1 The Variety and Quality of a Nation’s Exports By H UMMELS AND P ETER J. K LENOW * D AVID Large economies export more in absolute terms than do small economies. We use data on shipments by 126 exporting countries to 59 importing countries in 5,000 product categories to answer the question: Do big economies export larger How? margin), a wider set of goods (the extensive quantities of each good (the intensive margin), or higher-quality goods? We find that the extensive margin accounts for around 60 percent of the greater exports of larger economies. Within categories, richer countries export higher quantities at modestly higher prices. We compare these findings to some workhorse trade models. Models with Armington national product differentiation have no extensive margin, and incorrectly predict lower prices for the exports of larger economies. Models with Krugman firm-level product differentiation do feature a prominent extensive margin, but overpredict the rate at which variety responds to exporter size. Models with quality differentiation, mean- while, can match the price facts. Finally, models with fixed costs of exporting to a given market might explain the tendency of larger economies to export a given ( JEL F12, F43) product to more countries. (1987) and Gene M. Grossman and Helpman Virtually every theory of international trade predicts that a larger economy will export more (1991), feature a quality margin, namely richer countries produce and export higher-quality in absolute terms than a smaller economy. goods. Trade theories differ, however, in their predic- larger economies export more. These divergent predictions imply very dif- tions about how ferent consequences for welfare. If larger econ- Models that assume Paul S. Armington’s (1969) omies intensively export more of each variety, intensive national differentiation emphasize the margin: an economy twice the size of another the prices of their national varieties should be lower on the world market. In large-scale Com- will export twice as much but will not export a putable General Equilibrium (CGE) models wider variety of goods. Monopolistic competi- with distinct national varieties, the simulated tion models in the vein of Paul R. Krugman welfare changes associated with trade liberal- (1981) stress the extensive margin: economies ization are dominated by such terms-of-trade twice the size will produce and export twice the effects (see Drusilla K. Brown, 1987). In Daron range of goods. Vertical differentiation models, such as Harry Flam and Elhanan Helpman Acemoglu and Jaume Ventura (2002), these effects prevent real per capita incomes from diverging across countries with differing invest- * Hummels: Krannert School of Management, Purdue ment rates. These authors argue that richer University, 403 West State Street, West Lafayette, IN 47907 countries face lower export prices, and that this and National Bureau of Economic Research (e-mail: [email protected]); Klenow: Department of is the critical force maintaining a stationary 1 Economics, Stanford University, 579 Serra Mall, Stanford, world income distribution. CA 94305 and National Bureau of Economic Research To the extent larger economies export a wider (e-mail: [email protected]). We are grateful to Oleksiy array of goods or export higher-quality goods, Kryvtsov and Volodymyr Lugovskyy for excellent research lower export prices are no longer a necessary assistance, and to the Purdue University Center for Interna- tional Business and Research for funding data purchases. Hummels acknowledges the assistance of the National Sci- 1 ence Foundation (Grant 0318242). We thank Mark Bils, Donald R. Davis and David E. Weinstein (2002) build on the Acemoglu and Ventura model in estimating terms- Russell Cooper, Eduardo Engel, Thomas Hertel, Russell of-trade-driven welfare losses to U.S. natives from in- Hillberry, Tom Holmes, Valerie Ramey, Richard Rogerson, migration. and three anonymous referees for helpful comments. 704
2 VOL. 95 NO. 3 HUMMELS AND KLENOW: THE VARIETY AND QUALITY OF A NATION’S EXPORTS 705 consequence of size. Rather than sliding down sive margins are dominated by higher quantities of each good rather than higher unit prices. world demand curves for each variety, bigger economies may export more varieties to more Richer countries export higher quantities of each good at modestly higher prices, consistent countries. Or they may export higher-quality goods at higher prices. If variety and quality with higher quality. Countries with more work- ers export higher quantities of each good, but margins dominate, then growth and develop- ment economists must rely on other forces— not at higher prices. These patterns hold for both the U.N. data with broad geographic cov- such as technology diffusion and diminishing erage, and U.S. data with more detailed product returns to capital—to tether the incomes of coverage. high- and low-investment economies. Further- The large extensive margins are inconsistent more, the welfare effects of trade liberalization with Armington models, which have no exten- could be very different than is typically found in many CGE models. sive margin and imply that larger economies face lower export prices. In contrast, Krugman- In this paper we use highly detailed 1995 style models with firm-level product differenti- United Nations data on exports to assess the ation predict that larger economies will produce importance of the extensive, intensive, and quality margins in trade. The data cover exports and export more varieties, consistent with the from 126 countries to each of 59 importers in large extensive margins we find (assuming a over 5,000 six-digit product categories. To strictly increasing relationship between variet- check robustness we also examine exports by ies produced and varieties exported). However, these models predict that variety will expand in 124 countries to the United States in 1995 in proportion to exporter size, which overstates the over 13,000 ten-digit product categories. We size of the observable extensive margin in the decompose a nation’s exports into contributions data. Further, the Krugman model predicts that from intensive versus extensive margins, and a country will export to all markets if it exports further decompose the intensive margin into price and quantity components. We then relate to any markets in a category, a prediction strik- ingly at odds with the evidence. Countries typ- each margin to country size (PPP GDP) as well ically export to a strict subset of markets, with as to its components: workers, and GDP per larger economies exporting to decidedly more worker. markets. This suggests that fixed costs of ex- Of special interest are the extensive and qual- ity margins. There are many possible ways to porting a given product to a given market, as modeled by Paul Romer (1994), may be define the extensive margin (counting catego- important. ries exported, counting categories over a certain Our investigation builds on the empirical size, weighting categories in various ways). We work of many predecessors. Feenstra (1994) measure the extensive margin in a manner con- applied his method to U.S. imports of six man- sistent with consumer price theory by adapting the methodology in Robert C. Feenstra (1994), ufactured goods and found evidence of substan- tial import variety growth. Michael Funke and which appropriately weights categories of goods by their overall importance in exports to Ralf Ruhwedel (2001) found that the variety of a given country. The quality margin is not di- both exports and imports are positively corre- lated with per capita income across 19 OECD rectly observable but can be inferred by exam- countries. Keith Head and John Ries (2001) ining projections of price and quantity on GDP looked for home-market effects in U.S. and and its components. That is, if large exporters systematically sell high quantities at high Canadian trade in order empirically to distin- prices, this is consistent with these exporters guish increasing returns and national product differentiation models. They found the evidence producing higher-quality goods. We also show alternatively how to interpret the projections of mostly consistent with national product differ- price and quantity in terms of unmeasured, entiation. By comparison, we examine model implications for extensive (increasing returns) within-category variety. versus intensive (national product differentia- Our findings are as follows. The extensive margin accounts for about 60 percent of the tion) margins, along with price and quantity effects that each implies. Peter K. Schott (2004) greater exports of larger economies. The inten-
3 JUNE 2005 706 THE AMERICAN ECONOMIC REVIEW ABLE 1—M ODEL P REDICTIONS FOR E XPORT M ARGINS T Intensive Extensive Price Quantity ( V ) ) px p ) ) ( x ( ( Armington 1 0 1/( 1) /( 1) Acemoglu & Ventura / Y L 10 1.6 0.6 L 01 0 0 Krugman 0 0 1 0 Quality 10 differentiation / L 10 Y 01 L Notes: For discussion of each model, see Section I in the text. Entries are model predictions for how exports increase with respect to exporter size. A single entry indicates the same elasticity with respect to both Y / L (GDP per worker) and L (employment). The Acemoglu and /( 1) and are equal to 1/( L / Y Ventura price and quantity elasticities with respect to 0.6 and 1.6 for their case of 2.6. 1), but these take on the values found that richer countries export to the U.S. at within the intensive margin are more subtle. In higher unit prices within narrow categories. the exposition below we describe the implica- tions of the models for prices and therefore the Countries more abundant in physical and hu- value of output for each exporter. Then, a pro- man capital likewise export a given variety at jection of each margin on output (or on output higher unit prices. Like Schott, we use data on export prices in narrow categories for countries per worker and number of workers) provides information on how well the models describe of differing income levels. Unlike his study, we the data. examine a broad range of importers and use To help explain the Table 1 entries, consider quantity data along with price data to extract information about quality differences. the following general environment. Consumers The rest of the paper proceeds as follows. In countries in each J buy from up to m in country observable categories of goods. Goods are I of Section I we briefly outline the predictions of differentiated both across categories and across some trade models for the various margins. We categories. For ex- within discuss the data we use in Section II, and this producing countries ample, midsize cars and trucks may be distinct guides how we define the extensive and inten- sive margins (and the latter’s price and quantity observable categories, but within a category components) in Section III. In Section IV we Japanese midsize cars are differentiated from present our empirical findings, and in Section V German midsize cars. For simplicity we adopt a Dixit-Stiglitz formulation with a single elastic- we offer conclusions and possible directions for future work. ity of substitution 1 between goods in different categories and goods from different countries. Consumers maximize utility given by I. Export Margins in Various Models 1 / J I In Table 1 we summarize what four trade 1/ 1 (1) U x N Q m jmi jmi jmi models predict for the size of the intensive and j 1 i 1 extensive margins, and for the price and quan- tity components of the intensive margin. In all of these models, exporter variation in workers subject to and productivity will cause variation in the quantity of output and exports, but along differ- J I ent margins. The predictions for the intensive (2) p N . x Y and extensive margins are stark and well jmi jmi jmi m 1 i j 1 known, but the price and quantity variations
4 VOL. 95 NO. 3 HUMMELS AND KLENOW: THE VARIETY AND QUALITY OF A NATION’S EXPORTS 707 Quality likewise does not vary across countries Q Here of varieties exported by quality is the jmi 2 i . m to country country in category j Q ( N 1 for all j ). A country with more workers is the j jmi or higher productivity simply produces more of number of symmetric varieties exported from j to i . (We assume for simplic- each variety ( x category m within ). This intensive margin L A j j j ity that these within-category varieties are sym- results in lower prices for each variety. The x metric.) effect on export prices is smaller the larger the is the number of units (quantity) jmi per variety in category i , exported from to j m elasticity of substitution between varieties: 1/ and p p is the price of each of the units. If ) L A ( j . Country ’s GDP is Y j j j jmi j 1 1/ p m - does not buy from country j country in cate- L ) x . Taking logs and rearrang V ( A j j j j j j j ing, country (say, because ’s export quantities and prices can i does not produce any gory i be expressed as x varieties in category ), then N 0 and jmi jmi Y 0. ’s income. is country m m If midsize car models are an observable cat- egory, then Japan’s exporting of multiple, dif- (3) ln x L / ln Y ln L j j j j 1 1 ferentiated midsize car models to the United N States would be an example of 1. Of jmi course, the more disaggregated the trade data, and the more cross-category variety is captured by the observable categories (the ’s). In the data I 1 1 . L / p ln (4) ln L Y ln j j j j section we examine the sensitivity of our results 1 1 to changing levels of aggregation. And although are not N unobserved, within-category varieties These expressions are the basis of the price and , we will directly distinguishable from quality Q quantity entries in the first row of Table 1. In be able to draw some inferences about the role this Armington world, larger economies inten- 4 of each using price and quantity data. sively export higher quantities at lower prices. 3 We now explain the entries in Table 1. In Many CGE models of trade liberalization employ a modified Armington structure that j doing so we focus on exporter variation that differs from the stark assumptions in this base feeds into proportional variation across all mar- . That is, market-specific i and categories m kets model. In particular, they employ exporter- specific weights in the utility function calibrated and category-specific proportional constants are so that exporter prices and country size do not omitted. We also express all objects relative to systematically co-vary in the cross section. an exporter for which the following variables , I are normalized to 1: (productiv- A , or Q These weights can be thought of as quality p , x , N , Q (employment), and N unobserved variety . Since the weights are ity), A . We assume that Y L and fixed, however, the implications of the base L differ exogenously across countries. We summarize variety within and across categories Armington model still apply to time series. That V is, changes in exporter size or income are pre- NI ( 1 in the reference country). as dicted to yield changes in output and prices as in A. Armington first-differenced versions of equations (3) and (4). In Armington’s (1969) national differentia- tion model, each country produces a single va- 4 riety in each category ( V 1 for all 1/ , given the ) p ( ln An alternative to expression (4) is j j j - ln ( A ) L ). For empirical estimation, this expres 1/ ln ( j j normalization), so there is no extensive margin. sion would allow consistent estimation of the effect of exogenous variables. With (4), in contrast, the effect of employment on prices will be biased downward (upward in 2 absolute terms). Higher L lowers income per worker for a We let quality enter the utility function without an A requires a higher exponent so that it is in “price units,” i.e., equivalent to a . The A given L , so controlling for Y / coefficient on L in (4) therefore captures the effect on export lower price. This is purely a normalization. Quality is a to keep prices of higher L , combined with enough higher A demand shifter in (1), raising the quantity a country can export to a market at a given price. Y / L unchanged. As we discuss below, we focus on (4) 3 because Y / L is directly observable, whereas one must know We refer the reader to Hummels and Klenow (2002) for a more detailed exposition. (and quality if it varies across countries) to derive A .
5 JUNE 2005 THE AMERICAN ECONOMIC REVIEW 708 exports this variety to all other markets. A cor- B. Acemoglu and Ventura ollary is that, conditional on exporting in a category, a country exports in this category to Acemoglu and Ventura (2002) add endoge- all other countries. In models with fixed costs of nous capital accumulation and an endogenous exporting to each market, such as Romer number of varieties to the Armington model. (1994), a country may instead export to a strict They posit constant returns to capital in the subset of markets, or even to no markets at all production of each variety, and a fixed labor despite producing in the category. When we requirement for producing each variety. The discuss our empirical findings in Section IV number of varieties a country produces is then below, we will present evidence on export des- proportional to its employment ( V ). A L j j A country with higher productivity ( tinations to address this issue. , broadly j construed to include physical capital) produces x more of each variety ( - Quality Differentiation D. ). Higher produc A j j tion of each variety translates into lower prices 1/ p for each variety: Suppose quality varies across countries ( Q ) ( j A ’s GDP . Country j j j 1/ 1 is Y ) but productivity and variety do j differs across p ) x V ( A L , but . Greater Y / L j j j j j j not greater L , is associated with producing not ( A ). Countries with V 1 and j 1 for all j j more workers produce more of each variety higher quantities of each variety and selling ( x them at lower unit prices: L ). A country’s unit prices reflect both the j j level of employment and the level of quality: 1 1/ 1/ p ( L Q ( Q Y . GDP is ) . Also, L ) j j j j j j 1/ ln (5) x / L ln Y j j j Y p . Quantity per variety ) L Q ( L / j j j j j 1 should positively project on exporter employ- ment but be unrelated to exporter GDP per and worker; prices for varieties should project pos- itively on GDP per worker but be unrelated to 1 (6) . L / ln p Y ln j j j employment. These results are shown in the 1 final row of Table 1. More generally, we can use consumer first- order conditions from (1) and (2) to express The second row of Table 1 summarizes this quality and within-category variety in terms of model’s predictions. the observed prices and quantities and the elas- ticity of substitution between varieties: C. Krugman 1 1 Krugman (1979, 1980, 1981) modeled coun- Q ln (7) ln N p ln ln . x N j j j j j tries as producing an endogenous number of 5 With fixed output costs of producing varieties. each variety, the number of varieties produced Note that the observed quantities per category in a country is proportional to the size of the Nx , rather than the theoretically are actually V economy ( ideal . Also note that quality and within- x ). In this simplest Y L A j j j j Krugman world, all countries export the same category variety are isomorphic (up to a scalar) x quantity per variety ( in this expression. We return to this issue when ) and export 1 for all j j at the same unit prices ( p discussing the empirical results. 1 for all j ). Neither j unit prices nor quantity per variety vary with GDP per worker or the number of workers. II. Data Description These results are stated in the third row of Ta- ble 1. We draw export data from two sources. We use The Krugman model has the property that, worldwide data from the United Nations Confer- conditional on producing a variety, a country ence on Trade and Development (UNCTAD) and Information Analysis Trade System (TRAINS) CD-ROM for 1995. The TRAINS 5 project combines bilateral import data collected See also Wilfred J. Ethier (1979, 1982).
6 VOL. 95 NO. 3 HUMMELS AND KLENOW: THE VARIETY AND QUALITY OF A NATION’S EXPORTS 709 by the national statistical agencies of 76 import- countries shipping to the United States, a total of 222 in 1995. We have data on employment ing countries, covering all exporting countries 7 and output in 1995 for 124 of these exporters. (227 in 1995). The data are reported in the Harmonized System (HS) classification code at In both datasets, we measure prices as unit the six-digit level, or 5,017 goods, and include values (value/quantity). Quantity (and therefore price) data are missing for approximately 16 shipment values and quantities. For a subset of these countries (126 of the 227 exporters and 59 percent of U.S. observations and 18 percent of 8 worldwide observations for 1995. of the 76 importers), we have matching employ- When the ment and GDP data (discussed below). The 59 U.S. data include multiple shipments from an exporter in a ten-digit category, we aggregate importers represent the vast majority of world values and quantities. The resulting prices are imports, so total shipments for each exporter reported in TRAINS closely approximate quantity-weighted averages of prices found within shipments from that exporter category. worldwide shipments for that exporter. Data on national employment and GDP at Our calculation of the extensive margin may 1996 international (PPP) prices come from Alan be sensitive to the level of aggregation at which we measure exports. That is, a country may Heston et al. (2002). We use PPP GDP, as opposed to GDP at current market exchange , but only categories NI export total variety V I are observable. As data become more aggre- rates, to avoid any mechanical association be- tween an exporter’s price and GDP through the gated, variety shifts from the observable I to the unobservable within-category N value of its market exchange rate. All of our . For example, empirical results are robust to using GDP val- were we to use the output data available in internationally comparable form at roughly the ued at market exchange rates instead of PPP two-digit level, we would find that most coun- exchange rates. 6 We tries produce and export in all sectors. would then obtain much smaller extensive mar- III. Decomposition Methodology gins, with most variety differences relegated to the intensive margin. By using more detailed We now construct empirical counterparts to ), the category exten- px the intensive margin ( export data with 5,017 six-digit HS categories, p ), and the price ( sive margin ( we can do a better job of assigning variety ) and quantity I differences to the extensive margin. x ( ) components of the intensive margin. To do so, we adapt Feenstra’s (1994) methodology for We also use U.S. data with more product incorporating new varieties into a country’s im- detail from the “U.S. Imports of Merchandise” CD-ROM for 1995, published by the U.S. Bu- port price index when preferences take the form of our equation (1). Feenstra shows that the reau of the Census. The data are drawn from electronically submitted customs forms that re- import price index is effectively lowered when port the country of origin, value, quantity, the set of goods expands. freight paid, duties paid, and HS code for each Instead of comparing varieties imported over shipment entering the United States. The ten- time, we compare varieties imported from dif- ferent exporters at a point in time. In this case, digit HS scheme includes 13,386 highly de- j tailed goods categories. The data include all comparing export prices for country relative to k a reference country requires an adjustment for the size of each exporter’s goods set. The ap- 6 propriate adjustment is the extensive margin. Prominent CGE models typically feature fewer than 50 manufactured goods, primarily because of the dearth of For the case when m are a ’s shipments to j output data, and therefore include no extensive margins in their analysis. The most disaggregated model we could find in the CGE literature is that employed by the U.S. Interna- 7 tional Trade Commission. It has roughly 500 sectors, still an The remaining 98, primarily very small or former order of magnitude fewer than the six-digit HS codes we Soviet-bloc countries, comprise only 5 percent of U.S. trade in 1995. use. Also, its structure is fundamentally different from the 8 - models under consideration here. It has a United States The likelihood that quantity data are missing is uncor related with aggregate employment and GDP per worker, so versus rest-of-world focus, contains no data on rest-of- our analyses should not be biased by dropping these world output, and allows for no cross-country differences in observations. varieties either in cross section or over time.
7 JUNE 2005 THE AMERICAN ECONOMIC REVIEW 710 equals the product of the m exports to k country , the extensive subset of ’s shipments to m k two margins: margin is defined as I x p ¥ kmi kmi x p ¥ jmi jmi i I jm i 1 EM (8) . jm . EM IM (10) jm jm ¥ x p kmi kmi I i I ¥ p x kmi kmi i 1 This is a cross-exporter analogue of Feenstra’s new varieties adjustment to an import price To see a simple example of the intensive versus index. I extensive decomposition, compare German and is the set of observable categories in jm Belgian exports to the United States, using k j which country has positive exports to , i.e., m rest-of-world for the reference country in each x 0. (In our empirical implementation, the jmi case. Given the size of each, it is not surprising I categories will be 5,017 six-digit U.N. HS that Germany’s exports to the United States are has posi- k product codes.) Reference country 6.2 times larger than Belgium’s. Some of this tive exports to I categories. (In our m in all comes through a greater number of categories will be rest-of- k empirical implementation, shipped—Germany ships in 79 percent of the world.) EM m equals country k ’s exports to in jm 5,017 six-digit HS codes, while Belgium ships I ’s exports to m I in all k relative to country jm in 51 percent. Were all categories of equal categories. weight, this would yield an extensive margin for The extensive margin can be thought of as a Germany that is 1.55 times larger than Bel- j ’s categories relative to ’s k weighted count of gium’s. This leaves an intensive margin (i.e., categories. If all categories are of equal impor- exports per category) for Germany that is four tance, then the extensive margin is simply the times larger than Belgium’s. However, not all exports to j fraction of categories in which . m categories are of equal weight. Germany ships More generally, categories are weighted by in categories that are a larger share of rest-of- ’s exports to m k . An advan- their importance in world exports to the United States, the numer- tage of evaluating a category’s importance with- ator in equation (8). Incorporating the weighted j ’s exports is that it prevents a out reference to sive margin is 1.65 counts, Germany’s exten category from appearing important solely be- times greater than Belgium’s, and its intensive cause j (and no other country) exports a lot to m 9 margin is 3.75 times larger. in that category. We now turn to decomposing the intensive The corresponding intensive margin com- margin into price and quantity indices. Suppose pares nominal shipments for and k in a com- j that quality ( ) N ) and within-category variety ( Q mon set of goods. It is given by vary across categories m . i for each importer This encompasses preferences that place more x p ¥ jmi jmi weight on some goods than on others. As a i I jm baseline case, assume further that quality and IM (9) . jm ¥ x p kmi kmi within-category variety do not vary by exporter. i I jm (In our empirical analysis, we will test these assumptions.) For this baseline case, Feenstra (1994) derives an exact price index for the in- IM j equals ’s nominal exports relative to k ’s jm m j ’s imports from tensive margin of country in those categories in which j nominal exports versus k : exports to m I ( ). The ratio of country to j jm w jmi p jmi 9 A disadvantage is that a country may appear to have P (11) . jm p kmi a large extensive margin because it exports a small I i jm exports a lot. As we amount in categories in which k discuss in the next section, we do not find this to be the case empirically. w In (11), is the logarithmic mean of s (the jmi jmi
8 VOL. 95 NO. 3 HUMMELS AND KLENOW: THE VARIETY AND QUALITY OF A NATION’S EXPORTS 711 ploy equations (8) through (12) with the import j ’s exports to m ) share of category i in country market m the United States. The U.N. TRAINS s and (the share of category ’s exports to i k in kmi data contain 59 import markets. We summarize i m I , where ): jm each exporter’s margins across all the markets as follows. We first decompose country ’s exports to j x x p p jmi kmi kmi jmi , s , s kmi jmi M m each market is the set of M , where j p ¥ x p ¥ x jmi kmi jmi kmi countries for which import data are available. We I i i I jm jm j ’s then take the geometric average of country s s M decompositions across the markets to get jmi kmi j ln s ln s jmi kmi a a jm jm w . IM EM EM IM jmi jm j j jm s s kmi jmi M m M m j j ¥ ln s ln s jmi kmi I i jm a a jm jm P P . X X j j jm jm We decompose the intensive margin into the M m m M j j 10 price index (11) and an implicit quantity index: a The weight is the logarithmic mean of the jm shares of m in the overall exports of j and . P (12) X IM jm jm jm W ’s a , respectively (normalized so that j m jm M sum to 1 over the set ). Expressions (8) through (12) define our de- j ’s exports to a given j composition of country IV. Empirical Results ). The market m k ’s exports to m (relative to functional forms in (8) through (12) are all For each exporting country, we construct over- based on Feenstra’s (1994) theory. To imple- all exports, the intensive margin, the extensive ment the decompositions, we need to choose a margin, and the price and quantity components of m to k reference k , we choose . For each market the intensive margin. We then regress the natural . In our dataset other than j be all exporters to m log of each margin on the exporter’s log GDP we always find I I j exports to , i.e., country jm relative to rest-of-world log GDP. Separately, we to m k in a subset of the categories exported by regress each margin on exporter GDP per worker j ’s extensive margin, it means . For measuring m and log employment, both expressed relative to the importance of different categories is deter- the rest of the world. The regression samples are mined by the broadest possible set of other cross sections of exporting countries in 1995. Ta- countries. Similarly, for measuring the price ble A1 in the Appendix presents each of these and quantity components of the intensive mar- variables for all 126 countries. ’s prices and quantities are compared to gin, j This approach has two advantages. First, be- . those of all other countries exporting to m cause OLS is a linear operator, the regressions For our implementation with U.S. data, we em- additively decompose the margins along which larger economies export more. Second, by pro- 10 jecting each margin on GDP, etc., our conclu- Feenstra incorporates the extensive margin into a broader price index. The analogue for cross-country exports sions are more robust to measurement error. For is example, the level of the extensive margin can be sensitive to the inclusion of very small trade I flows that one might argue should rightly be ¥ x p jmi jmi i 1 1 1/ 1 / ignored. But a projection of the extensive mar- EM X . EM P jm jm jm jm I gin on log GDP is not sensitive to this unless p x ¥ kmi kmi there is a systematic relationship between the i 1 11 measurement error and exporter GDP. The first bracketed term is a price index that reflects how higher EM lowers the cost of obtaining utility through jm . The second bracketed term is a imports from country j 11 quantity index incorporating the impact of the lower effec- We experimented with discarding small trade flows, with cutoffs at various levels in absolute and percentage- tive price on demand for country exports. j
9 JUNE 2005 712 THE AMERICAN ECONOMIC REVIEW ARGINS M T I XTENSIVE AND 2—E ABLE NTENSIVE Independent variable 3 2 2 Adj. R Adj. LL / Y 2 Dependent variable R Y 0.89 0.86 1.00 0.83 1.29 Overall exports (0.04) (0.07) (0.04) Intensive margin 0.44 0.36 0.60 0.38 0.60 (0.03) (0.05) (0.03) 34% 41% 38% 0.61 0.79 0.53 0.85 Extensive margin 0.74 (0.05) (0.03) (0.03) 66% 59% 62% number of Notes: All variables are in natural logs. Number of exporting countries observations 126. Standard errors are in parentheses. For definitions of each margin, see equations (8), (9), and (10). Percentages describe the contribution of each margin to the overall export elasticity. L 1995 employment in the exporting country relative to the sum of employment in the other 125 exporters. Y 1995 PPP GDP in the exporting country relative to the sum of GDP in the other 125 exporters. Y / L is simply the ratio of these two variables. Sources: UNCTAD for 1995 exports to 59 countries by 126 countries in 5,017 six-digit categories. Heston et al. (2002) for employment and PPP GDP. Table 3 breaks the intensive margin into its Although we will be comparing our findings to each model’s predictions, we are not aiming price and quantity components. Within catego- ries and to a given market, countries with twice to test each model formally. The models were the GDP per worker export 34 percent higher deliberately stark and can all easily be rejected. quantities at 9 percent higher prices. Countries Our goal instead is to identify model ingredients that may help explain the facts. This, we hope, with twice the employment tend to export 37 percent higher quantities at no higher or lower will prove useful for future work developing a single model consistent with the facts. prices. Economies with twice the GDP export Table 2 presents the extensive and intensive 36 percent higher quantities at 2 percent higher margins in the 1995 U.N. data. The data cover prices, not far from the elasticities with respect exports by 126 countries to 59 markets in 5,017 to employment. categories. Each regression has 126 observa- The results in Tables 2 and 3 do not conform tions, one for each exporting country. All of the to the predictions of any single model in Ta- ble 1. This is not surprising, given that the coefficients in the tables are significantly differ- -values below 1 percent unless ent from zero ( models are polar cases. The Armington model p has no extensive margin, omitting a channel that otherwise noted). The first row shows that larger economies export substantially more to constitutes more than half the exports of larger the typical market. The second and third rows economies. Within the intensive margin, Arm- ington predicts higher quantities per variety and report that, with respect to GDP, around 38 lower prices. Countries with higher GDP do percent of this occurs on the intensive margin export higher quantities per category, but not and the other 62 percent on the extensive mar- gin. Figure 1 plots the extensive margin against nearly to the extent predicted by the model. Just as striking, larger economies do not export their GDP for the 126 countries. Table 2 shows fur- varieties at lower prices. Richer countries ex- ther that the extensive margin plays a more port at modestly higher prices, and countries prominent role for richer economies (66 per- cent) than for economies with more workers (59 with more workers export at neither higher nor percent). lower prices. Typical elasticities of substitution estimated at the six-digit level are between five and ten (see Hummels, 1999). The Armington model therefore predicts price elasticities in the of-exports terms. The cutoffs did not materially alter exten- range 0.25. The coefficient on GDP 0.11 to sive margin projections on GDP or its components.
10 VOL. 95 NO. 3 HUMMELS AND KLENOW: THE VARIETY AND QUALITY OF A NATION’S EXPORTS 713 F IGURE 1 ABLE ARGIN M NTENSIVE I UANTITY C Q RICE AND 3—P T OMPONENTS OF THE Independent variable 3 2 2 Y 2 Dependent variable R Y Adj. R Adj. LL / 0.01 0.01 0.02 0.14 0.09 Prices (0.01) (0.01) (0.02) 0.58 0.34 Quantities 0.37 0.58 0.36 (0.03) (0.03) (0.05) number of All variables are in natural logs. Number of exporting countries Notes: observations 126. Standard errors are in parentheses. For definitions of the price and L 1995 employment in the exporting quantity components, see equations (11) and (12). country relative to the sum of employment in the other 125 exporters. Y 1995 PPP GDP in the exporting country relative to the sum of GDP in the other 125 exporters. Y L / is simply the ratio of these two variables. UNCTAD for 1995 exports to 59 countries by 126 countries in 5,017 six-digit Sources: categories. Heston et al. (2002) for employment and PPP GDP. with respect to GDP per worker to be large and per worker is ten standard errors away from this, and the coefficient on employment is eight 0.6). In contrast, the empir- negative (around ical price effects are small and go in the wrong standard errors away. Like the Armington model, the Acemoglu direction. This suggests that diminishing returns and Ventura model predicts richer countries will and technology diffusion may be needed to en- export higher quantities of each variety at a sure a stationary world income distribution. lower price. To match facts about the world The Krugman model does feature a promi- income distribution, Acemoglu and Ventura nent extensive margin, consistent with evidence assuming there is a strictly increasing relation- (2002) need the elasticity of substitution to be small (around 2.6), and the elasticity of price ship between production of varieties and
11 JUNE 2005 714 THE AMERICAN ECONOMIC REVIEW T XTENSIVE L V EVELS OF M ARIOUS E HE 4—T ABLE ARGIN AT exporting of observable categories. The Krug- A GGREGATION man model’s predictions for prices and quanti- ties are also closer to the data than are the L 3 Y Regressor Y L / j j j j predictions of the Armington and Acemoglu 66% 59% 62% 6 digit and Ventura models, but discrepancies remain. 64% 56% 59% 5 digit Richer economies and those with more workers 62% 4 digit 49% 54% have notably higher export quantities. This 48% 34% 39% 3 digit could be consistent with the Krugman model if 30% 2 digit 39% 25% 11% 9% 15% 1 digit within larger economies produce more varieties six-digit categories. The level of aggregation at Notes: number of obser- Number of exporting countries which one measures the extensive margin vations 126. For the definition of the extensive margin, clearly affects how it varies according to coun- see equation (8). The percentages describe the contribution L 1995 of each margin to the overall export elasticity. try size. If variety differences exist at more employment in the exporting country relative to the sum of disaggregated levels (e.g., eight-digit or ten- 1995 PPP GDP employment in the other 125 exporters. Y digit), then we will capture only some of the in the exporting country relative to the sum of GDP in the variety differences in the extensive margin with is simply the ratio of these two L / Y other 125 exporters. six-digit data, and some will be in the intensive variables. Sources: UNCTAD for 1995 exports to 59 countries by 126 margin. This can be seen most clearly by rede- countries in 5,017 six-digit categories. Heston et al. (2002) fining the extensive margin at more aggregated for employment and PPP GDP. data levels. Table 4 displays the covariation of country size with extensive margins measured Putting the polar quality differentiation model at the six-, five-, four-, three-, two-, and one- aside, how much does quality vary with exporter digit levels. As we aggregate, the size of the income? We would like to extract quality from the extensive margin (and its covariation with 12 price and quantity margins using equation (7) country size) naturally declines. One fact difficult for the Krugman framework above. We cannot disentangle quality from with- in-category variety, however, unless we have de- to explain is the higher price of rich country exports. Quality differentiation would seem to tailed data on the precise number of varieties per good from another source. For an example of this be necessary. The simple quality differentiation model described above has no extensive mar- sort of calculation, consider Japanese versus South gin, which is at odds with the large extensive Korean car exports to the United States. In 1995, dollar sales of Japanese models in the United margins documented in Table 2. But this model has an ingredient which can help explain some States exceeded dollar sales of South Korean 13 models by a factor of 28. Japan exported 56 of the price and quantity facts in Table 3. By exporting higher-quality goods, richer econo- different car models to the United States in 1995, 14 mies can export higher quantities without low- whereas South Korea exported 8 car models. ering the prices of their varieties on world We would therefore attribute a factor of 7 out of markets. Quality is a demand shifter in our the 28 total to more Japanese varieties (a 58- specification of utility in (1), raising the quan- percent extensive margin in log terms), and the tity a country can export to a market at a given remaining factor of 4 to the intensive margin. The price. The polar version of the quality model— average unit price of Japanese models was almost with no variety margin— overstates the price 2.4 times the average unit price of South Korean margin for richer economies, but fits the (ab- sence of any) price margin for countries with 13 The data are from Ward’s Motor Vehicle Facts & more workers. It understates the quantity mar- Figures. In our calculations below, we include domestic gin with respect to GDP per worker, and over- production of models exported to the United States in sales states it with respect to employment. of models exported to the United States. 14 - There are only 7 six-digit categories covering passen ger motor vehicles in the U.N. data, so Japan exported an average of 8 car models to the United States per six-digit car 12 category. This illustrates that a country can be exporting In an earlier draft (Hummels and Klenow, 2002), we used ten-digit data for U.S. imports to disaggregate further more than one variety to a given market in a six-digit and obtain still-larger extensive margins. category, i.e., within-category variety.
12 VOL. 95 NO. 3 HUMMELS AND KLENOW: THE VARIETY AND QUALITY OF A NATION’S EXPORTS 715 T , RICES P DJUSTED -A ,Q UALITY Q MPLY FOR I UANTITIES Q RICES AND P HAT 5—W ABLE UALITY ARIETY ATEGORY W ITHIN -C V AND LL Y / .14 Quality If all quality .23 2.6 5 .07 .16 10 .13 .03 2.6 .13 Prices .14 quality 5 .07 .07 10 .03 .04 .59 If all variety Variety 2.6 .35 5 .33 .82 .29 1.29 10 Prices .00 .09 NA quality Some of each quantity) NA .34 .37 Variety ( .01 price) NA .09 Quality ( Notes: Entries in the last two columns are elasticities with respect to Y / L and L . These are the elasticity of substitution between based on using estimates in Table 3 in equation (7). . L different varieties. NA means independent of 1995 employment in the exporting country relative to the sum of employment in the other 125 exporters. 1995 PPP GDP in Y the exporting country relative to the sum of GDP in the other 125 exporters. Y / L is simply the ratio of these two variables. Sources: UNCTAD for 1995 exports to 59 countries by 126 countries in 5,017 six-digit categories. Heston et al. (2002) for employment and PPP GDP. 5 and 10. These values models ($18,371 versus $7,768). The number of consider cars sold per Japanese model exceeded units sold correspond to markups of 25 percent and 11 per South Korean model by a factor of 1.7 (38,800 percent, respectively. We also entertain versus 22,900). Based on these figures, the 42- 2.6, which is Acemoglu and Ventura’s (2002) required value. The top panel of Table 5 reports percent intensive margin consisted of 26 percent higher prices and 16 percent higher quantities. the implied quality elasticities: countries with Y 5 and twice the tend to export 13 percent to 23 L Using an elasticity of substitution / percent higher-quality varieties, and countries equation (7), we would infer that Japanese models were 2.6 times the quality of South Korean mod- with twice the L tend to export 3 percent to 14 percent higher-quality products. If the quantity els. And we would say Japanese cars had lower margin is increasing in country size, our con- quality-adjusted prices (2.6 times the quality for struction implies that their quality-adjusted 2.4 times the price), explaining their higher unit sales per model. prices must be lower. This is in the spirit of the Armington and Acemoglu and Ventura models. Without similar data on within-category va- riety, we cannot do these calculations for all 0.13 (standard error 0.02), however, the At categories of goods. But we can ask what our quality-adjusted price elasticity with respect to / Y 0.63 required by estimates imply for quality and within-category is a long way from the L variety under particular assumptions. We start 2.6. Acemoglu and Ventura with their by supposing that within-category variety does We next suppose that within-category variety varies with exporter size, but quality does not. not vary with exporter Y / L or L . Then we can The middle panel of Table 5 applies (7) to this use expression (7), our estimates in Table 3, and case. Under the three values for , economies estimates of the elasticity of substitution taken 15 / L with twice the Y from the literature to infer quality variation. export 59 to 129 percent Based on estimates in Hummels (1999), we more varieties per category, and economies export 29 to 35 percent more with twice the Y varieties. These elasticities are large compared 15 to the increase in the extensive margin when This assumes the existence of a single elasticity of substitution, whereas this elasticity surely varies by category. going from 6 digits to 10 digits in the U.S.
13 JUNE 2005 THE AMERICAN ECONOMIC REVIEW 716 import dataset. Moreover, the quality-adjusted markets, conditional on exporting in a category. Interestingly, export destinations relate more price facts are problematic for the “all variety” assumption: if quality does not co-vary with closely to a country’s number of workers (elas- Y / exporter size, why are prices higher for high L ticity 14 percent, standard error 1 percent) than to its income per worker (elasticity 4 percent, exporters? Finally, suppose both quality and within- standard error 2 percent). A model that could match these facts would category variety are a function of exporter size, need three characteristics: firms can produce but but quality-adjusted prices are not. In this case, not export, firms can export to some but not all observed prices perfectly capture quality varia- markets, and the number of markets an exporter tion so the quality elasticities simply equal the reaches should co-vary positively with exporter price elasticities, as is typically assumed in the size. Bernard et al. (2003) use the Ricardian literature. If a country’s exports are representa- model developed by Eaton and Samuel Kortum tive of their production, the implication would (2002) to get extensive margins in trade as a be that quality differences are the proximate cause of 9 percent of differences in across L / Y result of trade barriers and the distribution of countries. Within-category variety is left to ex- productivity. What is not clear to us is whether such a Ricardian model can explain why larger plain the quantity elasticities. As the third panel have larger extensive export of Table 5 shows, the implication would be that economies economies twice the size export about 34 per- margins. Fixed costs of exporting each variety to each cent more varieties within categories. In this case, a hybrid of the Krugman model and the market, as in Romer (1994), combined with quality differentiation model could potentially some Ricardian heterogeneity might explain fit all of the facts. this phenomenon of larger economies exporting to more markets. The logic of such a model A. Fixed Costs of Exporting? might work as follows. Bigger countries pro- duce more distinct varieties because of fixed costs of production (Krugman). Because of As we mentioned while describing the polar models in Section I, the Krugman model has an fixed costs of exporting to each market (Ro- mer), only those varieties with sufficiently low extensive margin because of fixed costs of pro- duction. But the Krugman model has no exten- marginal cost (relative to quality) will be prof- itable to export to a given market. Some desti- sive margin in exports conditional on production. nation markets will have lower thresholds for This does not accord well with the profitable entry than others, say due to variation facts. Andrew B. Bernard et al. (2003) report that most manufacturing plants in the United in their size. In any particular category, a larger country will be more likely to produce at least States do not export, which means that the fail- one variety that can profitably be exported to a ure to export in a category need not imply zero given market. Hence larger countries should be production in that category. Jonathan Eaton et al. (2003) find similar results for French manu- more likely to export to smaller markets in each facturing firms. Further, they find that, condi- category. (Large and small countries alike tional on exporting, firms may export to only a should export to the largest markets.) strict subset of markets. We find similar patterns at the national level in our data. Conditional on Robustness Checks B. exporting in a category, countries export to, on average, fewer than 13 percent of the destina- We carried out a number of checks to see if our results are robust to the sample of countries, tion countries actively importing in that cate- gory. (When we weight destination markets by the sample of goods, and the inclusion of addi- tional covariates. First, we decomposed exports GDP, the mean number of destinations rises to 27 percent.) The larger the economy, however, for a sample of 124 countries exporting to the the more destinations for its exports in each United States in 1995. The U.S. data contain more commodity detail, reporting ten-digit HS category. Across the 125 exporters, those with categories (13,386 categories compared to the twice the GDP tend to export to 11 percent 5,017 in the six-digit U.N. data). We estimated (standard error 1 percent) more GDP-weighted
14 VOL. 95 NO. 3 HUMMELS AND KLENOW: THE VARIETY AND QUALITY OF A NATION’S EXPORTS 717 T RICES AND Y UANTITIES :T OP VERSUS B OTTOM Q / L 6—P ABLE 3 Independent variables 2 Dependent variable Sample Y / LL Adj. R 2 Richest 61 countries Prices 0.39 0.00 0.37 (0.06) (0.02) 0.61 Quantities 0.03 0.39 (0.17) (0.04) Poorest 60 countries 0.05 0.04 0.06 Prices (0.04) (0.02) Quantities 0.39 0.38 0.49 (0.11) (0.05) Notes: All variables are in natural logs. Standard errors are in parentheses. For definitions of L 1995 employment in the the price and quantity components, see equations (11) and (12). exporting country relative to the sum of employment in the other 125 exporters. Y 1995 PPP is GDP in the exporting country relative to the sum of GDP in the other 125 exporters. Y / L simply the ratio of these two variables. Sources: UNCTAD for 1995 exports to 59 countries by 126 countries in 5,017 six-digit categories. Heston et al. (2002) for employment and PPP GDP. ated with higher quantity per variety with no that the extensive margin accounts for 53 per- significant differences in prices. These results cent of the additional exports to the United 16 suggest markedly different patterns of special- States by larger economies. As in the U.N. ization across the samples. In the high sam- L / Y data, the extensive margin is more prominent ple we see results consistent with pure quality for richer exporters (64 percent) than for export- ers with more workers (47 percent). Economies differentiation— higher prices but no higher quantities per variety. In the low Y with twice the Y / L export 47 percent higher L sample, we / see a story consistent with within-category quantities at 13 percent higher prices to the U.S. variety— higher quantity per variety at the same market. Countries with twice the L export 62 prices. percent higher quantities at 5 percent lower Our second set of robustness checks exam- prices. Overall, exports by 124 countries to the ined sensitivity to the goods included in the U.S. tell a broadly similar story to the exports sample. One possibility is that the models of by 121 countries to 59 countries. trade described in Section I apply only to dif- Returning to the larger set of importers, we / split our exporter sample by Y ferentiated products. Many countries may sim- , performing the L ply lack the natural resources to export in regressions separately for the richest 61 and certain commodity categories. To address this poorest 60 exporters. The relative importance of we examined two samples designed to isolate the extensive and intensive margins is quite differentiated goods. First, we included only similar for the top and bottom Y / L samples. More interesting is the behavior of the price and those HS codes that correspond to manufactur- quantity margins, as reported in Table 6. In both ing categories, as defined by Standard Industrial L samples, countries with higher Trade Classification (SITC) categories 5 to 8, export higher quantities per variety with no significant differ- omitting commodity categories 0 to 4. Second, we included only those HS codes belonging to ences in prices. This is very similar to our findings with the pooled sample. However, the James E. Rauch’s (1999) differentiated products / L Y coefficient on four-digit SITCs. The excluded products are does vary across samples. In L the top sample, higher Y / those Rauch classified as reference priced or is associated with 17 higher prices and no greater quantity per vari- Results were traded on organized exchanges. Y / L is associ- ety. In the bottom sample, higher 17 The mapping from six-digit HS codes to Rauch’s 16 version of four-digit SITCs (revision 2) was not perfect. We When calculated at the six-digit level, the extensive margin accounts for 45 percent of the total. could not determine a Rauch classification for about 25
15 JUNE 2005 THE AMERICAN ECONOMIC REVIEW 718 similar using either differentiated-goods classi- at market exchange rates, then summed over fication scheme. markets. Including this crude proxy had no ma- When looking at differentiated goods only, terial effect on any of the exporter size we found much larger overall export elasticities coefficients. Y Y with respect to / and L than we found for the entire sample of goods. A likely explanation is V. Conclusion that the share of differentiated goods in exports Larger economies export more in absolute , and our elasticity picks this up. L / Y is rising in terms than smaller economies. In this paper we However, the contribution of the intensive and decompose a country’s exports into margins extensive margins to the overall export elastic- that account for these differences. We analyze ity was very similar to the full sample of goods, the extent to which larger economies export as was the contribution of the price and quantity components to the intensive margin. higher volumes of each good (the intensive mar- We also explored the robustness of our re- gin), export a wider set of goods (the extensive margin), and export higher-quality goods. sults to the inclusion of additional covariates. It may be that certain margins co-move with cer- Using 1995 trade data for many countries in many product categories, we find that the ex- / Y tain factors that contribute to L more than Y into compo- others. Accordingly, we broke tensive margin accounts for 62 percent of the L / ), human capital ( greater exports of larger economies. Within cat- nents: physical capital ( K / H / L egories, richer countries export more units at TFP L ), and . The relative size of the extensive / L and intensive margins was very similar for higher prices to a given market, consistent with Y producing higher quality. Our estimates imply and each component separately. Point estimates that quality differences could be the proximate on the price and quantity components did reveal cause of around 9 percent of country differences some differences. The standard errors on these in real income per worker. estimates were quite large, however, and the These calculations are useful in distinguish- differences were not statistically significant. ing features of trade models that correspond We next explored whether the results more or less well to the data. Such distinctions changed when we included measures of trade barriers. For the U.S. dataset, we used (total can be extremely important in determining the welfare consequences of access to trade. Arm- total freight)/(nominal exports). We duties ington models of national product differentia- then calculated a trade barrier index for each tion include no extensive margin, and so fail to exporter relative to the rest of the world, aggre- gating over categories in a manner identical to explain the largest margin by which the exports of large and small economies differ. Because the exporter price indices in equation (11). The they lack this margin, these models also imply coefficient on this barrier index was negative that the greater exports of larger economies will and highly significant in all but the price regres- be accompanied by lower export prices. In the sions. Roughly 70 percent of the barrier index’s Acemoglu and Ventura model (2002), these effect on exports was on the intensive margin and, within that, all on the quantity component. terms-of-trade effects result in a stationary world income distribution despite disparate in- When we added this variable as a control, how- ever, none of the coefficients on ,or Y was Y / vestment rates. L , L altered by even one standard error. Krugman-style models with products differ- For the U.N. countries and categories, data on entiated by firms come closer to fitting the facts on intensive/extensive export margins. To tariffs and freight costs are not readily available. In their stead we deployed distance to markets match the positive relationship between prices (and quantities) and exporter income per as a crude proxy for transport costs. For ex- porter j , we calculated distance to market m worker, however, requires modifying these , m weighted by ’s share of world output in 1995 models to include quality differentiation. Also, the simplest Krugman model implies that a country exports each variety to all other coun- tries. In the data, in contrast, countries export to percent of the 5,000 HS categories, so we excluded these as well. a strict subset of actively importing destinations
16 VOL. 95 NO. 3 HUMMELS AND KLENOW: THE VARIETY AND QUALITY OF A NATION’S EXPORTS 719 in most categories. But larger economies export and Kim J. Ruhl (2002) use a Ricardian model to more destinations (conditional on exporting to fit the large extensive component in post- (e.g., after growth trade liberalization in a category), perhaps reflecting fixed costs to 18 A Ricardian model, or even a exporting a variety to each foreign market. NAFTA). We considered only a few models, chosen factor-proportions model, might be constructed because they had the clearest predictions for that could generate covariation between ex- porter size and the extensive margin. We leave how exports should vary with an economy’s size. Other models feature an extensive margin this question for future work. and could perhaps match some or all of the facts we have documented. For example, the Ricard- ian model developed by Eaton and Kortum (2002) predicts variation in the extensive mar- 18 Russell Hillberry and Christine McDaniel (2002) also gin as a function of trade barriers and the dis- document a large extensive margin in post-NAFTA trade growth. tribution of productivity. And Timothy J. Kehoe
17 JUNE 2005 THE AMERICAN ECONOMIC REVIEW 720 A PPENDIX ABLE T A1 PX Y LY / L Country Overall IM EM 0.6652 0.0028 0.0006 0.4629 0.0950 0.00017 ALBANIA 0.0018 0.00027 0.0150 0.0141 0.1825 0.0872 0.00123 ANGOLA 0.00040 0.9452 0.0022 0.01055 1.0298 0.0514 0.3507 0.01804 0.0061 ARGENTINA 0.0499 1.7285 0.01192 3.1791 0.0371 1.0468 0.0388 0.5375 0.02085 AUSTRALIA 0.0038 0.00497 0.0016 3.1426 AUSTRIA 0.01432 0.5004 0.0286 1.3682 0.0209 0.00516 0.00207 0.4208 BANGLADESH 0.0788 0.0263 0.7543 0.0349 0.0123 0.0065 0.0001 1.9531 0.9700 0.0063 0.0145 0.00009 BARBADOS 0.00011 BELGIUM 0.00624 0.0018 3.5045 0.01185 0.4697 0.0252 1.5210 0.0166 0.9913 0.00004 0.0000 1.3108 BELIZE 0.00011 0.0216 0.0052 0.0053 0.0060 0.0044 0.1532 0.0241 0.00011 BENIN 0.00017 0.7453 0.0011 0.0169 0.0012 0.4613 0.5812 0.0098 0.0628 0.00062 BOLIVIA 0.00057 BOTSWANA 0.00025 0.0002 1.2177 0.00005 0.0309 0.0016 0.8325 0.0019 0.9109 0.0519 0.03256 0.0251 1.2948 0.0472 0.4688 0.02215 BRAZIL 0.7101 0.0126 0.00166 0.0017 0.9717 BULGARIA 0.00320 0.3566 0.0090 0.0060 1.1832 0.00025 0.0020 0.1253 BURKINA FASO 0.00010 0.0172 0.0051 0.8316 0.1226 0.0027 0.2614 0.00185 CAMEROON 0.0182 0.0151 0.00072 0.1117 0.1471 3.1747 0.8237 0.12114 CANADA 0.02009 1.3168 0.0063 0.0011 0.0001 0.5753 1.2534 0.0013 0.0095 0.00001 CAPE VERDE IS. 0.00004 0.0056 0.00010 0.0007 0.1598 CENTRAL AFR.R. 0.00007 0.0180 0.0041 0.7375 0.0653 CHAD 0.0010 0.1789 0.00012 0.0019 0.0622 0.9531 0.00018 0.8575 0.00354 0.0023 1.5316 0.0439 0.1630 0.00715 CHILE 0.0512 0.5627 0.2356 0.11010 0.4409 CHINA 0.09336 0.7043 0.1326 0.2497 0.0337 0.0355 0.00622 0.0074 0.8387 COLOMBIA 0.00775 0.2297 0.9497 2.7082 0.0003 0.0001 0.2595 0.00001 COMOROS 0.0093 0.0253 0.00002 0.5932 0.00013 0.0005 0.2490 0.0082 0.0963 0.00079 CONGO 0.0138 0.9932 0.0215 0.00052 0.0005 COSTA RICA 0.00276 0.1294 0.0213 0.9588 0.0038 0.0036 0.00034 0.0001 2.4111 CYPRUS 0.00073 0.1919 1.0713 1.4531 0.0165 0.00363 0.0012 3.0929 0.3556 0.00855 DENMARK 0.0240 0.0022 0.9544 0.0023 0.00001 0.0000 1.0981 0.00002 DOMINICA 0.0109 0.1476 1.0724 0.0221 0.00085 0.0010 0.8239 DOMINICAN REP. 0.00349 0.0237 0.0250 0.00290 0.00130 0.0015 0.8853 ECUADOR 0.8954 0.1160 0.0280 0.7040 0.0224 0.0072 0.8579 0.2460 0.00387 EGYPT 0.0157 0.00617 0.0190 0.0178 0.9317 0.1014 0.00181 EL SALVADOR 0.00069 0.9374 0.0007 0.0177 0.0104 0.0839 1.0100 0.0179 0.0207 0.00037 ETHIOPIA 0.00087 FIJI 0.00012 0.0001 1.0730 0.00041 0.0300 0.0138 1.0772 0.0128 1.6114 0.0156 0.00281 0.0011 2.6613 0.0251 0.2984 0.00750 FINLAND 1.7267 0.0302 0.03616 0.0113 3.2003 FRANCE 0.03479 0.6675 0.0521 0.0194 0.8697 0.00029 0.0002 1.2634 GABON 0.00163 0.0838 0.0223 0.8616 0.0056 0.0002 0.1608 0.00002 GAMBIA 0.0050 0.0043 0.00003 1.6029 0.05288 0.0173 3.0620 0.1096 0.7864 0.08620 GERMANY 0.0684 1.0232 0.0181 0.00064 0.0035 GHANA 0.00118 0.0636 0.0186 0.1834 0.0091 0.0134 0.00383 0.0018 2.1361 GREECE 0.00109 0.1194 0.6785 0.9074 0.0023 0.00001 0.0000 0.7563 0.0059 0.00001 GRENADA 0.0021 0.0199 0.5810 0.0342 0.00110 GUATEMALA 0.9171 0.00310 0.1559 0.0012 0.0339 0.6755 0.0160 0.00051 0.0013 0.3796 GUINEA 0.00037 0.0108 0.0037 0.9338 0.00002 0.0002 0.1172 0.00004 GUINEA-BISS 0.0102 0.0039 0.7461 0.00006 0.0001 0.4984 0.0117 GUYANA 0.00027 0.0232 0.0157 0.8935 0.0031 0.00031 0.0013 0.2428 HAITI 0.00016 0.0593 0.0027 0.0210 0.9313 0.00034 0.0007 0.4626 HONDURAS 0.00188 0.0893 0.0226 0.0340 0.0430 0.00464 0.0013 3.5626 0.5653 0.01922 HONG KONG 0.7898 0.0230 0.8746 0.0263 0.00259 0.0017 1.4888 HUNGARY 0.01208 0.5248 0.0543 0.0264 0.0250 0.00016 0.0001 2.6059 ICELAND 0.00143 1.0553 0.0261 0.01167 0.05684 0.1802 0.3155 INDIA 0.7478 0.4468 0.0349 0.0424 0.0495 0.02110 0.0331 0.6378 0.4510 0.01913 INDONESIA 0.8574 0.0662 0.9334 0.0710 0.00871 IRAN 1.1735 0.01111 0.1676 0.0074 0.2553 2.6907 0.0076 0.00182 0.0006 3.1197 IRELAND 0.00520 0.0204 0.0175 1.7373 0.0101 0.00263 0.0009 2.9943 0.00704 ISRAEL 0.4028
18 VOL. 95 NO. 3 HUMMELS AND KLENOW: THE VARIETY AND QUALITY OF A NATION’S EXPORTS 721 A1— Continued. T ABLE EM Country / L IM Overall PX Y LY 0.0508 0.03330 0.03520 0.0098 3.6090 ITALY 1.4725 0.6559 0.0345 0.9943 0.1723 0.0022 0.3395 0.00343 IVORY COAST 0.0200 0.0199 0.00076 0.0189 0.0176 0.5353 0.0643 0.00114 JAMAICA 0.00028 0.9345 0.0005 0.1775 0.0346 2.7007 1.5742 0.2795 0.7245 0.20249 JAPAN 0.09345 0.0079 0.00046 0.0004 1.1307 JORDAN 0.00036 0.0630 0.0057 0.7170 0.0171 KENYA 0.0056 0.1775 0.00116 0.0698 0.0166 0.9699 0.00100 0.7441 0.00008 0.0004 0.2008 0.0037 0.0223 0.00008 LESOTHO 0.0050 1.3189 0.0084 0.00042 0.0001 LUXEMBOURG 0.00032 0.0291 0.0110 5.5689 0.0063 0.0072 0.00026 0.0004 0.7295 MACEDONIA 0.00095 0.1510 0.8711 1.0592 0.0701 0.0024 0.1315 0.00062 MADAGASCAR 0.0083 0.0088 0.00032 0.8715 0.00019 0.0018 0.1096 0.0113 0.0209 0.00024 MALAWI 0.0130 0.7632 0.0773 0.00527 0.0031 MALAYSIA 0.03207 0.5439 0.0590 1.7290 0.0081 0.0087 0.00023 0.0020 0.1175 MALI 0.00009 0.0112 0.9216 1.0070 0.0068 0.00014 0.0001 2.3851 0.1758 0.00120 MALTA 0.0068 0.0283 0.7817 0.0362 0.00009 MAURITANIA 0.1827 0.00053 0.0189 0.0005 0.0710 1.2357 0.0129 0.00037 0.0002 1.7430 MAURITIUS 0.00113 0.0160 0.0728 0.05416 0.01945 0.0133 1.4661 MEXICO 0.9614 0.7441 0.0757 1.0825 0.2582 0.0035 0.7505 0.00703 MOROCCO 0.0251 0.0272 0.00265 1.0780 0.00037 0.0033 0.1143 0.0076 0.0197 0.00015 MOZAMBIQUE 0.0070 1.1762 0.0132 0.00012 0.0013 MYANMAR 0.00028 0.0179 0.0155 0.0930 0.0145 0.0119 0.00020 0.0002 1.0254 NAMIBIA 0.00069 0.0479 1.2190 0.8302 0.0078 0.00077 0.0037 0.2065 0.0366 0.00024 NEPAL 0.0065 0.0244 1.3072 0.0186 0.00953 NETHERLANDS 3.1340 0.01405 0.5770 0.0030 0.2317 1.1116 0.0268 0.00187 0.0007 2.5738 NEW ZEALAND 0.00689 0.0297 0.0076 0.00038 0.00024 0.0006 0.4067 NICARAGUA 0.9370 0.0503 0.0081 0.0031 0.0029 0.00022 0.0019 0.1185 0.0454 0.00014 NIGER 1.0653 0.0522 1.0612 0.0492 0.00302 NIGERIA 0.1322 0.00753 0.1442 0.0229 0.6081 1.0653 0.0629 0.00305 0.0009 3.3580 NORWAY 0.04077 0.0670 0.0252 0.8230 0.0306 0.00686 0.0146 0.4710 0.00323 PAKISTAN 0.1280 0.1128 PANAMA 0.9338 0.0202 0.00043 0.0004 1.0565 0.00212 0.0188 0.00161 0.0435 1.0555 0.0412 0.00045 0.0009 0.5245 PAPUA N.GUINEA 0.0369 0.0435 0.0342 0.0418 0.00076 0.0009 0.8515 PARAGUAY 0.00149 0.8176 0.0212 0.5710 0.00306 0.0043 0.7105 0.00227 PERU 0.1072 0.0371 0.8916 0.00626 0.0120 0.5235 0.0211 PHILIPPINES 0.00748 0.3538 0.0237 0.6757 0.0475 0.00829 0.0073 1.1401 POLAND 0.01859 0.5796 0.0321 0.0080 0.7786 0.00379 0.0019 2.0199 PORTUGAL 0.00185 0.2320 0.0102 0.0132 0.0194 0.00315 0.0048 0.6614 0.3539 0.00467 ROMANIA 0.6788 0.0626 0.7140 0.0876 0.03207 0.0316 1.0150 RUSSIA 0.02952 0.4717 0.0132 0.0038 0.0039 0.00014 0.0013 0.1048 RWANDA 0.00005 0.9657 0.0122 0.00056 0.00035 0.0017 0.2120 SENEGAL 0.8845 0.0463 0.0138 0.0064 0.0070 0.00002 0.0000 1.4883 0.0054 0.00003 SEYCHELLES 0.9198 0.0023 0.9766 0.0024 0.00012 SIERRA LEONE 0.1706 0.00005 0.0200 0.0007 0.5684 1.2248 0.0452 0.00234 0.0009 2.7300 SINGAPORE 0.03144 0.0553 0.0245 0.7214 0.0340 0.00149 0.0011 1.4020 0.01188 SLOVAK REPUBLIC 0.4845 0.4371 0.0175 0.9052 0.0194 0.00074 0.0004 1.8846 SLOVENIA 0.00766 0.01276 0.4025 0.9216 0.0344 0.00833 0.0056 1.4882 SOUTH AFRICA 0.0317 0.6480 SOUTH KOREA 0.0612 0.01820 0.0080 2.2864 0.0639 0.04143 1.0438 0.0208 0.00879 0.01909 0.0066 2.8861 SPAIN 1.2794 0.4232 0.0162 0.0114 0.0114 0.00163 0.0032 0.5121 0.1271 0.00145 SRI LANKA 0.9960 0.0088 1.0039 0.0088 0.00002 ST.VINCENT&GRE 1.0275 0.00006 0.0068 0.0000 0.4803 1.7166 0.0193 0.00537 0.0019 2.7783 SWEDEN 0.01593 0.0332 0.0754 2.0872 0.0361 0.00508 0.0016 3.0915 0.04794 SWITZERLAND 0.6361 0.1377 0.0246 0.9140 0.0269 SYRIA 0.0015 1.1121 0.00339 0.00162 0.04611 0.0719 0.7758 0.0927 0.00926 0.0039 2.3607 TAIWAN 0.6414 0.0495 TANZANIA 0.0080 0.00040 0.0059 0.0681 0.0074 0.00037 0.9282 0.5016 0.8367 0.0425 0.01189 0.0134 0.8865 0.01784 THAILAND 0.0356 0.0244 0.0040 0.6217 0.0065 TOGO 0.0007 0.1513 0.00010 0.00011 0.00134 0.0158 0.7757 0.0203 0.00034 0.0002 1.6386 TRINIDAD&TOBAGO 0.0847 0.3365 0.0190 1.1116 0.0171 0.00145 0.0012 1.1681 TUNISIA 0.00640
19 JUNE 2005 722 THE AMERICAN ECONOMIC REVIEW ABLE A1— Continued. T IM L / PX Y LY EM Overall Country 0.01449 0.0354 0.0378 0.9373 0.4092 0.01114 0.0114 0.9807 TURKEY 0.0123 0.03478 0.0291 1.4843 0.0432 0.7645 0.03305 U.K. 2.8229 0.0594 4.7114 0.27972 0.2500 1.3582 0.3395 0.9121 0.30966 U.S.A. UGANDA 0.0211 0.8922 0.0250 0.00053 0.0237 0.00045 0.0039 0.1153 0.1257 0.0227 1.0327 0.0220 0.00083 0.0006 1.3687 0.00286 URUGUAY 0.8223 0.01382 0.2073 0.0667 1.4221 0.0811 0.00455 0.0032 VENEZUELA YEMEN 0.00047 0.0439 0.0107 0.8394 0.0127 0.00039 0.0016 0.2472 0.5389 ZAIRE 0.00132 0.0939 0.0140 0.0260 0.00041 0.0091 0.0451 0.0013 0.00021 0.0159 0.9948 0.0158 0.0253 0.00040 ZAMBIA 0.1693 ZIMBABWE 0.0126 1.0044 0.0125 0.00085 0.0022 0.3795 0.0966 0.00122 0.01345 0.2282 0.0093 1.0119 0.0294 0.0087 mean 1.2476 0.0299 Notes: All variables are for 1995. Overall EM*IM. EM Extensive margin, a cross-market geometric average of equation (8) in the text. IM Intensive margin, a cross-market geometric average of equation (9) in the text. P Prices, as cross-market geometric average of equation (11) in the text. Quantity margin, a cross-market geometric average of equation (12) in the text (IM X P * X ). Y Country GDP relative to the rest of the world. L Country employment relative to the rest of the world. Y / L Country GDP per worker relative to rest-of-world GDP per worker. The variables are shown here in levels; for the regressions, the variables were logged. REFERENCES Evidence from French Firms.” Unpublished Paper, 2003. “Internationally Decreasing Ethier, Wilfred J. Acemoglu, Daron and Ventura, Jaume. “The World Income Distribution.” Costs and World Trade.” Journal of Interna- Quarterly Jour- tional Economics 117 (2), pp. 659 –94. , 2002, (1), pp. 1–24. 9 , 1979, nal of Economics “National and International Armington, Paul S. “A Theory of Demand for Ethier, Wilfred J. Products Distinguished by Place of Produc- Returns to Scale in the Modern Theory of tion.” American Economic International Trade.” International Monetary Fund Staff Pa- , 1969, 16 (1), pp. 159 –78. (3), pp. 389 – 405. 72 , 1982, Review pers Bernard, Andrew B.; Eaton, Jonathan; Jensen, J. Feenstra, Robert C. “New Product Varieties and Bradford and Kortum, Samuel. “Plants and the Measurement of International Prices.” Productivity in International Trade.” (1), pp. American American Economic Review , 1994, 84 Economic Review 157–77. (4), pp. 1268 –90. 93 , 2003, “Vertical Brown, Drusilla K. “Tariffs, the Terms of Trade, Flam, Harry and Helpman, Elhanan. Product Differentiation and North–South and National Product Differentiation.” Jour- Trade.” American Economic Review (3), pp. 503– 9 nal of Policy Modeling , 1987, , 1987, 77 26. (5), pp. 810 –22. Davis, Donald R. and Weinstein, David E. “Product “Tech- Funke, Michael and Ruhwedel, Ralf. Variety and Economic Growth: Empirical nological Superiority and the Losses from Evidence for the OECD Countries.” Migration.” National Bureau of Economic IMF (2), pp. 225– 42. 48 , 2001, Staff Papers Research, Inc., NBER Working Papers: No. Grossman, Gene M. and Helpman, Elhanan. In- 8971, 2002. . novation and growth in the global economy “Technol- Eaton, Jonathan and Kortum, Samuel. ogy, Geography, and Trade.” Cambridge, MA: MIT Press, 1991. , Econometrica (5), pp. 1741–79. 70 2002, Head, Keith and Ries, John. “Increasing Returns versus National Product Differentiation as an Eaton, Jonathan; Kortum, Samuel and Kramarz, Francis. Explanation for the Pattern of U.S.–Canada “An Anatomy of International Trade:
20 VOL. 95 NO. 3 HUMMELS AND KLENOW: THE VARIETY AND QUALITY OF A NATION’S EXPORTS 723 , 2001, Trade.” American Economic Review “Scale Economies, Product Krugman, Paul R. (4), pp. 858 –76. Differentiation, and the Pattern of Trade.” 91 70 Heston, Alan; Summers, Robert and Aten, Bet- (5), pp. , 1980, American Economic Review tina. 950 –59. . Center Penn World Table Version 6.1 Krugman, Paul R. for International Comparisons: University of “Intraindustry Specializa- tion and the Gains from Trade.” Journal of Pennsylvania, CICUP, 2002. , 1981, “A (5), pp. 959 – 89 Political Economy Hillberry, Russell and McDaniel, Christine. 73. Decomposition of North American Trade Rauch, James E. “Networks versus Markets in Growth since NAFTA.” International Eco- Journal of International International Trade.” nomic Review , USITC Publication No. 3527, 2002, (May/June), pp. 1– 6. Economics , 1999, 48 (1), pp. 7–35. “Toward a Geography of Romer, Paul. Hummels, David. “New Goods, Old Theory, and the Welfare Costs of Trade Restrictions.” Jour- Trade Costs.” Unpublished Paper, 1999. Hummels, David and Klenow, Peter J. “The Va- nal of Development Economics (1), 43 , 1994, riety and Quality of a Nation’s Trade.” Na- pp. 5–38. tional Bureau of Economic Research Inc., “Across-Product versus Within- Schott, Peter K. International in Specialization NBER Working Papers: No. 8712, 2002. Product Kehoe, Timothy J. and Ruhl, Kim J. “How Im- , Quarterly Journal of Economics Trade.” 2004, (2), pp. 647–78. 119 portant Is the New Goods Margin in Interna- tional Trade?” Unpublished Paper, 2002. United Nations Conference on Trade and Devel- opment. “Increasing Returns, Monop- Krugman, Paul R. “Trade Analysis and Information System (TRAINS),” 1995 CD-ROM. olistic Competition, and International Trade.” Journal of International Economics U.S. Bureau of the Census. “U.S. Imports of , 1979, Merchandise,” 1995 CD-ROM. 9 (4), pp. 469 –79.
American Economic Review 2013, 103(6): 2121–2168 http://dx.doi.org/10.1257/aer.103.6.2121 The China Syndrome: Local Labor Market Effects † of Import Competition in the United States By David H. Autor,...More info »
United States International Trade Commission Small and Medium- Sized Enterprises: Characteristics and Performance Investigation No. 332-510 USITC Publication 4189 November 2010More info »
The Rise of Offshoring: It’s Not Wine for Cloth Anymore Gene M. Grossman and Esteban Rossi-Hansberg Introduction In 1817, when David Ricardo penned his celebrated treatise on The P rinciples of Politi...More info »
EC08CH08-Hanson ARI 29 September 2016 16:27 ANNUAL Further REVIEWS to view this article's Click here online features: • Download figures as PPT slides • Navigate linked references The China Shock: Lea...More info »