Trade, Technological Change, and Wage Inequality: The Case of Mexico

Transcription

1 Trade, Technological Change, and Wage Inequality: The Case of Mexico Andrea Waddle University of Richmond January 2018 Abstract In the decade following the Mexico-U.S. trade integration, the manufacturing skill premium rose by roughly 50 percent in Mexico while also rising in the U.S. Standard trade theory predicts that when countries with different levels of skilled labor integrate, the skill premium should fall - not rise - in the skill-scarce country. In this paper, I reconcile theory and data by building a model in which intermediate goods are produced using rented technology. After integration, producers in Mexico begin to rent technologies from the United States, which are more advanced and, hence, more skill-intensive. This increases the skill premium in Mexico due to the adoption of the more advanced technology. Furthermore, the skill premium in the U.S. rises modestly due to increased investment in this technology, which is driven by the increased marginal return on the technology arising from its adoption in Mexico. The mechanism is supported by plant- and industry-level evidence: Mexican plants and industries which are more integrated into the U.S. supply chain have higher skill premia than their nonintegrated counterparts. The calibrated model can account for about 12% of the increase in the skill premium in Mexico. I am indebted to Tim Kehoe and Ellen McGrattan for valuable advice and guidance. I am grateful for comments and suggestions from Ariel Burstein and two anonymous referees. I would also like to thank Cristina Arellano, Evelyn Birnbaum, Alessandro Dovis, Larry Jones, oshinori Kurokawa, Fabrizio Perri, Collin Rabe, Joe Steinberg, David Wiczer, and the participants of the Trade workshop at the University of Minnesota for their useful comments and discussions. I d like to thank Tyler Pike for excellent research assistance, and I acknowledge the financial support of the University of Minnesota Doctoral Dissertation Fellowship. The usual disclaimers apply.

2 1 Introduction Standard trade theory has stark predictions for how factor prices should respond to trade integration between a skill-scarce and a skill-abundant country. In particular, models that are based on the Heckscher Ohlin (henceforth H O) theory predict that the ratio of wages paid to skilled versus unskilled workers (the skill premium) should rise in the skill-abundant country and fall in the skill-scarce country when the two countries open to trade with one another. A puzzle that has arisen in the context of this prediction is that when integrating with the world economy, many skill-scarce countries instead experience rising skill premia. Mexico is the canonical example of a country whose skill premium not only rose, but rose by much more than that of its more-developed counterpart, the United States, during the period in which Mexico opened its borders to trade with the United States. These observations have led many researchers to conclude that skill-biased technological change (SBTC), not increased openness to trade, has driven changes in developing economies skill premia. In this paper, I argue that trade liberalization, by stimulating investment in skill-biased technologies and facilitating cross-border adoption of these technologies, plays a significant role in explaining the aforementioned facts. I modify a standard trade model to include trade in technology, which occurs through the integration of supply chains across borders. I use the case of the Mexican trade liberalization and integration into the supply chain of American companies to explore the impact that technological transfer, which takes place as a part of this integration, has on the wages of workers in Mexico and the United States. I calibrate the model using surveys of the Mexican and U.S. manufacturing sectors, and I find that a reduction in barriers to trade in goods and technology that matches the change in trade flows observed during the liberalization between the two countries in the late 1980 s can account for approximately12% of the observed increase in the skill premium in Mexico and produces results for the United States that are consistent with the data. I furthermore find that if there were complete erosion of barriers to using foreign technology, the skill premium would have risen by 30% in Mexico. To support the quantitative analysis, I provide empirical evidence that Mexican plants and industries that become more integrated with the United States via more trade have higher skill premia on average. Moreover these plants and industries have greater increases in their skill premia in the late 1980s and early 1990s than their non-trading counterparts. These results suggest that trade connections are an important determinant of skill premia and that supply chains are a channel 1

3 through which technology is transferred. I show that the use of intermediate exports is an important predictor of the skill premium, indicating that supply chain relationships play an important role in determining the skill premium in a given industry. In order to assess the quantitative importance of supply chains on the skill premium, I adapt a standard trade model to allow for trade liberalization to increase both trade in goods and trade in ideas. I model ideas as technology capital, similar to the model in McGrattan and Prescott (2009), but I allow for technology capital to be rented from final goods producers, who own and invest in the stock of technology capital, to intermediate goods producers, who use it. I model trade liberalization as a reduction both in tariffs on goods and in distortions to the flows of royalties. I discipline my exercise using manufacturing data from Mexico and the United States. My model differs from those in the existing literature by incorporating two key ingredients. First, I allow for skill-biased technology to be endogenously accumulated by permitting firms to invest in a stock of technology that is assumed to be skill-augmenting. I consider final goods producers who own and invest in technology capital. Intermediate suppliers rent this technology capital in order to produce an intermediate product that will be a component of the final good. Consider for the moment a two-country world in which both countries are in autarky. When the countries open to trade, the final goods producer does not need to open a plant in the foreign country in order to use his technology capital there. Instead, the producer can rent his technology to an intermediate goods producer that is already operating in the foreign country. This, in turn, increases the marginal product of a unit of technology capital since it can now be used by additional intermediate goods producers. It is key that technology capital is non-rivalrous so that using it in multiple locations actually increases its marginal productivity. Opening to trade, therefore, increases the return to technology capital, as in McGrattan and Prescott (2009). Firms respond to these increased returns to their technology capital by investing more. I refer to this as the investment channel, and it is the main channel that drives the increase in the skill premium seen in the United States. This phenomenon is consistent with recent empirical work in Goel (2012) that provides evidence that firms in the United States respond to increased trade opportunities by increasing spending on innovation. Moreover, it is consistent with extant literature that concludes that the rise in the skill premium in the United States is driven primarily by technological change. Note that this does not mean that opening to trade plays no role in increasing the skill premium, but rather, that its role manifests as an increase in technology, driven by increased returns to investment in that technology. Second, I allow technology capital to be rented across borders. I provide plant- and industrylevel evidence that royalties paid as a percentage of output in Mexico co-varies positively with the skill-premium, meaning that industries that use more rented technologies are those that have larger 2

4 skill premia; I take this as evidence of the transfer of technology. Allowing for technology to be transferred through rental is the key to having the skill premium rise in both countries because it causes the skill premium to rise in the United States via the investment channel, as discussed above, while it causes the skill premium to rise in Mexico by what I will call the adoption channel. The adoption channel arises when intermediate goods-producing firms in one country begin to produce for the supply chain of the final goods producer in the other and, therefore, adopt the technology of the final goods producer from the foreign country. In the model, intermediate goods producers in Mexico choose to adopt U.S. technology and supply U.S. final goods producers much more than vice versa. This is because, in the initial steady state, U.S. technology is much more productive than Mexican technology. Therefore, the adoption channel is the dominant force driving the increase in the skill premium in Mexico. In my calibrated model, I find that moving from the pre-reform levels of tariffs and barriers to technology to the post-reform levels induces a skill premium increase of 5 percent in Mexico and less than 1 percent in the United States. This accounts for about 12% of the observed rise in the Mexican skill premium and 6% of the rise in the skill premium in the U.S. The adoption channel is key to obtaining these results. If tariff alone are lowered, leaving the barriers to the use of foreign technology in place, there is no significant rise in the skill premium. Furthermore, this policy change results in trade flows that increase by only a small fraction of the change that is observed in the data. This is consistent with a large literature on trade liberalizations, which shows that even large reductions in tariffs rarely result in significantly increase trade on impact. In the model, it is due primarily to the fact that the barriers to the use of U.S. technology capital in Mexico are much higher than tariffs, and so the size of this friction swamps the impact of tariffs. My quantitative results are disciplined by manufacturing data for the United States and Mexico. Because my interest lies primarily in how the skill premium changed in Mexico, I target the level of the skill premium in the initial period. I use aggregated industry data from Mexico on royalty payments to identify the parameter that governs the importance of technology capital in production. In particular, I match royalty payments as a percentage of payroll payments in the period before trade liberalization. I use as the pre-reform period; as documented below, the majority of Mexican trade reforms began in I also set the relative productivity of the manufacturing sectors in the two countries, the relative supply of skilled workers in each country, and the skill intensity of each industry to match the data in In order to analyze the impact of trade reform on the skill premium, I then conduct an experiment where tariffs on goods and frictions on the use of foreign technology capital are lowered. While direct observation of the reduction in tariffs that occurred in the data is possible, I am not able to directly observe a 3

5 measure of the distortions on foreign technology capital. This is because things such as the protection of intellectual property would have a strong impact on a firm s willingness to rent proprietary information to other firms and these protections changed substantially over the period of interest. Therefore, I set initial distortions to match the industry-by-industry fraction of intermediate goods that are traded between the two countries in both the pre- and post-reform periods. My baseline experiment includes both a change in tariff rates between the U.S. and Mexico and a reduction in the aforementioned distortions. I find that both policy changes increase the skill premium in both countries; however, the impact of a change in the former is much smaller than the impact of a change in both. This has implications for how we think about the impact of trade liberalization on the skill premium. My findings suggest that, even when tariffs are reduced by the same margins in two developing countries, the impact of these liberalizations could be very different. In particular, a country that integrates more fully into the supply chain of a more advanced country is the beneficiary of the diffusion of technology capital. This diffusion is the primary force that drives changes in the skill premium. Countries that simply lower tariffs but do nothing to protect the intellectual property rights of firms that are developing more productive technologies will be less able to use these technologies, which will in turn imply that the skill premium will not increase by as much in these countries. This results is significant, as part of the puzzle facing this literature has been the fact that developing countries experience very different impacts of trade liberalization on the skill premium. I conduct an additional experiment in which I consider a liberalization between two developed countries, such as the U.S. and Canada. I use the calibrated parameters of the baseline model, but set skill abundance, productivity, and tariffs to the observed levels in these two countries in I then exploit the change in trade that occurred after the signing of the Canada-U.S. Trade Agreement to test the implications of the model for this kind of liberalization. I find that, in part because the H O forces are no longer playing a large role, this has a larger impact on the U.S. skill premium than the integration with Mexico did. This is primarily due to the investment channel. U.S. firms face fewer distortions to using their technology capital in Canada than they do to using it in Mexico. We can think of this as being due to stronger intellectual property protection laws in Canada, which look more similar to those in the U.S. than their Mexican counterparts. Therefore, both U.S. and Canadian firms exploit the increased opportunity to use their non-rivalrous technology capital in the foreign country, thus increasing the incentive to invest in it. This causes the stock of technology capital to rise in both countries, inducing an increase in the skill premia. Because this exercise is not fully calibrated, I do not place a heavy weight on its quantitative implications; however, qualitatively, trade volumes between the U.S. and Canada increase more than they do with the 4

6 liberalization between Mexico and the U.S. and the skill premium in the U.S. is more heavily impacted than in after the liberalization with Mexico. Contribution to Related Literature There is a large body of literature dealing with the rise of the skill premium in the United States, and a somewhat smaller literature on the rise of the skill premium in Mexico. Studies such as Feenstra and Hanson (1996), Feenstra and Hanson (1997), and Grossman and Rossi-Hansberg (2008) have shown that increasing imports of intermediate goods from less-developed countries can increase skill premia in advanced economies. For a useful summary of articles that have explored the behavior of the skill premium of developing countries as they open to trade, see Goldberg and Pavcnik (2007). The papers that are most closely related to my own are Feenstra and Hanson s 1996 empirical and theoretical work on the importance of foreign direct investment (FDI) in Mexico, and Bustos 2011 exploration of skill upgrading in Argentina following the country s accession to MERCO- SUR. Empirically, Feenstra and Hanson show that regions with a higher proportion of inward FDI from the United States have greater increases in the relative demand for skilled labor. Furthermore, they build a theoretical model which rationalizes this prediction; capital is complementary with skilled labor, and as capital flows from the United States to Mexico via FDI, demand for skilled labor rises in Mexico. The Mexican subsidiary of the multinational in Mexico produces a less-skilled intermediate which is then substituted for less-skilled workers in the United States. Thus, the relative demand for unskilled workers falls in the United States as well. I see my paper as a complement to Feenstra and Hanson s work. At the aggregate level, flows of foreign direct investment between Mexico and the United States did not rise substantially until the mid-1990s, after the North American Free Trade Agreement (NAFTA) was established (GAO, 2005), and, as such, post-date the observed growth in the skill premium in Mexico. I focus on the transfer of technology through non-ownership channels precisely because trade increases substantially before NAFTA but direct investment does not. I provide evidence that supply chains are an important channel through which this transfer occurs. The mechanism proposed in Feenstra and Hanson s paper is also similar to what I propose. However, in their setup, the investment channel that I describe is not present. This is because the type of capital they consider is physical capital, which can be only used in one location at a time. I instead consider technology capital, which can be used in multiple locations at once. Therefore, once a firm has more than one location in which to use its technology, it has an increased incentive to invest in it. This is the primary driver of the increase of the skill premium in the United States in my model, whereas in the Feenstra and Hanson model, 5

7 the increase in the skill premium in the United States is primarily driven by Stolper Samuelson effects. In Bustos (2011), the author analyzes firm-level panel data covering the period of the Argentinian trade liberalization with Brazil. She documents that there is a steep increase in the demand for skill in Argentina following said liberalization, which is driven primarily by skill-upgrading within firms. She further finds that exporters engage in upgrading activities more quickly than non-exporters and that this skill-upgrading is accompanied by investments in new technologies. She then presents a modified Melitz model that can rationalize these observations through a mechanism that is similar but distinct from the one that I consider here. In Bustos model, there is a fixed cost to exporting, as well as a fixed cost to operating a technology that is both more productive and more skill-intensive than the alternative. The fixed-cost setup causes there to be a cut-off level of productivity above which firms will choose to both export and use the better, more skill-intensive technology. Thus, the main driver of skill-grading in this model operates through the standard Melitz logic: a reduction in tariffs causes the per-unit profit of exported goods to increase, thus reducing the cut-off level of productivity needed to export and upgrade to the better technology. Therefore, more firms export and operate this advanced technology after the trade liberalization. There is a similar observation as a result of trade liberalization in my model; however, my model differs both in terms of the key assumption and the implications for the impact of a trade liberalization. First, the key assumption in my model is that there is a stock of technology, which grows through investments made by the final goods producer and is rented to the intermediate goods producer, which will produce a part for the owner of that particular technology. The producer of an intermediate variety, i.e., a part used to produce the final good, is allowed to produce for both the domestic and the foreign the final goods producer but must operate a technology that is specific to the particular final goods in order to do so. A trade liberalization impacts the intermediate goods producer by making the part that they produce less expensive for the foreign final goods producer, thus inducing them allocate more of their production capacity towards producing intermediates for the foreign final goods producer. Because they are forced to use the technology of the foreign goods producer in order to produce a part for this supply chain, the domestic firm will become more skill-intensive if the foreign technology is skill-augmenting and there is a larger stock of technology capital in the foreign country. This, in turn, will cause the domestic skill premium to increase. Moreover, because technology capital is non-rivalrous, the fact that the foreign final goods producer s technology is now being used more intensively means that there is a greater return to investing in that technology, thus causing the final goods producer to increase investment and therefore the stock of this skill-augmenting technology. The adoption of foreign technologies thus induces an increase in skill intensity in both countries. The asymmetry of the two countries is one of the primary drivers of the greater increase in the skill premium in the less skill-abundant 6

8 country; the stock of technology capital in the skill-abundant country is much larger and, therefore, when the skill-scarce country begins to produce for the supply chain of the skill-abundant one, the skill premium in the skill-scarce country will increase dramatically. This asymmetric response will not occur in the framework proposed by Bustos and so, while her model can generate increasing skill premia in both countries, it is ill-equipped to explain why Mexico s skill premium increased by so much more than that of the United States. Additionally, in my analysis, technology diffusion is the main driver of increases in the skill premium, whereas in her paper, firms are choosing to operate technologies that already existed in their country. The mechanism proposed in her paper implies that when different countries open to trade, they should experience increases in the skill premium that are of the same magnitudes, all else equal. On the other hand, in my analysis the extent to which countries are integrate into the supply chain of a foreign country, thereby allowing them to access foreign technology, is the key determinant of increases in the skill premium; therefore, countries that open to trade but do not attract foreign technologies via supply chain integration will see much lower increases in their skill premia than those that do. I also contribute to the emerging literature on the interaction between trade, technology, and inequality. My work complements papers such as Acemoglu (2003), Acemoglu, Gancia, and Zilibotti (2012), and Burstein and Vogel (2017), which all address the idea that trade and technological innovation are linked. Acemoglu (2003) and Acemoglu, Gancia, and Zilibotti (2012) both build theoretical models in which globalization, in the form of trade or offshoring, can induce skill-biased technical change through increasing the price of skill-intensive goods and, thus, the profitability of investing in the technologies that are needed in order to produce these goods. The logic which supports the increase in skill-biased technology in these papers is the same as the one that I present here: trade makes it such that investing in these technologies is more profitable. Moreover, the driver of the increasing skill premia in less developed countries in these models is the same as the one considered here: technologies that are more skill-biased are adopted by the less developed countries. My paper adds to these two by proposing a mechanism by which these technologies may be adopted by firms in less developed countries and, in doing so, provides a way to understand why skill premia in less developed countries do not follow identical patterns following trade liberalizations with advanced countries. Burstein and Vogel (2017) build a quantitative multi-country model in which trade induces reallocation within economies towards more productive and skill-intensive firms. This model is able to generate increasing skill premia in all countries following a reduction in tariffs. My model differs from the one considered by Burstein and Vogel in that a firm s productivity is determined by the stock of technology capital it uses for production, while it is the result of an exogenous draw in theirs. Both models feature endogenous increases in productivity, as in Burstein and Vogel s model, firms with lower productivity draws are forced to shut down following a trade liberalization due to increased foreign competition. However, in 7

9 my model, final goods producers must choose to invest in this technology in order for the stock of technology capital to increase. This choice lends itself to considering the case of vertically integrated supply chains, while the setup in Burstein and Vogel (2017) does not lend itself as readily to considering how parts producers interact with the final goods producer for whom they are part of the supply chain. This paper is also related to the literature that has explored the impact of globalization on Mexican labor markets. A number of studies (for example, see Esquivel and Rodriguez-Lopez, 2003; Harrison and Hanson, 1999; and Robertson, 2004) explore this question using the Stolper Samuelson theorem as their basis, and find the correlation between changes in output prices and wages at the industry level to be very low. The conclusion from this strand of literature was that skill-biased technological change, and not trade, was responsible for the observed increase in the skill premium. Verhoogen (2008) explores both the overall increase in inequality and the between-plant inequality in Mexico, and hypothesizes that exporting opportunities increase wage dispersion across plants due to quality upgrading. Riano (2009) builds a model in which SBTC is embodied in capital equipment and measures the effect of increasing imports of capital equipment on the skill premium in Mexico. The idea in his paper is similar to what I model here, but importantly, the capital that is traded in my model is technology capital or ideas. The non-rivalrous nature of technology capital creates an environment such that even as the capital begins to be used in Mexico, firms in the United States have an incentive to invest more deeply in it. In fact, it is because the ideas are being used in an additional location that their marginal product increases. Also related to this paper is the literature on the skill premia in developing countries. Ripoll (2005) builds a model in which the skill premium in the developing country responds non-monotonically to trade liberalization and depends heavily on the initial conditions in the economy. Trefler and Zhu (2005) show that those countries with the largest increase in skill premia following a trade liberalization are those which export relatively more skill-intensive goods, and they build a model akin to Feenstra and Hanson (1996), but allowing the South to catch up to the technology of the North instead of receiving FDI flows. They do not propose a mechanism for how this catchup occurs. Burstein, Cravino, and Vogel (2013) and Parro (2013) each propose capital-embodied technology as an avenue by which skill-biased technological change crosses borders. I contribute to this literature by proposing an alternative way that this technology is accumulated and then transmitted from one country to the next, and I provide evidence of my hypothesis. The paper is organized as follows: In Section 2, I provide a description of the data and brief background information on the trade liberalization experience in Mexico in the late 1980s; in Section 3, I provide evidence for the importance of trade linkages for the diffusion of technology and for the skill premium; in Section 4, I present the model ; Section 5 contains comparative 8

10 statics exercises that highlihgt the mechanisms that are built into the model; Section 6 contains my calibration and results; and Section 7 concludes. 2 Background and Data This section briefly describes the data that will be used in the empirical analysis and establishes the relevant patterns of skill premia. It then describes the liberalization policies that were implemented in Mexico in the mid-1980s and establishes that trade began opening between the United States and Mexico before the implementation of NAFTA. 2.1 Data Description Data for Mexico s manufacturing sector comes from INEGI (Instituto Nacional de Estadística y Geografía), Mexico s national statistics bureau. I gather aggregate skill premium data from the EIA (Encuesta Industrial Anual), which is an annual survey of manufacturers which covers about 80 percent of the manufacturing sector. Aggregate data from 1980 through 2004 is publicly available on INEGI s website. I utilize data on production and non-production employees and payments to these two groups and construct the skill premium as the ratio of non-production wages to production wages, as is standard in the literature. Industry-level data is available by request for the years 1984 through 1994, and plant-level data is available also by request for 1984 through Both data sets include information on production and non-production employment, hours, and wages, as well as information on royalties paid, the value of production, and various production costs. The plant-level data includes information on imports, exports, and the percentage of trade conducted with the United States for the years 1986 to The trade-related information was gathered in a special survey conducted by the World Bank. For a more detailed description of the plant-level data, see Tybout and Westbrook (1995). The plant-level data also includes information on tariffs and license coverage for the period from 1984 to 1990, as well as foreign ownership. In addition to analyzing the plant-level data on its own, I will exploit these data to get industry-level measures of trade costs, which are not available for the years from other sources, to my knolwedge. I will collapse the plant-level observations on tariffs paid into weighted average industry-level tariffs and then combine them with the tariff data discussed below in order to construct a complete time series of trade frictions. Aggregate trade data for Mexico is obtained from the World Bank World Development Indicators Database (WDI) and information on bilateral trade between the U.S. and Mexico is gathered from the database compiled in Schott (2008). This database provides bilateral imports and exports, 9

11 as well as customs duties paid, between the U.S. and its trade partners (including Mexico) for the years 1972 through 2005 for SIC industries. I construct a concordance in order to match this data ( ) with the industry-level data for Mexican manufacturing. Bilateral tariffs between Mexico and the United States are not available at the industry-level (or good-level) for the entire period of interest, with most tariff data beginning in Therefore, I estimate trade costs as in Bernard, Jensen, and Schott (2006) as a proxy for industry-level tariffs between the U.S. and Mexico. This data is available from 1989 to 2005 and I use the observations from 1989 to For 1989 and 1990, I have two sources of data for trade costs: the plant-level data from INEGI and the Schott (2008) database. For those two years, the trade cost data does not match up exactly, so I average across the two data sets. Data for the U.S. manufacturing sector is obtained from the NBER-CES Manufacturing Productivity Database (Bartelsman and Gray, 1996). This data is available from 1959 through Again, the database provides information on production and non-production employees, as well as payments to each group. I then construct the skill premium as the ratio of non-production wages to production wages and measure skill-intensity using data from this database for robustness checks. 2.2 Skill Premia in Mexico and the United States For the remainder of the paper, I will use the term skill premium to mean the ratio of the wages of non-production workers to the wages of production workers, as is typical in the literature that examines the skill premium in developing countries. The U.S. experienced similar timing in the rise of the same variable. 1 In Mexico, the skill premium was stable with non-production wages being about twice as high as production wages during the late 1970s and early 1980s, but began to rise around It grew for the next decade and peaked in 1996, with non-production wages that were about 3.1 times higher than production wages. This can be seen in Figure 1a. As can be seen in Figure 1a, the skill premium in the manufacturing sector in the United States also began to rise in the mid-1980s. Figure 1a also shows that the skill premium in Mexico was substantially higher than that in the United States and rose by much more over the period of interest. Figure 1b shows that the timing of the increases in the two skill premia largely coincided. It also highlights that the increase in Mexico was substantially bigger than that in the United States. In particular, over the course of the decade from 1986 to 1996, the skill premium in Mexico rose by about 60%, while the skill 1 Amiti and Cameron (2011) use data that includes the educational attainment and production/non-production status of workers in Indonesian manufacturing over several years; they show that the production/non-production breakdown is a good proxy for skill or educational attainment. Bustos (2011) shows that in Argentina, non-production employment is 3.2 times more skill-intensive than production employment. 10

12 (a) Skill Premium Levels (b) Relative to 1980 Figure 1: Skill Premia in U.S. and Mexico premium in the United States rose by about 10 to 15%. 2 Therefore, the skill premium in Mexico rose by about four times as much as its American counterpart from 1986 to Mexico s Trade Liberalization During the 1950s, Mexico began to pursue a set of policies based on the theory of import substitution. As such, during this time, Mexico became one of the most closed economies in the world, with more than 90 percent of its domestic production subject to import licenses by Import licenses are commonly viewed as the main source of restricted trade flows (Kehoe, 1995, TenKate 1992), though, in practice, Mexico utilized three instruments to restrict these flows: (i) ad-velorum tariffs, (ii) official minimum prices for custom valuation, and (iii) quantitative restrictions such as quotas and the aforementioned import licenses. As a result of the balance of payments crisis in 1982, the Mexican government decided to pursue a large-scale liberalization of the Mexican economy, including a massive trade liberalization (apertura), in order to restart economic growth. In 1985, the Mexican government undertook a number of structural reforms, including reducing the import license coverage from 92 percent to 47 percent between June and December of that year. Many of these reforms were requirements of the debt restructuring agreement that Mexico entered with its international creditors in the wake of the debt crisis in the early 1980s. The government continued to phase out import licenses over the course of the decade, with coverage falling to 19 percent by Most of the remaining import licenses covered agricultural and petroleum refining 2 Note that the college premium, measured as the ratio of college to non-college wages, began to rise earlier in the 1980s. The college premium is the measure which is most frequently the focus of papers dealing only with the United States, but I will concentrate on comparable measures of the skill premium in this paper. 11

13 products. Over the same period, ad-velorum tariffs fell as well. In 1985, the maximum tariff was 100 percent; only a year later, in 1986, it was reduced to 50 percent. By 1987, the maximum tariff was 20 percent and the production-weighted average tariff was 11 percent (Esquivel and Tornell, 1995). Furthermore, in 1986, Mexico joined the General Agreement on Tariffs and Trade (GATT), which ensured that tariffs faced by its exporters also fell. Figures 2a and 2b show the progression of production-weighted tariffs and license coverage for the average manufacturing plant in Mexico. 3 (a) Average Tariff Rates (b) Average License Coverage Figure 2: Measures of Trade Distortions Figure 3a documents the ensuing increase in Mexican imports and exports of manufactured goods as a percentage of total value added in manufacturing. Figure 3b shows the percent of merchandise trade that was taking place with high-income countries. The solid lines represent the total amount of merchandise imports (blue) and exports (red) as a percentage of value added and the dotted lines show the merchandise trade occurring with high-income OECD countries. The United States accounted for roughly 70% of all of Mexico s manufacturing trade throughout this period (Patrick, 1994). As can be seen in the figures, the increase in trade came predominantly from increased trade with developed countries, namely the United States. During this period, Mexico also entered into trade negotiations with the United States, which culminated in a four-part understanding known as the Framework of Principles and Procedures for Consultation Regarding Trade and Investment Relations or more commonly, the Bilateral Accord. This Accord was the first-ever formal bilateral agreement governing commercial relations between the two countries, and it included a statement of principles, a mechanism for consultations, an agreement on data exchange, and an Immediate Action Agenda. The Immediate Action Agenda was the start of negotiations on a number of matters, including technology transfer. In particular, Mexico was interested in obtaining help from developed nations to develop its intellectual property rights protection laws so that technological transfer from companies in the United States would be 3 Note that these tariffs rates have been inferred from actual duties paid and are not officially recorded tariff rates. 12

14 (a) Manufacturing Trade (b) Merchandise Trade Figure 3: Mexican Manufacturing Trade more forthcoming. Mexico argued that access to new technologies was of utmost importance and was a necessary component to any improved trade arrangement between the two countries (DuMars, 1991). The recognition of intellectual property rights was an important step to allowing for the transfer of technology between the two countries. The government also began to loosen its restrictions on foreign ownership over this period; however, the process was slower to change than other policies, and significant restrictions remained in place for the next decade. In particular, foreign companies were not allowed to acquire existing Mexican firms without submitting to a lengthy approval process. Establishing a new foreign-owned business was somewhat easier, but only if the business fit certain criteria, which included a requirement that the business have at least a non-negative net export balance over the first three years of its existence. Maquiladora firms were exceptions to these rules, but the process for obtaining a license establishing a firm as a maquiladora was viewed as relatively cumbersome until the process was reformed in December of However, to the extent that foreign ownership was allowed, this also served as a channel for technological transfer across the two countries. In 1992, the Mexican government signed an agreement to enter into the North American Free Trade Agreement (NAFTA) with the United States and Canada on January 1, As part of NAFTA, all remaining tariffs on goods traded between the two countries would be phased out over the next decade. Moreover, the three countries agreed to abide by the intellectual property rights laws of the United States. Contrary to commonly held beliefs, the trade liberalization that took place during the 1980s was more substantial and impactful than NAFTA. By the time that NAFTA was signed, approximately 95% all of imports into Mexico from the U.S. were subject to import tariffs of 20% or less and 80% of imports into the U.S. from Mexico were subject to tariffs of 5% 13

15 or less. This stands in stark contrast to the 100% tariffs that were common in Referring back to Figure 2a, one can see that the vast majority of the decline in tariff rates pre-dates the implementation of NAFTA. 3 Evidence on Supply Chains and Technology Transfer In this section, I present evidence of the mechanism that I will consider, starting with a case study in which technology is transferred through supply chains and continuing with empirical evidence at the plant- and industry-levels. 3.1 Trade and Technology Transfer: A Case Study of an Auto Plant As an example of the transfer of technology capital, I now turn to a case study of one company s vertical integration of its supply chain, including a plant in Northern Mexico, conducted by Shaiken and Herzenberg (1987). This particular plant was partially owned by an undisclosed American auto manufacturer to make engines and engine parts for cars that would be exported, in keeping with Mexican regulations that required that a large proportion of output from foreign-owned plants be produced for export. The case study consists of interviews with both production and nonproduction workers, and it directly compares the Mexican plant with its U.S. counterpart in terms of production techniques, technology used, management practices, and the quality and quantity of output. The Mexican plant was a brand new facility, which utilized some of the most advanced production technologies available at the time of its construction, and it was built to compete with the most successful engine plants in the world. The study finds that the Mexican plant was able to achieve efficiency comparable to that of the U.S. plant, due in large part to its utilization of technology capital that was the same across the two locations. The company invested significant resources in training the Mexican workers to operate state of the art machinery and to manage operations according to the company s best practices. The case study details how supply chains served as a way to transfer blueprints for the production of parts, technology embedded in intermediate parts and in machines, best practices for efficient production, and organizational capital, such as worker training programs. These types of technology capital impacted production and non-production workers differentially. One place that this is evident is in the worker training programs implemented by the company. Non-production workers, such as technicians, had to acquire many skills that were not part of their previous skill sets. In order to do so, there were task certification programs, in-plant and on-the-job training programs and apprenticeships, and even trips to tool suppliers in the United States, all to ensure that 14

16 the technicians understood how to maintain the plant. Production workers also received training, but to a lesser extent. (a) Standard (b) Modified Figure 4: Production Line Layouts Another example of the transfer of technology capital is the modification of the production line layout. Figure (4a) shows the standard layout for a production line, where each circle with a cross in it represents a worker and each rounded solid box is a machine that needs to be operated. Figure (4b) displays the new layout used by the company. As can be seen by examining the areas enclosed with the dotted blue lines, in the standard layout each worker is in charge of two machines, whereas in the new layout, each worker is in charge of three machines. This changes the mix of workers necessary in the plant; because each production worker is now operating more machines, all else equal, fewer production workers are required. However, because this may lead to more breakdowns on the line, there are more technicians (non-production workers) required. The technology in this production line modification augments the skilled non-production workers and substitutes for the unskilled production workers. In general, this case study serves to highlight the role of integrated supply chains in transferring skill-augmenting technology capital across borders. 3.2 Plant-level Evidence In order to explore how trade integration impacts the transfer of technology and the skill premium, I use a balanced panel of 2,598 Mexican manufacturing plants, whose cumulative output accounts for about 70% of the country s manufacturing output. Data in the panel covers output from 1984 to I have eliminated plants in which the skill premium fluctuates by more than 1000% from one year to the next and those whose production values of exports exceeds their total value of production. This panel includes information on employment, wages, production value, tariffs, license coverage, and ownership for all years, as well as information on trade, including the proportion of trade done with the United States, for the years 1986 through Information on imports is limited to imports of machinery and of raw materials. Here, I investigate the relationship between 15

17 integration with the United States, technological transmission, and the skill premium. I begin by estimating the following regression by OLS: ( ) R = α + β 1 τ jt + β 2 US j + β 3 (τ US) jt + SI j0 + γ j + T t + ε jt, jt where j indexes plants and t indexes the year. Furthermore, ( R ) jt is royalties paid by the plant as a percentage of output, τ jt is the average export tariff levied on the plant s output (measured in percentage points), US j is a dummy variable indicating whether the plant is a wholly U.S.- owned entity, SI j0 is the skill intensity of the plant in the initial year of observation, and γ j and T t are vectors of industry 4 and year dummies, respectively. The left-hand side variable is the total royalty payment made by a given plant as a fraction of the value of the plant s output, but there is no information on the entity to which these royalties are paid. Specifically, the data does not distinguish between royalties paid to domestic firms versus foreign firms. However, the data also has information on royalties received, and so I can compare the sector-wide average of these numbers to determine that royalties paid are 10 times as large as royalties received, indicating that either royalties are being paid to firms that are outside of the manufacturing industry or outside of the country. Over the entire sample, the average value for royalties paid as a fraction of output is 0.38%, with a standard deviation of 1.62%, whereas average royalties received as a fraction of output is 0.033%, with a standard deviation of I use royalty payments as an indicator of the renting of technology, 5 and I relate first relate these payments to tariff rates. In Column (1) of Table 1, I investigate the simple relationship between tariff rates and royalties. The coefficient on the tariff rate can be interpreted as meaning that a one percentage point reduction in tariffs increases royalty payments by 1 percent (0.003 percentage points), when controlling for the plant s initial skill intensity, as well as industry and year fixed effects. Over the entire sample, tariffs fell by an average of 18 percentage points, which would translate into an increase in royalty payments of 18%. In Columns (2) and (3), I further explore the impact that trade liberalization with the United States has upon royalty payments. The data includes information on the percentage of the plant that is owned by various countries, including the United States. I use this variable to create a dummy variable that is equal to one if the plant is wholly U.S.-owned and zero otherwise. 6 The percent ownership of any given plant is constant across the periods of the sample so I am not able to test the extensive margin of U.S. ownership, but rather the differential impact of a 4 The industries are classified using the 4-digit Mexican classification of manufacturing industries (CMAP). 5 The variable name in the original data is gastos por transferencia de tecnología, which translates as expenditures for the transfer of technology. 6 The reported results are robust to less strict definitions of what constitutes U.S. ownership, ranging from 51% U.S.-owned up to 100% U.S.-owned. 16

18 change in tariffs on the intensive margin for U.S.-owned plants. When I consider the impact of U.S. ownership jointly with tariff reductions, I find that U.S.-owned plants pay, on average, royalty payments that are 0.47 percentage points higher than plants that are not owned by a U.S. entity, while the impact of tariffs remains unchanged. When I include an interaction between the dummy for U.S. ownership with tariff rates, we see that U.S.-owned plants that experience a 1 percentage point decrease in tariffs pay royalties that are, on average, 0.01 percentage points higher, in addition to the percentage point increase experienced when tariffs fall by the same amount. Again, the coefficient ( on ) tariff rates remains statistically significant and changes only slightly. In Column (4), I use Exp, the fraction of output that is exported to the United States as my measure of jt integration with the U.S. In order to explore the interaction between increased exporting and tariffs, I now estimate the following equation: ( ) ( ) ( R USExp = α β 1 τ jt + β 2 + β 3 τ USExp ) + SI j0 + γ j + T t + ε jt, jt jt jt reversing the sign of tariffs because increased tariffs and increased exports are expected to have opposing effects on royalty payments. As we can see from Column (4), firms that experienced decreases in tariffs but did not respond by increasing exports see no impact on their royalty payments, as the coefficient on tariff rates alone is not statistically significant. However, firms that increased exports and experienced tariff reductions pay royalty payments that are twice as high as those of their counterparts who do not increase exports to the United States. Combined, these results imply that the firms that were more integrated with the United States paid higher royalties on average, and that this effect became more pronounced as tariffs fell. Moreover, even plants that were not U.S.-owned experienced changes in royalty payments as tariffs fell, indicating that decreasing tariffs induced plants to increase adoption of technologies. Next, I will relate changes in royalty payments, as well as measures of trade liberalization, to changes in the skill premium. 17

19 U.S. Co U.S. Co U.S. Exports (1) (2) (3) (4) Constant (0.133) (0.133) (0.133) (0.133) τ jt (0.001) (0.001) (0.001) (0.004) US j (0.056) (0.097) (US τ) jt (0.004) USExp jt ( ) τ USExp jt (0.342) (0.011) SI j (0.083) (0.083) (0.083) (0.104) Industry Fixed Effects? es es es es Time Fixed Effects? es es es es R Table 1: Royalties and Tariffs Standard errors in parentheses. * denotes significance at the 10% level, ** significance at the 5% level, and *** significance at the 1% level. Next, I relate these observations to changes in the skill premium. In order to do this, I first estimate the following equation: ( ) ( ) ( ) Exp Imp R SP jt = α + β 1 + β 2 + β 3 jt jt jt ( ) Exp + β 4 US j + β 5 US jt ( ) ( ) Imp R + β 6 US + β 7 jt US + SI j0 + γ j + T t + ε jt, jt where SP jt is the plant-level skill premium, Exp is the value of the plant s exports divided by its jt output value, Imp is the value of the plant s imports of machinery divided by its output value, and jt the other variables are as defined above. I use exporting as a fraction of output instead of tariff rates in order to control for trade liberalization because, as we saw above in Table 1, a change in exporting as a result of tariff reductions has more predictive power than a change in tariff rates alone. I first restrict my attention to trade with the United States, then expand to trade with only non-u.s. entities, and finally, I expand the analysis to trade with all countries. Initial skill-intensity 18

20 at the plant level, as well as industry and year fixed effects, are included in all specifications in order to control for the skill content of production, as well as changes in skill content over the sample period. Sample U.S. Only No U.S. All U.S. Only No U.S. All (1) (2) (3) (4) (5) (6) Constant (0.115) (0.114) (0.115) (0.116) (0.114) (0.116) Exp jt (0.122) (0.122) (0.096) (0.130) (0.183) (0.100) Imp jt (0.078) (0.008) (0.065) (0.079) (0.123) (0.065) Royalties jt ( Royalties (0.008) (0.008) (0.008) (0.008) (0.008) (0.008) US jt ) US jt ( Exp US ) ( Imp US ) jt jt (0.057) (0.063) (0.057) (0.067) (0.065) (0.069) (0.032) (0.032) (0.032) (0.032) (0.032) (0.032) (0.360) (0.751) (0.323) (0.765) (1.027) (0.503) SI j (0.078) (0.032) (0.078) (0.078) (0.078) (0.078) Industry Fixed Effects? es es es es es es Time Fixed Effects? es es es es es es R Table 2: Skill Premium and Integration with the U.S. Standard errors in parentheses. * denotes significance at the 10% level, ** significance at the 5% level, and *** significance at the 1% level. I will focus on my preferred specification, which is reported in Columns (4) through (6) of Table 2. Notice first that the coefficient on exporting as a fraction of output is positive and statistically significant in all three columns and that the effect is larger when I focus only on exports to the United States. The coefficients can be interpreted as a 100% increase in Exp to the United States is associated with skill premia that are higher than that of the average plant, controlling for 19

21 all other variables. When we compare these coefficients to the ones in Columns (5) and (6), we see that when I restrict attention to trade with only non-u.s. countries (Column 5) or expand to all countries (Column 6), the magnitude of the coefficient falls; an increase in Exp of 100% to non-u.s. destinations increases the skill premium by an average of points. If we examine exports to all entities, this number is similar. This observation is reinforced by the coefficient on the interaction between exporting and the dummy for U.S. ownership. U.S.-owned plants that export 100% more to the United States have, on average, skill premia that are points higher (30% higher than the average plant), whereas a 100% increase in overall exporting for U.S.-owned plants is associated with an increase in the skill premium of points. Contrast this with U.S.-owned plants that export to entities outside of the U.S.; in these cases, an increase in exports of 100% is associated with a reduction of the skill premium, though this estimate is no longer statistically significant. These plants are clearly more integrated with the United States and are likely a part of a vertically integrated supply chain. Notice that I am also controlling for imports of machinery as a fraction of output, as well as its interaction with the U.S.- owned dummy, in order to account for the possibility of imported capital-embedded skill-biased technical change. Both of these variables are found not to be statistically significant, indicating that imported capital-embedded technical change is not the channel through which the skill premium is rising during this period. 7 I will therefore abstract from imported capital in the model, as it does not appear to have a statistical relationship with the skill premium. The coefficient on the dummy for U.S.-owned plants and the coefficient on royalty payments as a fraction of output are positive and statistically significant, indicating that U.S. ownership and payments for royalties have positive impacts on plant-level skill premia. This is true even though I control for initial skill intensity of the plant, meaning that these results are not merely capturing that U.S.-owned plants or plants that pay more royalties are more skill-intensive and therefore pay higher skill premia. I interpret royalties as being a proxy for arms-length relationships between non-u.s. owned plants and their U.S. counterparts. As we know from Table 1, U.S. ownership has a statistically significant and positive effect on royalty payments, but does not explain the entire variation in that variable. 8 Moreover, even plants that are not U.S.-owned respond to changes in tariff rates by increasing their rental of technology, as seen by increased royalty payments. I include the interaction between royalties paid and U.S. ownership, which is found not to be statistically significant, with p-values just above 0.1. Focusing for the moment on the magnitude of the coefficient on the interaction term, we can see that it almost exactly offsets the effect of royalties on the skill premium, indicating that if a firm is U.S.-owned, the impact of royalty payments is already captured by the 7 Coefficient estimates on other variables are similar if I omit Imp and its interaction with the U.S.-owned dummy. 8 The R 2 on the regression estimated in Table 1 is so a substantial amount of variation in royalties paid is not accounted for by U.S.-ownership. 20

22 coefficient on the U.S.-owned dummy. 9 I use this finding to validate my interpretation of royalties paid as being a proxy for transference of technology through ownership channels. In combination, these results indicate that the plants that are more closely related to the United States, though trade and ownership channels, have on average higher skill premia. Lastly, I restrict my attention to those plants that trade exclusively with the United States and are wholly U.S.-owned, as an attempt to analyze only those firms that are most clearly linked to the supply chain of a U.S. producer. There are only 28 such firms with observations in all years from 1986 to Although these plants trade exclusively with the United States, they do not necessarily export all of their output to the U.S. On average, these plants export about 40 percent of the values of their production. In this context, it is not clear what effect, if any, royalties should have on the skill premium, as non-rivalrous technology can be shared directly, without royalty payments. I therefore focus on a simple equation relating exports to the skill premium for this set of plants by estimating the following equation: ( ) ( ) USExp R SP jt = α + β 1 + β 2 + β 3 SI j0 + γ j + T t + ε jt. jt jt The results of this estimation are included below. Levels % Change (1) (2) Constant ( ) USExp jt ( R ) (0.328) (0.058) (0.422) (0.006) jt (0.061) (0.049) SI j (0.915) (0.130) Industry Fixed Effects? es No Time Fixed Effects? es No R Table 3: Trade and the Skill Premium Standard errors in parentheses. * denotes significance at the 10% level, ** significance at the 5% level, and *** significance at the 1% level. 9 In the Appendix, I include the specification in which I exclude the U.S.-owned dummy and include the interaction between trade variables and royalties paid. Results are similar. 21

23 As we can see, even when we concentrate on this subset of plants, exporting twice as much is associated with skill premia that are almost twice as high as the plants that export less. This is true even when controlling for the initial skill intensity of the plant, as well as for industry and year fixed effects. In Column (2) of Table 3, I instead examine the impact of the percent change in exports and royalties over production on the percent change in the the skill premium from 1986 to 1990, again controlling for initial skill intensity of the plant. Again, we see that a larger increase in exporting is associated with greater increases in the skill premium. In both cases, royalties over output is not statistically significant, but this is attributable to the fact that the plants are wholly U.S.-owned and therefore may not need to pay for the transfer of technology. 3.3 Industry-level Evidence Plant-level data is only available from 1984 to 1990, which only captures the first 3 years of increases in the skill premium and trade. However, industry-level data that is comparable to the plant-level data analyzed in the previous section is available from 1984 to I create a crosswalk between the CMAP industry classification to SIC 1987 industry codes in order to combine the Mexican industry-level with U.S. trade data, as well as trade cost data from Schott (2008). I further exploit an Input-Output matrix from 1989 provided by INEGI in order to construct a measure of intermediate trade with the United States as a fraction of output in each industry. Due to data limitations, I will assume that the I/O matrix is constant across the sample. The industry-level data does not provide information on ownership, so I will be unable to control for the percent of a given industry that is U.S.-owned. Instead, I will concentrate on royalty payments and trade as my measures of industry integration with the United States. I begin by examining the impact of the change in tariffs on royalty payments and the skill premium by estimating the following equations: ( ) R it = α + β 1 τ it + γ i + T t + ε it SP it = α + β 1 τ it + γ i + T t + ε it, where, as before ( R is royalties over the value of production paid by firms in industry i in year )it t, SP it is the industry-level skill premium in year t, τ it is the average tariff faced by industry i in year t, and γ i and T t are vectors of industry and year dummies. Table 4 reports the results of 10 Because industry classifications and the Survey of Manufacturers changed substantially in 1995, it is not advisable to combine industry-level data from before 1995 with data thereafter. 22

24 this estimation. In Column 1 of Table 4, all else equal, industries that experienced a 1 percentage point decrease in tariffs paid, on average, percentage points more royalties than those that experienced no change in tariffs. Here, I have controlled for industry and year fixed effects, so I am controlling for skill- or technology-intensity of production for the given industry, as well as time trends in royalties. In Column 2, we see that tariffs are also negatively correlated with the skill premium. Industries with 1-percentage-point higher tariffs saw, on average, skill premia that were points lower. Therefore, industries with bigger decreases in tariff rates saw increased royalty payments, as well as increased skill premia. 11 The average industry experienced a 22 percentage point drop in the tariff rate over the sample, which translates into an increase in royalty payments over production value of 0.22 percentage points. The average royalty payment as a fraction of the value of production is 0.66 percent. This change in tariff rates also translated into an increase in the skill premium of 0.06 points for the average industry, after controlling for skill and technologyintensity via an industry fixed effect. Variable R it SP it (1) (2) Constant (0349) (2.333) τ it (0.004) (0.001) Industry Fixed Effects? es es Time Fixed Effects? es es R Table 4: Industry Level Impact of Tariff Changes Standard errors in parentheses. * denotes significant at the 10% level, ** significant at the 5% level, *** significant at the 1% level. In order to further explore my hypothesis, I explore how trade, in conjunction with royalties, interacts with the skill premium. Falling export tariff rates are associated with increased export flows in the data, so I will explore the impact that exports have on the skill premium as I did in the plant-level data by estimating the following equation, ( ) ( ) ( ) Royalties Exports Imports SP it = β 1 + β 2 + β 3 + β 4 SI i0 + γ i + T t + ε it, it it it 11 For comparison, in the appendix, I include the regression in which my independent variable is change in tariffs from 1984 to 1994 and the dependent variables are also expressed in changes. 23

25 ( ) ( ) where Exports and Imports it are now exports and imports of intermediate goods as a fraction it of output in industry i and time t, and the rest of the variables are as defined above. Recall that at the plant level, imports of machinery were measured but not imports of other goods. Column (1) of Table 5 displays the results for the level of the skill premium, controlling for industry and year fixed effects, while Column (2) reports results for the impact of changes in all the independent variables on the change in the skill premium from 1984 to In both cases, increasing exports of intermediate goods and royalty payments as a fraction of output at the industry level is associated with higher skill premium on average, after controlling for the initial skill intensity of the industry in These industry effects are consistent with the plant-level evidence that firms that become more integrated with the United States via trade pay higher royalties and have higher skill premia on average. Level Change (1) (2) Royalties it (0.010) (0.018) Exports it (0.077) (0.054) Imports i,t (0.040) (0.050) SI i (0.392) (0.065) Industry Fixed Effects? es No Time Fixed Effects? es No R Table 5: Impact of Integration into U.S. Supply Chain * significant at the 10% level, ** significant at the 5% level, *** significant at the 1% level I interpret these results as supporting the idea that supply chains are an important determinant of skill premia. In light of this evidence, I build a model in which exporting plays a role in determining the skill premium. Those plants that export intermediate goods need to use the foreign technology capital in order to produce intermediate goods for the final goods producer in the other country, and they pay royalties for the rental of the technology capital. In the model, there is a single market for each type of labor and no adjustment costs, so all plants and industries experience the same increase in the skill premium. There is within-industry reallocation of labor from producing 24

26 intermediates for domestic final goods producers to producing intermediates for the foreign final goods producer. The extent to which this reallocation occurs differs across industries and these differences is driven by differential changes in tariffs and protection for intellectual property. 4 The Model Environment The model economy features two countries, Mexico (denoted by M) and the United States (denoted by U), and I industries, indexed by i. I include only the U.S. and Mexico because the primary focus of this paper is the change in Mexico s skill premium and during the period of interest, Mexico traded almost exclusively with the United States. In each country and industry, there is a nontradeable final good produced by perfectly competitive firms. This final good is a CES aggregate of a continuum of varieties, some of which are produced domestically and some of which are imported and subject to an iceberg transportation cost. The final good can be used either for consumption or for investment in the stock of technology capital, which is used in the production of the intermediate varieties. The production of these varieties requires skilled and unskilled labor, as well as the aforementioned technology capital. The producer of an individual variety is a monopolistic competitor and may produce that variety for a domestic final goods producer and for a foreign final goods producer at the same time, meaning that they may operate two different technologies within a single firm. I will refer to the operation of different technologies within a single variety as production lines within that variety or firm. Aggregate labor of each type is inelastically supplied and is immobile across countries, though it is perfectly mobile across industries and production lines. I will allow certain parameters to vary across industries but the producers of varieties within an industry will be identical. I now provide details of the model environment. Final Goods Producers In each industry and each country, there are perfectly competitive producers of a non-tradeable final good which seek to maximize the infinite sum of discounted dividends. The final good in country k and industry i is denoted ki and is composed of a CES aggregation of intermediate varieties y d k ji (ω), where the variety is denoted by ω and the country of origin for that particular variety is indexed by j. The final goods producer owns and invests in a stock of technology capital, Z ki, which it rents to the producers of the intermediate varieties of which its final good is composed. It earns returns r kki and r k ji from renting its technology capital to domestic and foreign producers of intermediate varieties, respectively. To economize on notation, I will suppress the time subscript. These final goods producers in country k {U,M} and industry i solve the following problem: 25

27 max t=0 P ki D ki (1) s.t. ( ) D ki = ki I ki + Z ki rkki + r k ji p k ji (ω)y ˆΩ d k ji (ω)dω (2) i I ki = Z ki (1 δ)z ki [ˆ ] 1/ρ (3) ki = y d k ji (ω)ρ dω Ω i (4) 1 where 1 ρ is the elasticity of substitution between varieties within each sector. A variety ω is producer country-specific, and the final producer will purchase varieties from both countries, substituting towards more domestic varieties if the cost of foreign intermediates is higher. Technology Capital: Investment and Rental The technology capital that I consider, Z ki, can be thought of as a stock of ideas, blueprints, or production techniques and, as such, is non-rivalrous in nature. These ideas and production techniques are the ones that are necessary in order to make an intermediate part for the final product so that when the final goods producer begins to purchase more intermediates from foreign producers, the foreign producer will have to rent more of this technology capital in order to produce these parts. This causes the rate of return on technology capital to increase for the final goods producer, thus inducing them to increase investment in it. Moreover, the rental of technology capital in order to produce the appropriate part for the foreign final goods producer is the channel by which this technology capital will be transferred across borders. I assume that this technology is high skill-augmenting and low-skill saving. This assumption is based on the large literature that shows that technological advancements since the early 1980s have favored high skill workers. I extend this observation to technology capital, assuming that production techniques have changed the mix of high- and low-skill workers. The extent to which this is true in the model will depend on parameter values, which will be discussed in more detail in the following sections. 26

28 Production of Varieties The intermediate goods producer that produces intermediate variety ω for industry i in country k can produce both for the domestic market (k) and for the foreign market ( j). He chooses the output for the domestic market (y kki (ω)), the output for the foreign market (y jki (ω)), skilled labor to produce for the domestic market (h kki (ω)), skilled labor to produce for the foreign market (h jki (ω)), unskilled labor to produce for the domestic market (l kki (ω)), unskilled labor to produce for the foreign market (l jki (ω)), and an amount of domestic and foreign technology (Z ki, Z ji ) to maximize profits, taking the inverse demand function, wages (w H k,wl k ), and the rental rates for technology (r jki,r kki ) as given. The first subscript refers to the country for which the intermediate variety is produced and the second refers to the country in which production takes place. The producer of the intermediate good must use the technology of the supply chain that they are integrating with in order to produce the variety for that firm. The intermediate-goods producing firm therefore solves the following problem: s.t. max p jki (ω) y jki (ω) w H k τ h jki(ω) w L k l jki(ω) Z ji r jki τ z jki (5) j {U,M} jki y jki (ω) = A k [ θ i ( Z α jih 1 α jki ) σ 1 ] σ σ (ω) + (1 θ i )l jki (ω) σ 1 σ 1 σ j {U,M} (6) y jki (ω) = ( p jki (ω) P ji ) 1 ρ 1 ji j {U,M}, (7) where σ > 0 is the elasticity of substitution between high- and low-skilled labor. The skill-intensity of production will be jointly governed by α (0,1) and θ i (0,1), as will be discussed in more detail below. A k > 0 is the level of the country-specific total factor productivity (TFP) and P ji is the cost of the final good produced in industry i in country j, the country for whose supply chain the intermediate variety is being produced. There are two wedges on the production, τ jki and τ z jk, which are both equal to one if the variety is produced for the domestic final goods producer ( j = k) and greater than one otherwise ( j k). The first of these is an iceberg trade cost on goods, and the second can be thought of as an implicit tax on or distortion to the use of foreign technology. This implicit tax is a stand-in for imperfect intellectual property protection. As detailed in Section 2, part of the liberalization between the United States and Mexico was the adoption of stricter protection of intellectual property by Mexican firms; τ z jki is included in order to capture this feature. 27

29 5 Skill Premium The skill premium in country k in this model can be expressed as w H k w L k ( θ i Z ji = (1 α) 1 θ i h jki ) α σ 1 σ ( h jki l jki ) 1 σ, (8) where I denote by h jki and l jki the equilibrium choices for high- and low-skilled labor that are made by all firms in industry i producing in country k for the supply chain of country j. Recall that producers of all varieties ω in a particular industry are identical in their productivities and the prices they face, so their equilibrium choices will be the same; therefore, I suppress the notation indicating the variety. Because labor is perfectly mobile across all industries, wages will equalize across industries and this equation will hold for every industry i. In the expression above, we can ( ) Z α σ 1 ( ) 1 think of the term ji σ h h jki as representing the relative demand for high-skill labor and jki σ l jki as representing the relative supply of labor in industry i. The demand will be affected by the amount of technology being utilized in a given country and industry and the supply will be subject to the normal Stolper Samuelson forces. In what follows, for simplicity, I will consider only two industries i {1,2} with Industry 1 being relatively high-skill intensive and Industry 2 being relatively low-skill intensive (θ 1 > θ 2 ). The results and logic will extend to the multi-industry case. Standard H O Mechanism Suppose for the moment that α = 0 and that I = 2, so there are only two industries, with the first being high-skill intensive and the second being low-skill intensive, so that θ 1 > θ 2. In this case, technology capital is not used at all in production and the expression for the skill premium simplifies to w H k w L k = θ i 1 θ i ( h jki l jki ) 1 σ, (9) which is equivalent to the skill premium in the standard H O model. The logic from the 2x2x2 H O model follows. Low-skill labor is relatively abundant in Mexico while high-skill labor is 28

30 relatively abundant in the United States. Therefore, absent any role for technology (which is true when α = 0), when countries are in autarky, the relative wages are completely determined by the aggregate supply of factors in the country and the skill premium will be higher in Mexico. Once the countries open to trade, because it is relatively less expensive to make the low-skill intensive good in Mexico (the low-skill abundant country), Mexico will move towards specializing in lowskill intensive goods. Likewise, the United States will move towards specializing in the production of goods for the more skill-intensive industries since it is less expensive to produce that good in the United States than in Mexico. If both countries were to completely specialize in the production of their respective goods, meaning that the entire labor force would be dedicated to production in that industry, the skill premia in the United States and Mexico would become wu H wu L w H M w L M = = θ 1 1 θ 1 θ 2 1 θ 2 ( HU L U ( HM L M ) 1 σ (10) ) 1 σ. (11) Because there would have been some portion of the labor force producing for each industry in autarky in order to satisfy demand for both goods, the movement to specialization drives the skill premium down in the United States and up in Mexico. This is true even in the case of incomplete specialization, as the result is driven by the fact that θ 1 > θ 2 and that after trade liberalization, reallocation of labor in the United States moved towards the high-skill intensive industry while in Mexico it moved towards the low-skill intensive industry. Role of Technology Capital Now, if α > 0, technology capital will be a necessary part of the production process, the skill premium will be expressed as in Equation (8), and the relative demand for labor will therefore be affected by technology capital. In order to understand how the skill premium is impacted by a trade liberalization, first note that if σ > 1 and Z ji increases following liberalization, then the skill premium will increase, all else equal. Therefore, an increase in technology capital may offset or completely overturn the pressure on the skill premium that arises due to labor reallocation across industries. Moreover, the increased availability of more productive technology due to a decrease in trade barriers will decrease the incentive for labor to reallocate across industries. There are two forces that may cause technology capital in the world to increase. The first is what I refer to as the adoption channel, whereby Mexican producers upgrade to using the more skill- 29

31 intensive technology of the U.S. final goods producers in order to produce for the U.S. supply chain. This occurs as the tariff on the intermediate good produced for the U.S. (τ jki ) falls or the distortion to the use of the U.S. technology capital (τ z jki ) falls. A reduction in either distortion results in the Mexican intermediate goods producer shifting towards producing more for the U.S. final goods producer than they did pre-liberalization. The second force is what I refer to as the investment channel, which is the increase in investment in technology capital by the final goods producer that will occur as a result of an increase in the rental rate for that technology. As barriers to trade fall, the technology of the United States will begin to be used more intensively in Mexico, as discussed above. This increases the return to investing in that technology for the American final goods producers. All else equal, these producers will increase investment in their technology capital. This is a secondary driver of increased technology in Mexico and the primary driver of increased technology in the United States. I will explore each of these pieces through some simple comparative statics. I will provide the intuition the following sections, while details can be found in the Appendix. Adoption Channel In order to understand how the adoption channel operates, suppose for the moment that there is only one industry, i, so high- and low-skilled laborers are fully employed therein and there is no reallocative pressure from the H O mechanism discussed above. Changes both to the tariff, τ jki, and to the distortion to the use of foreign technology capital, τ z jki, will contribute to the adoption channel. Suppose first that there is a reduction in the tariff on the good, and consider the problem of the final goods producer in the United States (Equations (1) through (4)). Because the intermediate goods producers are monopolistic competitors, they will charge a constant mark-up over marginal cost, c(w H k,wh k,r kki,r jki ) : p jki = ρτ jki c(w H k,wh k,r kki,r jki ), which is proportional to the tariff charge on the intermediate good. When the tariff falls, all else equal, the effective price paid by the final goods producer for the Mexican intermediate will also fall, therefore increasing the amount that the U.S. final goods producer demands from the Mexican intermediate firm. In order to meet this demand, the Mexican intermediate goods producer must rent more of the U.S. technology capital, Z Ui, and allocate more labor toward producing for the U.S. production line, which is more skill-intensive (Z Ui > Z Mi ) since the tartaric levels of technology capital will be determined by the skill level in country. This reallocation towards producing for the 30

32 U.S. final good will, therefore, increase the skill premium in Mexico. To see the second way in which the adoption channel might be activated, consider the intermediate goods producer s problem in Mexico (Equations (5) through (7)) and suppose that there is a reduction in the distortion to foreign technology capital. The intermediate goods producer is effectively paying a tax on his use of foreign technology capital, Z Ui. As the distortion on that technology capital falls, it becomes less costly for him to produce for the foreign final goods producer. Therefore, all else equal, he will reallocate resources toward that production line. This will again cause the producer to adopt more U.S. technology capital, thus affecting the skill premium as before. Investment Channel In order to understand the investment channel, consider the expression for the rental rate for the U.S. technology (Z Ui ) being used in Mexico by an intermediate goods producer for Industry i: where r jki = ρθ i α p [ ( jki(ω) A j θ i τ Zki α h1 α jki ) σ 1 σ ] 1 + (1 θ i )l σ 1 σ jki σ 1 α(σ 1) Z σ 1 ki h (1 α)(σ 1) σ jki, (12) τ = τ z jki τ jki. As can be seen in Equation (12), the tariff on the good and the tax on technology capital serve to reduce the return to renting technology across borders. Therefore, as either of these distortions decrease, the return to investing in the technology capital increases, thus increasing the U.S. final goods producer s investment. The key assumption here is that the technology capital is nonrivalrous and so the total return to investing in it is equal to the sum of the return in each country. All else equal, the domestic return will be unaffected by the reduction of the trade distortions, while the foreign return will increase. This means that the total return increases, incentivizing the final goods producer to increase the stock of technology capital. 31

33 6 Calibration and Quantitative Results The main goal of this paper is to examine the changing skill premium in Mexico. To facilitate this analysis, I focus on the liberalization between the United States and Mexico, as the United States is the dominant trade partner for Mexico and their trade relationship is dominated by the supply chain channels considered here. As such, the calibration will match features of the Mexican manufacturing sector. Here, I first calibrate the model and discuss the ability of the calibrated model to match key features of the data. I then conduct a number of counter-factual exercises and analyze the implications of the trade liberalization of interest for features of the Mexican economy. I will report results for the United States; however, the model will not be calibrated to match characteristics for specific U.S. industries. Instead, I will use the results generated for the United States in order to validate my calibration. I discuss several counter-factual exercises. 6.1 Calibration The model is parameterized to fit features of the Mexican manufacturing sector and its trade with the United States averaged over the pre-reform period, 1984 to As stated above, I consider only the United States and Mexico because trade with the United States represents roughly 70% of Mexico s total trade and close to 75% of Mexico s total exports in I restrict attention to only the manufacturing sector and set the number of industries to 20, which corresponds to the 20 (2-digit) broad manufacturing industries in Mexico (CMAP classification). Each period in the model corresponds to one year. My general calibration strategy is to group parameters into those that can be directly assigned and those that will be assigned such that endogenous outcomes of the model match certain observations in the data. I begin by assigning parameters directly. Table 6 presents the selected parameter values, as well as the source for these parameter selections. Because I have assumed that one model period equals one year, I set the risk-free interest rate equal to 4%, which implies a discount rate of This is an innocuous assumption for the initial calibration, as it plays no role in determining relative steady-state values and simply acts as a scale factor. It does, however, play some role in the speed of the transition from one steady state to the next. I follow McGrattan and Prescott (2009) and set the depreciation rate of technology capital to be Again, this plays little role in the determination of the stead state values and acts as a factor that scales steady-state values of technology capital up or down. The results hinge more on the relative stocks of technology capital in Mexico versus the U.S., so as long as I assume no asymmetry in depreciation across countries, δ can be set to any value and it will not change the relative steady-state values. The value selected 32

34 for δ will, again, affect the speed of transition between steady states. The parameters ρ and φ respectively determine the elasticity of substitution across intermediate goods for final goods production and across final goods for the consumer. These are set to match the median 5-digit SITC elasticity of substitution between 1990 and 2001, estimated by Broda and Weinstein (2006). These estimates are consistent with others used frequently in the trade literature. 12 I calculate the relative total factor productivity (TFP) in Mexico (A M ) to be Note that I normalize TFP in the United States to be 1. I then calculate the relative value added per worker in the manufacturing sector in Mexico in I choose the manufacturing sector instead of the overall economy because my skill premium data pertains to the manufacturing sector only. For the relative supply of high-skilled workers, I use household surveys that are available for both countries. While both countries have similar ratios of non-production to production employees in manufacturing, the ratio of college to non-college individuals differs substantially across the two. I use this difference in order to rationalize the large observed difference in initial skill premia. I follow the existing literature (see, for example Bustos, 2011) in defining skilled workers as those who have some college education (some enrollment in tertiary education). The data analog to high-skilled workers are non-production employees. In manufacturing, nonproduction workers include managers and technicians. As such, many have two-year technical degrees, and so the sample is extended to include those workers with some college in order to capture individuals with technical training. I do not want to rely on the ratio of non-production to production employees in manufacturing because this is an equilibrium outcome which is reflective of the skill intensity of manufacturing. I set the skill ratio in the model to match the percentage of the population aged 25 or older who have completed some college, as calculated using the Current Population Survey (accessed via IPUMS) in the case of the United States and La Encuesta Nacional de Ocupacion y Empleo (ENOE) in the case of Mexico. Following Burstein and Vogel (2017), I set the skill intensity parameters, θ i, to be equal to the share of hours worked by those with some college education in Industry i in Mexico in 1985, again using data from ENOE. My approach differs from Burstein and Vogel s in that, as previously discussed, I choose to include those with some college education to capture those with technical degrees. 13 Here, I only report the minimum and maximum values of θ for parsimony; the other values are reported in the online appendix. The last set of parameters that I impose exogenously are industry-level tariffs on exports 12 See, for example Burstein and Vogel (2017). 13 Similar to Burstein and Vogel, the share of high-skill workers in any given industry implied by the model will not exactly match the data. This is due, in part, to the fact that I am using the proportion of all high-skilled individuals in the economy as the supply of high-skilled workers, while the proportion of high- to low-skilled workers in manufacturing in Mexico is higher than this proportion in the general population. However, the skill intensity of each industry relative to some baseline industry looks similar in the model and in the data. 33

35 from Mexico to the U.S., τ y MUi, and on exports from the U.S. to Mexico, τy UMi. I use the tariffs described and exploited in the empirical section above. I allow these tariffs to differ across the countries and across industries. In my initial calibration, I use tariff rates that have been tradeweighted and averaged from the 4 digit industry to the 2 digit industry and then averaged from 1984 to The vector of tariff rates can be found in the online appendix. Parameter Value Source β 0.96 Annual return on risk-free bonds δ 0.08 McGrattan and Prescott (2009) ρ 0.63 Broda and Weinstein (2006) φ 2.7 Broda and Weinstein (2006) A M 0.25 Relative value-added per worker in 1985 H U H U +L U 0.28 CPS Fraction of Population with Some College 1985 H M H M +L M 0.09 ENOE Fraction of Population with Some College 1985 (θ min,θ max ) (0.256,0.573) ENOE Fraction of Industry i with Some College 1985 Table 6: Exogenously Set Parameter Values I now turn to calibration of the parameters that are assigned in order to match endogenous outcomes from the model with moments from the data. These include the importance of technology capital in the production of intermediate goods, α; the substitutability of high- and low-skilled labor, σ; and the distortions on technology capital from the United States being used in Mexico, as well as those from Mexico being used in the United States, τumi z and τz MUi, respectively. I assume that α and σ do not differ by country or by industry, but I allow τum z, and τz MU to vary by industry. If I set the number of industries, I, to be equal to 20, then I have 42 free parameters. I choose 42 moments of the data to match: the average skill premium in Mexico in the period from 1984 to 1986 (the pre-reform period), the ratio of royalties to payroll over the same time period, industry-by-industry intermediate exports from Mexico to the U.S. over total output of intermediates in Mexico, and industry-by-industry exports of intermediate goods from the U.S. to Mexico over total output of intermediates in the United States. I jointly match these moments in the data and the model via the Generalized Method of Moments (GMM), jointly minimizing the distance between the endogenous outcomes of the model and the observations from the data. I solve for these parameters iteratively in three stages. In the innermost loop, I assume values for all parameters and solve for the equilibrium of my model, producing model observations for skill intensity, trade flows, skill premia, and royalty payments. In a middle loop, I use the bilateral industry-level distortions on technology capital moving between the United States and Mexico 34

36 (τmui z and τz UMi ) in order to match industry-level bilateral exports between the two countries. In an outer loop, I take these distortions as given and then solve for α and σ in order to match the skill premium in Mexico and royalty payments over payroll. I use the inner two loops to solve for my counterfactual exercises. 6.2 Model Fit Table 7 presents the most parameter values that most impact the results and the model generated value for the moments that were targeted in order to calibrate these. I begin by matching the skill premium in Mexico and the average ratio of royalties to payroll in Mexican manufacturing. The estimated elasticity of substitution between high- and low-skill labor is consistent with estimates from the literature, though it is on the high side of the range of estimates. The model matches both of these moments exactly. I do not target the skill premium in the United States, but the model predicts this skill premium reasonably well, with a simulated value of I will analyze how the model performs relative to the data for the United States as check of its implications. Moment Parameter Data Model Value Ratio of Royalties to Payroll - Mexico α 0.12 Skill Premium - Mexico σ 1.92 Table 7: Target Moments Figures 5a and 5b present the moments generated in the model and those observed in the data for exports of intermediates from Mexico to the U.S. and from the U.S. to Mexico, respectively; these are the additional moments that I target for my calibration. I have included the 45 degree line for reference. As can be seen in the figures, the model is general able to replicate the industry-byindustry percentage of intermediate output that is exported from Mexico to the United States. The model replicates those industries that export a large proportion of their output well, such as industries dedicated to the production of auto parts, though it over predicts exports over output at the low end. the model matches the average exports over output, when averaged across industries in Mexico. On average, 5.3% of output in intermediates was exported from Mexico to the United States in The model predicts that 5.27% of output is exported from Mexico to the United States but under-predicts exports of intermediates from the United States to Mexico. This is due, in part, to the fact that I have assumed that the input-output structure of United States industries 35

37 (a) Mexican Exports by Industry (b) U.S. Exports by Industry Figure 5: Model vs. Data - Targeted Moments is equal to that in Mexico; if this is not true, the measure of intermediates for the United States may be inaccurate. The magnitude of this under-prediction by the model is, on average, about 0.3 percentage points. 14 I have not targeted the skill premium in the United States and the model under-performs on this metric as well. In the data, the skill premium in the United States is roughly 1.38 in 1984, whereas the model predicts that it should be closer to 1.14 for the same year. Again, I have assumed that the production technologies are the same across the two countries. However, one can imagine that technology capital is more important in production in the United States than it is in Mexico. This would imply higher royalty payments relative to payroll, which in turn, would imply a higher value for α. I do not have information on royalty payments for the United States, nor is it the focus of the paper. However, I will report results for the United States to show that the model s predictions qualitatively match the observed outcomes for the United States after its integration with Mexico. 6.3 Counterfactual Experiments I now conduct a series of counterfactual experiments. I begin by simply changing the tariff rates from their observed level in 1986 level to the level observed in I choose these two dates to match the empirical analysis done above. Furthermore, the skill premium in Mexico stops increasing in 1996, just after these reforms are completed. I then turn to an analysis that allows me to analyze the full effect of the liberalization, by including changes in tariffs and in the frictions 14 The model predicts 0.7% of output of intermediates from the U.S. are exported to Mexico, while in the data, this is closer to 1%. 36

38 on the use of technology capital, τ z. Next, I analyze what would happen if the U.S. were the barriers to the use of foreign technology completely eroded, to get a sense of what would happen, for example, if there were an international court with jurisdiction over all intellectual property. I complete the counterfactuals by comparing my results to what occurred when the U.S. and Canada opened to one another. Change in Tariff Rates Only I begin with an experiment where I hold fixed all parameters of the model but vary the tariff rates to match the observed tariff rates in The observed tariffs faced by Mexican intermediate goods producers who exported to the United States in 1984 were 30% and for U.S. intermediate goods producers who exported to Mexico, they were about 24%. By 1994, these had fallen to an average of 7% for Mexican intermediate goods producers and 8% for U.S. intermediate goods producers. This is a large reduction in tariffs relative to most trade liberalizations. Some industries see an 80% drop in their tariff rates. However, as Table 8 shows, this decrease in tariffs alone has very little impact on the skill premium in either country; when tariffs fall, the skill premium in Mexico increases by 0.005, from 2.03 to and average exports from Mexican producers of intermediates over their total output changes by just under one percentage point. This is roughly a 17% increase in these flows. In the data, exports of intermediates as a percentage of their total output increases by nearly fourfold over this time period. The fact that the drop in tariffs does not produce large increases in trade is not entirely surprising, given that many papers that study decreases in tariffs of a similar size find very small impacts of these tariff reductions on trade flows. The skill premium in the United States only increases by when the new bilateral tariff schedule is imposed, though exports of intermediates from the United States to Mexico increase by 9% in the model following this liberalization. The small change in the skill premium in Mexico can be accounted for by the fact that exports of intermediates change by very little in the case that only tariffs fall. Because the Mexican intermediate-goods-producing firm must be part of the supply chain of U.S. final-goods producer in order to use the more advanced technology capital, low exports of intermediates from Mexico to the U.S. predict low adoption of U.S. technology capital by Mexican intermediate goods producers. Likewise, low adoption of U.S. technology capital by Mexican producers implies that the change in the return to that capital will also be low, therefore predicting that the skill premium in the United States should remain virtually unchanged. 37

39 Figure 6: U.S. Exports by Industry: Post-Reform Change in Tariffs and Frictions on Use of Foreign Technology Capital I now instead evaluate the case where both tariffs and barriers to the use of foreign technology capital change. In this exercise, tariffs are lowered as observed in the data and the the frictions on foreign technology capital are re-calibrated to match the exports of intermediate goods as a fraction of their output in I re-calibrate the frictions both on the use of U.S. technology capital in Mexico and on the use of Mexican technology capital in the United States. 15 Figure 6 shows the fit of the model to the data in the post-reform period for Mexican exports to the U.S. As can be seen, the model generates trade flows that are commensurate with the observations in the data. Average exports of intermediates as a fraction of their output from Mexico to the United States are roughly 16% post-reform, in both the model and the data. The model continues to underpredict exports from the United States to Mexico by approximately 7%. This under-prediction has a similar magnitude in the pre- and post-reform periods. 16 Table 8 reports the impact of these reforms on the skill premium. Notice that, as discussed above, when only tariffs change, we see very little impact on the skill premium in both countries. However, when we change both frictions, the model is able to generate about 10% of the observed increase in the skill premium in Mexico. Moreover, the model produces an increase in the skill premium in the United States. This increase in the skill premium in the U.S. is substantially lower than the observed increase in the data; however, the goal of the model is to explore the impact on Mexico, while producing results that do not move in a counterfactual direction in the United States. 15 If I instead only change the frictions on the use of U.S. technology capital in Mexico, the results are quantitatively very similar. 16 In the post-reform period, average exports from the United States to Mexico are 2.7% in the data and 2.5% in the model. 38

40 Pre-Reform Post-Reform Data Model Data Model - Change in Tariffs Only Model - Change in Both (1) (2) (3) (4) (5) Mexico U.S Table 8: Results - Trade Liberalization Figure 7 presents the time path for the changes in the skill premium post-reform. The data is represented by the line with circles, while the baseline reform is represented by the solid line. Here, I assume that the policy is implemented immediately in 1986, with no phase in. This largely mimics the reform that was implemented in Mexico, as tariffs were reduced over the latter half of Because I use the trade flows to discipline the frictions on technology capital and no direct measure of these frictions, I will also assume that the frictions fall all at once, at the end of 1986 when the other reforms are put in place. This is in line with what DuMars (1991) finds; part of the 1986 negotiation included discussions of intellectual property protections that resulted in Mexico stating that it was adopting U.S. intellectual property protections. As such, we can see that the change in the skill premium in Mexico occurs primarily in the first few periods after the reform. This is due to the operation of the adoption channel. When the barriers to producing for the foreign final goods producer fall, Mexican intermediate goods producers immediately switch part of their production line from producing for the domestic final goods producer to producing for the foreign one. The rest of the adoption comes from increased investment in skill-augmenting technology capital, both in Mexico and in the United States. As revenues for the final goods producer rise as a function of renting more of its technology capital, the returns to renting this increase. This induces the final goods producer to invest more heavily in technology capital, and results in additional increases in the skill premium. Initial adoption accounts for roughly 60% of the increase in the skill premium that is generated by the model, with the investment channel generating the rest of the increase. The model predicts impacts on labor markets that match previous studies. Several other authors, including Bustos (2011) have found that trade liberalizations primarily cause labor reallocation within industries, and not across them. My model is largely consistent with this finding. In the case where tariffs change in isolation, there is no labor reallocation across industries. In the case where both tariffs and frictions to technology capital fall, there is some reallocation across industries, as the relative barriers to the use of foreign technology change across industries. However, most industries see less than 10% change in the fraction of labor allocated towards their industry, 39

41 Figure 7: Skill Premium in Mexico Post-Reform with the one exception being the production of automobiles and auto-parts. Employment in this industry increases by 38%. In the pre-reform period, this industry accounted for roughly 8% of the total labor market employed in manufacturing in the model and by 1994, it accounts for 11%. This growth in the fraction of employment that can be accounted for by this industry is similar in the data; in 1986, this the auto industry accounted for roughly 10% of Mexican manufacturing employment. By 1994, it accounted for roughly 13%. The model generates results mainly through reallocation of labor from domestic production to foreign production. There are some industries in which Mexican intermediate goods producers reallocate towards domestic production. All Barriers to Use of Foreign Capital Erode In order to explore the impact of international law protecting intellectual property, I move to the extreme case of complete erosion of the barriers to the use of foreign technology capital. In this case, I allow tariffs to remain at their observed rates in This causes the skill premium in Mexico to increase dramatically, as can be seen in Figure 7. In this figure, the hatched line represents the case tariff rates fall as they did in the data and distortions to the use of foreign technology capital are completely eroded. This results in a 30% increase in Mexico s skill premium, which is roughly half of the increase that was observed in the data. Similarly, the United States sees an increase in its skill premium, moving it from 1.14 in the baseline calibration to 1.2, a 5% increase, after the barriers to technology capital are removed. This extreme case serves to highlight the role of the transmission of technology capital in determining the skill premium. If technologies are not transmitted as part of a trade liberalization, we should not expect to observe large increases in the skill premium. This result is related to the finding highlighted in Ripoll (2005) that some liberalizations between developed and developing countries are accompanied by large increases in the skill premium, while others are not. My model 40