Changes in the Demand for Skilled Labor within Japan s Manufacturing Sector: Effects of Skill-Biased Technological Change and Globalization

Transcription

1 Bank of Jaan Workng Paer Seres Changes n the Demand for Sklled Labor wthn Jaan s Manufacturng Sector: Effects of Skll-Based Technologcal Change and Globalzaton tosh Sasak * htosh.sasak@boj.or.j Kench Sakura ** kench.sakura@boj.or.j No.05-E-12 Setember 2005 Bank of Jaan Nhonbash ongoku-cho, Chuo-ku, Tokyo * Research and Statstcs Deartment ** Research and Statstcs Deartment Paers n the Bank of Jaan Workng Paer Seres are crculated n order to stmulate dscusson and comments. Vews exressed are those of authors and do not necessarly reflect those of the Bank. If you have any comment or queston on the workng aer seres, lease contact each author. When makng a coy or reroducton of the content for commercal uroses, lease contact the Publc Relatons Deartment (webmaster@nfo.boj.or.j) at the Bank n advance to request ermsson. When makng a coy or reroducton, the source, Bank of Jaan Workng Paer Seres, should exlctly be credted.

2 CANGES IN TE DEMAND FOR SKILLED LABOR WITIN JAPAN S MANUFACTURING SECTOR: EFFECTS OF SKILL-BIASED TECNOLOGICAL CANGE AND GLOBALIZATION * tosh Sasak Research and Statstcs Deartment, Bank of Jaan. and Kench Sakura Research and Statstcs Deartment, Bank of Jaan. Setember, 2005 Abstract Ths aer examnes the demand shft toward workers who are unversty graduates n terms of skll-based technologcal change (SBTC) and globalzaton, both of whch utlze relatvely more workers wth secfc sklls. Accordng to the results of our emrcal study, whch used the major grous of Jaan s manufacturng anel data from the , t s clear that the ncrease n the relatve demand for unversty graduates s closely related to the R&D exendtures rato (SBTC) and the mort rato from the East Asan countres or the foregn roducton rato (globalzaton). Ths suggests that there has been a demand shft toward hghly educated workers n Jaan s manufacturng sector due to both SBTC and globalzaton. Keywords: sklled workers; skll-based technologcal change; globalzaton JEL Classfcaton Number: F16; J23; O30 * We are esecally grateful to Izum Takagawa for her caable assstance. We also thank Kyoj Fukao, Kojro Sakura, Ech Tomura, Masahro go, Takesh Kmura, semnar artcants at the Jaan Economc Assocaton 2005 Srng meetngs, and other offcals of the Bank of Jaan for ther helful comments and suggestons. Of course, we are solely resonsble for any remanng errors n ths aer. The vews resented n ths aer are those of the authors, and not those of the Bank of Jaan. E-mal: htosh.sasak@boj.or.j E-mal: kench.sakura@boj.or.j 1

3 1. INTRODUCTION In many advanced countres, ncludng the Unted States, the wage nequalty between sklled and unsklled workers has ncreased from the 1980s to 1990s. 1 It has been onted out that the nequalty s caused by the demand shft from unsklled to sklled workers. Ths aer s motvated by our nference that the demand shft toward sklled workers has also been n rogress recently n Jaan. One man factor that caused the demand shft toward sklled workers s skll-based technologcal change (SBTC, hereafter). SBTC s one artcular tye of technologcal change that utlzes relatvely more workers wth secfc sklls. It ncludes sklls such as, for examle, IT-related technque, whch s necessary to use comuters and software, nstallaton of manufacturng equments, whch requres advanced technques and sklls, and research and develoment (R&D, hereafter). Another mortant factor s the effects of globalzaton, esecally the ncrease n trade wth develong countres. They corresond to the exandng morts of chea and unsklled-labor-ntensve roducts due to the ncrease n trade wth develong countres, and to foregn outsourcng of a art of domestc manufacturng rocesses whch requre relatvely less-sklled domestc labor to those countres, where frms develo global roducton actvtes. Ths aer examnes the effects of both SBTC and globalzaton on the demand shft toward sklled labor n Jaan s manufacturng. Before erformng a concrete analyss n the next secton, we frst revew several revous studes from the Unted States and Jaan. Many studes have been carred out n the Unted States, snce the wage nequalty there between sklled and unsklled workers has recently grown, as we noted above. Esecally, the evaluaton of the effects of SBTC and globalzaton on the demand shft toward sklled workers has been a great concern, because SBTC s closely assocated wth educatonal and/or technologcal olces, whle globalzaton has much to do wth trade olcy. These revous studes have reorted that both SBTC (such as comuter nvestment rato) and globalzaton (such as foregn outsourcng to develong countres or trade exanson wth those countres) have caused the demand shft from roducton workers (unsklled workers) to non-roducton workers (sklled workers). There s no consensus yet, however, on the relatve macts of these two effects, whch deend on the secfcatons of the estmated models and on the 1 The rse n wage nequalty s clearly observed, esecally n the Unted States and the Unted Kngdom, n several ndustralzed countres accordng to Freeman and Katz (1994). These dfferences n wage nequaltes across develoed countres arse from the dfferences n labor demand and suly factors and some nsttutonal asects concernng wage-settng such as unonzaton, mnmum wage systems, and favorable treatments for unsklled labor such as unemloyment nsurance. 2

4 varables n use. 2 In contrast, there have been only a few related studes n Jaan, artly because wage nequalty has not been an oen queston thus far. Among the mortant studes n recent years, Sakura (2001) has found that recent technologcal changes such as IT are key n causng the demand shft from roducton workers to non-roducton workers. 3 As related to the effects of globalzaton, Sakura (2000) has reorted that the relatonsh between the rogress of foregn outsourcng and the demand shft toward non-roducton workers seems to be unclear. Meanwhle, ead and Res (2000) have onted out that ncreased emloyment n the overseas afflates has worked as a key factor for the demand shft toward non-roducton workers. The above-mentoned studes have analyzed the erods of the 1980s and/or the begnnng of the 1990s, and have lkewse nvestgated the effects of SBTC and globalzaton on the demand shft toward sklled workers, searately. In ths resect, Ito and Fukao (2004) have extended the data to 2000, and have also analyzed the demand shft toward sklled workers, takng nto account both SBTC and globalzaton smultaneously. They have reorted that the effects of SBTC, such as IT and R&D, on the demand shft are sgnfcant, whle those of globalzaton ncludng the vertcal ntra-ndustry trade rato and the foregn outsourcng rato wth the East Asan countres vary, deendng on the defnton of sklled labor or on the varables emloyed n the emrcal study. On the bass of these revous studes, ths aer has two man features: frst, we have examned the erods from the 1990s to early 2000s, when the effects of SBTC and globalzaton are lkely to have mostly ervaded n Jaan; and second, we have tred to evaluate the relatve macts of both SBTC and globalzaton on the demand shft toward sklled workers, takng account of them smultaneously. These two onts have not yet been suffcently examned by the revous Jaanese studes. Ths study covers the manufacturng sector only; t excludes the non-manufacturng sector to examne the relatonsh between the effects of globalzaton and the demand shft toward sklled workers. Moreover, t focuses on regular emloyment, snce the rato of rregular emloyment to total emloyment s stll low n Jaan s manufacturng. 4 Fnally, we assume that worker sklls are classfed by ther educatonal levels; therefore, we regard the unversty graduates as 2 For examle, Berman et al. (1994) and Autor et al. (1998) have reorted that SBTC has greatly contrbuted to the demand shft toward sklled labor whle the effects of globalzaton have been trval. On the other hand, Sachs and Shatz (1994), Wood (1994), Bernard and Jensen (1997) and Feenstra and anson (1996a, 1996b, 1999) have reorted that globalzaton too, along wth SBTC, has been a key factor. 3 Sakura (2004) has develoed the analyss of Sakura (2001), reortng that recent technologcal change had been an mortant factor n the demand shft toward hghly educated workers. 4 The ercentage of art-tme workers n the manufacturng ndustry s only about 10 ercent accordng to the Monthly Labor Survey 2003 of the Mnstry of ealth, Labor and Welfare. 3

5 sklled, and the other workers (non-unversty graduate workers) as unsklled. 5 The remander of ths aer s organzed as follows. In Secton 2, we descrbe the rocess of the share n the wage bll of sklled workers (roorton of the wage bll ad for sklled workers to the total wage bll) rsng when the relatve demand for sklled workers ncreases, due to the effects of SBTC and/or globalzaton. In Secton 3, we examne whether the relatve demand for unversty graduates, defned as sklled workers n ths study, have recently ncreased n Jaan by usng the wage data comled n the Basc Survey on Wage Structure (BSWS, hereafter) by the Mnstry of ealth, Labor and Welfare. In Secton 4, we erform an emrcal analyss to nvestgate whether both SBTC and globalzaton have ncreased the relatve demand for unversty graduates wth the use of anel data by ndustry of Jaan s manufacturng sector from Secton 5 concludes. 2. TE DEMAND SIFT FROM UNSKILLED TO SKILLED WORKERS Ths secton descrbes the rocess n whch the share n wage bll of sklled workers ncreases when the demand shft from unsklled to sklled workers occurs wthn each ndustry due to SBTC and/or globalzaton. An ndustry s assumed to roduce ts outut usng both sklled and unsklled labor L as nut factors whose unt rces (.e., wages) are descrbed as w and w L, resectvely. Takng sklled labor on the vertcal axs and unsklled labor on the horzontal axs, the soquant curve of ndustry should be shown as y 0 y0 n Fgure 1(1). When the economy s n equlbrum at ont A under erfect cometton, the relatve wage and the relatve labor rato are descrbed as w / w and / L, resectvely. When a technologcal change takes lace, the soquant curve nwardly shfts to the orgn; the ndustry s now caable of roducng a gven amount of outut wth less nuts. If ths technologcal change s skll-based (SBTC), the soquant curve shfts nward so as to save the unsklled labor relatve to the sklled one. Ths shft s exressed as an nward shft from y 0 y0 to y y 1 1 n Fgure 1(2). After the L 5 We can aly a roducton/non-roducton classfcaton nstead of an educatonal classfcaton; we can regard roducton/non-roducton workers as unsklled/sklled workers. owever, the roducton/non-roducton classfcaton n the major grous of manufacturng n the standard ndustral classfcaton s not avalable n the BSWS; t s only avalable n the dvson n the standard ndustral classfcaton (the roducton/non-roducton classfcaton n the major grous of manufacturng s avalable n the Census of Manufactures by the Mnstry of Economy, Trade and Industry only before 1990). Accordng to the Basc Survey on Wage Structure 2003CY by the Mnstry of ealth, Labor and Welfare, the roorton of unversty graduates n roducton workers s less than 10 ercent n the manufacturng as a whole, whle the roorton n non-roducton workers s more than 50 ercent. Therefore, t would not become an ssue to classfy both the hghly educated and non-roducton workers as sklled. 4

6 adjustments of labor nuts and wages, 6 the economy wll fnally reach equlbrum at ont B. Comarng the relatve labor ratos and relatve wage ratos at both economes A and B, the relatve labor rato ncreases ( / L< / L ) whle the relatve wage rato declnes ( w L / w > w L / w ). Consequently, the share n the wage bll of sklled workers at economy B (the rght sde of the nequalty below) exceeds that at economy A (the left-hand sde of the nequalty below): 7 wage bll w w < ( wl L + w ) ( w L L + w ) share of sklled workers at Economy A wage bll share of sklled workers at Economy B The above analyss has demonstrated that the share n the wage bll of sklled workers rses as a result of the ncrease n relatve demand for sklled workers wthn a same ndustry. Such a demand shft toward sklled workers wthn an ndustry occurs not only by SBTC but also by globalzaton, artcularly, by foregn outsourcng actvtes to develong countres where unsklled workers are abundant; frms become multnatonal and transfer a art of roducton rocesses, whch use relatvely more unsklled labor, to develong countres; then, they change the sources of rocurement of ntermedate and fnal goods from domestc sules that used to be rovded domestcally, to morts from those countres, resultng n the ncrease (decrease) n the relatve demand for sklled (unsklled) labor wthn each ndustry. 8 Fnally, t s worth mentonng that the foregn outsourcng effects stated above and the standard effects of changes n relatve rces of tradable goods,.e., the rce effects of Stoler-Samuelson theorem, are the same n that the trade exanson wth develong countres ncreases the relatve demand for sklled labor; although the ways of the demand shft toward sklled labor are dfferent n both cases. The demand shft takes lace toward sklled labor wthn an ndustry n the former, whle t does so between ndustres n the latter where the shft arses from the ndustres wth relatvely abundant unsklled labor to those wth relatvely abundant sklled ones. Therefore, the demand shft to sklled labor by globalzaton takes lace both wthn and between ndustres. 6 ere, the number of sklled and unsklled workers s exogenously gven, snce we focus on the change n the labor demand. Thus, the labor nut wth an ncrease n the relatve demand for sklled labor s adjusted by workng hours, not by number of workers. 7 When the cks-neutral technologcal change occurs, the soquant curve shfts nward, whch nevertheless affect nether the relatve labor rato nor the relatve wage rato. Consequently, the share n the wage bll of sklled workers remans unchanged after the technologcal change. 8 Concernng foregn outsourcng effects, strctly seakng, we need to ncororate n the model the transactons of arts and fnal goods n each roducton stage between ndustralzed and develong countres, rather than dscussng on the bass of roducng unque good, as shown n Fgure 1. For examle, Feenstra and anson (1996a) have develoed a theoretcal model that descrbes the rocess n whch foregn outsourcng rases the relatve demand for sklled labor wthn the same ndustry. 5

7 Takng nto account the above analytcal framework, the analyss of ths aer focuses on the effects of SBTC and globalzaton on the demand shft from unsklled (non-unversty graduate workers) to sklled workers (unversty graduate workers). 3. AN INCREASE IN TE DEMAND FOR IGLY EDUCATED WORKERS IN JAPAN S MANUFACTURING SECTOR We examne whether the relatve demand for hghly-educated workers has recently ncreased n Jaan before analyzng the relatonsh between SBTC and/or globalzaton and the demand shft toward workers wth college degrees. We use the wage data of regular male workers of Jaan s manufacturng from the BSWS. Secton 3.1 measures and examnes the hourly wage dfferentals between unversty graduates and workers wthout a college educaton snce 1985 by estmatng the Mncer-tye wage functon, takng nto account worker characterstcs, such as ther workng years. In Secton 3.2, we decomose the changes n the wage bll share of unversty graduate workers nto the wthn and the between shft to obtan mlcatons n terms of the effects of SBTC and globalzaton. 3.1 Measurng the ourly Wage Dfferentals Between Workers Wth and Wthout Unversty Degrees Accordng to the human catal theory, educaton s assumed to work for accumulatng the human catal of workers as ts nvestment effect, consequently augmentng ther wages. The educaton ncludes both schoolng and job tranng through busness. The Mncer-tye wage functon s derved on the bass of ths human catal theory (Mncer, 1974). In ths framework, the logarthm of worker s hourly wage s descrbed as a functon of hs duraton of schoolng, workng years, and ts square: lnw 2 = α + β t + γ 1K + γ 2K (3-1) where ln w reresents the logarthm of worker s hourly wage; t denotes years of schoolng; K s hs workng years; α s a constant term; and β, γ 1 and γ 2 are arameters, resectvely. In ths secton, we estmate the followng equaton (3-2) for 1985, 1990, 1995, 2000, and 2003, based on (3-1), wth the wage samle of regular male workers of 17 manufacturng ndustres 9 n the BSWS. The estmaton s erformed by weghtng 9 The 17 ndustres nclude Food, Beverages and Tobacco, Textle Mll Products, Aarel and Other Fnshed Products, Lumber and Wood Products, Furnture and Fxtures, Pul, Paer and Paer Products, Prntng and Alled Industres, Chemcal and Alled Products, Rubber Products, 6

8 each wage samle by ts number of workers (weghted least squares), dealng wth a ossble heteroscedastcty n the error term: 10 ln( wage ) = a + = b ( g g kn 2 1 ) g 3 Dh 2 Ej 16 h h D + e j j D + = 1 = 1 k = 1 Fk + d f k D + u (3-2) where denotes the subscrt of the wage samle; ln( wage ) s the logarthm of scheduled hourly wage, whch s equvalent to the rato of the scheduled cash earnngs (thousands of yen er month) to scheduled actual hours worked (hours er month); ( kn ) s the length of servce (years); Dh Ej D, D and D Fk reresent the school career, the sze of enterrse and the ndustry dummes, resectvely; a reresents a constant term; b g, d h, e j and f k are arameters; and u denotes the error term. Three knds of dummy varables are defned as follows: Dh School career dummes: D D1 D unversty graduate dummy (unversty graduate workers 1, others 0); D2 D hgher rofessonal school and junor college graduate dummy (hgher rofessonal school or junor college graduate workers 1, others 0); D hgh school graduate dummy (senor hgh school graduate workers 1, others 0). D3 Ej Sze of enterrse dummes: D E1 D large enterrse dummy (enterrses wth 1,000 or more emloyees 1, others 0); D medum enterrse dummy (enterrses wth emloyees 1, others 0). E 2 Fk Industry dummes: D Industry dummy varables consst of 16 tyes of manufacturng ndustres; the base ndustry s the manufacture of food, beverages and tobacco. By estmatng (3-2), we can quanttatvely evaluate the wage dfferentals among workers wth dfferent educatonal backgrounds, after controllng worker characterstcs contaned n the ndvdual wage samle,. 11 The estmaton results are shown n Table 1. For examle, the wage dfferental between unversty graduate Ceramc, Stone and Clay Products, Iron and Steel, Non-Ferrous Metals and Products, Fabrcated Metal Products, General Machnery, Electrcal Machnery, Equment and Sules, Transortaton Equment, and Precson Instruments and Machnery. 10 We can obtan 1,836 wage samles at most: that s, 17 (number of ndustres) 9 (number of age classes from 20 to 65 years old) 4 (number of educatonal background classes) 3 (number of frm sze classes). 11 As a related study, Sakura (2004) has evaluated the wage dfferentals among dfferent educatonal backgrounds durng the by estmatng the Mncer-tye wage functon, whch s smlar to equaton (3-2), ndcatng ther nterretatons and roblems. 7

9 workers and hgh school graduate workers s calculated as the dfference between the estmated arameter on the hgh school dummy and that on the unversty dummy (.e., dˆ ˆ 1 d 3 ). 12 Fgure 2 descrbes the chronologcal changes n the measured hourly wage dfferentals between unversty graduate workers and non-unversty graduate workers from the These wage dfferentals have slghtly ncreased durng the erod, even though t s stll necessary to take estmaton errors nto account. An ncrease n the suly of unversty graduate workers accomaned by the recent oularzaton of hgher educaton would work to lower the hourly wages of unversty graduate workers, at least relatvely. The above estmaton results nevertheless suggest that the magntude of the ncrease n relatve demand for unversty-educated workers has been large enough to cancel out the downward ressure on hourly wages due to such a suly factor. 3.2 Decomoston of the Increase n the Wage Bll Share of Unversty Graduates nto Wthn and Between Shfts We now focus on the changes n the wage bll share of unversty graduates snce Tycally, the wage bll data from whch the effects of varous worker characterstcs should be elmnated, followng the analyss conducted n the Secton 3.1 s used. owever, for smlcty, we calculate the wage bll as the average scheduled cash earnngs tmes the number of male workers. Frst, accordng to the chronologcal change n the wage bll share of unversty graduates n overall manufacturng n Table 2(1), the wage bll share has consstently ncreased snce 1985, and n artcular the rate of ncrease has somewhat accelerated snce The average rate of annual change from the s ercentage onts, whch s comarable to that n the U.S. manufacturng sector from ,.e., ercentage onts. 13 Furthermore, the wage bll share has ncreased n all ndustres, excet for a art of erods, as shown n Table 2(2). As the backgrounds of these movements, t should be onted out that there exsts an ncrease n the relatve demand for unversty graduate workers by ndustry, takng account of the results obtaned from the analyss n 3.1, n addton to the ncrease n the number of 12 A few onts should be noted regardng the estmaton results excet for the school career dummes n Table 1. Frst, the arameter estmate on workng years, ˆb, s ostve, whle that on ts square, 1 ˆb, 2 s negatve; ths suggests that the scheduled hourly wages of workers ncrease wth the workng years, although the rate of ncrease lowers wth the years. In addton, among the sze of enterrse dummes, the arameter on large enterrse dummy s estmated to be ostve; ths ndcates that the scheduled hourly wages of workers belongng to large enterrses are hgher than those of workers belongng to both medum and small enterrses. Fnally, the arameters on ndustry dummes mean the ndustral remums on the bass of the Beverages, Tobacco and Food ndustry. 13 Accordng to Autor et al. (1998), the rate of change n the wage bll share of unversty graduates n U.S. manufacturng s ercentage onts durng the erod, whle that durng the erod s ercentage onts. 8

10 unversty graduates, couled wth the oularzaton of hgher educaton. Let us now consder the backgrounds of the ncrease n the wage bll share of unversty graduates; the ncrease n the wage bll share of unversty graduates s decomosed nto the between and the wthn shfts; the former s aroached by the shft of unversty graduate workers across ndustres whle the latter s by the ncrease of unversty graduates wthn each ndustry. 14 To be more recse, the decomoston s shown n the followng equaton (3-3):.e., the between shft reresents the ncrease n the wage bll of ndustres where the wage bll shares of unversty-educated workers are hgh, whereas the wthn shft reresents the ncrease n the wage bll share of unversty graduates wthn the same ndustry: s = n s = between shft + n s = wthn shft (3-3) = 1,, n : ndustres ncluded n the major grous of manufacturng (n = 17); s = W / W : wage bll share of unversty graduates n all manufacturng; s = W / W : wage bll share of unversty graduates n ndustry ; = W / W : wage bll share of ndustry s workers n all manufacturng. where reresents the changes n varables n ercentage onts; s the average of two erods; W s the wage bll of all manufacturng; W s the wage bll of unversty graduates n all manufacturng; W s the wage bll of ndustry ; W s the wage bll of unversty graduate workers of ndustry. The between shft exressed as a frst term n the rght sde of (3-3) s led to by the change n ndustry structure accomaned by the changes n ether domestc and nternatonal demands for roducts or sector roductvty shocks. Meanwhle, the wthn shft exressed as the second term s aroached by the ncrease n the relatve demand for unversty graduates wthn the same ndustres accomaned by the effects of SBTC and/or those of globalzaton. If our hyothess that ether SBTC or globalzaton serves to ncrease the relatve demand for unversty graduate workers s true, the change n the wage bll share of unversty graduate workers should be brought about by the effects of the wthn shft rather than by those of the between shft. Table 3(1) shows the results of the decomoston nto the between and the wthn shfts of the wage bll share of unversty graduates n Jaan s manufacturng sector accordng to (3-3). It shows that the contrbuton rate of the wthn shft (aroxmately 90 ercent) largely exceeds that of the between shft (aroxmately 14 Ths rocedure s based on the analyss of Berman et al. (1994). 9

11 10 ercent) n any erod after Table 3(2) shows the results of further decomoston of the wthn and between shfts of the wage bll share of unversty graduates n Jaan s manufacturng sector durng the erod nto the arts of contrbutons by ndustry. As for the wthn shft, the wage bll shares of unversty graduates have ncreased n all ndustres. Above all, the rses are romnent n the ndustres such as general machnery, electrcal machnery, transortaton equment, and recson machnery. Subsequently, as for the between shft, the wage bll shares of unversty graduates have declned n many ndustres, ncludng textles, aarel, lumber and wood, furnture and fxtures, ceramcs, stones and clay, ron and steel, nonferrous metals, and fabrcated metals, all of whch are the mort cometng ndustres under the ressure of recently ntensfed nternatonal cometton. Meanwhle, the wage bll share has ncreased n ndustres such as general machnery, electrcal machnery, transortaton equment and recson machnery,.e., the exortng ndustres mentoned above. Summng u, t can be concluded that the wage bll share of unversty graduate workers has ncreased mostly by the effects of the wthn shft, n whch the rato of unversty graduate workers rses wthn the same ndustry, accomaned by the effects of the between shft, n whch unversty graduate workers have shfted toward the hghly comettve ndustres from the ndustres whch have lost ther nternatonal comettveness. Ths develoment s consstent wth our hyothess that the relatve demand for unversty graduate workers wthn the same ndustry has ncreased due to SBTC and the exanson of foregn outsourcng. 4. EMPIRICAL ANALYSIS The analyses u to the revous sectons have made t clear that, n Jaan, the demand for unversty-educated workers has ncreased as comared to the demand for workers wthout a unversty degree, and that the ncrease n the relatve demand for unversty graduates has occurred manly wthn the same ndustry. In ths secton, we erform an emrcal analyss on the effects of SBTC and globalzaton, both of whch are, n general, regarded as the man causes of the demand ncreases for unversty graduates, usng the anel data by ndustry of Jaan s manufacturng sector. 4.1 Econometrc Model Issues (1) Dervaton of the Cost Share Equaton The cost share equaton, on whch the followng emrcal analyss s based, s 10

12 derved based on the framework adoted n the semnal study of Berman et al. (1994). Let us assume that an ndustry wthn the manufacturng sector roduces outut wth ts nut factors of catal, K, unsklled labor, L and sklled labor,. The catal s assumed to be a fxed factor n the short run whle both tyes of labor are to be varable factors. 15 The structural factor, such as SBTC and/or globalzaton, s denoted as Z, whch shfts the roducton and cost of ndustry. Under these V assumtons, we set the followng varable cost functon, C, as the translog form: V ln C = α 0 + α ln w + α K ln K + α Y lny + α Z ln Z + α t t + 0.5( α YK KK ln K 2 + α ty YY lny 2 + α YZ ZZ ln Z 2 + α t tt 2 tk + q γ q KZ ln w ln w + α lny ln K + α t lny + α lny ln Z + α t ln K + α ln K ln Z + α t ln Z + Y K Z ρ lny ln w + ρ ln K ln w + ρ ln Z ln w + q ) t tz ρ t ln w (4-1) where w denotes the rce of varable factor; t s the tme trend effect. The followng cost share equaton s obtaned by dfferentatng the translog varable cost functon (4-1) wth resect to the logarthm of the varable factor rce, ln w, and alyng Sheard s Lemma: lnc ln w V = w d V C = s Y K Z q t = α + ρ lny + ρ ln K + ρ ln Z + γ ln w + ρ t q q (4-2) where d reresents the demand for varable factor wth ts factor rce, w. The condton for the translog varable cost functon n equaton (4-1) to be homogeneous of degree one wth resect to the varable factor rce, w, s for the followng equaton (4-3) to hold: 16 Y K Z t = γ q = ρ = ρ = ρ = ρ = γ and q 0 q 15 ere we assume the labor and catal to be, a ror, varable and quas-fxed, resectvely, rresectve of ther tyes. In ths resect, some authors have exressed doubts on whether ths assumton actually holds n Jaan. Nshmura and Mnetak (2004), for nstance, classfes the labor by educatonal backgrounds (hgh/low-educated) and age (old/young), and the catal by IT or non-it, then examnng ther varablty, ndvdually. Consequently, old and oorly educated workers are reorted to be a short-term fxed factor, snce aged sklls are requred under Jaan s long-term emloyment ractce, and on the sho-floor; IT-related catal, meanwhle, s determned to be short-term varable factor, snce software and comuters can be relaced n the short run. 16 The symmetry of arameters s also assumed,.e., γ q = γ. q 11

13 α =1 (4-3) In addton, we mose the followng restrcton of the constant returns to scale on the cost share equaton (4-2). Y ρ ρ = 0 (4-4) + K The varable factors n the cost mnmzaton roblem here are unsklled and sklled labors,.e., L and, resectvely. Takng account of the restrctons of both equatons (4-3) and (4-4), the cost share equaton (4-2) can thus be rewrtten by regardng the varable factor rce, w, ncluded n the translog varable cost functon (4-1), as the factor rce of sklled labor, w, as follows: s w K = α + γ ln ( ) + µ ln ( ) + λ ln Z + δ t (4-5) L w Y where s reresents the rato of the wage bll of sklled workers,.e., a varable factor, to the total varable cost:.e., the wage bll share of sklled workers, L L w /( w + w L ). ln( w / w ) s a logarthm of the relatve wage rato between sklled and unsklled workers; ln( K / Y ) s the logarthm of catal ntensty; ln Z s the logarthm of structural varable; α s the constant term; γ, µ, λ and δ are arameters. The feature of the cost share equaton s that the wage bll share of sklled workers s reresented as the lnear functon of structural and other controllng varables. In Secton 2, we have llustrated the rocess that the ncrease n the relatve demand for sklled workers due to the effects of SBTC and/or foregn outsourcng leads to a rse n the wage bll share of sklled workers. In ths resect, we can erform an emrcal study to examne the above mechansm by estmatng (4-5) wth the use of ndustry-level data under the assumton that the cost functons (4-1) are dentcal across ndustres. One sulementary ont should be noted before secfyng the estmaton model n the followng (2, below). The catal ntensty and the structural varables of the thrd and fourth terms ncluded n the rght sde of (4-5) are both assumed to fluctuate exogenously. owever, t s hardly ossble to assume that the relatve wage,.e., the second term of (4-5), vares exogenously as long as t reresents the relatve qualty between sklled and unsklled workers. In such a case, we can ossbly deal wth the roblem by estmatng the model wth the nstrumental varables correlated wth relatve wage. owever, snce t s n fact dffcult to fnd the arorate nstrumental 12

14 varables, we then exclude the term from the estmaton model after the revous studes, ncludng Berman et al. (1994). (2) Model Secfcaton We use the balanced anel data of the major grous of manufacturng (14 ndustres 17 ) from The estmaton model s based on the cost share equaton (4-5), whch however excludes the relatve wage of the second term n the rght sde of the equaton, addng the error term and fourth as follows: s t K = α + µ ln ( ) t + λz t + α + φ t + ε t (4-6) Y where s the subscrt of ndustry ncluded n the major grous of manufacturng; t s the tme subscrt; s t s the wage bll share of unversty graduate workers; α s the constant term; both µ and λ are arameters assumed to be dentcal across ndustres ; α s the ndvdual effects by ndustry; φ t s the tme effects; and ε t s the error term,.e., the dosyncratc shock. The structural varable, Z t, of the thrd term n the rght sde of (4-6) ncludes the terms that reresent the effects of globalzaton and SBTC. The mort rato from East Asa, M t, and the foregn roducton rato, F t, are used to denote the effects of globalzaton. 19 Moreover, the IT nvestment rato, comuter nvestment rato, and R&D exendtures rato are consdered to be suted for reresentng the effects of 17 The 14 ndustres nclude Food, Beverages, and Tobacco, Textle Mll Products Includng Aarel and Other Fnshed Products, Pul, Paer, and Paer Products, Prntng and Alled Industres, Chemcal and Alled Products, Rubber Products, Ceramc, Stone, and Clay Products, Iron and Steel, Non-Ferrous Metals and Products, Fabrcated Metal Products, General Machnery, Electrcal Machnery, Equment and Sules, Transortaton Equment, and Precson Instruments and Machnery. 18 Regardng the revous Jaanese studes based on the major grous of manufacturng, the samle number used n those emrcal studes s unlkely to be suffcent, snce there s a lmted number of ndustres. In ths resect, the revously cted Ito and Fukao (2004) has used the data of 35 ndustres of Jaan s manufacturng sector durng the erod , whch greatly mroves the small samle roblem on these emrcal studes. Ths aer has also constructed the balanced anel data consstng of 14 ndustres over 16 years, allowng us to secure, at most, 224 samles. 19 Because the East Asan countres ossess relatvely more abundant unsklled labor than does Jaan, Jaan s multnatonal comanes romote foregn outsourcng actvtes toward the regon wth an am to advance vertcal dvsons of labor. Consequently, morted roducts from East Asa nclude not only unsklled labor-ntensve roducts based on the dfferences n the factor roorton between Jaan and those countres, but both ntermedate and fnal roducts brought by the foregn outsourcng actvtes. In ths way, the mort rato from East Asa s a roxy for foregn outsourcng n a relatvely broader sense, comared wth the roxy of foregn outsourcng, whch s based only on the trade of ntermedate goods. Moreover, the varable of foregn roducton rato.e., another varable that reresents a globalzaton factor s suggested to be the roducton rato n the develong countres, takng nto account ths aer s urort. owever, snce the data of the foregn roducton rato n develong countres s not avalable, that of an entre world s used for t: t covers 13 ndustres, excludng Prntng and Alled Industres from the 14 ndustres lsted n footnote

15 SBTC. owever, as far as we know, there are no data on Jaan s IT catal, whch s dsaggregated by ndustry and covers the erod untl the early 2000s. 20 We have therefore decded to use the R&D exendtures rato, R t, as the roxy varable standng for the effects of SBTC. Because the deendent varable n the model s the wage bll share, the varables ncluded n the structural vector are denoted n rato, not logarthmc-transformed, makng t easer to nterret the estmaton results. In ths case, the estmated arameter vector λ on the structural varables should be nterreted as the values of elastcty. The mort and foregn roducton ratos are not smultaneously ncluded n the model at the stage of actual estmaton; they are searately ncluded wth the R&D exendtures rato as the followng (4-6 ) and (4-6 ): s s t t K M R = α + µ ln ( ) t + λ M t + λ R t + α + φ t + ε t (4-6 ) Y K F R = α + µ ln ( ) t + λ F t + λ R t + α + φ t + ε t (4-6 ) Y Regardng the nterretatons of the arameters ncluded n (4-6), µ reresents the comlement/substtute relatonsh between a worker s skll and catal. A ostve µ denotes ther comlement relatonsh, whle the negatve one reresents ther substtute relatonsh. The ostve arameter λ on the structural varable mles that the R&D exendtures rato as an SBTC factor and ether mort rato or foregn roducton rato as a globalzaton factor cause an ncrease n demand for unversty graduates. Moreover, t s also ossble to quanttatvely assess the relatve macts of SBTC and globalzaton from the estmated values of the arameters n the model Regardng IT catal data, for examle, the JIP database n Fukao et al. (2004) may be qute useful; t contans detaled data, ncludng catal stock by ndustry. Unfortunately, ts end erod s 1998; thus, we cannot obtan data from subsequent years. The rates of ncrease n the hourly wage dfferentals across dfferent educatonal backgrounds and the wage bll share of unversty graduates have been accelerated, artcularly snce 2000, as shown Secton 3 (Fgure 2 and Table 2 (1) cted above), from whch we can nfer that the demand for unversty graduates further ncreased around the corresondng erods. In such a sense, we have decded to nclude a samle coverng the erods u to the frst half of the 2000s, dealng wth the R&D exendtures rato, whch s avalable untl the latest erod, as an SBTC factor. 21 As Wood (1994) ndcates, when develoed countres exand ther morts of unsklled labor-ntensve roducts by ncreasng the volume of trade wth develong countres, t may romote technologcal develoments to further augment the added values of ther domestc roducts to comete wth ther chea morted roducts n qualty. Takng nto account that the technologes develoed n ths way tend to be unsklled labor-savng on the whole, SBTC s consdered to have the asects of defensve nnovaton, not of autonomous one, nduced by a globalzaton; for nstance, Lawrence (2000) has erformed an emrcal study on the relatonsh between ntensfed nternatonal cometton and technologcal rogress, fndng a sgnfcant relatonsh between them. It would therefore be desrable to measure the SBTC of the autonomous nnovaton, whch excludes the effects of globalzaton, to recsely evaluate the macts of SBTC and globalzaton; ths study wll be a art of our future work. 14

16 (3) Estmaton Method We estmate the lnear model (4-6) for anel ndustry data wth weghtng varable of the wage bll share of each ndustry to control characterstcs and sze dfferences across dfferent ndustres. Ths estmaton rocedure also deals wth the otental heteroscedastcty n the error term. It s mortant to decde how to handle the ndvdual effects across ndustres, α n (4-6), on estmatng the lnear model for anel data. We erform the estmaton of both fxed-effects and random-effects models, and then select an arorate one by the ausman test: the test results have suorted the fxed-effects models for all secfcatons. Moreover, n addton to the ndvdual effects, the tme effects, φ t, n (4-6) are also ncluded to control the shocks revalng over all ndustres at each erod. 22 The tme effects contan not only the technology shocks revalng over all ndustres and busness cycle effects, but suly sde factors such as the ncrease n the rato of unversty graduate workers accomaned by the recent oularzaton of hgher educaton, are also consdered mortant. Concernng the nterretaton that the tme effects nclude the labor suly factor, we nvestgate the valdty of the nterretaton by ncororatng the ncrease n the relatve suly of unversty graduate workers as the dfferent form of the tme effects nto the estmaton: that s, we estmate the model n whch the rato of unversty graduates to all school-leavers, NL t, s added to the rght sde of (4-6). 23 Fnally we check the robustness of the estmaton results n terms of the followng ways. () Parameter Consstency For the arameter estmates to be consstent, the exlanatory varables n the model need to be strctly exogenous, condtonal on the constant term and the unobservable ndvdual and tme effects:.e., the condtonal exectaton of the error term,, of the rght sde of (4-6) must be zero: ε t E[ ε α X α φ ] = 0 for all and u ( u = { 1, 2, L, t, L, T } ) t u t where E [ ] s the exectaton oerator; X u s the vector of exlanatory varables 22 Arorate models are selected by the secfcaton tests:.e., the F-test and the lkelhood rato test. 23 The estmaton model of (4-6) s derved from the demand sde, and the labor suly s not endogenzed. Therefore, t s an ad hoc way to add the rato of unversty graduates as a labor suly factor to the rght sde of the equaton. In ths resect, however, snce the wage bll share of unversty graduates,.e., the deendent varable of the equaton, should be affected by the labor suly factor, we thnk that t s necessary to control the effects of the labor suly factor to judge the emrcal results. 15

17 ncluded n the rght sde of (4-6),.e., ( ) X u = ln( K u / Y u ), Z u. We examne whether the vector of exlanatory varables, X t, n the rght sde of (4-6) satsfes a strct exogenety condton as follows. 24 Frst, we set u the estmaton model n whch the vector of subset varables, W s, of the vector of exlanatory varables, X t, for the erods excet for current erod t ( t s ) s added to the rght sde of (4-6): s t = α + Φ X + Ψ W + α + φ + ε (4-7) t s t t where both Φ and Ψ reresent the arameter vectors. Then, (4-7) s estmated, and fnally the F-test s conducted under the null hyothess that the arameter vector of the subset varables, W s,.e., the thrd term n the rght sde of (4-7), s equvalent to zero ( 0 : Ψ = 0). Unless the null hyothess could be rejected, the strct exogenety condton of the exlanatory varable, X t, would be satsfed. 25 In our ractcal test rocedures, we have not erformed the test for all the subset varables, W s : we have only conducted the test for the subset varables before and after one erod ( W t 1 and W t +1 ), whch are most lkely to be correlated wth the error term n erod t. Furthermore, we have estmated (4-6) wthout the term of relatve wage, L ln ( w t / wt ),.e., the second term n the rght sde of (4-5), as has already been descrbed above (1). owever, the exogenety condton of the exlanatory varables on the rght sde of (4-6) would be unsatsfed, f the omtted varable roblem was serous. Therefore, we also estmate the model n whch the term of relatve wage s added to the rght sde of (4-6), just for reference. () Test for the Seral Correlaton n the Error Term ε t The seral correlaton of the error term, ε t, n the rght sde of (4-6) could affect the arameter estmates. 26 We then estmate the model by assumng the autocorrelaton n the error term,.e., a anel AR(1) estmaton, comarng the result wth the one obtaned from a standard anel estmaton. 4.2 Data for Emrcal Analyss 24 See Wooldrdge (2002), Adversely, when the null hyothess s rejected, the lagged exlanatory varables should be added to the rght sde of (4-6) as the control varables n case of s < t, because the error term of (4-6) at erod t, ε, s correlated wth the exlanatory varables at erod t ( t s). On the contrary, n case of s > t, t s necessary to estmate the frst dfferenced model wth nstrumental varables to control the feedback effects from the error term at erod t of (4-6) to the exlanatory varables at ( s t) erod ahead. 26 Wth the seral correlaton n the error term, ether the ordnary least square or anel estmaton method.e., fxed-effects or random-effects models wll generate neffcent arameters; n such a case, the arameters of the frst dfferenced model would be rather effcent. 16

18 The ndustral classfcaton here corresonds to the major grous of manufacturng n the standard ndustral classfcaton for Jaan. 27 The data s on a calendar year bass, and the erod s from The defntons and sources of the varables used n estmatons are as follows: Wage bll share of unversty graduates ( s t ): the rato of wage bll of unversty graduates to total wage bll ( 100) from the BSWS. The wage bll s calculated by the average scheduled cash earnngs tmes the number of male workers. Catal stock ( K t ): tangble fxed assets excludng land, evaluated by book value, retrement and derecaton adjusted, from the Census of Manufactures of the Mnstry of Economy, Trade and Industry (METI). Ths ales to establshments wth 30 or more emloyees. The deflator of gross domestc catal formaton (lant and equment of rvate sectors) s from the System of Natonal Accounts of the Cabnet Offce. Value added ( Y t ): value added from the Census of Manufactures (METI). Ths ales to establshments wth 30 or more emloyees. The deflator of gross domestc exendture classfed by economc actvtes s from the System of Natonal Accounts. Imort rato from the East Asan countres ( M t ): the rato of mort value from East Asa to the sum of domestc shment value and total mort value ( 100). The varable s created by usng the Summary Reort on Trade of Jaan (Jaan Tarff Assocaton), the Census of Manufactures (METI), the Inut-Outut Table (Mnstry of Internal Affars and Communcatons), and the Cororate Goods Prce Index (Bank of Jaan). 29 Foregn roducton rato ( F t ): foregn roducton rato n the Annual Survey on Cororate Behavor from the Cabnet Offce. It s equvalent to the rato of foregn roducton value to the sum of domestc roducton and foregn roducton values ( 100). R&D exendtures rato ( R t ): the rato of comany research exendtures to sales ( 100) obtaned from the Reort on the Survey of Research and Develoment of the Mnstry of Internal Affars and Communcatons. Unversty graduates rato ( NL t ): the rato of unversty graduates and ostgraduates to total new graduates ( 100), whch s the sum of rmary school, junor hgh school, hgh school or junor school under the old system of educaton, junor colleges, techncal colleges, unverstes, and ostgraduate studes for male workers calculated from the Emloyment 27 The varables used n ths aer are n rncle based on the data by ndustry, corresondng to the major grous of manufacturng n the standard ndustral classfcaton for Jaan. owever, as the varable of mort rato s made by aggregatng the ndvdual tems comled n the trade statstcs by ndustry, the ndustral classfcaton n the varable may not be strct. 28 The samle erod begns from 1988, because the classfcaton of goods n the Summary Reort on Trade of Jaan, on whch the mort rato varable s based, s avalable only from The mort ratos of the manufacturng by ndustry are avalable n the Indces of Industral Domestc Shments and Imorts of the METI. owever, ths ndcator s lmted to several ndustres that only cover mnng and manufacturng roducton. In addton, the mort rato from East Asa s not avalable. We have therefore decded to create a new ndcator of mort rato from East Asa. See the Data Aendx for the concrete rocedure to create the varable. 17

19 Status Survey of the Mnstry of Internal Affars and Communcatons. The survey s ublshed every fve years (the latest s 2002) so that the data of unublshed years are lnearly sulemented. L Relatve wage rato ( w t / wt ) : the rato of average scheduled cash earnngs of unverstygraduate male workers to those of other male workers calculated from the BSWS. Table 4 shows the descrtve statstcs and correlatons of these varables. The average rates of change n the varables of both mort rato and foregn roducton rato as globalzaton factors (+0.30 and +0.67, resectvely) largely exceed those n the varable of R&D exendtures rato as an SBTC factor (+0.04). In addton, the unversty graduate rato also shows a hgh rate of ncrease (+0.42). Above these statstcs suggest that the exanson of trade wth East Asa and the oularzaton of hgher educaton n Jaan have wdely develoed n the 1990s. Fgure 3 shows the relatonshs between the wage bll share of unversty graduate workers and the R&D exendtures rato as an SBTC factor or the mort rato as a globalzaton factor. It s observed that the wage bll share of unversty graduate workers by ndustry s ostvely correlated wth the R&D exendtures rato and mort rato by ndustry, although the strength of these relatonshs slghtly dffers n each year. 4.3 Emrcal Results Table 5(1) reorts the results of emrcal analyss. The arameter estmates on mort rato, M t, and foregn roducton rato, F t, and those on R&D exendtures rato, R t, both of whch reresent the effects of globalzaton and SBTC factors, resectvely, are estmated to be ostve and statstcally sgnfcant for all secfcatons from the model (1) to (5). Furthermore, the tme effects are all statstcally sgnfcant n those secfcatons (two-way effects models are chosen by model secfcaton tests). In ths resect, when the term of unversty graduates rato, NL t, as a labor-suly factor s added to the exlanatory varables, the tme effects turn to be statstcally nsgnfcant (the one-way effects model s selected by model secfcaton test), and the term of unversty graduates rato s estmated to be ostve and statstcally sgnfcant (the model (6) and (7)). The estmated values of tme effects generally show the ever-ncreasng trend wth some fluctuatons, corresondng to the movements of the ncreases n unversty graduates rato couled wth the recent oularzaton of hgher educaton (Fgure 4). An ncrease n the relatve suly of unversty graduate workers wth the recent oularzaton of hgher educaton contrbutes to the rses n the wage bll share of unversty graduates used as a deendent varable. Nonetheless, accordng to the above emrcal results, both 18

20 globalzaton and SBTC have rased the wage bll share of unversty graduate workers n Jaan s manufacturng even after controllng the effects of such a suly factor. Fnally, although the arameter estmate on the catal ntensty, ln( K / Y ) t, used as a control varable, s sgnfcantly estmated to be negatve n the standard anel estmaton as shown n Table 5(1), t s not sgnfcantly estmated n the anel AR(1) secfcaton, as shown n Table 5(3). Moreover, the arameter estmates somewhat dffer by the model secfcatons and the estmaton methods. It s thus dffcult to fnd clear comlement or substtute relatonshs between catal ntensty and worker sklls n Jaan s manufacturng. 30 We now resent the results for verfyng the robustness of the above estmaton results. Frst, the strct exogenety test (F-test) for the exlanatory varables n the model s erformed. We obtaned the result that the null hyothess that both arameters on the subset varables, whch are one year lagged and one year forward of the thrd term n the rght sde of (4-7),.e., W t 1 and W t + 1, are equvalent to zero has not been rejected, as shown n Table 5(2). 31 Moreover, when we estmate the L model wth addng the term of relatve wage rato, ln( w t / wt ), to the rght sde of (4-6), the arameter estmates excet the constant term are hardly affected, as shown n the model (8) and (9) of Table 5(1):.e., the omtted varable roblems do not aear to be serous. Both results suort the consstency of the estmated arameters based on (4-6). Next, among the arameter estmates on the globalzaton and SBTC factors obtaned by the anel AR(1) estmates of Table 5(3), although the sgnfcance of the arameter on foregn roducton rato s merely weakened, the arameter estmates on mort rato and R&D exendtures rato are almost the same as those obtaned n the standard anel estmaton of Table 5(1). It s therefore hard to conclude that the exstence of the seral correlaton n the error term serously affects the estmaton results. The above test results suort the robustness of the anel estmaton results 30 Accordng to the revous Jaanese lterature, the relatonsh between catal ntensty and worker skll dffers by ther feld study. For examle, whle Sakura (2001) reorts that a clear relatonsh between them has not been found, Sakura (2004) found a sgnfcant ostve relatonsh between them. Moreover, ead and Res (2000) has reorted to fnd a negatve relatonsh. These results are dfferent from those of many revous U.S. studes, whch found clear ostve relatonshs between them; see, for examle, the revously cted Berman et al. (1994), Feenstra and anson (1996a, 1996b, 1999), Bernard and Jensen (1997), Autor et al. (1998), and so forth. In recent manufacturng remses, the roducton rocess has been moved from assembly to a batch system. Consequently, snce such a catal deeenng s lkely to substtute unsklled labor, both catal and worker skll are consdered n general to be comlementary; see Goldn and Katz (1998). Such a comlementary relatonsh has not been observed n Jaan accordng to the results of our emrcal studes; one lausble reason for ths s the steady ncrease n the weght of the all-urose ndustres such as the electrcal ndustry, whch roduce homogenous roducts and stresses the urchase and nstallaton of the latest ready-made machnes; another reason s smly due to the roblem of the measurement error of catal. 31 We attemt some atterns of varable sets; see Table 5(2). Further, we also erform the statstcal tests on the arameters of the subset varables, whch are two or three years lagged and forward, wth the result that the null hyothess cannot be rejected (the test results are abbrevated). 19