Labor Mismatch and Skill Premia

Transcription

1 Eastern Economc Journal, 214, 4, ( ) r 214 EEA /14 Labor Msmatch and Skll Prema Sherf Khalfa Department of Economcs, Steven G. Mhaylo College of Busness and Economcs, Calforna State Unversty, Fullerton, CA, 92834, USA. E-mal: skhalfa@fullerton.edu Contrary to the predctons of the Heckscher Ohln model, emprcal evdence shows that trade lberalzaton causes the skll premum to ncrease n some developng countres and to decrease n others. Ths paper develops a North South model n whch complex and smple goods are produced. The former s produced usng sklled workers, whle the latter s produced usng sklled and unsklled workers. Labor markets feature search frctons. In ths context, trade lberalzaton ncreases the skll premum n the North, whle the mpact on the skll premum n the South depends on skll abundance. The hgher the skll abundance n a developng country, the hgher the average productvty among sklled workers compared wth unsklled workers. Ths leads to a hgher employment of sklled workers n smple occupatons, or labor msmatch. Therefore, the ncrease n the prce of the smple good n the South, due to trade lberalzaton, leads to an ncrease n the wage of the msmatched workers. Ths offsets any negatve mpact on the skll premum caused by the decrease n the wage of sklled workers n complex occupatons or the ncrease n the wage of unsklled workers n smple occupatons. The opposte mpact on the skll premum takes place n countres that are less skll abundant n the South. A threshold emprcal estmaton, on 5 developng and developed countres, shows that there s a statstcally sgnfcant skll abundance threshold, below whch the coeffcent on the relatonshp between openness and wage nequalty s negatve and above whch the estmate s postve. Eastern Economc Journal (214) 4, do:1.157/eej ; publshed onlne 18 March 213 Keywords: trade lberalzaton; technologcal change; wage nequalty JEL: F16; J31; O34 INTRODUCTION The Heckscher Ohln model predcts that trade openness nduces countres to export the good that ntensvely uses the relatvely abundant factor of producton, and to mport the good that ntensvely uses the relatvely scarce factor of producton. Accordngly, skll-abundant developed countres are expected to export the good that ntensvely uses sklled workers. Ths contrbutes to an ncrease n the relatve prce of the sklled-ntensve good, a rse n the relatve demand for sklled workers, and consequently an ncrease n the skll premum. The model also predcts that skllscarce developng countres are expected to export the good that ntensvely uses unsklled workers. Ths contrbutes to an ncrease n the relatve prce of the unsklledntensve good, a rse n the relatve demand for unsklled workers, and consequently a decrease n the skll premum. Theoretcal predctons, however, are not supported by emprcal evdence. Some developng countres experenced an ncrease n the skll premum after trade lberalzaton, whle others wtnessed a decrease n the skll premum after trade lberalzaton. Ths s documented by Hanson and Harrson [1995], Robbns [1996], Wood [1997], Freeman and Oostendorp [2], and Goldberg

2 25 Sherf Khalfa Labor Msmatch and Skll Prema Cyprus Sngapore Premum.12 Premum SouthAfrca Premum Premum Korea Ghana Hungary Premum Premum Premum Israel Premum HongKong Premum Chle Premum Turkey Fgure 1. Openness and the skll premum n developng countres. and Pavcnk [24]. Fgure 1 shows the relatonshp between trade openness and the skll premum n several developng countres. 1 Fgures 1(a) 1(e) show the skll premum decreasng wth trade openness n some countres, whle Fgures 1(f ) 1( j) show the skll premum ncreasng wth trade openness n other countres. Eastern Economc Journal 214 4

3 Sherf Khalfa Labor Msmatch and Skll Prema 251 Several studes attempted to address ths puzzle n order to resolve the nconsstency between the theoretcal predctons and the emprcal evdence. The frst stream of studes attrbuted the ncrease n the skll premum n the South to outsourcng and technology transfer. For nstance, Feenstra and Hanson [1995] argue that outsourcng shfts a porton of the nput producton from the North to the South. As ths porton s the most sklled-ntensve n the South, outsourcng ncreases skll demand and the skll premum n both countres. Smlarly, Zhu [24] and Zhu and Trefler [25] argue that f the North loses compettveness n unsklled-ntensve products, a process of technology transfer s nduced. Ths process mples that the producton of the unsklled-ntensve products s relocated to the South. The relocated goods are the most sklled-ntensve by Southern standards. Ths Southern catchup rases the relatve demand for sklled workers and exacerbates wage nequalty. Xu [23] shows that, n a framework where there are non-traded goods whose range s endogenously determned by the level of trade barrers, a tarff reducton causes an expanson n the South s mport range. Ths ncreases the demand for sklled workers n the North. The ncrease n the sklled labor cost n the North leads the South to expand ts export range as well. The ncrease n the export ranges of both countres leads to an ncrease n skll demand and n the skll premum. Beauleu et al. [24] argue that a reducton n trade barrers wthn the hgh-tech sector can rase the demand for these products n both countres, and ncrease the demand for sklled labor and the skll premum. Other studes argued that trade nduces skll-based technologcal change. Acemoglu [22, 23] shows that trade causes the relatve prce of sklled-ntensve goods to ncrease n the North. The prce ncrease makes the technologes used n the producton of these goods more proftable to develop. Ths skll-based techncal change contrbutes to the ncrease n wage nequalty n the North. Snce the South mtates the North technologes that are becomng more skll-based, t experences an ncrease n the skll premum as well. Thoeng and Verder [23] argue that when globalzaton trggers an ncreased threat of technologcal leapfroggng, frms respond by basng the drecton of ther nnovatons toward sklled-ntensve technologes. Thus, openness causes defensve skll-based techncal change n the North and techncal upgradng n the producton of the mtated goods n the South toward more sklled-ntensve ones. Ths causes an ncrease n wage nequalty n the North and the South. These studes provde nsghts nto the factors generatng an ncrease n the skll premum n developng and developed countres. However, they do not address the asymmetry of the response of the skll premum to trade openness wthn developng countres. Ths paper contrbutes to the lterature by ntroducng a framework that features search frctons wth heterogeneous workers and occupatons. In ths context, the paper develops a setup wth two countres: the North and the South. In each country, the labor force s dvded nto sklled workers and unsklled workers. There are two types of fnal goods: a complex good and a smple good. The former s produced utlzng sklled workers. The latter s produced utlzng both sklled workers and unsklled workers. Labor markets n the two countres feature search frctons. There are two types of frms: those that produce complex goods and those that produce smple goods. The former post, complex vacances, can be flled by sklled workers only. The latter post, smple vacances, can be flled by both sklled workers and unsklled workers. Ths s motvated by the observaton that developed and developng countres experence a hgh level of msmatch between the Eastern Economc Journal 214 4

4 252 Sherf Khalfa Labor Msmatch and Skll Prema level of educaton of workers and the educatonal requrement of jobs they are occupyng. Ths feature of the labor markets s referred to as overeducaton. The ntroducton of the aspect of skll msmatch s justfed by ample documented observatons. The evdence on the ncdence of overeducaton across developed and developng countres s dscussed n detals n the next secton. Wages of the sklled workers n complex occupatons, the sklled workers n smple occupatons, and the unsklled workers n smple occupatons are determned by a sharng rule as n Gauter [22]. The skll premum s the rato of the weghted average wage of all sklled workers to the wage of unsklled workers. Trade lberalzaton ncreases the skll premum n the North, whle the mpact on the skll premum n the South depends on the level of skll abundance. Developng countres that are relatvely more skll abundant experence an ncrease n the skll premum after trade lberalzaton, whle others that are less skll abundant experence a decrease n wage nequalty. In the context of endogenous skll acquston, the hgher the skll abundance the hgher the average productvty of sklled workers compared wth unsklled workers. Ths hgher skll bas leads to a hgher employment of sklled workers n smple occupatons, snce they are more productve compared wth unsklled workers. Ths mples that the ncrease n the prce of the smple good n the South, due to trade lberalzaton, wll have a postve mpact on the skll premum through the ncrease n the wage of the sklled workers n smple occupatons. If the proporton of ths labor type s hgh, the ncrease n the prce of the smple good leads to an ncrease n ther total wage. Ths wll offset any negatve mpact on the skll premum caused by the decrease n the wage of the sklled workers n complex occupatons or the ncrease n the wage of the unsklled workers n smple occupatons. The opposte mpact on the skll premum takes place n countres that are less skll abundant n the South. Accordngly, the response of the skll premum to trade openness n developng countres s condtonal upon skll abundance. An emprcal analyss s undertaken usng the threshold estmaton technque ntroduced by Hansen [1999]. The estmaton results suggest the presence of a statstcally sgnfcant skll abundance threshold, below whch the coeffcent estmate of the relatonshp between openness and wage nequalty s sgnfcantly negatve, and above whch the pont estmate s sgnfcantly postve. Thus, the emprcal estmaton provdes evdence that supports the theoretcal fndngs. The remander of the paper s organzed as follows: the secton Observatons presents the evdence on the ncdence of overeducaton, the secton Model presents the model, the secton Estmaton ncludes the emprcal estmaton, the secton Smulaton ncludes the smulaton, the secton Concluson ncludes the concluson, and the secton Appendx ncludes the data and dervatons appendces. References, tables, and fgures are ncluded thereafter. OBSERVATIONS Overeducaton occurs when the educatonal qualfcatons of the worker s hgher than the educatonal requrement of the job. Thus, an ndvdual who has educaton n excess of that requred to do the job s consdered overeducated. The ncdence of overeducaton across developed and developng countres has been documented n several studes. 2 Ths evdence justfes the ntroducton of the aspect of overeducaton n the model developed n ths paper. Eastern Economc Journal 214 4

5 Sherf Khalfa Labor Msmatch and Skll Prema 253 Several studes documented the ncdence of overeducaton n the Unted States. For nstance, Duncan and Hoffman [1981] provde evdence from the Panel Study of Income Dynamcs (PSID) that the ncdence of overeducaton reached 42 percent n Burrs [1983] uses the Natonal Sample Survey to show that 21.7 percent of full-tme workers were overeducated n ther current jobs n Rumberger [1987] uses the Qualty of Employment Survey to show that there were 27 percent overeducated workers n 1973 and 32 percent n Verdugo and Verdugo [1989] estmate a 1.9 percent overeducaton n all occupatons usng the 198 census. Scherman [1991] uses the PSID waves of 1976 and 1978 to show that 4 percent of the workers n the sample report themselves are overeducated. Robst [1995] uses the PSID waves of 1976, 1978, and 1985 to show that percent of the sample s overeducated. Cohn and Khan [1995] use the PSID to estmate a 33 percent ncdence of overeducaton n Daly et al. [2] use the PSID to show that 38.5 percent of men and 36.8 percent of women were overeducated n The authors also provde evdence that 31.8 percent of men and 33.5 percent of women were overeducated n Evdence on overeducaton s also avalable for the Unted Kngdom. For nstance, Groot (1996) uses the Brtsh Household Panel Survey to show that 13 percent of the males and 1 percent of the females were overeducated n Alpn et al. [1998] use the Labor Force Survey to show a 27 percent overeducaton n Battu et al. [1999] use the Careers of Hghly Qualfed Workers Survey to show that, among the 1985 male graduates, overeducaton reached 37.6 percent n 1986, 39.6 percent n 1991, and 41.5 percent n The authors also show that, among the 1985 female graduates, overeducaton reached 46.4 percent n 1986, 39 percent n 1991, and 4.1 percent n Battu et al. [1999] also show that, among the 199 male graduates, overeducaton reached 41.6 percent n 1991 and 41.3 percent n 1996, whle, among the 199 female graduates, overeducaton reached 45.3 percent n 1991 and 39.3 percent n Green et al. [1999] fnd 46 percent overeducaton n 1998 accordng to the New Castle Survey, 29 percent overeducaton n 1986 accordng to the Socal Change and Economc Lfe Intatve, 32 percent overeducaton n 1997 accordng to the UK Sklls Survey, and 47.4 percent overeducaton n 1995 accordng to the Natonal Chld Development Study. Dolton and Vgnoles [2] use the Natonal Survey of Graduates and Dplomates to show that 38 percent of all graduates surveyed were overeducated n 198, whle 3 percent were overeducated n Chevaler and Lndley [29] use the Careers of Hghly Qualfed Workers Survey to show that the male cohort n 1985 had overeducaton of 13 percent, percent, and percent usng three measures of overeducaton. The author also shows that the female cohort n 1985 had overeducaton of percent, percent, and 2.2 percent, usng three measures of overeducaton. Chevaler and Lndley [29] also fnd that the male cohort n 199 had overeducaton of percent, 3.87 percent, and percent, whle the female cohort n 199 had overeducaton of percent, 3.93 percent, and percent. Dolton and Slles [21] use the New Castle Alumn Survey n 1998 to show that 22 percent of the graduates are overeducated for ther jobs and 42 percent are overqualfed for ther frst job. Mavromaras et al. [27] use the Brtsh Workplace Employment Relatons Survey 24 to show that 31.9 percent have sklls hgher than needed n ther job and 21.1 percent have sklls much hgher than needed. In the developng world, the evdence on overeducaton s not as abundant as n the developed world, due to the lack of detaled labor market data necessary to conduct such an analyss. However, some studes were able to provde evdence on Eastern Economc Journal 214 4

6 254 Sherf Khalfa Labor Msmatch and Skll Prema overeducaton n some developng countres. For nstance, Cohn and Ng [2] use the 1986 Hong Kong Census to show that 38 percent of the males and 32 percent of the females are overeducated. The authors also use the 1999 Hong Kong Census to show that 37 percent of the males and 31 percent of the females are overeducated. Hung [28] and Tsay et al. [25] use the Tawan Socal Change Survey n 1997 and 22 to show that 45 percent of the workers are overeducated. Ln and Wang [25] use the Survey of Famly Income and Expendture n Tawan to show that the percentage of workers who are overeducated ncreased from 27 percent n 1993 to 35 percent n There s an evdence of overeducaton n Chna as well. For nstance, Mayston and Yang [28] use a survey conducted by the Research Center for the Economcs of Educaton at Pekng Unversty to show that 2.5 percent of the graduates across Chna reported themselves as beng overeducated. The authors also show an 18.8 percent rate of overeducaton for females and 21.5 percent for males. Xangrong [28] uses the Survey of Natonal Income Dstrbuton to show that percent of the whole sample, percent of all the males, and 14.3 percent of all the females n the sample are overeducated n It s obvous that the evdence on overeducaton n developng countres s not as abundantly avalable as n the developed world. However, t s reasonable to assume that the level of skll msmatch s hgher n the developng countres, snce the unemployment rate s hgher than that n the developed world. Ths s because sklled workers accept unsklled jobs temporarly, to avod unemployment, untl they are matched wth a sklled job. Snce some developng countres have hgher unemployment rates than others, the level of labor msmatch dffers between developng countres as well. MODEL Assume there are two countres: the North and the South. The two countres are ndexed by the subscrpt A(N, S), respectvely. In country, the labor force s dvded nto sklled workers and unsklled workers. The North s relatvely abundant n sklled labor, whle the South s relatvely abundant n unsklled labor. In country, there are two types of frms: those that produce complex goods and those that produce smple goods. The former post complex vacances that can be flled by sklled workers only. The latter post smple vacances that can be flled by both sklled workers and unsklled workers. A sklled worker n a smple occupaton s allowed to conduct on-the-job search for a complex occupaton. The feature of onthe-job search s essental, as t allows sklled workers to accept a smple occupaton temporarly and contnue to search for a complex one. Indvduals We follow Dnopoulos and Segerstrom [1999] and assume country has N ndvduals ndexed by ther ablty aa[, 1]. The dstrbuton of abltes across ndvduals s unform. Each ndvdual knows hs own ablty, as do all frms that can potentally hre hm. The lfetme of each ndvdual s fnte and gven by D>. The frequency wth whch ndvduals are born and de s exogenous. Assume that N ndvduals are born at each nstant n tme. When an ndvdual des, he s replaced by a new ndvdual wth the same ablty. Thus, the country s populaton level s statonary and equals N ¼ DN. These assumptons ensure that the economy s populaton s Eastern Economc Journal 214 4

7 Sherf Khalfa Labor Msmatch and Skll Prema 255 Not Acqure sklls Acqure sklls 1 a ~ Fgure 2. Threshold ablty level. statonary. An ndvdual can enter the labor force as unsklled and earn the unsklled wage W ls for the duraton of hs lfe D. Alternatvely, an ndvdual wth ablty a can enter the labor force after spendng a gven perod of tme ToD n tranng to become sklled. A sklled worker wth ablty a earns a wage W h, whch s the average wage of all sklled workers for a perod DT>. The worker does not earn any ncome durng the perod of tranng. The wages W ls and W h are determned later n a search framework. The tranng process does not requre any real resources. Thus, the opportunty cost of becomng sklled equals the dscounted value of the foregone unsklled wage. Therefore, an ndvdual wth ablty a undergoes tranng to become a sklled worker f and only f ð1þ Z tþd t e d ð st Þ W ls dso Z tþd tþt e d ð st Þ aw h ds where d s the steady state dscount rate. The left-hand sde of the nequalty equals the dscounted wage of an unsklled worker earnng W ls from tme t untl death at tme t þ D. The rght-hand sde s the lfetme wage of a sklled worker. The sklled worker earns ncome durng the tranng perod and aw h from tme t þ T untl tme t þ D, dscounted to tme t. Snce the rght-hand sde s ncreasng n ablty a and the lefthand sde s ndependent of a, there exsts a threshold ablty denoted ~a such that (1) holds as an equalty. Therefore, all ndvduals wth ablty a4~a choose to acqure sklls, whle those wth ablty ap~a choose not to acqure sklls. Settng (1) as an equalty yelds the threshold ablty level ~a ¼ð1e dd Þ=ðe dt e dd Þð1=o Þ. Fgure 2 shows the threshold ablty level ~a. Thus, an ncrease n the skll premum o ¼ W h /W ls reduces the fracton ~a of ndvduals that choose not to become sklled. Consder the steady state supply of unsklled and sklled labor at tme t. There are ndvduals who reman unsklled and s comprsed of a fracton ~a of the populaton. Thus, the supply of unsklled labor s L ðo Þ ¼ ~a ðo ÞN. The supply of unsklled labor s decreasng n the skll premum snce ~a ðo Þo. The remanng fracton 1 ~a makes up the two remanng groups: one conssts of ndvduals engaged n tranng and the other conssts of ndvduals who completed ther tranng. Because N ndvduals are born at each nstant n tme, the number of potental tranees conssts of those younger ndvduals born between tme t and tt. Thus, the number of potental tranees s TN. Smlarly, the number of potentally sklled workers conssts of those older ndvduals born between tme td and tt and s equal to (DT)N. Thus, the number of ndvduals n tranng s ð1 ~a ÞTN ¼ ð1 ~a ÞT=DN, whereas the number of sklled workers equals ð1 ~a ÞðD TÞN ¼ ð1 ~a Þð1 T=DÞN. The average ablty of workers that completed tranng s ð1=2þþðea =2Þ and the supply of sklled labor measured n effcency unts s H ðo Þ ¼ ½N ð1 ~a ðo ÞÞð1 T=DÞŠ½ð1 þ ~a ðo ÞÞ=2Š ¼ N lo ð Þ. Eastern Economc Journal 214 4

8 256 Sherf Khalfa Labor Msmatch and Skll Prema Thus, the country s relatve endowment of sklled labor s gven by H ðo Þ= L ðo Þ ¼ lo ð Þ=~a ðo Þ. Ths reveals that the relatve endowment s ndependent of the populaton level N and s ncreasng n the skll premum snce l (o)> and ~a ðoþo. In ths context, we assume that the productvty of the sklled workers s equal to ther average ablty gven by A h ¼ð1þ ~aþ=2. Smlarly, the productvty of the unsklled workers s gven by ther average ablty A l ¼ ~a 2. H þ L s consdered the labor force n ths economy, whch we normalze to 1. Thus, the labor force n country s of measure 1, and s dvded nto y I ¼ (H /H þ L ) sklled and (1y ) ¼ (L /H þ L ) unsklled. Search frctons n the labor market allow for equlbrum unemployment. Therefore, the sklled and the unsklled household members are dvded nto those employed and those unemployed as follows ð2þ þ U h ¼ y ð3þ N ls þ U l ¼ 1 y Where hs s the number of sklled workers n complex occupatons, N s the number of sklled workers n smple occupatons, and N ls s the number of unsklled workers n smple occupatons. U h l and U denote the number of the sklled unemployed and the unsklled unemployed n country, respectvely. Frms Frms n country produce two goods: the complex good Y c and the smple good Y s. The aggregate producton s gven by a constant elastcty of substtuton composte of the two goods as follows ð4þ h Y ¼ g Y s e þð1 gþ Y c e1 Where ea(, N) s the elastcty of substtuton between the two goods. ga(, 1) s a dstrbuton parameter that determnes the mportance of the two goods n the aggregate producton. The complex good n country s produced usng sklled workers only accordng to the followng producton functon e1 e e e1 ð5þ Y c ¼ A c A h b Nhc The smple good n country s produced usng both unsklled workers and sklled workers accordng to the followng producton functon ð6þ Y s ¼ A s A l Nls þ A h Nhs b Where ba(, 1). A h s the sklled labor augmentng technology, A l s the unsklled labor augmentng technology, A c s the technology n the complex sector, and A s s the technology n the smple sector. A h /A l s defned as the skll technologcal bas or the relatve average productvty between sklled and unsklled workers. A c /A s s Eastern Economc Journal 214 4

9 Sherf Khalfa Labor Msmatch and Skll Prema 257 defned as the sector technologcal bas. The prce of the complex good n country s gven by 1 Y e ð7þ ¼ ð1 gþ P c Whle the prce of the smple good n country s gven by P s ¼ g Y 1 e ð8þ Y s the prce of the aggregate good n country s normalzed to 1 as follows h ðþ g e P s 1eþ ð1 gþ e P c 1 1e 1e ð9þ ¼ 1 Frms producng complex goods post complex vacances V c that can be matched s wth sklled workers only. Frms producng smple goods post smple vacances V that can be matched wth both sklled workers and unsklled workers. The sklled unemployed use a porton s of ther tme to search for a complex occupaton and (1s ) to search for a smple occupaton. The unsklled unemployed use all ther tme to search for a smple occupaton. The sklled workers n smple occupatons can undertake on-the-job search wth a search ntensty of o. The matchng functons are constant returns to scale, homogeneous of degree one, functons of the vacances, and effectve searchers as follows M c ¼ m c V c a s U h þ o N hs 1a ð1þ M s ¼ m s V s a ð 1 s ÞU h þ U l 1a ð11þ M c and M s represent the number of complex and smple occupaton matches, respectvely. a s the elastcty of matchng wth respect to vacances. m c and m s are the effcency parameters of the complex and smple matchng processes, respectvely. In ths context, r c ¼ M c =ðs U h þ o N hs Þ and r s ¼ðM s Þ=ðð1 s ÞU h þ U l Þ are the endogenous probabltes that a searcher s matched wth a complex or a smple occupaton, respectvely. Smlarly, q c ¼ M c /V c and q s ¼ M s /V s are the endogenous probabltes that a complex or a smple vacancy s matched wth a worker, respectvely. Y c Wages The standard approach to solve equlbrum search models s to attach an asset value to every possble worker and job state as n Gauter [22]. The expected ncome Uh stream for the sklled unemployed and the unsklled unemployed s denoted by ro and ro Ul, respectvely. Furthermore, there are three possble employment states ro Ehc, ro Ehs, ro Els, three possble job states ro Fhc, ro Fhs, ro Fls, and three possble dsagreement whle barganng states ro Dhc, ro Dhs and ro Dls. The asset values for the unemployed are gven by ð12þ ð13þ ro Uh ¼ b þ r c O Ehc O Uh þ r s O Ehs ro Ul ¼ b þ r s O Els O Ul O Uh Eastern Economc Journal 214 4

10 258 Sherf Khalfa Labor Msmatch and Skll Prema Where b s the unemployment beneft. The expected ncomes for employed workers can be expressed as ro Ehc ¼ W hc w c O Ehc O Uh ð14þ ro Ehs ¼ W hs þ r c O Ehc O Ehs w s O Ehs O Uh ð15þ ro Els ¼ W ls w s O Els O Ul ð16þ where W hc, W hs, and W ls denote the wages of the sklled workers n complex occupatons, the sklled workers n smple occupatons, and the unsklled workers n smple occupatons, respectvely. w c and w s denote the exogenous separaton rates from complex and smple occupatons, respectvely. The asset values of the dfferent matches from the frm s pont of vew are gven by ð17þ ð18þ ð19þ ro Fhs ro Fhc ¼ P s ro Fls ¼ P c qy c q W hc w c O Fhc O Vc qy s qn hs ¼ P s W hs qy s qn ls w s þ rc W ls w s O Fls O Fhs O Vs O Vs Fnally, the three dfferent dsagreement payoffs n the barganng state are gven by ð2þ ro Dhc ¼ b þ w c O Uh O Dhc ð21þ ro Dhs ¼ b þ r c O Ehc O Dhs þ w s O Uh O Dhs ð22þ ro Dls ¼ b þ w s O Ul O Dls The model can be closed by assumng that vacances are opened untl the expected ncome stream s as follows ro Vc ¼ q c O Fhc O Vc ð23þ C c ¼ ð24þ ro Vs ¼ q s U l OFls þ U h OFhs O Vs C s ¼ Where C c and C s denote the flow costs for complex and smple vacances, respectvely. We follow the lterature n assumng that a realzed match shares ths surplus. Therefore, the wage of a sklled worker n a complex occupaton n country s gven by 3 ð25þ W hc ¼ ð1 x Þb þ x P c qy c q Where x s the frm s share of the surplus. Smlarly, the barganed wage of the sklled worker n a smple occupaton s gven by 4 ð26þ Eastern Economc Journal W hs ¼ ð1 x Þb þ x P s qy s qn hs

11 Sherf Khalfa Labor Msmatch and Skll Prema 259 Fnally, the barganed wage of the unsklled worker n a smple occupaton s gven by 5 W ls ¼ ð1 x Þb þ x P s qy s ð27þ qn ls In all these cases, the wage s a weghted average of two terms: the frst term ndcates that the worker s compensated by a fracton (1x ) for the foregone unemployment beneft. The second ndcates that the worker s rewarded for a fracton x of the frm s revenues from the value of the worker s margnal productvty. Gven these wages, the skll premum o, n country, s gven by the rato of the average wage of the sklled workers n both occupatons W h ¼ð W hc W hs Þ=ð Þ to that of the unsklled workers, as follows ð28þ o ¼ Wh W ls In ths context, the mpact of trade lberalzaton on wage nequalty n the South s descrbed n the followng proposton., Proposton 1: ( a threshold skll bas, A h =Al ¼ S, below whch the skll premum decreases after trade lberalzaton, and above whch the skll premum ncreases after trade lberalzaton. Proof. Included n Appendx secton Proof of Proposton 1. Ths proposton provdes a possble explanaton for the observed patterns of skll prema n the South. As the skll bas s endogenzed, the condton for the ncrease n the skll premum n the South can be determned n terms of skll abundance as well. In ths context, A h s assumed the average ablty among sklled workers and A l s assumed the average ablty among unsklled workers. Then, the skll bas s related to skll abundance accordng to the followng proposton. Proposton 2: The relatve average productvty of sklled workers to unsklled workers, or the skll bas A h /A l, s ncreasng n relatve skll abundance H /L. Proof. Included n Appendx secton Proof of Proposton 2. In ths context, Propostons 1 and 2 mply that there exsts a threshold level of skll abundance (H /L )* n the South, ¼ S, below whch countres experence a decrease n the skll premum after trade lberalzaton and above whch countres experence an ncrease n the skll premum after trade lberalzaton. Fgure 3 shows the threshold skll abundance. Therefore, developng countres that are more skll abundant experence an ncrease n the skll premum after trade lberalzaton, whle others that are less skll abundant experence a decrease n the skll premum after trade lberalzaton. The hgher the skll abundance n a developng country, the hgher the average productvty of sklled workers compared wth unsklled workers. Ths hgher skll bas leads to a hgher level of employment of sklled workers n smple occupatons, snce they are more productve relatve to unsklled workers. Ths means that the ncrease n the prce of the smple good, due to trade Eastern Economc Journal 214 4

12 26 Sherf Khalfa Labor Msmatch and Skll Prema Skll Premum Skll Premum (H/L) * Fgure 3. Threshold skll abundance. lberalzaton, wll have a postve mpact on the skll premum through the ncrease n the wage of the sklled workers n smple occupatons. If ther proporton s hgh, the ncrease n the prce of the smple good leads to an ncrease n ther total wage. Ths ncrease offsets any negatve mpact on the skll premum caused by the decrease n the wage of the sklled workers n complex occupatons or the ncrease n the wage of the unsklled workers n smple occupatons. The opposte mpact on the skll premum takes place n countres that are less skll abundant n the South. Accordngly, we see an asymmetrc response of the skll premum to trade openness n developng countres, condtonal upon skll abundance. As for the North, the followng proposton apples. Proposton 3: The skll premum ncreases n the North after trade lberalzaton. Proof. Included n Appendx secton Proof of Proposton 3. Ths s consstent wth the observaton that the skll premum ncreases n all developed countres after trade lberalzaton. Therefore, we can conclude that there s a threshold level of skll abundance below whch all countres are developng. These developng countres experence a decrease n wage nequalty after trade lberalzaton. Above ths threshold, the remanng developng countres besdes all developed countres experence an ncrease n wage nequalty after trade lberalzaton. Ths concluson s tested emprcally n the followng secton. Fnally, we have a relatonshp between the sector bas and the threshold skll bas as follows. Proposton 4: A hgher sector bas A c /A s corresponds to a lower threshold skll bas (A h /A l )*. Proof. Included n Appendx secton Proof of Proposton 4. A hgher sector bas n the South leads to a hgher relatve productvty of the sklled workers n complex occupatons compared wth the sklled workers n smple occupatons. Accordngly, more of the sklled workers wll be employed n the complex sector rather than n the smple sector. Therefore, a hgher sector bas corresponds to a lower threshold skll bas. Ths mples that more countres n the South wll experence an ncrease n the skll premum after trade lberalzaton. ESTIMATION In ths secton, the proposton, that the mpact of trade openness on the skll premum depends on skll abundance, s tested emprcally usng the threshold Eastern Economc Journal 214 4

13 Sherf Khalfa Labor Msmatch and Skll Prema 261 estmaton technque developed n Hansen [1999]. The threshold estmaton model s gven by ð29þ Premum t ¼ m þ b 1 Openness t þ f 1 Abundance t þ f 2 RGDP t þ e t m þ b 2 Openness t þ f 1 Abundance t þ f 2 RGDP t þ e t f Abundance t ps f Abundance t 4s where the subscrpt ndexes the country and the subscrpt t ndexes tme. The dependent varable Premum t denotes the skll premum n country n year t. The varable Openness t s a measure of trade openness n country n year t. The threshold varable Abundance t s a measure of skll abundance n country n year t. The varable RGDP t denotes the logarthm of real gross domestc product (GDP) per capta n country n year t, and s added to control for macroeconomc developments that mght mpact wage nequalty. In ths context, the observatons are dvded nto two regmes dependng on whether the threshold varable Abundance t s smaller or larger than the threshold s. The regmes are dstngushed by dfferng regresson slopes, b 1 and b 2. Therefore, the threshold regresson determnes the exstence and sgnfcance of a threshold skll abundance level n the relatonshp between trade lberalzaton and wage nequalty rather than mposng a pror an arbtrary classfcaton scheme. Accordng to the predctons of the model, the coeffcent b 1 s expected to be negatve, whle the coeffcent b 2 s expected to be postve. As n Hansen [1999], another way of wrtng the equaton of nterest s ð3þ Premum t ¼ m þ b 1 Openness t IðAbundance t psþ þ b 2 Openness t IðAbundance t 4sÞ þ f 1 Abundance t þ f 2 RGDP t þ e t Þ where I(.) s the ndcator functon. A balanced panel annual data s used for 5 developng and developed countres and cover the perod from 1963 to A Thel ndex of wage nequalty compled by the Unversty of Texas Inequalty Project s used as a measure of the skll premum. Total trade as a percentage of real GDP from the Penn World Tables 6.2 s used as a measure of trade openness. As n Forbes [21], the average years of total educaton n the populaton aged over 15, from Barro and Lee nternatonal data on educatonal attanment, s used as a measure of skll abundance. Fnally, real GDP per capta s extracted from the Penn World Tables 6.2. Detaled data descrpton s ncluded n the appendx. Summary statstcs of the varables used n the estmaton are provded n Table 1. To determne the number of thresholds, the model s estmated by least squares allowng for zero, one, two, and three thresholds. The test for a sngle threshold s sgnfcant, whle the tests for a double and a trple threshold are not sgnfcant. Thus, we conclude that there s evdence for one threshold n the regresson relatonshp. The pont estmate of the threshold s , and the asymptotc 99 percent Table 1 Summary statstcs Mnmum 25 percent quantle Medan 75 percent quantle Maxmum Premum t RGDP t , , , , Openness t Abundance t Eastern Economc Journal 214 4

14 262 Sherf Khalfa Labor Msmatch and Skll Prema confdence nterval s [4.2144, 4.492]. More nformaton can be learned from the plot of the concentrated lkelhood rato functon dsplayed n Fgure 4. From the frst-step lkelhood rato functon, we see that the threshold estmate s the pont where the lkelhood functon equals. Ths occurs at s ¼ The regresson slope estmates, conventonal OLS standard errors, and whte-correlated standard errors are reported n Table 2. Real GDP per capta has a negatve effect on wage nequalty wth a coeffcent of Skll abundance has a postve mpact on wage nequalty wth a coeffcent of.395. The estmates of prmary nterest are those on trade openness. Trade openness has a statstcally sgnfcant negatve effect on wage nequalty, wth a coeffcent of.14, f skll abundance s below the threshold On the other hand, trade openness has a postve mpact on wage nequalty, wth a statstcally sgnfcant coeffcent of.11, f skll abundance s above the threshold. The same analyss s also conducted on a subsample of developng countres only. The test for a sngle threshold s sgnfcant, whle the tests for a double and a trple threshold are not sgnfcant. Thus, we conclude that there s evdence for one threshold n the regresson relatonshp. The pont estmate of the threshold s and the asymptotc 99 percent confdence nterval s [4.1458, 4.493]. More nformaton can be learned from the plot of the concentrated lkelhood rato Fgure 4. Confdence nterval constructon n sngle threshold model (entre sample). Table 2 Regresson estmates (entre sample) Regressor Coeffcent estmate OLS standard error Whte-correlated standard error RGDP t.1463* Abundance t.395* Openness t I (Abundance t p4.3686).14* Openness t I (4.3686oAbundance t ).11* *mples a statstcally sgnfcant coeffcent. Eastern Economc Journal 214 4

15 Sherf Khalfa Labor Msmatch and Skll Prema 263 Fgure 5. Confdence nterval constructon n sngle threshold model (developng countres subsample). Table 3 Regresson estmates (developng countres subsample) Regressor Coeffcent estmate OLS standard error Whte-correlated standard error RGDP t.18973* Abundance t.5166* Openness t I (Abundance t p4.493).146* Openness t I (4.493oAbundance t ).85* *mples a statstcally sgnfcant coeffcent. functon dsplayed n Fgure 5. From the frst-step lkelhood rato functon, we see that the threshold estmate s the pont where the lkelhood functon equals. Ths occurs at s ¼ It s noteworthy that the threshold skll abundance n the entre sample s wthn the confdence nterval of the threshold n the subsample. The regresson slope estmates, conventonal OLS standard errors, and whtecorrelated standard errors are reported n Table 3. Real GDP per capta has a negatve effect on wage nequalty wth a coeffcent of Skll abundance has a postve mpact on wage nequalty wth a coeffcent of The estmates of prmary nterest are those on trade openness. Trade openness has a statstcally sgnfcant negatve effect on wage nequalty, wth a coeffcent of.146, f skll abundance s below the threshold On the other hand, trade openness has a postve mpact on wage nequalty, wth a statstcally sgnfcant coeffcent of.85, f skll abundance s above the threshold. Thus, the emprcal estmatons provde evdence to support the theoretcal fndngs. SIMULATION Ths secton uses the coeffcents of the emprcal estmaton to smulate the effect of trade openness on the skll premum. The coeffcents used depend on the level of Eastern Economc Journal 214 4

16 264 Sherf Khalfa Labor Msmatch and Skll Prema Colomba Inda Premum PredctedPremum Iran Premum PredctedPremum Jordan Premum PredctedPremum Fgure Phllpnes Premum PredctedPremum skll abundance as shown n the threshold estmaton. Gven the skll abundance n every country, the coeffcents of the threshold estmaton are used to smulate the mpact of trade openness on the skll premum. The predcted skll premum s compared wth the actual skll premum observed n the data. Ths comparson s conducted n order to examne the success of the model and the estmaton n explanng the patterns observed n the real world. The actual and the predcted skll prema n selected countres are dsplayed n Fgures 6 and 7. The fgures, for most of the countres, show that the predcted and the actual skll premum show smlar patterns. Ths ndcates that the predcted skll premum s successful n explanng most of the pattern of skll prema that we observe n the data Premum Premum Openness, the skll premum, and the predcted skll premum. PredctedPremum PredctedPremum Venezuela CONCLUSION The Heckscher Ohln model predcts that trade openness nduces countres to export the good that ntensvely uses the relatvely abundant factor of producton and to mport the good that ntensvely uses the relatvely scarce factor of producton. Accordngly, skll-abundant developed countres are expected to export the good that ntensvely uses sklled workers. Ths leads to an ncrease n the relatve prce of the sklled-ntensve good, a rse n the relatve demand for sklled workers, and consequently an ncrease n the skll premum. On the other hand, Eastern Economc Journal 214 4

17 Premum PredctedPremum Cyprus Sngapore SouthAfrca Korea Ghana Hungary Israel HongKong Chle Turkey Premum PredctedPremum Premum PredctedPremum Premum PredctedPremum Premum PredctedPremum Premum PredctedPremum Premum PredctedPremum Premum PredctedPremum Premum PredctedPremum Premum PredctedPremum Fgure 7. Openness, the skll premum, and the predcted skll premum. Sherf Khalfa Labor Msmatch and Skll Prema 265 Eastern Economc Journal 214 4

18 266 Sherf Khalfa Labor Msmatch and Skll Prema skll-scarce developng countres are expected to export the good that ntensvely uses unsklled workers. Ths leads to an ncrease n the relatve prce of the unsklledntensve good, a rse n the relatve demand for unsklled workers, and consequently a decrease n the skll premum. Emprcal evdence, however, demonstrates that although some developng countres wtnessed a declnng skll premum after trade lberalzaton, others experenced a wdenng wage gap after trade lberalzaton. Ths paper attempts to address the nconsstency between the theoretcal predctons and the observatons. In ths context, the paper ntroduces a two-country setup: the North and the South. In each country, the labor force s dvded nto sklled workers and unsklled workers. There are two types of fnal goods: a complex good and a smple good. The former s produced utlzng sklled workers. The latter s produced utlzng sklled workers and unsklled workers. Labor markets n the two countres feature search frctons. The skll premum s the rato of the average wage of all sklled workers n both occupatons to the wage of unsklled workers. Trade lberalzaton ncreases the skll premum n the North. In the South, the mpact of trade lberalzaton on the skll premum depends on skll abundance. Developng countres that are more skll abundant experence an ncrease n the skll premum after trade lberalzaton, whle others that are less skll abundant experence a declne n wage nequalty. Ths can be ntutvely explaned as the hgher the skll abundance n a developng country, the hgher the average productvty of sklled workers compared wth unsklled workers. Ths hgher skll bas leads to a hgher level of employment of sklled workers n smple occupatons, snce they are more productve relatve to unsklled workers. Ths means that the ncrease n the prce of the smple good, due to trade lberalzaton, wll have a postve mpact on the skll premum through the ncrease n the wage of the sklled workers n smple occupatons. If ther proporton s hgh, the ncrease n the prce of the smple good leads to an ncrease n ther total wage. Ths wll offset any negatve mpact on the skll premum caused by the decrease n the wage of the sklled workers n complex occupatons or the ncrease n the wage of the unsklled workers n smple occupatons. The opposte mpact on the skll premum takes place n countres that are less skll abundant n the South. Accordngly, the response of the skll premum to trade openness n developng countres s condtonal upon skll abundance. An emprcal analyss s mplemented usng the threshold estmaton technque ntroduced by Hansen [1999]. The results suggest the presence of a statstcally sgnfcant skll abundance threshold, below whch the pont estmate of the coeffcent of the relatonshp between openness and wage nequalty s negatve, and above whch the pont estmate s postve. Acknowledgements I thank Danel Trefler, two anonymous referees, and partcpants n the Eastern Economc Assocaton Annual Meetng 29. Remanng errors are my own. Appendx Data The estmaton uses a balanced panel of annual data that covers the perod from 1963 to 1999 for 5 developng and developed countres. The developed countres n the sample nclude: Australa, Austra, Canada, Denmark, Fnland, Greece, Iceland, Eastern Economc Journal 214 4

19 Sherf Khalfa Labor Msmatch and Skll Prema 267 Ireland, Italy, Japan, the Netherlands, New Zealand, Norway, Span, Sweden, the Unted Kngdom, and the Unted States. The developng countres n the sample nclude: Algera, Chle, Colomba, Cyprus, Ecuador, El Salvador, Ghana, Honduras, Hong Kong, Hungary, Inda, Indonesa, Iran, Israel, Jamaca, Jordan, Kenya, Korea, Malaw, Malaysa, Maurtus, Mexco, Pakstan, Panama, Phlppnes, Sngapore, South Afrca, Syra, Tawan, Tunsa, Turkey, Venezuela, and Zmbabwe. The varables used n the estmaton are descrbed n detals as follows: Skll premum The skll premum or wage nequalty data set s compled by the Unversty of Texas Inequalty Project. The orgnal data comes from the Unted Natons Industral Development Organzaton statstcs, whch provde average manufacturng pay by ndustry. From these average ndustral wages, a Thel ndex of nequalty s calculated and used n ths analyss as a measure of wage nequalty. Detaled defnton of ths varable s ncluded n Galbrath and Kum [24]. Trade openness Trade openness data s extracted from the Penn World Tables 6.2. Exports plus mports dvded by real GDP s the total trade as a percentage of GDP. Ths s the constant prce equvalent of the total trade as a percentage of GDP. Skll abundance Informaton on the relatve supply of sklled and unsklled workers s avalable for only a few countres, whle data on educatonal attanment s wdely avalable and relatvely comparable across countres. Therefore, as a proxy for the relatve supply of sklled labor, we use average years of total educaton n the populaton aged over 15 years, as reported n Barro and Lee Internatonal Data on Educatonal Attanment. As the data s avalable only for the years 196, 1965, 197, 1975, 198, 1985, 199, 1995, and 2, we use lnear nterpolaton to derve the years n between. Real GDP per capta The data for real GDP per capta (Laspeyres) s extracted from the Penn World Tables 6.2, whch s obtaned by addng up consumpton, nvestment, government, and exports, and subtractng mports n any gven year. The gven year components are obtaned by extrapolatng the 1996 values n nternatonal dollars from the Geary aggregaton usng natonal growth rates. Dervatons Wage of sklled workers n complex occupatons The wage of sklled workers n complex occupatons s gven by W hc Wages can be derved from ¼ argmax O Ehc O Dhc x O Fhc ð1 x Þ O Ehc O Dhc ¼ x O Fhc 1x Eastern Economc Journal 214 4