NBER WORKING PAPER SERIES DISTANCE, SKILL DEEPENING AND DEVELOPMENT: WILL PERIPHERAL COUNTRIES EVER GET RICH? Stephen Redding Peter K.

Transcription

1 NBER WORKING PAPER SERIES DISTANCE, SKILL DEEPENING AND DEVELOPMENT: WILL PERIPHERAL COUNTRIES EVER GET RICH? Stephen Reddng Peter K. Schott Workng Paper NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue Cambrdge, MA January 2003 We thank Joshua Azenman, Mary Amt, Andrew Bernard, Sebastan Edwards, Davd Hummels, Jon Temple, Henry Thompson, and Tony Venables for valuable comments. We are also grateful for feedback receved durng semnars at the EIIT9, the London School of Economcs and the NBER IASE conference n Monterrey Mexco. Marco Schonborn and Martn Stewart provded able research assstance. The vews expressed heren are those of the authors and not necessarly those of the Natonal Bureau of Economc Research by Stephen Reddng and Peter K. Schott. All rghts reserved. Short sectons of text not to exceed two paragraphs, may be quoted wthout explct permsson provded that full credt ncludng notce, s gven to the source.

2 Dstance, Skll Deepenng and Development: Wll Perpheral Countres Ever Get Rch? Stephen Reddng and Peter K. Schott NBER Workng Paper No January 2003 JEL No. F12, F14, O10 ABSTRACT Ths paper models the relatonshp between countres dstance from global economc actvty, endogenous nvestments n educaton, and economc development. Frms n remote locatons pay greater trade costs on both exports and ntermedate mports, reducng the amount of value added left to remunerate domestc factors of producton. If skll-ntensve sectors have hgher trade costs, more pervasve nput-output lnkages, or stronger ncreasng returns to scale, we show theoretcally that remoteness depresses the skll premum and therefore ncentves for human captal accumulaton. Emprcally, we explot structural relatonshps from the model to demonstrate that countres wth lower market access have lower levels of educatonal attanment. We also show that the world s most perpheral countres are becomng ncreasngly remote over tme. Stephen Reddng Peter K. Schott London School of Economcs Yale School of Management Houghton Street, London 135 Prospect Street WC2A 2AE New Haven, CT Unted Kngdom and NBER s.j.reddng@lse.ac.uk peter.schott@yale.edu

3 Dstance, Skll Deepenng, and Development 2 1. Introducton Why some natons are rch and others are poor s perhaps one of the oldest and most fundamental questons n economcs. In 1996, the per capta ncome of the country at the 90th percentle of the world ncome dstrbuton was more than 30 tmes hgher than the country at the 10th percentle. The persstence of such dfferences s surprsng n lght of the ncreasng ntegraton of goods and fnancal markets n the post-war perod. Economsts have ponted to a number of factors whch may have prevented these ncome dfferences from beng arbtraged away, ncludng nsttutonal neffectveness, sluggsh technology dffuson and endowment dsadvantages. 1 A more recent lne of research has hghlghted the potental mportance of trade costs n reducng per capta ncome. 2 These trade costs nclude not only the expense of physcally movng products between locatons but also all nformaton, montorng and polcy (e.g. tarff) costs assocated wth transactng at a dstance. Because frms located n remote locatons pay greater trade costs on both ther sales to fnal markets and ther purchases of mported ntermedate nputs, they have less value added avalable to remunerate domestc factors of producton. In ths paper we focus on an addtonal penalty of remoteness. We demonstrate that beng located on the economc perphery can reduce the return to skll, thereby reducng ncentves for nvestment n human captal accumulaton. Ths penalty magnfes the effect that economc geography can have on cross-country per capta ncome; ncreasng a country s relatve trade costs not only reduces contemporaneous factor rewards, but also lowers gross domestc product by suppressng human captal accumulaton and 1 Engerman and Sokoloff (1997) and Leamer et al. (1999), for example, analyse the effect of nsttutons and resource endowments on educatonal attanment n Latn Amerca. Studes examnng the lnks between human captal, development, and growth nclude Benhabb and Spegel (1994), Bls and Klenow (2000), Echer and Garca-Penalosa (2001), Galor and Mountford (2001), Lucas (1988), and Mankw et al. (1992). 2 See, n partcular, Hanson (1998) and Reddng and Venables (2001). For earler, more nformal analyses of locaton and per capta ncome, see Hummels (1995) and Leamer (1997). Ths paper focuses on economc geography (the locaton of economc agents relatve to oneanother n space) rather than physcal geography (e.g clmate). For a dscusson of the latter, see for example Gallup et al (1998).

4 Dstance, Skll Deepenng, and Development 3 decreasng the supply of hgh-ncome sklled workers. 3 Ths result emerges from an extenson of the standard two sector (agrculture and manufacturng) Fujta et al (1999) economc geography model to allow unsklled ndvduals to endogenously choose whether to nvest n educaton. We beleve the role of economc geography n explanng persstent low levels of educatonal attanment n developng countres to be mportant and largely neglected n the exstng lterature. The paper reports three man theoretcal results. Frst, we show that countres located further from global economc actvty have a lower skll premum f manufactured goods are relatvely skll ntensve and face relatvely large trade costs. The ntuton for ths result can be conveyed va the well-known Stolper-Samuelson theorem: ncreased remoteness has the same affect as a reducton n the relatve prce of the manufactured good. Because manufacturng s relatvely skll ntense, the relatve wage of sklled workers and the ncentve to educate falls. Second, we demonstrate that ths result s robust to more general assumptons regardng trade costs, n partcular the assumpton that trade costs are hgher for agrculture than for manufacturng. 4 The result generalzes because of nput-output lnkages and ncreasng returns to scale n skll ntensve manufacturng. Input-ouput lnkages are mportant because trade costs must be pad on both mports and exports, wth the result that even relatvely small trade costs can be magnfed nto a relatvely large share of value added. Increasng returns to scale, on the other hand, emphaszes the mportance of proxmty to large markets. Frms that are remote from large markets have to charge a lower prce net of trade costs n order to export suffcent quantty to cover fxed costs. As a result, the equlbrum skll premum depends upon both physcal remoteness (.e. blateral trade costs) and economc remoteness (.e. the spatal dstrbuton of economc actvty). Thrd, we show how our model can be used to formalze the role of a number of other determnants of human captal nvestment, ncludng agr- 3 A wde range of emprcal studes for developed and developng countres provde evdence that sklled or educated workers receve hgher wages (see Psacharopoulos 1994 for a survey of ths lterature). 4 Bernard et al (2002), for example, show that whle US tarffs arehgherforskll ntensve manufactures (e.g. electroncs), freght and nsurance costs are hgher for bulk commodtes (e.g. Food).

5 Dstance, Skll Deepenng, and Development 4 cultural productvty and technology. We demonstrate that hgher agrcultural productvty (or, more generally, an abundance of natural resources) hnders manufacturng development and reduces ncentves to nvest n human captal. We also show that a transfer of manufacturng technology from developed to developng countres not only rases output per capta drectly but also has a postve ndrect effect through nduced human captal accumulaton. In general, the ndrect effect wll not be nternalzed by prvate sector agents, and the exstence of ths pecunary externalty provdes a potental ratonale for polces desgned to accelerate technology transfer. Whle the man focus of the paper s theoretcal, we explot structural relatonshps from the model to provde emprcal evdence that countres located far from centers of world economc actvty are characterzed by relatvely low levels of educatonal attanment. We also provde evdence that the world s most perpheral countres are becomng relatvely more remote from global economc actvty over tme. The paper proceeds as follows. Sectons 2 and 3 outlne the theoretcal model and explore the relatonshp between remoteness and equlbrum nvestments n skll. Secton 4 generalzes the analyss to allow for a more general specfcaton of trade costs. Secton 5 examnes the role of other determnants of human captal nvestments. Sectons 6 and 7 use the structure of the model to derve emprcal measures of market access and examne the lnk between market access and educatonal attanment. Secton 8 concludes. 2. Theoretcal Model Ths paper bulds upon exstng theoretcal research on new economc geography as syntheszed n Fujta et al (1999). 5 We extend the standard economc geography model by ntroducng endogenous human captal accumulaton. The analyss emphaszes the mportance of the nterplay between ncreasng returns to scale, transport costs, nput-output lnkages, 5 See also Krugman (1991), Krugman and Venables (1990, 1995), and Venables (1996). In the nterests of tractablty, much of ths lterature has assumed a sngle factor of producton, labor. Recent research by Amt (2001) and Strauss-Kahn (2001) has ntroduced consderatons of Heckscher-Ohln trade theory nto ths framework by endowng countres wth exogenous quanttes of multple factors of producton.

6 Dstance, Skll Deepenng, and Development 5 and human captal nvestments. 6 We derve predctons for the relatonshp between remoteness, nternatonal trade, human captal nvestment, and levels of per capta ncome Preferences and Endowments The world conssts of {1,...,R} countres. Each country s endowed wth a mass of L consumers. Consumers have one unt of labour whch s suppled nelastcally wth zero dsutlty. Ths unt of labour begns n an unsklled state, but ndvduals choose endogenously whether or not to nvest n becomng sklled. Consumer preferences are dentcal and homothetc, and are defned over consumpton of a homogenous agrcultural good and a varety of dfferentated manufacturng goods. For smplcty, the utlty functon s assumed to take the Cobb-Douglas form, U j = A (1 µ) j M µ j, 0 <µ<1 (1) where A denotes consumpton of the homogeneous agrcultural good and M corresponds to a consumpton ndex of dfferentated varetes. Gong forward, we use j to denote a country that s demandng or mportng a good and to denote a country that s producng or exportng a good. The consumpton ndex of dfferentated varetes takes the form M j = " R X =1 Z n 0 σ/(σ 1) " R # σ/(σ 1) X m C j(z) dz# (σ 1)/σ = n m C (σ 1)/σ j,(2) where σ>1sthe elastcty of substtuton between manufacturng varetes and the second equaton explots the fact that, n the equlbrum establshed below, all products produced n a country are demanded by country j n the same quantty. As a result, we dspense wth the ndex z and rewrte the ntegral as a product. n denotes the number of varetes produced n country and m C j denotes the amount of each varety produced n country for fnal consumpton n country j. 6 In terms of Marshall (1920) s three forces of agglomeraton (a pooled market for specalzed sklls, nput-output lnkages, and knowledge spllovers), our analyss focuses on nput-output lnkages. For a model of agglomeraton emphaszng search frctons n the labour market, see Amt and Pssardes (2002), whle Mor and Turrn (2002) emphasze complementartes between product varety, product qualty, and sklls. =1

7 Dstance, Skll Deepenng, and Development 6 Dual to the manufacturng goods consumpton ndex (M j )samanufacturng goods prce ndex (G j )defned over the prces of ndvdual varetes produced n and sold n j (.e. p M j ), " X R Z 1/(1 σ) " n R # 1/(1 σ) X G j = p M j (z) dz# 1 σ = n (p M j ) 1 σ, (3) =1 0 where the second equaton makes use of the symmetry n equlbrum prces Producton Technologes The homogenous agrcultural good s produced under condtons of perfect competton wth the followng constant returns to scale technology, Y = θ Y (S Y ) φ (L Y ) 1 φ, 0 <φ<1 (4) where Y denotes output of the agrcultural good; L Y denotes the amount of unsklled labour allocated to ths sector; S Y denotes the amount of sklled labour allocaton; and θ Y ndexes agrcultural productvty. 7 To facltate comparson of our results wth the standard economc geography model wthout endogenous human captal nvestments, we begn wth the conventonal assumpton that the homogenous agrcultural good s traded at no cost. We relax ths assumpton below to explore further how relatve trade costs across sectors nfluence ncentves for human captal accumulaton. 8 Varetes of traded goods are produced wth an ncreasng returns to scale technology usng a composte of prmary factors of producton (sklled and unsklled labour) and the output of all manufacturng goods (ntermedate nputs). The representatve country frm thus faces the followng cost functon, Γ = w S α (w U ) β G (1 α β) c [F + x ], (5) 7 Whle not explctly modelled here, θ Y may be thought of as capturng the effects of endowments of land, other factors of producton, and the nature of land holdngs on agrcultural productvty. Introducng these other factors of producton more explctly merely complcates the analyss wthout addng any nsght. 8 Our focus s on endogenous human captal nvestments. Davs (1999) examnes how agrcultural trade costs affectthe homemarketeffect, whle Venables and Lmao (2002) consder the relatonshp between relatve trade costs across sectors and ndustral structure. =1

8 Dstance, Skll Deepenng, and Development 7 where c denotes a constant margnal nput requrement; c F s a fxed nput requrement; and x = P R j=1 x j s the total output of the frm produced for all markets. We assume that the composte of prmary factors of producton and ntermedate nputs takes the Cobb-Douglas form, where w S s the wage of sklled workers (wth nput share α), w U s the wage of unsklled workers (wth nput share β), and G s the prce ndex for manufacturng goods from equaton (3) (wth nput share (1 α β)). The parameter c corresponds to an ndex of technologcal effcency that may potentally vary across countres. We assume trade costs take the ceberg form. 9 In order for one unt of atradedgoodtoarrvenlocatonj from locaton, Tj M > 1 unts must be shpped. Thus, when Tj M =1trade s costless, and Tj M 1 measures the proporton of output lost n shppng from to j. We assume the parameter Tj M captures all trade costs between locatons and j, ncludng physcal transportaton costs, nformaton, communcaton, and montorng costs, tarff barrers and non-tarff barrers Human Captal Investment An ndvdual z n country s endowed wth one unt of unsklled labour, whch can be converted nto a unt of sklled labour by ncurrng a fxededucatoncostofω (z) unts of unsklled labour. Denomnatng the educaton cost n terms of unsklled labour captures the dea that real resources are used n the process of becomng sklled. As a result, the cost of educaton s proportonal to the unsklled wage, reflectng the hgher opportunty cost of the real resources used n countres wth a hgher unsklled wage (see also Echer and Garca-Penalosa 2001). The amount of unsklled labour used n becomng sklled depends on two components: Ω (z) =h /a(z). Frst, the parameter h captures the effect of all aspects of the nsttutonal envronment and of government polces 9 We use ceberg trade costs for tractablty. All we requre s that remote locatons face greater trade costs. 10 Hummels (1999) and Lmao and Venables (2001) provde evdence on the mportance of transport costs and the role of geography n determnng ther magntude. See Leamer and Storper (2001) and Venables (2001) for a dscusson of the contnung and sometmes ncreasng mportance of locaton followng the ntroducton of new nformaton and communcaton technologes such as e-mal and the nternet.

9 Dstance, Skll Deepenng, and Development 8 that nfluence ndvduals prvate costs of educaton. These nclude, for example, the extent of publc versus prvate provson of educaton, compulsory schoolng laws and other government regulatons, as well as explct subsdes and grants. The parameter h s an nverse measure of the extent of publc provson, so that hgher values of h correspond to a hgher prvate cost of educaton. Second, an ndvdual s educaton cost depends on ther ablty a(z), and we assume that hgh ablty ndvduals face lower educaton costs. Ths captures the dea that hgh-ablty ndvduals have to work less hard n order to attan a gven level of educaton or are able to complete ther educaton n a shorter perod of tme. Although we have chosen here to make the costs of educaton a functon of ablty, one could equvalently assume that the cost of educaton s the same for all ndvduals but that the rate of return vares wth ablty. 11 We assume that there are upper and lower bounds to ablty determned by human bology and that an ndvdual s ablty s drawn from a dstrbuton over the nterval [a, a]. The probablty densty functon of ablty s denoted by λ(a), so that the mass of ndvduals n country wth a partcular level of ablty a 0 s λ(a 0 ) L. We assume the probablty densty functon of ablty s determned by human bology and s therefore the same n all countres. An ndvdual z n country wll choose to become educated f the wage dfferental between sklled and unsklled workers exceeds educaton costs, w S w U h a(z) wu. (6) We R assume that the cumulatve dstrbuton functon of ablty Λ(a) = a a λ(a)da s contnuous and monotoncally ncreasng n ablty. That s, as we consder successvely hgher and hgher levels of ablty, there wll be fewer ndvduals more able than ths level. The analyss s compatble 11 In ths alternatve specfcaton, the wage w S n frms cost functon s the wage per ablty-adjusted (qualty-adjusted) unt of sklled labour. An ndvdual z of ablty a(z) receves an actual wage of a(z)w S. In the alternatve formulaton, the ablty-adjusted sklled wage s the same for all ndvduals, but actual wages vary wth ndvdual ablty. All of the paper s results are robust to consderng the alternatve specfcaton. It s only necessary to slghtly modfy the condton for an ndvdual to become sklled n equaton (6) to: a(z)w S w U h w U.

10 Dstance, Skll Deepenng, and Development 9 wth a wde range of probablty densty functons for ablty. A partcularly tractable case s where ablty s unformly dstrbuted over the nterval [a, a], nwhchcaseλ(a) =1/[a a] General Equlbrum Consumer Equlbrum Consumers maxmze utlty subject to ther budget constrant. The frst-order condtons mply the followng demand-sde relatonshp between the relatve prce and relatve consumpton of the two goods, p Y j = 1 µ G j µ M j A j, (7) where p Y j s the prce of the agrcultural good and G j s the manufacturng goods prce ndex from above. Equaton (7) determnes country j s consumpton of all manufacturng goods (of the consumpton ndex M j ). Fnal consumpton demand for ndvdual varetes produced n may be derved by applyng Shepherd s Lemma to the manufacturng prce ndex (G j ), m C j =(p M j ) σ E C j G σ 1 j, (8) where Ej C denotes total consumer expendture on manufacturng goods n country j Equlbrum Supply of Sklls The ndvdual s educaton decson compares the wage dfferental between sklled and unsklled workers wth the costs of educaton. Equaton (6) mplctly defnes a crtcal value for ablty a such that all ndvduals wth levels of ablty a(z) a (z ) choose to become sklled. From above, ths crtcal value for ablty s, (S) a = h w S /w U. (9) 1 The margnal ndvdual wth the crtcal level of ablty a s ndfferent between becomng sklled and remanng unsklled, and equaton (9) s therefore termed the skll ndfference condton (S).

11 Dstance, Skll Deepenng, and Development 10 Proposton 1 The equlbrum crtcal level of ablty a above whch ndvduals become sklled s monotoncally decreasng n the relatve sklled wage (w S/wU ) and monotoncally ncreasng n the cost of educaton parameter (h ) Proof. Proposton 1 follows mmedately from the skll ndfference condton (S). Intutvely, as the relatve wage of sklled workers (w S/wU )ncreases,t becomes proftable for ndvduals of lower ablty to nvest n educaton. As the crtcal level of ablty a falls, the equlbrum number of sklled workers ncreases and the equlbrum number of unsklled workers decreases. The equlbrum masses of sklled and unsklled workers are, L = Z a a S = Z a a λ(a) L da Z a λ(a) L da (10) Z a a h a λ(a) L da (11) h S + L + a a λ(a) L da = L (12) Fgure 1 graphs the relatonshp between the relatve wage and the crtcal level of ablty above whch ndvduals become sklled (the rghthand sde of equaton (9)) Producer Equlbrum In the agrcultural sector, proft maxmzaton and constant returns to scale mply that prce equals unt costs of producton f the agrcultural good s produced, p Y =1= 1 θ Y (w S ) φ (w U ) 1 φ, (13) where we choose the agrcultural good for the numerare and hence p Y =1 for all. In the manufacturng sector, the representatve country frm maxmzes the followng proft functon, π = RX j=1 p M j x j T M j w S α (w U ) β G (1 α β) c [F + x ]. (14)

12 Dstance, Skll Deepenng, and Development 11 The frst-order condtons for proft-maxmzaton yeld the standard result that equlbrum prces are a constant mark-up over margnal cost, µ σ w p M S α = (w U σ 1 ) β G (1 α β) c, (15) where p M = p M j /T j M s the free on board or customs prce charged by the frm pror to trade costs. Substtutng ths prcng rule nto equaton (14), we obtan the followng expresson for the equlbrum proft functon, [x (σ 1)F ]. (16) π = µ p M σ In order to break even n a monopolstcally compettve equlbrum, the frm s output must equal a constant: x =(σ 1)F. The prce needed to sell ths many unts s determned by the frm s demand functon, where demand conssts of the sum of fnal consumpton and ntermedate demand across all markets. A frm n country wll therefore sell the quantty x when t charges a prce, 12 (p M ) σ = µ 1 x R X j=1 E j Gj σ 1 T M 1 σ j, (17) where E j = Ej C+EI j denotes total country j expendture (fnal consumpton and ntermedate) on manufacturng goods. Combnng the expresson n equaton (17) wth the fact that, n equlbrum, prces are a constant mark-up over margnal cost, we obtan the followng zero-proft condton, µ σ σ µ (W) w S α σ 1 (w U ) β G (1 α β) 1 X R c = E j G σ 1 j T M 1 σ x j.(18) Ths relatonshp s termed the wage equaton (W). It pns down the maxmum wages of sklled and unsklled workers that a frm n country can afford to pay, gven demand for ts products (as captured n the summaton on the rght-hand sde of the equaton), and gven the cost of ntermedate nputs (as captured n the manufacturng prce ndex on the left-hand sde of the equaton). 12 The transport cost term (T j )enterswthexponent1 σ and not σ because total shpmentstomarketj are T j tmesquanttesconsumed. j=1

13 Dstance, Skll Deepenng, and Development Market Clearng Condtons Factors are relatvely mmoble nternatonally, and we therefore make the standard trade theory assumpton of factor moblty across sectors wthn a country and mmoblty across countres. General equlbrum requres that each country s labour market clears, S Y + S M = S (19) L Y + L M = L (20) where {S M,L M }and{s Y,LY } denote sklled and unsklled employment n the manufacturng and agrcultural sectors respectvely. The total supples of sklled and unsklled labour {S,L } are determned accordng to equatons (10) and (11) above. In equlbrum, we also requre goods markets to clear at the world level, for manufacturng varetes and the homogeneous agrcultural good. 3. Geography and Skll Deepenng The full general equlbrum of the model combnes consumer optmzaton, educaton optmzaton, and producer optmzaton wth the market clearng condtons to solve for equlbrum prces, equlbrum expendtures, and the equlbrum locaton of producton. In ths secton, we use structural equatons of the model to characterze the nature of the relatonshp between locaton and ncentves to nvest n sklls that must hold n general equlbrum. We follow Reddng and Venables (2001) n usng the model to derve theory-consstent measures of a country s locaton relatve to ts markets and sources of supply. We then go on to demonstrate how market access and suppler access nfluence ncentves to nvest n human captal acquston. We begn by combnng fnal consumpton demand (from equaton (8)) andntermedatedemandtoobtananexpressonforblateraltradeflows between countres and j. Expressng ths relatonshp n aggregate value terms, yelds, (T) n p x j = n p 1 σ T M j 1 σ Ej G σ 1 j. (21) In ths gravty equaton (the trade equaton (T)), blateral exports depend on three sets of consderatons. Frst, on a measure of demand n the

14 Dstance, Skll Deepenng, and Development 13 mportng country j termed market capacty (m j E j Gj σ 1 )andcomprsed of total expendture on manufacturng goods n market j (E j )aswellas the number of competng frms and the prces they charge as summarzed n the manufacturng prce ndex (G j ). Second, on a measure of supply potental n the exportng country termed supply capacty (s n p 1 σ ) and comprsed of the number of manufacturng frms (n ) together wth the prces they charge (p ). Thrd, on blateral trade costs (Tj M)1 σ. For each exporter, we may sum market capactes n the mporters t serves, weghtng by blateral trade costs. Ths yelds a measure of the country s overall access to markets - market access (MA ), RX RX MA (Tj M ) 1 σ E j Gj σ 1 = (Tj M ) 1 σ m j. (22) j=1 j=1 Smlarly, for each mporter j, we may sum supply capactes n the exporters that t receves goods from, weghtng by blateral trade costs. Ths yelds a measure of the country s overall access to sources of supply - suppler access (SA j ), SA j RX RX (Tj M ) 1 σ n p 1 σ = (Tj M ) 1 σ s. (23) =1 =1 From the trade equaton (T), market and suppler access may be constructed from nformaton on blateral trade flows. We now show how the wage equaton (W), whch pns down the maxmum wages of sklled and unsklled workers that frmsneachlocatoncanafford to pay, may be wrtten as a functon of market and suppler access. Takng the manufacturng prce ndex over to the rght-hand sde of (W) and usng the defnton n equaton (22), we have, w S α (w U ) β = ξ 1 (MA ) 1 (α+β 1) σ G, (24) c where ξ absorbs earler constants. Now note from equatons (3) and (23) that the manufacturng prce ndex (G ) may be wrtten as a functon of a country s suppler access alone, 1 (P) G =[SA ] 1 σ. (25)

15 Dstance, Skll Deepenng, and Development 14 Combnng the expresson for the prce ndex (P) wth equaton (24), the maxmum sklled and unsklled wages that a frm n locaton can afford to pay can be wrtten as, (W ) (w S ) α (w U ) β = ξ 1 c (MA ) 1 σ (SA ) (1 α β) (σ 1). (26) Intutvely, countres whose locatons provde easy access to supples of manufacturng goods (a hgh value of SA ) are characterzed by low values of the manufacturng prce ndex. Ths tself reduces unt costs of producton and ncreases the maxmum wages that manufacturng frms n those locatons can afford to pay. If a country s locaton also provdes easy access to markets for manufacturng goods (a hgh value of MA ), ths ncreases the free on board prce that manufacturng frms can charge for ther products whle stll sellng enough output to cover the fxed costs of producton, thereby agan ncreasng the maxmum wages that the frms can afford to pay. Consder a country that s ncompletely specalzed n agrculture and manufacturng. We establsh below the condtons under whch ths occurs. 13 The zero proft condtons n agrculture (13) and manufacturng (26) together mplctly defne the equlbrum wages of sklled and unsklled workers. Manpulatng these zero proft condtons and combnng them wth the skll ndfference condton (S), we are able to completely characterze the equlbrum relatonshp between geographcal locaton and endogenous human captal nvestments. Takng logarthms and totally dfferentatng each zero proft condton, we have, 0=φ dws w S +(1 φ) dwu w U (27) 13 As n the standard economc geography model, ncomplete specalzaton wll occur for relatvely hgh values of trade costs and, emprcally, we fnd that countres produce both agrculture and manufacturng. It s straghtforward to also examne the complete specalzaton case. For a country completely specalzed n manufacturng, the relatve wage wll be determned by combnng the relatve supply of sklled workers from Proposton 1 wth a relatve demand relatonshp derved from the manufacturng wage equaton. Ths relatve demand relatonshp s a functon of market and suppler access, and hence geographcal locaton agan plays an mportant role n determnng relatve factor prces.

16 Dstance, Skll Deepenng, and Development 15 α dws w S + β dwu w U = 1 σ dma MA + (1 α β) dsa (28) (σ 1) SA Proposton 2 Suppose a country becomes more remote n the sense that equlbrum market and suppler access fall (dma /M A = dsa /SA = γ < 0). If manufacturng s skll-ntensve relatve to agrculture and the country remans ncompletely specalzed, the new equlbrum must be characterzed by a lower relatve wage of sklled workers. Proof. See Appendx. Intutvely, a fall n market and suppler access n the manufacturng zero proft condton acts exactly lke a fall n the prce of the skll-ntensve good n the Stolper-Samuelson Theorem of Heckscher-Ohln trade theory. A fall n market and suppler access leads to a volaton of the manufacturng zero proft condton at ntal equlbrum factor prces and results n a declne n the sze of the manufacturng sector. The declne n manufacturng releases relatvely more sklled labour than s demanded n agrculture at ntal equlbrum relatve factor prces. Hence, at the new equlbrum, the nomnal sklled wage s lower, the nomnal unsklled wage s hgher (so that the agrcultural zero proft condton contnues to be satsfed), and these together mply that the relatve wage of sklled workers s lower. A lower relatve wage of sklled workers unambguously reduces the ncentve to nvest n sklls. Hence, as a country s equlbrum values of market and suppler access fall, the number of sklled workers falls and the number of unsklled workers rses. Proposton 3 Suppose a country becomes more remote n the sense that equlbrum market and suppler access fall (dma /M A = dsa /SA = γ < 0). If manufacturng s skll-ntensve relatve to agrculture and the country remans ncompletely specalzed, the new equlbrum must be characterzed by: (a) a hgher crtcal level of ablty a above whch ndvduals become sklled (b) a reduced supply of sklled workers and an ncreased supply of unsklled workers Proof. See Appendx. Equlbrum relatve wages and employment n the two sectors for gven levels of market and suppler access are shown graphcally n Fgure 2. Ths

17 Dstance, Skll Deepenng, and Development 16 s the drect analogue of the soprce lne representaton of equlbrum n the Heckscher-Ohln trade model (the dual of the Lerner dagram). The more steeply-sloped curve s the unt cost functon n agrculture, whch must equal prce n an equlbrum where agrculture s produced. The more shallow-sloped curve s the margnal cost functon n manufacturng (b = (w S)(wU )G(1 α β) c ), whch s drawn n sklled wage-unsklled wage space for a gven level of market and suppler access. The level of suppler access 1 pns down the value of the manufacturng prce ndex (G =[SA ] 1 σ ). The manufacturng wage equaton (W ) mples that, n an equlbrum where manufacturng goods are produced, margnal costs are proportonal to market access (b = ξ(ma ) 1 σ ). A reducton n market and suppler access corresponds to an nward shft n the manufacturng soprce lne (less value added s avalable to remunerate the factors of producton). From Fgure 2, the new equlbrum must be characterzed by a lower sklled wage and hgher unsklled wage. The manufacturng producton technology mples that margnal and average relatve unt factor nput requrements are the same. Hence, the slope of each soprce lne corresponds to relatve employment of sklled and unsklled labour by a representatve frm n that sector. In order for both manufacturng and agrculture to be produced n equlbrum, we requre that, at the equlbrum relatve factor prces (ŵ S, ŵu ), the slope of a lne ndcatng the endogenous relatve supply of sklled and unsklled workers (S,L ) les n between the tangents to the two soprce lnes. In the full general equlbrum of the model, market and suppler access are endogenously determned by the dstrbuton of producton and expendture across locatons. Propostons 2 and 3 characterze relatonshps that hold n the full general equlbrum. They explot the equlbrum structure of producton and the supply of sklls to characterze the relatonshp between market access, suppler access, and human captal nvestments when countres are ncompletely specalzed. 14 In our emprcal work, we use the trade equaton (T) to measure market and suppler access from blateral trade data. We take as gven the locaton of producton and expendture, 14 Agan, note the analogy wth the Stolper-Samuelson Theorem. In general, relatve goods prces are endogenously determned. The Stolper-Samuelson Theorem explots the equlbrum structure of producton to derve a relatonshp between relatve goods and relatve factor prces when countres are ncompletely specalzed.

18 Dstance, Skll Deepenng, and Development 17 as revealed by blateral trade, and examne to what extent cross-country nvestments n human captal are consstent wth the equlbrum relatonshp predcted by the model. To keep the analyss as clean as possble and to facltate comparson wth the standard economc geography model, we have so far assumed that the agrcultural good s freely traded. In the next Secton, we relax ths assumpton. 4. Equlbrum wth Trade Costs n Manufacturng and Agrculture To ntroduce agrcultural trade costs n as tractable way as possble, we modfy consumers utlty functons slghtly. Specfcally, we assume that each country produces a sngle dfferentated agrcultural good. 15 These dfferentated goods enter a consumpton ndex A j n equaton (1) that takes the Dxt-Stgltz form, " R # σ/(σ 1) X A j =, σ > 1. (29) =1 A (σ 1)/σ j The dfferentaton of goods across countres may be nterpreted n two ways, whch for our purposes are equvalent. Frst, there are a number of dfferent agrcultural goods (e.g. wheat, corn, barley, maze) whch are mperfect substtutes for one another, and each country completely specalzes n a dfferent agrcultural good due to Rcardan dfferences n technology or unmodelled varaton n land endowments (as n Davs 1997; see also Deardorff 1998 and Krugman and Venables 2001). Second, there s Armngton dfferentaton by country of orgn. The frst s the more plausble nterpretaton and the one taken here. In order for one unt of an agrcultural good to arrve n locaton j from locaton, we assume that T A j > 1 unts must be shpped, so that T A j 1 s a 15 It s possble to analyze the role of agrcultural trade costs whle retanng the assumpton that agrcultural goods are homogenous. Ths complcates the analyss because, wth an homogenous product and trade costs, exporters of agrculture wll not generally export to all locatons and mporters wll not generally mport from all locatons (see for example Venables and Lmao 2002). As dscussed further below, one nterpretaton of the specfcaton here s that countres are specalzed n dfferent agrcultural goods whch are mperfect substtutes for oneanother.

19 Dstance, Skll Deepenng, and Development 18 measure of agrcultural trade costs. Wth perfect competton and constant returns to scale, the free on board (fob) prce of agrcultural goods n each locaton wll equal average cost, whle the cost nclusve of freght prce (cf ) charged n an mportng locaton j wll be a mark-up over average cost wth the sze of the mark-up determned by blateral trade costs, p Y j T A j = p Y = 1 θ Y (w S ) φ (w U ) 1 φ, (30) and we choose the cf prce of country s agrcultural good n one locaton (for example, j =1)asthenumerare(sothatp Y 1 =1and py =1/T1 A). The manufacturng zero proft condton (26) contnues to depend on market and suppler access, whle agrcultural trade costs (Tj A ) now enter drectly nto the agrcultural zero proft condton (30). To examne the effects of ncreased remoteness, we agan totally dfferentate the two zero proft condtons, assumng that a country remans ncompletely specalzed n agrculture and manufacturng, dt A 1 T A 1 = φ dws w S +(1 φ) dwu w U, (31) α dws w S + β dwu w U = 1 σ dma MA + (1 α β) dsa, (32) (σ 1) SA wherewehaveusedourchoceofnumerare(p Y =1/T1 A ) and manufacturng s assumed to be skll ntensve relatve to agrculture (α/β > φ/(1 φ)). Suppose that the country experences an equproportonate ncrease n the value of agrcultural and manufacturng trade costs to all locatons: dtj M/T j M = dtj A/T j A = γ>0. The ncrease n agrcultural trade costs (Tj A ) enters drectly nto the agrcultural zero proft condton and drectly shfts the soprce lne for ths sector nwards n Fgure 2. For gven values of producton and expendture n each locaton (.e. for gven values of market capacty, m j, and supply capacty, s ), the ncrease n manufacturng trade costs (Tj M ) reduces market and suppler access (snce σ>1n equatons (22) and (23)). As a result, the manufacturng soprce lne also shfts nwards n Fgure 2. Though nomnal wages both fall (n terms of the numerare), the effect of these shfts on the relatve wage and human captal accumulaton appears ambguous.

20 Dstance, Skll Deepenng, and Development 19 Ths analyss yelds an mportant nsght. When both manufacturng and agrculture face trade costs, the effect of ncreased remoteness on skll accumulaton depends upon both relatve skll ntensty and relatve trade costs. If, as n Secton 3, trade costs are more mportant n the skll ntensve sector, remote locatons wll experence reduced ncentves to nvest n skll. However, the same outcome can emerge even f ad valorem trade costs are lower n the hgh skll sector due to asymmetres n the affect of trade costs on agrculture and manufacturng. Frst, though shppng costs may be lower for skll-ntensve sectors, other trade costs, ncludng all search, communcaton, and montorng costs, are lkely to be relatvely hgh n these sectors. 16 Second, a gven level of ad valorem trade costs has a stronger affect on manufacturng than agrculture because nput-output lnkages requre frms to ncur trade costs on both mported ntermedate mports as well as exports. Indeed, because ntermedate nputs account for a substantal proporton of costs, even relatvely small trade costs can become large as a proporton of value added. The relatve mportance of trade costs n manufacturng can be seen by notng that the change n trade costs enters once n the agrcultural zero proft condton (31) but twce n the manufacturng zero proft condton (32) va ts effect on both market and suppler access. 17 Thrd, skll-ntensve manufacturng s ncreasng returns to scale, whle low-skll agrculture s constant returns to scale. The presence of ncreasng returns to scale n manufacturng means that market sze s mportant: n equlbrum, frms must sell enough unts of output n order to cover the fxed costs of producton. Hence, trade costs do not enter the manufacturng zero proft condton (32) drectly, but nstead enter through market and suppler access, whch respectvely weght market sze and supply capacty n all of a country s trade partners by blateral trade costs. In agrculture n contrast, the presence of constant returns to scale means that t s perunt trade costs whch are mportant, and these enter drectly nto the 16 New technologes (e.g. the nternet) may reduce trade costs n some skll ntensve sectors. One promnent example s the emergence of Bangalore as a software programmng centre n Inda. See Leamer and Storper (2001) and Venables (2001) for further dscusson of geography and new technologes. 17 See Radelet and Sachs (1998) for further dscusson of how trade costs whch are small as a share of gross output can have very large effects on value-added.

21 Dstance, Skll Deepenng, and Development 20 agrcultural zero proft condton. The mportance of ths dstncton can be seen by holdng a country s blateral trade costs constant but movng producton and expendture from neghborng to dstant locatons. Ths wll have no effect n the agrcultural zero proft condton where only per unt trade costs matter (equaton (31) where dtj A/T j A =0). However, reducng market and supply capacty at neghborng locatons and ncreasng them by the same amount at dstant locatons (.e. ncreasng a country s economc remoteness rather than ts geographcal remoteness) unambguously reduces a country s market and suppler access (.e. dma /M A < 0 and dsa /SA < 0 n the manufacturng zero proft condton (32)). As above, the fall n market and suppler access reduces the relatve wage and equlbrum supply of sklled workers f manufacturng s skll ntensve relatve to agrculture. Fnally, n the full general equlbrum of the model, changes n agrcultural and manufacturng trade costs wll themselves nfluence the dstrbuton of producton and expendture across locatons (of market capacty, m j, and supply capacty, s ), wth the resultng changes n economc remoteness nfluencng the skll premum and ncentves to nvest n human captal n the way dscussed above. The mportance of nput-output lnkages and ncreasng returns to scale n skll-ntensve manufacturng suggests that our earler fndng that remote countres have lower ncentves to nvest n human captal accumulaton carres over to a world where trade costs are pad on both agrcultural and manufacturng goods. The analyss also suggests that reductons n trade costs n relatvely skll-ntensve sectors (through, for example, trade lberalzaton) may be partcularly mportant n elevatng human captal nvestments n perpheral countres. 5. Other Determnants of Human Captal Investment The dscusson so far has emphaszed the mportance of geographcal locaton for ncentves to nvest n human captal. In ths secton, we examne the effects of changes n other parameters of the model whch are related to potental determnants of human captal nvestment emphaszed n the exstng lterature. To solate the effects of these other varables, we consder the effect of parameter changes holdng constant a country s mar-

22 Dstance, Skll Deepenng, and Development 21 ket and suppler access. We contnue to assume countres are ncompletely specalzed n agrculture and manufacturng (the condtons for whch are derved above). Proposton 4 The crtcal level of ablty a above whch ndvduals become sklled s monotoncally ncreasng and the equlbrum supply of sklled workers S s monotoncally decreasng n (a) productvty n agrculture θ Y (b) the cost of manufacturng producton parameter c (c) the cost of educaton parameter h Proof. See Appendx. Intutvely, ncreases n agrcultural productvty θ Y act lke a rse n the prce of the agrcultural good. By analogy wth the Stolper-Samuelson Theorem, an ncrease n agrcultural productvty reduces the relatve sklled wage, and hence reduces equlbrum human captal nvestments. Our theoretcal framework, therefore, formalzes the dea that a productve agrcultural sector or, more generally, an abundance of agrcultural land and other natural resources may both hnder the development of manufacturng and mpede nvestments n human captal (see for example the analyss of Latn Amerca n Engerman and Sokoloff 1997 and Leamer et al 1999). The model also captures the dea that technology and, n partcular, the transfer of technology from advanced countres s mportant for economc development. Technology transfer that reduces manufacturng producton costs, c, rases the maxmum sklled and unsklled wage that a manufacturng frm n country can afford to pay gven market and suppler access. In terms of Fgure 2, the manufacturng soprce lnes shfts outwards away from the orgn. Snce manufacturng s skll ntensve relatve to agrculture, ths ncreases the relatve wage of sklled workers, and hence rases equlbrum human captal nvestments. Thus, there s an mportant general equlbrum complementarty between technology and sklls. The transfer of technology to skll ntensve manufacturng ndustres n developng countres not only drectly rases output per capta but also has postve ndrect effects through nduced hu-

23 Dstance, Skll Deepenng, and Development 22 man captal nvestment. 18 The ncentve to transfer technology depends, n part, on a country s nsttutons and polcy envronment (see for example Acemoglu et al 2001). Hence, the analyss s consstent wth an mportant effect of nsttutons on the process of economc development and wth a complementarty between nsttutons and human captal nvestment. Fnally, nsttutons are also mportant va ther effects on the cost of educaton parameter, h. Increases n the extent of publc versus prvate provson of educaton, or changes n government regulatons that reduce the prvate costs of educaton, h, wll ncrease equlbrum nvestments n human captal. As the supply of sklled workers rses, output of the skllntensve manufacturng sector wll also rse. 6. Emprcal Measurement of Market and Suppler Access Usng blateral trade flow data compled by Feenstra et al (1997), we construct theoretcally consstent measures of market access and suppler access for all countres at fve year ntervals from 1970 to 1995 usng equatons (22) and (23). To ensure that these measures are not drven by small countres that trade very lttle wth the rest of the world, we restrct our sample to the 137 countres that trade wth at least 5 partners. From the trade equaton (T), the model predcts that blateral trade flows depend upon exportng country characterstcs (.e. supply capacty, s n p 1 σ ), mportng partner characterstcs (.e. market capacty, m j E j G σ 1 j ), and blateral transportaton costs (Tj M). We use country dummy varables (denoted by d and d j, respectvely) to capture market and supply characterstcs for each par of countres and j. Ths has the advantage of controllng for all observed and unobserved varables that affect market and supply capacty. 19 The dummy varables also capture any component of transport costs or trade polcy that s common across all of a partcular country s export partners and mport supplers. We model the blateral component of transportaton costs as dependng upon dstance (dstance j ) and whether or not two countres share a common 18 See Reddng (1996) for an analyss of technology-skll complementarty wthn ndustres n advanced countres. 19 In partcular, the dummes capture the role of the manufacturng prce ndex, G j. They control, therefore, for what Anderson and Van Wncoop (2001) term multlateral resstance (a country s average trade barrer wth all partners).

24 Dstance, Skll Deepenng, and Development 23 border (border j ). Dstance s the great crcle dstance, n klometers, between the two countres largest ctes. Thus, the emprcal counterpart of equaton (21) s: ln(x jt )=α t +β 1t d +β 2t d j +δ 1t ln(dstance j )+δ 2t border j +u jt.(33) for each tme perod separately, where X jt denotes the value of exports from to j at tme t and u jt s a stochastc error. Table 1 presents the results of estmatng ths equaton on the sample of non-zero trade flows, by year. The dstance and common border varables have the expected sgn and are statstcally sgnfcant at the 1% level. The null hypothess that the coeffcents on ether the country dummes or the partner dummes are equal to zero s easly rejected at conventonal sgnfcance levels usng a standard F-test, and the model explans over 90% of the cross-sectonal varaton n blateral trade flows. The economc mportance of dstance appears to grow wth tme: whereas a 1% ncrease n dstance s assocated wth a 1.2% reducton n blateral exports n 1970, t s assocated wth a 1.5% reducton n exports by Estmated market access ( g MA t ) for exporter and suppler access ( f SA jt )formporterj can be constructed usng coeffcent estmates from equaton (33) and equatons (22) and (23) from Secton 3.: gma t = fsa jt = RX j=1 RX =1 β d b 2t j (dstance j ) b δ 1t (border j ) b δ 2t β d b 1t (dstance j ) b δ 1t (border j ) b δ 2t Emprcally, market and suppler access are hghly correlated. As a result, we use MA g t as a proxy of countres dstance from world economc actvty n our emprcal analyss below. Very smlar results are obtaned f SA f jt s used nstead. gma t and SA f jt have a number of advantages compared to tradtonal estmates of dstance. Most mportant, they derved from a general equlbrum model of nternatonal trade that ncorporates economc geography. Second, they rely upon blateral trade data to uncover revealed access to markets and are thereby able to ncorporate the effects of unobservable

25 Dstance, Skll Deepenng, and Development 24 transportaton costs, trade barrers, and determnants of market and supply capacty. Fnally, they capture n a sngle measure several dmensons of physcal dstance. Table 2 reports the results of regressng MA g t on countres great crcle dstance from the US, Japan and Belgum n klometers. Dstance from these three economc centers explans 90% of the varaton n market access. Of the three locatons, market access s most negatvely correlated wth dstance from Brussels; coeffcent estmates mply that a 1% ncrease n a country s dstance from that Brussels reduces market access by 0.77%. Fgure 3 provdes a vsual representaton of ths correlaton n Notable outlers n the fgure, ncludng Canada and several Asan countres, are markets whch are located near ether the US or Japan. Comparson of the dstrbutons of market access from 1970 to 1995 ndcates that perpheral countres are becomng more economcally remote over tme. Table 3 reports the ratos of varous percentles of these dstrbutons at fve year ntervals and ndcates that the dstrbutons are characterzed by ncreasng nequalty. All three ratos 90 th /10 th,80 th /20 th and 75 th /25 th ncrease wth tme. Ths trend s nterestng n ts own rght and worthy of further nqury. Potental explanatons nclude an ncreasngly uneven dstrbuton of world economc actvty and the rsng coeffcent on dstance n the trade equaton estmaton. 7. Emprcal Relatonshp Between Geography and Human Captal In ths secton, we check whether the human captal mplcatons of the model are supported by the data. Consstent wth the model, we provde evdence that educatonal attanment s hgher n countres wth greater market access. Data on educatonal attanment for a large cross-secton of developed and developng countres n 1990 s avalable from Barro and Lee (2001). These data record the percent of each country s over-15 populaton that has attaned secondary and tertary educaton. 20 Data on both market access and educatonal attanment are avalable for approxmately 100 countres, dependng upon the year. More detal on the data used n 20 For further nformaton, see Barro and Lee (1993), (2001). Domenech and de la Fuente (2000) provdes a complementary source of nformaton for OECD countres.