What Does Trade Openness Measure?

6656 2017 September 2017 What Does Trade Openness Measure? E Fu

Impressum: CESfo Workng Papers ISSN 2364 1428 (electronc verson) Publsher and dstrbutor: Munch Socet for the Promoton of Economc Research CESfo GmbH The nternatonal platform of Ludwgs Maxmlans Unverst s Center for Economc Studes and the fo Insttute Poschngerstr. 5, 81679 Munch, German Telephone +49 (0)89 2180 2740, Telefax +49 (0)89 2180 17845, emal offce@cesfo.de Edtors: Clemens Fuest, Olver Falck, Jasmn Gröschl www.cesfo group.org/wp An electronc verson of the paper ma be downloaded from the SSRN webste: www.ssrn.com from the RePEc webste: www.repec.org from the CESfo webste: www.cesfo group.org/wp

CESfo Workng Paper No. 6656 Categor 8: Trade Polc What Does Trade Openness Measure? Abstract An emprcal measure of trade openness s defned as the rato of total trade to GDP, and represents a convenent varable routnel used for cross-countr studes on a varet of ssues. However, the effects that the crude measure captures reman ambguous, makng t dffcult to nterpret the emprcal results. Drawng on several strands of the lterature, ths stud examnes the nformatonal content of the trade openness measure usng ntranatonal and nternatonal data. We fnd that, even for full ntegrated economes wthn a countr, trade openness s approxmatel half as varable as t s for segmented dverse countres around the world. The nformaton t conves s better characterzed as the extent of the economc remoteness and dosncratc dstrbuton of sectoral producton. The cross-countr varaton of trade openness derves more from the varablt n GDP than trade. JEL-Codes: F400, F140. Kewords: trade openness, specalzaton, gravt model, market ntegraton, prce devatons, remoteness. E Fu School of Economcs Kwanse Gakun Unverst 1-155 Uegahara Ichban-cho Japan Nshnoma, Hogo 662-8501 efu@kwanse.ac.p Ths verson: September 7, 2017 The author thanks Ken Fuwara and Mark Spegel for helpful dscussons and comments. All remanng errors are solel the author s.

1. Introducton An emprcal measure of trade openness, defned as the rato of exports plus mports to GDP, s a convenent varable routnel used n a varet of nternatonal macroeconomc studes. Use of the varable s pervasve n cross-countr studes on a broad range of topcs, ncludng growth analses (Levne and Renelt, 1992), real exchange rate dnamcs (Goldfan and Valdéz, 1999), government sze (Rodrk, 1998), output volatlt (Fatás and Mhov, 2001), soveregn debt defaults (Lev Yeat and Panzza, 2011), and the poltcal econom of electons and reforms (Cermeño, Grer, and Grer, 2010). 1 Albet an admttedl crude measure of openness, the fact that ths varable s wdel used n the lterature hghlghts the mportance of understandng the nformatonal content that t conves. The prevalent use n the lterature of trade openness seems partl a result of data avalablt. Because data on total trade as a percentage of GDP are avalable for man countres from standard nternatonal databases, researchers ma fnd them useful n conductng cross-countr analses. However, despte ths pervasveness, the effects that the trade openness varable actuall captures reman ambguous. Although t ma well be an ntended prox measure, the queston s for what exactl s the varable a good prox. For nstance, what effects are we effectvel controllng for b ncludng the trade-to-gdp ratos on the rght-hand sde of regresson equatons for cross-countr dfferences n growth, real exchange rate dnamcs, and so on? One exemplar macroeconomc nterpretaton of the trade openness varable s that t measures the degree to whch a domestc econom s exposed to external shocks. To the extent that an econom reles on export demand and mported nputs/products, t s 1 Some studes refer to the trade-to-gdp rato as a trade share (Frankel and Romer, 1999) or a trade ntenst rato (Leamer, 1988) nstead of trade openness. 1

subect to potental transmsson of dsturbances from abroad. However, the trade-to-gdp rato vares across countres because of dfferences n trade polces, factor endowments, and geographcal locatons rrespectve of realzatons of external shocks. Thus, usng varatons n trade-to-gdp ratos seems ll-suted for measurng the extent to whch external shocks affect domestc economes. 2 Conceptuall, trade openness ma be defned as the degree to whch an econom mantans ts outward orentaton n trade. However, emprcall, adoptng ths defnton s challengng because t requres detaled and consstent data for man countres on the extent of explct and mplct trade mpedments n varous forms that are product-, destnaton-, and orgn-specfc and tme-varant. Even f such data are avalable, an addtonal hurdle exsts. Aggregatng the detaled data nto an overall ndex that qualfes as a unversal measure of trade openness s dffcult (Harrson, 1996). In fact, although varous ndcators are nvented, Prtchett (1996) showed that the are vrtuall uncorrelated wth each other, castng doubt on ther consstenc and relablt. 3 Although the aforementoned ssues ma tempt researchers to seek a more practcal alternatve, usng trade-to-gdp ratos as a measure of trade openness ntroduces sgnfcant ambgut to the nterpretaton of ts effects. The sze of trade n relatve terms to GDP depends on a varet of factors. In addton to the extent of the outward orentaton of trade polc, the nclude the szes of the domestc and external markets, the dstances to consumers and from producers outsde one s own terrtor, factor endowments that nduce specalzaton n producton, and households preferences for a 2 To that end, measures based on changes n terms of trade and/or trade-weghted real effectve exchange rates seem more sutable. However, ther data avalablt s lmted, especall for less developed countres. 3 For a dscusson on the ssue, see also Harrson and Rodríguez-Clare (2010), Edwards (1993, 1998), and Rodrguez and Rodrk (2000). 2

varet n consumpton. As a composte of the numerous factors, the emprcal measure of trade openness, as convenent as t ma be, nflcts dffcult on the nterpretaton of ts effects as documented n varous contexts. 4 Ths dscusson motvates us to examne the nformatonal content of the trade openness varable b drawng on several strands of the lterature. In dong so, we analze both ntranatonal and nternatonal data. In partcular, we use Japan s prefecture-level data as, so to speak, a mnature world n whch assured free trade purges the ambgut arsng from falng to measure varous trade restrctons n nternatonal contexts. Furthermore, ntra-japan data on trade, sectoral output, product prces, and so on allow us to evaluate several canddates for the effects captured b the trade openness varable, whch s dffcult f not nfeasble to do n nternatonal contexts. Namel, the are economc remoteness (Anderson and van Wncoop, 2003; Baldwn and Harrgan, 2011), market ntegraton (Engel and Rogers, 1996; Parsel and We, 2001; Goldberg and Verboven, 2005), unqueness n sectoral producton dstrbutons (Krugman, 1991; Kaleml-Ozcan, Sørensen, and Yosha, 2003), and nclnaton to pursue economes of scale effects (Alesna, Spolaore, and Waczarg, 2000; Alesna and Waczarg, 1998). Through the exercses, ths stud contrbutes to the lterature b sheddng lght on the nformatonal content of the wdel used albet dffcult-to-nterpret emprcal varable. To antcpate, our chef fndngs are as follows. Even for the full ntegrated (and hence perfectl open) regonal economes n Japan, the measured trade openness s approxmatel half as varable as the trade openness of 171 segmented dverse countres around the world. The ntranatonal varaton of trade openness embodes two prmar 4 See Leamer (1988) for related problems regardng trade-to-gdp ratos as a measure of trade openness. 3

nformatonal contents: the extent of economc remoteness and that of dosncras n sectoral producton dstrbutons. We fnd lttle evdence of an assocaton between the extent of trade openness and market ntegraton n terms of prce converson n tradables. The varance decomposton exercse we conduct reveals that GDP sze varablt matters more than does trade to the varablt of the trade openness of the nternatonal data. Altogether, these results mpl that a substantal part of the nformaton conveed b the cross-countr varaton n the popular trade openness measure has lttle to do wth openness n the sense of polc outward orentaton or the extent of market ntegraton. The remander of ths paper s organzed as follows. Secton 2 dscusses the trade openness measure from the perspectve of the gravt model. B drawng on a few strands of the lterature, secton 3 consders four possble effects that the trade openness varable ma reflect and how the can be alternatvel measured. Secton 4 descrbes the data and presents a prelmnar analss. In secton 5, we examne the assocaton between trade openness and the alternatve measures suggested n secton 3. After conductng a varance decomposton analss n secton 6, we conclude n secton 7. 2. Trade openness from the gravt perspectve As an emprcal measure of trade openness, a plethora of studes adopt OPEN = x + m = x ( k,, k +,, k m ), (1) where x, m, and are countr s exports, mports, and GDP, respectvel. Wth and k denotng partner countres and sectors, respectvel, x,, ( k m, k, ) s s exports to 4

(mports from) n sector k. Tme subscrpts are suppressed for brevt. As (1) makes explct, the varable s constructed through aggregaton n multple dmensons: a domestc countr s exports and mports are combned whle also beng aggregated over all other countres (.e., the rest of the world) and all sectors. Ths measurement method s clearl dscordant wth the mcroeconomc approach of the emprcal trade lterature (Harrgan, 1996) that assesses the extent of openness b carefull dstngushng exports and mports, destnaton/orgn countres, and sectors. 5 For macroeconomc analses, the sectoral aggregaton over k n (1) ma be ustfed on the condton that the resultng measurement s taken at best as an approxmaton n aggregate terms. A crucal queston s what prox nformaton can be effectvel conveed through a varaton of (1) across s. As a startng pont to address the queston, we consder aggregate blateral trade. In partcular, a useful framework can be drawn from the gravt equaton of trade (Anderson and van Wncoop, 2003; Head and Maer, 2014). Specfcall, we refer to the gravt equaton of blateral trade derved b Anderson and van Wncoop (2003) 1 σ t, x =, (2), W p p where x, s total exports from countr to, t, denotes the blateral trade cost factors, ( ) and p ( p ) are the ncome and prce ndces, W s the world ncome, and σ s the CES parameter. 6 Rearrangng terms n (2) elds 5 Harrgan (1996) assessed the extent of openness to trade n manufacturers n OECD countres b estmatng mport equatons b subsector whle allowng for both mporter and exporter fxed effects. 6 Anderson and van Wncoop (2003) noted that resstance varables that depend on p and p are multlateral t, and should not be narrowl nterpreted as 5

6 σ = 1,, W p p t x. (2) Smlarl, we obtan σ = 1,, W p p t x. (3) From (2) and (3), σ σ + = + 1, 1,,, W W p p t p p t x x. (4) In the case that blateral trade barrers are smmetrc such that t t,, =, 7 σ = + = + 1,,,,, 2 W p p t m x x x. (5) When aggregated over all s, (5) elds an expresson equvalent to (1) that = + = + W p p t m x m x σ 1,,, 2. (6) Accordng to (6), s trade openness (1) s determned b the szes of ts tradng partners economes relatve to the world, trade-resstant factors (.e., the term n the brackets), and the elastct of substtuton parameter. Ths gravt exposton of trade openness consttutes a base for the analses we subsequentl conduct. We note that defnng as the rest of the world generates sgnfcant hurdles for emprcal mplementatons n nternatonal contexts. As dscussed n the ntroducton, the trade-resstant factors encompassng explct restrctons and mplct mpedments n varous forms are dffcult to measure consstentl. Further, the smmetr assumpton consumer prce ndces. 7 The smmetr assumpton, whle generall nvald for nternatonal trade, s vald for ntranatonal trade.

t t,, = generall does not hold. We get around these problems b conductng ntranatonal analses usng Japan as, so to speak, a mnature world n whch the prefectural economes engage n mutuall free trade. 3. Alternatve nterpretatons and measurements For a gven econom, (1) s smpl the rato between total trade and GDP. To allow for economc nterpretaton, t s necessar to consder what makes the rato of trade to GDP hgh or low, and what t means for an econom to have a hgh or low rato. We dscuss several canddates b drawng on dfferent strands of the lterature. For each, we also ntroduce an alternatve measurement that well fts the nterpretaton. 3.1 Economc remoteness Assumng that the CES parameter σ s constant across s, the gravt exposton of trade openness (6) conssts of two ke elements: the relatve sze of tradng partners and the trade-resstant factors. In other words, the extent of trade openness s determned b how large the partners are and the severt of the trade mpedments lng between domestc and partner economes. As dscussed n secton 1, t s not straghtforward to consstentl gauge nternatonal trade restrctons n a varet of forms. However, ntranatonall, no artfcal trade barrers exst and the sole mpedment s trade cost, whch tends to ncrease wth nterregonal dstance. The nformaton on partner sze and dstance thereto can be succnctl summarzed b the remoteness ndex proposed b Baldwn and Harrgan (2011) REM = [ ] 1 d, η, (7) 7

where d, s the dstance between economes and. Followng Baldwn and Harrgan (2011), we set η equal to unt. 8 A larger value of (7) ndcates a greater extent of economc remoteness n the sense that the tradng partners are smaller and/or located farther awa. 9 Thus, the gravt perspectve suggests a hpothess that the prmar nformaton emboded n (1) be the extent of economc remoteness. If so, (1) should be sgnfcantl negatvel assocated wth the drect measure of remoteness (7). 3.2 Market ntegratng forces n tradables Although the gravt model provdes sgnfcant nsght, other strands of the lterature offer dfferent perspectves on what can be conveed b varatons n (1). In nfluental studes on natonal prce levels, Kravs and Lpse (1987, 1988) use the trade openness to gauge the degree to whch nternatonal trade forces drve traded goods prces toward unformt across countres. Use of trade openness contnues to be popular n the subsequent lterature on natonal prce levels and real exchange rates. 10 From the aforementoned perspectve, trade openness ndcates the strength of the forces that ntegrate the domestc market n tradables wth external markets. In other words, (1) can be regarded as a quantt-based prox measure of market ntegraton of tradables. In general, the extent of market ntegraton s better measured b prce-based nformaton, such as ntermarket prce dfferentals and ther varances (Engel and Rogers, 1996; Parsel and We, 2001; Goldberg and Verboven, 2005). If the trade openness varable reasonabl captures the extent to whch the forces of arbtrage appl 8 See Head and Maer (2014) for ssues regardng varous remoteness measures proposed n the lterature. 9 An mplct assumpton s that tradng costs ncrease wth dstance. 10 See, for nstance, Cheung and La (2000), Broda (2006), and Fu (2015), among others. 8

convergng pressure on the prces of tradables, then t should be negatvel assocated wth the sze of ntermarket prce dfferentals. We gauge the prce devaton n the tradable sector k of econom from others b 1 2 DEV, k = ( P, k, ) 1 P k, (8) N where P, s the prce of k n and N s the number of all economes. The prevous k dscusson leads us to a hpothess that more trade-open economes tend to have smaller tradables prce devatons and, hence, (1) and (8) are sgnfcantl negatvel assocated wth each other. 3.3 Idosncratc factor endowment and sectoral specalzaton Unlke the gravt model, the Heckscher-Ohln model hghlghts the roles of heterogeneous factor endowments n motvatng specalzaton and trade. From ths conventonal perspectve, the trade openness varable can be vewed as an mplct ndcator of the degree of specalzaton n producton nduced b heterogeneous factor endowments. The pont s well artculated theoretcall b Leamer (1988) that, n the absence of trade barrers, (1) s essentall a measure of the peculart of the resource suppl. Whle drectl measurng factor endowment s dffcult, the extent of specalzaton can be quantfed b usng sectoral GDP data. More specfcall, we adopt the ndex of specalzaton proposed b Kaleml-Ozcan, Sørensen, and Yosha (2003) SP where =, k, k k N 1 1 2, (9), and k, respectvel, are the sector k and total outputs of. 11 11 The dea of the specalzaton ndex orgnates from Krugman (1991). The ndex he 9

In theor, specalzaton and trade go hand n hand. However, the relatonshp can be somewhat complcated n the emprcal measure. Precsel, (9) quantfes the extent of dosncras n s sectoral output dstrbuton vs-à-vs others. Thus, f economes specalze n dfferent sectors to engage n nterndustr trade wth each other, then the extent of specalzaton (9) should be postvel assocated wth the trade openness of (1). However, f economes specalze n the same sectors to engage n ntrandustr trade, then a hgh trade-to-gdp rato can occur wth a low specalzaton ndex value. In ths case, (1) and (9) wll not be postvel assocated wth each other. The pont s that f ntrandustr trade domnates nterndustr trade, then the trade-to-gdp rato ma not be postvel assocated wth the sectoral specalzaton ndex. In fact, the ma exhbt a negatve correlaton. A dsuncton between specalzaton and trade can potentall (though probabl less lkel) occur also when economes dffer substantall n ther preferences. If domestc producton n each econom takes place prmarl to meet the local demand shaped b regon-specfc heterogeneous preferences, then a dstnctve sectoral producton structure captured b (9) does not necessarl lead to a greater trade-to-gdp rato (1). 3.4 Inclnaton to pursue economes of scale Another nterpretaton of the trade openness varable derves from the lterature on the relatonshp between economc ntegraton and the equlbrum sze and number of countres (Alesna and Spolaore, 1997; Alesna, Spolaore, and Waczarg, 2000). To the proposed adopts sectoral shares n emploment rather than output and absolute values nstead of squares. We rel on output data because the correspondng data on sectoral emploment are not avalable. Kaleml-Ozcan et al. (2003) used (9) to assess the extent of specalzaton wthn manufacturers b lmtng k as manufacturng subsectors and defnng as total manufacturng output. In our ntra-japan context, products and servces other than manufacturers can also be tradeable. Thus, we consder all sectors for k. 10

extent that returns are ncreasng wth the scale of producton, smaller countres suffer more severel from trade restrctons because ther productvtes sta low gven lmted domestc market sze. In contrast, large countres can eno economes of scale n ther domestc markets. Therefore, smaller economes have stronger ncentves than larger ones to become more open to trade. 12 Therefore, an econom s sze matters to trade openness even after trade s scaled b GDP as (1). Ths argument suggests that the rato of trade to GDP can be drven b the nclnaton of a countr to expand ts market sze beond ts own terrtor to pursue economes of scale effects. Ths perspectve consttutes a credble explanaton for wh small economes such as those of Luxembourg and Sngapore are ver open, whereas large ones such as those of the Unted States and Japan are far less so based on trade-to-gdp ratos. When faced wth severe lmtatons on domestc market sze, economes have strong ncentves to pursue specalzaton and expand trade. Consequentl, the adopt a lberal trade polc. In the lterature, a standard market sze measure s populaton. 13 Followng ths conventon, we use logged total populaton as a measure of the nclnaton to explot the economes of scale effects. If the trade openness varable effectvel reflects the strength of the ncentves to explot the scale effects, t should be negatvel assocated wth domestc market sze as approxmated b populaton sze. 12 Alesna and Waczarg (1998) argued that the postve assocaton between openness and government sze hghlghted b Rodrk (1998) s drven b the countr sze effect and should not be vewed a drect causal relatonshp. Ram (2009) provded counter evdence to the argument. 13 We also consder GDP as a measure of market sze. Not surprsngl, GDP and total populaton exhbt strong postve correlatons, and ther replacement exerts lttle mpact on the emprcal results. 11

4. Data and prelmnares 4.1 Data descrpton In nternatonal contexts, (1) measures s trade (n relatve GDP terms) vs-à-vs the rest of the world. Thus, analzng ts contents requres data on the other varables also n relaton to the rest of the world, whch ma not be obtanable. For nstance, t s nfeasble to secure data on prce devatons n tradables and/or sectoral specalzaton for all countres around the world. The ntranatonal approach we pursue makes t feasble to examne the effects, whch s otherwse dffcult to do. The other mert of usng ntranatonal data s that t enables us to examne the behavor of the trade openness varable n a controlled envronment n whch trade polces/restrctons pla no role and all economes are assured to be unforml and full open to trade wth each other. Under these crcumstances, t s useful to examne the varaton that the measured trade openness stll exhbts and what t reflects. Ths nformaton should be taken nto account when nterpretng the effects of trade openness n nternatonal contexts n whch addtonal sources of varatons, such as dfferences n trade polc, come nto pla. The ntranatonal data we analze are Japanese prefecture-level data. Japan conssts of fort-seven prefectures, lsted n Table A-1 n the appendx, whch are geographcall defned admnstratve unts largel correspondng to the states n the Unted States. Although no artfcal trade barrers exst between the prefectures, natural mpedments do. In general, transport cost ncreases wth dstance and, hence, remoteness can become a maor trade-resstant factor. The Annual Report on Prefectural Accounts publshed b the Cabnet Offce of Japan provdes prefectural ncome accountng data. From the 2012 CD-ROM of the 12

report, we extract data on prefectural GDP n total and b sector, mport from and export to all other prefectures, and populaton. 14 The sample perod s from 1996 to 2009. The sectors for whch dsaggregated output data are avalable are lsted n the data appendx. Unfortunatel, the sectoral GDP seres for the Oknawa prefecture are ncomplete. Consequentl, the specalzaton ndex (9) s constructed for fort-sx prefectures except for Oknawa. To quantf the extent of economc remoteness b (7), we gauge the nterprefecture dstances (n klometers) b those between the prefectural captals usng a calculator avalable at the webste of the Geospatal Informaton Authort of Japan, Mnstr of Land, Infrastructure, Transport and Toursm. To measure prce devatons n tradables b (8), we use the Regonal Dfference Index of Prces (RDIP) publshed b the Statstcal Bureau of the Mnstr of Internal Affars and Communcatons. Usng nformaton collected b the natonal prce surve, the RDIP reveals regonal prce level dfferences as ndces to the natonal average. The surve takes places n fve-ear ntervals to consstentl record prces of precsel defned products and servces throughout the countr. Although the RDIP seres s avalable for varous product categores, our purpose s to measure prce devatons n tradables. 15 Thus, we adopt the RDIP of the followng hghl tradable categores: clothng and footwear; and furnture and household utensls. 14 Because of a lack of data, measurng the extent of openness to nternatonal trade b prefecture s not possble. However, we note that Japan s among the least open countres to nternatonal trade, wth an average trade-to-gdp rato of onl 24% for 1996 2009. The openness for nterprefectural trade s far greater, at approxmatel 133%. These facts suggest that trade vs-à-vs other prefectures generall has far more sgnfcant mplcatons for prefectural economes. 15 RDIP data are avalable for the followng categores: general; general except for fresh foods; general except for rents; foods; housng; utltes (.e., water and electrct); furnture and household utensls; clothng and footwear; medcal servces; 13

Durng the 1996 2009 perod covered b the prefectural ncome accountng data, the RDIP are avalable onl for 1997, 2002, and 2007. Hence, we use the observatons of these three ears for the analses requrng (8). We also use standard nternatonal databases, specfcall the World Bank s World Development Indcator (WDI) and the IMF s Drecton of Trade Statstcs (DOT). The WDI data on total trade as a percentage of GDP are based on trade vs-à-vs the rest of the world. We use DOT data to calculate trade wthn the OECD, the EU, and the Eurozone. 4.2 Trade openness b ntranatonal and nternatonal data Panel A of Table 1 presents the descrptve statstcs for the 1996 2009 average trade openness of the fort-seven Japanese prefectures. On average, ntra-japan trade openness s approxmatel 130%. Of the fort-seven prefectures, the least open s the remote sland of Oknawa (63%) and the most open s geographcall centrall located Me (215%). How do the fgures compare wth nternatonal counterparts? We report the correspondng nternatonal statstcs n panel B of Table 1. After extractng the data for all countres from the WDI database, we retan 171 countres that have no mssng observatons for 1996 2009. Not surprsngl, the average trade openness of the 171 countres around the world s substantall lower at 88%. Further, the dfference between the most and least open countres (Sngapore 373% and Brazl 24%) s larger than that of ntra-japan data. For addtonal nsghts, we also consder the OECD, EU, and Eurozone countr samples. For comparson purposes, Table 2 summarzes the features of the dfferent transportaton and communcaton; educaton; entertanment; and mscellaneous. 14

samples. The OECD sample precludes a large number of less developed countres ncluded n the full sample. These countres tend not to be on the same level as the developed countres n terms of trade restrctons. The also tend to have much smaller GDP. The EU sample conssts of countres clustered n Europe that have shared access to the common market. The Eurozone sample contans onl those that addtonall share a common currenc. These nternatonal subsamples are consdered mddle cases between the full world sample and the ntra-japan sample n whch prefectural economes are full ntegrated through a common market, currenc, and fscal sstem. In addton, Japanese prefectures are geographcall more condensed than the EU and Eurozone countres. The trade openness of the OECD, EU, and Eurozone countr subsamples are also reported n panel B of Table 1. Interestngl, lmtng the sample to OECD countres further lowers the average openness, albet modestl, to 83%. Gven that the OECD sample retans large economes, such as the Unted States and Japan, whle excludng numerous small less developed economes, the lower average openness ma n part reflect the sze effects. Nonetheless, the overall sample statstcs appear surprsngl smlar between the world and OECD samples. In other words, whether or not to nclude a large number of less developed economes n the sample does not seem to matter much, whch renforces our motvaton to queston the effects trul captured b the trade-to-gdp ratos n the name of trade openness. The openness of EU member countres s on average hgher, at approxmatel 100%, presumabl reflectng the effects of European market ntegraton and geographcal proxmt. Further restrctng the sample to the Eurozone countres elevates the average openness to 107%. In accordance wth the antcpaton, the extent of the EU and 15

Eurozone samples trade openness exceeds that of the world sample whle fallng short of that of the Japanese sample. 16 A cross-sample comparson of the maxma reveals that the prefecture most open to ntra-japan trade (Me, 215%) s substantall less open than countres such as Sngapore (373%) and Luxembourg (274%) are to nternatonal trade. As a whole, Japan s the least open OECD countr, wth an average trade openness of onl 24%, and s the second least open of the 171 countres, next to onl Brazl. 17 To be an nformatve measure, trade openness should have suffcent cross-sectonal varatons. As dsplaed n the fourth column of Table 1, the standard devaton of ntra-japan trade openness s approxmatel 33%, whereas the nternatonal counterparts fall between 46% and 55%. The ffth column contans the coeffcent of varaton (.e., the rato of the standard devaton to the mean). Usng ths measure, the ntranatonal and nternatonal comparson s 25% versus 47% 57%. Relatve to the szes of the standard devatons and the coeffcents of varaton, respectvel, trade openness wthn Japan s 66% and 45% as varable as t s across 171 countres around the world. That s, the trade openness of full ntegrated, equall open, and geographcall condensed economes s approxmatel half as varable as t s for dverse countres scattered around the world and s segmented b artfcal and natural trade mpedments. Agan, these observatons renforce our motvaton to ask about the effects that we actuall capture usng trade-to-gdp ratos. 16 Note that the trade openness seres extracted from the WDI database s based on the total trade vs-à-vs the rest of the world. Thus, the statstcs for the EU and Eurozone groups do not ndcate trade openness wthn each group. Nonetheless, to the extent that common markets and common currenc enhance trade between member countres, the EU and Eurozone groups are expected to have a hgher degree of trade openness than the world or the OECD samples. 17 The Unted States s the thrd least open countr, wth average openness of 24.5%. 16

5. Evaluatng the canddates 5.1 Parwse co-movement In secton 3, we provde four possble economc nterpretatons of the effects reflected b the trade openness varable. Namel, the are the extent of economc remoteness, market ntegraton n tradables, dosncratc sectoral specalzaton, and nclnaton to explot economes of scale. We also ntroduced alternatve measures of these effects. In ths subsecton, we use Japanese data to test whether the measures of the four effects exhbt sgnfcant co-movements wth the trade openness varable. Table 3 presents a correlaton matrx based on averages of the 1997, 2002, and 2007 observatons. 18 Among the four alternatves, the ndces of economc remoteness and dosncratc sectoral specalzaton exhbt sgnfcant correlatons of 0.67 and 0.73, respectvel, wth trade openness. The opposng sgns are as antcpated. A more trade-open prefecture tends to be less remote (.e., closer to larger markets) and more unquel specalzed, mplng that nterndustr trade rather than ntrandustr trade prevals among the prefectures. The negatve correlaton of -0.36 between the remoteness and specalzaton ndces, s also statstcall sgnfcant. Further, the remoteness ndex shares a sgnfcant postve correlaton wth the devatons of clothng and footwear prces. Thus, economcall remote prefectures also tend to have large prce devatons. Nonetheless, trade openness does not exhbt a sgnfcant correlaton wth the prce devatons of ether of the two tradable product categores. The correlaton between trade openness and populaton sze s nsgnfcant and sgned, n contrast to our expectatons. 18 Regonal prce data are avalable onl for these ears, as noted n sub-secton 4.1. We also generate ear-b-ear results that conve a smlar message. To conserve space, the are not reported but are avalable on request. 17

For robustness, we also examne the rank correlatons b orderng the prefectures accordng to trade openness and the other four factors. The mert of the nonparametrc rank correlaton test s that t does not requre a lnear relatonshp between the varables. The exercse helps us dentf the factors that well mmc trade openness when rankng the prefectures. Table 4 presents the Spearman rank correlatons on the bass of the three-ear average data. In accordance wth the correlaton coeffcent results, the trade openness varable exhbts statstcall sgnfcant rank correlatons wth the measures of economc remoteness and sectoral specalzaton. A prefecture ranked hgh b trade openness also tends to be ranked hgh b the extent of dosncratc specalzaton n producton whle ranked low n terms of remoteness. The results are consstent wth the vew based on the gravt equaton. The also suggest that a tandem relatonshp of trade and specalzaton holds well ntranatonall, mplng domnance of nterndustr trade wthn Japan. From the rank correlatons among the four factors, we addtonall observe that economc remoteness and market sze are negatvel assocated wth each other. Not surprsngl, economcall remote prefectures tend to be small n terms of populaton. Altogether, the results n ths subsecton ndcate potental ssues regardng usng (1) as a measure of openness. Whle assocated wth the ndces of economc remoteness and sectoral specalzaton, the trade openness varable does not exhbt a sgnfcant drect assocaton wth the extent of tradables prce devatons or market sze, unlke the ndex of economc remoteness. 5.2 Explanator power for the cross-sectonal varaton 18

In ths subsecton, we conduct a regresson analss to account for the cross-sectonal varaton of trade openness. We emphasze that the purpose of the exercse s lmted to evaluatng the four canddates n terms of ther explanator power for cross-sectonal varaton n the measured trade openness. The regressons presented are not ntended for an causal nference and should be vewed accordngl. We lmt the focus of our dscusson to the model sgnfcance and explanator power. Poolng the data for 1997, 2002, and 2007, we estmate ln OPEN, t φ t tdt + X, t ' Φ + ε, t =, (10) where D t denotes the ear-specfc dumm varables and X s a vector consstng of the four alternatves n logarthms, namel, the measures of economc remoteness, market ntegraton, sectoral specalzaton, and scale effects. To make the estmates comparable across all specfcatons, we exclude Oknawa from the sample gven mssng observatons on the specalzaton ndex. Table 5 presents the estmates. As a benchmark, we ntall regress the trade openness varable onl on the ear-specfc constants, whch elds an adusted R 2 estmate of.021 and an nsgnfcant F-statstc as dsplaed n column 1. Retanng the ear-specfc ntercepts, we then estmate b-varate specfcatons to compare the margnal explanator power of the four alternatves, each as a sole regressor. The results are dsplaed n columns 2 through 5. 19 The adusted R 2 estmates ndcate that the remoteness and specalzaton varables (specfcatons 2 and 4) possess substantal margnal explanator power for trade openness. The F-statstcs also endorse onl these two varables n terms of model sgnfcance. 19 In the results reported n Table 4, prce devatons are measured usng the prces of clothes and footwear. Replacng them wth the furnture prce devatons does not sgnfcantl affect the results. The results are avalable on request. 19

When combned, the remoteness and sectoral specalzaton ndces (along wth the ear-specfc ntercepts) account for nearl 70 % of the varaton n the measured trade openness as shown n column 6. As observed n columns 7 through 9, there s lttle margnal gan n the model s explanator power b addtonall usng the nformaton on tradables prce devatons and/or populaton sze. Overall, the results thus far suggest that, n the absence of artfcal trade barrers, cross-sectonal varaton n trade openness conves nformaton prmarl on economc remoteness and the dstnctveness of sectoral producton dstrbutons. The latter s consstent wth the nsght of Leamer (1988) that (1) measures the extent of the peculart of the resource suppl n the absence of trade barrers. In the meantme, we also fnd that approxmatel 30% of the total varaton n trade openness remans unaccounted for b the four effects we consder. In the next secton, we decompose the varablt of trade openness to understand ts component sources and ther relatve mportance. 6. Varance decomposton analss The varablt of trade openness depends on the varablt of trade, the varablt of GDP, and the co-varablt between trade and GDP. More specfcall, the followng decomposton apples to the varance of (1) n logarthm Var(lnOPEN) = Var(lnTRADE lngdp). (11) = Var(lnTRADE) + Var(lnGDP) 2Cov(lnTRADE,lnGDP) Var (ln GDP) can be taken as a measure of the heterogenet of the sze of the econom, whch depends on the nature of the sample under consderaton. For nstance, consderng advanced and nonadvanced economes together tends to generate a large 20

Var (ln GDP) value. In contrast, a sample consstng of smaller-scale regonal economes wthn a countr wll naturall make Var (ln GDP) small. The remanng two terms on the rght-hand sde of (11) contan trade. For them, t s useful to frst consder the determnants of trade sze (n absolute terms) of a gven econom. Abstractng from detaled condtons, trade sze depends prmarl on the sze of the econom. All else held constant, a large econom trades more n absolute terms than a small econom. Thus, n prncple, Var (lntrade) depends on Var (ln GDP), and Cov (lntrade,lngdp) should be postve. 20 Once we allow other condtons to var, dfferences n trade polc, factor endowments, geographcal locatons, and so on ntervene to alter trade sze for a gven GDP sze and, hence, alter Cov (lntrade,lngdp). The factors that can make the co-varance smaller than otherwse nclude restrctve trade polc, severe natural trade mpedments, homogenet n factor endowments, and sgnfcant scale econom effects that motvate smaller economes more than larger ones to expand ther trade. 21 A useful wa of analzng the decomposton s to measure the three components n relatve terms to the total varance. The relatve shares ndcate the contrbutons b component whle takng the heterogenet of GDP sze for the sample as gven. In other words, the nform us of the relatve mportance of GDP varablt, trade varablt, and ther co-varablt and whether such mportance dffers notceabl across dfferent samples. 20 In the long term, GDP can also depend on trade f trade causes growth. In ths case, Var (ln GDP) ma also depend on Var (lntrade). For an emprcal assessment of the trade-growth nexus, see Dollar and Kraa (2003), Edwards (1993, 1998), Frankel and Romer (1999), and Rodrguez and Rodrk (2000). 21 The effects of GDP sze, trade restrctons, and economes of scale are ntertwned. As artculated b Alesna, Spolaore, and Waczarg (2000), a smaller countr benefts more from less restrctve trade when economes of scale effects are present. 21

Table 6 summarzes the varance decomposton results. The relatve contrbutons of GDP varance, trade varance, and GDP-trade co-varance to total varance are denoted n percentage terms n columns (), (), and (), respectvel. The lack of varaton n GDP sze n the ntra-japan sample leads to the relatvel small contrbuton of Var (ln GDP) to the total varaton. The share of 922% s less than half the correspondng share of 2,067% for the world sample. Nonetheless, the OECD sample appears much closer to the ntra-japan sample than the world sample n terms of the contrbuton of Var (ln GDP), whch s 1,045%. The EU and Eurozone samples (1,681% and 1,616%, respectvel) le between the world and the OECD samples. A comparson of the world and OECD samples ndcates that excludng a large number of less developed economes substantall reduces the share of the GDP sze varablt, as antcpated. Interestngl, the shares of trade varance and co-varance declne even more overtl. The EU and Eurozone samples turn out to have greater varaton n GDP and trade than the OECD sample. Regardng the contrbuton of Var (lntrade), the ntra-japan sample appears largel comparable to the nternatonal subsamples. The share of 1,090% well exceeds the OECD s 751% but falls short of the EU and Eurozone fgures of 1,333% and 1,247%, respectvel. When measured n relatve terms to GDP varablt, trade varablt s greater wth the ntra-japan sample than wth the nternatonal counterparts. The rato dsplaed n column (v) of Var (lntrade) to Var (ln GDP) exceeds unt onl for the ntra-japan sample. For all nternatonal samples, Var (lntrade) s less than Var (ln GDP). The same apples to the rato dsplaed n column (v) of 22

Cov (lntrade,lngdp) to Var (ln GDP). Gven wdel used nternatonal data, varaton n (1) results more from the varablt of GDP than the varablt of trade. In panel C of the table, we report the decomposton results b consderng onl wthn-group trade for the Eurozone, EU, and OECD samples, such that the are more strctl comparable wth those of the ntra-japan data. 22 Ths recalculaton lowers the relatve shares of all components across the samples and hghlghts the dstnctveness of the OECD sample n terms of the small relatve shares of each component. The recalculaton also shfts the EU and Eurozone samples awa from the world sample and toward the Japanese sample n terms of smlart n the relatve component shares. In summar, we develop the man message of the results as follows. GDP sze heterogenet matters much to the varaton of trade openness. In fact, gven standard nternatonal data, cross-countr varatons n (1) derve more from GDP than from trade. Consequentl, (1) as a measure of trade openness ma not exhbt suffcent varablt even f trade s relatvel varable when the sample conssts of economes wth smlar GDP szes, whch has two mplcatons. Frst, the ntra-japan data we adopt can be deemed an exemplar sample wth whch trade openness varablt s restraned because of the lack of GDP sze varablt. In addton, the prefectural economes are full ntegrated and completel open to trade wth each other. However, even gven these extreme condtons, we stll observe trade openness b (1) to be approxmatel half as varable as t s wth the world sample. Ths varablt prmarl reflects the dfferences n economc remoteness and sectoral producton dstrbutons rather than openness n terms of outward polc orentaton. 22 To calculate, trade openness wthn the OECD, EU, and Eurozone countres (rather than vs-à-vs the ROW), we use the trade matrx for each group as extracted from the IMF s DOT and GDP n current U.S. dollar terms. 23

Second, the evdence on ntra-japan data ma understate the mportance of the nclnaton to explot economes of scale effects. The nsgnfcance of the own market sze effects n secton 5 ma be drven b the nsuffcenc of the GDP sze varablt specfc to the ntra-japan sample. 7. Conclusons Faced wth severe data constrants, a researcher ma choose to sacrfce some conceptual accurac over data avalablt when decdng on the emprcal varables to adopt. Ths apples partcularl to studes coverng a wde cross-secton of countres for whch desrable data are often dffcult to obtan. Although the rato of total trade to GDP s admttedl a crude measure of trade openness, data avalablt makes t a popular choce for emprcal analses. Despte the warnng b Leamer (1988), we contnue to see the varable used routnel n the lterature on a wde range of topcs. Although the varable ma well be ustfed as a prox measure, the queston s for what exactl the varable can be a good prox. Ths paper contrbutes to the lterature b emprcall explorng the nformatonal content of the heavl used albet often onl vaguel nterpreted measure of trade openness. For accurate nterpretaton of the emprcal results, understandng the effect that s actuall controlled for b takng nto account the dfferences n the trade-to-gdp ratos s ndspensable. Usng ntranatonal and nternatonal data, we examned the nature of the trade openness varable. In partcular, ntra-japan data enable us to stud the trade openness varable n an envronment n whch no artfcal trade restrctons nterfere. We observed substantal varablt n the extent of the measured trade openness across 24

prefectures wthn Japan despte the fact that the are unforml under free trade vs-à-vs each other and, hence, full open. Further, we fnd that the varablt s well assocated wth economc remoteness and dosncratc sectoral dstrbuton n producton. The varance decomposton results suggest for the nternatonal data that the varablt of trade openness derves more from the varablt of GDP sze than trade. B constructon, (1) s as much a measure about GDP as t s about trade. These fndngs together llustrate what one ma end up capturng b adoptng the trade-to-gdp rato n the name of trade openness. We acknowledge that the ntranatonal analses do not nform us drectl of the mportance of dfferences n trade polc because ther effect s preempted b the sample constructon. Nonetheless, the ntra-japan sample can be vewed as an exemplar case near the lower bound n terms of trade openness varablt gven the lack of polc dfferences and the low varablt of GDP sze. From ths perspectve, our fndngs mpl that a substantal porton of nternatonal trade openness s lkel to have lttle to do wth openness per se. The conveed nformaton can be more appropratel characterzed as the extent of economc remoteness and dosncratc dstrbuton of sectoral producton. In a nutshell, the name of the varable s deceptve. 25

Data Appendx Sources of the ntra-japan data Annual Report on Prefectural Accounts 2012 CD-ROM, Department of Natonal Accounts, Economc and Socal Research Insttute, Cabnet Offce, Government of Japan. Regonal Statstcs Database, Offcal Statstcs of Japan. Sources of nternatonal data World Development Indcators, the World Bank. Drecton of Trade Statstcs, the Internatonal Monetar Fund. Internatonal dstances are calculated as dstances between natonal captals b usng the calculator avalable at http://www.nfoplease.com/atlas/calculate-dstance.html. The computaton s for the great crcle dstance between ponts, and do not account for dfferences n elevaton. Notes Whle the Annual Report contans both nomnal and real data, real seres have an advantage of beng free from possble cross-prefectural dfference n relatve prce of government purchases. We thus use the real data. Data on the dsposable ncome (DI) are avalable onl n nomnal fgures. We convert them nto real seres usng the prefectural GDP deflators. Also, there are a couple of prefecture-specfc ncdents of mssng observatons. Frst, the observatons on Toko s DI are mssng altogether. We obtan the 2001-2009 nomnal fgures from Toko Metropoltan Government s ste (http://www.touke.metro.toko.p). These fgures are then converted nto real terms usng Toko s GDP deflator. Also, Ach prefecture s export and mport are avalable onl n nomnal fgures. The are also converted nto real seres usng Ach s prefectural GDP deflator. Sectors for the specalzaton ndex The specalzaton ndex (9) n the man text s constructed b usng the output data on the followng sectors: A. Industres A1. Agrculture, forestr and fshng

1) Agrculture, 2) Forestr, 3) Fshng A2. Mnng A3. Manufacturng 1) Food products and beverages, 2) Textles, 3) Pulp, paper and paper products, 4) Chemcals, 5) Petroleum and coal products, 6) Non-metallc mneral products, 7) Prmar metal, 8) Fabrcated metal products, 9) Machner, 10) Electrcal machner, equpment and supples, 11) Transport equpment, 12) Precson nstruments, 13) Others A4. Constructon A5. Electrct, gas and water suppl 1) Electrct suppl, 2) Gas and water suppl A6. Wholesale and retal trade A7. Fnance and nsurance A8. Real estate A9. Transport and communcatons A10. Servce actvtes B. Producers of government servces B1. Electrct, gas and water suppl B2. Servce actvtes B3. Publc admnstraton C. Producers of prvate non-proft servces to households C1. Servce actvtes

Table A-1. Lst of the prefectures n Japan 1 Hokkado 25 Shga 2 Aomor 26 Koto 3 Iwate 27 Osaka 4 Mag 28 Hogo 5 Akta 29 Nara 6 Yamagata 30 Wakaama 7 Fukushma 31 Tottor 8 Ibarak 32 Shmane 9 Tochg 33 Okaama 10 Gunma 34 Hroshma 11 Satama 35 Yamaguch 12 Chba 36 Tokushma 13 Toko 37 Kagawa 14 Kanagawa 38 Ehme 15 Ngata 39 Koch 16 Toama 40 Fukuoka 17 Ishkawa 41 Saga 18 Fuku 42 Nagasak 19 Yamanash 43 Kumamoto 20 Nagano 44 Ota 21 Gfu 45 Mazak 22 Shzuoka 46 Kagoshma 23 Ach 47 Oknawa 24 Me

Table A-2. Lst of countres b group OECD EU Euro-zone 1 Australa Austra Austra 2 Austra Belgum Belgum 3 Belgum Bulgara Cprus 4 Canada Croata Estona 5 Chle Cprus Fnland 6 Czech Republc Czech Republc France 7 Denmark Denmark German 8 Estona Estona Greece 9 Fnland Fnland Ireland 10 France France Ital 11 German German Latva 12 Greece Greece Lthuana 13 Hungar Hungar Luxembourg 14 Iceland Ireland Malta 15 Ireland Ital Netherlands 16 Israel f Latva Portugal 17 Ital Lthuana Slovak Republc 18 Japan Luxembourg Slovena 19 South Korea Malta Span 20 Luxembourg Netherlands 21 Mexco Poland 22 Netherlands Portugal 23 New Zealand Romana 24 Norwa Slovak Republc 25 Poland Slovena 26 Portugal Span 27 Slovak Republc Sweden 28 Slovena Unted Kngdom 29 Span 30 Sweden 31 Swtzerland 32 Turke 33 Unted Kngdom 34 Unted States Notes: The table lsts the sample countres used n the emprcal analses n the man text.

Table A-3. Lst of countres n the world sample Albana, Algera, Angola, Antgua and Barbuda, Argentna, Armena, Aruba, Australa, Austra, Azerbaan, The Bahamas, Bahran, Bangladesh, Barbados, Belarus, Belgum, Belze, Benn, Bhutan, Bolva, Bosna and Herzegovna, Botswana, Brazl, Brune, Darussalam, Bulgara, Burkna Faso, Burund, Cabo Verde, Camboda, Cameroon, Canada, Central Afrcan Republc, Chad, Chle, Chna, Colomba, Comoros, Dem. Rep. Congo, Rep. Congo, Costa Rca, Cote d'ivore, Croata, Cuba, Cprus, Czech Republc, Denmark, Domnca, Domncan Republc, Ecuador, Arab Rep. Egpt, El Salvador, Equatoral Gunea, Ertrea, Estona, Ethopa, F, Fnland, France, Gabon, The Gamba, Georga, German, Ghana, Greece, Grenada, Guatemala, Gunea, Gunea-Bssau, Hat Honduras, Hong Kong SAR, Hungar, Iceland, Inda, Indonesa, Ireland, Israel, Ital, Japan, Jordan, Kazakhstan, Kena, Krbat, Rep. Korea, Kuwat, Krgz Republc, Lao PDR, Latva, Lebanon, Lesotho, Lba, Luxembourg, Macao SAR, Macedona, FYR Madagascar, Malaw, Malasa, Maldves, Mal, Malta, Maurtana, Maurtus, Mexco, Moldova, Mongola, Morocco, Mozambque, Namba, Nepal, Netherlands, New Zealand, Ncaragua, Nger, Ngera, Norwa, Oman, Pakstan, Palau, Panama, Paragua, Peru, Phlppnes, Poland, Portugal, Puerto Rco, Qatar, Romana, Russan Federaton, Rwanda, Samoa, Saud Araba, Senegal, Serba, Sechelles, Serra Leone, Sngapore, Slovak Republc, Slovena, Solomon Islands, South Afrca, Span, Sr Lanka, St. Ktts and Nevs, St. Luca, St. Vncent and the Grenadnes, Sudan, Swazland, Sweden, Swtzerland, Takstan, Tanzana, Thaland, Togo, Tonga, Trndad and Tobago, Tunsa, Turke, Turkmenstan, Uganda, Ukrane, Unted Kngdom, Unted States, Urugua, Uzbekstan, Vanuatu, Venezuela, Vetnam,, West Bank and Gaza, Rep.Yemen, Zamba Zmbabwe.