Iner-regional Transporaion and Economic Developmen: A Case Sudy of Regional Agglomeraion Economies in Japan Jepan Wewioo a and Hironori Kao b a Deparmen of Civil Engineering, The Universiy of Tokyo 7-3-1, Hongo, Bunkyo-ku, Tokyo 113-8656, Japan Phone: +81-3-5841-7451; Fax: +81-3-5841-7496 E-mail: jepanw@ip.civil..u-okyo.ac.jp b Deparmen of Civil Engineering, The Universiy of Tokyo 7-3-1, Hongo, Bunkyo-ku, Tokyo 113-8656, Japan Phone: +81-3-5841-7451; Fax: +81-3-5841-7496 E-mail: kao@civil..u-okyo.ac.jp Absrac. This sudy invesigaes he benefi from ransporaion o economy hrough agglomeraion. We analyze empirically he impacs of agglomeraion on regional economic reurn using an economeric approach assuming hree ypes of agglomeraion economics: localizaion agglomeraion, urbanizaion agglomeraion, and mixed agglomeraion. We esimae he agglomeraion elasiciies of 11 indusries using iner-regional ransporaion nework daa and regional socio-economic panel daa for 1981, 1986, 1991, 1996, 001, and 006, covering 47 prefecures in Japan. Our resuls show ha, on average, he indirec benefi of produciviy improvemen hrough localizaion agglomeraion ends o be more significan han ha hrough urbanizaion agglomeraion. We also find ha while mining enjoys significan benefi from urbanizaion raher han localizaion agglomeraion and he ransporaion/communicaion indusry enjoys significan benefi from localizaion raher han urbanizaion agglomeraion, finance/insurance and real esae migh benefi from boh agglomeraion economies. We furher find negaive elasiciies in he agriculure and service indusries; his could be parly due o he indusries characerisics. Keywords: Iner-regional ransporaion, economic developmen, agglomeraion, Japan, panel daa 1
1. Inroducion Prolonged economic sagnaion has raised global concerns abou infrasrucure invesmen, wih beer abiliy o recover deb and higher rae of reurn from invesmen being he main policy agenda for he coming years. In line wih his agenda, ransporaion infrasrucure invesmen has been given high prioriy in boh he developing and developed world. Generally, he marginal gain from new ransporaion invesmen could o be smaller in developed regions ha already have well-esablished ransporaion neworks; hus, new ransporaion invesmen is less likely o be acceped hrough convenional appraisal. Consequenly, he addiional benefis o an economy no capured previously by he convenional mehod are now being proposed and are inroduced ino projec appraisal in some developed counries. For insance, in he Unied Kingdom, poenially addiional benefis are calculaed separaely and evaluaed using sensiiviy analysis in addiion o he convenional cos-benefi analysis (Deparmen for Transpor, 014a). In heir pioneering discussions, SACTRA (1999) porrayed cerain indirec benefis, which were laer incorporaed ino a guideline (Deparmen for Transpor, 014b), he so called Wider Impac (WI). The Unied Kingdom s WI relaes o he agglomeraion effec, addiional benefis from imperfec marke compeiion, and ax benefis from addiional labor from ransporaion improvemen. A muli-crieria analysis has been widely adoped in he US conex, whereby differen saes follow differen crieria and raings in he appraisal procedure (Weisbrod, 015). However, Weisbrod e al. (014) poined ou ha WI can help quanify he economic impacs of ransporaion ino moneary erms and hus make i more appealing o decision makers for projecs where economic developmen is he main invesmen arge. However, noe ha criicisms of bias and double-couning effec have been generally raised in WI esimaions. Thus, he concep iself needs o be exensively sudied and we sill require rigid evidence o prove is exisence in he various cos and benefi gains from ransporaion invesmen. As regards he relaionship beween ransporaion and economic developmen, Lakshmanan (011) showed he economic consequences of ransporaion invesmen o consis of gains from rade, echnology diffusion, and he coordinaion from he Big Push effec as well as from agglomeraion. Liman (010) discussed how ransporaion affecs economic developmen: as he ransporaion sysem s efficiency improves, he ransporaion cos decreases and producers can yield more oupu per uni. In a broader sense, he producion of more goods and services from he gains of ransporaion developmen leads o economic developmen. Rephann and Isserman (1994) caegorized he economic effecs of highways ino four ypes: he emporal effecs semming from consrucion processes; he indusrial effecs varying hrough ime and across ypes of indusries; he spaial effec in local and regional scales; and a synhesis of he emporal, indusrial, and spaial effecs. From hese sudies, we can summarize ha he economic impacs of ransporaion developmen emerge from he premium in accessibiliy and ransporaion cos, or, in general, from he reducion in generalized cos leading o a more producive economy. Numerous sudies have suppored wih evidence he hypohesis of ransporaion impacs on economic developmen. From ancien imes, Roman highways were buil primarily for miliary logisics, alhough hey also indirecly benefied he economy hrough he expansion of iner-regional rade and services, such as mail and privae ransporaion (Berechman, 003). Following he pioneering empirical work of Aschauer (1989) suggesing an expeced reurn of up o 4% from invesmen in he core US infrasrucure during 1949 1985, Canning and Fay (1993) unveiled he posiive and significan impac of ransporaion infrasrucure on naional economic growh by esimaing he oal facor produciviy in he Cobb Douglas producion funcion. Furhermore, Chandra and Thompson (000) used he age of inersae highway as ransporaion facor, suggesing ha he presence of highways affec indusrial growh in various secors, alhough economic aciviies remain closer o he highway. These resuls of posiive reurns from highways concur wih he resuls of Duranon and Turner (01), who saed ha an increase in inersae highway sock leads o ciy employmen growh by around 15%. A recen sudy by Farhadi (015) also invesigaed he ransporaion infrasrucure effec across he OECD counries, and concluded ha ransporaion invesmen resuls in posiive reurns o GDP, especially fuure GDP, alhough he effec is sill less compared o oher infrasrucure invesmens. Addiionally, he World Bank (1994) and Canning (1998) also showed a posiive relaionship beween GDP and infrasrucure sock, wih higher GDP per capia in counries wih higher infrasrucure sock per capia. Alhough several sudies addressed he relaionship beween ransporaion invesmen and economic developmen, is mechanism has no been explained. Agglomeraion could be one of he facors for he relaionship, as proposed in he Unied Kingdom s WI. An agglomeraion economy is ypically defined as he benefi from firms saying close ogeher. The concep of indusrial scale of economies in Marshall (190) has been furher formulaed ino hree facors ha lead o agglomeraion economies, all closely relaed o ransporaion service. Firs, agglomeraion creaes clusers of firms wherein producers, suppliers, and cusomers are locaed ogeher; his reduces he cos of goods, maerials, and even services. Beer ransporaion services could creae more opporuniies for firms o access beer and cheaper inpu maerial. Second, his effec is observed in he case of workers as well. A larger pool of workers for access by firms enables a beer maching of firms and workers, which improves produciviy,
because skilled workers can beer mach heir work wih heir skills. Since beer accessibiliy inspires workers o work away from home, larger agglomeraion can be aained in labor pooling hrough beer ransporaion. Third, he so-called knowledge spillover can be expeced in agglomeraed areas. One of he mos famous examples is he Silicon Valley; many firms including semiconducor manufacurers and IT firms are locaed ogeher here, leading o an environmen of muual learning and assisance. Again, beer ransporaion encourages more meeings, discussions, or even workshops for individuals, and his hasens he learning process, acceleraes firms echnology, and resuls in beer produciviy. One modern applicaion of agglomeraion o economic heory is he New Economic Geography (NEG), originally proposed by Krugman (1991). According o Ascani e al. (01), he NEG consiss of four imporan elemens. The firs is he increasing reurn o scale; his highlighs he spaial unevenness of economic aciviy. However, such agglomeraion should be carefully invesigaed since he NEG is modeled assuming an almos single region. The benefis of clusering ypically maer less if dispersion force from agglomeraion is dominaed (Brakman e al., 004). The second elemen comprises he economic erms defined in he funcions, including he number of varieies or firms, using he Dixi Sigliz monopolisic compeiion model (Dixi and Sigliz, 1977). The hird is ransporaion cos, defined as an iceberg-ype cos funcion (Samuelson, 1954); his plays a crucial role in he choice of locaion. The iceberg cos funcion in he NEG model implies ha he ransporaion cos increases exponenially wih disance, conradicing he evidence ha delivered prices end o be concave raher han convex wih disance (McCann, 005). The fourh elemen is he pecuniary exernaliies ha he NEG considers for indusry localizaion. Such agglomeraion exernaliies, as menioned earlier for he firs elemen, can be represened by he benefis of labor marke pooling, availabiliy of inermediaes, and echnological spillover effecs. However, how does he agglomeraion srucure influence produciviy in pracice? To undersand his mechanism from an empirical perspecive, pas sudies have caegorized agglomeraion in differen ways. For insance, from a ime scale perspecive, agglomeraion is caegorized ino saic and dynamic agglomeraion. McDonald and McMillen (007) explained ha saic agglomeraion indicaes a one-ime change in producion due o agglomeraion whereas dynamic agglomeraion means a coninuous effec of agglomeraion on produciviy over ime. From a variey-in-indusry viewpoin, agglomeraion may also be caegorized ino localizaion and urbanizaion agglomeraion. In localizaion agglomeraion, firms in he same indusry locaed ogeher gain from agglomeraion. From Marshall s economy of scale, firms benefi from supplier sharing or even echnology ransfer hrough localizaion. In urbanizaion agglomeraion, firms in general, for insance, in bigger ciies, improve heir produciviy as he oal marke expands hrough urbanizaion; his leads o larger labor pooling and cross-indusry aciviies, and furher o produciviy improvemen. Empirical sudies have repored he impacs of agglomeraion following hese caegories. For example, Henderson (003) found ha high-ech indusries benefi more from localizaion economies whereas machinery indusries do no. In conras, Gleaser e al. (199) claimed ha indusrial diversiy promoes ciy employmen growh raher han specializaion. Transporaion sudies such as Graham (007), Graham e al. (009), Melo e al. (01, 013) also examined he conribuion of ransporaion o produciviy, considering ransporaion as one of he facors for agglomeraion economies; hey showed ha improvemen in accessibiliy from ransporaion in erm of Effecive Densiy, could creae a beer agglomeraion environmen. However, mos of hese sudies invesigaed he firm- or naional-level effec of agglomeraion. Therefore, we analyze he regional-level effec of agglomeraion on economic produciviy raher han firm- or naional-level effec. This is mainly because many counries have recenly raised policy concerns abou he regional impacs of iner-regional ransporaion infrasrucure such as high-speed rail. This sudy examines hree ypes of agglomeraion hrough an empirical economeric analysis where he produciviy elasiciies of agglomeraion by indusry are esimaed using iner-regional Japanese daa. This paper is srucured as follows. The nex secion presens he mehodology used, including he formulaion of regional producion funcion and definiion of agglomeraion. Secion 3 presens empirical daa wih unconrolled relaionships beween agglomeraion and economic developmen. Secion 4 presens he resuls of economeric model esimaion of he impacs of agglomeraion on economic produciviy. Finally, Secion 5 summarizes our conclusions and furher issues. 3
. Mehodology.1. Producion Funcion This paper empirically analyzes he impac of agglomeraion on regional produciviy by esimaing he regional producion funcion. We assume a generalized Cobb Douglas funcion for he regional producion funcion as follows: k l GDPni A Kni Lni, (1) where GDP ni represens he GDP of zone i in indusry secor n; A represens echnology (oal facor produciviy or TFP); K ni and L ni represen respecively he capial and labor inpu of zone i in indusry secor n; and, k, and l represen he elasiciies peraining o echnology, capial, and labor, respecively. By using he naural log, we can re-wrie Eq. (1) as gdpni a kkni llni, () where he lowercase gdp ni, a, k ni, and l ni represen he logarihmic GDP, logarihmic echnology, logarihmic capial, and logarihmic labor, respecively. One issue o be addressed in economeric esimaion is he endogeneiy effec. This could arise wih reverse causaliy and omied variables. This sudy assumes ha agglomeraion affecs produciviy. On he oher hand, reverse causaion, which can be reasonably expeced when a region wih higher produciviy aracs more firms and workers, leads o furher agglomeraion. The mos popular echnique o deal wih he endogeneiy problem in regression analysis is he insrumenal variable (IV) approach; his echnique assumes ha agglomeraion can be explained by oher IV facors. Alhough we ried various IVs for our empirical analysis, including pas daa, as proposed by Arellano and Bond (1991), and he generalized mehod of momens (GMM) echnique, unforunaely, we could no find any appropriae IVs and GMM yielded unpromising resul. For more deails of oher model esimaion rials, see Appendix A. Anoher possible source of endogeneiy is omied variables. Following several rials and numerous errors in our esimaion, we finally could assume ha he echnology erm can be explained by agglomeraion; here, agglomeraion can be represened by effecive densiy, ED, and oher independen variables,. We define effecive densiy in he nex subsecion. We hen challenge he following semi-parameric approach, which is similar o Graham e al. s (009) mehod: gdpni A ED, kkni llni. (3) As for he TFP funcion A, capial and invesmen are he proxy variables, apar from effecive densiy, following he original work of Olley and Pakes (1996): kni vni kkni llni gdpni ED,, (4) where ni as a hird-order bivariae polynomial expansion of he Cobb Douglas funcion: v represens he invesmen of zone i in indusry secor n. In our regression process, k, gdpni ED kkni llni vvni kk ni v ni is specified kni vv vni kvknivni k v k v k 3 v 3 kkv ni ni kvv ni ni kkk ni vvv ni. (5) 4
.. Effecive Densiy This sudy assumes hree ypes of effecive densiies o represen agglomeraion. The firs follows he concep of urbanizaion agglomeraion; here, he benefis of agglomeraion, as described in Jacobs (1969), emerge from he differen secor s knowledge spillover supporing one anoher. Moreover, innovaion growh is believed o be simulaed by a variey of indusrializaion approaches since differen ideas and informaion can be synhesized hrough variey raher han specializaion. Gleaser e al. (199) showed ha he economic growh of ciies can be developed hrough urbanizaion agglomeraion; in sum, hey explained his by he cross-ferilizaion of ideas, implying ha urbanizaion can lead o more labor mobiliy. The effecive densiy used in his sudy applies a graviy-like model, as proposed by he Deparmen for Transpor (DfT) Wider Impac Guideline (Deparmen for Transpor, 014b) for incorporaing ransporaion ino agglomeraion. The effecive densiy of zone i is defined as he sum of he mass of employmen in anoher zone j and he ravel ime beween zone i and zone j. This formulaion depics agglomeraion in wo ways: he mass of employmen gives he amoun of aciviies generaed by a paricular zone j, and ravel ime represens he araciveness of zone j s aciviies from he viewpoin of zone i. The effecive densiy under urbanizaion agglomeraion can be formulaed as E j ED i, (6) j gij where ED i represens he effecive densiy of zone i a a ime, E j represens he oal employmen in zone j a ime, and g ij represens he ravel ime beween zone i and zone j a ime. In his case, he firs erm on he righ-hand side of Eqs. (4) and (5) saisfy ED ni ED i in he esimaion process. The second ype of effecive densiy follows he concep of localizaion agglomeraion. The concep of localized indusries was proposed by Marshall (190) and expanded ino a more sophisicaed formalizaion by Arrow (196) and Romer (1986); he accumulaion of knowledge spillover wihin he same indusry is now known as Marshall Arrow Romer exernaliies. The effecive densiy under localizaion agglomeraion can be formulaed as where Enj ED ni, (7) j gij ED ni represens he effecive densiy of zone i in indusry secor n and E nj represens he employmen of zone i in indusry secor n. Here, he firs erm on he righ-hand side of Eqs. (4) and (5) saisfy ED ni ED ni in he esimaion process. The hird ype of effecive densiy follows mixed agglomeraion, which includes urbanizaion and localizaion. Under Marshall s proposal, more ineracion beween indusries can lead o beer reurns for boh paries. However, localizaion considers he ineracion beween he same ype of indusries and ignores he ineracion beween differen ypes of indusries. On he conrary, urbanizaion considers he whole economy, ignoring he economic srucure. Zones wih differen indusries ypes and indusrial share can have differen effecs from agglomeraion as well. For a beer undersanding of he whole agglomeraion economy, we define he weighed effecive densiy under mixed agglomeraion by assuming a weigh parameer of nm for each pair of indusry as EDni j m nmemj gij, (8) where nm is he effecive densiy s weigh parameer o explain he degree of indusrial ineracion beween secor n and secor m. From his formulaion, we can explain agglomeraion a a poin beween localizaion and urbanizaion hrough he weigh nm, which roughly represens he produciviy of join aciviies and/or ineracions beween indusries n and m; weigh nm is formulaed modifying he co-agglomeraion index proposed by Ellison and Glaeser (1997) as 5
sni xi s mi xi i nm exp, (9) 1 xi i where s ni and s mi are he respecive shares of employmen in indusries n and m ou of he oal employmen in zone i, and x i is he mean share of employmen in zone i ou of he naional employmen across all indusries. Noe ha Ellison and Glaeser s co-agglomeraion index ignores he real spaial ineracion agglomeraion in erms of disance beween firms (Duranon and Overman, 005). Thus, he co-agglomeraion index in a spacious zone becomes he same as ha in a smaller zone if boh zones have he same number of firms, bu in realiy, he smaller zone can aain beer agglomeraion benefis from he shorer disance beween firms. Despie such mehodological disadvanages, our analysis uses his index for analyical simpliciy. In his case, we assume ha he firs erm on he righ-hand side of Eqs. (4) and (5) saisfy ED ni ED in he esimaion process, as in analysis of localizaion agglomeraion. 3. Daa ni We use he iner-regional ransporaion daa of Japan for our empirical analysis. Since iner-regional ransporaion connecs one region wih anoher, is impac on produciviy can be fel across regions raher han wihin a region. Thus, we obain daa a he prefecural level (firs-level adminisraive division in Japan, approximaely equivalen o NUTS 1 in he European Union) for our daase, alhough, in realiy, urbanizaion in he prefecural conex migh vary over prefecures. For insance, he buil-up areas in mega ciies such as Tokyo and Osaka could cover muliple prefecures whereas he buil-up areas in less urbanized prefecures migh cover only small owns in a single prefecure. Thus, agglomeraion in our daa may be regarded as macroscopic approximaion a he regional level. Our daase covers 11 indusrial secors (agriculure; mining; manufacuring; consrucion; elecriciy/gas/waer; reail; finance/insurance; real esae; ransporaion/communicaion; service; and governmen service) in 47 prefecures for six years a five-year inervals: 1981, 1986, 1991, 1996, 001, and 006. Socio-demographic and socioeconomic daa, such as prefecural populaion, GDP, employees, wage, capial and invesmen sock by indusry, ec., were derived from he Saisic Bureau and Cabine office of Japan. Noe ha all economic daa were adjused o he year 000. As for ransporaion daa, he ravel ime beween each prefecure pair was esimaed as he shores ravel ime incorporaing he six ravel modes of high-speed rail, convenional rail, air, ferry, iner-ciy bus, and privae car. We used he Naional Inegraed Transpor Analysis Sysem (NITAS) sofware developed by he Minisry of Land, Infrasrucure, Transpor, and Tourism (MLIT) of Japan o search for he shores pah. Noe ha he ransporaion nework varies over six imes since he ransporaion infrasrucure had been developed gradually over ime. Figure 1 illusraes he relaionship beween hree ypes of prefecural effecive densiy and prefecural GDP. For he localizaion and mixed agglomeraion cases, we presen he prefecural GDP for he manufacuring indusry as example. Alhough he laer years indicae less producion, a comparison of he daa for he same ime-period show he prefecures wih more effecive densiy o have higher GDP, implying agglomeraion leads o higher overall producion. This may be raher reasonable because effecive densiy includes he number of workers and hence influences posiively he prefecural GDP. Nex, Figure illusraes he relaionship beween prefecural effecive densiy and prefecural GDP per worker. The figure shows he prefecures wih higher effecive densiy o have higher GDP per worker. This could imply ha more agglomeraion leads o higher produciviy, concurring wih Tabuchi and Yoshida (000), who suggesed an expeced 10% wage increase when he ciy populaion of Japan doubles. These unconrolled for analyses clearly sugges a relaionship beween agglomeraion and produciviy. However, o find he reurn o produciviy ha can be expeced from agglomeraion, we need a conrolled analysis. 1 NUTS or Nomenclaure of Terriorial Unis for Saisics, a subdivision code uses in EU. NUTS level conains around 800,000 3,000,000 of populaion. Populaion in Prefecure Level ( 都道府県 ) of Japan covers a range of 600,000 1,000,000. 6
Manufacuring GDP by Prefecure (Trillion Yen) 14 1 10 8 6 4 0 0 0000 40000 60000 80000 100000 10000 140000 Localizaion Effecive Densiy in Manufacuring (Worker/Min.) Prefecure GDP (Trillion Yen) 100 80 60 40 0 0 0 100000 00000 300000 400000 500000 600000 700000 Urbanizaion Effecive Densiy (Worker/Min.) Manufacuring GDP by Prefecure (Trillion Yen) 14 1 10 1981 8 1986 1991 6 1996 4 001 006 0 0 1000000 000000 3000000 4000000 5000000 6000000 Mixed Effecive Densiy in Manufacuring (Worker/Min.) Figure 1 Localizaion, Urbanizaion, and Mixed Agglomeraions versus Prefecural GDP. Manufacuring GDP per worker by Prefecure (Million Yen/Worker) 0 18 16 14 1 10 8 6 4 0 0 0000 40000 60000 80000 100000 10000 140000 Localizaion Effecive Densiy in Manufacuring (Worker/Min.) Prefecure GDP/Worker (Million Yen/Worker) 14 1 10 8 6 4 0 0 100000 00000 300000 400000 500000 600000 700000 Urbanizaion Effecive Densiy (Worker/Min.) Manufacuring GDP by Prefecure (Trillion Yen) 0 18 16 14 1981 1 1986 10 1991 8 1996 6 001 4 006 0 0 1000000 000000 3000000 4000000 5000000 6000000 Mixed Effecive Densiy in Manufacuring (Worker/Min.) Figure Localizaion, Urbanizaion, and Mixed Agglomeraions versus Prefecural GDP per worker 7
4. Resuls 4.1. Esimaion Resuls We esimae hree models in regression processes, he prefecural fixed effec model ( prefecure conrolled ), he ime-period fixed effec model ( ime conrolled ), and he prefecural and ime-period fixed effec model ( wo-wayconrolled ), for each ype of producion funcion. Tables 1,, and 3 give he esimaion resuls, highlighing he elasiciies of effecive densiy for each model. See Appendix B for he enire resuls. Table 1 summarizes he esimaion resuls for he 3 regression models using urbanizaion agglomeraion in 11 indusries, assuming Eq. (5) for effecive densiy. For all indusries, from he degree of freedom, model finess is he highes in he ime-conrolled model, followed by he prefecure-conrolled model and he wo-way-conrolled model. Firs, he prefecure-conrolled model shows ha effecive densiy has significanly posiive impacs on mining and finance/insurance bu significanly negaive impacs on real esae and governmen service indusries. Nex, he imeconrolled model shows ha effecive densiy has a significanly posiive impac on real esae by a significanly negaive impac on agriculure indusry. Finally, he wo-way-conrolled model shows ha effecive densiy has no impac on any indusry. Table summarizes he esimaion resuls of he hree regression models using mixed agglomeraion in eleven indusries, assuming Eq. (8) for effecive densiy. Models assuming mixed effecive densiy end o perform beer han assuming urbanizaion agglomeraion, alhough he resuls are generally he same as for earlier models. Firs, he prefecure-conrolled model shows ha effecive densiy has significanly posiive impacs on mining, finance/insurance, and ransporaion/communicaion bu significanly negaive impacs on he service indusry. Nex, he ime-conrolled model shows ha effecive densiy has a significanly posiive impac on real esae bu negaive impacs on he agriculure indusry. Finally, he wo-way-conrolled model shows ha effecive densiy has a significanly posiive impac on governmen service. Table 3 summarizes he esimaion resuls of he hree regression models using localizaion agglomeraion in eleven indusries, assuming Eq. (7) for effecive densiy. Firs, he prefecure-conrolled model shows ha effecive densiy has significanly posiive impacs on consrucion, reailing, finance/insurance, and ransporaion/communicaion indusries bu significanly negaive impacs on manufacuring, elecriciy/gas/waer, and service indusries. Nex, he ime-conrolled model shows ha effecive densiy has a significanly posiive impac on real esae bu a significanly negaive impac on he agriculure indusry. Finally, he wo-way-conrolled model shows ha effecive densiy has a significanly posiive impac on mining indusry. Our major findings based on he above esimaion resuls can be summarized as follows: The prefecure-conrolled model shows ha (1) boh urbanizaion and localizaion agglomeraions have a posiive influence on regional produciviy in he finance/insurance indusry; () urbanizaion agglomeraion ends o have a posiive influence on regional produciviy in he mining indusry; (3) localizaion agglomeraion ends o have a posiive influence on regional produciviy in he ransporaion/communicaion indusry; and (4) localizaion agglomeraion ends o have a negaive influence on regional produciviy in he services indusry. The ime-conrolled model shows ha (5) boh urbanizaion and localizaion agglomeraions have a posiive influence on regional produciviy in he real esae indusry; and (6) boh urbanizaion and localizaion agglomeraions have a negaive influence on regional produciviy in he agriculure indusry. Noe ha he above findings assume ha a significan resul from he models of boh urbanizaion and mixed agglomeraions imply influence from urbanizaion agglomeraion, whereas a significan resul from he models of boh localizaion and mixed agglomeraions imply influence from localizaion agglomeraion. Noe also ha he prefecure-conrolled model excludes he impacs of he unique prefecure-relaed facor by inroducing consans o each prefecure whereas he ime-conrolled model excludes he impacs of he unique ime-relaed facor by inroducing consans o each ime. Findings (1) o (4) are based on observaions of he prefecure-conrolled model only, meaning ha he resuls could hold rue across prefecures bu could be affeced by he ime facor. Findings (5) and (6) are based on observaions of he ime-conrolled model only, meaning ha he resuls could hold rue across ime bu could be affeced by prefecural facor. 8
4.. Discussion From he resuls, he finess of he esimaed models assuming localizaion agglomeraion end o be higher han ha for he oher wo models in any indusry. The number of indusries wih significan esimaes for agglomeraion is also larges in he localizaion models. This could imply ha localizaion agglomeraion has a higher influence on economic producion han urbanizaion agglomeraion. However, he resuls also show ha agglomeraion has differen effecs for each indusry. Firs, he posiive impacs of boh urbanizaion and localizaion agglomeraion on regional produciviy in he finance/insurance and real esae indusries, or he so-called FIRE indusry, may be explained reasonably using Marshall s heory. Since he FIRE indusry should have cusomers from many oher indusries, a higher densiy of poenial cusomers from various indusries could give more business opporuniies o hem; his may be one of he sources of exernal benefi from urbanizaion agglomeraion. As he FIRE indusry paricularly needs he laes informaion abou local/regional/global markes, he social nework of workers in he same indusry could effecively conribue o sharing knowledge hrough meeings. Because communicaion opporuniies such as seminars and informal meeings could arac businesspeople from across he regions, a higher densiy of colleagues in he FIRE indusry could provide more knowledge spillover hrough communicaion; his is one of he sources of exernal benefi from localizaion agglomeraion. Localizaion agglomeraion also affecs he labor pool as well as procuremen of high-sandard service, because he FIRE indusry requires skillful labor and efficien business environmen for aaining higher produciviy. A significan impac in he finance/insurance indusry could be found only wih he prefecure-conrolled model, probably because is impac considerably varies across prefecures. Significan impac could be found in he real esae indusry wih he ime-conrolled model, probably because he real esae marke in Japan was influenced by condiions in he naional economic marke raher han by each prefecure s unique condiion, alhough he significance in he prefecure-conrolled model is relaively srong as well. Noe ha he esimaed elasiciies in he finance/insurance indusry wih respec o urbanizaion, mixed, and localizaion agglomeraions are 0.935, 1.64, and 0.750, respecively, and hose in real esae indusry are 0.91, 0.94, and 0.44, respecively. This could sugges ha urbanizaion agglomeraion may have a greaer influence on produciviy han localizaion agglomeraion in hose indusries. Second, he posiive impac of localizaion agglomeraion on regional produciviy in mining may be explained from he naural resource as well as marke perspecive. Mining producs usually come direcly from naural resources, which are ypically locaed in limied areas based on geographical condiions of resource availabiliy. Since he uni freigh ransporaion cos of mining producs is expeced o be higher han ha of oher goods because of he naure of large volume ranspor, mining indusries end o locae near he naural resource sies. This is he case in Japan oo, where he areas rich in naural resources arac more mining indusries. Thus, closeness o naural resource sies iself generaes higher produciviy, leading o localizaion agglomeraion in he mining indusry. On he oher hand, localizaion agglomeraion may also generae exernal effecs, such as benefis from cos savings in he join procuremen of machines and skilled labors and from echnology ransfer among mining firms. Knowledge sharing on local condiions may be criical for he mining indusry because heir business depends significanly on he unique local geographical environmen. A significan impac of agglomeraion on mining could be found only wih he prefecure-conrolled model because is impac varies considerably across prefecures because of he geographically uneven availabiliy of naural resources. However, reverse causaion from he effec of naural advanages could lead o beer produciviy, and agglomeraion could merely be he resul of ha produciviy. As Ellison and Glaeser (1997) show, naural resources can be reaed as naural advanages for he mining secor. As a place wih abundan naural resources could provide beer economies of scale, producers end o be araced, resuling in agglomeraion of he mining indusry. Third, he posiive impac of localizaion agglomeraion on regional produciviy in ransporaion/communicaion may reflec regional marke characerisics. For insance, when ransporaion firms are locaed closely, rucks/vans or drivers can be easily shared among hem, hus reducing heir poenial business risk due o demand flucuaion in he ransporaion marke. The nework economy may also work in ransporaion/communicaion businesses ha paricularly use physical nework. For example, muliple public ransi operaors working closely ogeher can form a wider ransporaion nework covering vas areas and hus enhance accessibiliy and he mobiliy of passengers; his could improve he produciviy of public ransi operaors from he complemenariy of services. A significan impac of agglomeraion was found in he ransporaion/communicaion indusry only wih he prefecure-conrolled model because is impac considerably varies over prefecures owing o he geographically uneven availabiliy of naural resources. Fourh, localizaion agglomeraion negaively influences regional produciviy in he service indusry. Generally, negaive produciviy elasiciies of agglomeraion are found when he cenrifugal forces semming from agglomeraion are sronger han he cenripeal forces (Fujia e al., 1999). The cenrifugal force or diseconomies from 9
agglomeraion may arise from higher land ren, an increase in living expenses, or even more congesion from a denser populaion. One possible reason for negaive elasiciy in he service indusry is ha agglomeraion of he same service firms causes serious marke compeiion among hem, which could lose he addiional benefi of he imperfec compeiive marke. Agglomeraion could even lead o overcompeiion, generaing negaive exernal effecs such as a weaker posiion in business conracs wih heir cliens or cusomers, while less agglomeraed firms could enjoy higher marke power. Negaive impac o some indusry can be suppored by Combes e al. (01), where he firm selecion process has no impac on spaial produciviy difference. Fifh, boh urbanizaion and localizaion agglomeraions have a negaive influence on regional produciviy in he agriculure indusry. One of he possible explanaions is ha he economy of geographical scale works well in agriculural business because i ypically requires larger land for beer producion. Larger area of land decreases he average cos of producion, meaning beer produciviy, and leads o less agglomeraion. Anoher possible reason paricularly for he poor impac of localizaion agglomeraion is he negaive exernal effec of agglomeraion. For example, densely agglomeraed agriculural businesses can consume excessive naural resources such as waer, wood, and fish and hus reduce he performance of agriculural producion. Finally, indusries oher han FIRE, ransporaion/communicaion, service, and agriculure may no have noable impacs from agglomeraion. Paricularly, he poor significance of agglomeraion in elecriciy/gas/waer, reail, and governmen service indusries could be explained by he characerisics of such services and/or goods. As hese are necessary goods/services for people s daily life, he indusries producing such commodiies are essenially required o be disribued evenly. Governmen service is a ypical case, and he reail and elecriciy/gas/waer indusries also have o run heir businesses even if heir profi is near zero. More posiively, hese indusries hemselves disribue evenly based on he disribuion of populaion, and so regional agglomeraion may make less sense in hese indusries. The firm selecion approach explains ha he beer produciviy from agglomeraion is due o he inensive compeiion in larger markes. Only he bes firms can survive compeiion, resuling in beer overall produciviy in a large marke compared o a smaller marke. 10
Table 1 Esimaed elasiciies of produciviy wih respec o effecive densiy based on urbanizaion agglomeraion (N=8) Semi-parameer (Eq. (5)) - Prefecure conrol Agriculure Mining Manufacuring Consrucion Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.090 0.155 0.58 0.561 1.67 0.178 7.16 0.000 *** -0.03 0.155-0.05 0.838-0.011 0.184-0.058 0.954 ln(l) 0.195 0.041 4.740 0.000 *** 0.86 0.033 8.779 0.000 *** 0.411 0.13 3.337 0.001 *** 0.574 0.050 11.447 0.000 *** Adj. R 0.49 0.510 0.77 0.64 Elec, Gas & Waer Reail Finance & Insur Real Esae Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.175 0.148 1.180 0.39-0.055 0.113-0.488 0.66 0.935 0.198 4.719 0.000 *** -0.417 0.138-3.06 0.003 ** ln(l) -0.13 0.076-1.613 0.108 0.69 0.05 10.848 0.000 *** 0.548 0.063 8.76 0.000 *** 0.636 0.075 8.460 0.000 *** Adj. R 0.706 0.745 0.699 0.730 Transpor & Comm Service Gov. Service Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) -0.051 0.155-0.330 0.74 0.066 0.083 0.795 0.47-0.195 0.054-3.641 0.000 *** ln(l) 0.0 0.064 3.448 0.001 *** 0.080 0.036.7 0.07 * 0.549 0.056 9.888 0.000 *** Adj. R 0.71 0.771 0.747 Semi-parameer (Eq. (5)) - Time conrol Agriculure Mining Manufacuring Consrucion Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) -0.96 0.059-5.00 0.000 *** 0.011 0.053 0.0 0.840 0.095 0.047.001 0.046 * -0.101 0.045 -.7 0.04 * ln(l) 0.331 0.038 8.783 0.000 *** 0.053 0.038 1.389 0.166 0.574 0.07 1.09 0.000 *** 0.53 0.060 4.49 0.000 *** Adj. R 0.777 0.885 0.98 0.915 Elec, Gas & Waer Reail Finance & Insur Real Esae Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.00 0.030 0.065 0.949 0.033 0.037 0.904 0.367 0.06 0.041 1.534 0.16 0.9 0.07 4.086 0.000 *** ln(l) 0.111 0.07 4.104 0.000 *** 0.1 0.043 4.948 0.000 *** 0.194 0.058 3.363 0.001 *** 0.090 0.181 0.496 0.60 Adj. R 0.97 0.930 0.96 0.893 Transpor & Comm Service Gov. Service Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.016 0.031 0.58 0.598 0.01 0.019 1.118 0.65 0.001 0.07 0.039 0.969 ln(l) 0.009 0.046 0.191 0.849 0.159 0.03 4.94 0.000 *** 0.963 0.040 4.367 0.000 *** Adj. R 0.930 0.936 0.99 Semi-parameer (Eq. (5)) - Two-way conrol Agriculure Mining Manufacuring Consrucion Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) -0.199 0.30-0.863 0.389 0.460 0.86 1.609 0.109 0.097 0.01 0.483 0.630-0.98 0.191-1.555 0.11 ln(l) 0.06 0.037 5.60 0.000 *** 0.158 0.046 3.457 0.001 *** 0.376 0.11 3.346 0.001 *** 0.17 0.074.341 0.00 * Adj. R 0.185 0.611 0.330 0.95 Elec, Gas & Waer Reail Finance & Insur Real Esae Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.06 0.149 0.418 0.676 0.057 0.115 0.501 0.617 0.064 0.116 0.55 0.58 0.164 0.188 0.873 0.384 ln(l) 0.073 0.047 1.549 0.13 0.064 0.038 1.671 0.096. 0.06 0.044 1.413 0.159 0.585 0.119 4.910 0.000 *** Adj. R 0.617 0.557 0.537 0.84 Transpor & Comm Service Gov. Service Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0. 0.118 1.870 0.063. 0.005 0.076 0.060 0.95 0.095 0.068 1.411 0.160 ln(l) 0.098 0.043.311 0.0 * 0.060 0.09.10 0.037 * 0.710 0.063 11.316 0.000 *** Adj. R 0.490 0.630 0.37 Noe: ***p<0.001; **p<0.01, and *<0.05. 11
Table Esimaed elasiciies of produciviy wih respec o effecive densiy based on mixed agglomeraion (N=8) Semi-parameer (Eq. (5)) - Prefecure conrol Agriculure Mining Manufacuring Consrucion Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.18 0.134 1.631 0.104 1.076 0.164 6.573 0.000 *** -0.037 0.057-0.654 0.514 0.447 0.135 3.315 0.001 ** ln(l) 0.197 0.041 4.85 0.000 *** 0.44 0.03 7.600 0.000 *** 0.40 0.108 3.740 0.000 *** 0.444 0.063 7.097 0.000 *** Adj. R 0.433 0.50 0.78 0.649 Elec, Gas & Waer Reail Finance & Insur Real Esae Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) -0.038 0.137-0.77 0.78 0.17 0.039 3.6 0.001 ** 1.14 0.150 8.100 0.000 *** -0.36 0.119-3.09 0.003 ** ln(l) -0.16 0.077-1.648 0.101 0.169 0.036 4.654 0.000 *** 0.355 0.065 5.489 0.000 *** 0.67 0.075 8.313 0.000 *** Adj. R 0.706 0.747 0.713 0.730 Transpor & Comm Service Gov. Service Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.664 0.13 5.014 0.000 *** -0.04 0.008-5.537 0.000 *** -0.05 0.075-0.699 0.486 ln(l) 0.198 0.061 3.67 0.001 ** 0.050 0.033 1.5 0.19 0.464 0.053 8.86 0.000 *** Adj. R 0.79 0.773 0.744 Semi-parameer (Eq. (5)) - Time conrol Agriculure Mining Manufacuring Consrucion Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) -0.97 0.059-5.005 0.000 *** 0.011 0.053 0.03 0.839 0.100 0.046.155 0.03 * -0.10 0.044 -.30 0.0 * ln(l) 0.331 0.038 8.790 0.000 *** 0.053 0.038 1.389 0.166 0.57 0.07 0.99 0.000 *** 0.53 0.060 4.45 0.000 *** Adj. R 0.777 0.885 0.98 0.915 Elec, Gas & Waer Reail Finance & Insur Real Esae Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.00 0.030 0.060 0.95 0.031 0.036 0.857 0.39 0.06 0.041 1.534 0.16 0.94 0.07 4.101 0.000 *** ln(l) 0.111 0.07 4.104 0.000 *** 0.13 0.043 4.991 0.000 *** 0.194 0.058 3.364 0.001 *** 0.089 0.181 0.493 0.6 Adj. R 0.97 0.930 0.96 0.893 Transpor & Comm Service Gov. Service Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.016 0.031 0.514 0.608 0.01 0.019 1.130 0.60 0.000 0.06 0.013 0.990 ln(l) 0.009 0.046 0.189 0.850 0.159 0.03 4.99 0.000 *** 0.963 0.040 4.369 0.000 *** Adj. R 0.930 0.936 0.99 Semi-parameer (Eq. (5)) - Two-way conrol Agriculure Mining Manufacuring Consrucion Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.000 0.000 -.307 0.0 * 0.000 0.000 1.380 0.169 0.000 0.000 0.581 0.56 0.000 0.000 0.590 0.556 ln(l) 0.181 0.094 1.914 0.057. 0.040 0.09 0.433 0.665 0.151 0.130 1.166 0.45 0.173 0.11 1.48 0.155 Adj. R 0.005 0.485 0.53 0.15 Elec, Gas & Waer Reail Finance & Insur Real Esae Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.000 0.000-0.685 0.494 0.000 0.000 0.48 0.631 0.000 0.000 1.598 0.11 0.000 0.000.56 0.01 * ln(l) 0.6 0.380 0.689 0.49-0.167 0.490-0.341 0.733-0.09 0.075-0.393 0.695 0.37 0.303 1.8 0.1 Adj. R 0.114 0.017 0.37 0.070 Transpor & Comm Service Gov. Service Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.000 0.000 1.007 0.315 0.000 0.000 1.005 0.317 0.000 0.000 3.404 0.001 *** ln(l) -0.09 0.300-0.698 0.486 0.09 0.049 0.585 0.560 0.311 0.14.51 0.013 * Adj. R 0.163 0.46 0.191 Noe: ***p<0.001; **p<0.01, and *<0.05. 1
Table 3 Esimaed elasiciies of produciviies wih respec o effecive densiy based on localizaion agglomeraion (N=8) Semi-parameer (Eq. (5)) - Prefecure conrol Agriculure Mining Manufacuring Consrucion Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) -0.11 0.1-0.570 0.569 0.060 0.091 0.657 0.51-0.74 0.067-4.078 0.000 *** 0.53 0.107 4.986 0.000 *** ln(l) 0.09 0.050 4.07 0.000 *** 0.180 0.078.90 0.03 * 0.570 0.11 5.091 0.000 *** 0.31 0.069 4.60 0.000 *** Adj. R 0.49 0.447 0.73 0.657 Elec, Gas & Waer Reail Finance & Insur Real Esae Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) -1.34 0.151-8.765 0.000 *** 0.03 0.058 3.50 0.001 *** 0.750 0.106 7.064 0.000 *** 0.153 0.11 1.59 0.09 ln(l) 0.117 0.07 1.636 0.103 0.131 0.043 3.01 0.003 ** 0.56 0.08 3.110 0.00 ** 0.647 0.076 8.470 0.000 *** Adj. R 0.78 0.747 0.709 0.78 Transpor & Comm Service Gov. Service Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.50 0.055 9.511 0.000 *** -0.478 0.057-8.366 0.000 *** -0.07 0.073 -.846 0.005 ** ln(l) 0.141 0.055.583 0.010 * 0.15 0.031 4.936 0.000 *** 0.563 0.063 8.985 0.000 *** Adj. R 0.74 0.776 0.745 Semi-parameer (Eq. (5)) - Time conrol Agriculure Mining Manufacuring Consrucion Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) -0.470 0.108-4.357 0.000 *** 0.14 0.090 1.571 0.117 0.109 0.044.453 0.015 * -0.106 0.053-1.991 0.048 * ln(l) 0.401 0.036 10.99 0.000 *** 0.045 0.038 1.196 0.33 0.567 0.07 0.71 0.000 *** 0.6 0.059 4.41 0.000 *** Adj. R 0.773 0.886 0.98 0.915 Elec, Gas & Waer Reail Finance & Insur Real Esae Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.008 0.03 0.55 0.799 0.04 0.036 0.678 0.499 0.061 0.040 1.58 0.18 0.44 0.066 3.685 0.000 *** ln(l) 0.111 0.07 4.070 0.000 *** 0.17 0.043 5.085 0.000 *** 0.190 0.057 3.318 0.001 ** 0.083 0.18 0.457 0.648 Adj. R 0.97 0.930 0.96 0.89 Transpor & Comm Service Gov. Service Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.013 0.031 0.434 0.665 0.018 0.00 0.939 0.349-0.007 0.09-0.35 0.814 ln(l) 0.007 0.046 0.161 0.87 0.161 0.03 5.008 0.000 *** 0.964 0.040 4.316 <e-16 *** Adj. R 0.930 0.936 0.99 Semi-parameer (Eq. (5)) - Two-way conrol Agriculure Mining Manufacuring Consrucion Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) -0.575 0.53 -.76 0.04 * 1.073 0.09 5.1 0.000 *** 0.74 0.195 1.406 0.161-0.54 0.07-1.3 0.19 ln(l) 0.67 0.045 5.860 0.000 *** 0.014 0.05 0.69 0.788 0.364 0.11 3.39 0.001 ** 0.177 0.074.398 0.017 * Adj. R 0.197 0.67 0.333 0.93 Elec, Gas & Waer Reail Finance & Insur Real Esae Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.155 0.155 1.004 0.317 0.09 0.111 0.6 0.794 0.060 0.104 0.576 0.565 0.9 0.187 1.566 0.119 ln(l) 0.057 0.050 1.141 0.55 0.066 0.038 1.70 0.087. 0.060 0.044 1.37 0.171 0.570 0.118 4.83 0.000 *** Adj. R 0.618 0.557 0.537 0.87 Transpor & Comm Service Gov. Service Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) Esimae Sd.Error Pr(> ) ln(ed) 0.1 0.097.68 0.04 * 0.06 0.075 0.350 0.77 0.054 0.074 0.737 0.46 ln(l) 0.071 0.045 1.576 0.117 0.060 0.09.109 0.036 * 0.711 0.063 11. 0.000 *** Adj. R 0.49 0.630 0.370 Noe: ***p<0.001; **p<0.01, and *<0.05. 13
5. Conclusion This sudy provided empirical evidence of he impacs of agglomeraion on regional developmen using Japanese hisorical daa. Our resuls showed ha on average, he indirec benefi semming from produciviy improvemen hrough localizaion agglomeraion ends o be more significan han ha hrough urbanizaion agglomeraion alhough heir robusness indicaes ha each indusry uilizes agglomeraion in differen ways. From our resuls for indusries, mining enjoys significan benefi from urbanizaion raher han localizaion, ransporaion/communicaion enjoys significan benefi from localizaion raher han urbanizaion, and FIRE could benefi from boh ypes of agglomeraion economies. Negaive elasiciies were found for agriculure and service indusries, bu his could be due parly o he indusries characerisics. This sudy also parly discussed he facors ha could lead o agglomeraion. As shown in our discussions on he mining indusry, he geographical disribuion of naural resources is one of he facors. Alhough we ried o analyze he poenial reverse causaliy and explain he agglomeraion wih oher facors, our aemps failed of our limied daase. This could be parly because of he unique policy implemened earlier by he naional governmen in he 1980s o 1990s in Japan. Alhough in he early sages afer World War II, a series of expressways and high-speed railways had been successfully inroduced o expand he ransporaion nework and mee he challenges of he rapid economic growh, he governmen gradually shifed is policy goal from naional economic developmen o regional economic developmen under he concep of he regionally balanced naional developmen policy in he 1980s o 1990s. During ha period in Japan, he invesmen of iner-regional ransporaion infrasrucure or developmen of regional indusries may have been deermined hrough poliical debaes raher han on a consisen decision-making process, hus making i difficul for us o inerpre he mechanism of regional agglomeraion in Japan. Noe ha he formal cos-benefi analysis guideline for ransporaion invesmen was inroduced in Japan around 000. Alhough his sudy conribued o validae he assumpion ha improved regional accessibiliy promoed economic developmen hrough agglomeraion, several furher issues remain o be addressed. Firs, from a echnical perspecive, one of he issues is he raionale for using effecive densiy o explain agglomeraion. Kanemoo (013) menioned ha he concep of effecive densiy migh no be jusified in some cases. For example, he effecive densiy in Eq. (6) follows he urbanizaion agglomeraion neglecing indusrial srucure. Thus, a problem could arise, for example, when a zone wih 90% employmen in indusry n and 10% employmen in indusry m has he same effecive densiy as anoher zone wih 10% employmen in indusry n and 90% in indusry m, alhough clearly he produciviy beween hem should be differen. This is he main reason we inroduce he weighed effecive densiy in our analysis, alhough he resul could imply ha applying he Ellison and Glaeser co-agglomeraion index is no promising, a leas wih our specificaion and daase. Furher examinaion would be required for he definiion of agglomeraion. Ye, our resul could give some suggesion o ransporaion planner regarding agglomeraion o a cerain exend. Relaionship beween ransporaion invesmen and economy, hrough agglomeraion, could be posiive, negaive, or no relaed, depending on he disribuion of indusrial secor Addiionally, our resuls could also sugges ha he exernaliies o producion may no be explained by agglomeraion only. In his analysis, because of our small sample size (N=8), our daa correlaion, which is always one of main concerns, resrics us from inroducing more independen variables in he model esimaion. We can have a more sophisicaed analysis by using firm-level daa raher han macroscopic daa, which would enable a more precise esimaion. However, such firm-level daa can ypically be obained for only a single ciy. I would no be reasonable o consider he agglomeraion impac in a single ciy since he benefi of agglomeraion in such a ciy could be he resul of a loss in oher ciies, paricularly in he conex of iner-regional ransporaion invesmen. However, as criicized in Duranon and Overman (005), our esimaion used daa based on adminisraive division, hus ignoring he acual spaial ineracion beween firms. Such spaial ineracion could play an imporan role in agglomeraion and macroscopic analysis because he spaial consideraion of firms can give more definie explanaions for agglomeraion economies. 14