The Revealed Competitiveness of U.S. Exports

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 1026 August 2011 The Revealed Competitiveness of U.S. Exports Massimo Del Gatto Filippo di Mauro Joseph Gruber Benjamin R. Mandel NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References to International Finance Discussion Papers (other than an acknowledgement that the writer has had access to unpublished material) should be cleared with the author or authors. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from Social Science Research Network electronic library at www.ssrn.com.

The Revealed Competitiveness of U.S. Exports* Massimo Del Gatto (G.d'Annunzio University and CRENoS) Filippo di Mauro (European Central Bank) Joseph Gruber (Federal Reserve Board) Benjamin R. Mandel (Federal Reserve Board) Abstract The U.S. share of world merchandise exports has declined sharply over the last decade. Using data at the level of detailed industries, this paper analyzes the decline in U.S. share against the backdrop of alternative measures of the competitiveness of the U.S. economy. We document the following facts: (i) only a few industries contributed to the decline in any meaningful way, (ii) a large part of the drop was driven by the changing size of U.S. export industries and not the size of U.S. sales within those industries, (iii) in a gravity framework, the majority of the decline in the U.S. export share within industries was due to the declining U.S. share of world income, and (iv) in a computed structural measure of firm productivity, average U.S. export productivity has generally maintained its high level versus other countries over time. Overall, our analysis suggests that the dismal performance of the U.S. market share is not a sufficient statistic for competitiveness. Keywords: Trade competitiveness, gravity model, firm productivity JEL classification: F14, F17 * Division of International Finance, Board of Governors of the Federal Reserve System, Washington, D.C. 20551 U.S.A. m.delgatto@unich.it; Filippo.diMauro@ecb.int; Joseph.W.Gruber@frb.gov; Benjamin.R.Mandel@frb.gov. The authors thank Brian Andrew for excellent research assistance. The views in this paper are solely the responsibility of the authors and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System, any other person associated with the Federal Reserve System, or the European Central Bank.

The Revealed Competitiveness of U.S. Exports∗ Massimo Del Gatto Filippo di Mauro Joseph Gruber Benjamin R. Mandel This version: March 18, 2011 G.d’Annunzio University and CRENoS. m.delgatto@unich.it. European Central Bank. Filippo.diMauro@frb.gov Federal Reserve Board. Joseph.W.Gruber@frb.gov. Federal Reserve Board. Benjamin.R.Mandel@frb.gov. 1 Introduction The U.S. share of world merchandise exports has declined sharply over the last decade. Using data at the level of detailed industries, this paper analyzes the decline in U.S. share against the backdrop of alternative measures of the competitiveness of the U.S. economy. Usual suspects for a given country’s decline in export share might include: unfavorable relative price movements, crowding out from the proliferation of low-cost exporters from developing countries, uneven reductions in trade costs and barriers around the world, or possibly the deterioriating productivity of exporting firms compared to foreign rivals. Disentangling these factors presents several complications. First, relative prices are only weakly correlated with U.S. market share, and are thus not very helpful in explaining its recent dynamics. This is evidenced by the accelerating drop in share during the 2000’s amidst a decline in the value of the broad real dollar. Second, in many instances, and particularly for international comparisons, trade costs and firm productivity are difficult to measure directly.1 And third, export shares may additionally reflect the idiosyncratic composition of the U.S. export bundle, which may have little to do with the ability of U.S. exporters within a given industry to compete. To tackle these issues, Section 2 begins by decomposing the decline in share into detailed industry groups; we find that only a few of these industries contributed to the decline in any meaningfulway. Moreover,alargepartofthedropwasdrivenbythechangingsizeofU.S.export industries and not the size of U.S. sales within those industries. This means that U.S. exporters arespecializedinindustriesthathappentohavebeengrowingrelativelyslowlyasashareofworld trade. These observations offer our first suggestion that the fall in aggregate U.S. share has little to do with the underlying productivity of U.S. exporting firms. ∗The authors thank Brian Andrew for excellent research assistance. The views in this paper are solely the responsibilityoftheauthorsandshouldnotbeinterpretedasreflectingtheviewsoftheBoardofGovernorsofthe Federal Reserve System, any other person associated with the Federal Reserve System, or the European Central Bank. 1Measures of aggregate tfp, comparable across countries, are usually obtained as the residual component of GDP growth that cannot be explained by the growth of production inputs. One of the drawbacks of the growth accountingapproachisthattheroleofthesectoralcompositionofoutputisruledoutbyassumption. Byassuming that GDP is produced by a single sector, one cannot disentangle tfp differences (across countries) due to sectoral specializationfromtfpdifferencesduetootherfactors. 1

Wethenpresenttwomeasuresoftradecompetitivenesswhich,insofarastheyareinferredfrom actual trade flows, we refer to as revealed competitiveness. The first measure, in Section 3, is derived as a residual from a standard gravity equation. The objective of the exercise is to purge bilateral trade flows of the effect of national income and geography, wherein the residual contains information about the relative productivity and unmeasured trade costs of exporters. We find that the majority of the decline in the U.S. export share is in fact due to the declining share of U.S. income in the world. The residual, which we view as a ‘purer’ measure of competitiveness, is declining but not as dramatically. Our second approach, in Section 4, is derived from a structural model and builds on the multi-country, multi-sector version of Melitz-Ottaviano (2008).2 In that framework, the overall competitiveness of a country in a given sector is the outcome of a process of firm selection driven by: (1) the degree of ’accessibility’ (i.e. trade costs) of the country and the size of its domestic market, as well as (2) the exogenous ability of the country to generate low cost firms, which depends on structural and technological factors. We extend previous empirical applications of that model by using richer product level detail, and additionally employ an innovative approach byNovy(2009)tocomputecompetitivenessindicatorswhicharecomparableovertime. Consistent with our gravity residual exercise we find that, notwithstanding significant heterogeneity across sectors,U.S.exportproductivityhasgenerallymaintaineditshighlevelversusothercountriesover time. Overall, our analysis suggests that the dismal performance of the U.S. market share is not a sufficient statistic for competitiveness. 2 The state of U.S. export share From 1980 to 2009 the U.S. share of world exports fell by almost one third, declining from about 11 percent to just over 8 percent of world exports. In this section we examine the decline in the U.S. share using NBER-UN bilateral trade data from Feenstra, Lipsey, Deng, Ma and Mo (2005). As shown in Figure 1, the United States’ share of world merchandise exports rose slightly from 1986 to 1999, increasing from about 101 to 121 percent of world exports, before falling 4 2 2 percentagepointsbetween1999and2009. Thebilateraltradedatarunthrough2004and,inFigure 2, weobservethateveryindustrygroupatSITC1-digitaggregationregisteredadecreaseoverthe period from 1984 to 2004, with many of the larger changes occuring in the early 2000’s. The largest declines in share were recorded among the basic materials categories (SITC 0 through 4), which account for approximately 25 percent of U.S. exports, and in machinery and transportation equipment (SITC 7), which account for almost half of U.S. export sales. It is interesting to note thatthetimingofthedeclineinU.S.sharediffersoverSITCcategories. Thefallinbasicmaterial sharesisgradualandpersistent,whiledeclineinmachinery&transportationequipmentisabrupt, primarily occurring after 1999. The decline in market share is machinery and transportation equipment is particular notable giventheimportanceofthissectorforoverallU.S.exports. ThefallintheU.S.shareofmachinery andtransportationequipmentisexaminedfurtherinTable1,whichbreaksthecategoryintoSITC 2-digitsubcategories. AlthoughthedeclineinU.S.shareisapparentacrossmost2-digitmachinery categories,thefallintheU.S.shareofofficemachineandcomputerexportsisparticularlystriking, with U.S. share of world exports falling from a third of the total to just under one tenth. As with overallexports,thereissomedispersioninthetimingofthedeclineinsharesacrosssubcategories. Whereas the fall in computers is steep and steady over the entire period, in most other categories of machinery the U.S. was able to maintain or expand export share through 1999 before shares plummeted sharply. 2ThatmodelwasfirstbroughttothedatabyDelGatto,MionandOttaviano(2006)andfurtherdevelopedby Ottaviano,TaglionianddiMauro(2009). 2

Amoremeaningfulwayofdecomposingthedeclineintheaggregateexportshareistocompute the appropriately weighted contribution from disaggregate categories of goods. The change in aggregate export share can be expressed as the sum of changes across product categories (i) as a ratio of the change in world exports: ∆X US = (cid:88) ∆X U i S ∆X ∆X WORLD WORLD i Figure 3 depicts the contributions to the change in aggregate export share for each 1 digit SITC code over the period from 1984 to 2004. Food & live animals provided the largest contribution to the decline in share, accounting for almost one fourth of the aggregate decline. Almost as large were the contributions of machinery & transportation and crude materials, also each contributing about one fourth to the overall decline in share. The importance of raw materials for the decline inU.S.shareraisesanoteofcautionininterpretingaggregateexportsharestatistics. Commodity pricesfellovermosttheperiodunderconsideration,andsincetheexportsoftheUnitedStatesare relatively commodity intensive, so did the U.S. share of world exports. The importance of commodities is further illustrated in Figure 4, which depicts the top 10 contributors to the aggregate decline among 4-digit SITC codes. Corn and soybeans contribute a combined one sixth of the overall decline. However, the 4-digit data also reveals that a number of categoriesofmanufacturedgoodsalsocontributedtothedecline,includingmotorvehiclepartsand digitalprocessingunits(computers). Thetakeawaymessageisthatatruemeasureofdevelopments in U.S. competitiveness is more likely to be found by looking at U.S. export performance within relatively narrowly defined categories. The importance of foods for the explaining the overall decline in U.S. share is somewhat surprising given foods relatively small share in U.S. exports and, as shown in Figure 2, the lack of an abnormally large fall in the U.S. share of food specific exports. However, it is important to note that the contribution of each individual category to the fall in the U.S. aggregate share occurs along both an intensive and an extensive margin. The decline in the U.S. aggregate share reflects both an intensive decline in market share within each category, as well as an extensive decline stemming from changes in the size of each category in world exports. For instance, corn (SITC0440)contributestothedeclineinU.S.aggregatesharebothastheU.S.capturesasmaller proportion of the corn-specific export market and also as corn’s share of overall world exports declines. Oneestablishedmethodofassessingtheimportanceofcompositionforchangesintradeshares is constant market share analysis (see ECB (2005) for a detailed description).3 Constant market share analysis separates changes in aggregate market share into two components, a commodity effect and competitiveness effect defined as follows:4 ∆Xi (cid:88) Xi (cid:18) Xi (cid:19) (cid:88) (cid:18) Xi (cid:19) Xi US = US . ∆ WORLD + ∆ US . WORLD ∆X Xi X Xi X WORLD i WORLD WORLD i WORLD WORLD (cid:124) (cid:123)(cid:122) (cid:125) (cid:124) (cid:123)(cid:122) (cid:125) Commodity Effect Competitiveness Effect The commodity effect measures the effect of composition on the change in the aggregate export share, by weighting the change in the composition of world exports by the initial composition of 3Constantmarketshareanalysisisbesetbyanumberofwelldocumentedtheoreticalproblems(seeRichardson (1971)foranoverview). However,theapproachremainsillustrativeandsimpletoimplementevenifinterpretation iscomplicatedbyrelativepricechangesandotherissues. 4The constant market share approach often includes an additional “market effect” related to the geographical pattern of trade. For ease of exposition we have focused only on the commodity effect, in a sense wrapping the market effect into our measurement of the competitiveness effect. With declining trade costs it is likely that the marketeffecthasbecomealesspronounceddeterminateofaggregateshareinanycase. 3

the U.S. export bundle. The competitiveness effect measures the portion of the change in the aggregate share that is due to changes in the within category share of U.S. exports. Figure 5 decomposes the contribution of each 1-digit SITC export category to the change in the aggregate export share (the blue bars) into components due to commodity (the green bars) and competitiveness (the red bars) effects over the 1984 to 2006 period. The large negative contributionsoffoodandliveanimalsandcrudematerialslargelyreflectthedecliningimportance of these goods in world exports (signified by negative commodity effects), although U.S. exports alsosufferedanegativecompetitivenesseffectineachcase. Incontrastthenegativecontributionto the aggregate recorded by the machinery and transportation sector is completely due to a decline in U.S. competitiveness, as the sector has greatly increased its weight in the world exports over the time frame under consideration. In summary, interpreting the decline in the U.S. export share is complicated by compositional effects. The primary drivers of the decline in aggregate U.S. share were raw commodities, with negative contributions that largely derived from their declining weight in the world export basket. Thatsaid,theU.S.didexperiencealargedeclineintheshareinthemachineryandtransportation sector, which was not reflected in the composition of U.S. exports but rather declines within detailed sub-categories. Here the evidence of a fall in U.S. competitiveness is more compelling. Againstthisbackground,thefollowingsectionsfocusonU.S.exportperformancewithinindustries and attmept to identify its drivers. We suggest two alternative empirical methodologies to parse out a narrower definition of competitiveness: exporter productivity purged of geographical and relative market size considerations. This strategy is termed, ”revealed competitiveness,” which derives from the fact that it is inferred from observable trade flows. 3 Reduced Form Revealed Competitiveness Onepossibleexplanationfor the declineinU.S.exportshare issimplythattheU.S. nowaccounts for a smaller share of global output. As China and other emerging economies expand rapidly and become more integrated into the global economy, it is natural that the U.S. share of world exportswouldfallwithoutnecessarilyindicatinganydeclineintheproductivityofU.S.exporters. As shown in Figure 6, the fall in the U.S. share of global exports of about 4 percentage points over the past decade corresponds to a decrease in the U.S. share of global output of about 3 1/2 percentage points on a PPP basis. The relatively tight correlation between export share and income holds true for many other countries as well. Across the G7, France, Italy, and Japan have experienceddeclinesinexportsharethatbroadlymatchtheirdecliningshareofworldoutput. On the other hand, Germany has more less maintained export share even as its share of world income hasdeclined,whileCanadaandtheUKhavesufferedmuchsteeperfalloffsinexportsthanincome. In percentage terms, the export share growth of China has outpaced its income share, and the same holds for India. Figure 6 strongly suggests that changes in market share may be conflating competitiveness effects with income dynamics. Specifically, country characteristics such as size may be influencing market share but have little to do with the underlying ability of a country’s exporters to compete. Tocontrolforsuchcharacteristics,ourfirstapproachistonon-parametricallyestimatetradeflows minus the contribution of country size, geography and certain trade costs. A derivative of the gravity equation is a natural candidate to do so. Previous studies such as Baier and Bergstrand (2001),andmorerecentlyWhalleyandXin(2009)andNovy(2009),usegravitytodecomposethe levels of bilateral trade flows into contributions from income, trade costs or otherwise. Each finds that exporter and importer income plays a substantial, even dominant, role in explaining trade. Our approach extends this logic to the case of relative trade performance, where the gravity equation is ‘folded’ by dividing through by a reference exporter. In the particular case where the reference country is the entire world, the gravity equation converts neatly into an expression 4

for market share in terms of relative exporter size, relative geographic characteristics and relative productivity. Ourapproachto‘decomposing’theshareintofactorsthathavetodowithcompetitivenessand those that don’t involves simply looking at the time variation in the residual of a panel gravity estimation. The intuition is that if a country is increasingly outperforming the average exporter’s performance (i.e., a country exports more relative to its own size and more to distant countries over time) then its residual will grow over time. We posit that this residual contains information about changes in the underlying productivity of exporters. In this section, we do not apply a structuralinterpretationtothatproductivity,itismerelycontainedintheresidual. Inthesection that follows, we apply a structure that allows for more specific interpretation of the residual and, moreover, is consistent with the reduced form gravity equation herein. To be concrete, define Tlh as country l’s exports to country h in sector s in a given period t: s Tlh =DlDhrl rhρlhφ (1) st t t st st s t Equation(1)correspondstoagenericgravitymodel, wherebilateraltradeisafunctionofcountry size(D),latentcountry-specificmultilateralresistance(r),geographiccharacteristics(ρ)andglobal shocks (φ). Exploiting the multiplicative form of the equation, we cancel out importer-specific terms by dividing through by total exports to country h in industry s. Tlh Dlrl ρlh st = t st s (2) (cid:80) Tlh (cid:80) Dlrl ρlh l st l t st s The intuition for this reduced form is that the change in a given importer’s income or multilateral resistancewillaffectthelevelofthatcountry’simportsbutnothowthenewimportsareallocated across exporters. Moreover, a global shock affecting all exporters will not affect their relative performanceandhencetheφtermscanceloutaswell. Themethodoftakingratiosofthegravity equationhasthreeostensiblebenefits. First,forourpurposeofrelatingtheshareofU.S.exportsto underlying productivity measures, equation (2) is expressed in the correct units of share owing to income, tradecostsandproductivity. Second, thesizeofthedatamatrixusedintheestimationis reducedbyfoldingintheimporter-specificterms. Third, multilateralresistanceterms(asdefined in Anderson and van Wincoop (2003)) associated with importers cancel out, sparing the need to approximate them using fixed effects.5 Denoting the geometric mean of a given variable by X = (cid:81) l Xn 1, taking logs, and allowing for a mean-zero perturbation (ε), we can rewrite the above expression as:6 Tlh 1 Dl ρlh rl ln st =ln +ln t +ln s +ln st +εlh (3) (cid:80) Tlh n D ρh r st l st s t s st The log of country l’s market share in destination market h is a positive function of its relative income, its geographic proximity and its relative productivity. Again, this specification is isomorphic to a standard gravity model, though specified in relative terms. With the additional assumptions that ρ and n are constant over time, variation in exporter multilateral resistance and productivityisidentifiedastheresidualofamodelwithexporterrelativeincomeandcountry-pair fixedeffectsontheright-handside. Thatis,theactualmarketsharechangesovertimerelativeto the changes in the gravity model prediction contains information about the evolution of relative 5Other examples of cancelling out the importer fixed effects in a gravity framework include: Head and Mayer (2000),Martin,MayerandThoenig(2008)andHead,MayerandRies(2010). 6Expression(3)imposesseparabilityacrossright-handsideratioswiththeassumptionthatln(cid:80)T =(cid:80)lnT. In practice,thismayhavetheeffectofoverestimatingtheshareofeachexporter(i.e.,sincethesharesasdecomposed ontheright-handsidewilladduptomorethan1),butlittleimpactontherelativesizeoftheshares. 5

exporter productivity. This is what we will refer to as the share-to-model ratio, or simply the growth in the model residual by exporting country, averaged across industries: (cid:32) (cid:92) (cid:33) rl Tlh Tlh (cid:52)ln t = (cid:52)ln st −(cid:52)ln st (4) r (cid:80) Tlh (cid:80) Tlh t l st l st The assumption of a time-invariant ρ is similar to standard gravity approaches using variables such as distance, common border and common language that don’t tend to change much. The implication of this assumption, however, is that decreases in trade costs due to changing trade policy will also be captured in the residual term. In our implementation we add dummies for significant shifts in policy (e.g., NAFTA, EMU) as well as over the course of our sample to try to control for changing trade costs, but nonetheless the residual likely captures elements of falling tradecostsinadditiontorelativeproductivity. Assuch,inthissectionwejointlyestimaterelative performance due to these factors, both of which fit into a reasonable, if broad, definition of export competitiveness; in the following section we use a structural model to parse the residual more finely. Theassumptionofaconstantnumberoftradingpartnersperimporter(n)maybelessbenign. Due to the seminal work of Feenstra (1994), there has been much study of the increase in product varietywithinevennarrowlydefinedproductcategories. Weaddressthisempiricallyintwoways. First, every specification below contains time fixed effects which would soak up a secular trend in varieties. Secondly, most specifications contain country-pair fixed effects or exporter-time fixed effectswhichwouldpickupatleastaportionoftheleveldifferencesinnbycountry. Wenotethat a disproportionate rise in relative product variety to certain countries over time would decrease thetermln(1/n)andhenceworkagainstthefindingofrisingproductivityintheresidual. Forthe mostprolifictradersintermsoftheirnumberoftradingpartners,whichincludestheU.S.,wethus take our estimates of the rising residual as an underestimate of the true change in productivity and trade costs. 3.1 Reduced form revealed competitiveness: data & specification The data used in the estimation are bilateral trade flows as described in the previous section, nominal GDP data from the Penn World Table which are converted into international dollars at PPPexchangerates,dummyvariablesforNAFTA,EUandEMItradeflows,thedistancebetween capital cities, as well as common border and common language dummies. We follow previous studies by truncating the data at $10,000 per annual bilateral flow to avoid potential distortions fromerrorsofunitsinthedataandimplausiblysmalltradevalues. Weruneachgravityregression at the SITC 4-digit level and constrain ourselves to products with over 1,000 exporter-importeryearobservations. Theamountofdatalostduetoconcordanceissuesforincomeanddistancedata willvarybyspecificationsincetheuseoffixedeffectsoftenobviatestheuseofthosevariables, but the most punitive cut of the data still accounts for over 83 percent of global trade value between 1980 and 2004. OurestimatorisOLSonthelog-linearspecificationof(3). Cognizantofthefactthatthereare many different ways to specify that equation, we try an array of five different panel specifications withvaryingdegreesofcontrolformultilateralresistanceterms. Again,ourobjectiveistocompute various indexes of the change in the residual (4) which will be informative of the portion of U.S. share decline not explained by gravity controls such as income and geography. The differences amongthesefiveregressionsarethetreatmentoftheρterms(whichinsomecasesarecountry-pair fixed effects and in others are the standard distance, border and language controls), the measure of country-specific variables D, as well as the subset of data used for the estimation. The specifications are described in Table 2. In specification (i), we regress the exporter’s share of global sales in each SITC product on the exporter’s relative nominal income (recall that 6

the importer-specific terms cancel by dividing by a reference exporter), exporter-importer fixed effects, year fixed effects, and dummies for NAFTA and EMU. An actual measure of income is used to control for the trends described in the previous section. The exporter-importer FE is a static measure of trade costs which wipes out variation in border, distance and language, and arguablyincludesmanymoreunmeasured(andunchanging)barrierstotrade. Tocontrolforsome large policy changes during our sample which we do not view as endogenous to competitiveness, dummiesforpost-NAFTAandpostEuroyearsareincludedfortheappropriatecountries. Finally, year fixed effects soak up secular trends in n. Specification(ii)usesthesameregressorsas(i),butonasubsetofthedatathathasobservations for at least 20 of the 25 years in the sample (i.e., within each exporter-importer-SITC cell). It is informative to constrain ourselves to the subset of bilateral trade flows that are balanced over the course of the sample for at least two reasons. First, the average results statistics reported across products may be skewed by compositional changes over time in the unbalanced panel. Secondly, our linear-in-logs specification potentially introduces selection bias by dropping the observations withzerotradeflows. Onepossiblewaytoassessthesensitivityoftheresultstolooseningthedata truncation at zero would be to tighten it further; that is, any selection bias caused by dropping zero values would be enhanced by dropping sporadic ones. Specification (iii) uses exporter-year fixed effects in the place of GDP. Since these fixed effects also approximate changes to the multilateral resistance terms of the exporter, they may in fact be soaking up some of the information on competitiveness intended to be measured. As such, the robustnessoftheresultconsistsofasimilarprofileofresidualchangesinspecifications(i)through (iii), due to the following trade-off: in the first two there is likely some omitted variable bias since implicit indexes of multilateral resistance (as defined in Anderson and van Wincoop (2003)) are themselves a function of geographic variables included in the regression. On the other hand, the appropriatecontrolformultilateralresistanceremovesfromtheresidualatleastsomeinformation on the relative performance of exporters. Specifications (iv) and (v) check the robustness of the results to more standard gravity specifications, by unfolding (3) into levels and incorporating conventional measures of static trade costs. Specification (vi) uses a an alternative data source on bilateral international trade flows aggregated into broader ISIC 2-digit sectors.7 3.2 Reduced form revealed competitiveness: results After controlling for model factors in several alternative formulations of the gravity model, we find that the U.S. export share is only in slight decline. In our benchmark specification (i), the majorityoftheroughly20percentdeclineinaggregateU.S.exportshareisexplainedbythemodel with about a 6 percent decline in the residual.8 Table 3 shows the estimates of control variables for (3) estimated across all products.9 As expected, exporter GDP share is positively related to export share, with a 1 percent decrease in relative income decreasing export share by roughly 0.4-0.6 percent. These magnitudes are similar to the coefficients on GDP in the level regressions and slightly lower than those using the ISIC data. The effect of NAFTA and the introduction of the euro are both positive and significant, with coefficients ranging from 0.4-1.5 and 0.1-0.5, respectively. Measures of distance, language and border have the expected sign. 7Specification(vi)confirmstheconsistencyofthereducedformresultswiththeempiricalexerciseinthefollowing section. WhilethereducedformregressionsusetheFeentraetal. (2005)datadescribedabove,themethodologyin thenextsectionadditionallyrequiresdataonsectoralintra-nationaltrade,whichnecessitatesusinganalternative dataset. Thosedataaredescribedbelow. 8ResultsfortheremainingfivespecificationscanbefoundinAppendixB. 9Asmentioned,thegravityresidualsareestimatedattheSITC4-digitlevelforspecifications(i)-(v)andatthe ISIC 2-digit level for specification (vi). In the table, due to computational constraints on such a large dataset, we present aggregate control variables estimated without product fixed effects. As such, the coefficients can be interpretedassimpleaveragesacrossSITCproducts,orinthecaseofspecification(vi),ISICproducts. 7

An index of market share changes for the U.S., along with an index of model predicted values, are shown in Figure 7. The index in each year is a geometric mean of share changes across U.S. destination countries and products, where each change in share is weighted by the SITC-importer value in the year 2000.10 Despite a widening of the gap between the two indices in the early period, the model prediction broadly follows the share trend. Since there are not many timevarying regressors in our gravity estimation, this result is closely related to the observation in Figure 6 that U.S. income share and trade share have similar dynamics. To construct a statistic for the overall percent change in market share due to the gravity residual, the ratio of actual to predicted share is averaged across time periods in the early part of thesample(1980-1992)andthelatterpart(1993-2004)andthelog-differenceofthesetworatiosis taken for each exporter-importer-SITC group. The average of those statistics across destinations and 4-digit product groups is shown in Figure 8 for the G20 plus Singapore, Taiwan and Hong Kong.11 Across all products, the U.S. is in the middle of the pack with decreases in its residual of 6 percent. This can be interpreted as a decrease in U.S. export market share of 6 percent that is not accounted for by the dynamics of income, and is notably smaller than the overall share declineofapproximately20percentoverthatperiod. ThissuggeststhatU.S.relativeproductivity competitiveness,albeitinslightdeclinebythismeasure,didnotdeclinebynearlyasmuchasitsfall insharemightsuggest. Thisresultisconsistentacrossproductcategories, showninTable4, even for SITC 7 (machinery and transportation) where U.S. share performance was particularly grave, as well as for other specifications shown in Appendix B. For other exporters, clear winners and losersemerge. Indonesia,China,IndiaandMexicohadamongthehighestincreasesintheirgravity residualbyalargemargin,astheirexportgrowthfaroutpacedtheincreaseintheirincomeshares. On the other hand, certain large Asian exporters had dramatic falls in their residuals presumably duetotheriseofChinaandlargeincreasesinMexicanexportstotheU.S.overthesampleperiod. European countries and Canada had more moderate changes in their export performance and, with a few exceptions, tended to lag behind the rest of the world. Insummary,thisreducedformexercisestronglysupportsthestorythatexporterincomeshares areanimportantdeterminantoftradeshares. Beyondthat,however,itisdifficulttoknowwhether thegravityresidualreflectstheactualevolutionintheunderlyingproductivityofexportersrather than other factors, such as evolving trade costs. In the following section we take a different approach to identifying relative cost competitiveness across countries by modelling the microfoundations of trade shares explicitly. 4 Structural Revealed Competitiveness In this section we build on a multi-country multi-sector version of the Melitz-Ottaviano (2008) model to obtain a (computable) structural equation for the relative competitiveness of a country. A full description of the reference model is reported in Corcos et al. (FEEM, 2010), although its main properties are summarized in Appendix A. Themodelyieldsthefollowingexpressionforaggregatebilateraltradefromcountryltocountry 10For the sake of comparability, the predicted and actual market share changes are aggregated over exactly the same SITC-destination pairs. The index of share change does not exactly match that in Figure 1, since: (a) it is a geometric index, whereas simply adding up share across products as in Figure 1 is analogous to an arithmetic mean,and(ii)becausetheindexismatchedineachperiod(i.e.,thetradeflowhadtooccurinbothtimetandt-1 forittobeincluded),thecompositionofitemsintheFigure2indexwillbeasubsetofthoseinFigure1. Overall, the magnitude of the drop of the geometric index seems reasonably close to the aggregate drop and the dynamics ofcontractionsintheearly1980’sand2000’sparalleloneanother. 11Thislistcorrespondswellwiththetoptwentyexportersbysizein1980. Inthetable,thecategory‘OtherEU’ includes:Austria,Belguim,Denmark,Finland,Ireland,Netherlands,PortugalandSweden. 8

h in a given sector s12: Tlh =Υ ρlh El [max(m)l]−γs Dh [mhh]γs+2 (5) s s s s s s where γ is the shape parameter of the (Pareto) marginal cost distribution in sector s; Υ ≡ s s 1 isabundlingsectoralparameterplayingnoroleinsubsequentanalysis;El isthenumber 2υs(γs+2) s of entrants in country l - sector s; max(m)l is the upper bound of the exogenous marginal cost s distribution in country l - sector s (exogenous cost cutoff); ρlh ∈ (0,1] is a measure of trade s freeness between country l and country h in sector s; Dh is country size (i.e population and, by extension, GDP); mhh is the endogenous maximum possible marginal cost for a generic domestic s firm producing and selling in country h - sector s (endogenous cost cutoff). Equation (5) expresses exports from l to h in a given sector as a function of bilateral trade freeness [ρlh] and a set of country characteristics specific to the exporting [max(m)l, El] or the s s s importing [Dh, mhh] country.13 s Itisworthnotinghow,bearinginmindequations(13)and(14),thevectorofinverseendogenous cutoffsM −1 (withgenericelement1/mhh)canbeinterpreted, onceordered, asacountryranking s s in terms of actual competitiveness. On the other hand, the vector of inverse exogenous cutoffs Ψ−1, with generic element 1/ψh ≡ (cid:2) ωhfh(cid:0) max(m)h(cid:1)γs (cid:3)−1 , can be thought of, once ordered, as S s s s s a country ranking in terms of the exogenous ability to generate low cost firms. Given Ψ−1, a S country’spositioninM −1 isaninversefunctionofhomemarketsize(D)andtradefreeness(P ). s S As explained in Appendix A, to stress this relationship between 1/max(m)h and 1/mhh, we refer s s to the former as the Producer (Marginal Cost) Competitiveness of country h and to the latter as its Overall (Marginal Cost) Competitiveness (henceforth OC and PC respectively). Equation (5) provides us with the chance to derive an analytical expression that can be used to infer the vector M −1 of the OC of the countries from observed bilateral trade flows. To this s aim,westartbynotingthatonlyTlh andDh areobservable. Thuswefirstofallneedtopurge(5) s of the unobservable terms. However, consider that the terms in (5) are specific to both the origin and the destination country [i.e. ρlh], or either to the former (i.e. [max(m)l]−γs El) or the latter s s s (i.e. [mhh]γs+2Dh) only. To isolate OC, we can therefore use country l’s exports to a reference s country f (UK in the application), to transform equation (5) into a prediction of relative (instead of absolute) trade flows: Tlh/Dh ρlh (cid:20) mhh(cid:21)γs+2 s = s s (6) Tlf/Df ρlf mff s s s Thisexpression,inwhichmeasurabletermsaregroupedonthelefthandside,expressesmeasurable (relative) trade flows as a function of trade freeness and OC, both in relative terms. Using a tilde to indicate that a variable is expressed in relative terms (ρ˜lh = ρlh/ρlf; D˜h = s s s Dh/Df; m˜hh ≡ mh s h ), relative average marginal costs in a given country-sector can be written as s mf s f (cid:32) (cid:33) 1 T˜lh 1 γs+2 m˜¯hh ≡ s (7) s D˜h ρ˜lh s where we also used the fact that, under the Pareto assumption, mh = γs mhh, and thus m˜¯hh ≡ s γs+1 s s mh s h . mf s f 12ThenumberofexportersfromltohamountstoEl (cid:20) ml s (cid:21)γs . Eachexporterfromltohgeneratesf.o.b. s max(m)l s h exportsalesequaltoplh(c)qlh(c). Aggregatingoverallexportersyieldsequation(5). 13Note that the adjustment of Tlh takes place along both the ‘extensive margin’ (number of exporters) and the s ‘intensivemargin’(percapitaexports). 9

Bilateraltradecosts-ormorepreciselythedegreeoftradefreenessρ˜lh -arehoweverunknown. s To deal with this issue, we derive - as suggested by Novy (2009) - a very simple form for bilateral trade freeness, which exploits the structure of the reference model without the need to estimate a gravity equation. From (6), bilateral trade freeness between country l and country h can be in fact expressed as T˜lh T˜hl ρ˜lh ρ˜hl Ω˜lh ≡ s s = s s . (8) s T˜ll T˜hh ρ˜ll ρ˜hh s s s s The intuition behind (8) is (Novy, 2009) straightforward. If bilateral trade flows between two countries increase relative to domestic trade flows, it must have become relatively easier for the two countries to trade with each other. This is captured by an increase in Ω(cid:101)lh, and vice versa. s Assuming ρ˜lh = ρ˜lh, (8) can be plugged into (7) in order to obtain the following measure of s s Revealed Overall Competitiveness (henceforth ROC)14: D˜h Ω(cid:101)lh ROC ≡(m˜¯hh)−1 = s . (9) s T˜lh s Equation (9) does not require econometrics. The advantage over gravity estimates15 is that Ω(cid:101)lh s can be calculated not only for cross-sectional data but also for time series and panel data. Thus, theevolutionoftheresultingcountryrankingscaninthiscasebetrusted. Notealsothat,although Ω(cid:101)lh =Ω(cid:101)hl, T˜lh normally differs from T˜hl. Thus, what equation (9) suggests is that the difference s s s s in T˜lh respect to T˜hl has to be traced back to differences in relative costs (m˜¯hh/m˜¯ll) and market s s s s size (D˜h/D˜l). Finally, it is worth noting that the idea of ”revealed” competitiveness associated with (7) is moregeneralthanmoreconventionalmeasuresofaggregatetotalfactorproductivity(tfp). Tosee this, consider equation (12): our measure of ”overall competitiveness” is a composition of ”inverse tfp”(c)andinputcosts(wl ), aswellasinputshares(βl ). Althoughacountrycouldhavehigh x,s x,s tfp (i.e. low c) in sector s, that may not be sufficient to be competitive in international markets. Itcouldbethatinternationaldifferencesininputcosts(suchascapital, labourandintermediates) are a disadvantage to that country. Moreover, a country’s domestic value added (DVA) content of exports might be low, which would dampen the link between a country’s tfp and its export performance. The importance of tfp in determining the international competitiveness of a country decreases with the degree of international fragmentation of the production process in the country. By definition, tfp is meant to measure the output differences which are not explained by different input choices and occurs, instead, through marginal product increases. Due to this physical nature, firms’ tfp (and thus a country’s tfp) is invariant to different choices concerning whether to outsource phases of the 14Since the exponent 1 plays no role in determining the country rankings, as it only entails a re-scaling by γs+2 sector,itwillbeomittedhereinafter. 15Equation(6)couldbeinterpretedasagravityequationandestimatedas (cid:32) T˜lh (cid:33) ln D˜ s h =imps −βslnX˜ s lh (10) wherevectorX˜lhincludesbilateraldistances,aswellasanumberofdummiescontrollingforthepresenceofborder s effects(contiguity,languageindicators,etc.),andimpsisa(destination)country-sectordummycapturingtheROC. Estimationof(10)providesuswithinformationontradecosts,throughβˆs,and,atthesametime,withinformation onROC.Moreprecisely,thefixedeffectsin(10)canbeestimated(seeFadingerandFleiss,2008)as m˜¯h s h=exp (cid:34) ln (cid:32) T D ˜ ˜ s l h h (cid:33) −βˆ slnX ¯˜ s lh (cid:35) (11) wherethebarreferstothemeanacrossexportingcountries. 10

production process and whether to buy intermediates domestically or abroad. Whilst tfp is not affected by these choices, marginal costs are. For given quantities of intermediate inputs used in production,thepossibilitytoimportthemfromabroadoffersachancetoreducemarginalcosts(see equation (12)). In the aggregate, this results in an improved capacity to target the international consumers of the final good s at relatively low prices. Since it is expressed in units of marginal costs, ROC is a measure of competitiveness which is ”naturally” linked to the concept of DVA. Moreover, since the international structure of ROC (vector M ) results from a combination of s forces (such as trade costs and market size) affecting the degree of international competition for final goods, ROC is informative of a given country’s ability to sell good s at low prices to the international market; in contrast, tfp is informative of that country’s ability to sell good s at low prices domestically. 4.1 Structural revealed competitiveness: data & specification As equation (9) derives country h’s ROC from its bilateral trade flows with a given country l, for each country h (and industry s) we compute m˜¯hh as many times as the number of its commercial s partners. In other words, our reported ROCs are obtained considering, for each country, all the country pairs for which bilateral trade flows are available. A single value for m˜¯hh is then obtained s as a weighted average in which each country is assigned its share on country h’s total imports as weight.16 Asinsection3,wefocusontwoperiods(1980-1991and1992-2004). Dataonbilateralflowsare obtainedfromtheCEPIITradeProd database. Thechoiceisdrivenbythefactthat,incontrastto thebilateraltradedatausedabove,TradeProd reportsreliableinternaltradeflows. Tradeflowsare providedinnominaldollarsatthe3-digitleveloftheISICRev.2classification. Again,wetruncate the data at $10,000 per annual bilateral flow; this has no remarkable effects on the results. As above, data on country GDP from the Penn World Tables are converted into international dollars at PPP exchange rates. We use United Kingdom as our reference country since it has the highest number of observations as importer or exporter. Consistent with the reducted form exercise, results are presented for the G20 country group with the exception of Saudi Arabia, for which information on internal trade flows is unavailable. 4.2 Structural revealed competitiveness: results In this section we focus on average percentage changes in ROC from the early to the late period. Our main results are synthesized in Figure 9, though readers are directed to Appendix B for additional detail on countries and industries. Figure9reportstheaverageROCpercentagechangeforthoseG20countriesforwhichROCis available for at least 23 out of our 28 sectors. For each country, the sectors are weighted using the product’s average share in the country’s export bundle during the late period. Standardization is by sector and with respect to the whole G20. A slight decline (-5.58%) characterizes the evolution oftheaverageROCvariationintheU.S.Overall,Figure9confirmstheexceptionalcompetitiveness growth of certain emerging market competitors such as China and Mexico but also that of other, moretraditional,competitorssuchasCanadaandAustralia. AmongEUcountries,onlyUK,Spain and Austria show a positive variation in ROC. In particular, Italy is the worst performing G20 country, followed by Portugal and traditional U.S. competitors like France and Germany. Throughout the paper we have interpreted the gravity residual and structural cost estimates as largely reflecting latent exporter productivity. However, other factors likely contribute to export performance in excess of what might be predicted by these frameworks. As mentioned in 16With this specification, zeros-missings in bilateral trade do not translate one-to-one into zeros in m˜¯hh. The s lattercaninsteadbeduetomissinginformationonGDPand/orinternaltradeincountryh. 11

the model description, the structure of production within certain regions of the global economy could be playing an important role. For example, regions that are relatively intensive in crossborder production sharing would record higher exports for a given unit of output independent of exporterproductivity. Indeed,thismaybebehindsomeofthehighmeasuresofperformancethat we estimate for China and Mexico over the sample period. In principle, though, the dynamic trade cost measure in the structural analysis (which compares international to intranational trade flows) captures some of the increasing incidence of production sharing. The fact that East Asian countries, excluding China, had competitiveness losses in the reduced form estimates and competitiveness gains in the structural estimates is consistent with the reality of large flows of goods passing through China for final assembly. 5 Conclusion TheU.S.shareofglobalexportshasfallenbyroughly20percentoverthelastdecade. Thispaper aimstodeconstructthedriversofthedeclineinshare. First,wedocumentthatthedistributionof thedeclineisquiteuneven,withaminorityofcategoriescontributingdisproportionately. Second, whencontrollingfortherelativedeclineintheU.S.shareofglobaloutput,drivenbyalargeextent by rapid growth in emerging market economies, the fall in share is far less pronounced. We formalize this notion within a gravity framework and assess changes in competitiveness through the evolution of the estimation residuals in an array of empirical specifications. We find that, accountingforincomeshareandothercontrols,thedeclineinU.S.exportshareislargelyexplained by model factors. We then take a more structural approach to examining the evolution of U.S. competitiveness, deriving an expression for U.S. export share from a heterogeneous firms model in the style of Melitz-Ottaviano (2005). This approach confirms the outcome of our gravity model exercise, that the U.S. has generally maintained its level of competitiveness within detailed product categories, despite the fall in the overall share. All together this analysis points to the inadequacy of the aggregate export share as an indicator of country export competitiveness. 12

References [1] Anderson,J.E.andE.vanWincoop(2003),”GravitywithGravitas: ASolutiontotheBorder Puzzle”, American Economic Review, 93(1), 170-92. [2] Baier, S.L. and J.H. Bergstrand (2001), ”The growth of world trade: tariffs, transport costs and income similarity”, Journal of International Economics, 53, 1-27. [3] Chen N. and Novy D. (2009) ”International Trade Integration: A Disaggregated Approach”, CEPR dp, 7103. [4] Corcos G., Del Gatto M., G. Mion and G.I.P. Ottaviano (2009), ”Productivity and Firm- Selection: Quantifying the ”New” Gains from Trade”, FEEM wp, 115. [5] DelGattoM.,diMauroF.,ForsterK.(2010),”Re-establishingcompetitivenessintheEURO area: insights from a computable partial equilibrium trade model with heterogeneous firms”. mimeo. [6] Del Gatto M., G. Mion and G.I.P. Ottaviano (2006), ”Trade Integration, Firm Selection and the Costs of Non-Europe”, CEPR Discussion Paper, 5730. [7] European Central Bank (2005), ”Competitiveness and the Export Performance of the Euro Area”, Occasional Paper Series, 30. [8] Fadinger H. and Fleiss P. (2008) ”Trade and Sectoral Productivity”. MPRA wp, 6938. [9] Finicelli A., Pagano P. Sbracia M. (2009) ”Trade-revealed TFP”. Banca d’Italia wp, 729. [10] Jacks D. Meissner C., and Novy D. (2008) ”Trade Costs, 1870-2000”. American Economic Review, Papers Proceedings, 98(2), 529-534. [11] Head, K. and T. Mayer (2000), ”Non-Europe: the Magnitude and Causes of the Market Fragmentation in the EU”, Weltwirtschaftliches Archiv, 136(2), 285-314. [12] Head, K., T. Mayer and J. Reis (2010), ”The erosion of colonial trade linkages after indepenence”, Journal of International Economics, 81, 1-14. [13] Martin, P., T. Mayer and M. Thoenig (2008), ”Make trade not war?” Review of Economic Studies, 75(3), 865-900. [14] Melitz M. and Ottaviano G. (2008), ”Market Size, Trade, and Productivity”, Review of Economic Studies, 75, 295-316. [15] Novy D. (2009) ”Gravity Redux: Measuring International Trade Costs with Panel Data”. mimeo. [16] Ottaviano G.I.P., Taglioni D. and di Mauro F. (2009). ”The euro and the competitiveness of European firms”. Economic Policy, 24(01). [17] Richardson J. D. (1971), ”Constant-Market-Shares Analysis of Export Growth”, Journal of International Economics, 227-239. [18] Waugh M.E. (2009) ”International trade and income differences”. Federal Reserve Bank of Minneapolis, Staff Report, 435. [19] Whalley, J. and X. Xin (2009), ”Regionalization, changes in home bias, and the growth of world trade”, Journal of Policy Modeling, forthcoming. 13

14 13 12 11 10 9 8 7 6 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 Figure 1: U.S. share of world merchandise exports tnecreP

20% 18% 16% 14% 12% 10% 8% 6% 4% 2% 0% 1984 1989 1994 1999 2004 Figure 2: U.S. share of world merchandise exports, by SITC 1-digit sector )tnecrep( stropxe dlrow fo erahS Food & Live Animals (0) Beverages & Tobacco (1) Crude Materials (2) Mineral Fuels (3) Animal & Vegetable Products (4) Chemicals (5) Manufactured Goods by Material (6) Machinery & Transportation (7)

Chemicals Misc. Manufactured Other Manufactured Goods by Material Animal & Vegetable Products Beverages & Tobacco Mineral Fuels Crude Materials Machinery & Transportation Food & Live Animals -1.0% -0.8% -0.6% -0.4% -0.2% 0.0% 0.2% Figure 3: SITC 1-digit contributions to the aggregate share decline (percentage points) Fuel oils Cotton, not carded or combed Aircraft parts Office and data processing machines Oscilloscopes & spectrum analyzers Digital processing units Lignite Soybeans Motor vehicle parts and accessories Corn -0.4% -0.3% -0.3% -0.2% -0.2% -0.1% -0.1% 0.0% Figure 4: Top 10 4-digit contributions to the aggregate share decline (percentage points)

Chemicals Misc. Manufactured Other Manufactured Goods by Material Animal & Vegetable Products Beverages & Tobacco Mineral Fuels Crude Materials Machinery & Transportation Food & Live Animals -2.0% -1.5% -1.0% -0.5% 0.0% 0.5% 1.0% 1.5% 2.0% Contribution to Aggregate Share Decline Commodity Effect "Competitiveness" Overall Figure 5: Commodity and competitiveness contributions to the aggregate share decline (percentage points)

Figure (cid:25): Export and GDP Shares

Figure (cid:25): Export and GDP Shares

120 110 100 90 80 70 60 50 Figure 7: Predicted and actual market share indices )sthgiew 0002 raey( erahS tekraM .S.U fo xednI cirtemoeG 0891 1891 2891 3891 4891 5891 6891 7891 8891 9891 0991 1991 2991 3991 4991 5991 6991 7991 8991 9991 0002 1002 2002 3002 4002 gravity model market share

60% 40% 20% 0% -20% -40% -60% Figure 8: Reduced form measure of competitiveness, change between early and late sample (percentage points) 100 80 60 40 20 0 -20 -40 -60 -80 Figure 9: Structural measure of competitiveness, change between early and late sample (index)

U.S. Export Share SITC Description 1984 1989 1994 1999 2004 71 POWER GENERATING MACHINERY AND EQUIPMENT 0.25 0.23 0.23 0.26 0.20 72 MACHINERY SPECIALIZED FOR PARTICULAR INDUSTRIES 0.16 0.12 0.14 0.16 0.13 73 METALWORKING MACHINERY 0.10 0.10 0.13 0.14 0.12 74 GENERAL INDUSTRIAL MACHINERY AND MACHINE PARTS 0.17 0.15 0.17 0.18 0.12 75 MACHINES 0.33 0.24 0.20 0.14 0.09 76 TELECOMMUNICATIONS AND SOUND EQUIPMENT 0.08 0.08 0.11 0.13 0.06 77 ELECTRICAL MACHINERY AND ELECTRICAL PARTS THEREOF 0.20 0.16 0.17 0.18 0.12 78 ROAD VEHICLES 0.14 0.10 0.11 0.10 0.08 79 TRANSPORT EQUIPMENT 0.21 0.30 0.29 0.32 0.24 Table 1: U.S. export share in machinery and equipment categories

(i) (ii) (iii) (iv) (v) (vi) Dependent Export share Export share Export share Export sales Export sales Export share Exporter Exporter Exporter Exporter Exporter-year GDP, GDP, Exporter D GDP share GDP share FE Importer Importer GDP share GDP GDP Distance, Country-pair Country-pair Country-pair Country-pair language, Country-pair ρ FE, NAFTA, FE, NAFTA, FE, NAFTA, FE, NAFTA, border, FE, NAFTA, EMU EMU EMU EMU NAFTA, EMU EMU n Year FE Year FE Year FE Year FE Year FE Year FE Sampple Full Balanced Full Full Full Full Data SITC-4 SITC-4 SITC-4 SITC-4 SITC-4 ISIC-2 Table 2: Gravity regression specifications.

Dependent var. → Export Share Export Volume Memo: (i) (ii) (iii) (iv) (v) ISIC industries Exporter GDP share 0.430 ** 0.644 ** 0.890 ** (0.001) (0.005) 0.008 Exporter GDP 0.682 ** 0.345 ** (0.005) (0.001) Importer GDP 0.522 ** 0.478 ** (0.004) (0.001) NAFTA 0.484 ** 0.424 ** 1.486 ** 0.922 ** 1.281 ** 1.192 ** (0.011) (0.011) (0.010) (0.012) (0.009) 0.055 EMU 0.417 ** 0.305 ** 0.074 ** 0.332 ** 0.461 ** 0.191 ** (0.004) (0.005) (0.005) (0.005) (0.004) 0.011 Distance -0.263 ** (0.001) Common Language 0.142 ** (0.002) Common Border 0.409 ** (0.002) Exporter-Importer FE Yes Yes No Yes No Yes Exporter-Year FE No No Yes No No No Year-FE Yes Yes Yes Yes Yes Yes Balanced panel No Yes No No No No N 11,638,401 4,526,163 12,672,551 11,253,727 10,101,064 2,998,339 R-squared 0.40 0.44 0.20 0.25 0.20 0.60 RMSE 1.65 1.37 1.89 1.67 1.74 1.69 Table 3: Estimates of control variables in the gravity regression

All SITC 0 1 2 3 4 5 6 7 8 Indonesia 50% 29% 9% 17% 36% 71% 36% 57% 91% 48% China 46% 26% 5% 27% 7% 9% 21% 45% 65% 64% India 39% 33% 18% 16% 41% 49% 66% 49% 9% 44% Turkey 37% 7% 10% 7% -19% 5% -4% 57% 48% 56% Mexico 36% 22% 62% 6% -17% 0% 24% 37% 53% 47% Saudi Arabia 31% 54% 29% 5% 5% 30% 46% 66% -16% 13% Spain 19% 13% 6% 22% 17% 31% 20% 15% 21% 24% Italy 15% 5% 10% 13% -21% 19% 10% 22% 12% 17% South Africa 14% 3% 40% 8% 50% 4% 11% 8% 31% 11% Rest of World 8% -2% 6% 4% 22% 2% 4% 11% 11% 11% Australia 5% 5% 27% 3% 2% -7% 3% 2% 4% 10% France 3% 2% -10% 2% 15% 3% 2% -2% 7% 2% Other EU 2% 0% 3% 5% 1% 7% 4% -4% 5% 2% Germany 1% 6% 6% 17% -1% -1% -8% -2% 7% -4% Brazil -6% -7% -6% 18% -40% -20% -5% -3% -9% -13% USA -6% -9% -6% -4% -16% 2% -2% -9% -6% -4% Canada -10% -12% 2% -4% -4% 7% -23% -19% -9% 5% Korea -12% -27% -23% -2% 50% -35% 17% -18% 13% -58% UK -12% -8% -1% -1% -19% 17% -21% -13% -11% -11% Argentina -12% 1% 27% -7% 16% 11% 8% -17% -48% -20% SSiinnggaappoorree -2299%% -3377%% -33%% -4422%% -3333%% -4477%% -66%% -4422%% -2222%% -3344%% Japan -29% -49% -23% -31% -10% -53% -9% -44% -20% -42% Taiwan -29% -100% -15% -25% -32% -29% -10% -23% -14% -57% Hong Kong -41% -52% -26% -63% -19% -61% -46% -42% -28% -46% Table 4: Evolution of the gravity residual (early sample to late sample)

A Appendix A: Short description of the reference model The theoretical background is the framework developed by Del Gatto et al. (2006), also used in Ottaviano et al. (2009). While the reader is redirected to those papers, and in particular to Corcos-DelGatto-Mion-Ottaviano (FEEM, 2010)17 for an extensive exposition, here we report a short description of the logic behind the model and the key equations for the application. Themodelisamulti-countrymulti-sectorversionofMelitzandOttaviano(2008)encompassing S industries (with no inter-industry linkages)18 active in N countries, indexed l = 1,...h..,N. Each country-industry is endowed with given amounts of labor Ll and capital Kl (factors are geographically immobile) and the output of each industry is horizontally differentiated in a large set of varieties. Consumers maximize a quasi-linear utility function with quadratic sub-utility, as in Ottaviano etal. (2002). Underthishypothesis, thedemandofagenericvarietyinagivencountryispositive only provided that its selling price is lower than a certain (cutoff) level max(p)l. This level is s higher when: consumers like the differentiated good a lot, varieties are very differentiated, the average price is high, the number of competing varieties is small. Firmscompeteinamonopolisticmarketandeachvarietyissuppliedbyoneandonlyonefirm. Eachfirmisnegligibletothemarketanddoesnotcompetedirectlywiththeotherfirms. However, given the demand structure, firms interact indirectly through an aggregate demand effect, as the total output of the industry has an influence on firms’ profit. Firms in a given sector share the same (Cobb-Douglas) technology but are heterogeneous in terms of Unit Input Requirement (UIR) c, defined as inverse ‘total factor productivity’ (tfp) (i.e. c = 1 ). c is used to identify the firm. Accordingly, the marginal cost faced by a generic firm c tfp active in country l and sector s is: ml ≡m(c)l =c ωl (12) s s s where ωl = B (cid:81) (cid:0) wl /β (cid:1)βx,s, with wl and β denoting input x’s cost and share (in s x∈X x,s x,s x,s x,s country l - sector s) respectively, and X = {k,l,m} (i.e. capital, labour, and intermediates) and (cid:80) β =1. B is the bundle of parameters associated with the Cobb-Douglas.19 x∈X x,s National markets are segmented but firms can export and, as production faces constant returns to scale, they independently maximize the profits earned in different destination countries. Exporting firms incur a per-unit trade cost, encompassing not only carriage in a strict sense, but all those ”impediments to trade” whose amount is related to the quantity exported. For each delivered unit from country l to country h, τlh > 1 units have to be shipped. Moreover, we also s allow for costly trade within a country with τlh >τll ≥1. s s Firm heterogeneity is modeled as follows. In order to enter the market, each firm has to make an irreversible investment in terms of labor and capital. This ”sunk cost of entry” amounts to ωlfl. Only once this cost has been payed, and production started, a firm is allowed to observe s s its own marginal cost ml. This is modeled as the outcome of a draw from a common and known s Pareto distribution (cid:104) ml s (cid:105)γs , with support [0,max(m)l] varying across countries.20 max(m)l s s 17Thepaperisdownloadableathttp://www.feem.it/userfiles/attach/2009121116814115-09.pdf. 18Asinter-industrylinkagesareruledout,thesindexcouldbeomittedandthemodelpresentedasan”industrymodel”,withalltheequationsreferringtoagenericindustry. However,thesindexwillrevealusefulinsubsequent analysis,ascountry,industry,andcountry-industryspecificvariables(parameters)coexistinthemodel. 19Equation(12)expressesthemarginalcostassociatedwithastandardCobb-Douglasproductionfunction Q(c)l s =c−1 (cid:89) (Mx)βx,s x∈X whereMx denotestheamountofinputxutilized. 20Inastrictsense,theParetoassumptionreferstoc(i.e. theUIR).However,asevidentfrom(12), (cid:20) ml s (cid:21)γs ≡ max(m)l s 14

Only those firms whose cost draw is good enough to enable them to sell to market h at a price below the price cutoff max(p)h earn non-negative profits and can afford to serve that market. s Let mhh denote the marginal cost inclusive of trade frictions faced by a producer in country hs industry s that is just indifferent between serving its local market or not. Then, by definition mhh = max(p)hh. A firm, wherever located, can serve market h only provided that its delivered s s costdoesnotexceedmhh. Inotherwords: firmcproducingincountryl isabletotargetmarketh s when τlhml <mhh, it is not able to target market h when τlhml >mhh, it is indifferent between s s s s s s serving or not market h when τlhml = mhh. Thus, mhh measures the ‘cutoff cost’ in country s s s s h-industry s. The analytical solution in terms of the N ×S equilibrium cost cutoffs is: M γ1+2 = Φ P −1 D−1 Ψ 1 1 1 1 . . . . . . . . . . . . . . . M γs+2 = Φ P −1 D−1 Ψ (13) s s s s . . . . . . . . . . . . . . . M γS+2 = Φ P −1 D−1 Ψ S S S S where: • M is the N ×1 vector of the equilibrium cost cutoffs in industry s, whose h-th element s mhh = ωh max(c)h denotes the maximum possible marginal cost for a generic (domestic) s s s firm active in industry s, producing and selling in country h; • Φ ≡2υ (γ +1)(γ +2)≡ γs+1 is a (scalar) positive bundling parameter21; s s s s Υs • P isaN×N ‘tradefreenessmatrix’whoseelementinrowlandcolumnhisρlh ≡ (cid:0) τlh(cid:1)−γs ∈ s s s (0,1]. ρlh denotes the degree of trade freeness between country l and country h, ; s • D is a N ×N diagonal matrix with population along its diagonal and zero elsewhere. In a wide sense, population can be thought of as a measure for the size of the domestic market; • Ψ is a N ×1 vector with h-th generic element ψh ≡ ωhfh(cid:0) max(m)h(cid:1)γs, where fh and s s s s s s max(m)h denoterespectivelythefixedcostofentryandtheupperboundofthe(exogenous) s marginal cost distribution in country h-industry s (exogenous cost cutoff). As presently discussed, ψh is an inverse measure for the ‘exogenous competitiveness’ of country h in a s given industry s; • γ is the shape parameter of the marginal cost distribution in sector s, with higher values s denoting a distribution which is more skewed towards high cost (less productive) firms. Each row of (13) states, for each country in a given sector, the marginal cost above which a firm is not productive enough to serve the domestic market from therein and, since max(m)lh = s mhh/τlh, from anywhere.22 s s (cid:104) c (cid:105)γs for [0,max(c)l]. Thus, there is no loss of generality in thinking (and solving) the model in terms of max(c)l s s marginalcosts. 21Parameter υs comes from the utility function and measures the degree of product differentiation between different varieties of good s. When υs = 0, consumers only care about their total consumption level over the varietiesofgoods. 22Sincetherearenointer-industrylinkages,rowsin(13)areindependentoneanother. 15

Overall (1/mhh) and producer (1/ψh) competitiveness. ByCramer’srule,theh-thgeneric s s element (i.e. the cutoff level in country h-industry s) of M can be expressed as s mhh = (cid:34) Φ s (cid:80)M l=1 (cid:12) (cid:12)R s lh (cid:12) (cid:12)ψ s l (cid:35) γs 1 +2 (14) s Dh |P | s (cid:12) (cid:12) where|P s |isthedeterminantofthetradefreenessmatrixinsectorsand(cid:12)R s lh(cid:12)isthecorresponding cofactor. Equation(14)entailsarelationshipbetweenmhh andψl, basicallytwomeasuresforthe”coms s petitiveness” of a country. In this relationship: • mhh is endogenously determined by a selection process in which the degree of ’remoteness’, s through the term (cid:80)M l=1 |R s lh|ψ s l , and the size of the domestic market, through Dh, play a key |Ps| role. • ψh captures the exogenous ability of country h to generate low cost firms in industry s, s abstracting from its market size and the degree of remoteness in that sector: low entry costs (lowfh),lowfactorprices[low (cid:0) wh (cid:1)βx,s],andlowprobabilityofinefficientdrawsbyentrants s x,s [i.e. low max(m)h] foster the creation of low cost firms. s Fromapracticalpointofview,(14)canbeusedtoseehowmuchoftheactualcompetitivenessof a country, measured in terms of marginal costs (i.e. 1/mhh), can be traced back to its exogenous s competitiveness, expressed in terms of a mixture of ”traditional” competitive advantages (i.e. factor prices and technology) and entry costs. To highlight this relationship, we refer, for each sector s, to 1/mhh and 1/ψh as respectively ”overall” and ”producer” competitiveness of country s s h (OC and PC respectively), with M −1 and Ψ −1 denoting (once ordered) the corresponding s s country-rankings. 16

All SITC 0 1 2 3 4 5 6 7 8 China 45% 24% 6% 19% -30% -5% 10% 41% 68% 83% 4,803 502 43 294 44 18 752 1,457 529 1,160 India 39% 31% 35% 11% 3% 76% 48% 42% 12% 61% 2,335 321 17 194 2 7 219 754 293 523 Turkey 38% 12% 37% -13% -10% -7% 17% 51% 68% 78% 1,228 337 25 121 2 8 68 330 70 265 Indonesia 38% 11% 16% 9% -7% 69% 23% 51% 29% 77% 1,010 227 23 141 18 34 76 246 21 221 Mexico 32% 12% 63% 6% -45% -30% 21% 31% 66% 34% 1,227 144 32 106 15 6 232 252 265 173 Spain 27% 20% 2% 26% 5% 11% 34% 30% 27% 24% 7,007 596 104 317 62 57 1,073 2,186 1,597 1,006 Italy 17% 4% 19% 6% -15% 32% 14% 25% 16% 16% 15,316 798 125 472 107 57 2,073 4,520 4,602 2,546 Australia 6% 3% 25% 10% 2% -7% -1% 1% 12% 12% 3,180 656 55 336 49 32 331 692 623 393 South Africa 4% -7% 91% -4% 47% -52% -20% -11% 57% 43% 1,166 204 15 231 25 3 108 358 123 98 Other EU 4% 0% -2% 2% -3% 5% 8% 0% 6% 8% 45,610 3,920 557 2,156 457 330 6,939 12,450 12,059 6,677 Saudi Arabia 4% 53% 92% -16% -23% . 27% 74% -27% 8% 179 7 2 23 41 0 47 12 31 15 Rest of World 3% -6% 0% -2% 11% 1% -3% 8% 8% 9% 29,912 5,154 417 2,612 371 211 3,512 6,783 5,018 5,730 France 3% 7% -16% -3% 20% 1% 4% 0% 6% 3% 17,937 1,476 303 709 152 101 2,878 4,755 4,762 2,791 Canada 2% -2% -5% 3% 34% 11% -10% -9% 5% 27% 3,773 448 37 420 34 18 433 840 1,044 480 Brazil 1% -8% -17% 23% -30% -7% 2% 2% 4% -11% 3,467 415 58 241 17 45 459 1,138 767 320 Argentina -5% -1% 46% -13% 48% 6% 11% -13% -34% -23% 1,155 320 27 107 17 37 198 236 128 82 USA -6% -4% 3% -8% -5% -5% -3% -7% -11% 0% 20,678 1,709 238 1,192 256 166 3,212 4,663 6,239 2,906 Germany -7% 0% 0% 8% -5% -10% -16% -7% -4% -8% 21,815 1,277 175 947 232 187 4,004 5,808 6,169 3,000 UK -9% -5% 2% 3% -25% 28% -17% -9% -10% -6% 18,277 1,187 291 675 207 76 3,045 4,828 5,113 2,834 Japan -25% -37% -5% -28% 6% -33% 3% -38% -20% -36% 13,711 306 55 358 79 36 1,849 3,601 5,205 2,192 Singapore -27% -28% -18% -42% -2% -54% -12% -42% -19% -31% 3,950 330 42 204 56 69 466 807 1,232 720 Korea -28% -39% 30% -17% 55% -144% 18% -25% -4% -75% 4,047 121 10 91 10 1 415 1,460 947 988 Taiwan -38% -95% -14% -20% -2% -3% -22% -30% -22% -65% 4,961 193 13 137 10 4 409 1,521 1,466 1,204 Hong Kong -41% -44% -34% -57% 15% -105% -47% -46% -22% -47% 3,595 160 14 67 7 8 165 881 825 1,460 Table B1: Percent change of the gravity regression residual from the early period (1980-1992) to the late period (1993-2004) - Specification (ii)

All SITC 0 1 2 3 4 5 6 7 8 South Africa 14% 12% 14% 8% 18% 13% 18% 16% 15% 8% 7,179 903 130 768 117 38 981 2,193 1,262 737 Indonesia 9% 3% 8% 8% 19% 5% 7% 12% 19% 4% 6,688 717 72 535 90 155 683 2,116 768 1,499 Mexico 9% 3% 15% 5% 3% 12% 11% 8% 11% 7% 6,748 492 109 461 93 29 1,367 1,618 1,597 948 Saudi Arabia 8% -8% -28% 12% -1% 22% 17% 12% 1% 7% 2,156 194 11 161 165 18 527 520 329 205 Turkey 6% -3% 1% 10% 11% -6% 11% 8% 15% -2% 7,665 1,113 103 547 55 62 729 2,531 1,277 1,211 Rest of World 6% 3% 9% 6% 7% 8% 7% 5% 10% 5% 181,193 21,917 2,353 13,804 3,044 1,485 22,931 44,440 38,135 31,393 Hong Kong 6% 1% 19% 5% 16% 10% 11% 6% 8% 2% 11,992 557 64 396 42 41 1,023 3,273 2,982 3,537 India 6% 3% 7% 5% 10% 4% 8% 5% 12% -2% 11,476 1,018 110 739 36 82 1,597 3,679 2,255 1,885 Singapore 5% -4% 16% 8% -6% -1% 5% 3% 8% 6% 14,478 1,156 135 659 244 350 1,691 3,096 4,598 2,424 Argentina 4% 1% 6% 7% -22% 1% 4% 3% 6% 7% 5,515 1,081 108 455 90 186 827 1,228 970 529 Korea 4% 3% -15% 5% -5% 31% -2% -1% 8% 8% 15,442 506 72 420 113 22 1,739 5,110 4,579 2,841 Australia 3% -1% -6% 1% 1% -4% 5% 3% 5% 3% 11,242 1,764 147 1,024 199 100 1,213 2,510 2,729 1,441 All 2% -1% 4% 3% 2% 4% 2% 1% 4% 0% 690,889 63,144 8,143 40,081 8,996 6,156 93,134 180,759 177,818 107,995 Other EU 1% -2% 1% 2% 2% 2% 1% -1% 3% -1% 127,706 10,737 1,550 6,073 1,520 1,282 18,771 33,177 36,519 17,294 Brazil 0% 0% 3% -2% -3% 4% 0% -1% 2% 3% 13,557 1,312 206 833 109 204 1,865 4,160 3,361 1,442 Taiwan 0% -6% 6% 2% -8% 1% 1% -1% 1% 1% 14,543 590 35 554 87 48 1,598 4,347 4,253 2,983 All 0% -2% 1% 0% -1% 0% 0% 0% 0% 0% 230,339 20,808 2,678 12,151 2,270 1,511 32,963 60,578 59,128 37,784 Japan 0% -4% -10% 1% -8% 7% -5% -1% 2% 1% 29,191 887 130 990 245 133 3,913 7,772 10,635 4,320 USA -1% -3% 3% 3% 1% 6% -2% -1% 0% -4% 45,837 4,353 646 3,059 678 586 6,281 10,739 13,016 6,173 UK -1% -6% -3% -1% -4% 2% 0% -2% 1% -2% 40,583 2,850 621 1,887 530 337 6,332 10,534 11,376 5,902 Canada -1% -5% -2% -4% -11% 4% 1% 0% 0% -7% 15,341 1,624 127 1,158 146 104 1,764 3,564 4,673 2,026 France -3% -9% 0% 2% -4% 1% -1% -4% -1% -3% 41,510 3,709 605 1,890 464 360 6,027 10,585 11,538 6,137 China -3% 3% 0% 3% 18% 9% 4% -3% -1% -14% 20,356 1,469 153 1,100 242 97 2,697 6,126 4,377 4,030 Italy -4% -6% -3% 2% -10% -6% -3% -6% -4% -5% 37,678 2,292 366 1,522 396 241 5,139 10,702 11,002 5,851 Spain -4% -10% 1% -1% 5% 6% -6% -9% 1% -1% 22,813 1,903 290 1,046 291 196 3,439 6,739 5,587 3,187 Table B2: Percent change of the gravity regression residual from the early period (1980-1992) to the late period (1993-2004) - Specification (iii)

All SITC 0 1 2 3 4 5 6 7 8 China 48% 31% 7% 17% 18% 1% 26% 49% 58% 69% 16,841 1,228 127 925 187 92 2,107 5,204 3,531 3,384 Mexico 31% 26% 53% 7% -24% -21% 15% 30% 48% 43% 6,872 500 106 463 91 27 1,363 1,694 1,609 984 Spain 18% 21% 16% 16% 1% 12% 18% 14% 18% 27% 23,001 1,881 287 1,057 275 190 3,384 6,888 5,621 3,284 Italy 17% 10% 29% 12% -17% -9% 8% 27% 12% 25% 36,339 2,214 344 1,553 380 257 4,893 10,444 10,443 5,643 Belgium 12% 9% 38% 16% 2% 7% 21% 2% 15% 21% 25,062 1,949 243 1,229 377 265 4,331 7,180 6,337 2,950 France 3% 1% -4% 0% 0% -9% 2% 1% 7% 3% 39,142 3,408 525 1,912 442 357 5,651 10,183 10,626 5,845 Rest of World 3% -1% -2% 5% 12% 11% 0% 6% 2% 2% 214,349 27,861 2,895 16,466 3,047 2,071 23,715 56,116 43,358 36,870 USA 2% -4% -12% -3% -7% 4% 1% -3% 9% 5% 41,024 3,859 592 2,771 571 585 5,456 9,836 11,444 5,658 All 0% -3% 0% 1% 1% 0% -1% -1% 2% -2% 622,628 56,551 7,112 36,199 7,559 5,704 80,308 166,556 158,213 100,215 Netherlands -1% 2% 0% 7% -28% -5% -10% -7% 6% 6% 29,022 3,179 466 1,559 482 597 4,976 6,914 7,260 3,438 Sweden -3% 8% 5% -11% 28% 28% 0% -9% -5% 4% 17,535 732 101 793 154 116 1,931 4,938 6,293 2,374 Austria -11% -10% 15% -11% 4% 23% -21% -15% -1% -18% 15,782 739 128 617 89 32 1,823 4,822 4,860 2,589 Korea -11% -28% -48% 3% 59% -35% 16% -15% 12% -57% 14,841 478 64 436 107 22 1,634 4,947 4,320 2,798 Switzerland -11% -26% -11% -14% 9% -43% -19% -16% -3% -6% 19,748 964 213 639 123 50 3,578 4,872 5,971 3,245 UK -13% -10% -3% -6% -15% 5% -21% -14% -11% -12% 39,904 2,817 564 2,006 530 391 6,107 10,533 10,857 5,871 Canada -19% -14% -16% -8% -4% -5% -36% -32% -14% -10% 15,718 1,569 125 1,131 148 103 1,772 3,773 4,782 2,169 Japan -20% -40% -17% -11% 0% -29% 4% -32% -16% -34% 27,744 884 117 1,013 233 145 3,663 7,494 9,772 4,274 Singapore -26% -29% -11% -21% -39% -45% -3% -40% -16% -36% 12,794 1,059 120 620 196 314 1,336 2,826 3,980 2,240 Hong Kong -35% -48% -27% -55% -8% -26% -52% -34% -24% -35% 12,270 605 61 419 42 39 1,011 3,458 2,993 3,565 Taiwan -37% -95% -49% -26% -20% -10% -21% -31% -24% -62% 14,640 625 34 590 85 51 1,577 4,434 4,156 3,034 Table B3: Percent change of the gravity regression residual from the early period (1980-1992) to the late period (1993-2004) - Specification (iv)

All SITC 0 1 2 3 4 5 6 7 8 China 61% 48% 17% 11% 12% 6% 26% 63% 74% 93% 18,461 1,313 133 1,055 217 98 2,428 5,594 3,989 3,566 Mexico 21% 31% 72% -20% -47% -40% 15% 19% 28% 36% 6,751 504 107 459 88 25 1,324 1,644 1,592 972 Spain 6% 30% 43% 2% -12% 57% 4% 5% -5% 16% 22,178 1,805 275 1,039 267 183 3,270 6,629 5,439 3,140 Rest of World 1% 12% 26% -6% 4% 26% 4% 4% -8% -3% 207,430 26,969 2,697 16,052 2,880 2,037 23,261 54,220 42,659 34,740 Italy 1% 23% 52% -2% -30% 22% -4% 12% -15% 8% 34,256 2,042 330 1,470 343 226 4,631 9,830 9,947 5,277 Belgium -2% 19% 58% 0% -15% 7% 9% -11% -9% 4% 24,799 1,909 243 1,226 385 264 4,286 7,123 6,279 2,889 All -5% 11% 29% -9% -11% 15% -4% -5% -13% -7% 613,553 55,439 6,848 36,080 7,449 5,659 79,968 164,164 156,967 96,809 USA -6% 9% 14% -13% -28% 16% -2% -8% -12% -1% 43,270 4,084 591 3,060 645 618 5,909 10,462 11,924 5,717 Korea -6% -10% -21% -6% 48% -16% 16% -8% 18% -53% 14,608 470 69 434 108 22 1,613 4,855 4,270 2,730 Sweden -8% 36% 44% -18% 25% 43% 1% -11% -19% -2% 17,348 731 103 799 154 116 1,941 4,860 6,235 2,306 France -10% 13% 27% -11% -14% 1% -7% -13% -17% -13% 38,088 3,307 507 1,894 438 355 5,498 9,950 10,316 5,633 Netherlands -11% 12% 29% -2% -39% 3% -19% -17% -15% -8% 28,499 3,105 453 1,533 480 595 4,867 6,802 7,168 3,349 Switzerland -16% 8% 33% -21% 12% -28% -18% -19% -20% -14% 19,667 954 209 643 122 51 3,649 4,865 5,904 3,170 Singapore -19% -10% 23% -30% -61% -24% 8% -29% -12% -34% 12,981 1,052 111 632 193 304 1,448 2,872 4,054 2,218 UK -20% 13% 32% -10% -28% 23% -29% -19% -31% -20% 39,476 2,810 561 2,005 518 396 6,028 10,482 10,742 5,703 Austria -24% -2% 54% -23% -10% 10% -26% -26% -24% -34% 15,024 678 121 589 81 26 1,719 4,616 4,651 2,459 Canada -26% -1% 19% -24% -28% -14% -29% -33% -33% -17% 15,643 1,550 123 1,130 153 103 1,771 3,767 4,780 2,125 Taiwan -28% -74% -18% -34% -42% -3% -17% -18% -20% -50% 15,017 636 36 601 91 54 1,619 4,575 4,280 3,071 Japan -32% -20% 13% -18% -8% -19% -1% -41% -36% -45% 27,598 909 122 1,017 239 146 3,654 7,496 9,689 4,180 Hong Kong -34% -28% 6% -68% -34% -41% -48% -30% -31% -35% 12,459 611 57 442 47 40 1,052 3,522 3,049 3,564 Table B4: Percent change of the gravity regression residual from the early period (1980-1992) to the late period (1993-2004) - Specification (v)

Industrial Other COUNTRY All Sectors Food Beverages Tobacco Textiles Apparel Leather Footwear Wood Furniture Paper Printing Chemicals Chemicals Petroleum China 80.6 95.6 23.4 77.1 45.0 84.3 40.6 - 147.0 185.8 106.8 152.6 70.1 128.5 123.6 Australia 67.1 21.4 50.2 6.0 34.8 0.4 46.5 -9.7 -73.5 -56.8 -23.9 11.1 189.6 63.5 15.3 Canada 57.4 27.0 37.8 -3.9 17.5 19.7 -70.3 -6.9 114.6 -32.6 66.5 34.6 64.9 0.3 30.4 Mexico 56.6 -33.7 -33.0 -334.3 38.7 93.9 62.3 9.2 87.9 214.6 -152.2 -16.4 -68.8 -58.2 160.3 Indonesia 50.6 44.8 -86.7 - 31.4 9.7 -75.5 8.3 86.9 161.9 125.1 -132.3 43.9 5.5 - Taiwan 47.2 56.0 13.4 126.7 61.2 61.0 - -34.1 32.3 55.6 62.0 131.2 99.2 36.9 49.1 Korea 46.8 70.2 -26.0 52.2 36.0 -158.5 36.1 -438.6 88.7 55.4 99.1 113.3 82.0 53.5 94.7 India 38.9 24.1 -215.7 25.6 29.0 95.2 129.0 42.3 44.2 13.4 19.2 -42.2 14.5 28.8 -66.4 Austria 38.8 20.6 -78.9 -7.0 12.6 66.1 87.2 59.6 -58.1 -31.0 25.8 19.7 -3.2 78.0 26.8 Spain 27.7 25.1 55.9 2.5 11.9 14.4 -12.9 20.1 49.6 1.0 17.5 44.3 14.1 11.0 29.7 UK 20.1 9.6 53.3 13.0 2.5 15.8 20.2 43.0 38.4 -15.1 0.1 33.2 16.2 24.8 10.0 Finland 0.5 -6.2 -19.4 126.7 16.6 -249.0 87.2 0.4 45.8 -112.3 -29.5 -12.1 0.7 12.5 -18.3 Turkey 0.4 -15.1 55.7 108.1 17.0 34.5 -67.2 14.2 16.3 -180.4 -38.4 -240.1 -54.0 -44.5 -175.0 Greece 0.0 -0.4 -44.3 45.1 18.8 - -81.5 14.3 47.3 -65.8 -44.2 -115.1 -142.8 -10.7 -21.7 South Africa -2.3 -74.2 -157.8 -154.5 -494.7 21.2 14.2 -56.3 -221.6 -44.6 -105.0 -289.3 89.2 -150.8 -7.5 Ireland -2.7 86.4 118.4 24.3 44.8 40.1 106.7 82.2 -206.2 -59.0 -66.9 82.6 - 140.8 - Denmark -4.5 24.2 31.9 -13.9 29.4 - - 65.3 -7.6 203.2 -2.0 14.1 -106.5 78.9 -19.8 USA -5.6 21.2 45.7 37.2 2.1 8.9 -8.6 42.4 50.0 1.5 5.7 51.2 -46.9 -13.4 42.1 Sweden -7.6 2.5 -28.2 -3.7 25.0 84.0 -324.3 73.8 48.5 162.2 -32.3 -4.5 -30.5 82.4 99.3 Germany -9.5 10.4 33.1 28.0 38.2 27.1 102.1 41.0 43.5 -27.6 23.5 12.4 -168.6 26.1 21.0 Argentina -9.6 12.0 -12.6 - 37.4 -24.8 55.4 8.0 -203.6 -38.5 -53.5 -99.7 -52.4 -65.7 42.1 France -18.3 1.2 17.9 21.9 2.9 12.6 79.2 27.8 30.4 -21.0 -5.9 19.0 -57.2 -21.6 11.6 Portugal -33.2 13.4 -103.7 -19.1 7.1 -328.7 -39.4 -40.5 -18.6 -101.1 48.8 -42.8 6.3 -9.5 79.2 Japan -50.2 -10.7 14.6 10.0 -12.8 -13.6 -69.9 25.1 21.3 -33.8 -47.7 15.4 -82.3 -53.5 17.3 Italy -65.5 -8.1 5.8 -29.0 -11.9 -14.4 -117.2 -72.5 10.1 -113.5 -24.4 9.5 -24.1 -37.3 -18.0 Belgium - 42.4 54.7 36.0 31.3 47.3 - 81.8 -67.8 -0.8 118.1 13.1 172.2 - -34.5 Brazil - -281.3 -214.3 -224.7 -66.7 -23.1 - - -205.0 27.2 -356.9 -46.5 -204.4 - - Hong Kong - 172.3 288.0 - - 76.0 - - - - - 93.7 - - - Netherlands - 24.4 120.8 49.8 -5.2 - - - 59.1 -62.5 88.0 39.7 179.0 80.7 -156.9 Singapore - -375.0 - - - - - - - -85.4 176.5 150.6 - -387.1 -334.4 Table B5: Structural estimates of revealed overall competitiveness, change from early to late period.

Nonmetal Fabricated Other Electric Scientific Other COUNTRY All Sectors Fuels Rubber Plastic Pottery Glass Minerals Iron & SteelOther Metal Metal Machinery Machinery Transport Equipment Mnfg. China 80.6 - 107.2 125.9 77.0 84.1 106.7 94.1 33.9 185.9 108.8 59.7 11.8 63.1 41.1 Australia 67.1 25.4 -33.6 -226.9 -10.6 -120.2 28.0 145.4 183.2 11.8 34.8 -27.5 -99.3 55.7 10.2 Canada 57.4 46.6 19.7 -63.3 27.1 32.1 -17.8 -13.6 133.8 20.1 49.2 30.9 55.8 42.6 30.1 Mexico 56.6 35.2 -24.3 -4.5 93.1 31.6 -40.9 -18.8 78.1 266.2 149.9 - 84.8 - - Indonesia 50.6 - 33.1 6.1 -26.0 90.6 -34.9 -208.3 - -96.6 147.5 103.4 -131.1 56.8 41.7 Taiwan 47.2 3.3 111.2 51.5 -478.4 68.2 40.5 76.2 5.7 148.5 - 60.6 13.8 69.6 40.9 Korea 46.8 56.1 122.2 107.1 -11.7 55.5 81.1 48.0 92.5 76.5 48.1 55.9 44.7 51.9 31.4 India 38.9 -294.6 40.3 49.1 -11.3 -65.3 67.4 7.6 -28.8 107.6 23.5 -15.6 25.3 42.2 - Austria 38.8 32.2 68.2 47.9 15.0 41.9 -3.2 36.6 -13.9 1.3 92.7 -21.1 121.0 -0.9 - Spain 27.7 42.1 67.9 61.1 25.9 28.4 29.3 -2.3 -24.7 -9.7 -9.2 14.5 59.1 26.0 32.2 UK 20.1 85.7 3.6 37.5 24.2 9.3 15.9 11.9 0.6 -32.7 22.9 66.2 12.2 -55.8 37.1 Finland 0.5 -188.5 93.4 -37.7 9.5 4.5 -2.8 51.7 17.7 -35.3 -10.8 13.0 38.1 26.1 30.8 Turkey 0.4 64.5 -85.1 3.6 6.7 35.0 16.1 -20.3 -20.2 -128.0 16.1 10.2 -0.6 -25.7 38.1 Greece 0.0 - -223.6 33.2 -72.5 22.9 166.0 -118.6 152.7 -22.6 36.3 -35.8 -19.2 61.9 36.7 South Africa -2.3 -175.4 -133.9 -357.8 -13.4 -449.5 -441.1 114.8 78.9 -200.0 12.2 -176.9 -46.9 37.4 12.9 Ireland -2.7 - 136.3 91.3 47.4 27.1 47.8 -205.9 - 74.2 -273.2 152.9 98.4 -22.7 - Denmark -4.5 66.6 33.5 -113.0 54.9 -27.3 3.7 150.5 - -31.6 -74.8 56.2 57.7 -425.1 23.2 USA -5.6 49.3 -10.8 55.8 16.1 25.6 24.0 -18.2 -50.4 0.1 -9.4 -19.9 -1.9 34.6 29.9 Sweden -7.6 33.2 68.7 6.6 77.5 12.6 12.8 23.2 41.3 -43.8 -55.9 13.9 -28.6 -7.7 -378.9 Germany -9.5 31.4 45.7 38.0 17.9 24.1 -0.2 12.9 -12.3 -28.0 5.8 -4.3 -6.5 30.8 39.6 Argentina -9.6 - -93.5 -138.8 10.0 -19.9 -61.5 -104.8 -58.3 -54.3 -29.0 -238.5 14.5 32.0 24.1 France -18.3 40.8 -66.0 30.2 - 0.9 -4.3 3.5 -48.6 -59.0 -3.8 -26.0 -33.2 -0.1 30.7 Portugal -33.2 -91.5 59.5 38.9 29.2 25.6 19.4 9.8 -81.9 10.4 -75.7 53.4 66.9 -149.7 16.5 Japan -50.2 41.8 -62.2 21.3 -2.8 6.8 -7.0 -64.4 -82.6 -64.9 -49.9 -88.7 -32.7 29.0 25.6 Italy -65.5 - -21.4 9.6 47.5 -1.4 -58.4 -16.4 -102.4 -160.6 -149.5 -64.3 -0.3 22.2 -193.8 Belgium - 51.8 - 52.0 4.7 -17.3 -97.6 - -294.3 -6.7 172.1 126.7 - - - Brazil - - -256.2 -5.6 -14.4 - - -176.8 - - -178.6 -247.8 -413.5 5.7 - Hong Kong - - - - - - 27.5 - - - - - - - - Netherlands - 44.0 - 31.5 8.3 74.1 -2.3 182.4 - -26.2 - 148.8 124.5 - - Singapore - - - 49.4 48.8 - 86.0 - - 97.2 - - -14.5 - - Table B5 cont'd.: Structural estimates of revealed overall competitiveness, change from early to late period.

80% 60% 40% 20% 0% ‐20% ‐40% ‐60% UN‐NBER (SITC 4‐digit) CEPII (ISIC 2‐digit) Figure B1: Comparison of gravity residual changes using UN-NBER and CEPII data