market potential and the location of...

MARKET POTENTIAL AND THE LOCATION OF JAPANESE INVESTMENTIN THE EUROPEAN UNION

Keith Head and Thierry Mayer

AbstractmdashThis paper develops a theoretical model of location choiceunder imperfect competition to formalize the notion that firms prefer tolocate ldquowhere the markets arerdquo The profitability of a location depends ona term that weights demand in all locations by accessibility Using asample of Japanese firmsrsquo choices of regions within European countrieswe compare the theoretically derived measure of market potential with thestandard form used by geographers Our results show that market potentialmatters for location choice but cannot account entirely for the tendency offirms in the same industry to agglomerate

I Introduction

We want to build our plants where the markets are

IN December 1997 Hiroshi Okuda chairman of Toyotaused the statement above to justify Toyotarsquos decision to

build a factory in northern France At that time analysts inthe press largely attributed that decision to the low marketshare of the Japanese car manufacturer in France (1) andin Europe in general (3)1 The Toyota example suggeststhat even in the zero-tariff internal market of Europe firmsstill seek locations with superior market access The man-agers of Toyota apparently thought that their existing pro-duction sites in Great Britain were not close enough to theFrench market The Wall Street Journal reported thatldquoToyota hopes to capture 3 [of the French market]after opening its factory here in 2001rdquo

This paper connects two previously disparate strands ofthe economic geography literature The first strand demon-strates a statistical tendency of firms to make the samelocation decision as previous firms with similar attributes(such as industry and national origin)2 Though such ag-glomeration effects appear regularly in empirical work theyare consistent with a variety of explanations The secondstrand comprises a large number of theoretical papers thatfocus on a particular mechanism of agglomeration namely

that producers concentrate where demand is highest andserve smaller markets via exporting3

We link the two strands by showing how to derive thefirmrsquos location choice probabilities as a function of produc-tion costs and a demand variable closely linked to themeasure of ldquomarket potentialrdquo introduced by Harris (1954)We then take the model to the data investigating whetherlocation choices of Japanese affiliates in Europe are drivenby market-access motivations a la Krugman or some otherform of agglomeration effect We find that the demand-pullmechanism has some explanatory power but it does notappear to explain away the entire empirical agglomerationeffect

The influence of market potential on the location ofproducers and the wages they pay has been the subject ofseveral recent papers Davis and Weinstein (1999 2003)find that production increases on a more than one-for-onebasis with demand (the ldquohome market effectrdquo) in manyindustries As in their 1999 paper we focus on how demandaffects the intranational location of production As in their2003 paper we construct a market potential measure thataggregates demand from multiple locations while discount-ing for distance using a parameter obtained from a first-stepestimation using bilateral trade flows In addition to distanceeffects our measure incorporates the effect of borders ontrade as well as a theoretically derived adjustment forcompetition

Hanson (2001) estimates the relationship between county-level wages in the United States and market potentialStructural estimation of this equation reveals that wages ina county are increasing in demand emanating from allAmerican counties with weights declining exponentially inbilateral distance Redding and Venables (2004) follow asimilar line of reasoning for international data using abilateral trade equation of a theoretical model to obtainestimates of bilateral trade costs and of each countryrsquosmarket and supply accessibility They find that internationalinequality is closely linked to differences in market accessCrozet (2004) uses the same theoretical framework andshows that migration flows also respond to market potential

The literature on firm location choice has not previouslyestimated models directly derived from the Krugmanmodel Prior work has of course considered demand buttypically only local demand4 Knowing the size of demandin each of the districts a firm might choose is not sufficientfor firms can export to nearby locations Some studies such

Received for publication June 6 2002 Revision accepted for publicationJanuary 6 2004

University of British Columbia and Universite de Paris I (TEAM)CEPII CERAS-ENPC and CEPR respectively

This paper was produced as part of a CEPR Research Network on TheEconomic Geography of Europe Measurement Testing and Policy Sim-ulations funded by the European Commission under the Research Train-ing Network Programme (contract no HPRN-CT-2000-00069) We wouldlike to thank participants at the CEPR Workshop Economic Geography ofEurope Measurement Testing and Policy Simulations and at the Empir-ical Investigations in International Trade 2000 Conference for theirsuggestions We also thank participants at seminars at the CatholicUniversity of Leuven (KUL) University of Paris XII and XIII andUniversity of Geneva Richard Baldwin Bruce Blonigen Pierre-PhilippeCombes Matthieu Crozet Gianmarco Ottaviano Jean Marc Siroen andtwo referees provided helpful comments on earlier versions for which weare particularly grateful

1 Douglas Lavin Interactive Edition (December 10 1997) APndashDowJones News Service (December 9 1997)

2 Devereux and Griffith (1998) and Head Ries and Swenson (1999) arerecent examples from this literature

3 Krugman (1980 1991) wrote the seminal papers in this literature themonograph of Fujita Krugman and Venables (1999) thoroughly analyzesthe basic model and its extensions

4 See for instance Coughlin Terza and Arromdee (1991)

The Review of Economics and Statistics November 2004 86(4) 959ndash972copy 2004 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology

as Friedman Gerlowski and Silberman (1992) HendersonKuncoro and Turner (1995) and Head et al (1999) con-sider nonlocal demand but not using measures derived fromtheory5 In particular theory suggests that nonlocal demandmust be discounted for bilateral trade impediments Further-more a given amount of market access contributes less toprofits when a firmrsquos competitors have access to the samemarkets We follow Krugman (1992) in adjusting the marketpotential measure to take into account the location ofcompetitors

The paper proceeds as follows In section II we derive alinear-in-logs equation that relates the profitability of alocation to a prospective foreign investor to a measure ofthat locationrsquos access to demand We then show how toestimate the distance and border effects that impede marketaccess using bilateral trade data In section III we report theresults from the trade equation and show how we use themto calculate market potential We then discuss our sample ofJapanese investors and the set of possible location choicesOur location-choice results are detailed in section IV andwe conclude and propose directions for future work insection V

II The Model

Let Er denote expenditure in a representative industry(we omit industry subscripts for notational simplicity) inregion r Consumers (who may be firms or individuals)allocate their expenditures across differentiated varieties inthe representative industry They have constant-elasticity-of-substitution subutility functions for each industry Max-imizing this subutility function subject to expenditure Erand the delivered prices from all R possible product originswe obtain the demand curve for the representative variety inthe representative industry as

qij pij

yenr1R nrprj

1 Ej (1)

where pij is the delivered price faced by consumers inregion j (destination) for products from region i (origin) Itis the product of the mill price pj and the trade cost factorij Trade costs include all transaction costs associated withmoving goods across space and national borders

A The Profit Equation for Foreign Affiliates

Each firm sets its mill prices to maximize profits Fol-lowing Dixit and Stiglitz (1977) firms treat the elasticity ofsubstitution as if it were the price elasticity of demand(this may also be interpreted as the assumption that eachfirm has infinitesimal market share) The resulting mill

prices are simple markups over marginal costs denoted crthus pr cr[( 1)] Substituting into equation (1) weobtain the quantity that a firm producing in region i woulddeliver6 to each destination region j

qij 1

ciij

GjEj (2)

where Gj yenr nr(crrj)1 The gross profit earned in eachdestination region j for a firm producing in region i is

ij pi ciijqij ciij

1

GjEj (3)

This gross profit is an increasing function of the expenditureof country j on the considered industry The fraction mul-tiplying Ei depends on the costs of the representative firmrelative to its competitors from all R regions In the numer-ator we see that profits are decreasing in local (region i)production costs Lower trade costs to reach region j (that isa low ij) also raise profits Because the effect of trade costsis always moderated by the elasticity of substitution weintroduce the notation of ij ij

1 to measure the accessof exporters from i to market j7 The denominator containsthe corresponding characteristics of competing suppliersNote that the denominator contains a factor capturing theidea that competition is fiercer and profits are thereforelower when varieties are less differentiated from each other

Summing the gross profits earned in each market andsubtracting the fixed costs Fr necessary to establish a plantin region r we obtain the aggregate net profit r to beearned in each potential location r

r cr

1

j1

R

rj

Ej

Gj Fr

cr1

Mr Fr (4)

where

Mr j

rjEj

Gj

We will refer to Mr as the Krugman market potentialbecause an expression for it first appeared in Krugman(1992) The profit equation suggests that firms face atradeoff between low production costs and high marketpotential8

5 Friedman et al (1992) use a distance-weighted sum of per capitaincomes Henderson et al (1995) use distance to nearest major businessdistrict and Head et al (1999) sum the personal incomes of adjacentstates

6 The iceberg convention implies that to deliver q units the firm mustship q units

7 Baldwin et al (2003) refer to as ldquofreeness of traderdquo and mainlyconsider cases of symmetric trade costs (ij ji) The term marketaccess is better suited to our case where we find asymmetries in tradecosts

8 The wage equation analyzed by Fujita et al (1999 p 53) can bederived from the profit equation by assuming that free entry sets equation(4) equal to 0 In their specification production requires 1-unit of labor perunit of output and F units of labor as overhead Denoting the cost of a

THE REVIEW OF ECONOMICS AND STATISTICS960

When a firm chooses its location the only relevantinformation is the ordering of the profits Invariant fixedcosts do not affect the profit ordering of regions and cantherefore be omitted For tractability therefore we assumethat fixed costs do not differ across locations (Fr F r)As monotonic transformations also leave the ordering un-changed we will make four of them to create a simple andintuitive expression for profitability Namely we add Fmultiply the result by raise the result to the power 1( 1) and take natural logs Denoting the result as Ur weobtain

Ur ln lnr F

1 ln cr 11 ln Mr

(5)

Equation (5) expresses the profitability for a firm of locatingin region r as a very simple function that is decreasing inproduction costs and increasing in the Krugman marketpotential term

Considering the cost term first let us suppose that thevariable cost function is Cobb-Douglas with constant re-turns using labor at cost wr and other inputs (such as landand intermediates) at cost vr Laborrsquos share is and Ar

represents total factor productivityThus log marginal costs are given by

ln cr ln wr 1 ln vr ln Ar (6)

Substituting (6) into (5) and rearranging we have

Ur ln wr 11 ln Mr

1 ln vr ln Ar(7)

We observe wages wr and will calculate Mr using a methoddescribed in section II B We do not observe vr and Ar andthey will be captured with several proxies (specified insection III C) and a random term observed by firms but notby the econometrician (detailed in section III B)

The Krugman market potential aggregates the expendi-tures of all regions while adjusting for region rrsquos access rj

and for competition from firms located in other regions GjAnalyzing the function we can determine the assumptionsthat were implicit in the original Harris (1954) formulationof market potential Specifically if we set Gj 1 and rj 1drj then Mr reduces to yenjEjdrj that is the inverse-distance-weighted sum of incomes In the Krugman marketpotential function Mr the denominator Gj takes into the

account the competition that firms from region r face fromrival firms exporting from other regions to serve the demandin each export market j This competition adjustment isincreasing in the number of rivals and decreasing in theirtrade and production costs

The competition adjustment can help explain why other-wise identical firms would not all select the same locationAs more firms choose one region the market there becomesmore crowded lowering Mr until another region is moreprofitable This is not the only mechanism causing disper-sion however Firms are not identical and will thereforediffer in their views of the prospective profitability of eachregion This heterogeneity is analogous to matching in labormarkets In the context of our model this heterogeneity canbe thought of as firm-specific variation in regional produc-tivity (ln Ar)

The Krugman market potential has the advantage ofbeing derived rigorously from theory However unlike theHarris form its calculation requires estimates of the un-known parameters ij and Gi Our strategy will be to useinformation from international trade flows to estimate theseparameters

B The Trade Equation

We do not observe trade flows between regions and mustinstead rely upon trade between nations to estimate theparameters that determine trade costs Results from Wolfrsquos(2000) study of intranational trade in the United States arereassuring in this respect for he finds distance effectsresembling those found using international trade data Wereinterpret equation (2) as giving the quantity exported by arepresentative firm in country I to country J (we reserve thelowercase i and j to denote origin and destination regions)The aggregate value of country Irsquos exports to country Jdenoted XIJ is given by the quantity exported by a repre-sentative variety firm from I multiplied by the price and thenumber of varieties from J

XIJ pIJqIJnI nI

cI1 IJEJ

GJ

Taking natural logs and grouping variables according tosubscripts we obtain

ln XIJ lnnIcI1 lnEJGJ ln IJ (8)

Following Redding and Venables (2004) we will estimatethe first two terms using exporter and importer fixed effectsdenoted EXI and IMJ9 Bilateral market access (IJ) will beestimated as a function of distance (dIJ) borders (BIJ 1labor unit as wr this implies marginal costs of cr wr and fixed costs of

Fr Fwr Solving for wr we have

wr 1

FMr 1

That is for firms to be indifferent between locations wages must be apower function of market potential Hanson (2001) and Redding andVenables (2004) estimate variations of this relationship

9 An earlier version of this paper [Head and Mayer (2002) available atceprorg] used the number of Japanese firms to measure n and calculatedGj at the regional level It constructed the ci

1 term using wage data andestimates of obtained from a first-step trade equation Our currentapproach suggested by a referee imposes fewer and more plausibleassumptions but arrives at the same empirical conclusions

MARKET POTENTIAL AND JAPANESE INVESTMENT 961

for I J) sharing a common language (LIJ 1 if I andJ share a language and I J and 0 otherwise) and an errorterm 13IJ Parameters capturing the effect of distance bor-ders and language on trade volumes are denoted J and respectively IJ dIJ

exp[(J LIJ) BIJ 13IJ]We interpret border effects as comprising home bias inconsumer preferences and government procurement differ-ential technical standards exchange rate uncertainty andimperfect information about potential trade partners10 Thespecification allows border effects to differ across importingcountries which is largely supported by the empirical evi-dence (see Chen 2004) The estimated equation will there-fore be

ln XIJ EXI IMJ ln dIJ JBIJ LIJBIJ 13IJ

(9)

The estimated parameters on trade costs and importersrsquofixed effects are then used to construct the market potentialvariable that will be included in the location choice analysisof Japanese firms in Europe Recall that the market potentialof region i is Mi yenj ij(EjGj) where ij is theaccessibility of market j to goods shipped from region iThe formulas for calculating inter- and intraregional accessare

ij expJ LIJ dij

when i I j J and I J

ij dij

when i and j belong to the same country and

ii dii 2

3areai

for intraregional tradeThe last equation models the average distance between a

producer and a consumer based on a stylized geographywhere all producers are centrally located and the consumersuniformly distributed across a disk-shaped region

The second component of market potential calculation isregional-level competition-weighted expenditure Takingthe exponential of importerrsquos fixed effectsrsquo coefficients weobtain EJGJ exp(IMJ) that is the competition-weightedexpenditure of country J We calculate EjGj for each regionj of country J by allocating EJGJ to the different regions inproportion to their share of national GDP ( yjyJ) that isEjGj ( yjyJ) exp(IMJ)

We also compare the Krugman market potential with thesimpler version proposed by Harris (1954) This variablesums distance-weighted industry-level expenditures yenj Ejdij Again we allocate national expenditure to regions

according to GDP shares of regions [Ej ( yjyJ) EJ]National expenditure is calculated using apparent consump-tion that is EJ production exports imports in theconsidered industry It comprises final and intermediatedemand from all sectors

III Econometric Model and Data

We estimate a model of location choice of 452 Japanese-owned affiliates that were established in 57 regions belong-ing to nine European countries (Belgium France GermanyIreland Italy the Netherlands Spain Portugal and theUnited Kingdom) during the period 1984ndash1995 As can beseen in equation (5) we hypothesize that market potential isa key determinant of this decision We construct the marketpotential for 18 industries using the parameters estimatedfrom the trade equation (9)

A Estimation of the Trade Equation

We estimate equation (9) using Eurostat data on bilateraltrade matched with production at the NACE70 two-digitlevel Production data are needed because internal tradeflows XII are constructed by subtracting total exports (yenJI

XIJ) from national production Our sample runs from 1980to 1995 and includes the fifteen 1995 members of theEuropean Union plus Switzerland and Norway Althoughthe set of EU countries in the location choice analysis onlyincludes nine members it is important to incorporate thedemand emanating from the rest of the European EconomicArea in the calculation of the market potential For instanceSwiss and Austrian consumers are not trivial components ofthe southern German and northern Italian regionsrsquo marketpotential that we consider in the location choice of Japanesefirms Incorporating the demand from those countries in-volves obtaining estimates of their fixed effects as importersand of bilateral trade cost parameters and we thereforeincorporate them in the trade equation

Both internal (dII) and external (dIJ) distances areweighted averages of point-to-point distances between sub-national regions Head and Mayer (2000) provide greaterdetail on their construction and the distance matrix for theEU12 countries (coordinates and population shares of re-gions of Switzerland Austria Norway Sweden and Fin-land have been collected using a combination of Eurostatand national sources) The common language variable LIJ

takes a value of 1 for the United Kingdom and IrelandGermany and Austria Germany and Switzerland Franceand Switzerland Belgium and France and Belgium and theNetherlands The regressions are run for each of the 16years and 18 industries Each component of the marketpotential gathered through the trade equation (the fixedeffects of importing countries and bilateral trade cost esti-mates) are thus industry- year- and country-specific Moredetails on the data and implementation of those regressionscan be found in the data appendix

10 See Anderson and Van Wincoop (2003b) for a survey of the now largeliterature measuring and explaining border effects


Table 1 summarizes the border distance and languageeffects estimated for each two-digit industry This tablegives the average coefficients over two subperiods of oursample 1980ndash1987 and 1988ndash1995 chosen because of thestart of the Single Market Programme in 1987 which wasexpected to yield a fall in border effects The border effectspresented average over the four large sources of demandinside our sample Germany France the United Kingdomand Italy

We find distance effects that average 14 across the twoperiods This number aligns closely with the results ofRedding and Venables (2004) it is slightly superior to usualestimates of gravity equations11 and suggests Harrisrsquos as-sumption of an inverse distance rule is a rough but reason-able approximation Where Harrisrsquos specification appearsinappropriate for Europe is in its omission of the effect ofnational borders12

Border effects for the four core EU countries average 184 inthe first period and 163 in the second period Expressing theirmagnitude in the McCallum (1995) manner within-bordertrade after 1987 remains more than five [exp(163) 51]times as large as cross-border trade after controlling for theeffect of relative distance and characteristics of the tradingpartners (in particular their economic size) Though sizablethese effects are considerably smaller than the value of 20 firstreported by McCallum for the Canada-US border but morecomparable to the rescaled estimates obtained by Andersonand van Wincoop (2003a) when using a specification moreclosely linked to theory In contrast with distance effects there

appears to be a downward trend in the effects of nationalborders in the EU all but one of the border effects declinedWe also find large common-language effects The trade-creating effect of common language is slightly increasing overtime from an average of 039 in the first period to 042 in thesecond one Interestingly we observe large variation acrossindustries in the effect of sharing a language Goods boughtmainly as intermediate inputs tend to have smaller languageeffects than goods destined for final consumption For instancecountry pairs sharing a language trade 25 times more clothingand footwear than pairs lacking a common language

B Specification of the Location Choice Model

We estimate the parameters of the profit equation (7)using a discrete choice model As we do not observe thepotential profitability of each location we rely upon theassumption that firms choose the country yielding the high-est profit The location choice literature makes extensive useof the conditional logit model (CLM) This model requireserror terms that are independent across locations As itseems likely that the unobserved component of profitabilityis correlated among regions in the same nation we use ageneralization that permits such a structure of the randomterm the nested logit model (NLM)13

For estimation purposes it is useful to decompose equa-tion (7) into components that are observable at the nation-state level denoted Vs and at the region level denoted Wras well as remaining random variation that the econometri-cian does not observe denoted r

Ur Vs Wr r11 Anderson and van Wincoop (2003a) also use this fixed-effects spec-

ification in the sensitivity analysis reported in their table 6 p 187Interestingly they mention a substantial rise in the (absolute value of the)distance decay parameter which becomes 125 for US-Canada tradeflows

12 His pioneering study considered the market potential of countieswithin the United States

13 Train (2003) provides a clear description of the nested logit method-ology Mayer and Mucchielli (2002) use the same sample used here toconfirm the validity of a structure nesting region choice under nationchoice

TABLE 1mdashESTIMATES OF BORDER AND DISTANCE EFFECTS AVERAGES FOR THE LARGE EU NATIONS (FRANCE GERMANY ITALY AND THE UNITED KINGDOM)

Industry NameNACECode

1980ndash1987 1988ndash1995

Border()

Dist()

Lang()

Border()

Dist()

Lang()

Metalmdashprimary 22 13 15 04 080 155 17Nonmetallic mineral products 24 229 167 24 211 168 22Chemicals and fibers 25 26 181 119 06 162 122 06Metalmdashfabricated 31 272 154 39 261 147 48Machinery 32 156 108 26 142 106 32Office machines 33 075 083 15 085 079 15Electronics 34 209 099 32 176 101 33Motor vehicles and parts 35 095 168 44 089 152 37Cycles 363 096 212 27 066 187 53Precision instruments 37 151 100 34 105 098 15Food drink and tobacco 41 42 294 129 59 277 138 48Textiles 43 272 128 46 245 126 59Leather 44 176 143 62 123 149 62Clothing and footwear 45 195 151 92 187 148 91Wood and wooden furniture 46 256 191 74 24 196 68Paper printing and publishing 47 266 155 6 254 146 56Rubber and plastics 48 191 135 20 185 136 23Toys and sports 494 072 123 52 052 123 74


Vs includes national policiesmdashcorporate tax rates and socialcharges in this studymdashthat affect all regions Wr includeswages and market potential as well as proxies for otherinput prices and productivity (detailed in section II C) Weenvision the random term r as a shock to ln Ar that isspecific to firm-region pairs (we continue to suppress firmsubscripts for readability) It also contains any other influ-ences on the attractiveness of a location that matter to firmsbut are not included as controls by the econometricianMcFadden (1978) shows that if the distribution of r isgiven by a multivariate extreme value with parameter then the conditional probability that firms choose region rconditional on choosing state s is Prs exp(1Wr Zs)where Zs ln yenis exp(1Wi) is termed the inclusivevalue for state s The probability of choosing nation s isPs exp(Vs Zs Z) where Z ln[yenI exp(VI ZI)] The parameter measures the degree of indepen-dence between the unobserved portion of profitability ofregions in a given state For 0 regions are perfectsubstitutes inside a nation whereas for 1 there is fullindependence and patterns of substitution are the samewithin and between nations We also consider this non-nested version of the model corresponding to 1 In thatcase the probability of choosing a region r is Pr PrsPs exp(Vs Wr Z) Consequently the NLM collapses tothe CLM

The parameters on the components of Ur will be esti-mated by substituting the probabilities (Pr) of the actuallocation choices made by Japanese firms in Europe into alog likelihood function and maximizing

C Implementation of the Location Choice Model

The sample of Japanese firms is extracted from the 1996Survey of Current Manufacturing Operations of JapaneseFirms in Europe issued by the Japan External Trade Orga-nization (JETRO) More than 700 Japanese manufacturinginvestments are listed in this survey with correspondingdate when operation started country of location employ-ment and other details including a detailed description ofthe product manufactured In order to assign investments tosubnational regions the Directory of Japanese-AffiliatedCompanies in the EU 1996ndash1997 also issued by JETROwas used to determine the precise city where the plant waslocated Almost all explanatory variables come from indus-trial statistics issued by Eurostat either at the national or atthe regional level The selection of Japanese investments(452 retained location decisions over the years 1984ndash1995)was essentially driven by the availability of regional andnational data Further details concerning the data can befound in the data appendix which includes table A1 show-ing the regions in our choice set and the number of Japanesefirms each one received

Figure 1 uses these data to plot the Japanese affiliates inthe NUTS 1 region where they invested14 Several importantfeatures of Japanese investment patterns are immediatelyapparent The strong attractiveness of the United Kingdomas a whole the agglomeration in the northern part ofEurope and a tendency of investors to locate in the eco-nomic core of each country (Japanese investors cluster inLondon in the United Kingdom Paris in France Milan inItaly and Barcelona in Spain)

Although the theoretical model of the location decisionfocuses on wages and market potential the large literature onlocation choice includes a number of other explanatory vari-ables We include the standard controls For wages we use thetotal wage bill divided by number of employees in the two-digit industry region Wages do not provide a complete de-scription of labor costs because the functioning of the labormarket (measured by unemployment rates) and government-imposed charges also contribute to the true cost of workersGovernment subsidies and taxes affect the cost of capital (partof v in the model) we control for those through corporate taxrates and eligibility for EU regional policy funding Last wefollow Coughlin et al (1991) and add land area of the regionintended to control for differences in land supply and thereforeland prices which also enter v

Much evidence suggests that related firms tend to cluster inthe same regions We consider three forms of relatedness15 (1)establishments in the same industry (2) affiliates in the sameindustry originating from the same country (Japan) (3) affili-ates owned by the same parent company or affiliated in thesame supplier groups (known as keiretsu in Japan) Clusters ofrelated firms may form regional production networks sellingintermediate inputs to each other and thereby lowering vr Theymay also share knowledge raising Ar It is also likely thatclusters will form around the same exogenous sources of lowinput costs or high productivity Of particular interest to thispaper is the hypothesis that clusters form in areas with highmarket potential in the relevant industry This hypothesiswould receive support if after controlling for market potentialthe presence of same industry firms lowers the attractiveness ofa region Regardless of the underlying mechanism we willrefer to the three cluster variables as agglomeration effects Allvariables used in our specification of production costs aredescribed in table 2

IV Location Choice Results

We begin with an assessment of the performance of ourmarket potential variable in a conventional specificationused in the literature We therefore start with the nonnestedconditional logit estimation of the region choices of Japa-nese firms in Europe (table 3) We then turn to a nested logit

14 NUTS is the official classification for EU regions it has five levelsof geographical detail ranging from 15 NUTS 0 nations to more than100000 NUTS 5 areas

15 Head et al (1999) find evidence that all three forms matter forJapanese location choices in the United States


specification in which we first estimate the choice of regionwithin a given nation and then estimate the choice of nationtaking into account the attractiveness of its constituentregions (table 4)

A Region Choice Nonnested Logit

Table 3 provides results for six different conditional logitestimations of the location choice of Japanese affiliates

FIGURE 1mdashJAPANESE INVESTORS IN EUROPE AT THE END OF 1995


among EU regions Specification (1) shows coefficients forthe standard set of production cost variables excluding thedifficult-to-interpret agglomeration effects Specifications(2) to (4) add successively more sophisticated measures ofdemand culminating with the Krugman market potentialColumn (5) adds nation-level fixed effects to the analysiswhich is a first way to capture unobserved correlations inthe characteristics of regions belonging to the same nationColumn (6) adds the counts of various types of related firms

Though conventional wisdom would predict a negativeeffect for wages and a positive effect for unemployment theresults in specification (1) yield just the opposite As weshall see neither ldquoperverserdquo result is robust to the inclusionof appropriate controls The effect of wages is small andstatistically insignificant after using any of the controls for

demand The absence of a significant negative effect ofregional wages on location choice in all specifications isdisappointing However the result is not out of line withother studies (such as Devereux amp Griffith 1998 and Headet al 1999) Because the standard model of wage determi-nation (the Mincer equation) explains differences in wageswith differences in human capital (education experienceand ability) that are presumably valuable to the firm am-biguous results should perhaps be expected16

16 In unreported regressions we attempted to control for this effect ofhuman capital differences by using the wage of the clothing industry as ameasure of unskilled wages and that of the chemical industry as a measureof skilled wages The idea (for which we thank a referee) was to purge thewage measure of cross-regional human capital variation Unfortunately

TABLE 3mdashCONDITIONAL LOGIT MODEL OF REGION CHOICE

Specification

452 Firms Choosing between 57 Regions

(1) (2) (3) (4) (5) (6)

ln wages 047c 020 012 017 050 013(025) (026) (028) (025) (034) (036)

Unemployment rate 890a 450a 157 322c 434c 135(169) (170) (195) (178) (228) (243)

Obj 1 eligibility 025 012 025 001 022 024(021) (022) (022) (022) (024) (025)

ln regional area 031a 005 058a 059a 058a 021b

(005) (006) (006) (006) (007) (008)Social charges rate 226a 228a 225a 156a 024 001

(038) (038) (038) (038) (183) (186)Corporate tax rate 482a 480a 503a 496a 040 034

(059) (058) (060) (061) (236) (234)ln regional GDP 080a

ln yr (008)ln Harris market potential 188a

ln yenj Ejdrj (021)ln Krugman market potential 111a 107a 034b

ln Mr (013) (014) (016)ln(1 domestic industry count) 052a

(008)ln(1 Japan industry count) 086a

(011)ln(1 network count) 124a

(022)National fixed effects No No No No Yes YesLikelihood ratio index 0054 0079 0077 0073 0079 0126

Standard errors in parenthesesa Significant at 1 levelb Significant at 5 levelc Significant at 10 level

TABLE 2mdashPRODUCTION COST SPECIFICATIONS

Geographic Level Variable Definition

Region-specific (Wr) Wages Total wage bill divided by number of employees in the two-digit industry regionUnemployment rate Unemployed as percentage of the labor forceObj 1 eligibility Equals 1 when the per capita income of the region is less than 75 of the EU

average the critical value for qualifying for EU structural fundsArea Land area of regionDomestic industry count Number of establishments in the two-digit industry regionJapan industry count Number of Japanese affiliates in the three-digit industry regionNetwork count Number of affiliates owned by the same Japanese parent or members of same

vertical keiretsu

Nation-specific (Vs) Corporate tax rate Statutory rate of taxation for profits made by foreign investorsSocial charges rate Nonwage labor costs such as payroll taxes and pension contributions divided

by the total labor cost (wage bill social charges)


Larger regions attract significantly more investors thansmall ones in all specifications except (2) The elasticity issignificantly below the unit value that Coughlin et al (1991)refer to as the ldquodartboardrdquo approach to location decision Wesee these results which appear in the nested estimation aswell as indirect support for the importance of land costs inlocation decisions (using land area rather than land valuesprobably makes sense for the latter is likely to reflectunobserved qualities of the location) Regions eligible forObjective One subsidies have by definition low per capitaincomes and these two effects seem to cancel each otherout yielding insignificant effects

The social charge rate enters with a consistently negativesign until nation-specific fixed effects are added in column(5) The negative and mainly significant effects of socialcharges make sense in that this variable incorporates vari-ation in labor costs that is unrelated to human capitalUnfortunately it does not vary across regions within na-tions and its time-series variation is inadequate to identifya clear result The effect of corporate taxation appears to belarge in line with recent estimates surveyed in the meta-analysis on FDI and corporate taxes by Mooij and Ederveen(2001) The coefficient in column (1) means that a 1-pointrise in the national corporate tax rate yields a fall ofapproximately 5 in the probability that regions in thatcountry will be chosen The comparable average semielas-ticity in the literature reported in table 31 in Mooij andEderveen (2001) is 48 As with social charges the taxeffect is not robust to the inclusion of country fixed effectsThe evolution of corporate tax rates over the period did notseem to influence the location choices by Japanese inves-tors

The main results of interest lie in columns (2) to (5)where we introduce different demand variables and com-pare results of the Krugman market potential with those ofalternative proxies for demand We start in column (2) byadding the most widely used demand measure regionalGDP As expected the coefficient is positive and highlysignificant The measure is hardly an adequate proxy fordemand for few firms would go to the trouble of setting upan overseas factory to serve a single region Column (3)substitutes the simple Harris market potential calculation forregional GDP The coefficient is again significantly positiveand its magnitude is more than double that of local GDPreflecting the attractiveness of central regions in Europe(the ones with a combination of high local demand andproximity to other important sources of demand) The over-all fit of the model is however slightly reduced The Harrismeasure does not take into account border effects variationin distance costs or market crowding due to local compe-tition The Krugman market potential which handles allthree issues is introduced in column (4) The coefficient hasthe expected sign and significance but its magnitude is

lower than the coefficient in Harrisrsquos version and the overallfit of the model is worse (as seen in the 0004 decline of thelikelihood ratio index) The coefficient is however robust tothe inclusion of national fixed effects in column (5)

How substantial is the effect of market potential onlocation choice There are two ways to evaluate the size ofthe effect First the coefficient itself is closely tied to theprobability elasticity which is given by b(1 Pr) whereb is the coefficient and Pr is the probability of choosingregion r On average Pr is the reciprocal of the number ofchoices Using column (5)rsquos estimate of b we find aprobability elasticity of 107(1 157) 105 Thus 10increases in market potential increase the probability ofattracting investment by approximately 10 Though use-ful this way of expressing economic significance does nottake into account the actual variation in the explanatoryvariable

A second approach is to imagine a hypothetical regionwith the mean level of market potential Denote this regionrsquosinitial probability of being chosen as P Then redistributedemand so as to raise the regionrsquos market potential by 1standard deviation Denote the new probability as P Sup-pose that the demand redistribution leaves the overall at-tractiveness of European regions unchanged that is theinclusive value Z is fixed Then the rise in the probabilityof attracting investment will be given by

P

P expblnmeanMr stdevMr

lnmeanMr

1 cvMrb

where cv(Mr) stdev(Mr)mean(Mr) is the coefficient ofvariation of market potential For electronics (NACE 34)the recipient of the most Japanese manufacturing we havea coefficient of variation of Mr of 058 in the year 1995Based on b 107 from specification (5) this implies thata 1-standard-deviation shock would increase the attractive-ness of the average region by 63 The effects on otherindustries which mainly had greater variance in Mr aresometimes considerably larger (for motor vehicles the cor-responding increase is 159)

The agglomeration variables introduced in specification(6) may allow for alternative mechanisms of spatial con-centration such as human capital externalities or techno-logical spillovers which have been the primary emphasis ofpast work on location choice using this type of variablesThis paper is the first to control for market potential usingthe exact functional form dictated by theory Thus if priorfindings of agglomeration effects merely reflected the ab-sence of adequate controls for variation in demand theagglomeration terms would not have significant positiveeffects in our specification and could even enter negativelyto the extent that firms wish to avoid overcrowded markets

the two wages were highly correlated and neither singly nor jointlyyielded negative and significant effects


Although estimated agglomeration effects are often inter-preted as evidence of spatial externalities this is not theonly possible interpretation Even with the many controlsand national fixed effects employed in specification (5) it islikely that omitted variables remain a problem Fixed effectsfor each region are not feasible because they cannot beestimated for regions that received no investment Counts offirms in the same industry will partly reflect omitted vari-ables which form part of the unobserved attractiveness ofregions and therefore correlate with large numbers of do-mestic and Japanese firms Adding agglomeration variablestherefore also has the advantage of mitigating the inevitableomitted variable bias in this type of estimation

The data reject the hypothesis of zero agglomerationeffects despite the presence of the Krugman market poten-tial term Market potential retains a significant positive signbut the coefficient is divided by more than 3 and the overallfit of the estimation is very much improved by including thecounts of related firms Market potential remains an eco-nomically significant factor however A 1-standard-deviation shock now increases attractiveness by 17 inelectronics and 35 in motor vehicles We interpret theresult that counts of related firms have a strong effect evenafter controlling for market potentialmdashwhich recurs in the

nested logitsmdashas indicating that spatial concentration arisesin large part from mechanisms other than the demandlinkages emphasized by Krugman

Another interesting result is that the influence of previouslocation choices by similar firms is increasing in the degreeof relatedness of those competitors All three variables havea large and positive influence but the domestic firms in eachregion have a notably weaker attractive power than otherJapanese affiliates which are themselves superseded bymembers of the same vertical keiretsu This last variablecertainly captures input-output linkages of the Venables(1996) type Its large coefficient is a sign that this type ofvertical linkages might offer more solid empirical explana-tory power than the simple version of the Krugman (1991)model primarily based on final demand linkages

B Region Choices Nested within Nation Choices

The use of the country dummies in specifications (5) and(6) of table 3 helps to mitigate the problem associated withnonindependent errors across regions belonging to the samenation However the country fixed effects do not solveproblems associated with cross-industry and intertemporaldifferences in the attractiveness of nations By considering

TABLE 4mdashNESTED LOGIT MODEL OF REGION AND NATION CHOICE

Specification (1) (2) (3) (4) (6)

421 Firms Choosing within 7 Nations

ln wages 121b 053 012 010 034(048) (053) (050) (050) (052)

Unemployment rate 977a 397 215 103 053(239) (252) (276) (263) (276)

Obj 1 eligibility 089a 071b 081b 074b 054(033) (035) (035) (035) (036)

ln regional area 029a 007 059a 069a 026b

(006) (007) (007) (008) (010)ln regional GDP 082a


ln yenj Ejdrj (026)ln Krugman market potential 141a 051b

(ln Mr) (017) (022)ln(1 domestic industry count) 053a



(026)

Likelihood ratio index 0054 0097 0095 0093 0146

421 Firms Choosing among 7 Nations

Social charges rate 266a 254a 268a 211a 227a

(043) (043) (043) (044) (045)Corporate tax rate 320a 369a 279a 332a 208a

(065) (063) (063) (062) (055)Inclusive value 059a 078a 053a 065a 069a

Zs (007) (007) (006) (006) (006)




the choice of region for a given choice of nation wecondition on all aspects of the nation that do not vary acrossits constituent regions from the perspective of a giveninvestor The drawback is that we must omit the national taxand social charges variables However we reintroduce themwhen estimating the upper level of the decision tree (nationchoice)

The nested region choice forces us to remove Portugaland Ireland because they lack subnational regions in ourdata set This reduces the set of choosers from 452 to 421The choices sets vary in size from 4 (Belgium and theNetherlands) to 11 (Germany Italy and the United King-dom) On average there are 922 choices per Japaneseinvestor

Table 4 reestimates the same specifications as table 3 in anested structure The results are broadly similar As beforecontrolling for demand eliminates a spurious positive effectfor wages Controlling for both market potential and ag-glomeration the wage and unemployment effects conformto conventional wisdom albeit without statistical signifi-cance Market potential has a statistically significant effecteven after controlling for agglomeration effects howeverthe inclusion of the latter improve the fit considerably Bothmarket potential variables have larger effects on regionchoice within nations than they do on nonnested regionchoice17

We again find that ldquotheory doesnrsquot payrdquo in the sense thatthe Harris market potential outperforms the Krugman mar-ket potential in both magnitude and fit In contrast Hanson(2001) finds that when he augments the Harris function toinclude relationships derived from the Krugman model thefit improves However Hansonrsquos simulations show thatincome shocks in the Harris-based formulation of marketpotential have larger effects on the wages paid in nearbycounties than in the Krugman-based formulation The broadsimilarity in results for the two formulations suggests thatwe should be cautious in interpreting the positive effect ofmarket potential as evidence for the particular mechanismsof Krugman-style models of economic geography Othermodels of economic geography could generate observation-ally equivalent results

The lower panel of table 4 reports estimates for the choiceof nation These estimations directly consider the nationalvariables (taxes and social charges) but indirectly considerall the regional determinants of attractiveness as they enterthe inclusive value Zs for each nation We do not includecountry dummies because there would not be sufficientremaining variation in the data

We find consistently negative effects of social chargesCorporate income taxes have large negative effects Like the

results on the nonnested specification those on socialcharges and tax are fragile Including a dummy for Irelandand the United Kingdom makes both effects become muchsmaller and insignificant This result raises the question ofwhether low charges and taxes or some other effect (the useof English) made these countries so attractive to Japaneseinvestors

The inclusive value Zs an index of the maximum ex-pected profitability from locating in a given country con-sidering the underlying characteristics of its regions obtainsreasonable values in all specifications differing quite sig-nificantly from the value of 1 that would obviate the needfor nesting and the value of 0 in which investors areindifferent between regions inside a given country In otherwords the coefficient on the inclusive value supports thevalidity of our country-region nesting structure

V Conclusion

We analyze the determinants of location choices by Jap-anese firms in Europe Our work is particularly concernedwith the appropriate way to take into account the spatialdistribution of demand in location decisions We rigorouslylink the optimal location choice of Japanese investors to atheoretical model of imperfect competition in a multiloca-tion setting The underlying profit equation incorporates aterm that is closely related to the market potential indexoriginally introduced and used by geographers (Harris1954) Our theory-derived term aggregates the spatial dis-tribution of demand weighted by trade costs and the locationof potential competitors We estimate the border and dis-tance effects that determine market accessibility using abilateral trade equation implied by the same model thatgenerates the profit equation

We find that demand does matter for location choice A10 increase in our market potential term raises the chanceof a region being chosen by 3 to 11 depending on thespecification Despite the fact that we bring theory to em-pirical implementation in a structural way the ldquocorrectrdquomeasure of market potential actually underperforms theatheoretical Harris (1954) measure Moreover nonstructuralagglomeration variables retain a robust influence Theseresults suggest that the downstream linkages emphasized inKrugman (1991) are not the only or even the main cause ofagglomeration Future research should probably considerother reasons why firms cluster It does not seem possible tofalsify the hypothesis that observed agglomeration effectsmerely reflect omitted exogenous location attributes How-ever a natural follow-up to this paper would be to estimatestructural location choice models that implement the Ven-ables (1996) setup with upstream and downstream linkagesbased on an input-output matrix

REFERENCES

Anderson J and E van Wincoop ldquoGravity with Gravitas A Solution tothe Border Puzzlerdquo American Economic Review 931 (2003a)170ndash192

17 Due to the smaller number of alternatives in the nested logit (922 onaverage versus 57 in the nonnested estimation) direct inspection of thecoefficients is inadequate to make the comparison Adjusting coefficientsfrom specification (4) shows 141(1 1922) 126 111(1 157) 109 The difference is larger for specification (6)


ldquoTrade Costsrdquo Boston College manuscript (2003b)Baldwin R R Forslid P Martin G Ottaviano and F Robert-Nicoud

Public Policies and Economic Geography (Princeton NJ Prince-ton University Press 2003)

Chen N ldquoIntra-national versus International Trade in the EuropeanUnion Why Do National Borders Matterrdquo Journal of Interna-tional Economics 631 (2004) 93ndash118

Coughlin C C J V Terza and V Arromdee ldquoState Characteristics andthe Location of Foreign Direct Investment in the United Statesrdquothis REVIEW 68 (1991) 675ndash683

Crozet M ldquoDo Migrants Believe in Market Potentialrdquo Journal ofEconomic Geography 44 (2004)

Davis D and D Weinstein ldquoEconomic Geography and Regional Pro-duction Structure An Empirical Investigationrdquo European Eco-nomic Review 432 (1999) 379ndash407

ldquoMarket Access Economic Geography and Comparative Advan-tage An Empirical Assessmentrdquo Journal of International Econom-ics 591 (2003) 1ndash23

Devereux M and R Griffith ldquoTaxes and the Location of ProductionEvidence from a Panel of US Multinationalsrdquo Journal of PublicEconomics 683 (1998) 335ndash367

Devereux M R Griffith and A Klemm ldquoCorporate Income Tax Re-forms and International Tax Competitionrdquo Economic Policy 35(2002) 449ndash488

Dixit A and J Stiglitz ldquoMonopolistic Competition and Optimum Prod-uct Diversityrdquo American Economic Review 673 (1977) 297ndash308

Dodwell Marketing Consultants Industrial Groupings in Japan (TokyoDodwell Marketing Consultants 1988)

Friedman J D Gerlowski and J Silberman ldquoWhat Attracts ForeignMultinational Corporationsrdquo Journal of Regional Science 324(1992) 403ndash418

Fujita M P Krugman and A Venables The Spatial Economy CitiesRegions and International Trade (Cambridge MA MIT Press1999)

Hanson G ldquoMarket Potential Increasing Returns and Geographic Con-centrationrdquo University of California San Diego manuscript(2001)

Harris C ldquoThe Market as a Factor in the Localization of Industry in theUnited Statesrdquo Annals of the Association of American Geogra-phers 64 (1954) 315ndash348

Head K and T Mayer ldquoNon-Europe The Magnitude and Causes ofMarket Fragmentation in Europerdquo Weltwirtschaftliches Archiv1362 (2000) 285ndash314

ldquoMarket Potential and the Location of Japanese Investment in theEuropean Unionrdquo Centre for Economic Policy Research discus-sion paper no 3455 (2002)

Head K J Ries and D Swenson ldquoAttracting Foreign ManufacturingInvestment Promotion and Agglomerationrdquo Regional Science andUrban Economics 292 (1999) 197ndash218

Henderson J V A Kuncoro and M Turner ldquoIndustrial Development inCitiesrdquo Journal of Political Economy 1035 (1995) 1067ndash1090

Krugman P ldquoScale Economies Product Differentiation and the Patternof Traderdquo American Economic Review 70 (1980) 950ndash959

ldquoIncreasing Returns and Economic Geographyrdquo Journal of Po-litical Economy 993 (1991) 483ndash499

ldquoA Dynamic Spatial Modelrdquo National Bureau of EconomicResearch working paper no 4219 (1992)

Mayer T and J-L Mucchielli ldquoHierarchical Location Choice and Mul-tinational Firmsrsquo Strategyrdquo (pp 133ndash158) in J Dunning and J-LMucchielli (Eds) Multinational Firms The Global and LocalDilemma (London Routledge 2002)

McCallum J ldquoNational Borders Matter Canada-US Regional TradePatternsrdquo American Economic Review 85 (1995) 615ndash623

McFadden D ldquoModelling the Choice of Residential Locationrdquo (pp75ndash96) in A Karlquist et al (Eds) Spatial Interaction Theoryand Residential Location (Amsterdam North-Holland 1978)

Mooij R A and S Ederveen ldquoTaxation and Foreign Direct InvestmentA Synthesis of Empirical Researchrdquo CPB discussion paper no 003(2001)

Redding S and A Venables ldquoEconomic Geography and InternationalInequalityrdquo Journal of International Economics 621 (2004) 53ndash82

Strange R Japanese Manufacturing Investment in Europe Its Impact onthe UK Economy (London Routledge 1993)

Train K Discrete Choice Methods with Simulation (Cambridge UKCambridge University Press 2003) pp 81ndash90

Venables A J ldquoEquilibrium Locations of Vertically Linked IndustriesrdquoInternational Economic Review 372 (1996) 341ndash359

Wolf Holger C ldquoIntranational Home Bias in Traderdquo this REVIEW 824(2000) 555ndash563

Yamawaki H J-M Thiran and L Barbarito ldquoUS and Japanese Multi-nationals in European Manufacturing Location Patterns and HostRegionCountry Characteristicsrdquo in K Fukasaku F Kimura andS Urata (Eds) Asia and Europe Beyond Competing Regionalism(Brighton Sussex Academic Press 1998)

DATA APPENDIX

1 Trade Equation Estimation

Most data used in estimating equation (9) come from Eurostat data-bases and were in part already used in Head and Mayer (2000) wheremore details can be found The COMEXT database provides bilateraltrade flows The VISA database provides the production data used tocalculate internal trade flows of a country subtracting its value from thevalue of total exports of the country Production values are adjusted toallow for the fact that some countries reported data only for firms largerthan 20 employees in the years we are considering whereas trade flowsare exhaustive Trade and production figures are then both converted intoNACE two-digit industries in order to match the level of detail of thesubsequent location choice estimation Though this is straightforward forproduction data and trade flow data after 1987 (provided by Eurostat inNACE three-digit) a large concordance work is needed to convert tradeflows for the previous period from NIMEXE to NACE three-digit Thishas been done using a correspondence available from the site httpwwweiitorg

Distance calculations are crucial in this paper for both the tradeequation and profit equation estimations We calculate the distance of onenation to anothermdashor itselfmdashas a weighted average of subnational dis-tances Considering two countries I and J (the origin and destinationcountries of a given flow) respectively consisting of regions indexed i I and j J the following formula provides both external and internaldistances

dIJ iI

jJ

jdij i

We define dij as the distance between the centers of regions i and j andi as the weight of region i calculated as the share of population in 1990for both origin and destination weights The distance from a region toitself is obtained using a simple geographical approximation Each regionis approximated as a disk in which all production concentrates at thecenter and consumers are uniformly distributed throughout the rest of thearea The average distance between a producer and a consumer is thengiven by

dii 0

R

r2r

R2 dr

where R denotes the radius of the disk and 2rR2 is the density ofconsumers at any given distance r to the center We obtain R as the squareroot of the area A divided by Integrating we obtain dii 2

3R 2

3A 0376AThe estimation procedure consists of 288 OLS regressions (18 indus-

tries 16 years) providing the estimates to be used for the constructionof market potentials Each regression yields the importersrsquo fixed effect theestimated effects of bilateral distance and common language and a set ofimporter-specific border effects Those border effects are identified usinginternal trade flows and therefore cannot be estimated for observationswhere production figures are missing This is in particular the case fornon-EU countries for which Eurostat does not report data We apply theestimated German border effect to northern European countries (SwedenFinland Norway Switzerland and Austria) and the estimated Frenchborder effect for southern European countries Greece is assigned the


French border effect for the whole period Spain and Portugal havemissing data before their membership in 1986 For this year we calculatea ratio of French to Spanish and Portuguese border effects and apply thisratio to the French border effect to get Spanish and Portuguese values forthe preceding years After those adjustments there are a few remainingholes in the data mainly resulting from the well-known confidentialityissues in Belgium and the Netherlands for production data Those missingfigures are filled in taking an average of the industry and country averageborder effects for the corresponding year

2 Location Choice Estimation

2a Regions and years used

The regional level choice sets incorporate 57 regions in Europe usingthe NUTS 1 level of detail for Germany (11 regions) France (8 regions)Italy (11 regions) the United Kingdom (11 regions) Spain (6 regions) theNetherlands (4 regions) and Belgium (4 regions) Ireland and Portugal areconsidered as single-region countries Out of those 57 regions 50 werechosen at least once by Japanese investors

Industry-level regional data availability limits the sample to the years1984ndash1995 Although there are some Japanese investments in the lateseventies and early eighties the vast majority of the investments tookplace in the late eighties

2b Affiliates

The location choices of Japanese affiliates are mainly extracted fromJETROrsquos Survey of Current Manufacturing Operations of Japanese Firmsin Europe 1996 This source provides in particular the country chosen andthe date on which operations started for all manufacturing affiliates whichhad established operations in Europe by the end of 1995 Droppinginvestments in a set of countries (Luxembourg Denmark Greece AustriaFinland Sweden Norway Switzerland and Iceland) and years (before1984 and after 1995) for which explanatory variables were not availablewe obtain 452 location choices to be explained

We then identified for each firm the city in which the production unitwas located This information appears in a larger document also issued byJETRO The Directory of Japanese-Affiliated Companies in the EU1996ndash1997

A crucial matter for our study is the quality of this information It hadto be checked that in the directory the affiliatersquos location reported was notthat of the headquarters but that of the actual production unit Fortunatelythe directory almost always specifies both the location of the headquartersand the location of the plant However the information was doublechecked using three alternative sources The database used by Yamawakiet al (1998) mostly using data from Toyo Keisai and kindly madeavailable by Hideki Yamawaki was of great help Table 54 in Strange(1992) also confirmed the locations of Japanese subsidiaries in the UnitedKingdom Finally a document from the DATAR helped to check locationsfor France

The Japanese affiliatesrsquo location choice is of course our dependentvariable but is also used to construct the Japanese counts agglomerationvariable This variable consists for each choice of a cumulative count ofsame three-digit industry affiliates that chose each region from the firstyear where Japanese FDI started in Europe until the year preceding thechoice under consideration We also use these data to calculate thenetwork counts For each prospective investor it counts the number ofaffiliated firms that already chose the region in question We definedldquoaffiliatedrdquo so that it includes investments with the same Japanese parentcompany (for example two different Sony factories) or investments fromparent companies that are members of the same industrial group (forexample a Toyota assembly factory and a Nippondenso automotiveelectronics factory) We defined industrial groups using Dodwellrsquos (1988)lists of vertical keiretsu

3 Market Potential Calculation

As detailed in the text the calculation of regional Krugman marketpotential involves (1) a regional allocation of competition-weighted ex-penditure estimated at the nation and industry levels in the trade equationsdetailed above and (2) measures of bilateral distances between regionsacross the European Union

We allocate the national competition-weighted expenditure of eachindustry among its regions according to the share of national GDPRegional GDP shares come from Eurostatrsquos REGIO database Bilateralregional geodesic distances are calculated using the coordinates of themain city in each region which were collected manually Distance insidea region only requires data on the land area of the region also obtainedfrom the REGIO database

Note finally that market potential calculation involves six countries thatare not present in the location choice set of Japanese affiliates AustriaDenmark Finland Norway Sweden and Switzerland are included in theanalysis in order to allow for the fact that some regions in the choice sethave their market potential enhanced more than others by demand ema-nating from those six countries

The calculation of the Harris market potential also involves bilateraldistances between all regions and national apparent consumption in eachindustry allocated to regions using the same regional shares of GDP as inthe Krugman market potential National apparent consumption is obtainedusing the same Eurostat data as were used in the trade equations anddetailed above

4 Industry-Level Regional Data

The main source of industry-level regional data is the Eurostat publi-cation Structure and Activity of Industry Annual Inquiry Principal Re-sults Regional Data It consists of two-digit NACE data essentiallyavailable for NUTS 1 regions This database contains the number ofestablishments employment statistics and the wage bill for each industry-region combination For single-region countries national-level data areused

An electronic version exists with regional data for the years 1989 to1992 but in fact 1992 has many missing values We additionally used theprinted version for 1984 and 1987 Observations for 1984 to 1986 arematched with the 1984 data Observations for 1987 and 1988 are matchedwith the 1987 data 1989 1990 and 1991 observations are matched withsame-year data Observations from 1992 to 1995 are matched with 1991data NACE 26 (man-made fibers industry) was excluded from the samplebecause too few data were available When the data were missing for aparticular NUTS 1 region the following procedure was adopted Themissing values are often due to missing values in small NUTS 2 subre-gions (for instance Corsica in Mediterranee or Val drsquoAoste in Nord Ovesthave many missing employment and wage values because there are onlyone or two firms In this case we just sum the remaining NUTS 2 regionsto get what appears to be a very precise approximation of the true data Inother cases too many data from subregions were missing for a particularyear we then replaced the figure with its value for the nearest yearavailable As a general pattern the main problems in data availabilityconcerned Netherlands and (even more) Belgium

5 National-Level Data

Variables at the national level are the corporate tax rate and the socialcharges rate Social charges rates use Eurostat data on nonwage labor costs(such as payroll taxes and pension contributions) at the two-digit industry-country level The source is the national version of the database Structureand Activity of Industry also used to obtain regional data The variableconsists of the share of those charges in the total labor cost of the industryThe corporate tax data are the national statutory tax rates taken from thedata set put together by Devereux et al (2002) and made available athttpwww1ifsorgukcorptaxindexshtml


TABLE A1mdashTHE LOCATION OF JAPANESE AFFILIATES IN EUROPEAN REGIONS IN 1996

Region Code Jpn Firms Region Code Jpn Firms

Belgium

Brabant BE0 7Bruxelles-Brussels BE1 4Vlaams Gewest BE2 17Region Wallonne BE3 8

Germany

Baden-Wuerttemberg DE1 8Bayern DE2 16Berlin DE3 1Bremen DE5 0Hamburg DE6 6Hessen DE7 12Niedersachsen DE9 10Nordrhein-Westfalen DEA 25Rheinland-Pfalz DEB 6Saarland DEC 0Schleswig-Holstein DEF 4

United Kingdom

North UK1 18Yorkshire and Humberside UK2 10East Midlands UK3 9East Anglia UK4 3South East (UK) UK5 51South West (UK) UK6 13West Midlands UK7 27North West (UK) UK8 11Wales UK9 31Scotland UKA 21Northern Ireland UKB 2

SpainNoroeste ES1 1Noreste ES2 5Madrid ES3 8Centro (e) ES4 2Este ES5 34Sur ES6 3

Netherlands

Noord-Nederland NL1 0Oost-Nederland NL2 6West-Nederland NL3 12Zuid-Nederland NL4 19

ItalyNord Ovest IT1 6Lombardia IT2 19Nord Est IT3 2Emilia-Romagna IT4 4Centro (i) IT5 4Lazio IT6 3Abruzzi-Molise IT7 0Campania IT8 0Sud IT9 1Sicilia ITA 0Sardegna ITB 0

France

Ile de France FR1 32Bassin Parisien FR2 18Nord-Pas-de-Calais FR3 2Est FR4 16Ouest FR5 6Sud-Ouest FR6 11Centre-Est FR7 7Mediterranee FR8 2

Ireland

Ireland IE 30

Portugal

Portugal PT 12


as Friedman Gerlowski and Silberman (1992) HendersonKuncoro and Turner (1995) and Head et al (1999) con-sider nonlocal demand but not using measures derived fromtheory5 In particular theory suggests that nonlocal demandmust be discounted for bilateral trade impediments Further-more a given amount of market access contributes less toprofits when a firmrsquos competitors have access to the samemarkets We follow Krugman (1992) in adjusting the marketpotential measure to take into account the location ofcompetitors

The paper proceeds as follows In section II we derive alinear-in-logs equation that relates the profitability of alocation to a prospective foreign investor to a measure ofthat locationrsquos access to demand We then show how toestimate the distance and border effects that impede marketaccess using bilateral trade data In section III we report theresults from the trade equation and show how we use themto calculate market potential We then discuss our sample ofJapanese investors and the set of possible location choicesOur location-choice results are detailed in section IV andwe conclude and propose directions for future work insection V

II The Model

Let Er denote expenditure in a representative industry(we omit industry subscripts for notational simplicity) inregion r Consumers (who may be firms or individuals)allocate their expenditures across differentiated varieties inthe representative industry They have constant-elasticity-of-substitution subutility functions for each industry Max-imizing this subutility function subject to expenditure Erand the delivered prices from all R possible product originswe obtain the demand curve for the representative variety inthe representative industry as

qij pij

yenr1R nrprj

1 Ej (1)

where pij is the delivered price faced by consumers inregion j (destination) for products from region i (origin) Itis the product of the mill price pj and the trade cost factorij Trade costs include all transaction costs associated withmoving goods across space and national borders

A The Profit Equation for Foreign Affiliates

Each firm sets its mill prices to maximize profits Fol-lowing Dixit and Stiglitz (1977) firms treat the elasticity ofsubstitution as if it were the price elasticity of demand(this may also be interpreted as the assumption that eachfirm has infinitesimal market share) The resulting mill

prices are simple markups over marginal costs denoted crthus pr cr[( 1)] Substituting into equation (1) weobtain the quantity that a firm producing in region i woulddeliver6 to each destination region j

qij 1

ciij

GjEj (2)

where Gj yenr nr(crrj)1 The gross profit earned in eachdestination region j for a firm producing in region i is

ij pi ciijqij ciij

1

GjEj (3)

This gross profit is an increasing function of the expenditureof country j on the considered industry The fraction mul-tiplying Ei depends on the costs of the representative firmrelative to its competitors from all R regions In the numer-ator we see that profits are decreasing in local (region i)production costs Lower trade costs to reach region j (that isa low ij) also raise profits Because the effect of trade costsis always moderated by the elasticity of substitution weintroduce the notation of ij ij

1 to measure the accessof exporters from i to market j7 The denominator containsthe corresponding characteristics of competing suppliersNote that the denominator contains a factor capturing theidea that competition is fiercer and profits are thereforelower when varieties are less differentiated from each other

Summing the gross profits earned in each market andsubtracting the fixed costs Fr necessary to establish a plantin region r we obtain the aggregate net profit r to beearned in each potential location r

r cr

1

j1

R

rj

Ej

Gj Fr

cr1

Mr Fr (4)

where

Mr j

rjEj

Gj

We will refer to Mr as the Krugman market potentialbecause an expression for it first appeared in Krugman(1992) The profit equation suggests that firms face atradeoff between low production costs and high marketpotential8

5 Friedman et al (1992) use a distance-weighted sum of per capitaincomes Henderson et al (1995) use distance to nearest major businessdistrict and Head et al (1999) sum the personal incomes of adjacentstates

6 The iceberg convention implies that to deliver q units the firm mustship q units

7 Baldwin et al (2003) refer to as ldquofreeness of traderdquo and mainlyconsider cases of symmetric trade costs (ij ji) The term marketaccess is better suited to our case where we find asymmetries in tradecosts

8 The wage equation analyzed by Fujita et al (1999 p 53) can bederived from the profit equation by assuming that free entry sets equation(4) equal to 0 In their specification production requires 1-unit of labor perunit of output and F units of labor as overhead Denoting the cost of a



Ur ln lnr F

1 ln cr 11 ln Mr

(5)






Ur ln wr 11 ln Mr

1 ln vr ln Ar(7)









XIJ pIJqIJnI nI

cI1 IJEJ

GJ





wr 1

FMr 1








(9)


ij expJ LIJ dij


ij dij


ii dii 2

3areai




























1980ndash1987 1988ndash1995

Border()

Dist()

Lang()

Border()

Dist()

Lang()

























Specification


(1) (2) (3) (4) (5) (6)

ln wages 047c 020 012 017 050 013(025) (026) (028) (025) (034) (036)


Obj 1 eligibility 025 012 025 001 022 024(021) (022) (022) (022) (024) (025)




(059) (058) (060) (061) (236) (234)ln regional GDP 080a












vertical keiretsu










P


lnmeanMr

1 cvMrb












Specification (1) (2) (3) (4) (6)


ln wages 121b 053 012 010 034(048) (053) (050) (050) (052)




(006) (007) (007) (008) (010)ln regional GDP 082a






(026)






Zs (007) (007) (006) (006) (006)












V Conclusion



REFERENCES




































DATA APPENDIX




dIJ iI

jJ

jdij i


dii 0

R

r2r

R2 dr


3R 2









2b Affiliates


















Belgium


Germany


United Kingdom



Netherlands



France


Ireland

Ireland IE 30

Portugal

Portugal PT 12



Ur ln lnr F

1 ln cr 11 ln Mr

(5)






Ur ln wr 11 ln Mr

1 ln vr ln Ar(7)









XIJ pIJqIJnI nI

cI1 IJEJ

GJ





wr 1

FMr 1








(9)


ij expJ LIJ dij


ij dij


ii dii 2

3areai




























1980ndash1987 1988ndash1995

Border()

Dist()

Lang()

Border()

Dist()

Lang()

























Specification


(1) (2) (3) (4) (5) (6)

ln wages 047c 020 012 017 050 013(025) (026) (028) (025) (034) (036)


Obj 1 eligibility 025 012 025 001 022 024(021) (022) (022) (022) (024) (025)




(059) (058) (060) (061) (236) (234)ln regional GDP 080a












vertical keiretsu










P


lnmeanMr

1 cvMrb












Specification (1) (2) (3) (4) (6)


ln wages 121b 053 012 010 034(048) (053) (050) (050) (052)




(006) (007) (007) (008) (010)ln regional GDP 082a






(026)






Zs (007) (007) (006) (006) (006)












V Conclusion



REFERENCES




































DATA APPENDIX




dIJ iI

jJ

jdij i


dii 0

R

r2r

R2 dr


3R 2









2b Affiliates


















Belgium


Germany


United Kingdom



Netherlands



France


Ireland

Ireland IE 30

Portugal

Portugal PT 12





(9)


ij expJ LIJ dij


ij dij


ii dii 2

3areai




























1980ndash1987 1988ndash1995

Border()

Dist()

Lang()

Border()

Dist()

Lang()

























Specification


(1) (2) (3) (4) (5) (6)

ln wages 047c 020 012 017 050 013(025) (026) (028) (025) (034) (036)


Obj 1 eligibility 025 012 025 001 022 024(021) (022) (022) (022) (024) (025)




(059) (058) (060) (061) (236) (234)ln regional GDP 080a












vertical keiretsu










P


lnmeanMr

1 cvMrb












Specification (1) (2) (3) (4) (6)


ln wages 121b 053 012 010 034(048) (053) (050) (050) (052)




(006) (007) (007) (008) (010)ln regional GDP 082a






(026)






Zs (007) (007) (006) (006) (006)












V Conclusion



REFERENCES




































DATA APPENDIX




dIJ iI

jJ

jdij i


dii 0

R

r2r

R2 dr


3R 2









2b Affiliates


















Belgium


Germany


United Kingdom



Netherlands



France


Ireland

Ireland IE 30

Portugal

Portugal PT 12















1980ndash1987 1988ndash1995

Border()

Dist()

Lang()

Border()

Dist()

Lang()

























Specification


(1) (2) (3) (4) (5) (6)

ln wages 047c 020 012 017 050 013(025) (026) (028) (025) (034) (036)


Obj 1 eligibility 025 012 025 001 022 024(021) (022) (022) (022) (024) (025)




(059) (058) (060) (061) (236) (234)ln regional GDP 080a












vertical keiretsu










P


lnmeanMr

1 cvMrb












Specification (1) (2) (3) (4) (6)


ln wages 121b 053 012 010 034(048) (053) (050) (050) (052)




(006) (007) (007) (008) (010)ln regional GDP 082a






(026)






Zs (007) (007) (006) (006) (006)












V Conclusion



REFERENCES




































DATA APPENDIX




dIJ iI

jJ

jdij i


dii 0

R

r2r

R2 dr


3R 2









2b Affiliates


















Belgium


Germany


United Kingdom



Netherlands



France


Ireland

Ireland IE 30

Portugal

Portugal PT 12
























Specification


(1) (2) (3) (4) (5) (6)

ln wages 047c 020 012 017 050 013(025) (026) (028) (025) (034) (036)


Obj 1 eligibility 025 012 025 001 022 024(021) (022) (022) (022) (024) (025)




(059) (058) (060) (061) (236) (234)ln regional GDP 080a












vertical keiretsu










P


lnmeanMr

1 cvMrb












Specification (1) (2) (3) (4) (6)


ln wages 121b 053 012 010 034(048) (053) (050) (050) (052)




(006) (007) (007) (008) (010)ln regional GDP 082a






(026)






Zs (007) (007) (006) (006) (006)












V Conclusion



REFERENCES




































DATA APPENDIX




dIJ iI

jJ

jdij i


dii 0

R

r2r

R2 dr


3R 2









2b Affiliates


















Belgium


Germany


United Kingdom



Netherlands



France


Ireland

Ireland IE 30

Portugal

Portugal PT 12








P


lnmeanMr

1 cvMrb












Specification (1) (2) (3) (4) (6)


ln wages 121b 053 012 010 034(048) (053) (050) (050) (052)




(006) (007) (007) (008) (010)ln regional GDP 082a






(026)






Zs (007) (007) (006) (006) (006)












V Conclusion



REFERENCES




































DATA APPENDIX




dIJ iI

jJ

jdij i


dii 0

R

r2r

R2 dr


3R 2









2b Affiliates


















Belgium


Germany


United Kingdom



Netherlands



France


Ireland

Ireland IE 30

Portugal

Portugal PT 12









Specification (1) (2) (3) (4) (6)


ln wages 121b 053 012 010 034(048) (053) (050) (050) (052)




(006) (007) (007) (008) (010)ln regional GDP 082a






(026)






Zs (007) (007) (006) (006) (006)












V Conclusion



REFERENCES




































DATA APPENDIX




dIJ iI

jJ

jdij i


dii 0

R

r2r

R2 dr


3R 2









2b Affiliates


















Belgium


Germany


United Kingdom



Netherlands



France


Ireland

Ireland IE 30

Portugal

Portugal PT 12










V Conclusion



REFERENCES




































DATA APPENDIX




dIJ iI

jJ

jdij i


dii 0

R

r2r

R2 dr


3R 2









2b Affiliates


















Belgium


Germany


United Kingdom



Netherlands



France


Ireland

Ireland IE 30

Portugal

Portugal PT 12


































DATA APPENDIX




dIJ iI

jJ

jdij i


dii 0

R

r2r

R2 dr


3R 2









2b Affiliates


















Belgium


Germany


United Kingdom



Netherlands



France


Ireland

Ireland IE 30

Portugal

Portugal PT 12







2b Affiliates


















Belgium


Germany


United Kingdom



Netherlands



France


Ireland

Ireland IE 30

Portugal

Portugal PT 12




Belgium


Germany


United Kingdom



Netherlands



France


Ireland

Ireland IE 30

Portugal

Portugal PT 12


market potential and the location of...

Documents