The New Economic Geography of Prefecture Cities in China:

Transcription

1 The New Economc Geography of Prefecture Ctes n Chna: The Relevance of Market Access and Labor Moblty for Agglomeraton Maarten Bosker, Steven Brakman, Harry Garretsen & Marc Schramm 1 Ths Verson: Aprl 29 th 2010 Abstract Based on a data set of 264 Chnese Prefecture ctes for the perod , we use a general new economc geography (NEG) model to determne the long-run equlbrum allocaton of economc actvty across Chna. The model s used to analyse the role of labor moblty and the assocated Hukou system that regulates mgraton n Chna. Our fndngs can be summarzed as follows. Frst, and n lne wth related studes, we fnd that (nternal) market access s an mportant determnant of Chnese cty-wages, our estmatons show that ctes wth hgher market access also have hgher wages. Second, and usng these estmaton results as an nput, the long-run equlbrum smulatons wth the NEG model ndcate that ncreased labor moblty would lead to more pronounced core-perphery outcomes for Chna. The resultng core-perphery patterns that go along wth ncreased labor moblty are such that the (new) agglomeratons are not necessarly along the coastal regons, but they are typcally located n the populous heartland of Chna. Internatonal market access seems to be less relevant. Our analyss ndcates that the economc geography of Chna that would result from ncreased labor moblty s frst and foremost determned by wthn Chna dfferences n market access, populaton and (arable) land. In a qualtatve sense, the NEG model predcts that ncreased labor moblty would prmarly beneft centrally located, non-coastal Prefecture ctes n the large and populous Honan and Anhwe provnces. 1 Bosker, Brakman and Garretsen: Faculty of Economcs and Busness, Unversty of Gronngen, The Netherlands; Schramm: Utrecht Unversty School of Economcs, Utrecht Unversty, The Netherlands. Please send all correspondence to:.h.garretsen@rug.nl We thank semnar partcpants at the Spatal Economcs Research Centre at the London School of Economcs; the Dept. of Land Economy at the Unv. of Cambrdge; the 2009 ETSG Conference n Rome; Utrecht Unversty; the 2009 NARSC conference n San Francsco, the EBRD n London as well as Vernon Henderson, Bart van Ark, Bernard Fngleton, Laura Herng, Ralph de Haas, Ron Martn, Ana Moreno Monroy, Charles van Marrewk, Henry Overman, Perre Pcard, Sandra Poncet, Mark Roberts and Tuo Sh for ther comments and suggestons. 1

2 1 Introducton Accordng to the World Development Report 2009 (World Bank, 2008), economc development s to a large extent drven by economc geography. Economc geography refers not only to physcal characterstcs of the landscape, but also and most mportantly to man-made economc geography n the sense of a regon s access to upstream and downstream markets or the degree of urbanzaton. The analytcal underpnnngs for ths (re)dscovery of the relevance of economc geography come n no small part from the new economc geography lterature ntated by Krugman (1991) n whch the balance between agglomeratng and spreadng forces determne the locaton of economc actvty. The relevance of ths approach has been shown for the cases of the US, and Europe, and only recently attenton has shfted towards developng countres lke Chna. Chna s one of the fastest growng economes n the world and t s also home to a huge (potental) nternal market and a large number of very large and fast-growng ctes. Chna s a textbook case to analyze how agglomeratng and spreadng forces can shape the economc landscape of Chna, and how economc geography can foster economc development. Recent studes (see Au and Henderson, 2006a,b), however, argue that, due to the severe restrcton on labor moblty mposed by the Hukou system (see Chan, 2008 and 2009), Chnese ctes are stll underszed and, as a result, Chna does not reap the full benefts of agglomeraton. A smple llustraton of ths fact s that the Gn coeffcent for the Chnese cty sze dstrbuton was 0.43 n 2000, as compared to a world-average Gn coeffcent of 0.564, whch ndcates a more evenly szed cty sze dstrbuton n Chna compared to the rest of the world (Futa et al, 2004, p. 2955). In ths paper we consder the mpact of the Hukou system on the spatal dstrbuton of economc actvty wthn Chna. In partcular, we use a New Economc Geography (NEG) model (Puga, 1999) to analyse the agglomeraton-labor moblty nexus for the case of Chna. Inspred by Krugman (2009,2010), we thnk that NEG models are partcularly suted to analyse the case of rapdly ndustralzng economes lke Chna. 2 In the NEG model of Puga (1999) the balance between agglomeraton and spreadng forces and hence the equlbrum spatal allocaton depends explctly on the degree of nterregonal labor moblty, makng t a natural startng pont to assess the effect of loosenng the restrcton posed by the Hukou system on Chna s nternal economc geography. Based on a sample of 264 Chnese Prefecture ctes, we frst estmate the equlbrum wage equaton that s central n NEG 2 Chnese economc geography s hghly remnscent of the economc geography of the advanced natons crca 1900 [that formed the backdrop for the Krugman (1991) NEG model]- and t fts gratfyngly well nto the new economc geography framework (Krugman, 2010, p. 14). 2

3 models. The estmated wage equaton not only emprcally establshes the lnk between a cty s wages and ts market access (see also Herng and Poncet, 2006); t also provdes two key model parameters that together make up for the so called free-ness of trade whch s central n the subsequent smulaton analyss of the relatonshp between agglomeraton and labor moblty for the 264 Chnese ctes n our sample. Our paper crucally dffers from related recent NEG studes for Chna (e.g. Ma, 2006, Ln, 2003, Herng and Poncet, 2006, 2007, De Sousa and Poncet, 2007, Amt and Javorck, 2008) to the extent that we do not take the spatal allocaton of labor (and frms) as gven. Instead, we wll make use of the complete NEG model and not ust the equlbrum wage equaton. Ths enables us to analyse the relevance of labor moblty for the equlbrum agglomeraton for Chnese ctes. Ths s the frst contrbuton of the paper to the lterature. A second contrbuton s that we go beyond the smple mgraton dynamcs that underle NEG models and look also at more ntrcate mgraton dynamcs. Our analyss shows that the economc geography of Chna that would result from ncreased labor moblty s frst and foremost determned by wthn Chna dfferences n market access, populaton and (arable) land. In a qualtatve sense, the NEG model predcts that ncreased labor moblty would prmarly beneft centrally located, non-coastal Prefecture ctes lke Zhoukou or Luan n the large and populous Honan and Anhwe provnces. By focusng on the long run verson of the NEG model where the spatal allocaton of economc actvty s no longer gven, we are also able to show that by takng ths spatal allocaton and the resultng market access as gven, one may end up wth rather dfferent conclusons. In partcular, Prefecture ctes lke Shangha that have a very hgh market access for a gven spatal allocaton of frms and labor do not necessarly end up beng the centers of economc actvty once nter-cty frm and labor moblty s taken nto account. The paper s organzed as follows. In the next secton we brefly ntroduce the Hukou system that characterzes Chnese nternal mgraton polcy and dscuss the mportance of ths system from a NEG perspectve. Secton 3 ntroduces the background NEG model as developed by Puga (1999) and explans the relevance of ths model for Chna. Secton 4 dscusses our data for Chnese Prefecture ctes for the perod and gves estmates of the NEG wage equaton. The estmaton wll not only be used to establsh f Chnese wages are hgher for ctes wth a good market access, t also yelds estmates for the substtuton elastcty and the transport cost parameter. These parameter values wll be used as nput n secton 5 where we 3

4 conduct a smulaton analyss of the full NEG model for Chnese ctes. The man am of ths analyss wll be to fnd out whether nterregonal, or n our case, nter-cty labor moblty n Chna changes the relatve economc poston of ctes. The results provde nsghts nto the possble effect of a loosenng of the Hukou system on Chna s nternal economc geography. Fnally, secton 6 concludes. 2. Labor moblty n Chna and the mplcatons from a NEG perspectve Snce the 1950s the Chnese authortes have been much concerned wth nternal labor mgraton flows and n partcular wth rural-urban labor mgraton. The Chnese government alternated between perods of more and less restrctve mgraton polces (Zhao, 2004, Futa et al, 2004, Chan and Buckngham, 2008; World Bank, 2008, ch 5, and Chan, 2009) but ever snce the 1950s the so called Hukou system has been a man feature of the nternal mgraton polces. The Hukou system s equvalent to an nternal vsa system that s meant to regulate mgraton. In recent decades, the system has been qute restrctve n the sense of lmtng (offcal) mgraton flows from rural to urban areas and by also puttng a brake on nter-urban mgraton flows (Chan and Bucknham, 2008, Au and Henderson, 2006a,b; Poncet, 2006, Henderson, 2009). Wthout a vsa for a partcular locaton, a Chnese ctzen has no or only lmted rghts to housng, educaton, food or socal securty n that locaton. Those rghts are ted to one s offcal place of resdence and a change n resdency (f a ctzen for nstance would try to move from a rural area to a cty) wll only be matched wth a transfer of these rghts f the (local) authortes hand out a vsa or permt for the new place of resdence. Untl recently (see n partcular Chan, 2009 or Chan and Buckngham, 2008, pp ), mgraton under the Hukou system had two equally mportant dmensons. The frst concerns the restrctons regardng preferred locatons alluded to above. To be granted permsson to lve and work n a cty or any other locaton and to be enttled wth the aforementoned rghts to publc provsons, households need a local Hukou. The second dmenson s functonal and refers to the dstncton between agrcultural and non-agrcultural workers. Wthn each locaton, a further dstncton s made between agrcultural and non-agrcultural workers. Non-agrcultural workers wth a local Hukou are tradtonally enttled to more rghts than agrcultural workers. Taken together, the local and functonal classfcaton thus defnes 4 possble categores of resdents (Chan, 2009, p.202). 4

5 In recent years, the Hukou system has been changed and has become somewhat less restrctve (Chan and Buckngham, 2008). Wth urban wages outstrppng rural wages, the result has been an ncrease n offcal (temporary) mgraton and also n llegal mgraton nto the boomng (coastal) ctes. One mportant change s that nowadays n many cases the local nstead of the central government decdes upon the permts, whch gves local governments some degree of freedom wth respect to the lenency of grantng these permts. A second change s that the dstncton between a non-agrcultural and an agrcultural Hukou has become relatvely less mportant (Chan, 2009). Ths mples that the key ssue for a mgrant s now foremost whether she has a local Hukou, whch s prmarly concerned wth the locaton of a worker. Despte that the system has become less strngent and more dfferentated, the fact that temporary and llegal mgraton flows to the man ctes have been surgng also reflects the fact that the handng out of a cty Hukou s stll restrcted and that the transacton costs of obtanng such a urban Hukou are consderable:. all these restrctons sharply reduce the benefts and rase the costs of mgraton, partcularly nto large ctes. Mgraton s lmted and most mgraton s short-term, or return mgraton. ( ). Overall the hukou system holds hundreds of mllons of people n locatons where they are not explotng ther earnngs potental. (Futa et al, 2004, p. 2957). Chan (2008) estmates that between 1982 and 2006, the annual volume of Hukou mgrants amounted to mllon people. The mpact of the labor moblty restrctons s clear from the stablty of ths annual offcal mgraton flows (Futa et al, 2004, p. 2957). The stablty of these mgraton flows also suggests that nether the scrappng of the functonal Hukou nor the decentralzaton of the mgraton polcy has had a substantal mpact on the flow of permanent (offcal) mgrants to the ctes. What are the consequences of the Hukou system for Chna? Au and Henderson (2006a,b) forcefully argue that Chnese ctes are too small as a consequence of the mgraton restrctons and that the agglomeraton rents assocated wth urbanzaton are therefore underutlzed (see also Futa et al, 2004). Fgure 1 llustrates the under-urbansaton of Chna n a smple way (Henderson, 2009). Based on the scatterplot, Chna s ncome per capta s assocated wth a below average degree of urbansaton, suggestng that, ceters parbus, a hgher ncome per capta s possble by lettng urbanzaton ncrease. 5

6 Fgure 1. Urbanzaton and gross natonal ncome (GNI) per capta for countres CHN Source: Henderson (2009) based on World Bank Development ndcators, 2006; urbanzed= degree of urbanzaton; lngn= PPP corrected GNI per capta (n logs) The assocated welfare losses of the lack of labor moblty, especally between rural-urban areas as well as ntercty labor moblty, are consderable (Whalley and Zhang, 2007). Nonetheless, by usng mgraton data on the provnce level and based on the emprcal NEG mgraton model of Crozet (2004), Poncet (2006) fnds that nter-provncal mgraton flows n Chna do respond to economc ncentves even though mgraton s subect a sgnfcant dstance decay (mgraton s amed at nearby ctes). Ths effect s stronger wthn than between provnces. Ths confrms the noton that the Hukou system lmts labor moblty, but not completely. However, apart from the general predcton that relaxng the mgraton lmts that exst n the Hukou system wll most lkely result n larger and/or more uneven-szed Chnese ctes (Au and Henderson, 2008), lttle work has been done on assessng the mpact of ncreased labor moblty on the nternal economc geography of Chna. Wll t result n very strong coreperphery patterns, wth most economc actvty concentrated n a few very large ctes? And f so, whch ctes or regons wll be able to attract economc actvty and whch ctes wll thereby become more perpheral? Or wll core perphery patterns become less pronounced? From an economc development perspectve ths s mportant gven the well establshed postve relatonshp between agglomeraton (urbanzaton) and economc development (see World Bank (2008) and Fgure 1 above). Au and Henderson (2006a, p. 568) e.g. estmate that 6

7 about 50-60% of the Chnese Prefecture ctes are underszed and that movng a cty from 100,000 to 1.27 mllon people would ncrease real output per worker by 83%. The man am of ths paper s therefore to shed lght on ths ssue by nvestgatng the mplcatons of dfferent labor moblty regmes for the spatal dstrbuton of people and economc actvty across Chnese ctes. Based on a general NEG model, we analyze theorybased predctons as to the relevance of changes n nterregonal labor moblty for Chna s nternal economc geography. We argue that the NEG approach s useful because labor moblty s plays a key role n NEG models, and smulatons can reveal how the present dstrbuton of economc actvty dffers from stuatons wth ncreased labor moblty. In NEG models a less than perfect nterregonal labor moblty results n wage dfferences: large agglomeratons have hgher wages than more perpheral areas. Economc centers wll offer frms (and workers) better access to upstream and downstream markets. Wth perfect labor moblty, that s, a perfectly elastc nterregonal labor supply, economc centers wll smply attract more workers (and frms) untl (real) wages wll be equalzed across locatons (or that all footloose workers have mgrated to the centre before complete real wage equalty s reached). It s only for a gven spatal dstrbuton of producton and workers or when labor s not perfectly moble that a hgher market access wll result n hgher wages. 3 The Chnese Hukou system s responsble for at least some degree of labor mmoblty, whch accordng to NEG has the effect that Chnese ctes wth a hgh market access have hgher wages than ctes wth a low market access. 4 In the next secton, we wll test ths wages-market access hypothess for our sample of Chnese ctes. But we frst wll brefly ntroduce the NEG model at hand. 3 The NEG model 5 Followng Bosker et al (2010), we use the NEG model by Puga (1999) as our benchmark model to carry out the smulatons for the equlbrum allocaton of economc actvty n and 3 See also our dscusson of the NEG model n secton 3 and n partcular of the wage equaton (2). 4 Indeed, the lnk between mperfect labor moblty and the wages-market access relatonshp s the startng pont for the recent NEG emprcal studes that try to establsh f Chnese wages do depend postvely on market access, see Herng and Poncet (2006, 2007), Au and Henderson (2006a), De Sousa and Poncet (2007), Moreno Monroy (2008), Ln (2003) and our own estmatons n secton 4. Market access s also a determnant of the locaton decsons of frms. Buldng on Head and Mayer (2004) and also nvokng mperfect labor moblty, Amt and Javorck (2008) show for nstance that FDI flows to Chnese regons depends postvely on the market access of those regons, for a smlar result usng an NEG model see Ma (2006). 5 The dscusson of the Puga (1999) model n ths secton closely follows the dscusson n Bosker et al. (2010). 7

8 between Chnese ctes n the next secton. Puga (1999) encompasses a varety of NEG models as specal cases and notably the core models by Krugman (1991), Krugman and Venables (1995) and Venables (1996). More mportantly (see also the dscusson n Hu, 2002), t explctly ncorporates labor moblty. The model enables us to dfferentate between dfferent levels of labor moblty and to llustrate to what extent the equlbra depend on the degree of labor moblty. The M (=1, M) regon verson of the model can be understood as follows (see the Appendx for further detals). Each of the M regons s populated by L workers and endowed wth K unts of arable land. Each regon s economy conssts of two sectors: agrculture and ndustry. Labor s used by both sectors and s moble between sectors wthn a regon and t s ether moble or mmoble between regons. Land on the other hand s used only by the agrcultural sector and s mmoble between regons. The result of ths assumpton s that producton n agrculture s subect to dmnshng returns, and any attempt of manufacturng frms to lure workers away from the agrcultural sector mples wage ncreases (see Chan, 2009, p. 208 on the wage elastcty of ntra-regonal manufacturng labor supply). The depcton of each regon, as beng home to a manufacturng and an agrcultural sector, approxmates the defnton of a Chnese Prefecture cty wth a non-urban part (producng manufacturng goods) and an urban part (producng agrcultural goods). In addton, see Chan and Buckngham (2008) and our dscusson n secton 2, the present-day Hukou system can be sad to be characterzed wth a hgh degree of ntra-prefecture labor moblty. In partcular, the declned mportance of the dstncton between the agrcultural and the non-agrcultural Hukou wthn each Prefecture mples that, n terms of the model, labor can move between the 2 sectors. 6 In equlbrum, wages between the 2 sectors are equalzed n each regon n the model. There s also emprcal evdence to back up the dea that nter-sector labor moblty wthn Chnese Prefecture ctes (our sample, see secton 4) s mportant. Futa et al (2004, p. 2958, Table 2) document for nstance that n the 1990s the non-agrcultural employment growth clearly outstrpped populaton growth whch nter ala sgnals the move wthn(!) prefecture ctes from labor the agrcultural to the manufacturng sector. 6 Ths holds for legal wthn-cty mgraton between the 2 sectors, n Chna there are very substantal temporary as wel as llegal mgraton flwos to ctes where rural, non-ruban workers end up n lvng n the rural part of the cty and have a ob n the manufacturng sector despte the fact that these workers do not a have a local Hukou of that cty. In our data set, we can only deal wth offcal (legal) cty populaton data. For an nstructve (and lucd) summary of the current Hukou system see see also Chan (2009) for a very good ntroducton to and assessment of the Hukou system. 8

9 The agrcultural good s produced under perfect competton and free entry and ext usng a Cobb-Douglas technology 7 and s freely tradable between regons (the exstence of unform agrcultural prces s realstc gven the subsdes on agrcultural prces). An mportant parameter n our analyss of the Chnese case wll be the Cobb-Douglas share of labor (θ) n agrcultural producton. The ndustral sector produces heterogeneous varetes of a sngle good under monopolstc competton and free entry and ext, ncurrng so-called ceberg trade costs when shpped between regons (τ 1 goods have to be shpped from regon to let one good arrve n regon ). Industral producton technology s characterzed by ncreasng returns to scale. The producton nput s a Cobb-Douglas composte of labor and ntermedates n the form of a composte manufacturng good, wth 0 µ 0 the Cobb- Douglas share of ntermedates. The composte manufacturng good s specfed as a CESaggregate (wth σ > 1 the elastcty of substtuton across varetes) of all manufacturng varetes produced. Consumers have Cobb-Douglas preferences over the agrcultural good and a CES-composte (also wth σ > 1 the elastcty of substtuton across varetes) of manufacturng varetes, wth 0 γ 0 the Cobb-Douglas share of the composte manufacturng good. Specfyng the composte manufacturng good n ths way ensures a postve demand from each regon for each manufacturng varety. Together wth the fact that each varety s produced by a sngle frm n a sngle regon, ths mples that trade takes place between Chnese Prefectures. Aganst ths background, we can dstngush between the short-run equlbrum and the longrun equlbrum verson of the model (see also the Appendx). In the short run, the nterregonal dstrbuton of workers (and frms) s fxed and the model reduces to the followng 3 equlbrum condtons for each regon (the 3 condtons refer to the (manufacturng) prce ndex q, the (manufacturng) wage w and total expendtures on manufactures, e ): ( ) 1/(1 σ ) µσ 1 σ (1 µ ) 1 σ ς τ σβ 1 q = Lq w (1) σ 1 ασ (1 µ ) 1/( σ (1 µ )) µ 1 σβ µ /( µ 1) β σ 1 1 σ w = q e q τ (2) σ 1 α( σ 1) 7 Puga (1999) defnes the agrcultural sector somewhat more general. However, when dervng analytcal results, a Cobb-Douglas producton functon n agrculture s used, see Puga (1999, p. 318). 9

10 e = γ ( w L + K r( w )) + µ /(1 µ ) wς L (3) where (n addton to the varables and parameters already ntroduced), α = fxed cost of producng a manufacturng varety, β = margnal cost of producng a manufacturng varety, ς = share of workers n manufacturng; γ = share of ncome spent on the manufactures and K = land endowment, r(w ) = rent earned per unt of land. For our present purposes, we are partcularly nterested n the equlbrum wage equaton (2) 8. Note that the prce ndex, q, has a mxed effect on wages. Wth ntermedate nputs (µ > 0), the term q µ /( µ 1) ndcates that frms n larger regons wll ncur lower transport costs for ther ntermedate nputs than frms n perpheral regons. Better (or cheaper) access to ntermedate nputs gves frms a cost advantage whch enables them to pay hgher wages. Ths so-called suppler access only plays a role when ntermedate nputs are used n the producton process. Prce ndces are also present n the last term between brackets, σ e q 1 τ, whch s also known as Real Market Potental or Real Market Access (Head and Mayer, 2004). As wll become clear n the next secton, we focus n our emprcal analyss on ths term. It ndcates that a regon can pay hgher wages f t s part of and/or surrounded by regons,, wth a hgh real market access. Real market access s postvely related to real ncome and negatvely to transportaton costs. As we already ponted out before, ths postve relatonshp between real market access and wages holds only when we ether assume a gven fxed spatal dstrbuton of workers or a lmted degree of nterregonal labor moblty. The dfference between the short-run and long-run equlbrum verson of the Puga (1999) model s that n the long-run labor can move between regons f nterregonal labor moblty s allowed. In the case of nterregonal labor mmoblty, the short-run verson of the model (eqs (1)-(3) above) also descrbes the long-run equlbrum. Nomnal wage equalty between the manufacturng and agrcultural sector s suffcent n ths case as all workers n a regon are confronted wth the same prce ndex. In the case of nterregonal labor moblty, labor mgraton s drven by nterregonal real wage dfferences and labor moves untl real wages ω are equalzed across all regons: ω = q w = ω (4) γ 8 Vrtually all emprcal contrbutons n the new economc geography lterature are based on the wage equaton. See e.g. Reddng and Venables (2004), Hanson (2005), Knaap (2006) or Herng and Poncet (2009). 10

11 In the model, as s common place n NEG models, the mgraton dynamcs that ensure that (4) wll hold are rather crude because they assume that mgraton s drven by nterregonal wage dfferences only. More specfcally the mgraton dynamcs are specfed as follows 9 : dλ = ψ ω ω λ (4 ) where ω = λ ω s the average real wage across all regons and λ s the share of manufacturng frms n regon ; ψ s the speed of adustment parameter. In secton 5 we wll also apply a more elaborate mgraton dynamcs. Followng for nstance a mgraton equaton estmated for the case of Chna by Poncet (2006) and as an alternatve to (4 ), we wll also allow mgraton to be also determned by addtonal varables, notably blateral dstances between Prefecture ctes and (provncal) border effects. Fnally, we wll also examne the mplcatons of allowng a long run equlbrum wth labor moblty to be reached when real wages are not (fully) equalzed by changng condton (4) such that a real wage dfferences of x% are allowed for a long run equlbrum to be reached,.e. max( ω ω / ω ) = x%, (4 ) 4 Data set, estmaton strategy and results 4.1 Data set At the hghest level of aggregaton, the People s Republc of Chna or Chna for short s composed of 33 admnstratve unts (22 provnces, 5 autonomous regons, 4 large muncpaltes (Beng, Shangha, Tann, and Chongqng) and 2 specal regons (Hong Kong and Macau). The 2 nd ter of regonal dvson, the so called Prefecture level, n Chna conssts of 333 regons that together make up the 33 largest unts. Of these 333 regons at the Prefecture level, 283 regons are Prefecture-ctes. In the Appendx, we lst for every regonal unt at the hghest level of admnstraton (we have data on 30 of the total of 33 regons) 10 the correspondng Prefecture ctes (264 max.) that are n the data set. The Appendx also ncludes a map of Chna wth the provnces and the captal ctes of the provnces. 9 Note that any NEG s not truly a dynamc model, so that talkng about dynamcs s somewhat artfcal. However when smulatng the model n secton 5, we assume that people respond to short-run wage dfferences accordng to some pre-specfed mgraton rule. 10 We have no data on the followng 3 regons out of the group of 33 regons: Hong Kong, Macau, and Tbet. The same underlyng data set s also used by for nstance Au and Henderson (2006a) or Moreno Monroy (2008). 11

12 Our data set for Chna conssts of data for a large subset of these Prefecture ctes: 264 of the 283 ctes. 11 Even though the data do not cover the whole of Chna, t covers the bulk of the populaton and economc actvty. The 264 prefecture ctes cover 86% of total populaton n Chna and 96% of total GDP (data sources: see below). The maorty of the Prefecture ctes can be found n the eastern and southern part of Chna. For our sample of 264 Prefecture ctes, Fgure 2 shows for each Prefecture cty ts share of the populaton (panel a) as well as ts share of secondary ndustry employment (panel b), whch serves as a proxy for the dstrbuton of manufacturng actvty (frms). What s clear from Fgure 2 s that Chnese Prefecture ctes do not only dffer wth respect to ther share n the dstrbuton of frms and people, but also that the dstrbuton of manufacturng actvty does not correspond one-to-one wth the dstrbuton of people. Secondary employment (panel b) s more spatally concentrated (Herfndahl ndex (HI)= 0,01) than the populaton (HI=0,006) wth a few notable Prefecture ctes lke Beng, Tann, Shangha and Chongqng (the large black dot n panel b) each beng home to 3-5% of secondary employment. 12 For for nstance these economc centers, ther respectve populaton shares amount to barely 1%. Fgure 2 also reveals that Prefecture ctes vary substantally n geographcal sze (km 2 ). 11 Note that the prefectures not only nclude the urban populaton of a cty, but also the rural area surroundng a partcular cty, n many Prefecture ctes n our data set the maorty of the populaton as well as the bulk of the land area s classfed as non-urban. Apart from the 283 Prefecture ctes the 2 nd ter of regonal admnstraton also conssts of 17 Prefectures (manly n Xnang and Tbet), 30 Autonomous Prefectures (n western Chna, regons wth a large share of ethnc mnortes) and 3 Leagues (regons n Inner Mongola), see 12 A perfect evenly spread of populaton or secondary employment would gve a HI of 0,0038. As we wll explan n secton 5 the Prefecture cty of Chongqng s a rather specal case. 12

13 Fgure 2 Dstrbuton of people and economc actvty across Prefecture ctes Panel a. Dstrbuton of populaton Source: our data set; N/A = not avalable and/or not a prefecture cty Panel b. Dstrbuton of sec. ndustry employment As to the dstrbuton of people (panel a), Fgure 2 shows a concentraton of dark(er) dots n an area that can be loosely defned as the part of Chna that les wthn the Shangha- Chongqng-Beng trangle. See also the map of Chna n the Appendx. Ths trangle manly encompasses the provnces of Honan, Hupeh, Anhwe and Santung of whch only the latter s a coastal provnce. Taken together these 4 provnces alone are home to 30% of the Chnese populaton and to almost 25% of the 264 Prefecture ctes. Combned wth the fact that the prefecture ctes n these (non-coastal) provnces are often also relatvely large n terms of land area, the presence of many relatvely populous ctes n ths Chnese heartland wll turn out to be mportant n the long-run equlbrum analyss of our NEG model for Chna n secton 5. The data on Chnese prefecture ctes are obtaned from the Chnese Data Center from the Unversty of Mchgan (see The orgnal data source s the Natonal Bureau of Statstcs of Chna. From ths database we have data on each prefecture cty s wages w, expendtures (ncome) e, the total number of people, the total avalable arable land (n km 2 ), and the (great-crcle) dstances each prefecture cty-par. Also, 13

14 as control varables n the wage equaton regressons, we have collected nformaton on each cty s dstance to the nearest maor port, secondary educaton enrolment, the share of employment n bankng and fnance, and ts total area. As to the dependent varable n our wage equaton, and n lne wth other studes that are faced wth data lmtatons, we use gdp per capta (see also Reddng and Venables, 2004) as proxy for these. We do have data on the average urban wage of workers n each the prefecture cty, where urban refers to the urban part of the Prefecture area, and also data exsts on the total wage of employees. Followng Herng and Poncet (2007, p. 13), who convncngly argue that the avalable prefectural wage data do not suffcently reflect wages n the prvate sector, we dd, however, not use ether of these two wage measures as our dependent varable (see Table 1 below), and opted for urban GDP per capta for each cty nstead. 4.2 Estmatng the wage equaton for Chnese ctes Equaton (5) s the emprcal specfcaton of equaton (2) that we use n the estmatons. However, due to lmted data avalablty we do not have cty data on ntermedate (nputs) or the prce ndex q. To partly crcumvent ths problem, we assume µ=0 n the estmatons, whch addtonally - solves the mult-collnearty problem as emphaszed by Reddng and Venables (2004) that arses when ncludng both suppler and market access n the NEG wage equaton 13 Furthermore, we use a smplfed measure of the prce ndex n our estmatons (see below). Wth these qualfcatons, the emprcal counterpart of equaton (2) becomes (n logs): 29 1 ln w (5) 1 σ σ 1 ( ) = ln e τ q σ = 1 The ndces (=1 264) and (=1 30) refer to the (max.) 264 Chnese ctes and 30 Chnese provnces respectvely for whch we have data. The provnce varables (=1, 30) n the real market access term, that s the term between brackets n eq. (5), are constructed by aggregatng the correspondng prefecture varables n the sample at the provnce level (see Hanson (2005) for a smlar aggregaton for hs sample of US countes). Wthn the market access term, and followng e.g. Hanson (2005), Brakman et al. (2004) or Bosker et al. (2007), 13 Even f we would have for nstance a detaled nput-output matrx on the cty level, the ncluson of the suppler access term alongsde the market access term n the emprcal wage equaton would be problematc snce suppler and market access are typcally found to be strongly multcollnear (Reddng and Venables, 2004). 14

15 we specfy trade costs as, τ =(D ) τ, D s defned as the dstance between the cty and the captal of the provnce. For nternal dstance n the market access term, we use: 2/3 (land area of the provnce/π). The basc mplcaton of eq. (5) s the same as wth the equlbrum wage eq. (2) from the NEG model: Chnese ctes wth a hgher real market access do have a hgher gdp per capta than ctes wth lower market access. But besdes beng able to verfy ths predcton of the NEG model, the estmaton of equaton (5) yelds estmates for 2 (key) model parameters, the substtuton elastcty, σ, and the transport cost or dstance parameter, τ. There are two basc emprcal strateges to estmate a NEG wage equaton lke equaton (5). The frst strategy, as ntroduced by Reddng and Venables (2004), s to use blateral trade data between regons to construct market access and to use ths trade based measure of market access n the estmaton of the wage equaton. Herng and Poncet (2006, 2007), Ln (2003) and Ma (2006) follow ths strategy and use data on nter-provncal trade. The second strategy, followng Hanson (2005), s to try to estmate wage equaton drectly. We opt for ths second strategy because there are no nter-cty trade data at the Prefecture-cty level that (suffcently) cover our sample. Ths s why we draw on the drect estmaton strategy, lke the related Chna market access studes by Amt and Javorck (2008) or Au and Henderson (2006a). Apart from a relatvely hgh real market access, gdp per capta n cty could be relatvely hgh for varous other reasons. In urban economcs, regonal (= cty) wages can be relatvely hgh f a regon has a hgher productvty level because of ncreasng returns assocated wth the well-known Marshallan externaltes of labor market poolng, knowledge or nput sharng. To control for ths, and followng Herng and Poncet (2006), Brenlch (2006) and Bosker and Garretsen (2008), we nclude populaton densty and a proxy for human captal as control varables. 14. To control for the possble relevance of the economc structure, we added for each cty the share of bankng and fnance n total Prefecture-cty employment. Our sample perod s and our panel thus ncludes (max.) 264 Chnese ctes at the Prefecture level. To control for cty-specfc fxed effects, Prefecture cty dummes are ncluded as well, and we added tme/year fxed effects to take care of common shocks that may have ht the sample of Chnese ctes. Instead of cty-specfc fxed effects, we also estmated the wage equaton by usng Provnce fxed effects (by ncludng 29 Provnce 14 For ther sample of 51 Chnese ctes, Herng and Poncet (2006) use mcro data whch allows for a much better control of varous wage determnants besdes market access. 15

16 dummes). Wth Provnce fxed effects, see columns c and d of Table 1, we could also add the (tme-nvarant) dstance to the nearest seaport as a control varable. The NEG wage equaton (5) stpulates that only market access matters. It s relatvely easy to rewrte the NEG model and hence the equlbrum wage equaton (2) such that t ncludes a cty-specfc technology effcency varable alongsde the market access term. In ths way the control varables ntroduced above can be gven a theoretcal groundng n our NEG model. In our case ths addtonal term n the wage equaton s approxmated by cty fxed effects and tme fxed effects to capture respectvely the cty-specfc but tme nvarant determnants of w as well as the cty-nvarant and tme-varyng determnants of w. The ncluson of human captal, populaton densty and the bankng&fnance employment share s meant to capture the possblty of the cty s technologcal effcency also beng drven by both cty and tme varyng varables. 15 Gven the well-establshed fact (Hu, 2002, Au and Henderson, 2006a) that coastal ctes n Chna can pay hgher wages than landlocked ctes because of ther superor market access to the rest of the world, we allow for the fact that Chna s not a closed economy and that market access wth respect to the rest of the world may also be relevant. To allow for ths, we make the basc assumpton that Chna trades wth the rest of the world by sea and that, for smplcty, the rest of the world conssts of the three economc blocs: the USA, Japan and the EU. Economc mass outsde Chna was then measured as the sum of dstance-weghted GDP of the USA, Japan and the EU 15, see Table 1. Next, we ether ncorporate the economc mass outsde of Chna to the overall the market-access term e τ σ q σ n equaton (5), = 1 or by addng economc mass outsde Chna as an addtonal control varable n the regressons: ln[(cty s dstance to the nearest seaport)*ln(dstance-weghted foregn gdp)]. A fnal specfcaton ssue concerns the treatment of the unobservable manufacturng prce ndex, q, that s present n the Real Market Access term To approxmate the prce ndex q n the wage equaton (2), we smplfy the prce ndex equaton (1) from the underlyng model as 15 To ntroduce a locaton specfc technology effcency parameter n the equlbrum wage equaton (2) (see Reddng and Venables (2004)): the cost functon for a manufacturng varety h changes to c (α+(x(h)) nstead of the standard formulaton, α+β(x(h)), where c s the locaton specfc technologcal effcency parameter, and α and β are, respectvely, the fxed and margnal cost parameters (see the appendx for the consequences of ths change). For a dervaton see the Appendx. 16

17 follows (Brakman, Garretsen and Schramm, 2004). 16 For each Chnese provnce (note that n our specfcaton (5) real market access s measured at the provnce level), we focus on wages n provnce of a manufactured good produced n that provnce, and the average wage of a manufactured good produced n other provnces. The determnaton of the smplfed prce ndex for manufactures requres a measure of average dstance between provnce and the other provnces. The dstance to the nearest economc centre s a straghtforward measure. The smplfed prce ndex becomes 1 1 σ 1 σ { λ + ( λ )( τ ) } W 1 w, centre σ q (6) = 1 where w s the average wage of all the prefectures n the sample, τ,centre s the ceberg transport cost parameter based on the dstance between the captal of provnce and the nearest economc centre; Beng, Shangha or Quandung. In ths shortcut of the prce ndex, λ s the share of number of employees n provnce, the outsde wage _ w s weghted by τ,centre Columns (a)-(d) n Table 1 gve the estmaton results of estmatng equaton (5) n four dfferent ways. The most relevant fndng for our purposes s that the substtuton elastcty σ and the transport cost parameter τ are sgnfcant. A substtuton elastcty of 5.8 s well wthn the range found n smlar NEG studes and t fulfls the theoretcal requrement that σ > 1. The estmated coeffcent for τ s 0.6. Taken together the two model parameters consttute the so called free-ness of trade parameter, φ =(D ) τ(1-σ) and n our preferred specfcaton (column a) ths amounts to φ =(D ) 0.63(1-5.88). Au and Henderson (2006a), and followng Poncet (2006), put the dstance coeffcent τ at 0.87 for ther sample of Chnese ctes. 17 A second man fndng based on Table 1 s that the market access (MA) coeffcent, 1/σ (see equaton 5), s 0.17 (1/σ =1/5.88=0.17). Ths s n lne wth Herng and Poncet (2006, MA-coeff. 0.1), Herng and Poncet (2007, MA-coeff. 0.07) and Moreno Monroy (2008, MA-coeff. 0.1), but t s somewhat lower than the MA-effect found by Au and Henderson (2006a). The 16 Two other optons to deal wth unavalablty of data on cty prced ndces are; to go for nomnal nstead of real market access (see Au and Henderson 2006a) whch takes us out of the world of NEG, or to opt for a soluton whereby other equlbrum condtons of the underlyng NEG model can be used (Hanson, 2005). The 2 nd route followed by Hanson (2005) has the drawback that real wage equalzaton must be assumed, whch s unrealstc for Chna. 17 In hs calbraton of a NEG model for Chnese regons followng the ntroducton of the Natonal Trunk Hghway System, Faber (2009) uses σ=7 and τ=0.33 whch yelds a free-ness of trade parameter that s qute close to one arrved at after estmaton our wage equaton (5). 17

18 sze of the market access coeffcent mples that a 1% ncrease n market access wll boost Chnese cty-wages by more than 0.25%. The ntroducton of control varables helps to separate spatal dependency from spatal heterogenety. In our preferred specfcatons wth cty fxed effects (columns a-b), the control varables, human captal and populaton densty, are sgnfcant but the share of bankng and fnance employment s not. Also foregn market access does not seem to be very relevant. Chna s nternal market access dwarfs the (dstance weghted) market access of the rest of the world. Ths s not to deny that coastal ctes lke Shangha are not specal but ths can be pcked up by the specfc fxed effects through our set of Prefecture dummes and for many coastal ctes the fxed effect coeffcent s sgnfcant (not shown here). Interestngly, the postve coeffcent for populaton densty suggests that negatve crowdng effects (congeston effects) are not an ssue (yet). As a further check as to the robustness of our fndngs, we also ran cross-secton estmatons of wage equaton (5) for 1999 and The reason beng that t s beleved that the relablty of the cty-populaton data may be less from 2000 onwards as the Chnese authortes allegedly manpulated these data. 18 When we estmate wage equaton (5), wthout prefecture dummes and usng model specfcaton (d) as shown n Table 1, for 1999 or 2000 for a sample of 233 ctes the estmaton results for the substtuton elastcty and the trade cost parameters are very close to the full-sample estmaton results from Table 1. For 1999 we fnd that the 2 correspondng coeffcents are (st-errors between brackets): (2.95) and (0.013). For 2000 these coeffcents are (st-errors between brackets): (2.74) and (0.012). 18 We thank our dscussant at the NARSC 2009 conference, Vernon Henderson, for pontng out ths problem and for the suggeston to use cross-secton estmatons for the ntal years of our sample as a robustness check. For the years 1999 and 2000 we only have data for 233 of our 264 ctes. 18

19 Table 1 Estmatng equaton (5) for Chnese Prefecture ctes Dependent varable: ln(urban gdp per cap.) (a) (b) (c ) (d) σ (.70) τ.632 (.04) (.51) (Enrollment secondary educaton) / populaton Bankng & fnance (as share n employment) Ln(Dstance to nearest seaport) Ln(Dstance to nearest seaport cty)*ln(dstanceweghted GDP) Ln(Populaton densty).795 (.47) (.70).636 (.04) (.51).791 (.04) (.77).589 (.04) (1.12) (1.03) (.77).584 (.04) (1.11) (1.02) (1.85) (.03) (.12).101 (0.02).101 (.02).153 (.02).144 (.02) Sample perod Observatons Ad. R Foregn GDP ncluded n market Yes No Yes No access term? Fxed effects? 263 prefecture dummes, 6 tme dummes 263 prefecture dummes, 6 tme dummes 29 provnce dummes, 6 tme dummes 29 provnce dummes, 6 tme dummes -Whte heteroskedastcty-consstent standard errors between brackets, coeffcents n bold are sgnfcant at least at the 5% level; # of ctes ncluded: (max.) 264 [unbalanced panel of 264 ctes], NLS, estmaton. -secondary educaton s the student enrolment n regular secondary schools as a fracton of cty s populaton; -populaton densty s measured as cty s populaton (n 10,000 persons) per km 2. -Dstance-weghted GDP s defned here as Y t /Dstance k, where s USA, Japan or EU 15, k s the maor seaport of USA, Japan or EU 15 (Los Angeles, Yokohama and Rotterdam, respectvely), s the maor seaport n Chna (Shangha). The measure of dstance s the great-crcle dstance and the great crcle dstance between Shangha and Rotterdam s calculated as the sum of the dstance between Shangha-Dbout, Dbout-Suez, Suez- Gbraltar, Gbraltar-Rotterdam. In our smulatons n the next secton, we wll use the estmated values n column (a) as our benchmark parameters, because these estmaton results are based on the specfcaton when foregn market access s ncluded n the theoretcally most consstent way. Ths means that we wll ntally set the substtuton elastcty at 5.88 and the dstance parameter at 0.63 n our smulatons wth the full-blown NEG model. The estmaton of wage equaton (5) s only a means to an end. The estmaton results serve as an nput for the smulaton analyss wth the complete NEG model of secton 3, n order to 19

20 fnd out what the equlbrum allocaton of economc actvty across our sample of Chnese ctes would look lke under varous regmes of nter-prefecture-cty labor moblty. The analyss of the long-run equlbrum consttutes the maor dfference wth the other NEG wage equaton or market access studes of Chna, e.g. Herng and Poncet (2006, 2007) or Au and Henderson (2006a) that bascally stck to a pre-fxed or gven spatal allocaton of workers and frms across Chna. In a stylzed way, t has the advantage of allowng us to assess the mpact of a loosenng of the Hukou restrctons wthn the well-defned boundares of our general NEG-model. Based on our estmaton results, we calculated for each cty for each year the mpled market access (n logs), so how large s each cty s market access term from equaton (5)? It turns out that (on average) Shangha has the hghest market access durng our sample perod Ths may not come as surprse but for each year there are qute a few ctes among the top-20 ctes n terms of market access that are non-coastal ctes, not part of coastal provnce and relatvely small wth respect to ther own-cty GDP. Ths ndcates that as far as the spatal dstrbuton of market access s concerned, t s not the case that coastal and/or large ctes wth a relatvely hgh own-cty GDP nevtably also have the hghest market access. Economc geography seems to matter. Havng sad ths, n the long-run equlbrum analyss n secton 5 market access s endogenous so the (short-run) dfferences n market access and wages that follow from the estmaton of wage equaton (5) do not amount to much n the sense that they can gve no clear predcton as to whch ctes wll be the economc centers. To be more specfc, a cty s hgh rankng n terms of the market access that result from the estmaton of a NEG wage or market access equaton lke our equaton (5) does not guarantee that ths cy wll be a (or the) economc center once frm and labor moblty s allowed for and the full NEG model of secton comes nto play to determne equlbrum spatal allocaton of economc actvty across our sample of Prefecture ctes. The mere fact, based on Table 1, that Shangha has the hghest market access s not a suffcent condton for Shangha to be as mportant when we allow for frm and labor moblty. In the long run analyss of the Puga (1999) NEG model, addtonal varables wll have to be taken nto account that have been gnored n the (short run) analyss n our estmaton of the wage equaton n the present secton. In partcular, see our dscusson n secton 3, the allocaton of the producton factors K (land) and L (labor) across the ctes comes nto play. Here, and ust lke wth market access results, t s thus not a pror clear why (f at all) a 20

21 partcular ntally large or centrally located (coastal) cty, lke Shangha, should also be an economc centre n the long run equlbrum. As we already for nstance ponted n our dscusson of Fgure 2 both populaton (labor, L ) and land area (land, K ) are not are not dstrbuted evenly across the Prefecture ctes. There s n other words no straghtforward a pror answer n terms of our NEG model to the queston whch Prefecture ctes or Chnese regons wll stand to gan from an ncrease n labor moblty and what the resultng equlbrum spatal allocaton of economc actvty wll look lke. To get an answer to ths queston, one has to let the full NEG model from secton 3 do the talkng and t s to ths topc that we turn n the remander of the paper. 5 Labor moblty and the long run agglomeraton equlbrum for Chnese ctes 5.1. Smulaton set-up We are now fnally n a poston to analyse the equlbrum spatal allocaton of economc actvty for our sample of 264 Chnese ctes. Analytcal solutons for the case of M > 2 regons are non-exstent n NEG (unless one assumes that all regons are at equdstance, whch s a mathematcal constructon wth no applcatons n any many regon realty), so for the case where trade costs or dstance s asymmetrc, one has, by necessty, to rely on smulaton analyses (Behrens and Thsse, 2007, Futa and Mor, 2005 and Bosker et al 2010) to establsh an equlbrum allocaton. For the smulatons, we need not only nformaton on the substtuton elastcty (σ = 5.88) and the trade cost parameter (τ = 0.63), but also on the share of ntermedate nputs (µ) n manufacturng producton, the share of ncome spent on manufactures (γ) and the Cobb-Douglas share of labor n agrculture (θ). Based on a regonal nput-output table for Chna for 2000 we can nfer µ and γ for Chna as a whole. To be specfc, we take µ=0.51 (I-O table: Chnese ntermedate demand for Chnese manufacturng by Chnese manufacturng frms) and γ = 0.34 (I-O table: Chnese fnal demand for Chnese manufacturng as share of total fnal demand for Chnese output). 19 Based on the Statstcal Yearbook of Chnese Bureau of Statstcs, we ntally set θ = 0.87, the Cobb-Douglas labor share n agrcultural producton, ndcatng that agrcultural producton s (stll) relatvely labor ntensve n Chna. 20 For the cty-specfc producton factor, K (land), we take the (2002) cultvated arable land share of each Prefecture cty, and for the cty specfc factor, L 19 Source: regonal I/O table from Insttute of Developng Economes (2003), Mult-Regonal Input-Output Table for Chna We arrve at ths Cobb-Douglas share by usng the rato (compensaton employees/valued added) for the Chnese agrcultural sector. 21

22 we take the (2002) populaton share of each Prefecture cty. 21 Fnally, n some of the smulatons, we wll also not only use the estmaton results for the two model parameters (σ and τ) but also take the results for control varables nto account, ncludng those for the cty fxed effects (not shown n Table 1). As stated above, we are manly nterested n the queston how varyng assumptons wth respect to nter-regonal or nter-cty labor moblty mght change the current spatal equlbrum allocaton for Chna. Note that wthn each Prefecture cty we assume ntersectoral labor moblty, that s, rural-urban labor moblty. 22 To analyse the relevance of changes n the degree of nter-cty labor moblty we bascally looked at two basc scenaros for labor moblty: () no or restrcted labor moblty across Chna [wth no labor moblty equlbrum eqs. (1)-(3) suffce to smulate the equlbrum; wth restrcted labor moblty we only allow for ntra-provncal labor moblty]; () labor moblty across Chna [as wth ntra-provncal labor moblty labor mgraton s governed by (alternatves to) mgraton dynamcs (4 ) such that the real wage condton (4) or (4 ) s realzed but as opposed to ntra-provncal labor moblty labor moblty now ncludes all Prefecture ctes as a possble locaton choce]. From a model perspectve, the presence of (some degree of) nterregonal labor moblty s mportant because t changes the mx of agglomeraton and spreadng forces that determnes the long run spatal equlbrum. Wthout any nterregonal labor moblty, and gven the (crucal) assumpton that agrcultural producton takes place under decreasng returns to scale, 21 We decded to consder the Prefecture cty of Chongqng as an outler n the followng way. In 1990, the then provnce of Chongqng was awarded the same status as Shangha, Beng and Tann that s to say one of (effectvely) a provnce that s also a prefecture cty. Ths admnstratve change creates a problem for our present purposes because the cty Chongqng s not only much larger than all other prefecture ctes t s also huge n terms of (arable) land area. (n 2002 Chongqng had a populaton of 31 mllon people and an arable land area of 2283 (x 1000 hectare)). In populaton terms ths about 3 tmes as large as the 2 nd and 3 rd largest ctes, Shangha and Beng, and n arable land terms ths s more than 4 tmes as large as these 2 other man ctes, Shangha and Beng, wth the same admnstratve status. These dfferences really matter n the long run analyss. Especally the land area dfference matters n the sense that because of the decreasng returns to food producton, see secton 3, labor s very scarce n Chongqng and ths puts an upward pressure on local wages to the effect that all Chnese footloose workers nvarably wll end up n Chongqng n the long run. A smlar effect would occur f the Chnese authortes would suddenly decde to upgrade the large and populous provnces of Honan or Anhwe to the prefecture cty status. To correct for ths artfcal sze of the Prefecture cty of Chongqng we took as the (2002) populaton ts urban populaton (9.9. mllon) and then 456 (x1000 hectare) as the correspondng arable land area. 22 In the sample of ctes the maorty of the Prefecture cty populaton s classfed as agrcultural populaton, and smlarly for most Prefecture ctes the largest part of the Prefecture area (n km 2 ) s classfed as non-urban. 22

23 the nflux of manufacturng frms to core or center regons has a prce tag n the form of hgher wages, because the addtonal workers needed n the manufacturng sector can only come from the local agrcultural sector. The excess labor demand from the manufacturng sector drves up wages n ths regon, and creates nterregonal wage dfferences between core and perpheral regons. The rsng wages also strengthen, ceters parbus, a spreadng force for manufacturng frms to return to regons wth lower wages. Wth perfect nterregonal labor moblty, ths upward pressure on wages s countered by the addtonal supply of labor that moves from perpheral to core regons. Before we present the long run equlbrum smulaton results n sectons 5.2 and 5.3, we would lke to stress that these model smulatons are certanly not amed at gvng a pn-pont predcton as to what wll happen to Chnese agglomeraton patterns once more labor moblty s ntroduced. The model used s obvously much too stylzed to be able to predct wth any degree of accuracy whch ctes wll gan or lose. But, and usng the no labor moblty case as a benchmark, we do beleve that n a qualtatve sense the model can be used to better understand the mplcatons of ncreased labor moblty for agglomeraton outcomes. It s here that our model smulatons have ther value added compared related NEG-market access studes that take the spatal allocaton of frms and workers as gven. 5.2 Smulaton results for nter-cty labor mmoblty and ntra-provncal labor moblty Gven the parameter choces above and the ntal shares of cty populaton L and arable land K, the long run equlbrum s found for the case of labor mmoblty when the allocaton of labor between the agrcultural and manufacturng sector s such that wages are equalzed between these sectors. Ths smulaton refers to an extreme Hukou scenaro and t calls for a soluton of eqs. (1)-(3). To characterze the resultng long run equlbrum for Chna, we can depct the equlbrum n terms of a cty s equlbrum share of manufacturng frms or ts populaton share (recall Fgure 2a,b). Note that the equlbrum dstrbuton of people n the case of labor mmoblty between the Prefecture ctes s by defnton the same as the actual or ntal dstrbuton shown n Fgure 2a. Wth nter-cty labor mmoblty, the smulaton results for the equlbrum spatal allocaton of frms by and largely resemble the actual dstrbuton of frms across our sample of Prefecture ctes as shown by Fgure 2b. The correlaton coeffcent between the smulated dstrbuton of manufacturng frms and the dstrbuton as shown n Fgure 2b s The 23

24 overall degree of spatal concentraton as measured by Herfndahl ndex s respectvely 0,006 (st dev 0,003) and 0,010 (st dev 0,005). The man dfference wth the actual dstrbuton s that ctes that are centrally located n for nstance the Honan provnce n the populous heartland of Chna (recall Fgure 2a) lke the Prefecture ctes of Zhoukou and Nanyang gan n terms of ther frm share at the expense of coastal ctes lke Beng, Tann or Shangha. Only 10 out of 264 prefecture ctes do not have a postve equlbrum share of the footloose frms. As wth all other smulaton cases reported below, we preformed three man senstvty analyses to check for the robustness of these fndngs for the case of labor mmoblty. Frst, we allowed for the fact, see Table 1, that market access s not the only determnant of Chnese wages and we ncluded the estmaton results for the cty fxed effects and human captal nto the smulatons as well. Second, we lowered trade costs to capture n a rather crude way what wll happen f due to for nstance large scale mprovements n the between-cty nfrastructure transport costs wll fall (see also Faber, 2009). Instead of the estmated τ=063, we lowered the trade cost parameter to 0.36 and 0.1. Thrd, gven the mportance of the (arable) land varable n the NEG model and the consderable (arable) land dfferences between the Prefecture ctes, we also re-ran all smulatons wth all ctes havng the same (arable) land sze. For the present case of nter-cty labor mmoblty, these senstvty analyses do not change the resultng equlbra. The only noteworthy change s that when trade costs are lowered, the agglomeraton of frms gets somewhat more pronounced whch s n accordance wth the NEG model n the sense that ceters parbus lower trade costs foster agglomeraton. What happens next f labor mgraton s allowed but restrcted to the Provncal level only? In ths case, real wages are equalzed wthn provnces, but not between provnces. Ths follows for nstance Futa et al. (2004) or Poncet (2006) who argue that provncal borders matter most for mgraton and that mgraton s restrcted to nearby ctes. In terms of the model ths mples that labor mgraton s determned by real wage equalzaton condton (4) and mgraton dynamcs (4 ) but wth mgraton by assumpton - beng restrcted to the ntra- Provncal level. Table 2 below llustrates the results when labor mgraton s restrcted to nclude ntra- Provncal labor moblty only. As we explaned n secton 4.1, we have data for 33 of the 30 admnstratve regons that exst at the hghest level of publc admnstraton n Chna (of the 33 regons there are 22 Provnces and 4 muncpaltes, the latter are also Prefecture regons 24

25 and denoted n bold n Table 2). Table 2 thus shows the equlbrum frm shares for each of the 30 regons (for smplcty denoted as Provnces). Even though all footloose actvty now ends up n a sngle cty wthn each Provnce, the resultng core-perphery s (by constructon) less extreme when compared to the case of perfect labor moblty. In terms of equlbrum populaton shares, the populaton moves n tandem wth frms to the 30 Provncal Prefecture ctes shown n Table 2 Table 2 Costly Mgraton I: Intra-Provncal Mgraton, Equlbrum Frm Shares Provnce cty Populaton Share Frm share Beng Beng ,006 Tann Tann 0,008 0,005 Hebe Baodng 0,061 0,066 Sans Yuncheng 0,024 0,018 Inner Mongolan Autonomous Reg Chfeng 0,011 0,006 Laonng Shenyang 0,038 0,038 Krn Changchun 0,022 0,016 Helunkang Harbn 0,034 0,029 Shangha Shangha 0,012 0,007 Kangsu Yancheng 0,065 0,070 Tsekang Wenzhou 0,041 0,039 Anhwe Fuyang ,055 Fuan Quanzhou 0,030 0,026 Kangs Ganzhou 0,035 0,030 Santung Lny 0,082 0,112 Honan Nanyang 0,087 0,111 Hupeh Jngzhou 0,047 0,044 Hunan Changde 0,058 0,061 Kwantung Zhanang 0,069 0,078 Kwangs Guln 0,035 0,029 Hanan Sanya 0,001 0,000 Chongqng Chongqng 0,009 0,005 Szetswan Chengdu 0,065 0,074 Kwetsou Zuny 0,014 0,009 Yunnan Zhaotong 0,018 0,012 Shaanx Wenan 0,031 0,025 Kansu Tanshu 0,014 0,009 Tsngha Xnng 0,002 0,001 Nngsa Wuzhong 0,003 0,001 Snkang-Ughurë Urumq 0,002 0,001 In bold, Prefecture ctes that are also an admnstratve unt at the Provnce level. The data for Chongqng only apply to the urban part of Chongqng. Fgure 3 llustrates the results shown n Table 2 for the equlbrum frm dstrbuton for the case of ntra-provncal mgraton. The 2 ctes wth largest frm (and populaton) shares are Lny (Santung provnce) and Nanyang (Honan provnce), the 2 ctes denoted n talcs n 25

26 Table 2 and n black n Fgure 3. The relatvely large populaton share for thee 2 ctes s a drect consequence of the fact that Santung and Honan are the two most populous provnces. The fact that the equlbrum frm shares for these two ctes exceed ther equlbrum populaton shares sgnals that footloose frms end up n centrally located ctes that offer a superor market access. Ths fndng s consstent wth prevous case of no labor moblty when frms already dsplayed a tendency to move to the ctes n the populous provnces. The 4 adacent provnces of Honan, Anhwe, Santung and Hupeh are home to 28% of the overall sample populatons and together are home to 61 Prefecture ctes. The case of ntra- Provncal labor moblty s also nterestng because t has been suggested that t may offer a feasble polcy opton for Chnese polcy makers for a (gradual) loosenng of the Hukou system (Henderson, 2007, p. 9). Fgure 3 Intra-provncal Mgraton frm dstrbuton 26

27 5.3 Agglomeraton outcomes for 4 cases of ntra-chnese labor moblty The two above cases of no nter-cty labor moblty and ntra-provncal labor moblty are useful benchmarks but n realty nter-regonal labor moblty n Chna s nether absent nor restrcted to wthn- Provnce moblty only. As we argued n secton 2, the Hukou system s stll restrctve but at the same tme allows (ncreasngly) for (temporary) nterregonal labor mgraton. In order to show the long run equlbra for varyng degrees of labor moblty across Chna, we ran smulatons for 4 stylzed cases of nterregonal labor moblty across Chna. The frst case takes the Puga (1999) NEG model at face value and assumes that labor mgraton s determned by real wage equalzaton condton (4) and mgraton dynamcs (4 ). Ths s the case of perfect labor moblty where workers mgraton behavor s only drven by nter-cty real wage dfferences, however small. Ths s the case of perfect, unrestrcted labor moblty. The second case assumes that mgraton s costly such that a threshold of 10% n wage dfferences needs to be crossed before workers mgrate. Ths threshold ncreases the number of possble long-run equlbra because real wages do not have to be equalzed. Any spatal allocaton that for whch the nter-cty real wage dfferences are 10% wll now be a long run equlbrum. All smulaton outcomes wthn ths 10% wage-bandwdth are thus equlbra: strct equalty of equaton (4) s no longer needed. Equaton (4) s therefore replaced by equaton (4 ) from secton 3 (wth x%=10%). The thrd case replaces the mgraton dynamcs (4 ) by (a smplfed verson of) the mgraton dynamcs as estmated by Poncet (2006) n her study on the mgraton between and wthn Chnese provnces: λd 0.87 exp( delta * D )( w w ) same provnce where D same provnce s a dummy varable ndcatng whether ctes and are located n the same provnce, and delta=3.22 the correspondng coeffcent (see eqs. (4) and (5) and Table, column (4) n Poncet (2006). As a result mgraton n ths thrd case not only depends on real wage dfferences but also (negatvely) on the blateral dstance between locatons and (negatvely) on a provncal border effect. Note that wth ths smulaton the mgraton dynamcs are dfferent than n the NEG model of secton 3: the Poncet equaton above wll be used nstead of (4 )), but the 27

28 condton that real wages have to be equalzed for a long run equlbrum, equaton (4), stll needs to hold. The fourth and last smulaton case s a mx between the two prevous ntermedate cases: we assume (a) that mgraton only takes place when real wage dfferences exceed the 10% wage dfference threshold (equaton (4 ) nstead of (4)), whle at the same tme (b) the mgraton dynamcs are based on our verson of Poncet (2006). Though the detals dffer, the upshot of the model smulatons for these 4 cases of nter-cty labor moblty across Chna s that one ends up wth very dfferent (and far more extreme agglomeraton outcomes) as compared to the labor mmoblty benchmarks form the prevous sub-secton. Agan, the results are not meant to be an actual predcton as to what more labor moblty mght ental but they do lead to two mportant qualtatve fndngs n our vew. Frst, the smulatons for all 4 cases of labor moblty clear show that ctes n the (non-coastal) provnces of Honan and Anhwe wll gan n mportance. Recall from Fgure 2a n secton 4 and see also Table 2 above that these provnces are large n terms of (ntal) populaton as well as land area. Ths fndng s robust to our three senstvty checks (ncludng the fxed effects and human captal results from Table 1; no dfferences n arable land; lower trade costs). The second man fndng s that our model smulatons show that ntal dfferences n market access, as reported at the end of secton 4 followng the estmaton of the wage equaton, do not need to be decsve as to where frms and workers wll locate once frm and labor moblty are allowed for. The market access dfferences across Chnese ctes, as alluded to at the end of secton 4, follow from a short-run nterpretaton of the model where market access dfferences between ctes are merely taken to be gven and fxed. To llustrate our pont, the ctes of Zhoukou and Nanyang n the Honan provnce or Luan n the Anhwe provnce are only md-table n the lstng of market access n secton 4 based on our estmaton results. But these same ctes are the man or even only centers of economc actvty n the long run equlbra wth labor moblty across Chna. In the case of perfect labor moblty, where labor moblty s bascally seen as frctonless and beng solely drven by nter-cty real wage dfferences, the Prefecture cty of Nanyang (Honan provnce) becomes the agglomeraton of Chna. Ths tendency of NEG models to 28

29 dsplay (near) full agglomeratons n case of labor moblty s well-known and too stark to be true (see Bosker et al, 2010), but ths agglomeraton bas does not concern us here. Our concern s wth the qualtatve search as to where n the urban landscape of Chna the core ctes are most lkely to pop up wth respect to the equlbrum agglomeraton. Or ven more specfcally, our qualtatve search s about the relatve shft n the spatal allocaton of economc actvty that the ntroducton of more labor moblty would mply The basc fndng s that for the regme of perfect labor moblty ths s typcally n ctes n the populous, noncoastal provnces of Honan or Anhwe that are home to a relatvely large number of Prefecture ctes wth n many cases an above average (arable) land area. In secton 4 we concluded that the coastal cty of Shangha ntally had the largest market access. In the case of perfect labor moblty ths only shows up n Shangha beng the center of agglomeraton when we, rather artfcally, do not allow for arable land dfferences between ctes. 23 But even n the case of no arable land dfferences, f we perform senstvty analyses by ether ncludng the estmaton results for cty-fxed effects and human captal nto the smulatons or by lowerng trade costs, Shangha ceases to be the agglomeraton and centre shfts back to the ctes n the Anhwe or Honan provnce (to be precse to the Prefecture ctes of Luan or Zhoukou respectvely). 24 The fact that wth perfect labor moblty the model gravtates towards (extreme) agglomeraton s a drect consequence of the NEG model. Wth perfect labor moblty (and opposed to the case of labor mmoblty), there are no nter-cty wage dfferences. Wth ths dsperson force beng non-actve, the resultng cty agglomeraton(s) do(es) not at some pont become to expensve for manufacturng frms n terms of wage costs such that they should relocate to more perpheral ctes. 23 Arable land dfferences are mportant n the NEG model of Puga (1999) because wth decreasng returns n the producton of the agrcultural good, more arable land mples that labor s more scarce and that manufacturng frms n that cty ceters parbus have to have to pay hgher wages to lure workers away from the agrcultural sector. 24 We also expermented wth the ntroducton of congeston, whch ntroduces an addtonal spreadng force nto the model. To explore ths possblty we changed the cost functon for manufacturng frms by lettng the fxed costs n cty be a functon of the total number of manufacturng frms n that cty, see the Appendx for modelng detals. Wth ctes allegedly stll beng underszed (Au and Henderson, 2006a,b) one may wonder f congeston s really a relevant force yet (see also the postve coeffcent for populaton densty n Table 1, whch suggests that congeston s not (yet) en ssue). Not surprsngly, ncludng congeston n the smulaton model for Chna, the result (f anythng) was that the equlbrum frm dstrbuton became more dspersed than wthout the congeston. 29

30 The basc fndngs for the case of perfect labor moblty are renforced for the 3 other cases of Chna-wde labor moblty when we allow for more realstc labor moblty regmes by changng the mgraton dynamcs (Poncet, 2006) and/or do not requre that real wages have to fully equalzed for a long run equlbrum to be reached. In these cases and takng the smulaton results for our 3 senstvty checks nto account as well, the qualtatve concluson s nvarably that footloose frms and workers end up n Prefecture ctes lke Nanyang, Zhoukou and Luan n the populous heartland of Chna. Ths concluson also holds when we assume that all ctes are equally szed n terms of (arable) land. Wth mgraton now, as opposed to the case of perfect labor moblty, somehow costly and not frctonless (e.g. beng a negatve functon of dstance), there s even a stronger mpetus for the equlbrum agglomeraton of both frms and people to gravtate to those Provnces and ctes where ntally the bulk of people and land s already concentrated. What s typcally decsve n NEG models as to whch locatons or ctes end up as beng the centers n the long run equlbrum s relatve locaton, that s the locaton of a cty vs a vs other ctes. The ctes of Nanyang, Zhoukou and Luan (the 3 dots n Fgure 4) are centrally located rght n the heart of the provnces of Honan, Hupeh, Anhwe and Santung (see Fgure 2a and the grey areas n Fgure 4 below) whch effectvely form the nsde of the trangle that can be drawn between the muncpaltes of Shangha-Chongqng-Beng/Tann, see Fgure 4. Each of these muncpaltes or mega ctes s larger than any of ctes n the 4 adacent provnces sngled out n Fgure 4. In a mult-regon verson of the core Krugman (1991) NEG model, Krugman (1993) already showed that ntal sze need not be decsve to determne whch locaton(s) end(s) up consttutng the agglomeraton. 25 In the case of Chna ths explans why a Prefecture cty lke Luan (the rght dot n Fgure 4) wth only 6.3 mllon people can become an agglomeraton. The reason s that Luan thrves upon ts centralty relatve to other ctes and n partcular to the large muncpaltes. Or n terms of the NEG model, when we allow for labor moblty a cty lke Luan offers footloose frms and workers n Chna a superor locaton n terms of market access to the rest of Chna. 25 For a further llustraton of the feature of NEG models that ntal sze need not be decsve for the fnal or equlbrum allocaton of footloose actvty, see for nstance Brakman, Garretsen, and Van Marrewk (2009), pp

31 Fgure 4 Labor Moblty and the Agglomeratons n the Muncpalty Trangle 6. Conclusons In ths paper we have used a new economc geography (NEG) model to analyse the relatonshp between market access, labor moblty and agglomeraton for Chna. The man am of the paper s to smulate based on estmates of key model parameters n the NEG wage equaton the dynamcs of agglomeraton n Chna under varous assumptons regardng labor moblty. The estmates use a data set for 264 Chnese prefecture ctes for the perod Next, and based on actual Chnese data, we performed smulatons wth the complete model, under dfferent assumptons regardng labor moblty. Our paper crucally dffers from related recent NEG studes for Chna (e.g. Ma, 2006, Ln, 2003, Herng and Poncet, 2006, 2007, De Sousa and Poncet, 2007, Amt and Javorck, 2008) to the extent that we do not take the spatal allocaton of labor (and frms) as gven. Instead, we wll make use of the complete NEG model and not ust the equlbrum wage equaton and ts real market access component. Ths enables us to analyse the relevance of labor moblty for the equlbrum agglomeraton for Chnese ctes. Ths s the frst contrbuton of the paper to the 31

32 lterature. A second contrbuton s that we go beyond the smple mgraton dynamcs that underle NEG models and look also at more ntrcate mgraton dynamcs. Our analyss shows that the economc geography of Chna that would result from ncreased labor moblty s frst and foremost determned by wthn Chna dfferences n market access, populaton and (arable) land. In a qualtatve sense, the NEG model predcts that ncreased labor moblty would prmarly beneft centrally located, non-coastal Prefecture ctes lke Zhoukou or Luan n the large and populous Honan and Anhwe provnces. By focusng on the long run verson of the NEG model where the spatal allocaton of economc actvty s no longer gven, we are also able to show that by takng ths spatal allocaton and the resultng market access as gven, one may end up wth rather dfferent conclusons. In partcular, Prefecture ctes lke for nstance Shangha that have a very hgh market access for a gven spatal allocaton of frms and labor do not necessarly end up beng the centers of economc actvty once ntercty frm and labor moblty s taken nto account. References Amt, M. and B. S. Javorck, (2008), Trade Costs and Locaton of Foregn Frms n Chna, Journal of Development Economcs, Vol. 85, pp Au, C-C. and J.V. Henderson, (2006a),Are Chnese Ctes Too Small? Revew of Economc Studes, Vol. 73, pp Au, C-C. and J.V. Henderson (2006b),How mgraton restrctons lmt agglomeraton and productvty n Chna, Journal of Development Economcs, Vol. 80, pp Bosker, M., S. Brakman, H. Garretsen, and M. Schramm (2010), Addng Geography to the New Economc Geography, Journal of Economc Geography, forthcomng {see also CESfo Workng Paper, 2038, CESfo, Munch, 2007} Brakman S., H. Garretsen, and M. Schramm (2006), Puttng new economc geography to the test: free-ness of trade and agglomeraton n the EU regons, Regonal Scence and Urban Economcs, 36(5), Brakman, S. H.Garretsen, C.van Marrewk (2009), The New Introducton to Geographcal Economcs, Cambrdge Unversty Press, Cambrdge. Chan, K.W. (2003), Urbanzaton n Chna n the 1990s: New Defntons, Dfferent Seres, and Revsed Trends, The Chna Revew, 3(2), pp Chan, K. W. (2008), Internal Labor Mgraton n Chna: Trends, Geographcal Dstrbuton and Polces, Proceedngs of Unted Natons Expert Group Meetng on Populaton Dstrbuton, Urbanzaton, Internal Mgraton and Development, ESA/P/WP.206, Unted Natons, pp Chan, K.W. and W.Bucknham (2008), Is Chna Abolshng the Hukou System, Forthcomng, The Chna Quarterly, No.195. Chan, K.W. (2009), The Chnese Hukou System at 50, Eurasan Geography and Economcs, 50(2), pp Crozet, M., Do mgrants follow market potentals? An estmaton of a new economc geography model. Journal of Economc Geography, Vol. 4,

33 De Sousa, J and S. Poncet (2007), How are Wages Set n Beng?, CEPII Workng Paper , CEPII, Pars. Faber, B. (2009), Integraton and the Perphery: The Unntended Effects of New Hghways n a Developng Country, paper presented at Urban Economcs Sessons at the NARSC 2009 conference November 2009 San Francsco. Futa M., T. Mor, J.V. Henderson and Y. Kanemoto (2004), Spatal Dstrbuton of Economc Actvtes n Japan and Chna n J. V. Henderson and J-F. Thsse (eds.), Handbook of Regonal and Urban Economcs, volume IV, North Holland, Amsterdam, Henderson, J.V, wth J. Qugley and E Lm, (2007), Urbanzaton n Chna: Polcy Issues and Optons, Chna Economc Research and Advsory Programme, avalable at Henderson, J.V. (2009), Ctes and Development, Journal of Regonal Scence, forthcomng, avalable at Herng, L. and S. Poncet (2006), Market Access and Indvdual Wages: Evdence from Chna, forthcomng 2010 n Revew of Economcs and Statstcs. Herng, L. and S. Poncet (2007), Economc Geography, Spatal Dependence, and Income Inequalty n Chna. mmeo. Unversté Pars I Panthéon-Sorbonne. Hu, D. (2002), Trade, rural-urban Mgraton, and Regonal Income Dsparty n Developng countres: a spatal general Equlbrum model nspred by the case of Chna, Regonal Scence and Urban Economcs, Vol.32, pp Krugman, P.R. (1991), Increasng returns and economc geography, Journal of Poltcal Economy, 99, Krugman, P.R. (1993), On the Number and Locaton of Ctes, European Economc Revew, 37, Krugman, P.R. and A.J. Venables (1995), Globalzaton and the nequalty of natons, The Quarterly Journal of Economcs, vol 110, Krugman, P.R. (2009), Increasng Returns n a Comparatve Advantage World, November 2009, avalable at Krugman, P.R. (2010), The New Economc Geography, Now Mddle-Aged, Aprl 2010, avalable at Ln, S. (2003), Internatonal Trade, Locaton, and Wage Inequalty n Chna, WIDER Dscusson Paper, no. 2003/61, WIDER Unted Natons Unversty. Ma, A.C. (2006), Geographcal Locaton of Foregn Drect Investment and Wage Inequalty n Chna, The World Economy, pp Moreno Monroy, A.I. (2008), The dynamcs of spatal agglomeraton n Chna: an emprcal assessment, mmeo, Unversty of Gronngen [forthcomng n Papers n Regonal Scence]. Poncet, S. (2006), Provncal mgraton dynamcs n Chna: Borders, costs and economc motvatons, Regonal Scence and Urban Economcs, 36, pp Puga, D. (1999), The rse and fall of regonal nequaltes, European Economc Revew, Vol. 43, Reddng, S. and A.J. Venables (2004), Economc geography and nternatonal nequalty, Journal of Internatonal Economcs, 62(1), Venables, A.J. (1996), Equlbrum Locatons of Vertcally Lnked Industres, Internatonal Economc Revew, 37, Whalley, J. and S.Zhang (2007), A numercal smulaton analyss of (Hukou) labor moblty restrctons n Chna, Journal of Development Economcs, Vol. 83, pp World Bank, 2008, World Development Report 2009, Washngton. 33

34 Appendx Basc set up of Puga (1999) model Producton The agrcultural good s produced under perfect competton and free entry and ext usng Cobb- Douglas technology and s freely tradable between regons. The ndustral sector produces heterogeneous varetes of a sngle good under monopolstc competton and free entry and ext, ncurrng so-called ceberg trade costs when shpped between regons (τ 1 goods have to be shpped from regon to let one good arrve n regon ). Industral producton technology s characterzed by ncreasng returns to scale,.e. producton of a quantty x(h) of any varety h requres fxed costs α and varable costs βx(h) that are assumed to be the same n each regon. Ths, together wth free entry and ext and proft maxmzaton, ensures that n equlbrum each varety s produced by a sngle frm n a sngle regon. The producton nput s a Cobb-Douglas composte of labor and ntermedates n the form of a composte manufacturng good, wth 0 µ 0 the Cobb-Douglas share of ntermedates. The composte manufacturng good s specfed as a CES-aggregate (wth σ > 1 the elastcty of substtuton across varetes) of all manufacturng varetes produced. The resultng mnmum-cost functon assocated wth the producton of a quantty x(h) of varety h n regon can be wrtten as: µ M1 C( h) = q w µ ( α + β x( h)) (1) where q s the prce ndex of the composte manufacturng good, and regon.. M w the manufacturng wage n Preferences Consumers have Cobb-Douglas preferences over the agrcultural good and a CES-composte (also wth σ > 1 the elastcty of substtuton across varetes) of manufacturng varetes, wth 0 γ 0 the Cobb-Douglas share of the composte manufacturng good. Specfyng the composte manufacturng good ths way ensures demand from each regon for each manufacturng varety, whch, together wth the fact that each varety s produced by a sngle frm n a sngle regon, mples that trade takes place between regons. Equlbrum Havng specfed preferences over and the producton technologes of the manufacturng and agrcultural good, the equlbrum condtons of the model can be calculated. Proft maxmzaton and free entry and ext determne the share of labor employed, L, the wage level w n agrculture, whch A equals the margnal product of labor, and the rent earned per unt of land r( w ). The former two n turn pn down the share of workers n manufacturng, ς. Gven the assumed Cobb-Douglas producton functon n agrculture, wth labor share θ, we have that: M A L L K θ ς = = 1 = 1 A L L L w where 0 θ 1 denotes the Cobb-Douglas share of labor n agrculture, and 1 1 θ A M A L and (2) A L the number of workers n manufacturng and agrculture respectvely. Equaton (2) shows that, n contrast to Krugman (1991), where agrculture uses only land (θ = 0), or to Venables (1996), where agrculture employs only labor (θ = 1), the share of a regon s labor employed n manufacturng s endogenously determned n ths model. It ncreases wth a regon s labor endowment and agrcultural wage level and decreases wth a regon s land endowment and wth the Cobb-Douglas share of labor n agrcultural producton. Consumer preferences n turn determne total demand for agrcultural products n regon as: x A = (1 γ ) Y (3) 34

35 In the ndustral sector, utlty maxmzaton on behalf of the consumers, combned wth proft maxmzaton and free entry and ext, gves the famlar result that all frms n regon set the same prce for ther produced manufacturng varety as beng a constant markup over margnal costs: σβ M (1 ) p q µ = w µ (4) σ 1 where q s the prce ndex of the composte manufacturng good n regon defned by: 1 1 σ 1 σ (1 σ ) q = τ n p (5) where n denotes the number of frms n regon and M ( σ 1) 1 w = (1 µ ) n p ( α + β x ) ( ς L ) σβ (6) s the manufacturng wage level n regon. It also gves total demand for each manufacturng varety produced (comng from both the home regon as well as foregn regons ) whch s the same for each varety n the same regon due to the way consumer preferences are specfed: σ ( σ 1) 1 σ x = p e q τ (7) where n (7) demand from each foregn regon s multpled by τ because (τ 1) of the amount of the product ordered from regon melts away n transt (the ceberg assumpton), and ( σ 1) e = γ Y + µ n p ( α + β x ) (8) σβ s total expendture on manufacturng varetes n regon (the frst term representng consumer expendture and the second term producer expendture on ntermedates), where Y A (1 ) M ( A = w ς L + w ς L + r w ) K + nπ (9) s total consumer ncome consstng of workers wage ncome, landowners rents and entrepreneurs profts respectvely. Due to free entry and ext these profts are drven to zero (π = 0), thereby unquely defnng a frm s equlbrum output at: x = α( σ 1) / β (10) Fnally, to close the model, the labor markets are assumed to clear: θ M A ( σ 1) M 1 θ L = L + L = (1 µ ) n p ( α + β x ) w + K w A σβ (11) where the demand for labor n agrculture, M L A L, follows from the assumpton of Cobb-Douglas technology n agrculture and the term between square brackets represents the manufacturng wage bll. Moreover equatng labor supply to labor demand n the ndustral sector gves an mmedate relatonshp between the number of frms and the number of workers n ndustry: ς L n = ασ (1 µ ) q w µ M µ Long run equlbrum Next, to solve for the long run equlbrum, Puga (1999) dstngushes between the case where labor s both nterregonally and ntersectorally moble and the case when t s only ntersectorally moble. Wthout nterregonal labor moblty, long run equlbrum s reached when the dstrbuton of labor between the agrcultural and the ndustral sector n each regon s such that wages are equal n both sectors. Ths s ensured by labor beng perfectly moble between sectors drvng ntersectoral wage dfferences to zero. When nstead labor s also nterregonally moble, not only ntersectoral wage A L 1 1 (12) 35

36 dfferences are drven to zero n all regons n equlbrum. Workers now also respond to real wage (utlty) dfferences between regons by movng to regons wth the hgher real wages (utlty) untl real wages are equalzed between all regons, hereby defnng the long run equlbrum Interregonal labor mmoblty The long run equlbrum n case of nterregonal labor mmoblty can now be shown to be a soluton {w,q } of three equatons that have to hold n each regon. In our case (when usng wage-worker space) these are, usng the fact that n equlbrum w M = w A = w : ( ) 1/(1 σ ) µσ 1 σ (1 µ ) 1 σ ς τ σβ 1 q = Lq w (13) σ 1 ασ (1 µ ) 1/( σ (1 µ )) µ 1 σβ µ /( µ 1) β σ 1 1 σ w = q e q τ σ 1 α ( σ 1) (14) e = γ ( w L + K r( w )) + µ /(1 µ ) wς L (15) where (13) s obtaned by substtutng (4) and (12) nto (5), (14) by substtutng (4) and (10) nto (7), and (15) by substtutng (4), (10) and (12) nto (8). Interregonal labor moblty In case of nterregonal labor moblty, a soluton to (13)-(15) merely consttutes a short run equlbrum. Wth nterregonal labor moblty, workers wll move between regons n response to real wage dfferences untl the nterregonal real wage dfferences, that are possble to persst when workers are unable (or unwllng) to move between regons, are no longer present. More formally, the LRE soluton {w,q } for each regon has to adhere to the addtonal condton that real wages, ω, are equal across all regons: γ ω = q w = ω (16) Changng the model: spatal Heterogenety and Congeston In the smulatons of secton 5.4 of the paper, we extend the Puga (1999) model n two ways: a) Followng Reddng and Venables (2004), we allow for a slghtly dfferent cost structure: µ 1 µ µ 1 µ C = q w c ( α + x ) as compared to the Puga where we have C = q w ( α + βx ), the man change beng that we now add a cty-specfc technology effcency parameter c that allows for the fact that, as the estmaton results n Table 1 also suggest, dfferences n w across Chnese ctes are not only drven by dfferences n market access but also by dfferences n ths effcency parameter. In the smulatons we can then plug n ths new cost structure and use the estmaton results for for nstance human captal or the cty fxed effects as an nput for c b) In the case of (Chnese) ctes, one drawback of the Puga (1999) model s that t does not nclude congeston as a spreadng force meanng that as the overall sze of the cty ncreases (n frm or worker share), the ncreased crowdedness has as such no costs as opposed what s normally assumed n urban economcs (see Henderson, 1974). Includng congeston n cost functon can be done n several ways. We can assume α, β n (1) to be a functon of n. We choose the most smple κ varant: α( n ) = αn Margnal costs congeston complcates matters consderably by makng (4) also a functon of congeston, that s, n, whch mples that (12) s also nserted n (4). Brakman, et al (1996, pp ) show, however, that n general the addtonal spreadng forces for ths specal case (fxed cost congeston) and the more general congeston case (fxed and margnal cost congeston) produce smlar results n a qualtatve sense n ther NEG model. So we choose κ α( n ) = αn So, when we combne extensons a and b we end up we followng amended cost µ 1 µ κ structure: C = q w c ( α n + x ). Ths cost structure and ts mplcatons for the equlbrum condtons (13)-(15) above are the bass for the model smulatons n secton

37 µ 1 µ κ More specfcally, by usng C = q w c ( αn + x ) one arrves at the followng 3 equlbrum condtons (settng κ = 0, gves the model wth spatal heterogenety, but wthout congeston): (13) σ 1 q = σ 1 ( ασ (1 µ ) [(1 σ ) 1/(1+ κ )] (1 σ ) 1/(1+ κ ) c τ ( ς ) 1 /(1+ κ ) L q [ µ /(1+ κ ) + µ (1 σ )] w [ µ /(1+ κ ) + (1 µ )(1 σ )] 1/(1 σ ) (14) w σ σ 1 κ /(1+ κ ) [ µκ σµ ( κ + 1)]/( κ + 1) [ κ σ ( κ + 1)]/( k + 1) κ /(1+ κ ) = ( ) [ α( σ 1)] [ ασ (1 µ )] q c ( ς L ) e σ 1 (15) = (15) above q ( σ 1) (1 σ ) τ ( κ + 1) µκ + σ (1+ κ )(1 µ ) 37

38 Provnces and the Prefecture ctes Beng Tann Hebe Sans Inner Mongolan Autonomous Regon Laonng Krn Helunkang Shangha Kangsu Tsekang Anhwe Beng Tann Shazhuang Tayuan Hohhot Shenyang Changchun Harbn Shangha Nanng Hangzhou Hefe Tangshan Datong Baotou Dalan Jln cty Qqhar Wux Nngbo Wuhu Qnhuangdao Yangquan Wuha Anshan Spng Jx Xuzhou Wenzhou Bengbu Handan Changzh Chfeng Fushun Laoyuan Hegang Changzhou Jaxng Huanan Xngta Jncheng Tonglao Benx Tonghua Shuangyashan Suzhou Huzhou Maanshan Baodng Shuozhou Dandong Changbashan Daqng Nantong Shaoxng Huabe Zhangakou Yuncheng Jnzhou Songyuan Ynchun Lanyungang Jnhua Tonglng Chengde Xnzhou Yngkou Bacheng Jamus Huaan Quzhou Anqng Cangzhou Lnfen Fuxn Qtahe Yancheng Zhoushan Huangshan Langfang Laoyang Mudanang Yangzhou Tazhou Chuzhou Hengshu Pann Hehe Zhenang Lshu Fuyang Telng Suhua Tazhou Suzhou Chaoyang Suqan Chaohu Huludao Luan Bozhou Fuan Kangs Santung Honan Hupeh Hunan Kwantung Kwangs Hanan Szetswan Kwetsou Yunnan Fuzhou Nanchang J'nan Zhengzhou Wuhan Changsha Guangzhou Nannng Hakou Chengdu Guyang Kunmng Xamen Jngdezhen Qngdao Kafeng Huangsh Zhuzhou Shaoguan Luzhou Sanya Zgong Lupanshu Qung Putan Pngxang Zbo Luoyang Shyan Xangtan Shenzhen Guln Panzhhua Zuny Yux Sanmng Juang Zaozhuang Pngdngshan Ychang Hengyang Zhuha Wuzhou Luzhou Anshun Zhaotong Quanzhou Xnyu Dongyng Anyang Xangfan Shaoyang Shantou Beha Deyang Baoshan Zhangzhou Yngtan Yanta Heb Ezhou Yueyang Foshan Fangchenggang Manyang Nanpng Ganzhou Wefang Xnxang Jngmen Changde Jangmen Qngzhou Guangyuan Longyan J'an Jnng Jaozuo Xaogan Zhangae Zhanang Gugang Sunng Nngde Ychun Taan Puyang Jngzhou Yyang Maomng Yuln Neang Shangrao Weha Xuchang Huanggang Chenzhou Zhaoqng Hezhou Leshan Rzhao Luohe Xannng Yongzhou Huzhou Base Nanchong Lawu Sanmenxa Suzhou Huahua Mezhou Hech Ybn 38

39 Lny Nanyang Loud Shanwe Dazhou Dezhou Shangqu Heyuan Yaan Laocheng Xnyang Yangang Bazhong Bnzhou Zhoukou Qngyuan Zyang Heze Zhumadan Dongguan Zhongshan Chaozhou Jeyang Yunfu Shaanx Kansu Tsngha Nngsa Xnang Uyghur Aut.Regon Chongqng X'an Lanzhou Xnng Ynchuan Urumq Chongqng Tongchuan Jayuguan Shzushan Karamay Bao Jnchang Wuzhong Xanyang Bayn Wenan Tanshu Yan'an Juquan Hanzhong Zhangye Yuln Wuwe Ankang Pnglang 39

40 Source: 40