FRAGMENTATION, PRODUCTIVITY AND RELATIVE WAGES IN THE UK: A GENERAL EQUILIBRIUM APPROACH * Alexander Hijzen. University of Nottingham

Transcription

1 FRAGMENTATION, PRODUCTIVITY AND RELATIVE WAGES IN THE UK: A GENERAL EQUILIBRIUM APPROACH Alexander Hjzen Unversty of Nottngham Abstract Feenstra and Hanson (1999) propose a two-stop method to analyse the role of outsourcng and skll-based technologcal change (SBTC) n the rse n wage nequalty. Ths paper apples ther methodology to UK manufacturng usng data for the 1990s and extends t n order to obtan addtonal nsght n the relatve mportance of the sector bas and the factor bas of outsourcng and SBTC. The results ndcate that outsourcng has sgnfcantly contrbuted to the rse n the domestc wage nequalty accountng for approxmately 12% of the ncrease n the UK n the 1990s. Factor-based outsourcng was about 2.5 tmes as mportant as sector-based outsourcng n explanng the ncrease n wage nequalty. JEL Classfcaton: F14, J31 Keywords: fragmentaton, outsourcng, productvty, mandated wage regressons Word count: 7,103 (ncludng text, tables, appendx, bblography) I would lke to thank my supervsors Bob Hne and Holger Görg for ther gudance and Doug Nelson, Geoff Reed and partcpants of the ETSG Annual Conference 2002 for helpful comments and suggestons. The usual dsclamer apples. The provson of the QLFS data by the UK Data Archve s hereby acknowledged. Address for correspondence: School of Economcs, Unversty of Nottngham, Unversty Park, NG7 2 RD, Nottngham, Unted Kngdom, lexafh@nottngham.ac.uk

2 One of the man consequences of the process of globalsaton s the ncreasng nternatonal fragmentaton of producton, that s, the splttng up of producton processes nto separate components so that they can be produced n dfferent locatons (Jones and Kerzkowsk, 2001). Internatonal fragmentaton of producton leads to the establshment of nternatonal producton networks, whch are assocated wth trade n ntermedates. In ths paper fragmentaton (or outsourcng) s related to the debate on trade and wages. In a prevous study Görg, Hjzen, and Hne (2001) use a partal equlbrum framework, whch focuses on the factor bas by estmatng the relatve demand for labour. The paper presents mxed results wth respect to the role of trade on domestc wage nequalty. Obvously, a partal equlbrum framework employed does not pck up the expanson and declne of ndustres n response to foregn competton. General equlbrum effects smlarly apply to the mpact of technologcal change and outsourcng. Haskel and Slaughter (2002b) fnd evdence ndcatng that the sectorbas of skll-based technologcal change s mportant. That s, the factor bas of technologcal change or outsourcng matters to relatve factor prces to the extent that relatve ndustry prces are affected. Ths can only be the case when those developments are unequally dstrbuted across ndustres. Thus, a general equlbrum framework s necessary n order to account for both factor and sector bas of varous structural forces. The theoretcal framework to assess the mpact of trade and outsourcng on wages s provded by the Stolper-Samuelson theorem, whch relates relatve ndustry prces to relatve factor prces through the zero-proft condtons. Emprcal studes that employ a general equlbrum approach by drectly applyng the Stolper-Samuelson theorem, generally take the form of so-called mandated wage regressons n whch the change n ndustry prces s regressed on the factor-cost shares n that ndustry. The coeffcents reflect the mpled factor prce changes followng the change n ndustry prces. 1 However, Feenstra and Hanson (1999) n ther study of the effect of outsourcng on US wages argue that f fully specfed the regresson becomes an dentty and can no longer be used to make nferences about the mpled factor prce changes. In order to solve ths problem they propose to endogense prces and total factor productvty (TFP) n a two-stage procedure. In the frst stage ndustry prces and TFP are regressed on expendture on computers and outsourcng. 2 In the second stage the estmated coeffcents from the frst stage regressons are nserted as the dependent varables n the mandated wage regressons. In fact Feenstra and Hanson estmate a reduced-form n whch they jontly estmate the drect and ndrect effect of ther structural varables on value-added prces and TFP. The ndrect effect s the mpact of the structural varables on productvty that s passed-through on value-added prces. Ths specfcaton s nterestng as t ncludes the ndrect effect of outsourcng and SBTC on factor prces, but does not requre a consstent pass-through coeffcent for TFP on prces (TFP and value added prces are correlated by constructon). 1 See Slaughter (2000) for a survey. 2 By ncludng productvty n the prce regresson they effectvely mpose the large country assumpton. 1

3 Haskel and Slaughter (2001) adopt a smlar methodology usng UK data for the 1970s and 1980s, but do only consder trade n fnal goods. They do not allow for pass-through from TFP on prces as the UK s consdered to be small relatve to the world economy. As a result they have a separate set of structural varables for both prces and TFP. Most mportantly, they fnd that prce effects (and not technology) were the man force behnd the ncrease n relatve wages durng the 1980s. However, they do not consder the mpact of foregn competton on productvty, or the mpact of productvty on prces. The contrbuton of the present paper s to provde a detaled analyss of the effects of outsourcng and technology on wages n the UK usng 3-dgt manufacturng data for the perod The use of recent data s thought to be crucal as fragmentaton s consdered to be prmarly a phenomenon of the last decade. The emprcal analyss apples and extends the method ntroduced by Feenstra and Hanson (1999). An mportant nnovaton s that ths paper explctly addresses the relatve mportance of factor and sector bas (or the drect and ndrect effect) of skll-based technologcal change and outsourcng. The sector bas captures the relatve cost-savng effect, whle the factor bas captures changes n total factor-use. Decomposng the total effect of technologcal change or outsourcng n ts factor bas and sector bas s nterestng for three reasons. Frst, the relatve mportance of those effects has been the topc of an ntensve debate n the theoretcal lterature. Leamer (1995) emphasses that n small-open economy the sector bas s all that matters. Krugman (2000) argues that t s justfed to emphasse the factor bas of technologcal change when technologcal change s global. However, no study seems to have addressed the ssue emprcally. Ths paper s the frst to emprcally address ths queston. Second, the reduced-form specfcaton developed by Feenstra and Hanson (1999) s only sensble when the factor bas and the sector bas affect effectve prces n the same drecton. In other words t could be rsky to nterpret the obtaned results wthout further knowledge on the factor bas and the sector bas. Thrd, decomposng the aggregate effect of structural varables allows one to nvestgate the role of outsourcng and foregn competton on productvty growth. The paper s structured as follows. Secton 1 presents a dscusson of the theory lnkng trade and fragmentaton to relatve wages. Specal emphass s gven to the ssue of sector versus factor bas. Secton 2 sets out the methodology employed n the econometrc analyss whle hghlghtng the key dfferences n the methodology used n Feenstra and Hanson (1999) and Haskel and Slaughter (2001). Secton 3 dscusses the data and presents some descrptve statstcs. Secton 4 apples the two-stage method proposed by Feenstra and Hanson, whch nvolves estmatng the mpact of SBTC and outsourcng on the sum of productvty and value-added prces whle controllng for foregn competton and market concentraton. Secton 5 extends the methodology to gan addtonal nsght n the factor and the sector bas, whch requres the estmaton of the two frst-stage regressons for productvty and prces (ncludng TFP growth). In order to account for the endogenety of TFP the regressons are estmated smultaneously usng three-stage least squares. Fnally, Secton 6 concludes. 2

4 1. Theory The theoretcal foundaton for emprcal studes analysng the mpact of trade on wages s provded by the Stolper-Samuelson theorem. Wth many goods and factors the most approprate verson states that for any vector of goods-prce changes, the accompanyng vector of factor-prce changes wll be postvely correlated wth the factor-ntensty-weghted averages of the goods-prce changes. Allowng for productvty growth and accountng for the use of ntermedate nputs yelds: (1.1) pˆ θ pˆ k k k = θ wˆ j j j TFˆ P Equaton (1.1) s derved from the zero-proft condtons equatng prces to average costs resultng from the assumpton of perfect competton. 3 The term on the left-hand sde reflects value-added prces whch equal fnal good prces, p, mnus the sum of the cost shares, θ k, of ntermedate nputs k=1,,k tmes p k. In equlbrum valueadded prces equal the sum of the prmary cost shares, θ j, tmes factor prces, w j wth j=1, J. Hats ndcate proportonal changes. Note that for factor prces the ndustry subscrpt s omtted as factor prces are equalsed throughout the economy due to the assumpton of perfect factor moblty across ndustres. As such an assumpton s only warranted over a suffcently large tme horzon ths should be reflected n the emprcal analyss. In a small-open economy factor prce changes can only result from a change n the relatve proftablty (short-run) across sectors. From (1.1) t follows that the relatve proftablty can change as a result of ether a change n relatve prces or productvty growth. In a small-open prce-takng economy domestc prce changes are solely due to changes n world prces. By the factor-prce nsenstvty theorem (Leamer, 1995) changes n relatve factor endowments leave relatve factor prces unaffected. The change n factor endowments wll nstead be accommodated by a magnfed change n output quanttes. The mplcatons of ths theorem are far-reachng. Any change affectng the relatve demand or relatve supply for prmary factors, whether due to skll-based technologcal change, fragmentaton of producton or an ncrease n the relatve supply of sklled labour, also leave relatve factor prces unaffected as long as those changes do not affect the relatve proftablty across sectors. The fact that even wthn a small open economy SBTC and fragmentaton are generally not neutral n ther effect on relatve factor prces comes entrely from ther sector bas of the assocated savng n producton costs (reflected by hgher productvty growth). At ths pont t s useful to clarfy some of the termnology that wll be used throughout ths paper. It was stated that n a small-open economy everythng that matters s the relatve proftablty across sectors. Technologcal change (whatever ts 3 In the presence of mperfect competton the Stolper-Samuelson theorem remans vald n two cases. Frst, when the market s charactersed by monopolstc competton the zero-proft condtons are stll satsfed. Second, even when the zero-proft condton are no longer satsfed the Stolper-Samuelson theorem remans vald as long as mark-ups are constant over tme. Haskel and Slaughter (2002a) fnd suggestve evdence that changes n ndustry-specfc rents account for 15% of the changes n wages. 3

5 source) only matters to the extent that t changes total factor costs between sectors (cost-savng effect). Ths wll be referred to as the sector bas of technologcal change. In Jones (1965) ths was called the 'dfferental sector effect'. In the 2x2 case the cost change, γ, n ndustry =1,2 as a result of technologcal change can be represented as n equaton (1.2) where l stands for labour, k for captal and θ reflects the respectve factor cost shares. (1.2) γ = θ cˆ + θ cˆ l l k k However technologcal change may also, as does a change n endowments, alter factor market equlbrum. Jones referred to ths as the 'dfferental factor effect' and n ths paper t wll be referred to as the factor bas of technologcal change. Formally for factors j=l, K, equaton (1.3) gves the total change n factor use. (1.3) j cˆ λ cˆ j1 j1 j 2 j 2 γ = λ + Equaton (1.3) s derved by fully dfferentatng the full-employment condtons equatng total endowments of factor j to the sum of ts use across all sectors (unt nput requrements tmes output). Changes n factor endowments or technologcal change can only be accommodated by adjustments n the relatve sze of ndustres. In the small-open dversfed economy any change n relatve demand wll leave relatve factor prces unaffected. Changes n relatve outputs do not affect goods prces as those are exogenously determned by world prces. The relatve factor demand curve s therefore nfntely elastc. The relatve mportance of the factor bas versus sector bas s subject to consderable. dsagreement. Leamer (1995, 1998) emphasses that n small-open economy the sector bas s all that matters. Krugman (2000) argues that t s justfed to emphasse the factor bas of technologcal change when technologcal change s global. Several ssues matter for the mportance of factor bas. Frst of all, t depends on the relatve sze of the country compared to world markets. Related s the argument presented by Krugman (2000) that technologcal change (or any other structural force) at home and abroad has a smlar effect as technologcal change n a closed economy. In a closed economy t s solely the factor bas of technologcal change that determnes what happens to relatve factor prces. In addton, the relevance of factor bas depends on the producton technology (Xu, 2001), and the relatve sze of the non-tradables sector where goods prces are endogenous by defnton. Second-order effects may be mportant whenever technologcal change s fnte (Leamer, 1998; Fndlay and Jones, 2000). Fnally, the factor bas matters whenever countres are completely specalsed. The focus of the present paper however s on trade n ntermedates resultng from the ncreasng nternatonal fragmentaton of producton processes rather than trade n fnal goods. The lterature on nternatonal fragmentaton s predomnantly charactersed by perfect competton. Fragmentaton n those models s generally drven by the presence of cross-border dfferences n relatve factor prces. Fragmentaton takes the form of movng unsklled ntensve manufacturng processes 4

6 from a developed country to a developng country. It s argued that fragmentaton has a smlar effect as skll-based technologcal change. Arndt (1997) analyses the mpact of fragmentaton n a small open developed economy n a standard 2x2x2 Heckscher-Ohln model. Consequently, Arndt emphasses the sector bas of outsourcng. He concludes that outsourcng of labourntensve components n the labour-ntensve ndustry actually reduces wage nequalty whereas outsourcng of labour-ntensve components n the captal/skllntensve ndustry ncreases wage nequalty. 4 Deardorff (2001) analyses fragmentaton across cones,.e. n the absence of factor prce equalsaton. He argues that the mpact of fragmentaton on relatve factor prces depends crucally on the relatve factor-ntensty of the fragment beng moved abroad and the average factor ntensty n the economy. The adjustment of the economy dependng on the relatve factor-ntensty of the fragment and the average factorntensty of the economy s reflected n a change n relatve factor prces. The dfference wth Arndt resdes n the fact that Arndt consders an open-dversfed economy, whle Deardorff consders a completely specalsed economy. 5 One can conclude that the debate on the relevance of factor bas extends to trade n ntermedates. However, t should be noted that the case for an mportant role for the factor bas s stronger when trade n ntermedates s concerned. Fragmentaton s only vable f t brngs suffcent savngs n factor costs to cover the cost of fragmentaton. Thus t requres the persstence of factor prce dfferentals (the lack of nternatonal factor prce equalsaton). Furthermore, the mpact of outsourcng on the producton technologes s unlkely to be margnal but nstead s expected to be qute radcal. Therefore t seems that n order to assess the mpact of fragmentaton on relatve wages one has to account both for ts factor and ts sector bas Methodology In order to nvestgate more formally the lnk between fragmentaton, productvty and factor prces a two-stage methodology based on Feenstra and Hanson (1999) and Haskel and Slaughter (2001) s employed. Feenstra and Hanson (1999) argue that estmatng equaton (1.1) n a fully specfed regresson yelds an dentty and cannot be used to make nferences about the mpled factor prce changes. In order to solve ths problem Feenstra and Hanson propose to endogense prces and total factor productvty (TFP). They therefore develop a twostage procedure. In the frst stage ndustry prces and TFP are regressed on expendture on computers and outsourcng. In the second stage the estmated 4 Jones and Kerzkowsk (2001) confrm these possbltes, but also stress the radcal nature of outsourcng for whch the Heckscher-Ohln framework may be ll equpped. A pror, therefore, t s very dffcult to predct how fragmentaton wll actually affect relatve wages. 5 Consequently, n the analyss conducted by Arndt relatve factor demand s nfntely elastc and only the sector bas matters. In Deardorff's analyss the relatve demand curve s downward-slopng. 6 The net cost savng effect appears n the data as productvty growth. In all lkelhood ths productvty growth has often been attrbuted to domestc SBTC nstead of outsourcng. 5

7 coeffcents from the frst stage regressons are nserted as the dependent varables n the mandated wage regressons. Haskel and Slaughter (2001) adopt a smlar methodology usng UK data for the 1970s and 1980s. The essental dfference resdes n the scope of pass-through from TFP to prces n the frst stage of the regresson analyss. Feenstra and Hanson allow for such a pass-through whereas Haskel and Slaughter do not. The dea of passthrough s plausble whenever the factor bas plays a role. 7 Haskel and Slaughter (2001) separately estmate a prce and a TFP-regresson n the frst-stage. (2.1a) ln p = δ ' z + ω (2.1b) lntfp = α ' + ε z When productvty pass-through s allowed for ths s reflected n the prce regresson where the change n value-added prces s regressed on TFP plus a vector of structural varables. Feenstra and Hanson (1999) therefore start off wth the followng set of equatons n the frst stage: (2.2a) ln p = λ ln TFP + β ' z + υ (2.2b) lntfp = α ' + ε z Whenever λ s negatve one should allow for pass-through from TFP to prces. 8 Includng TFP n the prce regresson (2.2a) accounts for the pass-through of the sector-based technologcal change. Its effect on goods prces equals λα, whle ts effect on factor prces equals (1+λ)α. The factor bas comes nto play when output affects goods prces. If one beleves n pass-through one should also account for the factor bas. The factor bas s gven by β. Thus, the total effect of technologcal change and outsourcng (factor and sector bas) on goods prces equals λα +β, whle ts effect on factor prces equals (1+λ)α+β. Note that the factor bas only matters to the extent that technologcal change drectly affects prces! Note that prce effects may also be due to exogenous forces affectng prces wthout affectng TFP (such as reductons n tarff and transportaton costs). 7 When allowng for prce-effects (large country assumpton, pass-through and factor bas) sectorbased TFP only has an mpact on factor prces when pass-through s ncomplete. In the presence of complete pass-through sector-based TFP growth leaves relatve factor prces unaffected. 8 Both Feenstra and Hanson (1999) and Haskel and Slaughter (2001, 2002a) estmate that the passthrough rate s sgnfcantly dfferent from zero. Applyng the F-test for λ=0 n (2.2a) s rejected, whereas the F-test of λ= -1 cannot be rejected (95% confdence nterval). However as Baldwn and Can (2000) note the emprcal measurement of TFP s endogenous to the prce settng process. 6

8 In fact Feenstra and Hanson estmate a reduced-form n whch they jontly estmate the drect and ndrect effect of the structural varables on value-added prces and TFP reflected by (2.3), whch s obtaned by addng TFP to both sdes of (2.2a). The ndrect effect s the mpact of the structural varables on productvty that s passedthrough on value-added prces. (2.3) ln p + ln TFP = γ ' z + η where γ=(1+λ)α+β and η =(1+λ)ε +υ. Feenstra and Hanson thus regress several structural varables on effectve prces,.e. the sum of prce changes and TFP growth. The advantage of ther specfcaton s that they only have to estmate a sngle parameter, γ. Addng TFP to both sdes of the equaton sngles out the net effect of technologcal change on relatve factor prces. However, t would be nterestng to decompose the drect effect (factor bas) and ndrect effect (sector bas) of factors such as outsourcng and SBTC on wage nequalty. Not only does ths contrbute to the academc debate, but t may also have some nterestng polcy mplcatons. Decomposng drect and ndrect effects nvolves the estmaton of the three parameters α, λ and β n equatons (2.2a) and (2.2b). The ncluson of TFP as an explanatory varable n equaton (2.4b) mples that OLS estmates would suffer from smultanety bas. One has to smultaneously estmate both equatons n order to deal wth the endogenety problem assocated wth TFP (TFP and error term wll be correlated) n (2.2b). The problem can be overcome by usng three-stage least squares (3SLS). In the second-stage the components explaned by each structural varable, k, n the frst-stage are regressed on the average factor shares, V j. Note that γ n equaton (2.4) ether reflects γ n (2.3) or the combnaton of α, λ and β from (2.4a) and (2.4b) as dscussed above. (2.4) γ ' z = δ ' V + ν k k k 3. Data and Descrptve Statstcs The labour market data are obtaned from Labour Force Survey (LFS). The LFS allows one to construct numerous skll measures. In ths paper skll s defned on the bass of the Standard Occupatonal Classfcaton (SOC), whch allows one to construct a more accurate measure of skll than the one based on the dstncton between manual/non-manual workers generally used n the lterature. In the QLFS workers are classfed accordng to 9 Major Groups. The SOC Major Groups are based on qualfcatons, tranng, sklls, and experence. Therefore, dstngushng skll groups on the bass of ther Major Group Codes allows one to construct a very accurate measure of skll. For the determnaton of skll groups the approach taken by Gregory, Zssmos and Greenhalgh (2001) s adopted. Apart from provdng a more 7

9 accurate measure of skll, ths approach allows one to dstngush three skll groups: sklled, ntermedate, and unsklled. 9 Producton data are obtaned drectly from the ONS. Data on R&D ntensty and outsourcng are obtaned from the Input-Output Tables. The trade data are obtaned from EUROSTAT. The perod under consderaton s lmted to The year 1993 s the frst for whch data are classfed accordng to SIC92 s Followng Feenstra and Hanson (1999) two measures of outsourcng wll be used: narrow (to the same ndustry) and dfferental outsourcng (to other ndustres). In order to emphasse outsourcng of low-skll fragments drven by the persstence of factor cost dfferentals, outsourcng s weghted by mports from developng countres. SBTC s measured by R&D ntensty. Foregn competton s measured by mport prces based on unt values. Fnally, concentraton ratos are used to control for ndustry characterstcs such as market concentraton. 10 Table 3.1 represents summary statstcs wth respect to factor prces and factor cost shares. Changes n factor prces are measured as the average annual change n the log of factor earnngs. Factor cost shares reflect the average of the start and end of perod factor cost shares. The data confrm the ncrease n wage nequalty. Sklled labour wages ncreased at an annual average of 4.2% over the perod , whle unsklled labour wages only ncreased by 2.9% a year. A recurrent element s that t s actually sem-sklled labour that experenced the lowest wage ncreases. Concernng the factor cost shares one should note that factor cost shares are relatvely stable over tme. The factor cost share of ntermedates seems to be an excepton, whch mght ndcate the ncreasng mportance of outsourcng. Table 3.1: Summary Statstcs, Average (%) Annual change Change n log factor prces (based on quanttes) 11 - Sklled labour Sem-sklled labour Unsklled labour Intermedates Captal Factor cost shares - Sklled labour Sem-sklled labour Unsklled labour Intermedates Captal Averages are computed over 1993 and Changes n factor prces are measured as the average annual change n the log of factor earnngs. Factor cost shares reflect the average of the start and end of perod factor cost shares. 9 Sklled workers are those classfed as Managers and Admnstrators, Professonal Occupatons, Assocate Professonal and Techncal Occupatons; sem-sklled workers as Clercal and Secretaral Occupatons, Craft and Related Occupatons, Personal and Protectve Servce Occupatons; and unsklled workers as Plant and Machne Occupatons, Other Occupatons. 10 See the data appendx for more detaled nformaton on the constructon of the varables. 11 The annual change n the earnngs of sem-sklled and unsklled workers together amounts The annual ncrease n wage nequalty was then 1.71%. 8

10 Table 3.2 depcts the trends n prces and TFP by low and hgh skll-ntensty as well as ther relatve change. Skll-ntensty s defned as the sklled labour cost share n value-added. Value-added prces n skll-extensve ndustres ncreased by 1.9% a year whle value-added prces decreased by 2.7% a year n skll-ntensve ndustres. The relatve value-added prce of skll-ntensve ndustres thus fell by 4.6% a year. Prce effects cannot explan the ncrease n wage nequalty n the UK, although t s possble that value-added prces reflect productvty pass-through. TFP growth was consderably hgher n skll-ntensve ndustres amountng to 5.5% a year compared to 0.7% n skll-extensve ndustres. One should therefore focus on the prce effect net of productvty pass-through to see whether the trade-based explanaton can be rejected. Import prces may clarfy ths puzzle. From Table 3.2 t follows that mport prces rose faster n skll-extensve ndustres. Note that the relatve prce rse of skllextensve mports s drven by the rse n the prce of mports from developed countres as the relatve prce of skll-extensve mports from developng countres fell sgnfcantly. In sum, prce effects can reflect two complementary developments. Frst, productvty pass-through may be responsble for a large part of the observed prce effects. The part of TFP growth that s not transmtted through lower prces affects relatve factor prces. Second, prces of mports from developed countres had a postve effect on the relatve prce of skll-ntensve products, whereas prces of mports from developng countres had a negatve mpact. Overall Stolper-Samuelson effects should have nduced a reducton n wage nequalty. The remanng explanaton for the ncrease n wage nequalty should therefore come from TFP growth not passed through onto prces. Table 3.2: Summary Statstcs by Use of Sklled Labour, Skll-ntensty lnp ln TFP ln p M (all mports) ln p M (developng countres) ln p M (developed countres) Low Hgh Relatve Emprcal Results Table 6.1 reports the results obtaned from estmatng (2.3) wth OLS. It s mportant to note that a pror the expected sgn on outsourcng or technologcal change s not clear. However, meanngful results may be expected when the processes are regular,.e. ts sector bas and factor bas effect affect the sum of prces and TFP n the same drecton. By defnton outsourcng should have a postve mpact on TFP as outsourcng s only proftable when t s cost-savng. The queston therefore s under what crcumstances a postve relatonshp between the factor-based outsourcng and prces becomes plausble. By assumpton outsourcng leads to an excess supply of unsklled workers (as t s assumed that the low skll-ntensve fragments move abroad). As the economy as a whole becomes more skll-ntensve the full employment condtons requre that on average unsklled ntensve ndustres expand and skll-ntensve ndustres contract. Ths wll be assocated wth a relatve fall n the prce of unsklled-ntensve ndustres f prces are determned endogenously (at least partly). Thus a postve relatonshp between outsourcng and prces n the prce 9

11 regresson can only occur when outsourcng s concentrated n the skll-ntensve ndustres. 12 Outsourcng of fragments wthn the same ndustry to developng countres s postve and statstcally sgnfcant, whereas outsourcng to other ndustres n developng countres s nsgnfcant. R&D ntensty s also postve and sgnfcant. Thus both processes seem ndeed to be concentrated n the skll-ntensve ndustres. Import prces are found to exert a postve and sgnfcant effect on effectve prces. In column (2) mports prces are splt nto mport prces based on mports from developng countres and mport prces based on mports from developed countres. The results ndcate that competton from other developed countres drves the relatonshp between mport prces and effectve prces, whch corresponds to the descrptve statstcs n Table 3.2. The coeffcent on the concentraton rato s negatve and sgnfcant. Although ts coeffcents are extremely small the concentraton rato seems to play an mportant role by controllng for ndustry characterstcs such as market concentraton. Table 4.1: Stage-I-Regressons wth OLS Dependent Varable: p +TFP (1) (2) Outsourcng (narrow) (4.29) (1.87) Outsourcng (-1.00) (1.02) (dfference) R&D Intensty (2.05) (2.80) Import prces (2.04) Import prces (0.47) (developng countres) Import prces (developed countres) (2.13) Concentraton Rato -0.0 (-2.14) -0.0 (-2.49) Constant (3.18) (2.85) N R-squared Robust T-statstcs n parentheses,,, ndcate sgnfcance levels of 10%, 5% and 1% respectvely. Observatons are weghted by sales. The second stage nvolves estmatng equaton (2.4),.e., the mpact of each structural force on relatve factor prces. The component explaned by each structural varable serves as the dependent varable of the mandated wage regressons. On the rght-hand sde the average cost share over of sklled and unsklled labour are ncluded. The estmated coeffcents are nterpreted as the mpled average annual factor prce changes resultng from the structural varable under consderaton. The dfference between the coeffcents on sklled and unsklled labour s nterpreted as the total mandated change n domestc wage nequalty. The results n Table 4.2 relate to the results n regresson (1) of Table Note that the regresson does not mply a drect causal relatonshp between outsourcng and prces but the correlaton between two varables reflectng the outcome of the mpled general equlbrum dynamcs. 10

12 Table 4.2: Stage-II-Regressons - Mandated wage changes, R&D Intensty Outsourcng (N) Outsourcng (D) Import Prces Sklled (-0.08) (-0.87) (0.46) (-3.54) Unsklled (-3.82) (-1.96) (1.77) (1.13) Constant (7.51) (1.84) (-1.45) (2.27) R-squared N Mandated change Robust T-statstcs n parentheses,,, ndcate sgnfcance levels of 10%, 5% and 1% respectvely. Observatons are weghted by employment. Outsourcng to developng countres s estmated to have reduced the wage of unsklled workers by on average 0.06% a year over the perod , whle no sgnfcant mpact s found for sklled wages. The total mandated change n domestc wage nequalty as a result of outsourcng therefore was 0.06 percent a year. R&D had no sgnfcant effect on sklled wages, whle t had a negatve sgnfcant mpact on unsklled wages. R&D reduced unsklled wages by percent a year. Thus, from regresson (1) t follows that both outsourcng and SBTC have contrbuted to the ncrease n wage nequalty. Wth the actual annual ncrease n wage nequalty around 1.7% (see Table 3.1), the proporton explaned by outsourcng amounts to 4% as does the proporton explaned by SBTC. Feenstra and Hanson fnd that outsourcng accounts for 15% of the annual ncrease n wage nequalty n the US. 5. The sector bas and factor bas of outsourcng and SBTC Table 5.1 represents the results of estmatng (2.2a) and (2.2b) smultaneously usng 3SLS. Results for system (1) ndcate that both narrow outsourcng and R&D ntensty are postvely and sgnfcantly related to TFP. Outsourcng to other ndustres n developng countres s found to have a negatve mpact but ts mpact s extremely small. Import prces and the concentraton rato both are nsgnfcant. If anythng the sgns of the coeffcents suggest that competton encourages technologcal progress. Foregn competton s assocated wth a postve sgn, whle the lack of domestc competton s assocated wth a negatve sgn. In the frst set of results value-added prces are regressed on overall mport prces n addton to TFP. Foregn prces have a postve but nsgnfcant mpact on valueadded prces. In the second set of results a dstncton s made between the prce of mports from developng countres and developed countres. Both are postve, but nsgnfcant. The coeffcents for both measures of outsourcng and R&D are all nsgnfcant. It mght be the case that the mpact of outsourcng s not the same across sectors. As a matter of fact the econometrc framework relates outsourcng wthn a partcular ndustry to prces changes n that ndustry. However, n theory there s no such drect relatonshp. Instead t s the excess supply of unsklled workers that has to be absorbed by relatve output adjustments. Thus, one would expect relatve prces to go 11

13 down n low skll-ntensve ndustres ndependent of the level of outsourcng n those specfc-ndustres but as a result of the total ncrease of outsourcng of fragments of low skll-ntensty. One way to capture ths s to use an nteracton term for narrow outsourcng wth skll-ntensty. In the second specfcaton the nteracton term s ncluded n the prce regresson. Includng the nteracton term n the prce regresson renders both outsourcng and R&D ntensty postve and sgnfcant. The nteracton term s negatve whch ndcates that the effect of outsourcng on prces s less the hgher the skll ntensty n the ndustry. The negatve sgn mght reflect that outsourcng n unsklled-ntensve ndustres tends to result n more job losses for unsklled workers (because for nstance the fragments moved abroad are even less skll-ntensve) and therefore requres a more radcal adjustment of output quanttes. 13 Table 5.1: Frst-Stage Regressons wth 3SLS (1) (2) Dependent varable p TFP p TFP TFP (-2.18) (-2.30) Outsourcng (N) (-0.01) (5.86) (2.26) (5.86) Outsourcng (D) (-0.33) (-1.75) (-0.84) (-1.75) R&D ntensty (1.37) (1.66) (1.83) (1.66) Import prces (1.76) (0.47) (1.46) (0.47) Concentraton -0.0 (-2.78) -0.0 (-1.01) -0.0 (-3.40) -0.0 (-1.01) rato Factor ntensty (-1.36) O(N) Constant (3.77) (-0.28) (4.00) (-0.28) R N T-statstcs n parentheses,,, ndcate sgnfcance levels of 10%, 5% and 1% respectvely. Observatons are weghted by sales. Table 5.2 reports the results for the mandated wage regressons for specfcaton (2) of the Frst-Stage Regressons. The total ncrease n the relatve wage of sklled workers mandated by outsourcng amounted to 0.21% a year (consderably hgher than the role attrbuted to outsourcng n the OLS regressons) whch amounts to 12% of the ncrease n domestc wage nequalty over the perod n the UK. SBTC nduced an ncrease 0.07% n wage nequalty, whch s the equvalent of 4% of the total. Import prces nduced a reducton n the relatve wage of sklled labour of 0.09% a year (5% of total). Nevertheless, the net effect of globalsaton defned as the sum of foregn prce competton and outsourcng, ponts at a sgnfcant role for globalsaton n explanng the rse n domestc wage nequalty accountng for an annual ncrease n the relatve wage of sklled workers of 0.12% (stll larger than SBTC). 13 Includng the nteracton term also n the TFP regresson yelds smlar results. However the nteracton term n the TFP regresson s postve and sgnfcant ndcatng that the hgher the skllntensty the hgher the cost-reducng effect of outsourcng. 12

14 Table 5.2: Second-Stage Regressons for Specfcaton (3) TFP O(N) total R&D total PM total Sklled (-1.14) (-0.87) (-0.08) (-3.54) Unsklled (0.63) Constant (-0.02) (-1.96) (2.20) (-3.82) (7.51) (1.13) (2.27) R&D FB (-0.08) (-3.82) (7.51) R&D SB (-0.08) (-3.82) (7.51) O(N) FB (-0.87) (-1.96) (2.20) O(N) SB (-0.87) (-1.96) (2.20) Mandated change R n Robust t-statstcs n parentheses,,, ndcate sgnfcance levels of 10%, 5% and 1% respectvely. Observatons are weghted by employment. The dependent varable for the sector bas outsourcng and R&D s based on (1+λ)α thus accountng for productvty pass-through, the factor bas s based on β. Ther total effect therefore s based on the sum of the factor and sector bas, (1+λ)α+β. Wth 3SLS t s possble to decompose the total effect of outsourcng nto the effect due to the factor and sector bas respectvely. The results are reported n the last two columns of Table 5.2. Results attrbute a 0.15% annual ncrease n wage nequalty to the factor bas of outsourcng, whle the sector-bas of outsourcng accounts for 0.06%. Thus t s suggested that the factor bas of outsourcng has been about 2.5 tmes as mportant as the sector bas of outsourcng for the rse n domestc wage nequalty. However these results should be nterpreted wth cauton as they results crtcally depend on the estmate of the pass-through rate. The estmated coeffcent on TFP n the prce regresson n Table 5.3 equals -.7. For dfferent specfcatons the passthrough estmate was typcally found to be n the range of 0.7 and -1. Ths s broadly n lne wth fndngs by Feenstra and Hanson (1999) and Haskel and Slaughter (2001). However, t s not clear to what extent ths fndng reflects the actual TFP passthrough or s drven by the emprcal measurement of TFP. It seems lkely that the estmated pass-through serously overestmates the actual pass-through rate as a result of the constructon of TFP. It s therefore nterestng to see how the relatve mportance of the factor and sector bas changes wth λ. The results are summarsed n Table 5.3. Table 5.3: Summary results outsourcng by constraned 3SLS λ Sector bas Factor bas Total mandated change The frst column reflects the mposed values on the pass-through rate. Wth zero passthrough as s the case for a small open economy the mandated change n relatve wages s almost domnated by the sector bas. For the ntermedate case wth λ=-0.5 both sector bas and the factor bas matter, the latter beng the domnant force. For the extreme case where pass-through s complete only the factor bas that matters. Thus n order to assess the relatve mportance of factor and sector bas the estmate of the pass-through rate s crtcal. 13

15 The result that TFP s hghly sgnfcant n explanng prces wthout nducng any change n relatve factor prces s consstent wth Haskel and Slaughter (2001). It may not be approprate to conclude, however, that sector-based TFP growth dd not affect relatve factor prces. As a matter of fact the total mpact of sector-based TFP growth s estmated to have reduced relatve wages by 0.08% a year (the sum of R&D and outsourcng). The coeffcent on TFP n the frst-stage regressons reflects to what extent sector-based productvty growth s neutralsed by a sector-based prce change. It should therefore not come as a surprse that TFP pass-through dd not nduce any change n relatve factor prces. 6. Concluson Ths paper analyses the mpact of trade and n partcular the mpact of trade n ntermedate goods resultng from the ncreasng nternatonal fragmentaton of producton processes on the ncrease n wage nequalty n UK manufacturng for the perod Theoretcally relatve factor prces can be affected by the factor bas and the sector bas of structural change. The factor bas captures the mpact of changes n the relatve demand and supply for factors on relatve factor prces through the mpact of restructurng on relatve goods prces. The sector bas reflects changes n the relatve proftablty across sectors due to foregn prce competton or productvty growth. In order to account for both the factor and the sector bas a general equlbrum approach s adopted followng Feenstra and Hanson (1999). The results obtaned n the present study usng ths approach ndcate that outsourcng and SBTC have sgnfcantly contrbuted to the rse n the domestc wage nequalty n the UK n the 1990s. Both processes explan about 4% of the ncrease n wage nequalty. In order to get addtonal nformaton on the relatve mportance of sector bas and factor bas the TFP and prces regressons are smultaneously estmated usng threestage least squares. Wth ths extended method t s possble to evaluate the mpact of structural varables such as outsourcng and technologcal change n cases where these processes are not regular,.e. the two effects do not affect prces and TFP n the same drecton. The analyss yelds qualtatvely smlar results to the OLS regressons. Outsourcng s estmated to have mandated an annual ncrease n the relatve wage of sklled labour of 0.21%, whch amounts to about 12% of the actual ncrease n domestc wage nequalty. Decomposng the total effect of outsourcng n the effect of factor-based outsourcng and sector-based outsourcng yelds that the factor bas of outsourcng mandated of an ncrease n the relatve wage of sklled workers of 0.15% a year, whle the sector bas of outsourcng mandated an ncrease of 0.06% a year. The factor bas s thus found to be about 2.5 tmes as mportant as ts sector bas. However n assessng the relatve mportance of the two effects the estmate of the productvty pass-through coeffcent s crucal. Further research s needed to look nto the actual level of the pass-through rate. 14

16 In contrast to what s generally assumed mport prces turn out to have reduced domestc wage nequalty. Import prces mandated a reducton n the relatve wage of sklled labour of 0.09% a year. Nevertheless, the net effect of globalsaton defned as the sum of foregn prce competton and outsourcng ponts at a sgnfcant role for globalsaton n explanng the rse n domestc wage nequalty accountng for an annual ncrease n the relatve wage of sklled workers of 0.12%. 15

17 I Data Appendx Value-added prces are constructed as follows: (A.1) ln p = ln p x X X Y X Y ( ) 1/ 2( / + / ) j t t t 1 t 1 j ln p Value added prces are obtaned by subtractng the value of the sum of ntermedate purchases. The weghts are obtaned from the combned-use matrx for The weghts sum up to unty across manufacturng ndustres excludng servces. I j Total factor productvty s measured by the prmal Tornqvst Index: lntfp = ln ( ln E V ) ( ln E V ) ( ln E V ) (A.2) ( ln K V ) k The factor cost shares are based on sales. The captal cost share s defned as the resdual after subtractng the labour cost share and the ntermedate cost share from unty. Captal payments are defned as the captal cost share tmes sales. Narrow outsourcng s defned as follows total ntermedate purchases (C) tmes the rato of mported ntermedate nputs (O) over total ntermedate purchases n 1995 weghted by the proportonal change n mport penetraton by developng countres over value added: (A.3) O C M C M 95 t t M 95 t N O = t t Dfferental outsourcng n turn: 95 (A.4) O N t = j O C j95 j95 C jt M jt t M jt M j95 j95 j95 j95 R&D ntensty s measured by the relatve mportance of spendng on external R&D as a share of total ntermedate purchases. The concentraton ratos reflect the share n sales of enterprses wth more than 500 employees. 16

18 References Arndt, S.W. (1997), Globalzaton and the Open Economy, North Amercan Journal of Amercan Economcs and Fnance, Vol. 8, No. 1, pp Baldwn, R.E. and G.G. Can (2000), Shfts n relatve U.S. wages: The role of trade, technology, and factor endowments, Revew of Economcs and Statstcs, Vol. 82, pp Deardorff, A.V. (2001), "Fragmentaton across Cones", n S.W. Arndt, and H. Kerzkowsk, Fragmentaton: New Producton Patterns n the World Economy, Oxford Unversty Press. Feenstra, R.C. (1998), Integraton of Trade and Dsntegraton of producton n the Global Economy, Journal of Economc Perspectves, Fall 1998, pp Feenstra, R.C. and G.H. Hanson (1999), The Impact of Outsourcng and Hgh-Technology Captal on Wages: Estmates for the Unted States, , Quarterly Journal of Economcs, Vol. 114, Iss. 3, pp Fndlay, R. and R.W. Jones (2000), Factor Bas and Techncal Progress, Economcs Letters, Vol. 68, pp Görg, H., A. Hjzen, and R.C. Hne (2001), Internatonal Fragmentaton and Relatve Wages, Leverhulme Centre for Research on Globalsaton and Economc Polcy Research Paper, 2001/33. Gregory, M., B. Zssmos, and C. Greenhalgh (2001), Jobs for the sklled: how technology, trade, and domestc demand changed the structure of UK employment, , Oxford Economc Papers, Vol. 53, No. 1, pp Haskel, J. and M.J. Slaughter (2001), Trade, Technology, and U.K. Wage Inequalty, Economc Journal, Vol. 111, pp Haskel, J.E. and M.J. Slaughter (2002a), Does the Sector Bas of Skll-Based Techncal Change Explan Changng Skll Prema?, European Economc Revew, Vol. 46, pp Haskel, J.E. and M.J. Slaughter (2002b, forthcomng), "Have fallng Tarffs and Transportaton Costs Rased US Wage Inequalty?", Revew of Internatonal Economcs, forthcomng. Jones, R.W. (1965), "The Structure of Smple General Equlbrum Models", Journal of Poltcal Economy, Vol. 73, Iss. 6, pp Jones, R.W. and H. Kerzkowsk (1990), The Role of Servces n Producton and Internatonal Trade: A Theoretcal Framework, n R.W. Jones and A. Krueger (eds.), The Poltcal Economy of Internatonal Trade, Oxford, Basl Blackwell. Jones, R.W. (2000), "Techncal Progress, Prce Adjustments, and Wages", Revew of Internatonal Economcs, Vol.8, Iss. 3, pp Jones, R.W. and H. Kerzkowsk (2001), A Framework for Fragmentaton, n S.W. Arndt, and H. Kerzkowsk, Fragmentaton: New Producton Patterns n the World Economy, Oxford Unversty Press. Krugman, P.R. (2000), Technology, Trade and Factor Prces, Journal of Internatonal Economcs, Vol. 50, pp Leamer, E.E. (1995), The Heckscher-Ohln Model n Theory and Practce, Graham Lecture, Prnceton Studes n Internatonal Fnance, No. 77, February. Leamer, E.E. (1998), In Search of Stolper-Samuelson Lnkages between Internatonal Trade and Lower Wages, n S.M. Collns (eds.), Imports, Exports, and the Amercan Worker, Washngton D.C., Brookngs Insttuton Press, pp Slaughter, M. (2000), What are the Results of Product-Prce Studes and What Can We Learn form ther Dfferences n R.C. Feenstra (ed.), The Impact of Internatonal Trade on Wages, The Unversty of Chcago Press. Xu, B. (2001), Factor Bas, Sector Bas, and the Effects of Techncal Progress on Relatve Factor Prces, Journal of Internatonal Economcs, Vol. 54, pp