American exporters lobbying in the US. Last, but not least, we provide a methodology to estimate the weight given to social welfare in the US government objective function (relative to lobbying contributions) in the case of non-differentiable contribution function. Indeed, in the case of tariff preferences in the US, the foreign contribution function is likely to be non-continuous as preferences are generally either fully granted or not granted at all in the United States. Relaxing the continuous contribution function assumption provided us with an estimate of the weight granted to social welfare (a) around 2, partly solving the empirical puzzle of the empirical literature, where estimates for the US are above 100. 21 References [1] Cadot, Olivier, Jaime de Melo and Marcelo Olarreaga (2003), "The pro- tectionist bias of duty drawbacks: an application to Mercosur", Journal of International Economics 59, 217-229. [2] Cadot, Olivier, Jean-Marie Grether and Marcelo Olarreaga (2003), "Pro- tection for sale in India: who buys and for how much?", mimeo, The World Bank. [3] Dickens, William (1990), "Error components in grouped data: is it ever worth weighting?", Review of Economics and Statistics 72, 328-333. [4] Estevardeordal, Antoni; Jaime de Melo, Olivier Cadot, Akiko Suwa- Eisenmann and Bolormaa Tumurchudur (2002) "Assessing the Effect of Nafta's Rules of Origin", Working Paper, World Bank. [5] Gawande, Kishore (1997), "Generated regressors in linear and non linear models"; Economic Letters 54, 119-126. [6] Gawande, Kishore and Bandhopadhyay, U. (2000), "Is protection for sale? A test of the Grossman-Helpman Theory of Endogenous Protec- tion"; Review of Economics and Statistics, 139-152. [7] Gawande, Kishore and Pravin Krishna (2002), "The political economy of trade policy: empirical approaches"; in J. Harrigan, ed., Handbook of International Trade, New York: Basil Blackwell. 22 [8] Gawande, Kishore, Sanguinetti Pablo and Bohara, Alok (2001), "Exclu- sion for sale: Evidence on the Grossman-Helpman theory of free trade agreements", mimeo, University of New Mexico. [9] Goldberg, Pinelopi and Giovanni Maggi (1999), "Protection for Sale: an Empirical Investigation"; American Economic Review 89, 1135-1155. [10] Grossman, Gene, and E. Helpman (1994), "Protection for sale", Amer- ican Economic Review 84, 833-850. [11] ­ (1995), "The Politics of Free Trade Agreements, American Economic Review 89, 667-690. [12] ­ (2001), "Special Interest Politics", MIT press. [13] Mitra, Devashish (1999), "Endogenous Lobby Formation and Endoge- nous protection: A Long-Run Model of Trade Policy Determination"; American Economic Review 89, 1116-1134. [14] Mitra, Devashish, Thomakos, Dimitrios and Ulubasoglu, Mehmet (2002), "Protection for sale in a developing country: democracy ver- sus dictatorship" Review of Economics and Statistics 84. 23 Table 1: Estimating returns to LAC lobbyinga (1) (2) (3) (4) (5) (6) (1 + a)/a 0.59 4.76 1.22 1.17 1.48 1.48 (.36) (1.02) (.60) (.64) (.47) ( ) Constant 0.57 0.55 0.31 0.31 (.20) (.20) (.13) ( ) Hetero. corr. No No No Yes Yes Yesb Country dum. No No No No Yes Yes Industry dum. No No No No Yes Yes # obs. 1062 1062 1055 1055 1055 1055 Pseudo R2 NA NA .002 .001 .27 0.27 a ^ -2.43 0.27 4.55 5.88 2.08 2.08 r^ 3.70 .22 .17 .48 .48 aFigures in parenthesis are standard errors. The estimation technique is a two-stage tobit to control for th endogeneity of our right-hand-side variable, except for the first column where a simple tobit is used to estimate th parameter (1 + a)/a. bStandard errors are estimated using boot-strapping. EXPLAIN HOW WE DID THIS. Table 2: Excluding Mexico (Nafta preferences)a (1) (2) (3) (4) (5) (6) (1 + a)/a 0.59 4.10 1.22 1.20 1.32 1.32 (.36) (0.91) (.57) (.51) (.41) ( ) Constant 0.57 0.54 0.34 0.34 (.20) (.21) (.13) ( ) Hetero. corr. No No No Yes Yes Yesb Country dum. No No No No Yes Yes Industry dum. No No No No Yes Yes # obs. 1025 1025 1025 1018 1018 1018 Pseudo R2 NA NA .003 .003 .28 0.28 a ^ -2.43 0.32 4.55 5.00 3.13 3.13 r^ 3.10 .22 .20 .32 .32 aFigures in parenthesis are standard errors. The estimation technique is a two-stage tobit to control for th endogeneity of our right-hand-side variable, except for the first column where a simple tobit is used to estimate th parameter (1 + a)/a. bStandard errors are estimated using boot-strapping. EXPLAIN HOW WE DID THIS. Data Appendix Foreign Contribution Data The US department of Justice provides data on foreign lobbying through the Foreign Agent Registration Act (FARA) annual reports. The FARA annual reports contain the name and address of foreign agents, the name of the principals, the purpose of the agency and the amount of money in return for the agent's activities. Following the information contained on the FARA website20, an agent of a foreign principal is any individual or organization whose activities are directed by a foreign principal filling one of the criteria below21: - Engages in political activities; - Acts in a public relations capacity for a foreign principal; - Solicits or dispenses anything of value within the United States for a foreign principal; - Represents the interests of a foreign principal before any agency or official of the U.S. government; Since this paper studies the role of Latin American contributions on trade preferences granted by the US, we need to eliminate from the original data those contributions not related to trade. To make the process easier we follow Krishna et al. (2001) and organize the original FARA contribution data for the years of 1997 through 2000 in six categories as follows: (1) Tourist boards or private and/or government chambers of commerce that encourage general business contacts; (2) Government to government contacts; (3) Service industries; 20The FARA, its annual reports and additional legislation on foreign lobbying can be found on the electronic address www.usdoj.gov/criminal/FARA. 21See also the helpfull "Q&A" document at http://www.usdoj.gov./criminal/fara/q_A.htm. 26 (4) Agriculture or raw material industries; (5) Foreign political parties that were campaigning among ethnic diaspo- ras or seeking U.S. government recognition for their cause; (6) Manufacturing industries; The FARA annual reports for the years of 1997-2000 totalized 619 entries for the 33 countries of Latin America. The average per year of all the entries reached 102 million dollars. From the six categories identified above only the ones with numbers (4) and (6) are clearly related to trade on goods. Then, we decided to eliminate contributions from categories (1), (2), (3) and also those purely from political purposes located on category (5). Contributions from categories (4) and (6) were mapped into 3-digit ISIC codes using the name of the contribution's principal and the purpose of the agency. The table below gives some interesting features about the compo- sition of the data set. It is interesting to note that from 619 entries in the FARA reports only 109 were related to lobbying from specific industries in the manufacturing or agriculture sectors. Besides, only 15 percent of total contributions from Latin American countries were done directly by private agents like those contained in categories (4) and (6). Category (4) acounted for 25 percent of the entries and categories (6) for the remaining 75 percent. Tariff Preferences and Trade Data Data on US's duties and import values from the world and for each Latin American country from 1997 to 2000 was obtained from the USITC web site (dataweb.usitic.gov). To convert the data from the 8-digit level of the harmonized system to the ISIC 4-digit classification, we used the filter built by Jerzy Rozanski from the World Bank 27 Instruments The GTAP database for the year of 1995 provides data on sectorial factor ex- penditure and output value for many countries in Latin America. When data is not available for a particular country, the GTAP database provides figures for the a group of countries (including the missing one). Factor expenditure in any sector for each Latin American country is divided in capital, skilled labor, unskilled labor, natural resources and land expenditures. Information contained in the GTAP manual was used to filter the data from the GTAP classification to the 3-digit ISIC level. When the data on factor expenditures and value of output was filtered to the 3-digit level of the ISIC classification, factor shares were calculated dividing each factor expenditure by the value of output. For those countries where particular information was not available the factor shares of the regions they belonged were used instead. List of Latin American Countries Antigua, Argentina, Aruba, Bahamas, Barbados, Belize, Bermuda Bolivia, Brazil, British Virgin Islands, Cayman, Colombia, Costa Rica, Chile, Do- minican Republic, Ecuador, El Salvador, Guatemala, Guyana, Haiti, Hon- duras, Jamaica, Mexico, Netherland Antilles, Nicaragua, Panama, Paraguay, Peru, Santa Lucia, Suriname, Trinidad and Tobago, Uruguay, and Venezuela. 6.1 ISIC 3-digit sectors ISIC Description 111 Agriculture and livestock production 112 Agriculture Services 113 Hunting, trapping and game propagation 121 Forestry 122 Lodging 130 Fishing 210 Coal Mining 220 Crude Petroleum and Natural Gas Production 28 230 Metal ore mining 290 Other Mining 311 Food Manufacturing 313 Beverage Industries 314 Tobacco Manufactures 321 Manufacture of Textiles 322 Manufacture of Wearing Apparel, except footwear 323 Manufacture of Leather and products of Leather, and substitutes 324 Manufacture of footwear, except vulcanized or moulded rubber 331 Manufacture of wood and wood products, including furniture 332 Manufacture of furniture and fixtures, except primarily of metal 341 Manufacture of paper and paper products 342 Printing, publishing and allied products 351 Manufacture of industrial chemicals 352 Manufacture of other chemical products 353 Petroleum Refineries 354 Manufacture of Miscellaneous products of petroleum and coal 355 Manufacture of Rubber Products 356 Manufacture of Plastic Products not elsewhere classified 361 Manufacture of pottery, China and Earthenware 362 Manufacture of Glass and Glass products 369 Manufacture of other non-metallic mineral products 371 Iron and Steel basic industries 372 Non-ferrous metal basic industries 381 Manufacture of fabricated metal products, except machinery 382 Manufacture of machinery except electrical 383 Manufacture of electrical machinery apparatus 384 Manufacture of transport equipment 385 Manufacture of professional and scientific, 390 Other manufacturing industries 29