reading discussion

profilexiang
MelitzTreflerJEP.pdf

Journal of Economic Perspectives—Volume 26, Number 2—Spring 2012—Pages 91–118

TT he gains from long-distance international trade have been understood and he gains from long-distance international trade have been understood and exploited since prehistoric times. Our pre-urban ancestors were benefi t-exploited since prehistoric times. Our pre-urban ancestors were benefi t-ting from long-distance trade in obsidian some 10,000 years ago; Plato’s ting from long-distance trade in obsidian some 10,000 years ago; Plato’s Academy was built on the profi ts of Athenian silver exports; and Rome was not built Academy was built on the profi ts of Athenian silver exports; and Rome was not built in a day partly because goods moved too slowly in the vast Roman trade network.in a day partly because goods moved too slowly in the vast Roman trade network.

But whereas trade was once dominated by the movement of goods that could But whereas trade was once dominated by the movement of goods that could only be produced, harvested, or mined regionally, the international trade landscape only be produced, harvested, or mined regionally, the international trade landscape is now dominated by two striking facts. The fi rst is the rise of intra-industry trade—is now dominated by two striking facts. The fi rst is the rise of intra-industry trade— that is, two-way trade in similar products. Chinese consumers can now buy a midsize that is, two-way trade in similar products. Chinese consumers can now buy a midsize car from Toyota (Japan), Kia (Korea), General Motors (United States), and Chery car from Toyota (Japan), Kia (Korea), General Motors (United States), and Chery (China). Ditto for consumers in Japan, Korea, and the United States. The second (China). Ditto for consumers in Japan, Korea, and the United States. The second striking fact is that world trade is dominated by huge, innovative, and extraordi-striking fact is that world trade is dominated by huge, innovative, and extraordi- narily productive fi rms. For example, Intel is so large that it is the largest industrial narily productive fi rms. For example, Intel is so large that it is the largest industrial employer in both Oregon and New Mexico and accounts for 20 percent of Costa employer in both Oregon and New Mexico and accounts for 20 percent of Costa Rica’s exports. China’s Foxconn infamously employs 450,000 workers in a Rica’s exports. China’s Foxconn infamously employs 450,000 workers in a single one one of its many export-oriented electronics factories. These are big companies . . . and of its many export-oriented electronics factories. These are big companies . . . and if you are reading this document on an Apple computer you know that there are if you are reading this document on an Apple computer you know that there are other large companies, too.other large companies, too.

The rising prominence of intra-industry trade and huge multinationals has The rising prominence of intra-industry trade and huge multinationals has transformed the way economists think about the gains from trade. In the past, we transformed the way economists think about the gains from trade. In the past, we focused on gains that stemmed either from endowment differences (wheat for focused on gains that stemmed either from endowment differences (wheat for

Gains from Trade when Firms Matter

■ ■ Marc J. Melitz is the David A. Wells Professor of Political Economy, Harvard University, Cambridge, Massachusetts. Daniel Trefl er is the J. Douglas and Ruth Grant Canada Research Chair in Competitiveness and Prosperity, Rotman School of Management, University of Toronto, Toronto, Ontario, Canada.

http://dx.doi.org/10.1257/jep.26.2.91. doi=10.1257/jep.26.2.91

Marc J. Melitz and Daniel Trefl er

92 Journal of Economic Perspectives

iron ore) or inter-industry comparative advantage (David Ricardo’s classic example iron ore) or inter-industry comparative advantage (David Ricardo’s classic example of cloth for port). Today, we focus on three sources of gains from trade: 1) love-of cloth for port). Today, we focus on three sources of gains from trade: 1) love- of-variety gains associated with intra-industry trade; 2) allocative effi ciency gains of-variety gains associated with intra-industry trade; 2) allocative effi ciency gains associated with shifting labor and capital out of small, less-productive fi rms and associated with shifting labor and capital out of small, less-productive fi rms and into large, more-productive fi rms; and 3) productive effi ciency gains associated with into large, more-productive fi rms; and 3) productive effi ciency gains associated with trade-induced innovation.trade-induced innovation.

Back in the 1980s, a “New Trade Theory” was developed that focused on intra-Back in the 1980s, a “New Trade Theory” was developed that focused on intra- industry trade in differentiated goods produced subject to increasing returns to industry trade in differentiated goods produced subject to increasing returns to scale. This theory centered on an elegant tension: consumers love variety and are scale. This theory centered on an elegant tension: consumers love variety and are willing to pay a premium for a desired product, but as the market fragments into willing to pay a premium for a desired product, but as the market fragments into niche products, producers struggle to attain the volumes needed to recoup their niche products, producers struggle to attain the volumes needed to recoup their product development costs. International trade creates a larger marketplace, which product development costs. International trade creates a larger marketplace, which means that each fi rm can operate at a larger scale and hence more fi rms can survive. means that each fi rm can operate at a larger scale and hence more fi rms can survive. The result reads like an advertisement for free trade: lower prices, more varieties. The result reads like an advertisement for free trade: lower prices, more varieties. Paul Krugman earned the Nobel Prize in 2008 in large part for his work highlighting Paul Krugman earned the Nobel Prize in 2008 in large part for his work highlighting how economies of scale and product differentiation lead to intra-industry trade, just how economies of scale and product differentiation lead to intra-industry trade, just as in our example above of midsize cars. See Krugman (1979, 1980), Helpman and as in our example above of midsize cars. See Krugman (1979, 1980), Helpman and Krugman (1985), and Helpman (2011, chap. 4) for a review of love-of-variety gains Krugman (1985), and Helpman (2011, chap. 4) for a review of love-of-variety gains from trade.from trade.

More recently, a second source of gains from trade has emerged from the More recently, a second source of gains from trade has emerged from the research of Melitz (2003) and Bernard, Eaton, Jensen, and Kortum (2003). This research of Melitz (2003) and Bernard, Eaton, Jensen, and Kortum (2003). This is the fi rm-level “reallocation” effect that arises when there is fi rm heterogeneity. is the fi rm-level “reallocation” effect that arises when there is fi rm heterogeneity. By fi rm heterogeneity we mean that even within narrowly defi ned industries some By fi rm heterogeneity we mean that even within narrowly defi ned industries some fi rms are much larger and more profi table than others because, for example, fi rms are much larger and more profi table than others because, for example, they are much more productive. Globalization generates both winners and losers they are much more productive. Globalization generates both winners and losers among fi rms within an industry and these effects are magnifi ed by heterogeneity. among fi rms within an industry and these effects are magnifi ed by heterogeneity. Better-performing fi rms thrive and expand into foreign markets, while worse-Better-performing fi rms thrive and expand into foreign markets, while worse- performing fi rms contract and even shut down in the face of foreign competition. performing fi rms contract and even shut down in the face of foreign competition. This generates a new source of gains from trade: as production is concentrated This generates a new source of gains from trade: as production is concentrated towards better-performing fi rms, the overall effi ciency of the industry improves. towards better-performing fi rms, the overall effi ciency of the industry improves. In this way, globalization raises average effi ciency In this way, globalization raises average effi ciency within an industry. Why is it that . Why is it that only the better-performing fi rms grow? Globalization expands markets but also only the better-performing fi rms grow? Globalization expands markets but also increases competition in those markets. This competition effect dominates for the increases competition in those markets. This competition effect dominates for the worse-performing fi rms while the increased market access dominates for the better-worse-performing fi rms while the increased market access dominates for the better- performing fi rms. Also, a fi rm’s international expansion—whether by exporting, performing fi rms. Also, a fi rm’s international expansion—whether by exporting, by offshore outsourcing of intermediate components and assembly, or by building by offshore outsourcing of intermediate components and assembly, or by building plants abroad (multinationals)—entails some up-front fi xed costs; and only the plants abroad (multinationals)—entails some up-front fi xed costs; and only the best-performing fi rms have the sales volumes needed to justify these fi xed costs.best-performing fi rms have the sales volumes needed to justify these fi xed costs.

Our third source of gains from trade comes from the positive impacts of larger Our third source of gains from trade comes from the positive impacts of larger markets on markets on innovation. New productivity-enhancing products and processes require . New productivity-enhancing products and processes require up-front development costs. Trade integration, by expanding the size of the market, up-front development costs. Trade integration, by expanding the size of the market, encourages fi rms to pony up these development dollars, and this in turn raises encourages fi rms to pony up these development dollars, and this in turn raises productivity. Theories of innovation-based gains from trade with homogeneous productivity. Theories of innovation-based gains from trade with homogeneous fi rms were developed by Grossman and Helpman (1991) and are supported by fi rms were developed by Grossman and Helpman (1991) and are supported by

Marc J. Melitz and Daniel Trefl er 93

country-level evidence (Helpman 2004, chap. 5.6). At the fi rm level, there is a strong country-level evidence (Helpman 2004, chap. 5.6). At the fi rm level, there is a strong relationship between exporting and innovation. For example, Intel and Apple are relationship between exporting and innovation. For example, Intel and Apple are major patent holders, and Foxconn holds 40 percent of all Chinese patents fi led in major patent holders, and Foxconn holds 40 percent of all Chinese patents fi led in the United States (Eberhardt, Helmers, and Yu 2011). Of course, this correlation the United States (Eberhardt, Helmers, and Yu 2011). Of course, this correlation between exporting and innovation does not provide evidence for causality and lacks between exporting and innovation does not provide evidence for causality and lacks a framing theory featuring heterogeneous fi rms. Recently, however, there has been a framing theory featuring heterogeneous fi rms. Recently, however, there has been a great deal of theoretical and empirical progress. Lileeva and Trefl er (2010) show a great deal of theoretical and empirical progress. Lileeva and Trefl er (2010) show theoretically and empirically how the market-expanding effects of international theoretically and empirically how the market-expanding effects of international integration causally encourage fi rms to innovate. Verhoogen (2008), Bustos (2011a, integration causally encourage fi rms to innovate. Verhoogen (2008), Bustos (2011a, b), and Aw, Roberts, and Xu (2011) assess other interesting channels through which b), and Aw, Roberts, and Xu (2011) assess other interesting channels through which trade promotes fi rm-level innovation. Note that this third source of gains deals trade promotes fi rm-level innovation. Note that this third source of gains deals with with within-fi rm effi ciency; in contrast, the second source of gains above deals with effi ciency; in contrast, the second source of gains above deals with between-fi rm or allocative effi ciency. or allocative effi ciency.

This paper reviews these three sources of gains from trade both theoretically This paper reviews these three sources of gains from trade both theoretically and empirically. Our empirical evidence will be centered on the experience of and empirically. Our empirical evidence will be centered on the experience of Canada following its closer economic integration in 1989 with the United States—Canada following its closer economic integration in 1989 with the United States— the largest example of bilateral intra-industry trade in the world—but we will also the largest example of bilateral intra-industry trade in the world—but we will also describe evidence for other countries.describe evidence for other countries.

The related literature is huge. Here we focus on fi rms that expand interna-The related literature is huge. Here we focus on fi rms that expand interna- tionally via exporting as in Melitz (2003) and Bernard, Eaton, Jensen, and Kortum tionally via exporting as in Melitz (2003) and Bernard, Eaton, Jensen, and Kortum (2003). Another related research topic analyzes how fi rm boundaries evolve across (2003). Another related research topic analyzes how fi rm boundaries evolve across borders as multinational fi rms look abroad to “outsource” key parts of their produc-borders as multinational fi rms look abroad to “outsource” key parts of their produc- tion chain. The interested reader is directed to surveys by Antràs and Rossi-Hansberg tion chain. The interested reader is directed to surveys by Antràs and Rossi-Hansberg (2009) and Helpman (2011, chap. 6).(2009) and Helpman (2011, chap. 6).

Gains from Love of Variety (Economies of Scale and Product Variety)

Our fi rst source of gains from trade is intimately related to intra-industry trade. Our fi rst source of gains from trade is intimately related to intra-industry trade. To measure intra-industry trade, one needs to start with a classifi cation system that To measure intra-industry trade, one needs to start with a classifi cation system that assigns trade fl ows to a particular “industry.” One can then categorize trade fl ows as assigns trade fl ows to a particular “industry.” One can then categorize trade fl ows as either intra-industry (two-way trade within the same industry classifi cation code) or either intra-industry (two-way trade within the same industry classifi cation code) or inter-industry (imports and exports in separate industry codes). The United Nations inter-industry (imports and exports in separate industry codes). The United Nations uses the Standard International Trade Classifi cation, or SITC, to categorize world uses the Standard International Trade Classifi cation, or SITC, to categorize world trade fl ows. In its most detailed form, the SITC contains 1,161 separate industry trade fl ows. In its most detailed form, the SITC contains 1,161 separate industry codes (that can be consistently traced back over time), but these industries are often codes (that can be consistently traced back over time), but these industries are often aggregated into a smaller subset of industries.aggregated into a smaller subset of industries.

Figure 1 shows the time trend for the share of intra-industry trade according to Figure 1 shows the time trend for the share of intra-industry trade according to this most detailed classifi cation, and a more aggregated version with only 59 industry this most detailed classifi cation, and a more aggregated version with only 59 industry codes. Mechanically, the share of intra-industry trade rises with the level of aggrega-codes. Mechanically, the share of intra-industry trade rises with the level of aggrega- tion for the industrial classifi cation system (after all, with a single aggregate industry tion for the industrial classifi cation system (after all, with a single aggregate industry code, all trade would be “intra” to this aggregated industry). However, the time code, all trade would be “intra” to this aggregated industry). However, the time trends for the two series are very similar: intra-industry trade grew rapidly from trends for the two series are very similar: intra-industry trade grew rapidly from 1962 to the mid 1990s, before stabilizing at a substantially higher level. As countries 1962 to the mid 1990s, before stabilizing at a substantially higher level. As countries

94 Journal of Economic Perspectives

industrialize, they tend to experience a higher share of intra-industry trade because industrialize, they tend to experience a higher share of intra-industry trade because they tend to produce and export differentiated manufactured goods that are similar they tend to produce and export differentiated manufactured goods that are similar to other brands of goods that are imported. However, some of the countries with to other brands of goods that are imported. However, some of the countries with the highest shares of intra-industry trade in 2000 were newly industrializing nations, the highest shares of intra-industry trade in 2000 were newly industrializing nations, such as the Czech Republic (77 percent), the Slovak Republic (76 percent), Mexico such as the Czech Republic (77 percent), the Slovak Republic (76 percent), Mexico (73 percent), and Hungary (72 percent); for comparison, the United States had (73 percent), and Hungary (72 percent); for comparison, the United States had a 69 percent share of intra-industry trade in 2000 (a 69 percent share of intra-industry trade in 2000 (OECD Economic Outlook 2002, 2002, chap. 6; based on the 59-industry level of aggregation). Most recently, China’s share chap. 6; based on the 59-industry level of aggregation). Most recently, China’s share of intra-industry trade has risen above the 50 percent mark.of intra-industry trade has risen above the 50 percent mark.

Why might a country both export and import goods that are similar? As a Why might a country both export and import goods that are similar? As a starting point, consider world trade in automobiles. Consumers in a car-producing starting point, consider world trade in automobiles. Consumers in a car-producing country are not limited to buying the car models that are produced domestically: country are not limited to buying the car models that are produced domestically: many of those consumers choose to buy models that are produced elsewhere and many of those consumers choose to buy models that are produced elsewhere and imported. The extent of this product differentiation is then limited by high fi xed imported. The extent of this product differentiation is then limited by high fi xed start-up costs for a new brand and by the related economies of scale.start-up costs for a new brand and by the related economies of scale.

We now highlight how the combination of product differentiation and econo-We now highlight how the combination of product differentiation and econo- mies of scale generates intra-industry trade using a theoretical example. Notice that mies of scale generates intra-industry trade using a theoretical example. Notice that this source of gains from trade provides a rationale for trade between two identical this source of gains from trade provides a rationale for trade between two identical countries, which provides a stark contrast with the gains from inter-industry trade countries, which provides a stark contrast with the gains from inter-industry trade

Figure 1 World Share of Intra-Industry Trade 1962–2006

Source: Data from Brühlhart (2009). We thank Marius Brülhart for generously sharing his data. Notes: Figure 1 shows the time trend for the share of intra-industry trade according to the most detailed Standard International Trade Classifi cation (1,161 separate industry codes) and a more aggregated version with only 59 industry codes.

.6

.5

.4

.3

.2

.1

Aggregated industry classification Detailed industry classification

200620021998199419901986198219781974197019661962

Gains from Trade when Firms Matter 95

that arise from exploiting differences across countries, such as differences in tech-that arise from exploiting differences across countries, such as differences in tech- nology (Ricardo) or differences in factor supplies (Heckscher–Ohlin).nology (Ricardo) or differences in factor supplies (Heckscher–Ohlin).

In our theoretical example, two identical countries produce differentiated In our theoretical example, two identical countries produce differentiated widget varieties subject to the same constant-returns-to-scale technology. Assume widget varieties subject to the same constant-returns-to-scale technology. Assume that one worker can produce one widget, but that production of any new variety of that one worker can produce one widget, but that production of any new variety of widgets requires four workers to cover fi xed overhead costs: this implies decreasing widgets requires four workers to cover fi xed overhead costs: this implies decreasing average costs of production as the fi xed cost is spread over an increasing number of average costs of production as the fi xed cost is spread over an increasing number of output units (hence the economies of scale). Also, to be specifi c, suppose that both output units (hence the economies of scale). Also, to be specifi c, suppose that both countries have a fi xed supply of 12 workers. If they do not trade, then each country countries have a fi xed supply of 12 workers. If they do not trade, then each country can produce: A) 8 units of one variety, or B) 2 units each of two different varieties. can produce: A) 8 units of one variety, or B) 2 units each of two different varieties.

Allowing countries to trade leads to a new possibility that is better than what Allowing countries to trade leads to a new possibility that is better than what either country can achieve on its own. Suppose that each country produces 8 units either country can achieve on its own. Suppose that each country produces 8 units of one variety and exports 4 of these units to the other country. Consumers are now of one variety and exports 4 of these units to the other country. Consumers are now consuming 4 units of the home variety and 4 units of the foreign variety. This is consuming 4 units of the home variety and 4 units of the foreign variety. This is preferred to either of the no-trade production plans above. Compared to choice B, preferred to either of the no-trade production plans above. Compared to choice B, there is the same number of varieties (2 varieties), but more of each variety (four there is the same number of varieties (2 varieties), but more of each variety (four versus two). Compared to choice A, there is the same number of units (8 units), but versus two). Compared to choice A, there is the same number of units (8 units), but more varieties (two versus one). Thus, trade expands the set of consumer choices more varieties (two versus one). Thus, trade expands the set of consumer choices and eases the tradeoff between consumption units and product variety. Economic and eases the tradeoff between consumption units and product variety. Economic integration allows production of each individual variety to be consolidated for the integration allows production of each individual variety to be consolidated for the whole integrated market; given increasing returns to scale, this reduces average whole integrated market; given increasing returns to scale, this reduces average production costs. At the same time, product variety increases because consumers production costs. At the same time, product variety increases because consumers can buy varieties produced anywhere in the integrated market.can buy varieties produced anywhere in the integrated market.

One of the most salient real-world examples of economic integration between One of the most salient real-world examples of economic integration between similar countries occurred between the United States and Canada. This integration similar countries occurred between the United States and Canada. This integration started with the signing of the North American Auto Pact in 1964. Before then, most started with the signing of the North American Auto Pact in 1964. Before then, most car models were produced in the United States for U.S. consumers and in Canada car models were produced in the United States for U.S. consumers and in Canada for Canadian consumers. High tariffs on auto trade made it uneconomical to export for Canadian consumers. High tariffs on auto trade made it uneconomical to export most car models across the border. Because the Canadian auto market was roughly most car models across the border. Because the Canadian auto market was roughly one-tenth the size of the U.S. market, this implied substantial scale disadvantages for one-tenth the size of the U.S. market, this implied substantial scale disadvantages for production in the Canadian market: labor productivity there was about 30 percent production in the Canadian market: labor productivity there was about 30 percent below the U.S. level. The U.S. market was large enough that assembly lines could be below the U.S. level. The U.S. market was large enough that assembly lines could be dedicated to one particular car model, while Canadian assembly lines had to switch dedicated to one particular car model, while Canadian assembly lines had to switch across models, involving costly downtime and reconfi guration costs, and also had to across models, involving costly downtime and reconfi guration costs, and also had to hold substantially higher inventory levels.hold substantially higher inventory levels.

The 1964 Pact established a free trade area for autos that allowed manufac-The 1964 Pact established a free trade area for autos that allowed manufac- turers to consolidate the production of particular car models in one country and turers to consolidate the production of particular car models in one country and export that model to consumers in the other country. For example, General Motors export that model to consumers in the other country. For example, General Motors cut in half the number of models assembled in Canada. However, total produc-cut in half the number of models assembled in Canada. However, total produc- tion of autos in Canada increased as the remaining models produced in Canada tion of autos in Canada increased as the remaining models produced in Canada supplied the U.S. market as well as the Canadian one. Canadian automotive exports supplied the U.S. market as well as the Canadian one. Canadian automotive exports to the United States increased from $16 million in 1962 to $2.4 billion in 1968. That to the United States increased from $16 million in 1962 to $2.4 billion in 1968. That same year, U.S. automotive exports to Canada were valued at $2.9 billion: intra-same year, U.S. automotive exports to Canada were valued at $2.9 billion: intra- industry trade in action. Today, $85 billion worth of automotive products cross the industry trade in action. Today, $85 billion worth of automotive products cross the

96 Journal of Economic Perspectives

U.S.–Canada border each year—roughly half in each direction. The productivity U.S.–Canada border each year—roughly half in each direction. The productivity gains associated with this consolidation were also substantial: by the early 1970s, the gains associated with this consolidation were also substantial: by the early 1970s, the Canadian auto industry’s 30 percent labor productivity shortfall relative to its U.S. Canadian auto industry’s 30 percent labor productivity shortfall relative to its U.S. counterpart had disappeared.counterpart had disappeared.

Later, this transformation of the automotive industry was extended to include Later, this transformation of the automotive industry was extended to include Mexico. In 1989, Volkswagen consolidated its North American operations in Mexico. In 1989, Volkswagen consolidated its North American operations in Mexico, shutting down its plant in Pennsylvania. This process continued with the Mexico, shutting down its plant in Pennsylvania. This process continued with the implementation of the North American Free Trade Agreement between the United implementation of the North American Free Trade Agreement between the United States, Canada, and Mexico. In 1994, Volkswagen started producing the new Beetle States, Canada, and Mexico. In 1994, Volkswagen started producing the new Beetle for the entire North American market in that same Mexican plant.for the entire North American market in that same Mexican plant.

This consolidation in response to closer economic integration with the United This consolidation in response to closer economic integration with the United States was not limited to the auto industry. Following the implementation of the States was not limited to the auto industry. Following the implementation of the Canada–U.S. Free Trade Agreement in 1989, each Canadian manufacturing industry Canada–U.S. Free Trade Agreement in 1989, each Canadian manufacturing industry experienced a dramatic reduction in its product offerings, concentrating on a smaller experienced a dramatic reduction in its product offerings, concentrating on a smaller number of products (Baldwin, Beckstead, and Caves 2002; Baldwin, Caves, and Gu number of products (Baldwin, Beckstead, and Caves 2002; Baldwin, Caves, and Gu 2005; Baldwin and Gu 2009; Bernard, Redding, and Schott 2011). Baldwin, Caves, 2005; Baldwin and Gu 2009; Bernard, Redding, and Schott 2011). Baldwin, Caves, and Gu (2005) also report that the decrease in product offerings was accompanied and Gu (2005) also report that the decrease in product offerings was accompanied by substantial increases in production runs for individual products. This process is by substantial increases in production runs for individual products. This process is even evident in the Canadian wine industry, an industry that exclusively produced even evident in the Canadian wine industry, an industry that exclusively produced low-end wines that could not possibly compete with Californian giants such as Gallo. low-end wines that could not possibly compete with Californian giants such as Gallo. In response to the Agreement, Canadian manufactures dramatically reduced the In response to the Agreement, Canadian manufactures dramatically reduced the number of varietals produced and focused on the varietals used to produce ice wine. number of varietals produced and focused on the varietals used to produce ice wine. The industry is now healthier than ever (Beamish and Celly 2003).The industry is now healthier than ever (Beamish and Celly 2003).

Another prominent example of economic integration began in 1957, when the Another prominent example of economic integration began in 1957, when the major countries of Western Europe established a free trade area in manufactured major countries of Western Europe established a free trade area in manufactured goods: the European Economic Community, or EEC. Many politicians offered an goods: the European Economic Community, or EEC. Many politicians offered an old-fashioned Ricardian prediction that German manufacturers would eradicate old-fashioned Ricardian prediction that German manufacturers would eradicate their European competitors. The facts did not treat such predictions kindly: trade their European competitors. The facts did not treat such predictions kindly: trade within the EEC grew twice as fast as world trade during the 1960s, and intra-industry within the EEC grew twice as fast as world trade during the 1960s, and intra-industry trade as a share of EEC trade more than doubled from 1960 to 1990. The benefi ts trade as a share of EEC trade more than doubled from 1960 to 1990. The benefi ts of the original European Community agreement were about 1 percent of GDP for of the original European Community agreement were about 1 percent of GDP for the largest economies and about 3 percent of GDP for mid-sized economies such the largest economies and about 3 percent of GDP for mid-sized economies such as Belgium (Harrison, Rutherford, and Wooton 1989). (These numbers capture as Belgium (Harrison, Rutherford, and Wooton 1989). (These numbers capture more than just pure love-of-variety gains.) Economic integration has continued in more than just pure love-of-variety gains.) Economic integration has continued in Europe as more countries have joined the free trade area that is today called the Europe as more countries have joined the free trade area that is today called the European Union and as a subset of EU countries adopted the euro as a common European Union and as a subset of EU countries adopted the euro as a common currency in 1999. Euro-zone members have experienced strong trade growth, espe-currency in 1999. Euro-zone members have experienced strong trade growth, espe- cially intra-industry trade growth, relative to non-EU countries and even relative to cially intra-industry trade growth, relative to non-EU countries and even relative to EU countries that have not adopted the euro.EU countries that have not adopted the euro.

A substantial portion of the increased trade that comes with economic integra-A substantial portion of the increased trade that comes with economic integra- tion also delivers increased product variety to consumers. Balistreri, Hilberrry, and tion also delivers increased product variety to consumers. Balistreri, Hilberrry, and Rutherford (2011) show that the worldwide elimination of all trade barriers would Rutherford (2011) show that the worldwide elimination of all trade barriers would raise the number of varieties available by about 3 percent, lower manufacturing raise the number of varieties available by about 3 percent, lower manufacturing prices by a similar amount, and raise world welfare by 2 percent. Most of these prices by a similar amount, and raise world welfare by 2 percent. Most of these

Marc J. Melitz and Daniel Trefl er 97

gains would accrue to developing countries such as China. Broda and Weinstein gains would accrue to developing countries such as China. Broda and Weinstein (2006) estimate that the number of products available to U.S. consumers through (2006) estimate that the number of products available to U.S. consumers through imports tripled between 1972 and 2001, resulting in welfare gains to U.S. consumers imports tripled between 1972 and 2001, resulting in welfare gains to U.S. consumers equivalent to a 2.6 percent rise in U.S. GDP. Feenstra (2010, table 2.1) examines equivalent to a 2.6 percent rise in U.S. GDP. Feenstra (2010, table 2.1) examines how worldwide welfare would change if all countries went from autarky to their how worldwide welfare would change if all countries went from autarky to their 1996 levels of trade. He estimates that the welfare gains from increased varieties 1996 levels of trade. He estimates that the welfare gains from increased varieties are comparable to a 12.5 percent rise in world GDP. While the exact magnitudes are comparable to a 12.5 percent rise in world GDP. While the exact magnitudes of the gains from increased variety differ across studies due to differences in what of the gains from increased variety differ across studies due to differences in what exactly is being modeled, the clear message is that the gains have been very large for exactly is being modeled, the clear message is that the gains have been very large for developed countries and continue to be large for developing countries.developed countries and continue to be large for developing countries.

Trade expands product variety both in fi nal goods (which benefi ts consumers) Trade expands product variety both in fi nal goods (which benefi ts consumers) as well as in specialized production inputs (which benefi ts the fi rms that use those as well as in specialized production inputs (which benefi ts the fi rms that use those inputs). Ethier (1982) showed that there is a close parallel between these two. Instead inputs). Ethier (1982) showed that there is a close parallel between these two. Instead of the love-of-variety that accrues to consumers, fi rms benefi t from the increased of the love-of-variety that accrues to consumers, fi rms benefi t from the increased productivity derived from an increased range of available production inputs. Recent productivity derived from an increased range of available production inputs. Recent fi rm-level research has confi rmed this product variety benefi t for fi rms that import fi rm-level research has confi rmed this product variety benefi t for fi rms that import intermediate inputs. Using Hungarian data, Halpern, Koren, and Szeidl (2005) intermediate inputs. Using Hungarian data, Halpern, Koren, and Szeidl (2005) show that importing many varieties of foreign inputs increases fi rm productivity show that importing many varieties of foreign inputs increases fi rm productivity by 12 percent. Using Indonesian data, Amiti and Konings (2007) show that a by 12 percent. Using Indonesian data, Amiti and Konings (2007) show that a 10 percentage point fall in input tariffs leads to a productivity gain of 12 percent 10 percentage point fall in input tariffs leads to a productivity gain of 12 percent for fi rms that import their inputs. Kasahara and Rodrigue (2008), Kasahara and for fi rms that import their inputs. Kasahara and Rodrigue (2008), Kasahara and Lapham (2012), Topalova and Khandelwal (2011), and Goldberg, Khandelwal, Lapham (2012), Topalova and Khandelwal (2011), and Goldberg, Khandelwal, Pavcnik, and Topalova (2010) show similarly large gains for Chile and India. In the Pavcnik, and Topalova (2010) show similarly large gains for Chile and India. In the context of the Canada–U.S. Free Trade Agreement, Lileeva and Trefl er (2010) fi nd context of the Canada–U.S. Free Trade Agreement, Lileeva and Trefl er (2010) fi nd that the fall in Canadian tariffs on inputs that Canadian fi rms purchased from the that the fall in Canadian tariffs on inputs that Canadian fi rms purchased from the United States resulted in a 0.5 percent rise in Canadian manufacturing productivity. United States resulted in a 0.5 percent rise in Canadian manufacturing productivity. The Canadian impacts are not nearly as large as effects from developing countries, The Canadian impacts are not nearly as large as effects from developing countries, which suggests that access to a variety of inputs is an important ingredient in the which suggests that access to a variety of inputs is an important ingredient in the process of economic development.process of economic development.

More variety means more competition, and more competition forces fi rms More variety means more competition, and more competition forces fi rms to lower their markups and prices. We see evidence of this after the Turkish and to lower their markups and prices. We see evidence of this after the Turkish and Cote d’Ivoire trade liberalizations of 1985 (Levinsohn 1993; Harrison 1994) and in Cote d’Ivoire trade liberalizations of 1985 (Levinsohn 1993; Harrison 1994) and in Belgium during the 1994–2004 period of increased integration (Abraham, Konings, Belgium during the 1994–2004 period of increased integration (Abraham, Konings, Vanormelingen 2009; De Loecker 2011). On the other hand, there was no evidence Vanormelingen 2009; De Loecker 2011). On the other hand, there was no evidence of falling markups in Mexico after the trade liberalization of the early 1980s (Tybout of falling markups in Mexico after the trade liberalization of the early 1980s (Tybout and Westbrook 1995). We turn next to the gains associated with reallocation of and Westbrook 1995). We turn next to the gains associated with reallocation of resources across heterogeneous fi rms within an industry.resources across heterogeneous fi rms within an industry.

Gains from Reallocation at the Firm Level

By the mid-1980s, a large body of theoretical work demonstrated that By the mid-1980s, a large body of theoretical work demonstrated that freer trade could affect productivity by forcing fi rms to move up or down their freer trade could affect productivity by forcing fi rms to move up or down their average cost curves. Much of the follow-on empirical work assumed that fi rms average cost curves. Much of the follow-on empirical work assumed that fi rms

98 Journal of Economic Perspectives

were identical, and it made a variety of assumptions that allowed inferences to be were identical, and it made a variety of assumptions that allowed inferences to be drawn from industry-level data (for example, Harris 1984). We now know that the drawn from industry-level data (for example, Harris 1984). We now know that the heterogeneity of fi rms even within narrowly defi ned industries is a central feature heterogeneity of fi rms even within narrowly defi ned industries is a central feature of the data that cannot be ignored (for example, Bernard, Jensen, Redding, and of the data that cannot be ignored (for example, Bernard, Jensen, Redding, and Schott 2007).Schott 2007).

Our second source of gains from trade is the result of shifting resources away Our second source of gains from trade is the result of shifting resources away from less-productive fi rms and towards more-productive fi rms. To analyze gains from less-productive fi rms and towards more-productive fi rms. To analyze gains from reallocation of trade between fi rms, we need a model of trade with hetero-from reallocation of trade between fi rms, we need a model of trade with hetero- geneous fi rms—that is, in which performance varies across different fi rms. We can geneous fi rms—that is, in which performance varies across different fi rms. We can then capture how fi rms with different characteristics respond differently to trade. then capture how fi rms with different characteristics respond differently to trade. Consider the case from the previous example where opening to trade leads to a Consider the case from the previous example where opening to trade leads to a transition from production plan A in which each country produces two varieties to transition from production plan A in which each country produces two varieties to production plan B in which each country produces one variety. In the real world, production plan B in which each country produces one variety. In the real world, those varieties are associated with the fi rms that produce them. Opening up to trade those varieties are associated with the fi rms that produce them. Opening up to trade therefore implies that one of the two fi rms in each country shuts down, while the therefore implies that one of the two fi rms in each country shuts down, while the remaining fi rm expands production from 2 units to 8 units. But what are the factors remaining fi rm expands production from 2 units to 8 units. But what are the factors explaining which fi rms expand and which ones exit?explaining which fi rms expand and which ones exit?

Monopolistic Competitors with Heterogeneous Costs Melitz and Ottaviano (2008) develop a model of trade that allows for differ-Melitz and Ottaviano (2008) develop a model of trade that allows for differ-

ences across fi rms; we use a simplifi ed version of that model for the discussion here. ences across fi rms; we use a simplifi ed version of that model for the discussion here. Consider a monopolistically competitive industry in which many fi rms compete by Consider a monopolistically competitive industry in which many fi rms compete by offering different products that are relatively close substitutes for one another—at offering different products that are relatively close substitutes for one another—at least as compared to products in other industries. For simplicity, we assume that least as compared to products in other industries. For simplicity, we assume that each fi rm produces a single product, that demand for all products is symmetric, each fi rm produces a single product, that demand for all products is symmetric, and that fi rms differ only with respect to productivity. Specifi cally, fi rms differ and that fi rms differ only with respect to productivity. Specifi cally, fi rms differ only with respect to their marginal costs of production only with respect to their marginal costs of production cii , where , where i indexes fi rms. indexes fi rms. (A number of authors have developed related models that allow fi rms to produce (A number of authors have developed related models that allow fi rms to produce multiple products: for example, Eckel and Neary 2010; Bernard, Redding, and multiple products: for example, Eckel and Neary 2010; Bernard, Redding, and Schott 2011; and Mayer, Melitz, and Ottaviano 2011. Also, demand need not be Schott 2011; and Mayer, Melitz, and Ottaviano 2011. Also, demand need not be symmetric: there can be product-quality differences across fi rms. Such product-symmetric: there can be product-quality differences across fi rms. Such product- quality differences lead to very similar predictions for fi rm performance as the quality differences lead to very similar predictions for fi rm performance as the ones we now discuss for cost differences.)ones we now discuss for cost differences.)

Panel A of Figure 2 illustrates the price and quantity choices for two monopolisti-Panel A of Figure 2 illustrates the price and quantity choices for two monopolisti- cally competitive fi rms. Both fi rms face the same downward-sloping cally competitive fi rms. Both fi rms face the same downward-sloping residual demand demand curve: residual demand is demand as perceived by the fi rm, and thus depends on curve: residual demand is demand as perceived by the fi rm, and thus depends on the behavior of other competing fi rms in the market.the behavior of other competing fi rms in the market.11 On the production side, On the production side,

1 The equation for the demand facing a fi rm that is used in what follows is Q = S[(1/n) − b( _ p − p)],

where Q is the quantity of output demanded, S is the total output of the industry, n is the number of fi rms in the industry, b > 0 is a constant term representing the responsiveness of a fi rm’s sales to its price, p is the price charged by the fi rm itself, and

_ p is the average price charged by its competitors. This demand

equation may be given the following intuitive justifi cation: If all fi rms charge the same price, each will have a market share 1/n. A fi rm charging more than the average of other fi rms will have a smaller market share, whereas a fi rm charging less will have a larger share.

Gains from Trade when Firms Matter 99

Figure 2 Performance Differences across Firms

Source: Authors.

A: Cost, Price

B: Profit

C os

t, Pr

ic e

Pr ofi

t

Quantity

MC2

MC1

DMR q1q2

c2

p2

c1

p1

c *

Operating profit

Net profit

c1 c2 c

cmaxc *

Marginal Cost (MC)

– f

π1 o

π2 o

100 Journal of Economic Perspectives

marginal costs for fi rm 1 are shown as lower than those for fi rm 2. In panel A, fi rm 1 marginal costs for fi rm 1 are shown as lower than those for fi rm 2. In panel A, fi rm 1 has a lower marginal cost (has a lower marginal cost (c11) than fi rm 2 () than fi rm 2 (c22). We also assume that economies of ). We also assume that economies of scale exist because of a fi xed cost that a fi rm must incur to develop a product and scale exist because of a fi xed cost that a fi rm must incur to develop a product and set up its initial production.set up its initial production.

In this setting, each fi rm maximizes profi t by choosing an output level In this setting, each fi rm maximizes profi t by choosing an output level q that that equalizes marginal cost and marginal revenue. Firm 1 chooses a higher output level equalizes marginal cost and marginal revenue. Firm 1 chooses a higher output level than fi rm 2 (than fi rm 2 (q11 >> q22), associated with a lower price (), associated with a lower price (p11 << p22). Firm 1 also sets a ). Firm 1 also sets a higher markup than fi rm 2: higher markup than fi rm 2: p11 − − c11 >> p22 − − c22 ; this is a consequence of the marginal ; this is a consequence of the marginal revenue curve being steeper than the demand curve. Thus, fi rm 1 earns a higher revenue curve being steeper than the demand curve. Thus, fi rm 1 earns a higher operating profi t than fi rm 2: operating profi t than fi rm 2: ππ 11

o >> ππ 22 o , as represented by the shaded areas in panel A , as represented by the shaded areas in panel A

of Figure 2. We assume that all fi rms face the same set-up cost of Figure 2. We assume that all fi rms face the same set-up cost f, so fi rm 1 also earns , so fi rm 1 also earns higher net profi ts (subtracting the fi xed cost higher net profi ts (subtracting the fi xed cost f for all fi rms). Of course, differences in for all fi rms). Of course, differences in fi xed costs would not affect marginal costs and thus would not affect fi rm decisions fi xed costs would not affect marginal costs and thus would not affect fi rm decisions concerning price and output. We can thus summarize the relevant performance concerning price and output. We can thus summarize the relevant performance differences that result from marginal cost differences across fi rms in the following differences that result from marginal cost differences across fi rms in the following way. Compared to a fi rm with higher marginal cost, a fi rm with a lower marginal way. Compared to a fi rm with higher marginal cost, a fi rm with a lower marginal cost will: 1) set a lower price but at a higher markup over marginal cost, 2) produce cost will: 1) set a lower price but at a higher markup over marginal cost, 2) produce more output, and 3) earn higher profi ts.more output, and 3) earn higher profi ts.

Panel B in Figure 2 shows how fi rm profi t varies with its marginal cost Panel B in Figure 2 shows how fi rm profi t varies with its marginal cost cii . Both . Both operating and net profi t will be decreasing functions of marginal cost, while the operating and net profi t will be decreasing functions of marginal cost, while the difference between the two is the fi xed set-up cost difference between the two is the fi xed set-up cost f. Going back to panel A, we see . Going back to panel A, we see that a fi rm can earn a positive operating profi t so long as its marginal cost is below that a fi rm can earn a positive operating profi t so long as its marginal cost is below the intercept of the demand curve on the vertical axis. Let the intercept of the demand curve on the vertical axis. Let c** denote this cost cutoff. denote this cost cutoff. A fi rm with a marginal cost A fi rm with a marginal cost cii above this cutoff is effectively “priced out” of the above this cutoff is effectively “priced out” of the market and would earn negative operating profi ts if it were to produce any output market and would earn negative operating profi ts if it were to produce any output (represented by the dotted segment for operating profi t in panel A). Such a fi rm (represented by the dotted segment for operating profi t in panel A). Such a fi rm would choose to shut down and not produce (earning zero operating profi t but would choose to shut down and not produce (earning zero operating profi t but incurring a net profi t loss –incurring a net profi t loss –f due to the fi xed cost). Why would such a fi rm enter due to the fi xed cost). Why would such a fi rm enter in the fi rst place? Clearly, it would not if it knew about its high cost in the fi rst place? Clearly, it would not if it knew about its high cost cii prior both to prior both to entry and to paying the fi xed cost entry and to paying the fi xed cost f..

We assume that entrants face some randomness about their future production We assume that entrants face some randomness about their future production cost cost cii . This randomness disappears only . This randomness disappears only after the set-up cost the set-up cost f is paid and is sunk. is paid and is sunk. Thus some fi rms will regret their entry decision, as their net profi t is negative (they Thus some fi rms will regret their entry decision, as their net profi t is negative (they cannot recover the sunk cost cannot recover the sunk cost f ). This is the case for fi rm 2 in panel B; even though ). This is the case for fi rm 2 in panel B; even though its operating profi t is positive, it does not cover the sunk cost its operating profi t is positive, it does not cover the sunk cost f. On the other hand, . On the other hand, some fi rms discover that their production cost some fi rms discover that their production cost cii is very low and earn a high (and is very low and earn a high (and positive) net profi t.positive) net profi t.

Firms consider all these possible outcomes, captured by the net profi t curve Firms consider all these possible outcomes, captured by the net profi t curve in panel B when they make their entry decision. Firms anticipate that there is a in panel B when they make their entry decision. Firms anticipate that there is a range of lower costs where net profi ts are positive (shaded area to the left above the range of lower costs where net profi ts are positive (shaded area to the left above the horizontal axis), and another range of higher costs where net profi ts are negative horizontal axis), and another range of higher costs where net profi ts are negative (shaded area to the right below the horizontal axis). In the long-run equilibrium, (shaded area to the right below the horizontal axis). In the long-run equilibrium, fi rms enter until their fi rms enter until their expected net profi t across all potential cost levels net profi t across all potential cost levels cii is driven is driven to zero. If every cost level to zero. If every cost level cii from 0 to from 0 to cmaxmax is equally likely, then this equilibrium is is equally likely, then this equilibrium is

Marc J. Melitz and Daniel Trefl er 101

reached when the two shaded areas are equal.reached when the two shaded areas are equal.22 Panel B of Figure 2 summarizes the Panel B of Figure 2 summarizes the industry equilibrium for a given market size. It shows which range of fi rms survive industry equilibrium for a given market size. It shows which range of fi rms survive and produce (with cost and produce (with cost cii below below c**) and how their profi ts will vary with their cost ) and how their profi ts will vary with their cost levels levels cii . .

What Changes When Economies Integrate? How will the situation faced by these heterogeneous fi rms alter when econo-How will the situation faced by these heterogeneous fi rms alter when econo-

mies integrate into a single larger market? A larger market can support a larger mies integrate into a single larger market? A larger market can support a larger number of fi rms than a smaller market, which implies more competition in the number of fi rms than a smaller market, which implies more competition in the larger market. Increased competition—absent any increase in market size—leads to larger market. Increased competition—absent any increase in market size—leads to an inward shift of each fi rm’s residual demand curve. On the other hand, holding an inward shift of each fi rm’s residual demand curve. On the other hand, holding competition fi xed, a larger market rotates out the residual demand curves for competition fi xed, a larger market rotates out the residual demand curves for all fi rms. Putting these two effects of increased competition and greater market all fi rms. Putting these two effects of increased competition and greater market size together gives us the combined effect of international trade on the residual size together gives us the combined effect of international trade on the residual demand curve perceived by fi rms. This change is depicted in panel A of Figure 3, demand curve perceived by fi rms. This change is depicted in panel A of Figure 3, as the shift from demand curve as the shift from demand curve D to to D ′′. The residual demand curve shifts in from . The residual demand curve shifts in from the perspective of the smaller fi rms with lower output levels that operate on the the perspective of the smaller fi rms with lower output levels that operate on the higher part of the demand curve: here, the effect of tougher competition domi-higher part of the demand curve: here, the effect of tougher competition domi- nates. However, from the perspective of the larger fi rms that operate on the lower nates. However, from the perspective of the larger fi rms that operate on the lower part of the demand curve, the residual demand curve has shifted out: here, the part of the demand curve, the residual demand curve has shifted out: here, the effect of the larger market size dominates.effect of the larger market size dominates.

Panel B of Figure 3 shows the consequences of this demand change for the Panel B of Figure 3 shows the consequences of this demand change for the operating profi ts of fi rms with different cost levels operating profi ts of fi rms with different cost levels cii . The decrease in demand for . The decrease in demand for the smaller fi rms translates into a new lower cost cutoff the smaller fi rms translates into a new lower cost cutoff c**′′: Firms with the highest : Firms with the highest cost levels (above cost levels (above c**′ ′ ) cannot survive the decrease in demand and are forced to ) cannot survive the decrease in demand and are forced to exit. On the other hand, the fl atter demand curve is advantageous to fi rms with exit. On the other hand, the fl atter demand curve is advantageous to fi rms with the lowest cost levels: they can adapt to the increased competition by lowering the lowest cost levels: they can adapt to the increased competition by lowering their markup (and hence their price) and gaining some additional market share. their markup (and hence their price) and gaining some additional market share. (Recall that the high-cost fi rms are already setting low markups, and cannot lower (Recall that the high-cost fi rms are already setting low markups, and cannot lower their prices to induce positive demand, as this would mean pricing below their their prices to induce positive demand, as this would mean pricing below their marginal cost of production.) Thus, the best-performing fi rms with the lowest cost marginal cost of production.) Thus, the best-performing fi rms with the lowest cost levels levels cii now earn increased operating and net profi ts. Panel B of Figure 3 illustrates now earn increased operating and net profi ts. Panel B of Figure 3 illustrates how increased market size generates both winners and losers amongst fi rms in an how increased market size generates both winners and losers amongst fi rms in an industry. Low-cost fi rms thrive and increase their profi ts and market shares, high-industry. Low-cost fi rms thrive and increase their profi ts and market shares, high- cost fi rms contract, and the highest cost fi rms exit.cost fi rms contract, and the highest cost fi rms exit.

In this model, economic integration through market expansion does not In this model, economic integration through market expansion does not directly affect fi rm productivity. Nevertheless, it generates an overall increase in directly affect fi rm productivity. Nevertheless, it generates an overall increase in aggregate productivity as market shares are reallocated from the low-productivity aggregate productivity as market shares are reallocated from the low-productivity fi rms with high marginal costs to the high-productivity ones with low marginal costs.fi rms with high marginal costs to the high-productivity ones with low marginal costs.

2 In contrast, when there is no uncertainty about marginal cost because all fi rms share the same cost c, then entry drives the realized net profi t to zero for all fi rms. With fi rm heterogeneity, expected net profi t is zero, but realized profi ts will vary as shown in panel B of Figure 2.

102 Journal of Economic Perspectives

Figure 3 Winners and Losers from Market Integration

Source: Authors.

A: Shift in a firm’s residual curve with international trade

B: Shift in operating profit with international trade

Pr ic

e Pr

ofi t

Quantity

D

D ′

Combined effect of bigger market size and more competition

More competition (market size fixed)

Winners Losers

Exit

Marginal Cost (MC)

c *

c *′

Operating profit

Gains from Trade when Firms Matter 103

Trade Costs, Export Decisions, and Trade Liberalization The discussion to this point has implicitly modeled economic integration as The discussion to this point has implicitly modeled economic integration as

a change in market size from a closed economy with no trade all the way to a a change in market size from a closed economy with no trade all the way to a single combined market with no trade barriers. In reality, initial trade costs are single combined market with no trade barriers. In reality, initial trade costs are rarely so high as to block all trade prior to liberalization, and liberalization reduces rarely so high as to block all trade prior to liberalization, and liberalization reduces trading costs without fully eliminating them. In a number of ways, this kind of trading costs without fully eliminating them. In a number of ways, this kind of partial trade liberalization has a very similar effect to the simpler case of full inte-partial trade liberalization has a very similar effect to the simpler case of full inte- gration. With partial trade liberalization, the better-performing fi rms expand, the gration. With partial trade liberalization, the better-performing fi rms expand, the worse-performing ones contract, and the worst performing ones exit. This gener-worse-performing ones contract, and the worst performing ones exit. This gener- ates the same type of reallocation effect previously described and leads to a rise in ates the same type of reallocation effect previously described and leads to a rise in aggregate productivity.aggregate productivity.33

However, adding trade costs also allows us to analyze an additional issue: However, adding trade costs also allows us to analyze an additional issue: whether fi rms choose to export. With trade costs, exporting is profi table only for whether fi rms choose to export. With trade costs, exporting is profi table only for a subset of better-performing fi rms. Some fi rms do not export, and instead only a subset of better-performing fi rms. Some fi rms do not export, and instead only serve domestic consumers. We now extend our theoretical model to incorporate serve domestic consumers. We now extend our theoretical model to incorporate trade costs and fi rms’ export decisions. For this purpose, we can no longer analyze a trade costs and fi rms’ export decisions. For this purpose, we can no longer analyze a single market: instead, we need to look at fi rms’ decisions in both the domestic and single market: instead, we need to look at fi rms’ decisions in both the domestic and export markets jointly. For simplicity, we consider a special case where both coun-export markets jointly. For simplicity, we consider a special case where both coun- tries are symmetric, so that demand conditions in both the domestic and export tries are symmetric, so that demand conditions in both the domestic and export markets will be identical.markets will be identical.

Assume that a fi rm must incur an additional trade cost Assume that a fi rm must incur an additional trade cost t for each unit of output for each unit of output that it sells to customers across the border. As a result of this trade cost, each fi rm that it sells to customers across the border. As a result of this trade cost, each fi rm will set a different price in its export market relative to its domestic market, which will set a different price in its export market relative to its domestic market, which will lead to different quantities sold in each market, and ultimately to different profi t will lead to different quantities sold in each market, and ultimately to different profi t levels earned in each market. Because we are assuming that each fi rm’s marginal levels earned in each market. Because we are assuming that each fi rm’s marginal cost is constant and does not vary with production levels, the decisions regarding cost is constant and does not vary with production levels, the decisions regarding pricing and quantity sold in each market can be separated: a decision regarding the pricing and quantity sold in each market can be separated: a decision regarding the domestic market will have no effect on the profi tability of different decisions for domestic market will have no effect on the profi tability of different decisions for the export market.the export market.

Consider the case of fi rms located in Home. Their situation regarding their Consider the case of fi rms located in Home. Their situation regarding their domestic (Home) market is exactly as was illustrated in Figure 2, except that all the domestic (Home) market is exactly as was illustrated in Figure 2, except that all the outcomes such as price, output, and profi t relate to the domestic market only. Now outcomes such as price, output, and profi t relate to the domestic market only. Now consider the export (Foreign) market. The fi rms face the same demand curve in consider the export (Foreign) market. The fi rms face the same demand curve in Foreign as they do in Home (the two countries are identical). The only difference Foreign as they do in Home (the two countries are identical). The only difference is that each fi rm’s marginal cost in the export market is shifted up by the trade is that each fi rm’s marginal cost in the export market is shifted up by the trade cost cost t. What are the effects of the trade cost on the fi rms’ decisions regarding the . What are the effects of the trade cost on the fi rms’ decisions regarding the export market? A higher marginal cost induces a fi rm to raise its price, which leads export market? A higher marginal cost induces a fi rm to raise its price, which leads to a lower output quantity sold and to lower profi ts (as highlighted in Figure 2). to a lower output quantity sold and to lower profi ts (as highlighted in Figure 2).

3 The more general version of this model analyzed by Melitz and Ottaviano (2008) allows for multiple countries of different sizes and for arbitrary trade costs between any country pair (though the trade costs are proportional to production costs instead of per output unit as in the current version). That paper shows more formally that the effects of multilateral liberalization (all countries proportionally reduce trade costs) are very similar to the case of full economic integration that leads to a single larger market.

104 Journal of Economic Perspectives

We also know that if marginal cost is raised above a threshold level We also know that if marginal cost is raised above a threshold level c**, then a fi rm , then a fi rm cannot profi tably operate in that market. Thus, when there are trade costs, some cannot profi tably operate in that market. Thus, when there are trade costs, some fi rms will fi nd it profi table to operate in the domestic market but not in the export fi rms will fi nd it profi table to operate in the domestic market but not in the export market because the trade cost pushes their marginal cost for that market above the market because the trade cost pushes their marginal cost for that market above the threshold threshold c**..

Figure 4 helps to visualize the production and export decisions for all fi rms Figure 4 helps to visualize the production and export decisions for all fi rms based on their marginal cost based on their marginal cost cii. Panel A of Figure 4 separates a fi rm’s operating profi t . Panel A of Figure 4 separates a fi rm’s operating profi t into a portion earned from domestic sales, and a portion earned from export sales. into a portion earned from domestic sales, and a portion earned from export sales. (Both portions are functions of a fi rm’s marginal cost (Both portions are functions of a fi rm’s marginal cost cii as in Figure 2.) Because the as in Figure 2.) Because the only difference between the domestic and export markets is the additional per-unit only difference between the domestic and export markets is the additional per-unit trade cost trade cost t,, the horizontal distance between the two curves is equal to the trade the horizontal distance between the two curves is equal to the trade cost cost t. Firm 1 earns positive operating profi ts from sales in both the domestic and . Firm 1 earns positive operating profi ts from sales in both the domestic and export markets: it will export and reach consumers in both markets. This will be export markets: it will export and reach consumers in both markets. This will be the case for all fi rms with cost below the case for all fi rms with cost below c** – – t. On the other hand, fi rm 2 only earns . On the other hand, fi rm 2 only earns positive operating profi ts from sales in the domestic market—and thus chooses not positive operating profi ts from sales in the domestic market—and thus chooses not to export. Any fi rm with cost above to export. Any fi rm with cost above c** – – t will be in this sa me situation and therefore will be in this sa me situation and therefore will not export: those fi rms only serve their domestic market. As before, the worst will not export: those fi rms only serve their domestic market. As before, the worst performing fi rms with cost above performing fi rms with cost above c** cannot profi tably operate at all (even in their cannot profi tably operate at all (even in their domestic market) and therefore exit.domestic market) and therefore exit.

Panel B of Figure 4 summarizes the effects of trade liberalization—a reduction Panel B of Figure 4 summarizes the effects of trade liberalization—a reduction in the trade cost in the trade cost t—for those fi rm decisions. The fi gure shows the same two operating for those fi rm decisions. The fi gure shows the same two operating profi t curves from panel A both before and after (dashed curves) trade liberaliza-profi t curves from panel A both before and after (dashed curves) trade liberaliza- tion. The operating profi t for the domestic market shifts down due to the increase tion. The operating profi t for the domestic market shifts down due to the increase in competition (which shifts the residual demand curve for the domestic market in competition (which shifts the residual demand curve for the domestic market inward as explained earlier). Some of the higher-cost fi rms that used to produce for inward as explained earlier). Some of the higher-cost fi rms that used to produce for domestic consumption no longer earn a positive operating profi t after trade liberal-domestic consumption no longer earn a positive operating profi t after trade liberal- ization and exit. On the other hand, the operating profi t for the export market shifts ization and exit. On the other hand, the operating profi t for the export market shifts up due to the lower trade cost. (Increased competition in the export market tends to up due to the lower trade cost. (Increased competition in the export market tends to reduce operating profi ts there, but this effect is dominated by the direct effect of the reduce operating profi ts there, but this effect is dominated by the direct effect of the trade cost reduction.) A key empirical prediction is that some fi rms start exporting. trade cost reduction.) A key empirical prediction is that some fi rms start exporting. Specifi cally, among the fi rms that did not export prior to trade liberalization, only Specifi cally, among the fi rms that did not export prior to trade liberalization, only the most productive of these start exporting.the most productive of these start exporting.

Evidence on Gains from Inter-Firm Reallocations In many ways, the Canada–U.S. Free Trade Agreement is a useful natural In many ways, the Canada–U.S. Free Trade Agreement is a useful natural

experiment for considering the effects of trade integration. The policy experiment experiment for considering the effects of trade integration. The policy experiment is clearly defi ned: it dealt only with market integration and was not part of a larger is clearly defi ned: it dealt only with market integration and was not part of a larger package of macroeconomic reforms that often accompany trade liberalization. The package of macroeconomic reforms that often accompany trade liberalization. The enactment of the agreement was largely unanticipated: a Canadian general election enactment of the agreement was largely unanticipated: a Canadian general election was fought on the issue one month before the agreement was to be signed into law was fought on the issue one month before the agreement was to be signed into law and pollsters unanimously predicted that Canada’s ruling party—along with the and pollsters unanimously predicted that Canada’s ruling party—along with the free trade agreement—would be defeated (Brander 1991; Thompson 1993). Thus, free trade agreement—would be defeated (Brander 1991; Thompson 1993). Thus, evidence about the extent of aggregate productivity changes as a result of realloca-evidence about the extent of aggregate productivity changes as a result of realloca- tions among heterogeneous fi rms can be sought by looking at the distribution of tions among heterogeneous fi rms can be sought by looking at the distribution of

Marc J. Melitz and Daniel Trefl er 105

Figure 4 Export Decision and Trade Liberalization

Source: Authors.

A: Operating profits from domestic and export sales

B: Effects of trade liberalization on firm decisions

Operating profit (domestic)

ExitNonexportersExporters

c2c1

c *

c

Operating profit (export)

Marginal Cost (MC)

Marginal Cost (MC)

Operating profit (domestic)

Operating profit (export)

New exit

New exporters

c

c *

c * – t

Pr ic

e Pr

ofi t

106 Journal of Economic Perspectives

productivity across Canadian manufacturing plants before and after the agreement, productivity across Canadian manufacturing plants before and after the agreement, at entrants before and after the agreement, and at the productivity distribution of at entrants before and after the agreement, and at the productivity distribution of exporters and nonexporters.exporters and nonexporters.

The agreement came into effect on January 1, 1989. Panel A of Figure 5 shows The agreement came into effect on January 1, 1989. Panel A of Figure 5 shows the distribution of labor productivity as measured by value-added per employee the distribution of labor productivity as measured by value-added per employee across Canadian manufacturing plants both before the agreement in 1988 and in across Canadian manufacturing plants both before the agreement in 1988 and in 1996, when there had been time for fi rm adjustments to occur. For example, the 1996, when there had been time for fi rm adjustments to occur. For example, the 1996 curve summarizes the productivity distribution of all 35,000 Canadian manu-1996 curve summarizes the productivity distribution of all 35,000 Canadian manu- facturing plants in that year. Clearly, the distribution of fi rms shifted rightward: facturing plants in that year. Clearly, the distribution of fi rms shifted rightward: between 1988 and 1996, the share of low-productivity plants in manufacturing between 1988 and 1996, the share of low-productivity plants in manufacturing declined and the share of high-productivity plants rose.declined and the share of high-productivity plants rose.

The horizontal axis is based on a measure of the log of labor productivity. The horizontal axis is based on a measure of the log of labor productivity. However, to ensure that dispersion is driven by within-industry rather than between-However, to ensure that dispersion is driven by within-industry rather than between- industry differences in labor productivity, we scale each plant’s log productivity by industry differences in labor productivity, we scale each plant’s log productivity by subtracting from it the median log productivity of the plant’s four-digit SIC industry. subtracting from it the median log productivity of the plant’s four-digit SIC industry. Thus, the median plant in each industry has a score of zero on the horizontal axis. Thus, the median plant in each industry has a score of zero on the horizontal axis. The vertical axis shows the share of plants with the indicated level of productivity. The vertical axis shows the share of plants with the indicated level of productivity. These frequencies are weighted by plant employment; otherwise, tiny plants that These frequencies are weighted by plant employment; otherwise, tiny plants that account for only a tiny fraction of total employment would dominate the fi gure.account for only a tiny fraction of total employment would dominate the fi gure.

To get a sense of the degree of productivity dispersion, consider the horizontal To get a sense of the degree of productivity dispersion, consider the horizontal axis of Figure 5 and suppose that log productivity at plant A is one unit higher than at axis of Figure 5 and suppose that log productivity at plant A is one unit higher than at plant B. This is equivalent to saying that plant A is three times more productive than plant B. This is equivalent to saying that plant A is three times more productive than plant B. If A is four units higher than B, then A is 50 times more productive than B.plant B. If A is four units higher than B, then A is 50 times more productive than B.44

Obviously, labor productivity as shown in Figure 5 is not an identical concept Obviously, labor productivity as shown in Figure 5 is not an identical concept to the horizontal lines showing levels of marginal cost in the theoretical discussion. to the horizontal lines showing levels of marginal cost in the theoretical discussion. When marginal costs are low, we typically expect productivity to be high. Therefore, When marginal costs are low, we typically expect productivity to be high. Therefore, the inverse of marginal costs (1/the inverse of marginal costs (1/c) is often proxied in empirical work by productivity.) is often proxied in empirical work by productivity.

The productivity heterogeneity shown for Canadian manufacturing fi rms in The productivity heterogeneity shown for Canadian manufacturing fi rms in Figure 5 is a pervasive feature of all economies including, for example, the United Figure 5 is a pervasive feature of all economies including, for example, the United States (Bernard, Eaton, Jensen, and Kortum 2003), many European economies States (Bernard, Eaton, Jensen, and Kortum 2003), many European economies (Mayer and Ottaviano 2008; Bartelsman, Hatiwanger, and Scarpetta 2009), as (Mayer and Ottaviano 2008; Bartelsman, Hatiwanger, and Scarpetta 2009), as well as China and India (Hsieh and Klenow 2009). Wagner (2007) surveys related well as China and India (Hsieh and Klenow 2009). Wagner (2007) surveys related studies from countries all around the world, reporting similar patterns regarding studies from countries all around the world, reporting similar patterns regarding widespread fi rm heterogeneity within industries. Thus, the lessons derived from this widespread fi rm heterogeneity within industries. Thus, the lessons derived from this example are not specifi c to the Canadian experience.example are not specifi c to the Canadian experience.

What caused the change from 1988 to 1996 in the productivity distribution of What caused the change from 1988 to 1996 in the productivity distribution of Canadian manufacturing fi rms? It was largely due to the reallocation mechanisms Canadian manufacturing fi rms? It was largely due to the reallocation mechanisms across plants described above. The fi rst of these mechanisms that we examine is across plants described above. The fi rst of these mechanisms that we examine is

4 Let φA and φB be productivities of A and B and suppose that they are 1 unit apart i.e., ln(φA) – ln(φB) = 1. From the property of logs, φA/φB = e

1 = 2.7 ≈ 3. For a difference of 4 units, e4 ≈ 50. On a more technical level, the fi gure is constructed starting with “standardized” log productivities (see the formula in Lileeva 2008, p. 369), which we then multiply by a single scale factor to transform the standardized log produc- tivities into log productivities that are directly comparable with log productivities reported in studies from other countries.

Gains from Trade when Firms Matter 107

Figure 5 Distribution of Productivity across Canadian Manufacturing Plants before and after the Canada–U.S. Free Trade Agreement

Source: Authors’ calculations using data from the Canadian Annual Survey of Manufactures. Notes: The horizontal axes are based on a measure of the log of labor productivity. However, to ensure that dispersion is driven by within-industry rather than between-industry differences in labor productivity, we scale each plant’s log productivity by subtracting from it the median log productivity of the plant’s four-digit SIC industry. Thus, the median plant in each industry has a score of zero on the horizontal axis. The vertical axis shows the share of plants with the indicated level of productivity. These frequencies are weighted by plant employment; otherwise, tiny plants that account for only a tiny fraction of total employment would dominate the fi gure.

40%

30%

20%

10%

0%

Pre-FTA: Entrants during 1980–88

FTA: Entrants during 1988–96

–2 –1 0 1 2 3 4 5 6

B: Labor productivity distribution of entering Canadian manufacturing plants 1980–1988 and 1988–1996 (employment weighted)

A: Labor productivity distribution of all Canadian manufacturing plants 1988 and 1996 (employment weighted)

40%

30%

20%

10%

0%

1988

1996

–2 –1 0 1 2 3 4 5 6

–2 –1 0 1 2 3 4 5 6

Exporters, 1996

C: Labor productivity distribution of exporters and nonexporters, 1996 (employment weighted)

Nonexporters, 1996

40%

30%

20%

10%

0%

35%

25%

15%

5%

108 Journal of Economic Perspectives

the fall in the survival threshold of marginal cost—that is, the leftward shift of the fall in the survival threshold of marginal cost—that is, the leftward shift of c** in in panel B of Figure 4. The empirical counterpart to a fall in panel B of Figure 4. The empirical counterpart to a fall in c** is a rise in the break-is a rise in the break- even level of productivity. One can examine this mechanism by looking either at even level of productivity. One can examine this mechanism by looking either at exit rates or at entry rates. Since a plant may not exit until it has completed the exit rates or at entry rates. Since a plant may not exit until it has completed the multiyear process of depreciating its fi xed capital, it is best to look at entry rates, multiyear process of depreciating its fi xed capital, it is best to look at entry rates, which adjust more quickly to shocks such as a free trade agreement. Panel B of which adjust more quickly to shocks such as a free trade agreement. Panel B of Figure 5 displays the productivity levels of new entrants to Canadian manufacturing Figure 5 displays the productivity levels of new entrants to Canadian manufacturing both for the pre-agreement period (1980–1988) and for the period immediately both for the pre-agreement period (1980–1988) and for the period immediately after the agreement came into force (1988–1996). There was a striking decline in after the agreement came into force (1988–1996). There was a striking decline in the entry rates of plants with productivity near or below the median. To use a sports the entry rates of plants with productivity near or below the median. To use a sports analogy, in the earlier period many low-productivity plants made the cut and joined analogy, in the earlier period many low-productivity plants made the cut and joined the team while in the later period a number of such low-productivity plants no the team while in the later period a number of such low-productivity plants no longer made the cut.longer made the cut.55

The pattern here is confi rmed by more formal econometric analysis (specifi -The pattern here is confi rmed by more formal econometric analysis (specifi - cally, binary-outcome regressions of exit on plant characteristics as well as industry cally, binary-outcome regressions of exit on plant characteristics as well as industry and time controls). For example, Baggs (2005) and Baldwin and Gu (2006) show and time controls). For example, Baggs (2005) and Baldwin and Gu (2006) show that the free trade agreement tariff cuts raised exit by a large amount. Lileeva that the free trade agreement tariff cuts raised exit by a large amount. Lileeva (2008) estimates that the free trade agreement tariff cuts raised exit rates by as (2008) estimates that the free trade agreement tariff cuts raised exit rates by as much as 16 percent, with all of the increase involving the exit of nonexporters. much as 16 percent, with all of the increase involving the exit of nonexporters. Bernard, Jensen, and Schott (2006) fi nd similar results for U.S. plants faced with Bernard, Jensen, and Schott (2006) fi nd similar results for U.S. plants faced with U.S. tariff reductions.U.S. tariff reductions.

Trade Costs and the Export Decision A central prediction of the theory is that in the presence of trade costs, only A central prediction of the theory is that in the presence of trade costs, only

low-cost, high-productivity fi rms export. Panel C of Figure 5 shows the distribution low-cost, high-productivity fi rms export. Panel C of Figure 5 shows the distribution of Canadian plants separately for exporters and nonexporters. Clearly, the distribu-of Canadian plants separately for exporters and nonexporters. Clearly, the distribu- tion for exporters is to the right of that for nonexporters. On average, Canadian tion for exporters is to the right of that for nonexporters. On average, Canadian exporters are 40 percent more productive than nonexporters in the same industry exporters are 40 percent more productive than nonexporters in the same industry (Baldwin and Gu 2003). Since the seminal work of Bernard and Jensen (1995), a (Baldwin and Gu 2003). Since the seminal work of Bernard and Jensen (1995), a huge body of research covering dozens of countries has found this same pattern of huge body of research covering dozens of countries has found this same pattern of higher productivity for exporters relative to nonexporters.higher productivity for exporters relative to nonexporters.66

5 Bernard and Jensen (1999), Trefl er (2004), Lileeva (2008), and Lileeva and Trefl er (2010) all point out that one must look not just at pre-Free Trade Agreement levels (as in Figure 5), but also at pre-FTA trends. All of the FTA results reported here hold with pre-FTA controls for both levels and trends. For example, variants of some of the panels in Figure 5 with pre-FTA trend controls appear in Lileeva (2008). 6 In examining panel C of Figure 5, a critical reader may wonder why there are so many highly productive nonexporters and whether this contradicts the theory. A simple but prominent example may help to explain. Highly productive auto parts plants often cluster around a giant auto assembly plant—Ford, General Motors, and Honda all have major auto assembly plants near Toronto, Canada, that are surrounded by parts suppliers. These parts suppliers are highly productive, but do not directly export. This pattern is clearly not a challenge to the theory: these highly productive plants are supplying parts that are assembled into the autos that are ultimately exported to the United States: highly productive parts suppliers are “indirect” exporters. Of course, there are surely other factors outside the scope of our model that explain why some very productive fi rms do not export—and conversely—why some low- productivity fi rms export.

Marc J. Melitz and Daniel Trefl er 109

A much more demanding prediction of the theory deals with who will A much more demanding prediction of the theory deals with who will start exporting in response to falling trade costs. Panel B of Figure 4 shows that those who . Panel B of Figure 4 shows that those who start exporting start exporting will be among the most productive of those who never exported be among the most productive of those who never exported before. To test this prediction, Lileeva and Trefl er (2010) examined a sample before. To test this prediction, Lileeva and Trefl er (2010) examined a sample of over 5,000 Canadian manufacturing plants that had never exported prior to of over 5,000 Canadian manufacturing plants that had never exported prior to the Canada–U.S. free trade agreement. A very large percentage of these plants the Canada–U.S. free trade agreement. A very large percentage of these plants (40 percent) started exporting after the agreement came into force on January 1, (40 percent) started exporting after the agreement came into force on January 1, 1989. Lileeva and Trefl er examine whether these plants started exporting because 1989. Lileeva and Trefl er examine whether these plants started exporting because of the U.S. tariff cuts and, more importantly, whether those that started exporting of the U.S. tariff cuts and, more importantly, whether those that started exporting because of the tariff cuts were more productive than nonexporters. To this end, because of the tariff cuts were more productive than nonexporters. To this end, Lileeva and Trefl er divide up their sample into quartiles of the 1988 distribution Lileeva and Trefl er divide up their sample into quartiles of the 1988 distribution of labor productivity (with the quartiles defi ned separately for each industry to net of labor productivity (with the quartiles defi ned separately for each industry to net out industry characteristics). Only 20 percent of the plants in the bottom quar-out industry characteristics). Only 20 percent of the plants in the bottom quar- tile of labor productivity started exporting because of the tariff cuts, compared to tile of labor productivity started exporting because of the tariff cuts, compared to nearly 60 percent of the plants from the top quartile of labor productivity. (These nearly 60 percent of the plants from the top quartile of labor productivity. (These estimates are from a probit regression in which the dependent variable is 1 if the estimates are from a probit regression in which the dependent variable is 1 if the plant started exporting and 0 if the plant remained a nonexporter. The key inde-plant started exporting and 0 if the plant remained a nonexporter. The key inde- pendent variable is a plant-specifi c measure of the change in the U.S. tariff. This pendent variable is a plant-specifi c measure of the change in the U.S. tariff. This measure is described below.) The key conclusion is that, among fi rms that did not measure is described below.) The key conclusion is that, among fi rms that did not export before trade liberalization, the most productive of these were three times export before trade liberalization, the most productive of these were three times more likely to start exporting in response to the U.S. tariff cuts. This is as predicted more likely to start exporting in response to the U.S. tariff cuts. This is as predicted in panel B of Figure 4.in panel B of Figure 4.77

Quantifying the Gains from Trade Due to Reallocation across Heterogeneous Plants In the wake of the Canada–U.S. free trade agreement, Canadian manufacturing In the wake of the Canada–U.S. free trade agreement, Canadian manufacturing

productivity rose sharply. We have shown that part of this productivity gain was due productivity rose sharply. We have shown that part of this productivity gain was due to the reallocation mechanisms highlighted by the theory. But how important were to the reallocation mechanisms highlighted by the theory. But how important were these in quantitative terms for productivity growth and overall welfare?these in quantitative terms for productivity growth and overall welfare?

In the conventional approach to estimating the gains from trade, the focus In the conventional approach to estimating the gains from trade, the focus is on measuring welfare, or more specifi cally, on the income a society would be is on measuring welfare, or more specifi cally, on the income a society would be willing to pay for lower tariffs. These “compensating variation” gains are typically willing to pay for lower tariffs. These “compensating variation” gains are typically derived from models that 1) make a large number of parametric assumptions derived from models that 1) make a large number of parametric assumptions (assuming very specifi c functional forms for preferences that determine the (assuming very specifi c functional forms for preferences that determine the extent of product differentiation, as well as for the utility derived from love-extent of product differentiation, as well as for the utility derived from love- of-variety), and 2) make use of parameter estimates about which we are highly of-variety), and 2) make use of parameter estimates about which we are highly uncertain. In short, a lot of uncertainty surrounds welfare-gain estimates. In the uncertain. In short, a lot of uncertainty surrounds welfare-gain estimates. In the heterogeneous-fi rms literature, the focus has shifted to estimating productivity heterogeneous-fi rms literature, the focus has shifted to estimating productivity gains rather than welfare. The last two decades have seen major improvements gains rather than welfare. The last two decades have seen major improvements in our ability to estimate productivity gains, both because of the creation of in our ability to estimate productivity gains, both because of the creation of

7 On a related note, profi ts play a key role in all the mechanisms of our model. Baggs and Brander (2006) confi rm that profi ts move in the expected directions. In particular, they fi nd that falling Canadian tariffs are associated with declining Canadian profi ts, especially for import-competing fi rms, while falling U.S. tariffs are associated with increasing Canadian profi ts, especially for export-oriented fi rms.

110 Journal of Economic Perspectives

high-quality, fi rm-level longitudinal data and because of methodological develop-high-quality, fi rm-level longitudinal data and because of methodological develop- ments aimed at exploiting these data. Thus, although productivity gains are not ments aimed at exploiting these data. Thus, although productivity gains are not the same as welfare gains, we have much greater confi dence in our estimates of the same as welfare gains, we have much greater confi dence in our estimates of the productivity gains associated with freer trade.the productivity gains associated with freer trade.

The productivity gains associated with the reallocation of market shares across The productivity gains associated with the reallocation of market shares across fi rms following the Canada–U.S. free trade agreement are usefully broken into fi rms following the Canada–U.S. free trade agreement are usefully broken into two components. First, two components. First, the fall in the U.S. tariffs allowed Canadian plants to export allowed Canadian plants to export more. This shifted the composition of output towards high-productivity exporters more. This shifted the composition of output towards high-productivity exporters and away from low-productivity nonexporters. Lileeva and Trefl er (2010) estimate and away from low-productivity nonexporters. Lileeva and Trefl er (2010) estimate that the fall in U.S. tariffs causally raised Canadian manufacturing productivity by that the fall in U.S. tariffs causally raised Canadian manufacturing productivity by 4.1 percent via this export-composition channel. Second, 4.1 percent via this export-composition channel. Second, the fall in the Canadian tariffs led to a shift in domestic market shares—exporters gained market share at led to a shift in domestic market shares—exporters gained market share at the expense of nonexporters. In the extreme, many nonexporters simply went out the expense of nonexporters. In the extreme, many nonexporters simply went out of business. Trefl er (2004) calculates that this selection effect increased overall of business. Trefl er (2004) calculates that this selection effect increased overall Canadian manufacturing productivity by 4.3 percent.Canadian manufacturing productivity by 4.3 percent.88

Putting these numbers together, we see that the reallocation and selection Putting these numbers together, we see that the reallocation and selection effects induced by the free trade agreement generated a productivity increase of effects induced by the free trade agreement generated a productivity increase of 8.4 percent (4.1 8.4 percent (4.1 ++ 4.3) for Canadian manufacturing. This represents a massive 4.3) for Canadian manufacturing. This represents a massive productivity increase in just a short time—especially when one considers that this productivity increase in just a short time—especially when one considers that this productivity gain did not come from productivity improvements at the plant level: it productivity gain did not come from productivity improvements at the plant level: it only came from the shifting of market shares from less- to more-productive plants. only came from the shifting of market shares from less- to more-productive plants.

Canada is not the only country to have experienced such a substantial produc-Canada is not the only country to have experienced such a substantial produc- tivity boost from reallocations driven by trade liberalization. Bernard and Jensen tivity boost from reallocations driven by trade liberalization. Bernard and Jensen (2004) fi nd that almost half of all U.S. manufacturing productivity growth during (2004) fi nd that almost half of all U.S. manufacturing productivity growth during 1983–1992 is explained by the reallocation of resources towards exporters. Pavcnik 1983–1992 is explained by the reallocation of resources towards exporters. Pavcnik (2002) studies the response of the Chilean manufacturing sector to a massive trade (2002) studies the response of the Chilean manufacturing sector to a massive trade liberalization episode that took place from 1979 to 1986. She fi nds that two-thirds liberalization episode that took place from 1979 to 1986. She fi nds that two-thirds of the ensuing 19 percent increase in productivity (another example of a massive of the ensuing 19 percent increase in productivity (another example of a massive increase in aggregate productivity) was generated by composition changes within increase in aggregate productivity) was generated by composition changes within industries due to a reallocation of market shares towards more-effi cient producers. industries due to a reallocation of market shares towards more-effi cient producers. Surveys by Greenaway and Kneller (2007) and Wagner (2007) summarize the Surveys by Greenaway and Kneller (2007) and Wagner (2007) summarize the connections between trade liberalization and aggregate productivity—including connections between trade liberalization and aggregate productivity—including this reallocation effect across heterogeneous fi rms—for a wide range of studies this reallocation effect across heterogeneous fi rms—for a wide range of studies and countries.and countries.

8 Specifi cally, Trefl er (2004) regressed labor productivity growth in the period after the free trade agreement (relative to the pre–agreement period) on U.S. and Canadian tariff cuts mandated by the agreement. He then showed that the Canadian tariff cuts raised productivity at the industry level, but not at the plant level. This means that the gains in productivity were coming from selection, rather than from improvements at the plant level. Using this approach, he fi nds that the free trade agreement raised Canadian manufacturing labor productivity by 5.8 percent of which 4.3 percent was due to the exit associated with the Canadian tariff cuts.

Gains from Trade when Firms Matter 111

Gains from Rising Within-Plant Productivity

In this section, we move from this “between-plant” effect in which productive In this section, we move from this “between-plant” effect in which productive plants expand at the expense of less-productive plants to our third source of gains plants expand at the expense of less-productive plants to our third source of gains from trade: a “within-plant” effect in which trade raises productivity of individual from trade: a “within-plant” effect in which trade raises productivity of individual plants by raising the returns to innovation.plants by raising the returns to innovation.

At least as far back as Schmookler (1954), economists have known that the larger At least as far back as Schmookler (1954), economists have known that the larger the market, the more profi table it is for fi rms to invest in productivity-enhancing the market, the more profi table it is for fi rms to invest in productivity-enhancing activities. Firms in large markets have the large sales volumes needed to justify incur-activities. Firms in large markets have the large sales volumes needed to justify incur- ring the high fi xed costs of innovation. The U.S. tariff cuts that were part of the ring the high fi xed costs of innovation. The U.S. tariff cuts that were part of the U.S.–Canada free trade agreement greatly increased the size of the market faced by U.S.–Canada free trade agreement greatly increased the size of the market faced by Canadian fi rms. It should therefore have encouraged Canadian fi rms to increase Canadian fi rms. It should therefore have encouraged Canadian fi rms to increase their exporting their exporting and to increase their investments in productivity-enhancing tech- to increase their investments in productivity-enhancing tech- nologies. We start here with a short extension to the theoretical model that captures nologies. We start here with a short extension to the theoretical model that captures how larger markets generate incentives for some fi rms to innovate, and then turn how larger markets generate incentives for some fi rms to innovate, and then turn to empirical evidence.to empirical evidence.

A Theory of Market Size and Firm Innovation Suppose that an innovation process requires an up-front fi xed cost Suppose that an innovation process requires an up-front fi xed cost fII and and

in return generates a in return generates a reduction in marginal cost in marginal cost ΔΔcII . That is, innovation reduces . That is, innovation reduces marginal cost from marginal cost from c to to c − − ΔΔcII . A fi rm that produces . A fi rm that produces q units of output and engages units of output and engages in innovation will lower its production costs by in innovation will lower its production costs by q ×× ΔΔcII . The fi rm will weigh this cost . The fi rm will weigh this cost saving against the fi xed innovation cost saving against the fi xed innovation cost fII , and innovate if , and innovate if q ×× ΔΔcII >> fII or or

q > f I _

Δ c I .

In words, only fi rms with large volumes q (that is, those with initially lower levels of marginal cost) will fi nd it profi table to innovate. What happens to this fi rm-level innovation decision when trade is liberalized? Lower trade costs increase an export- er’s sales in the export market, and thus increase the exporter’s overall output level q. For some exporters, this increase in output will tip the balance in favor of inno- vating. For some nonexporters, trade liberalization will tip the balance in favor of exporting and innovating.

Evidence on Within-Firm Productivity Growth and Trade For evidence on the link from growth of trade to within-fi rm productivity, For evidence on the link from growth of trade to within-fi rm productivity,

we turn again to Canada’s experience with the free trade agreement. Lileeva and we turn again to Canada’s experience with the free trade agreement. Lileeva and Trefl er (2010) look at their sample of 5,000 Canadian manufacturing plants that did Trefl er (2010) look at their sample of 5,000 Canadian manufacturing plants that did not export prior to 1988 and divide these plants into those that started exporting not export prior to 1988 and divide these plants into those that started exporting after the passage of the free trade agreement and those that did not. In the raw data, after the passage of the free trade agreement and those that did not. In the raw data, the labor productivity of those that started to export rose 29 percent more than for the labor productivity of those that started to export rose 29 percent more than for nonexporters; starting to export was highly correlated with within-plant productivity nonexporters; starting to export was highly correlated with within-plant productivity growth. Of course, this 29 percent number does not take into account a serious growth. Of course, this 29 percent number does not take into account a serious

112 Journal of Economic Perspectives

problem of reverse causality: does exporting lead to increased productivity or does problem of reverse causality: does exporting lead to increased productivity or does increased productivity lead to exporting?increased productivity lead to exporting?

To answer this question, one needs an instrument for exporting: that is, one To answer this question, one needs an instrument for exporting: that is, one needs an event that causes a fi rm to export but that does not directly affect its needs an event that causes a fi rm to export but that does not directly affect its productivity growth. As Lileeva and Trefl er (2010) show, “plant-specifi c” tariff cuts productivity growth. As Lileeva and Trefl er (2010) show, “plant-specifi c” tariff cuts fi t the bill as an instrument. Consider a single Canadian plant called Lumber-fi t the bill as an instrument. Consider a single Canadian plant called Lumber- jack and the many products it produces. Empirically, products are defi ned very jack and the many products it produces. Empirically, products are defi ned very narrowly, at the six-digit level of the Harmonized System product classifi cation, so narrowly, at the six-digit level of the Harmonized System product classifi cation, so that there are thousands of products in manufacturing. For each product produced that there are thousands of products in manufacturing. For each product produced by Lumberjack, one can calculate the U.S. tariff cut. Averaging these tariff cuts by Lumberjack, one can calculate the U.S. tariff cut. Averaging these tariff cuts across all of Lumberjack’s products yields a plant-specifi c tariff cut. This plant-across all of Lumberjack’s products yields a plant-specifi c tariff cut. This plant- specifi c tariff cut has enormous power in predicting whether a Canadian plant specifi c tariff cut has enormous power in predicting whether a Canadian plant starts exporting and how much it exports. The tariff cut also has no direct impact starts exporting and how much it exports. The tariff cut also has no direct impact either theoretically or empirically on a plant’s productivity growth. It is thus a novel either theoretically or empirically on a plant’s productivity growth. It is thus a novel and valid instrument.and valid instrument.

Lileeva and Trefl er (2010) actually do something fancier than instrumental Lileeva and Trefl er (2010) actually do something fancier than instrumental variables—they estimate the local average treatment effect (LATE). This is the variables—they estimate the local average treatment effect (LATE). This is the effect on productivity of starting to export effect on productivity of starting to export for those plants that started exporting because of the tariff cuts. Thus, unlike previous studies of the causal impacts of exporting on . Thus, unlike previous studies of the causal impacts of exporting on productivity, their work only uses information drawn from plants that were likely productivity, their work only uses information drawn from plants that were likely to be affected by the free trade agreement. Using their plant-specifi c tariff instru-to be affected by the free trade agreement. Using their plant-specifi c tariff instru- ment and the local average treatment effect estimator, they establish that the free ment and the local average treatment effect estimator, they establish that the free trade agreement trade agreement caused the productivity of new exporters to rise by 15.3 percent. the productivity of new exporters to rise by 15.3 percent. Since this 15.3 percent rise occurred in plants that accounted for 23 percent of Since this 15.3 percent rise occurred in plants that accounted for 23 percent of Canadian manufacturing output, the 15.3 percent rise in labor productivity Canadian manufacturing output, the 15.3 percent rise in labor productivity of new exporters raised of new exporters raised overall Canadian manufacturing productivity by 3.5 percent Canadian manufacturing productivity by 3.5 percent (3.5 (3.5 == 15.3 15.3 ×× 0.23; see Table 2 below). 0.23; see Table 2 below).

Lileeva and Trefl er (2010) then trace the sources of this productivity gain Lileeva and Trefl er (2010) then trace the sources of this productivity gain back to investments in productivity. The authors examine advanced manufacturing back to investments in productivity. The authors examine advanced manufacturing technologies and rates of innovation at these plants. Table 1 presents the results. technologies and rates of innovation at these plants. Table 1 presents the results. Consider the fi rst row, which deals with management techniques essentially associ-Consider the fi rst row, which deals with management techniques essentially associ- ated with lean manufacturing. In the period immediately after implementation of ated with lean manufacturing. In the period immediately after implementation of the free trade agreement, 16 percent of new exporters adopted such techniques, the free trade agreement, 16 percent of new exporters adopted such techniques, 10 percentage points more than nonexporters did. The fi nal column, which reports 10 percentage points more than nonexporters did. The fi nal column, which reports local average treatment effect (LATE) estimates, shows that 7 of the 10 percentage local average treatment effect (LATE) estimates, shows that 7 of the 10 percentage points was attributable to the effects of increased exporting resulting from the U.S. points was attributable to the effects of increased exporting resulting from the U.S. tariff cuts. As shown in Table 1, similar results hold for the adoption of other tech-tariff cuts. As shown in Table 1, similar results hold for the adoption of other tech- nologies and for innovation.nologies and for innovation.

These results break with the discussion of Bernard, Jensen, Redding, and Schott These results break with the discussion of Bernard, Jensen, Redding, and Schott (2007) in this journal, who correctly argue that most careful studies show exporting (2007) in this journal, who correctly argue that most careful studies show exporting does not raise productivity. Over the years, however, a few careful studies have found raise productivity. Over the years, however, a few careful studies have found otherwise, as in Canada (Baldwin and Gu 2003, 2004; Lileeva 2008), in nine African otherwise, as in Canada (Baldwin and Gu 2003, 2004; Lileeva 2008), in nine African countries (Van Biesebrock 2005), and in Slovenia (De Loecker 2007). López (2005) countries (Van Biesebrock 2005), and in Slovenia (De Loecker 2007). López (2005) provides an overall survey.provides an overall survey.

Marc J. Melitz and Daniel Trefl er 113

What has recently buttressed the minority view that a rise in exporting can What has recently buttressed the minority view that a rise in exporting can lead to a rise in productivity is a spate of papers isolating the causal mechanisms lead to a rise in productivity is a spate of papers isolating the causal mechanisms through which exporting affects productivity. We have already seen the market-size through which exporting affects productivity. We have already seen the market-size mechanism of Lileeva and Trefl er (2010). Bustos (2011a, b) develops a related mechanism of Lileeva and Trefl er (2010). Bustos (2011a, b) develops a related model of scale-biased technology choice, which she takes to Argentinean fi rm-level model of scale-biased technology choice, which she takes to Argentinean fi rm-level data for the 1992–1996 period. Bustos (2011b, table 6) shows that fi rms that began data for the 1992–1996 period. Bustos (2011b, table 6) shows that fi rms that began exporting between 1992 and 1996 also increased their technology spending. Bustos exporting between 1992 and 1996 also increased their technology spending. Bustos (2011a) shows that technology spending increased most in sectors that experi-(2011a) shows that technology spending increased most in sectors that experi- enced improved access to Brazilian product markets (through the tariff cuts in the enced improved access to Brazilian product markets (through the tariff cuts in the Mercosur regional trade agreement). In a series of studies of Taiwanese electronics Mercosur regional trade agreement). In a series of studies of Taiwanese electronics exporters, Aw, Roberts, and Winston (2007) and Aw, Roberts, and Xu (2008, 2011) exporters, Aw, Roberts, and Winston (2007) and Aw, Roberts, and Xu (2008, 2011) show empirically that there is a complex dynamic interplay between the decisions show empirically that there is a complex dynamic interplay between the decisions to export and conduct research and development, with today’s decisions about to export and conduct research and development, with today’s decisions about one affecting tomorrow’s decisions about the other—and both affecting future one affecting tomorrow’s decisions about the other—and both affecting future productivity. Extending this dynamic methodology to general equilibrium, Shen productivity. Extending this dynamic methodology to general equilibrium, Shen (2011) also fi nds strong complementarities between exporting and productivity-(2011) also fi nds strong complementarities between exporting and productivity- enhancing investments among Spanish fi rms. Bloom, Draca, and Van Reenen enhancing investments among Spanish fi rms. Bloom, Draca, and Van Reenen (2011) show that import competition from China has led to increases in R&D, (2011) show that import competition from China has led to increases in R&D, patenting, information technology, and total factor productivity among more tech-patenting, information technology, and total factor productivity among more tech- nologically advanced European fi rms. Atkeson and Burstein (2010) are the only nologically advanced European fi rms. Atkeson and Burstein (2010) are the only ones to examine exporting and investing in productivity within a full-blown general ones to examine exporting and investing in productivity within a full-blown general equilibrium analysis. They fi nd the general equilibrium feedbacks are important.equilibrium analysis. They fi nd the general equilibrium feedbacks are important.

A New Dimension of Heterogeneity In our theoretical model above, fi rms below a certain productivity threshold In our theoretical model above, fi rms below a certain productivity threshold

should not be exporting. Yet in the empirical work reviewed above, we saw that should not be exporting. Yet in the empirical work reviewed above, we saw that many low-productivity Canadian plants started exporting in response to U.S. tariff many low-productivity Canadian plants started exporting in response to U.S. tariff cuts. There is a second puzzle that we have not yet noted: Lileeva and Trefl er (2010, cuts. There is a second puzzle that we have not yet noted: Lileeva and Trefl er (2010, table 3) report that the plants that gained most from starting to export (both in table 3) report that the plants that gained most from starting to export (both in terms of productivity gains and increased innovation) were primarily plants that terms of productivity gains and increased innovation) were primarily plants that

Table 1 Innovation Response to Free Trade Agreement by New Exporters

Raw adoption and innovation rates LATE

New exporters Nonexporters Difference Difference

Manufacturing information systems 16% 6% 10% 7% Inspection and communications 18% 10% 8% 8% Any product or process innovation 30% 20% 10% 8% Any product innovation 26% 14% 12% 7%

Source: Lileeva and Trefl er (2010). Note: “LATE” is the local average treatment effect.

114 Journal of Economic Perspectives

initially had low productivity. That is, among plants that started to export, the initially had low productivity. That is, among plants that started to export, the benefi t was greatest for the least-productive plants.benefi t was greatest for the least-productive plants.

To see why, consider a fi rm that is just indifferent between investing and not To see why, consider a fi rm that is just indifferent between investing and not investing. From the earlier equation, indifference means that investing. From the earlier equation, indifference means that q == fII / /��cII , where , where ��cII is the reduction in marginal cost or the increase in productivity. Rearranging is the reduction in marginal cost or the increase in productivity. Rearranging ((��cII == fII / /q) and noting that sales ) and noting that sales q are increasing in initial productivity, we arrive are increasing in initial productivity, we arrive at a simple conclusion. at a simple conclusion. Among the set of fi rms that are just indifferent between innovating and not innovating, the less-productive, low-, the less-productive, low-q fi rms must expect larger productivity fi rms must expect larger productivity gains gains ��cII from innovation. Lileeva and Trefl er’s (2010) results strongly confi rm from innovation. Lileeva and Trefl er’s (2010) results strongly confi rm this prediction.this prediction.

Conclusions

Recent research into the welfare gains from intra-industry trade have focused Recent research into the welfare gains from intra-industry trade have focused on three sources of gains: 1) gains from increased variety and economies of scale, on three sources of gains: 1) gains from increased variety and economies of scale, 2) productivity gains at the industry level from shifting resources away from low-2) productivity gains at the industry level from shifting resources away from low- productivity fi rms and towards high-productivity fi rms, and 3) productivity gains at productivity fi rms and towards high-productivity fi rms, and 3) productivity gains at the fi rm level from innovating for a larger market. Each of these mechanisms have the fi rm level from innovating for a larger market. Each of these mechanisms have proven to be highly important empirically in the context of the exhaustively studied proven to be highly important empirically in the context of the exhaustively studied Canada–U.S. free trade agreement, and also appear important in many other less-Canada–U.S. free trade agreement, and also appear important in many other less- studied contexts. Indeed, Balistreri, Hillberry, and Rutherford (2011) show that studied contexts. Indeed, Balistreri, Hillberry, and Rutherford (2011) show that adding fi rm heterogeneity to standard computable equilibrium models of trade raises adding fi rm heterogeneity to standard computable equilibrium models of trade raises the gains from trade liberalization by a multiple of four. Empirical confi rmation of the gains from trade liberalization by a multiple of four. Empirical confi rmation of the gains from trade predicted by models with heterogeneous fi rms represents one the gains from trade predicted by models with heterogeneous fi rms represents one of the truly signifi cant advances in the fi eld of international economics.of the truly signifi cant advances in the fi eld of international economics.

We summarize the causal effects of the free trade agreement on overall We summarize the causal effects of the free trade agreement on overall Canadian manufacturing productivity in Table 2. As the last row shows, Canadian Canadian manufacturing productivity in Table 2. As the last row shows, Canadian manufacturing labor productivity rose by 13.8 percent. The idea that a single manufacturing labor productivity rose by 13.8 percent. The idea that a single government policy could raise productivity by such a large amount and in such a government policy could raise productivity by such a large amount and in such a short time-span is truly remarkable.short time-span is truly remarkable.

In writing t his review, we have focused on the net gains from trade. Yet the model In writing t his review, we have focused on the net gains from trade. Yet the model we have developed highlights how intra-industry trade will generate both winners we have developed highlights how intra-industry trade will generate both winners and losers. For example, in the context of the Canada–U.S. free trade agreement, and losers. For example, in the context of the Canada–U.S. free trade agreement, Trefl er (2004) shows that 12 percent of the workers in low-productivity fi rms lost Trefl er (2004) shows that 12 percent of the workers in low-productivity fi rms lost their jobs. Recent research suggests that American workers are similarly struggling their jobs. Recent research suggests that American workers are similarly struggling in response to the Chinese import surge (Liu and Trefl er 2011; Autor, Dorn, and in response to the Chinese import surge (Liu and Trefl er 2011; Autor, Dorn, and Hanson 2011). Clearly, this suggests an important role for policies that provide an Hanson 2011). Clearly, this suggests an important role for policies that provide an adequate safety net and t ransitional assistance for those affected workers. The blow adequate safety net and t ransitional assistance for those affected workers. The blow to those workers could be cushioned by policies that impede the reallocation process to those workers could be cushioned by policies that impede the reallocation process across fi rms. However, such policies—as opposed to policies that provide some form across fi rms. However, such policies—as opposed to policies that provide some form of direct assistance to the affected workers—would also entail a substantial long-run of direct assistance to the affected workers—would also entail a substantial long-run cost. After all, it is precisely this reallocation process that generates some of the cost. After all, it is precisely this reallocation process that generates some of the long-run gains that we have described. In addition, policies that seek to impede the long-run gains that we have described. In addition, policies that seek to impede the

Gains from Trade when Firms Matter 115

reallocation process by making fi rm contractions and expansions costlier would also reallocation process by making fi rm contractions and expansions costlier would also reduce the potential gains to fi rm innovation and hence lead to less innovation and reduce the potential gains to fi rm innovation and hence lead to less innovation and further depress the potential long-run gains from trade. Nonetheless, it is important further depress the potential long-run gains from trade. Nonetheless, it is important to remember that there are winners and losers from trade liberalization not just to remember that there are winners and losers from trade liberalization not just among fi rms, but also among their employees.among fi rms, but also among their employees.

■ Trefl er gratefully acknowledges funding from the Social Sciences and Humanities Research Council of Canada (SSHRC) and the Canadian Institute for Advanced Research (CIFAR). His research would not have been possible without the tremendous support of John Baldwin and Alla Lileeva. We thank the editors (David Autor, Chad Jones, John List, and Timothy Taylor) for their invaluable comments.

References

Table 2 Overall Effect of Free Trade Agreement on Canadian Manufacturing Productivity

Selection/reallocation (between plants) Growth of exporters (most-productive plants) 4.1% Contraction and exit of least-productive plants 4.3%

Within-plant growth New exporters invest in raising productivity 3.5% Existing exporters invest in raising productivity 1.4% Improved access to U.S. intermediate inputs 0.5%

Total 13.8%

Source: Trefl er (2004) and Lileeva and Trefl er (2010).

Abraham, Filip, Jozef Konings, and Stijn Vanormelingen. 2009. “The Effect of Global- ization on Union Bargaining and Price-Cost Margins of Firms.” Review of World Economics (Weltwirtschaftliches Archiv) 145(1): 13–36.

Amiti, Mary, and Jozef Konings. 2007. “Trade Liberalization, Intermediate Inputs, and Produc- tivity: Evidence from Indonesia.” American Economic Review 97(5): 1611–38.

Antràs, Pol., and Esteban Rossi-Hansberg. 2009. “Organizations and Trade.” Annual Review of Economics 1(1): 43–64.

Atkeson, Andrew, and Ariel Tomás Burstein. 2010. “Innovation, Firm Dynamics, and Interna- tional Trade.” Journal of Political Economy 118(3): 433–84.

Autor, David H., David Dorn, and Gordon H. Hanson. 2011. “The China Syndrome: Local

116 Journal of Economic Perspectives

Labor Market Effects of Import Competition in the United States.” August. http://economics.mit .edu/fi les/6613.

Aw, Bee Yan, Mark J. Roberts, and Tor Winston. 2007. “Export Market Participation, Investments in R&D and Worker Training, and the Evolution of Firm Productivity.” World Economy 30(1): 83–104.

Aw, Bee Yan, Mark J. Roberts, and Daniel Yi Xu. 2008. “R&D Investments, Exporting, and the Evolution of Firm Productivity.” American Economic Review 98(2): 451–56.

Aw, Bee Yan, Mark J. Roberts, and Daniel Yi Xu. 2011. “R&D Investment, Exporting, and Produc- tivity Dynamics.” American Economic Review 101(4): 1312–44.

Baggs, Jen. 2005. “Firm Survival and Exit in Response to Trade Liberalization.” Canadian Journal of Economics 38(4): 1364–83.

Baggs, Jen, and James Brander. 2006. “Trade Liberalization, Profi tability, and Financial Leverage.” Journal of International Business Studies 37(2): 196–211.

Baldwin, John R., Desmond Beckstead, and Richard Caves. 2002. “Changes in the Diversifi cation of Canadian Manufacturing Firms (1973–1997): A Move to Specialization.” Statistics Canada, Analytical Studies Branch Research Paper 2002179.

Baldwin, John R., Richard E. Caves, and Wulong Gu. 2005. “Responses to Trade Liberalization: Changes in Product Diversifi cation in Foreign- and Domestic-Controlled Plants.” Chap. 10 in Governance, Multinationals and Growth, edited by Lorraine Eden and Wendy Dobson. Cheltenham, UK: Edward Elgar Publishing.

Baldwin, John R., and Wulong Gu. 2003. “Export-Market Participation and Productivity Performance in Canadian Manufacturing.” Cana- dian Journal of Economics 36(3): 634–57.

Baldwin, John R., and Wulong Gu. 2004. “Trade Liberalization: Export-Market Participa- tion, Productivity Growth and Innovation.” Oxford Review of Economic Policy 20(3): 372–92.

Baldwin, John R., and Wulong Gu. 2006. “Plant Turnover and Productivity Growth in Canadian Manufacturing.” Industrial and Corporate Change 15(3): 417–65.

Baldwin, John R., and Wulong Gu. 2009. “The Impact of Trade on Plant Scale, Production-Run Length and Diversifi cation.” Chap. 15 in Producer Dynamics: New Evidence from Micro Data, edited by Timothy Dunne, J. Bradford Jensen, and Mark J. Roberts. University of Chicago Press.

Balistreri, Edward J., Russell H. Hillberry, and Thomas F. Rutherford. 2011. “Structural Estima- tion and Solution of International Trade Models with Heterogeneous Firms.” Journal of International Economics 83(2): 95–108.

Bartelsman Eric J., John C. Haltiwanger, and Stefano Scarpetta. 2009. “Cross-Country Differ- ences in Productivity: The Role of Allocation and Selection.” NBER Working Paper 15490.

Beamish, Paul W., and Nikhil Celly. 2003. Vincor and the New World of Wine. HBR Case Study, Harvard Business Review.

Bernard, Andrew B., Jonathan Eaton, J. Brad- ford Jensen, and Samuel Kortum. 2003. “Plants and Productivity in International Trade.” American Economic Review 93(4): 1268–90.

Bernard, Andrew B., and J. Bradford Jensen. 1995. “Exporters, Jobs, and Wages in U.S. Manufac- turing: 1976–1987.” Brookings Papers on Economic Activity Microeconomics: 67–112.

Bernard, Andrew B., and J. Bradford Jensen. 1999. “Exceptional Exporter Performance: Cause, Effect, or Both?” Journal of International Economics 47(1): 1–25.

Bernard, Andrew B., and J. Bradford Jensen. 2004. “Exporting and Productivity in the U.S.” Oxford Review of Economic Policy 20(3): 343–57.

Bernard, Andrew B., J. Bradford Jensen, Stephen J. Redding, and Peter K. Schott. 2007. “Firms in International Trade.” Journal of Economic Perspectives 21(3): 105–130.

Bernard, Andrew B., J. Bradford Jensen, and Peter K. Schott. 2006. “Trade Costs, Firms and Productivity.” Journal of Monetary Economics 53(5): 917–37.

Bernard, Andrew B., Stephen J. Redding, and Peter K. Schott. 2011. “Multi-Product Firms and Trade Liberalization.” Quarterly Journal of Economics 126(3): 1271–1318.

Bloom, Nicholas, Mirko Draca, and John Van Reenen. 2011. “Trade Induced Technical Change? The Impact of Chinese Imports on Innovation, IT and Productivity.” NBER Working Paper 16717.

Brander, James A. 1991. “Election Polls, Free Trade, and the Stock Market: Evidence from the 1988 Canadian General Election.” Canadian Journal of Economics 24(4): 827–43.

Broda, Christian, and David E. Weinstein. 2006. “Globalization and the Gains from Variety.” Quar- terly Journal of Economics 121(2): 541–85.

Brülhart, Marius. 2009. “An Account of Global Intra-industry Trade, 1962–2006.” World Economy 32(3): 401–459.

Bustos, Paula. 2011a. “Trade Liberalization, Exports, and Technology Upgrading: Evidence on the Impact of MERCOSUR on Argentinian Firms.” American Economic Review 101(1): 304–40.

Bustos, Paula. 2011b. “The Impact of Trade Liberalization on Skill Upgrading: Evidence from Argentina.” http://www.crei.cat/people/bustos /Trade_Skill_PaulaBustos.pdf.

De Loecker, Jan. 2007. “Do Exports Generate

Marc J. Melitz and Daniel Trefl er 117

Higher Productivity? Evidence from Slovenia.” Journal of International Economics 73(1): 69–98.

De Loecker, Jan. 2011. “Product Differentiation, Multiproduct Firms, and Estimating the Impact of Trade Liberalization on Productivity.” Econometrica 79(5): 1407–51.

Eberhardt, Markus, Christian Helmers, and Zhihong Yu. 2011. “Is the Dragon Learning to Fly? An Analysis of the Chinese Patent Explosion.” CSAE Working Paper WPS/2011-15.

Eckel, Carsten, and J. Peter Neary. 2010. “Multi-Product Firms and Flexible Manufacturing in the Global Economy.” Review of Economic Studies 77(1): 188–217.

Ethier, Wilfred. 1982. “National and Interna- tional Returns to Scale in the Modern Theory of International Trade.” American Economic Review 72(3): 389–405.

Feenstra, Robert C. 2010. Product Variety and the Gains from International Trade. Cambridge MA: MIT Press.

Goldberg, Pinelopi Koujianou, Amit Kumar Khandelwal, Nina Pavcnik, and Petia Topalova. 2010. “Imported Intermediate Inputs and Domestic Product Growth: Evidence from India.” Quarterly Journal of Economics 125(4): 1727 –67.

Greenaway, David, and Richard Kneller. 2007. “Firm Heterogeneity, Exporting and Foreign Direct Investment.” Economic Journal 117(517): 134–61.

Grossman, Gene M., and Elhanan Helpman. 1991. Innovation and Growth in the Global Economy. Cambridge, MA: MIT Press.

Halpern, László, Miklós Koren, and Adam Szeidl. 2005. “Imports and Productivity.” CEPR Discussion Paper 5139.

Harris, Richard. 1984. “Applied General Equi- librium Analysis of Small Open Economies with Scale Economies and Imperfect Competition.” American Economic Review 74(5): 1016–32.

Harrison, Ann E. 1994. “Productivity, Imperfect Competition and Trade Reform: Theory and Evidence.” Journal of International Economics 36(1–2): 53–73.

Harrison, Glenn W., Thomas F. Rutherford, and Ian Wooton. 1989. “The Economic Impact of the European Community.” American Economic Review 79(2): 288–94.

Helpman, Elhanan. 2004. The Mystery of Economic Growth. Cambridge, MA: Belknap Press of Harvard University Press.

Helpman, Elhanan. 2011. Understanding Global Trade. Cambridge, MA: Harvard University Press.

Helpman, Elhanan, and Paul R. Krugman. 1985. Market Structure and Foreign Trade. MIT Press.

Hsieh, Chang-Tai, and Peter J. Klenow. 2009. “Misallocation and Manufacturing TFP in China

and India.” Quarterly Journal of Economics 124(4): 1403–48.

Kasahara, Hiroyuki, and Beverly Lapham. 2012. “Productivity and the Decision to Import and Export: Theory and Evidence.” February 13. http://faculty.arts.ubc.ca/hkasahara/working papers/productivity.pdf.

Kasahara, Hiroyuki, and Joel Rodrigue. 2008. “Does the Use of Imported Intermediates Increase Productivity? Plant-level Evidence.” Journal of Development Economics 87(1): 106–118.

Krugman, Paul R. 1979. “Increasing Returns, Monopolistic Competition, and International Trade.” Journal of International Economics 9(4): 469–79.

Krugman, Paul R. 1980. “Scale Economies, Product Differentiation, and the Pattern of Trade.” American Economic Review 70(5): 950–59.

Levinsohn, James. 1993. “Testing the Imports- as-Market-Discipline Hypothesis.” Journal of International Economics 35(1–2): 1–22.

Lileeva, Alla. 2008. “Trade Evidence and Productivity Dynamics: Evidence from Canada.” Canadian Journal of Economics 41(2): 360–90.

Lileeva, Alla, and Daniel Trefl er. 2010. “Improved Access to Foreign Markets Raises Plant- level Productivity . . . For Some Plants.” Quarterly Journal of Economics 125(3): 1051–99.

Liu, Runjuan, and Daniel Trefl er. 2011. “A Sorted Tale of Globalization: White Collar Jobs and the Rise of Service Offshoring.” NBER Working Paper 17559.

López, Ricardo A. 2005. “Trade and Growth: Reconciling the Macroeconomic and Micro- economic Evidence.” Journal of Economic Surveys 19(4): 623–48.

Mayer, Thierry, Marc J. Melitz, and Gianmarco I. P. Ottaviano. 2011. “Market Size, Competition, and the Product Mix of Exporters.” NBER Working Paper 16959.

Mayer, Thierry, and Gianmarco I. P. Ottaviano. 2008. “The Happy Few: The Internationalisation of European Firms: New Facts Based on Firm- Level Evidence.” Intereconomics: Review of European Economic Policy 43(3): 135–48.

Melitz, Marc J. 2003. “The Impact of Trade on Intra-Industry Reallocations and Aggregate Indus- try Productivity.” Econometrica 71(6): 1695–1725.

Melitz, Marc J., and Gianmarco I. P. Ottaviano. 2008. “Market Size, Trade, and Productivity.” Review of Economic Studies 75(1): 295–316.

OECD. 2002. OECD Economic Outlook. June 2002, no. 71.

Pavcnik, Nina. 2002. “Trade Liberalization, Exit, and Productivity Improvements: Evidence from Chilean Plants.” Review of Economic Studies 69(1): 245–76.

118 Journal of Economic Perspectives

Schmookler, Jacob. 1954. “The Level of Inven- tive Activity.” Review of Economics and Statistics 36(2): 183–190.

Shen, Leilei. 2011. “Product Restructuring, Exports, Investment, and Growth Dynamics.” September 24. http://individual.utoronto.ca /lshen/dynamic.pdf.

Thompson, Aileen J. 1993. “The Anticipated Sectoral Adjustment to the Canada–United States Free Trade Agreement: An Event Study Analysis.” Canadian Journal of Economics 26(2): 253–71.

Topalova, Petia, and Amit Khandelwal. 2011. “Trade Liberalization and Firm Productivity: The Case of India.” Review of Economics and Statistics 93(3): 995–1009.

Tybout, James R., and M. Daniel Westbrook. 1995. “Trade Liberalization and the Dimensions

of Effi ciency Change in Mexican Manufacturing Industries.” Journal of International Economics 39(1–2): 53–78.

Trefl er, Daniel. 2004. “The Long and Short of the Canada–U.S. Free Trade Agreement.” American Economic Review 94(4): 870–95.

Van Biesebroeck, Johannes. 2005. “Exporting Raises Productivity in Sub-Saharan African Manu- facturing Firms.” Journal of International Economics 67(2): 373–91.

Verhoogen, Eric. 2008. “Trade, Quality Upgrading and Wage Inequality in the Mexican Manufacturing Sector.” Quarterly Journal of Economics 123(2): 489–530.

Wagner, Joachim. 2007. “Exports and Produc- tivity: A Survey of the Evidence from Firm-level Data.” World Economy 30(1): 60–82.

This article has been cited by:

1. Andrew B. Bernard, J. Bradford Jensen, Stephen J. Redding, Peter K. Schott. 2018. Global Firms. Journal of Economic Literature 56:2, 565-619. [Abstract] [View PDF article] [PDF with links]

2. Robert C. Feenstra. 2018. Alternative Sources of the Gains from International Trade: Variety, Creative Destruction, and Markups. Journal of Economic Perspectives 32:2, 25-46. [Abstract] [View PDF article] [PDF with links]

  • Gains from Trade when Firms Matter
    • Gains from Love of Variety (Economies of Scale and Product Variety)
    • Gains from Reallocation at the Firm Level
      • Monopolistic Competitors with Heterogeneous Costs
      • What Changes When Economies Integrate?
      • Trade Costs, Export Decisions, and Trade Liberalization
      • Evidence on Gains from Inter-Firm Reallocations
      • Trade Costs and the Export Decision
      • Quantifying the Gains from Trade Due to Reallocation across Heterogeneous Plants
    • Gains from Rising Within-Plant Productivity
      • A Theory of Market Size and Firm Innovation
      • Evidence on Within-Firm Productivity Growth and Trade
      • A New Dimension of Heterogeneity
    • Conclusions
    • References

<< /ASCII85EncodePages false /AllowTransparency false /AutoPositionEPSFiles true /AutoRotatePages /All /Binding /Left /CalGrayProfile (Gray Gamma 2.2) /CalRGBProfile (sRGB IEC61966-2.1) /CalCMYKProfile (U.S. Web Coated \050SWOP\051 v2) /sRGBProfile (sRGB IEC61966-2.1) /CannotEmbedFontPolicy /Warning /CompatibilityLevel 1.4 /CompressObjects /Tags /CompressPages true /ConvertImagesToIndexed true /PassThroughJPEGImages false /CreateJobTicket false /DefaultRenderingIntent /Default /DetectBlends true /DetectCurves 0.1000 /ColorConversionStrategy /UseDeviceIndependentColor /DoThumbnails false /EmbedAllFonts true /EmbedOpenType false /ParseICCProfilesInComments true /EmbedJobOptions true /DSCReportingLevel 0 /EmitDSCWarnings false /EndPage -1 /ImageMemory 1048576 /LockDistillerParams false /MaxSubsetPct 100 /Optimize true /OPM 1 /ParseDSCComments true /ParseDSCCommentsForDocInfo false /PreserveCopyPage true /PreserveDICMYKValues true /PreserveEPSInfo false /PreserveFlatness false /PreserveHalftoneInfo false /PreserveOPIComments false /PreserveOverprintSettings true /StartPage 1 /SubsetFonts true /TransferFunctionInfo /Apply /UCRandBGInfo /Remove /UsePrologue false /ColorSettingsFile () /AlwaysEmbed [ true ] /NeverEmbed [ true ] /AntiAliasColorImages false /CropColorImages false /ColorImageMinResolution 100 /ColorImageMinResolutionPolicy /OK /DownsampleColorImages true /ColorImageDownsampleType /Bicubic /ColorImageResolution 72 /ColorImageDepth -1 /ColorImageMinDownsampleDepth 1 /ColorImageDownsampleThreshold 1.50000 /EncodeColorImages true /ColorImageFilter /DCTEncode /AutoFilterColorImages true /ColorImageAutoFilterStrategy /JPEG /ColorACSImageDict << /QFactor 0.40 /HSamples [1 1 1 1] /VSamples [1 1 1 1] >> /ColorImageDict << /QFactor 1.30 /HSamples [2 1 1 2] /VSamples [2 1 1 2] >> /JPEG2000ColorACSImageDict << /TileWidth 256 /TileHeight 256 /Quality 10 >> /JPEG2000ColorImageDict << /TileWidth 256 /TileHeight 256 /Quality 10 >> /AntiAliasGrayImages false /CropGrayImages false /GrayImageMinResolution 150 /GrayImageMinResolutionPolicy /OK /DownsampleGrayImages true /GrayImageDownsampleType /Bicubic /GrayImageResolution 72 /GrayImageDepth -1 /GrayImageMinDownsampleDepth 2 /GrayImageDownsampleThreshold 1.50000 /EncodeGrayImages true /GrayImageFilter /DCTEncode /AutoFilterGrayImages true /GrayImageAutoFilterStrategy /JPEG /GrayACSImageDict << /QFactor 0.40 /HSamples [1 1 1 1] /VSamples [1 1 1 1] >> /GrayImageDict << /QFactor 1.30 /HSamples [2 1 1 2] /VSamples [2 1 1 2] >> /JPEG2000GrayACSImageDict << /TileWidth 256 /TileHeight 256 /Quality 10 >> /JPEG2000GrayImageDict << /TileWidth 256 /TileHeight 256 /Quality 10 >> /AntiAliasMonoImages false /CropMonoImages false /MonoImageMinResolution 300 /MonoImageMinResolutionPolicy /OK /DownsampleMonoImages true /MonoImageDownsampleType /Bicubic /MonoImageResolution 72 /MonoImageDepth -1 /MonoImageDownsampleThreshold 1.50000 /EncodeMonoImages true /MonoImageFilter /CCITTFaxEncode /MonoImageDict << /K -1 >> /AllowPSXObjects true /CheckCompliance [ /None ] /PDFX1aCheck false /PDFX3Check false /PDFXCompliantPDFOnly false /PDFXNoTrimBoxError true /PDFXTrimBoxToMediaBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXSetBleedBoxToMediaBox true /PDFXBleedBoxToTrimBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXOutputIntentProfile () /PDFXOutputConditionIdentifier () /PDFXOutputCondition () /PDFXRegistryName () /PDFXTrapped /False /CreateJDFFile false /Description << /ENU ([Based on '[Smallest File Size]'] Use these settings to create Adobe PDF documents best suited for on-screen display, e-mail, and the Internet. Created PDF documents can be opened with Acrobat and Adobe Reader 5.0 and later.) >> /Namespace [ (Adobe) (Common) (1.0) ] /OtherNamespaces [ << /AsReaderSpreads false /CropImagesToFrames true /ErrorControl /WarnAndContinue /FlattenerIgnoreSpreadOverrides false /IncludeGuidesGrids false /IncludeNonPrinting false /IncludeSlug false /Namespace [ (Adobe) (InDesign) (4.0) ] /OmitPlacedBitmaps false /OmitPlacedEPS false /OmitPlacedPDF false /SimulateOverprint /Legacy >> << /AddBleedMarks false /AddColorBars false /AddCropMarks false /AddPageInfo false /AddRegMarks false /BleedOffset [ 0 0 0 0 ] /ConvertColors /NoConversion /DestinationProfileName (sRGB IEC61966-2.1) /DestinationProfileSelector /UseName /Downsample16BitImages true /FlattenerPreset << /PresetSelector /MediumResolution >> /FormElements false /GenerateStructure false /IncludeBookmarks false /IncludeHyperlinks false /IncludeInteractive false /IncludeLayers false /IncludeProfiles true /MarksOffset 6 /MarksWeight 0.250000 /MultimediaHandling /UseObjectSettings /Namespace [ (Adobe) (CreativeSuite) (2.0) ] /PDFXOutputIntentProfileSelector /NA /PageMarksFile /RomanDefault /PreserveEditing false /UntaggedCMYKHandling /UseDocumentProfile /UntaggedRGBHandling /UseDocumentProfile /UseDocumentBleed false >> << /AllowImageBreaks true /AllowTableBreaks true /ExpandPage false /HonorBaseURL true /HonorRolloverEffect false /IgnoreHTMLPageBreaks false /IncludeHeaderFooter false /MarginOffset [ 0 0 0 0 ] /MetadataAuthor () /MetadataKeywords () /MetadataSubject () /MetadataTitle () /MetricPageSize [ 0 0 ] /MetricUnit /inch /MobileCompatible 0 /Namespace [ (Adobe) (GoLive) (8.0) ] /OpenZoomToHTMLFontSize false /PageOrientation /Portrait /RemoveBackground false /ShrinkContent true /TreatColorsAs /MainMonitorColors /UseEmbeddedProfiles false /UseHTMLTitleAsMetadata true >> ] >> setdistillerparams << /HWResolution [600 600] /PageSize [612.000 792.000] >> setpagedevice