Employment Protection Legislation, Multinational Firms and. Innovation

Transcription

1

2 Employment Protecton Legslaton, Multnatonal Frms and Innovaton Rachel Grffth and Gareth Macartney Unversty College London and Insttute for Fscal Studes November 009 Abstract The theoretcal effects of labour regulatons such as employment protecton legslaton (EPL) on nnovaton s ambguous, and emprcal evdence has thus far been nconclusve. EPL ncreases job securty and the greater enforceablty of job contracts may ncrease worker nvestment n nnovatve actvty. On the other hand EPL ncreases adjustment costs faced by frms, and ths may lead to under-nvestment n actvtes that are lkely to requre adjustment, ncludng technologcally advanced nnovaton. In ths paper we fnd emprcal evdence that both effects are at work - multnatonal enterprses locate more nnovatve actvty n countres wth hgh EPL, however they locate more technologcally advanced nnovaton n countres wth low EPL. JEL codes: D, F3, O3, J4 Keywords: Innovaton, employment protecton, multnatonal frm locaton. Acknowledgements: The analyss contaned n ths paper was funded by the ESRC/EPSRC Advanced Insttute of Management Research (AIM) and the ESRC Centre at the IFS. We would lke to thank Laura Abramovsky, Helen Mller and semnar partcpants n Berln, Florence, Helsnk and Mannhem for helpful comments. Responsblty for any results, opnons, and errors les wth the authors alone. Correspondence: rgrffth@fs.org.uk; gmacartney@onpontanalytcs.com; IFS, 7 Rdgmount Street, London WCE 7AE, UK.

3 Introducton Employment protecton legslaton has been a focus of polcy concern n the European Unon. There s now consderable evdence that ths type of labour market rgdty s assocated wth hgher unemployment. More recently attenton has focused on the mpact of labour regulatons on the ncentves for frms to nvest n productvty enhancng nnovaton and growth. Here the relatonshp s less clear. When makng ts nnovaton decson a frm consders two effects of employment protecton legslaton (EPL). Frst, EPL ntroduces a frng cost to any adjustment to employment made by the frm. Second, ths adjustment cost ncreases job securty for exstng workers as t reduces the probablty of beng fred n response to small fluctuatons n demand. Effcency wage arguments suggest that ths ncreases the value of employment for the worker and ncreases ther (unobservable) effort, whch n turn can ncrease the return to nnovaton for the frm. On the other hand, where nnovaton s radcally new and requres new sklls, and thus a drastc adjustment to employment, EPL may ncrease the cost of such nnovaton. Exstng models of radcal nnovaton suggest that countres wth low EPL have a comparatve advantage n radcal nnovaton. 3 The man contrbuton of ths paper s emprcal. To motvate our emprcal strategy we descrbe a model that ncorporates both postve and negatve effects of EPL on nnovaton ncentves for frms. We dstngush between ncremental nnovaton and radcal nnovaton: radcal nnovaton s potentally more proftable than ncremental nnovaton, but requres a large and drastc employment adjustment, because workers wth new sklls are needed to mplement the nnovaton (as n Chapter 8 of Aghon- Howtt 998). EPL ncreases ths cost of adjustment, but t also has postve effects on both types of nnovaton by ncreasng workers effort to further ncrease the See, nter ala, Lazear (990), Blanchard and Wolfers (000), Nckell, Nunzata and Ochel (005), and Grffth, Harrson and Macartney (007). See Shapro and Stgltz (984) for the effcency wage set-up and Boer and Jmeno (005) for an applcaton to EPL. Although not ts central pont, workers nvest more n general tranng n the presence of search frctons n the labour market when they are less lkely to be fred by ther present employer n Acemoglu (997). See also Akerloff (98), Agell (999) and Chapter 0 of Sant-Paul (996) for the postve effects of EPL. 3 See Sant-Paul (997, 00) and Samanego (006). Also Cunat and Meltz (007) provde theoretcal and emprcal evdence that countres wth flexble labour markets have a comparatve advantage n ndustres wth hgh demand volatlty. Caballero et al. (004) provde theoretcal and emprcal evdence that countres wth strong EPL are slow to adjust employment, and that ths s assocated wth low productvty growth. Also, Bartelsman et al. (008).

4 productvty of nnovatons. The model suggest that, for plausble parameter values, the optmal level of nvestment n radcal nnovaton decreases wth EPL but that the optmal level of nvestment n ncremental nnovaton ncreases wth EPL. The paper s related to several lteratures. It s drectly related to the growng lterature on the effects of labour market regulatons on productvty and by extenson to the papers on cross-country patterns of specalzaton and natonal nsttutons. 4 There s a related lterature on the product lfe-cycle that dstngushes between new product nnovaton and mature product nnovaton, where demand s more certan for the latter. 5 It also relates to the endogenous growth lterature and the model presented bulds heavly on the framework of Aghon-Howtt, where the dstncton between radcal and ncremental nnovaton s through the employment adjustment that s requred to mplement radcal nnovaton. 6 Our paper s also related to the lterature on the locaton of actvty by multnatonal frms. 7 There s an exstng emprcal lterature on the relatonshp between labour regulatons and productvty and nnovaton, based manly on cross-country evdence, whch remans nconclusve, wth studes fndng dvergent results. 8 Such studes struggle to deal wth two key dentfcaton problems. One s that the effect of EPL may depend on the nature of nnovaton, and n most data t s dffcult to dstngush between ncremental and radcal nnovaton. The other s that n the cross-secton labour 4 See, nter ala, Nunn (007), Carln (003). 5 See, nter ala, Klepper (996) and Bresch et al. (000), Audretsch (995), Puga and Trefler (005), and Sant-Paul (997, 00). 6 Ths s n contrast to the dstncton that radcal nnovaton s less lkely to succeed than ncremental nnovaton that s made n Sant-Paul (997, 00) and Bartelsman et al. (008). We argue that modellng radcal nnovaton as requrng adjustment to employment s approprate for our sample of large ncumbent frms, whereas modellng radcal nnovaton as more rsky wth hgh frng costs arsng n the event of falure seemed more approprate for small frms and consderatons of frm entry and ext. If radcal nnovaton were more rsky and the cost of falure (ext) ncreased wth EPL then ths would enhance our predctons. 7 See, nter ala, Dunnng (977), Caves(996), Ekholm and Hakkala (007), Devereux and Grffth (998). Haaland and Wooton (003) show that mult-natonal enterprses wll locate hgh rsk projects n countres wth low redundancy costs n the presence of ndustry or economy wde wage barganng, and when the rsk profle of the MNE s dfferent to that of domestc frms. 8 Both Storm and Nastepaad (007) and Buchele and Chrstansen (999) fnd that hgh EPL s assocated wth greater productvty growth. Bassann and Ernst (00) fnd that EPL has a negatve effect n less coordnated countres, n hgher coordnated countres workers and frms can algn ther nterests better. Smlarly, Scarpetta and Tressel (004) fnd a sgnfcant mpact of EPL on mult-factor productvty growth when nteracted wth barganng coordnaton, but no lnear result. Hall and Soskce (00) argue that dfferences n specalsaton between Germany and the US are due to the more market orentated fnancal and labour market nsttutons n the US. Acharya et al. (009) fnd that strong labor laws encourage nnovaton. Akkermans et al. (005) support the vew that lberal market economes specalze n radcal nnovaton. 3

5 regulatons may be correlated wth unobservable characterstcs of countres, ndustres and frms that determne nnovaton. We deal wth the frst challenge by usng an ntutvely appealng measure of radcal nnovaton: the proporton of ctatons on a patent applcaton made to scentfc journals (as opposed to other patents). We show that patents that are closer to the scentfc lterature are assocated wth more varablty n output and employment. We tackle the second challenge by basng our results on an dentfcaton strategy that uses varaton wthn multnatonal frms from countres on where they locate dfferent nnovatve actvtes, and therefore controls for unobservable characterstcs of the home country, ndustry and frm that affect the nnovaton decson. We fnd that multnatonal frms perform more overall nnovaton n hgh EPL countres, but that the same multnatonal frms perform more radcal nnovaton n low EPL countres. We see these basc relatonshps n the cross-country assocaton between EPL and nnovaton actvty. Fgure shows the average proporton of ctatons to the scentfc lterature plotted aganst EPL, usng data on all frms that appled for patents at the European Patent Offce. 9 The downward slopng relatonshp suggests that there s less radcal nnovaton performed n countres wth hgh EPL. In ths paper we focus on multnatonal frms. 0 Fgure shows the same negatve assocaton between EPL and radcal nnovaton across these frms. In Fgure 3, however, we see a postve effect of EPL on overall nnovaton. These aggregate pctures may be maskng many dfferent effects. We show below that these results are robust to controllng for frm fxed effects and for many country level regulatory and factor endowment characterstcs. Ths paper proceeds as follows: secton presents a smple model of ncremental and radcal nnovaton; secton 3 dscusses our dentfcaton strategy; secton 4 descrbes our emprcal specfcaton and data, explanng our measure of radcal nnovaton; secton 5 presents our results; and a fnal secton concludes. 9 Ths graph s based on patent applcatons made to the European Patent Offce by 37,350 lsted and unlsted frms n the prvate sector, see Macartney (009). These frms were responsble for the flng of 30,3 patents n the perod 997 to 003. The country s the country of regstraton of the applcant frm. 0 Ths s a sample of,378 subsdares of multnatonal frms, see Secton 4 for detals of the data used. 4

6 Theoretcal Background The current lterature on the effect of EPL on productvty suggests that the nature of nnovaton has a role to play. The endogenous growth lterature (Aghon-Howtt, 998) emphasze the dfference between radcal and ncremental nnovaton. Where successful, radcal nnovaton requres a drastc adjustment of employment as the human captal of exstng workers s rendered obsolete. EPL ncreases ths cost by way of frng costs. Radcal nnovaton s more valuable than ncremental nnovaton and more costly. If there s uncertanty n future demand then EPL also has a postve effect on the returns to both types of nnovaton, n that t ncreases worker commtment and ther efforts n makng the new technology more productve through learnng by dong. EPL wll ncrease ncremental nnovaton effort, but at suffcently hgh levels t wll decrease radcal nnovaton effort. Thus frms wll be more lkely to choose to perform radcal nnovaton n low EPL regmes and ncremental nnovaton n hgh EPL regmes, whch s the central predcton tested n ths paper. The underpnnngs of ths model are based on Aghon-Howtt (998). Innovaton mproves the productvty of ntermedate goods suppled by a frm for use n the producton of a fnal good. A further mprovement on ths productvty gan comes va the effort (or learnng by dong) of producton workers. Ths effort s hgher n the presence of employment protecton legslaton (EPL), whch takes the form of hgher frng costs per worker, as producton workers are less lkely to be fred and therefore more lkely to share n the surplus from ncreased productvty. However, EPL can also have negatve effects on nnovaton, dependng on whether nnovaton s radcal or ncremental. Radcal nnovaton s more productve, but makes exstng human captal obsolete. The mplementaton of a radcal nnovaton requres 5

7 that all producton workers are replaced, at a per worker frng cost. Incremental nnovaton ncreases productvty, but to a lesser extent than radcal nnovaton, and exstng producton workers are retaned. EPL s effect on worker effort wll have an ncreasng effect on the returns to both types of nnovaton, but due to the frng costs t wll also have a negatve effect on the returns to radcal nnovaton. In ths paper our man nterest s n the mpact of EPL on nnovaton ncentves, where the man mpact of EPL s on costs, and therefore to focus on ths effect we assume away any strategc mpact of nnovaton n the product market.. Model A fnal good s produced usng a contnuum of ntermedate goods produced by frms, each one of whch s a monopolst n ts market, usng the technology, y j j ( Z ( e ) A ) = 0 x d, () where y s fnal output, ndexes frms (and ntermedate ndustres, snce each frm s j a monopolst n ts ndustry), j= 0, I, R ndexes nnovaton type, Z( ) e s the level of nvestment n unobservable effort made by producton workers, j A s the ntermedate producers productvty level, and x s other nputs to producton. Profts of the ntermedate frm are gven by, j j j ( e ) A π = δz, () j where π s profts and δ reflects the extent of competton n the ntermedate goods market. We consder the followng tmng of events: 6

8 Intermedate producers draw an ntal productvty level 0 A. Frms decde whether to nvest n radcal or ncremental nnovaton, and how much to nvest (whch determnes the probablty of success R I µ µ, ). If successful, ncremental nnovaton leads to a productvty ncrease of γ > and, f radcal nnovaton s successful productvty ncreases by a factor of γ. Innovaton ncurs a fxed cost. Productvty s enhanced by the efforts of workers. However, n the case of radcal nnovaton exstng workers do not have the requred sklls to work wth the new technology and must be fred and replaced by more sklled workers. Producton workers decde the level of nvestment n unobservable effort j productvty by a factor Z ( ) e e j, whch ncreases. A demand shock occurs whch leads to the possblty of the worker beng fred. We assume that the future uncertanty n demand s small enough to be trval to the frm, although of mportance to the workers. Intermedate producton occurs, f the frm chooses ncremental nnovaton then they use exstng workers. If the frm chooses radcal nnovaton then exstng producton workers are fred at cost ϕ per worker. They are replaced at zero hrng costs by producton workers wth more approprate sklls. Output s sold and the surplus shared between the frm and ts workers, dependng n part on (exogenous) worker Ths mples that nnovaton and producton are co-located, so the effect EPL has on worker ncentves affects the frm s nnovaton ncentves. Such a co-locaton s more lkely when technology transfer costs are hgh relatve to product transport costs (see Ekholm and Hakkala 007). Ths s consstent wth a model where locaton s endogenous and determned by the effect EPL has on the benefts to nnovaton. That s, f transport costs are low so that producton can be located anywhere, frms may choose to locate nnovaton and producton n countres where the labour market envronment s conducve to ther ntended type of nnovaton. 7

9 barganng power β. We are nterested n the nnovaton ncentves for the ntermedate producers. To solve for the mpact of frng costs on frms' ncentves to nnovate we solve the problem by backward nducton: Output generates surplus for the frm. These are gven for each j technology by, where V V V ( β ) π 0 I I I 0 I I ( β ) π + ( µ )( β ) π c F R R R 0 ( β ) π f + µ β π = (3) 0 I R = µ (4) = R R ( ) ( )( ) c F µ (5) j j V s frm s surplus usng technology j, µ s the level of nnovaton effort by the frm, c j s varable costs of nnovaton for j technology, f s frng cost ncurred f radcal nnovaton s successful, and j F s fxed costs of nnovaton for j technology. Varable costs take the form, j ( ) j j c = A µ. (6) Intermedate producton occurs. Output of the ntermedate frm s gven by equaton (). If the frm has chosen not to nnovate or chosen ncremental nnovaton then t uses exstng workers. If the frm chose radcal nnovaton then exstng producton workers do not have the sklls to work wth the new technology and are fred by the frm. EPL s modeled as a frng cost of ϕ per worker (a bureaucratc cost, not a 8

10 transfer to the worker), that makes employment adjustment costly. These frng costs take the form, ( e ) A 0 f = ϕkz (7) where ( ) kz e A s the number of exstng workers employed by the frm. 3 New workers are hred at zero hrng costs. Demand shock occurs. There are shocks to demand whch mean that the worker wll be fred wth probablty s ( ϕ). Ths occurs after the worker has commtted to an effort level. We assume that the future uncertanty n demand s small enough to be trval to the frm, although of mportance to the workers (see Acemoglu 997, and Boer and Jmeno 005). The frng cost of ϕ per worker makes t more lkely that employment adjustment n the face of demand shocks s unproftable to the frm and, therefore, ( ϕ), s ( ) < 0 s = s ϕ ncreases ther effort.. In ths way EPL ncreases workers job securty, and therefore There are condtons where EPL wll be rrelevant to frm locaton, specfcally when EPL takes the form of a redundancy payment rather than a bureaucratc cost to the frm. Pssardes (00) and Lazear (990) show that redundancy costs are rrelevant to the frm locaton decson f wages are determned endogenously. The worker takes nto account both the probablty of frm bankruptcy and the sze of the redundancy payment when barganng over wages. We have assumed ths stuaton away by nterpretng EPL as regulaton that results only n a (bureaucratc) frng cost to the frm and not a transfer to the worker. However, EPL as redundancy wll affect locaton decsons f wage barganng s conducted at the ndustry level rather than at the frm level and the probablty of bankruptcy s prvate nformaton to the frm and s dfferent to the ndustry average (Haaland and Wooton 003). The worker, takng nto account the probablty of recevng a redundancy payment, accepts a low (hgh) wage f the ndustry average rskness s hgh (low). Therefore a frm that s more rsky than the average s worse off, as t stll has to pay the same wage as other frms but has a hgher probablty of payng a redundancy payment. Therefore rsky frms (or frms more lkely to make employment adjustments) have an ncentve to locate ther actvtes n a low EPL country. 3 Let k =. The number of exstng workers comes from smple proft maxmsaton. 9

11 Producton workers decde level of effort. Ths ncreases productvty by a factor j j j Z ( ) (where Z ( 0 ) =, Z ( e ) > 0, Z ( e ) < 0 e maxmse ther expected return, max j e j j [( s( )) βπ + s( ϕ) 0 e ] ). Workers wll choose effort to ϕ (8) j j We assume Z takes the form Z ( e ) = e +, so that Z dsplays dmnshng returns to workers effort and s equal to one f workers make zero effort. Usng ths, substtutng equaton () nto (8) and performng the maxmzaton we obtan an expresson for the worker s optmal effort, j* e : Z ( ) j j s ( ) ( ϕ) e * βδ = A. (9) Worker effort for target nnovaton type j s ncreasng n the productvty of that nnovaton, and ncreasng n EPL, Z j* ( e ) s( ϕ) ϕ = j βδa ϕ > 0, (0) ϕ snce s ( ) < 0,.e. the probablty of beng fred decreases wth EPL. Frm decdes level of nnovaton. The problem facng the frm s to choose the optmal level of nnovaton effort, condtonal on the type of nnovaton and on worker effort. For ncremental nnovaton the frm chooses nnovaton effort, substtuton of equaton () nto equaton (4)): max µ I I µ, such that (from I I* I I 0* 0 I I [ µ ( β ) δz( e ) A + ( µ )( β ) δz ( e ) A c F ]. () 0

12 Usng equaton (6), the fact that nnovaton effort wll be: 4 A I 0 I* 0* = γa and Z ( e ) = γz ( e ) the frm s optmal 0* ( β ) γ δz( e ) I* µ =. () γ Ths effort s ncreasng n EPL, snce learnng-by-dong s ncreasng n frng costs, as stated n equaton (0), and γ >. To nvestgate how radcal nnovaton vares wth frng costs we substtute equatons (), (6) and (7) nto equaton (5) and, usng the fact that A R 0 = γ A, we obtan: 5 R* 0 0* 0 R 0* 0 0 R R [( β ) δz ( e ). γ A ϕkz( e ) A ] + ( µ )( β ) δz ( e ) A γ A ( ) F R max µ R µ µ. (3) Solvng as before, the frm s optmal radcal nnovaton effort wll be: R* δ ϕk 0* µ = ( β ) δγ Z ( e ). (4) γ γ Innovaton ncentves are ncreasng n workers learnng-by-dong effort and therefore EPL has an ncreasng effect for both types of nnovaton. Due to the large employment adjustment requred n the case of radcal nnovaton, frng costs also have a decreasng effect on the ncentves for radcal nnovaton. 4 0* 0 0* 0 0 I* Maxmsaton of equaton () gves ( β )( δz ( e ) γ A δz ( e ) A ) γa µ = 0 after rearrangement results n equaton (). 5 Maxmsaton of equaton (3) gves 0* 4 0 0* 0 0* 0 0 R β δz e γ A δz e A ϕkz e A γ A µ * ( )( ( ) ( ) ) ( ) = 0 gves equaton (4)., whch, whch after rearrangement

13 . Robustness to assumptons The dea that EPL ncreases worker effort n makng nnovaton more productve s robust to changng a number of the assumptons of the model. For nstance, we have assumed that the workers return to learnng-by-dong effort s entrely ted to the frm,.e. ther efforts enhance the productvty of the frm s captal but do not enhance ther own productvty. However, say the worker ganed from ther efforts by way of acqurng general sklls. Becker (964) predcts an under-nvestment n general sklls, as workers are credt constraned and frms are reluctant to fund sklls that the worker may use elsewhere. As descrbed by Acemoglu (997) t s lkely that a contract could be wrtten to mtgate such a problem (penaltes for workers who tran and qut) and, for our purposes, t s not ntally clear what role EPL has to play: EPL wll not stop workers leavng once traned and offered a job elsewhere. Acemoglu (997) consders a model of tranng and nnovaton wth job market search frctons, where workers can exogenously lose ther job wth probablty s. 6 Costly job search means that when a worker and frm are matched they bargan over the surplus of the match, and therefore over any ncreased productvty that the worker has acheved through learnng-by-dong effort. Ths leads to an under-nvestment n tranng by workers, as there s a probablty of beng fred and then, after search, recevng only a part return to ther tranng efforts. Where EPL reduces ths probablty of beng fred, t wll mtgate ths problem of under-nvestment, whch would be qualtatvely consstent wth our model. We have also assumed that the worker s effort s unobservable, otherwse the frm and worker could wrte a contract specfyng e n return for a guaranteed wage n each 6 Our equaton (9) s nspred by equaton () n Acemoglu (997).

14 perod. We could relax ths assumpton and assume that such a contract can be wrtten and that there s a montorng technology avalable to the frm so that a worker can be caught shrkng wth some probablty. The effcency wage pad to the worker so that they do not shrk s ncreasng n the exogenous probablty of spontaneous dsmssal n the future ( economc dsmssal ), ncreasng n the exogenous probablty of once dsmssed gettng another job ( flow nto employment ) and decreasng n the probablty of gettng caught shrkng and subsequently beng dsmssed ( dscplnary dsmssal ), as n Shapro and Stgltz (984). EPL can then have two effects: t wll decrease the probablty of economc dsmssal, but t wll also decrease the probablty of dscplnary dsmssal. Boer and Jmeno (005), argue that for large frms (whch s what we consder n our emprcal applcaton), where montorng s very dffcult, the domnant effect of EPL s that t decreases the probablty of economc dsmssal and therefore ncreases the value of employment to the worker and reduces the effcency wage that the frm must pay. As n our model, EPL wll ncrease the frm s nnovatve effort, snce the lower effcency wage wll ncrease the return to the frm from nnovaton..3 Relevant ranges for frng costs to affect nnovaton effort Specfyng a functonal form for s ( ϕ) allows us to nfer how optmal nnovaton effort vares over reasonable ranges of frng costs. Specfcally, consder the case h l where there s a probablty p of a drop n demand from to then (see Appendx A.): h s ( ϕ ) = p. ( ϕ ). (5) l 3

15 ϕ Ths functon s decreasng n ϕ at an ncreasng rate: s ( ϕ) < 0, s ( ) < 0. Usng a lnear approxmaton for equaton (5) (see Appendx A.) and, along wth equaton (9), we note that the expresson for radcal nnovaton effort, equaton (4), s quadratc n frng costs. Furthermore, radcal nnovaton effort ntally ncreases wth ϕ and then decreases wth ϕ. Ths functonal form for s ( ϕ) s not necessary for the qualtatve predctons of our model, but t wll help n dscussng the domnant effect of EPL on radcal nnovaton effort for realstc values of ϕ. In Appendx A. we show that, usng our approxmaton for s ( ϕ) : I* µ = ϕ γ 0 ( β ) γ δ pbβδa (6) R* µ ϕ = γ ϕk γ ( β ) γ δ pbβδa pbβδa (( p + p( ) b) + pbϕ ) βδa k γ (7) Equaton (6) s strctly postve: hgher EPL, modeled as hgher frng costs, unambguously ncreases optmal ncremental nnovaton effort. In equaton (7) the frst term s postve for γ >, and the second two terms are negatve and ncreasngly so n ϕ. To fnd the pont at whch frng costs start to have a negatve effect on radcal nnovaton, R ϕˆ, set equaton (7) equal to zero and solve to obtan: 7 R ˆ ϕ 4 ( )( ) ( p) β γ = pb ( ). (8) 7 R 4 We obtan ˆ ϕ = [( β )( γ ) δpb k( p + p( ) b) ] δ k kpb = becomes equaton (8) after some rearrangement. 4, whch usng

16 Therefore, R ϕˆ s lower when the productvty gans from nnovaton are low (when γ s smaller), and when the frms gets a low proporton of the return to nnovaton (worker barganng power, β, s hgh). Also, R ϕˆ s lower when the extent to whch ϕ ncreases learnng-by-dong s lower: that s, when the probablty of a negatve demand shock, p, s low and therefore the relevance of EPL n makng workers feel secure n ther jobs s lower; and when the elastcty of fnal good output wth respect to ntermedate good nput s low (.e. ( ) s small) as ths reduces the ntermedate good adjustment requred n the face of a small demand shock and therefore the possblty of gettng fred and the relevance of EPL to job securty (see Appendx A.). We can show that for reasonable values of the parameters n our model we always expect to be n the range where EPL has a negatve mpact on ncentves for radcal nnovaton, that s R ϕˆ s less than the lower lmt of the lkely range of ϕ. The frng cost ϕ, whch s a bureaucratc cost of dsmssng a worker, s lkely to be no greater than the worker s reservaton wage whch s normalzed at one. Therefore, t s realstc to assume that ϕ s between zero and one. Settng R = /, b =, p = 0., β = / we can calculate that ˆ ϕ < 0 for γ. 8 and that for.8 < γ. 968, 0 < ˆ ϕ R <. 8 In ths second range of γ values, EPL ncreases the value of radcal nnovaton ntally, but wll start to decrease t agan as the radcal frng cost effect starts to outwegh the learnng-by-dong effect. Rememberng that n 8 By nsertng values nto the followng expresson and solvng for γ : 4 ( γ ) = ( β ) + ( p) pb + RI ( ) + ˆ ϕ 5.

17 ths model the productvty gan from an ncremental nnovaton s γ and that for a radcal nnovaton s γ, the values just mentoned are very large: γ =. 8 corresponds to a productvty gan from ncremental nnovaton of 8. percent and from radcal nnovaton of 3.6 percent. Therefore, n ths model t s lkely that R ˆ ϕ < 0 and, therefore, frng costs have a strctly decreasng effect on radcal nnovaton ncentves. Also, the model predcts that frng costs have a strctly ncreasng effect on ncremental nnovaton..4 Emprcal predctons The theoretcal dscusson suggests two emprcal predctons that we can take to the data: Predcton : Frm nnovaton actvty s hgher n regmes wth hgher EPL. Predcton : The proporton of nnovaton performed by frms that s radcal (and wll lkely requre sgnfcant adjustments n employment) s hgher n regmes wth low EPL. 3 Emprcal strategy In order to nvestgate the two emprcal predctons outlned above we consder the decsons of multnatonal frms over where to locate nnovatve actvtes across twelve OECD countres. As emphassed above, our dentfcaton strategy s to use varaton wthn multnatonal frms, whch allows us to control for a large range of potentally unobservable characterstcs at the frm, ndustry and home country level. We use data on multnatonals from countres. We consder both the total level of nnovatve actvty, and the level of actvty on the most technologcally new projects, whch we nterpret as beng those most assocated wth employment adjustment and volatlty (we show evdence to support ths nterpretaton). 6

18 Our man measure of the level of nnovatve actvty s a count of patents. We follow the lterature (Hausman et al (984), Pakes (986), Blundell et al (999)) and model the count of patents wth a lnear exponental model. Consder a multnatonal frm (m), wth a number of subsdares (s) each of whch operates n (a potentally dfferent) ndustry () and s located n country (c). We model the level of nventve actvty, measured by patent applcatons (P), n each locaton as a functon of EPL, a range of other covarates (X), multnatonal effects (η ) and an dosyncratc error (u): P = ( β EPL + X + η + u ). (9) ms exp c c m ms Our nterest s n the sgn and magntude of β. Recall from the dscusson above that the theoretcal lterature s ambguous about what we expect the sgn to be - a postve sgn would suggest that the domnant effect of EPL s to ncrease both frms nvestment n workers and worker commtment, whle a negatve sgn would support the dea that hgher EPL makes employment adjustments more costly. Whle the theoretcal lterature s ambguous about the mpact of EPL on the overall level of nnovatve actvty, t clearly ponts to a detrmental effect of EPL on more technologcally advanced or rsky nvestments. To emprcally nvestgate ths predcton we estmate NPL ms ( β EPL + γ CITWP + X + η + ν ) = exp ln (0) c ms c m ms where NPL s a weghted count of patents that gves a greater weght to patents that are more technologcally advanced. More specfcally, NPL s a count of patents weghted by the number of ctatons made to non-patent lterature, manly scentfc journals (the specfc defnton s dscussed further n the next secton). CITWP s the count of all patents weghted by all ctatons made, to control for dfferences across patents n the amount of ctatons made. Agan, our nterest s the sgn and magntude of β - a negatve sgn would ndcate that hgher technologcally advanced patentng, as a proporton of overall patentng, s assocated wth lower EPL. 7

19 One concern we mght have wth estmatng equatons (9) and (0) s that EPL s correlated wth other nsttutonal varables that also affect nnovaton ncentves. We control for country level measures of other labour market regulatons, a measure of product market regulaton, concentraton n the bankng sector and a measure of the effcency of the courts, all of whch may affect the value of holdng a patent. 9 Full defntons of all varable and sources are n Table A. Some of these varables are hghly correlated wth EPL, as we can see from the correlaton matrx n Table, and t s therefore challengng to separately dentfy the effect of EPL, as wth all studes usng cross-country varaton n ths way. EPL s partcularly hghly correlated wth collectve barganng coverage and barganng coordnaton, whch have been found to be determnants of worker barganng power, 0 and wth the OECD overall measure of product market condtons. Nevertheless, our results are robust to controllng for these nsttutonal varables. A further concern we mght have s that dfferences n country-ndustry specalzaton may nfluence our results. The trade lterature emphasses that countres wth a large endowment of captal or sklls have an advantage n ndustres that are captal or skll ntensve, whch may nclude hgh-tech ndustres. We follow Nunn (007) and use captal abundance and nvestment n sklls at the country level, nteracted wth estmates of ndustry captal and skll ntensty. Another concern s that country sze may be correlated wth EPL, and producton actvty locates n large countres to access the product market, and where ths producton s hghly sklled t drves up 9 See Aghon et al. (005) for the effect of product market regulatons that determne competton on nnovaton. See Carln (003) for the effect of bankng concentraton on specalsaton n hgh tech nnovatve sectors. 0 See Calmfors and Drffl (988), Flanagan (999) and Grffth, Harrson and Macartney (007) for further dscusson and evdence that coordnaton ncreases worker barganng power. 8

20 wages for hgh sklled workers n those countres (e.g. see Ekholm and Hakkala, 007). As market access s less mportant for R&D ths may crowd out hghly sklled nnovaton to smaller countres. To control for country sze we nclude populaton. These consderatons lead to the followng structure for X c : X c k + kkc + 3Kc + h + 4h H c + 5H c + 6 = Pop, () c where k s the captal per unt output n ndustry based on US data (the US s not n the sample), ntensty of ndustry, K c s the natural log of the captal per worker n country c, h s the skll H c s the natural log of the proporton of GDP spent on hgher educaton n country c, and Pop c s the workng populaton of country c averaged over the sample perod. 4 Data In order to estmate equatons (9) and (0) we need nformaton on the geographc locaton and level of technologcal sophstcaton of multnatonal frms nnovatve actvty, along wth nformaton on EPL and other country and ndustry characterstcs. 4. Measurng the nnovatve actvty of multnatonal frms The data on patents come from the European Patent Offce PATSTAT dataset, whch we have matched to nformaton on corporate ownershp structure and fnancal accounts from BVD Amadeus (see Macartney 009). Patent applcatons fled at the European Patent Offce (EPO) are an attractve measure of nnovatve actvty for a number of reasons. The advantage of ths measure s that t s admnstratve n nature wth well defned rules that are ndependent of the locaton of the patent applcant. Furthermore, t s measured at the frm-locaton level (n contrast to data on frm level 9

21 R&D expendture, whch s not wdely avalable for frms n many European countres, and where t s reported t s almost always at the world-wde level). Patents data has been wdely used and found to be closely related to R&D expendture measures (see Grlches et al. 984 and 990), and ths s also true for our data at the ndustry level (see Macartney 009). There are of course also drawbacks to usng patents as a measure of nnovatve actvty, ncludng that frms n dfferent ndustres and countres have dfferent propenstes to patent, and that the value of a patent s heterogenous across frms. Our dentfcaton strategy, of lookng wthn the frm, helps to control for many of these potental drawbacks. Our man sample contans,378 subsdares of 343 multnatonal frms. These subsdares fle at least one patent that makes at least one ctaton. Condtonng on only patents that make a ctaton s convenent for estmatng the functonal form of equaton (0), whch ncludes as a regressor the log of the total number of ctatons made by patent held by a subsdary. In the results secton we show that the results we fnd for equaton (9) are smlar for the larger sample that also ncludes subsdares that fle patents that make no ctatons. Table shows how the frms and patent applcatons are dstrbuted across countres. Column () shows the total number of frms (ncludng domestc and foregn subsdares) n each country, and column () shows the total number of patents fled by corporatons. Column (3) shows how the,378 subsdares of multnatonal frms that make up our baselne sample are dstrbuted across countres and column (4) gves the count of ther patent applcatons. The baselne sample ncludes all patent Voth AG was removed form the sample due to concerns about the data. 0

22 applcatons whether or not they have been granted (we show the results are robust to consderng only granted patents). To estmate equaton (9) we measure nnovatve actvty as a smple count of patents (P). We use smple counts rather than weghtng patents by ctatons receved as many of the patents are relatvely new and therefore ctatons are severely truncated. However, our results are robust to usng ctatons weghted patents, suggestng that the effect s sgnfcant for economcally valuable patents. To estmate equaton (0) we measure radcal nnovaton actvty ( NPL ms ) as a count of non-patent lterature (NPL) ctatons made by patents fled by subsdary s n multnatonal frm m over the sample tme perod, and we control for the total number of ctatons made by the same patents. Ths measure s an ndcator of the newness of the nnovaton, snce NPL ctatons are typcally ctatons to scentfc journals. Table 3 shows how the proporton of all ctatons made that are to NPL vares across ndustres. We can see that ndustres whch we mght expect to requre hghly scentfc nnovaton, such as food producton, transport and communcatons, fnance and chemcal (ncludng pharmaceutcals) have the hghest proporton of NPL ctatons, and ndustres whch we mght expect to nvolve less scentfc nnovatons, such as lght manufactures, have the lowest proporton of NPL ctatons. Our nterest n ths paper s on the effect of labour market regulatons that affect job securty for workers and adjustment costs for employees. Increased job securty ncreases worker ncentves to nvest n nnovaton and therefore ncreases the return to nnovaton for employers. However, where nnovaton s uncertan or sgnfcantly new, n that t requres an adjustment n the skll mx of employees whch may nvolve the replacement of exstng workers wth external workers, regulatons that protect exstng employment ncrease the cost of nnovaton. Our expectaton s that the

23 second effect wll domnate the frst when nnovaton s sgnfcantly technologcally advance, as measured by the proporton of ctatons to the scentfc lterature (NPL). In Table 4 we show a number of correlatons that supports the approprateness of ths measure. We start n column () by showng the postve correlaton between hgh NPL ctatons and the average number of nventors per patent, a possble ndcator of the complex nature of the technology. Column () shows that NPL nnovaton s postvely correlated wth employment adjustment wthn frms and column (3) that NPL nnovaton s correlated wth country-ndustry sales volatlty, a measure of uncertanty. 4. Employment Protecton Legslaton We use an ndex of EPL from the OECD (OECD Economc Outlook 999 Chapter ) and wdely used n the lterature on the determnants of unemployment (see, nter ala, Nckell et al. 005, Ncolett et al 000). Our preferred measure s the ndcator of the legslaton relatng to regular contracts (coverng procedural nconvenences, drect cost of dsmssal, notce and tral perod). Our results are also robust to usng the hgher level ndcator that also ncludes legslaton for temporary contracts (coverng types of work admssble under temporary contracts and maxmum cumulatve duraton allowed). Key for our purposes s that there s real varaton n ths measure across the countres n our sample, as s evdent from Fgure. 5 Results To recap, we have derved emprcal predctons of an effect of EPL on nnovaton that s dfferental across the nature of nnovaton. On the one hand EPL may ncrease overall patentng, but on the other hand t may reduce radcal nnovaton. We consder

24 whether there s emprcal support for these predctons, controllng for multnatonal frm fxed effects, country and ndustry characterstcs. 5. Man results The results for total nnovaton are presented n Table 5. Column () shows results for a smple specfcaton wth multnatonal frm fxed effects. The postve coeffcent on EPL ndcates that, wthn multnatonal frms, more nnovaton s performed by subsdares n countres wth hgh employment protecton for workers. To ensure that the result s not drven by patterns of ndustral specalzaton n countres wth abundant captal, column () ncludes as control varables the captal ntensty of the ndustry n whch each subsdary resdes, the captal abundance of the country n whch t s regstered and the nteracton between these two varables. The coeffcent on the nteracton of captal abundance and captal ntensty s postve as we would expect, and the coeffcent on EPL remans postve and statstcally sgnfcant. Also ncluded s the workng populaton of each country to ensure that the result that total patentng occurs more n countres wth hgh EPL s not drven by market sze effects that may be correlated wth employment regulaton. To ensure that our results are not drven by patterns of ndustral specalzaton n countres wth abundant sklls, column (3) ncludes as control varables the skll ntensty of each ndustry n whch subsdares resdes, the skll level of the country n whch t s regstered and the nteracton between these two varables. Although the coeffcent on ths nteracton s negatve, counter to ntuton, column (3) shows that our results are not drven by patterns of comparatve advantage wth relaton to sklls. Column (4) ensure that our result s not drven by other natonal regulatons, by ncludng a set of product market, We do not have data on hgher educaton expendture n Portugal so two observatons drop out. 3

25 fnancal market and legal varables, as well as other labour market regulatons. However, as shown n column () of Table, many of these varables are hghly correlated wth our measure of EPL. Ths, and the fact that the coeffcent n column (4) ncrease substantally n magntude relatve to the prevous three columns and slghtly loses statstcal sgnfcance, suggests problems of mult-colnearty n ths specfcaton. Table 6 show the results for radcal nnovaton. The dependent varable s the number of ctatons to the non-patent lterature, wth the log of the total number of ctatons as a control, so that we can nterpret the results as the effect of EPL on the proporton of ctatons that are to NPL. In column () the negatve coeffcent on EPL ndcates that, wthn multnatonal frms, the more technologcally advanced nnovaton s performed by subsdares n countres wth low employment protecton for workers. Column () presents the results wth controls ncluded for ndustry captal ntensty, natonal captal abundance and ther nteracton. The coeffcent on the nteracton of captal abundance and captal ntensty s postve, as we would expect. The coeffcent on EPL remans negatve and statstcally sgnfcant at the 0% level. Agan, the workng populaton of each country s ncluded as a control for market sze effects. Column (3) ncludes control varables for ndustry skll ntensty, country skll abundance and ther nteracton. The coeffcent on ths nteracton s postve as expected, but s not statstcally sgnfcant. Nevertheless, the coeffcent on EPL n column (3) remans negatve and statstcally sgnfcant. Column (4) ensures that our key result s robust to the ncluson of regulatory varables n product, fnancal and labour markets, as well as the effcency of legal nsttutons. The coeffcent on EPL remans negatve and statstcally sgnfcant at the 0% level. 4

26 5. Robustness The specfcaton so far only used data on subsdares that fle patents that make ctatons. However, we can use a larger sample that also ncludes subsdares that fle patents that make no ctatons, as well as those that fle patents that do make ctatons, to look at the mpact of EPL on total nnovatve actvty. The coeffcents (standard errors) for ths sample of,870 subsdares for the equvalent specfcatons shown n Table 5 are shown n Table 7. The results reman robust. Weghtng patents by the number of ctaton that they receve captures ther economc mportance. The sample of patents that we have used n ths paper have been fled relatvely recently, and therefore ctatons are truncated. Nevertheless, we re-run the regressons from column () n Table 5, weghtng the patent count by ctatons so far receved. The resultng coeffcent (standard error) on the EPL varable s 0.57 (0.94). Also, the regresson was re-run usng the total number of ctatons made as the dependant varables, n whch case the coeffcent (standard error) on the EPL varable s 0.34 (0.33). The results thus far have reled on patent flngs rrespectve of whether or not those patents have been granted. The motvaton for usng all patent flngs s agan the relatve youth of the patents n the sample. Patents may not have been granted yet, although they wll be n the future. Furthermore, the length of tme t takes for a patent to be granted may be related to the nature of that nnovaton,.e. how radcal the nnovaton s, and therefore condtonng on only granted patents may ntroduce nonclasscal measurement error nto the dependant varable. Nevertheless, our results are broadly robust to condtonng on only patents that have been granted. Ths reduces the sample to just 80 observatons. The coeffcent (standard error) for the specfcaton prevously presented n column () of Table 5 when we use only granted 5

27 patents s (0.49) and the coeffcent (standard error) for the specfcaton prevously presented n column () of Table 6 when we use only granted patents s (0.086). It s not possble to dentfy the effect of EPL on nnovaton and nclude sklls, captal and other regulatory varables on ths reduced sample. We use a measure of employment protecton that concerns only regular full tme contract. We can also use a measure that ncludes the protecton afforded to temporary workers as well as regular workers, and ncludes a measure of protecton aganst collectve dsmssals. The results that employment protecton ncrease ncremental nnovaton by frms and decreases radcal nnovaton s robust to usng ths hgher level measure of employment protecton as shown n columns () and columns (4) of Table 8. Columns () and columns (5) show the two specfcatons usng the employment protecton legslaton afforded to temporary workers as the varable of nterest and columns (3) and (6) shows the results wth both EPL for regular contracts and EPL for temporary contracts ncluded as separate varables. The fact that the coeffcent on EPL for regular contracts s n general the larger and more statstcally sgnfcant suggests that t s the rghts of permanent employees that affect frms ncentves wth regard to nnovaton, as we would expect gven the ntuton of the model presented n ths paper. 5.3 Economc Sgnfcance What s the economc sgnfcance of these estmates? To consder ths we look at the mpact of movng each country to the mean EPL ndex of.3. Consder countres such as Italy and Germany, whch have relatvely strong employment protecton legslaton wth an EPL ndex of.8. Reducng ther EPL to the mean n our sample of.3 would result n approxmately a 0% fall n overall patents (usng the coeffcent estmates n 6

28 column () of Table 5 evaluated at the mean level of patentng), but an ncrease n radcal nnovatons of around 5%. On the other hand, consder a country lke Denmark wth a low amount of employment protecton, whch has an EPL ndex of.6. Increasng ther EPL ndex to.3 would lead to an ncrease n overall patentng of around 37%, but a fall n radcal nnovatons of around 6%. These are substantal effects. 6 Concluson Ths paper has nvestgated the relatonshp between employment protecton legslaton and nnovaton actvty across twelve European countres. We use new data on the actvtes of multnatonal frms operatng across dfferent jursdctons. Our fndngs suggest that multnatonal frms do more ncremental patentng actvty n hgh EPL countres and more radcal patentng actvty n low EPL countres. Ths s consstent wth a varant of an Aghon-Howtt style growth model that ncorporates the two effects of EPL - ncreases job securty for exstng workers, and thus ncreased effort, and ncreased frng costs leadng to hgher adjustment costs for the frm. As a fnal caveat, however, we should note that our emprcal fndngs are also consstent wth other theoretcal models, such as Sant-Paul s model of comparatve advantage, and wth the deas put forward n Hall and Soskce. We are not able to emprcally dstngush these alternatve models. Care must be taken n nterpretng these results. Whle we have attempted to control for a number of other characterstcs that vary across countres, and for frm specfc characterstcs, dentfcaton s stll from cross-sectonal data. We do not observe suffcent tme seres varaton n EPL and our data to dentfy the effects of changes n labour market regmes. Nonetheless, 7

29 ths evdence s suggestve and appears to be robust to a number of standard concerns put forward n the lterature. 8

30 Appendx A. In a smlar ven to Boer and Jmeno (005) consder that there s a small probablty p that demand for the fnal good wll drop from hgh, S level of worker effort output s then gven by y = ( ZA ) 0 h, to low, l. 3 For a gven x d, where S = h, l. 4 On the realsaton of the demand shock the frm wll wsh to adjust employment from h x to the new optmal level x l by frng workers. The probablty of beng fred for each worker s then gven by: h l x x s = p.. (A) h x In the presence of EPL t costs the frm ϕ per worker to adjust employment downwards. The loss to the frm of a non-optmal level of employment, x, s gven by: l l l ( ( ) x ) x ( ZA ) ( x ) ( ) x l π = ZA, (A) where the frst term s the level of profts gven low demand but wth employment l x > x and the second term s the level of profts gven low demand and the optmal level of employment. When ϕ h ( x x ) x =, π = 0. The frm faces frng costs gven by l x. The frm wll adjust the employment level untl the margnal gan from 3 Ths s smpler than Boer and Jmeno (005) n that we consder that demand s normally hgh, but there s a small possblty that t drops. The frm ntally chooses employment levels assumng demand wll be hgh. 4 For smplcty the j superscrpt and the varable for workers effort, e j, are omtted from the dscusson n ths secton. 9

31 30 dong so equals the margnal cost of frng an employee. The optmal level of employment, xˆ, gven frng costs s then gven by: ( ) ϕ = ˆ l x ZA (A3) Therefore, l x ZA ˆ = ϕ. (A4) Ths expresson s ncreasng n ϕ. Note that t reduces to l x n the absence of frng costs ( 0 = ϕ ). There wll also exst some level of ϕ where no adjustment occurs. Substtutng ths nto A gves the probablty of beng fred faced by each worker n the presence of frng costs: ( ) = =.. ϕ ϕ l h h l h p ZA ZA ZA p s. (A5) Ths probablty decreases as h l as we would expect. 5 Wrtng out ( ) ϕ s followed by ts frst and second dervatves, gves: ( ) ( ) =. ϕ ϕ l h p s, (A6) 5 The probablty of beng fred s non-postve when h l =. Note that <.

32 3 ( ) ( ) ( ). = ϕ ϕ l h p s, (A7) ( ) ( ) ( ) ( ) 3. = ϕ ϕ l h p s. (A8) As < and restrctng [ ) 0, ϕ we have ( ) ( ) 0, 0 < < ϕ ϕ s s, that s the probablty of a worker losng ther job s decreasng n ϕ and at an ncreasng rate. 6 Usng a Taylor expanson around 0 = ϕ we can now wrte A as: ( ) ( ) ( ) ϕ ϕ = l h l h l h p p p s (A9) Appendx A. From dfferentaton of equatons () and (4) n the man text, ncremental nnovaton and radcal nnovaton effort vary wth frng costs as descrbed n A0 and A. ( ) ( ) ϕ δ γ γ β ϕ µ = 0* * I e Z, (A0) ( ) ( ) ( ) ( ) 0* 0* 0* * R e Z k e Z k e Z γ ϕ γ ϕ ϕ δ γ γ β ϕ µ =. (A) 6 Restrctng frng costs to be between zero and one s natural here as the workers reservaton wage s normalsed to one and t s lkely that frng costs wll be some proporton of that. s tends to negatve nfnty as frng costs tend to one, but we just exclude ths and say that at some pont frng costs are so hgh that the frm does not adjust employment at all.

33 For small ϕ we can wrte (from equaton A9): h h s ( ϕ ) = p.. ϕ p ( ). (A) l l Insertng ths nto equaton (9) from the man text and lettng h l b = we ( ) ( ) can wrte: 7 Z 0 0 ( p( ( ) b bϕ) ) βδa = (( p + p( ) b) + pbϕ) βδa 0* ( e ) =. (A3) Inserton of expresson (A3) nto (A0) and (A), after trval manpulaton results n equatons (6) and (7) n the man text. 7 Note that b s decreasng n the severty of the shock. Its range s between zero and 3.

34 References Acemoglu, D. (997). Tranng and Innovaton n an Imperfect Labour Market, Revew of Economc Studes, vol. 64, pp Acharya, V., R. Bagha-Wadj and K. Subramaran (009) Labor Laws and Innovaton, CEPR Dscusson Paper, no. 77. Agell, J (999) On the Benefts from Rgd Labour Markets: Norms, Market Falures, and Socal Insurance The Economc Journal, 09 (453), Features (February), pp. F43-F64. Aghon, P., Bloom, N., Blundell, R., Grffth, R. and Howtt, P. (005). Competton and Innovaton: An Inverted-U Relatonshp, Quarterly Journal of Economcs, vol. 0, no., pp Aghon, P. and Howtt, P. (998). Endogenous Growth Theory, Cambrdge: MIT Press. Akerloff G (98) Labor Contracts as Partal Gft Exchange The Quarterly Journal of Economcs, vol. 97, no. 4, (November), pp Akkermans, D., Castald, C. and Los, B. (005). Do Lberal Market Economes Really Innovate More Radcally than Coordnated Market Economes? Hall and Soskce Reconsdered, Research Polcy, vol. 38, no., pp Audretsch, D. (995). Innovaton and Industry Evoluton, MIT Press. Bartelsman, E., Perott, E., and Scarpetta, S. (008). Barrers to Ext, Expermentaton and Comparatve Advantage, OECD Workng Paper. Bassann, A. and Ernst, E. (00). Labour Market Regulaton, Industral Relatons, and Technologcal Regmes: A Tale of Comparatve Advantage, Industral and Corporate Change, vol., no. 3, pp Becker, G. (964). Human Captal, Chcago: Unversty of Chcago Press. Blanchard, O. and Wolfers, J. (000). The role of shocks and nsttutons n the rse of European unemployment: the aggregate evdence, Economc Journal, vol. 3, pp. C 33. Blundell, R, R Grffth and J Van Reenen (999) Market Share, Market Value and Innovaton n a Panel of Brtsh Manufacturng Frms Revew of Economc Studes, vol. 66, pp Boer and Jmeno (005) The effects of employment protecton: Learnng from varable enforcement European Economc Revew vol. 49, pp Bresch, S., Malerba, F. and Orsengo, L. (000). Technologcal Regmes and Schumpeteran Patterns of Innovaton, The Economc Journal, vol. 0, pp Buchele, R. and Chrstansen, D. (999). Employment and Productvty Growth n Europe and North Amerca: The Impact of Labour Market Insttutons, Internatonal Revew of Appled Economcs, vol. 3, no. 3, pp Caballero, R., Cowan, D., Engel, E. and Mcco, A. (004). Effectve labor regulaton and mcroeconomc flexblty, Economc Growth Centre Yale Unversty, Dscusson Paper No

35 Calmfors, L. and Drffll, J. (988). Centralsaton of Wage Barganng and Macroeconomc Performance: A Survey, The Economcs of Unemployment Vol III, (ed) PN Junankar. Carln, W. and Mayer, C. (003). Fnance, Investment and Growth, Journal of Fnancal Economcs, vol. 69, pp Caves, R. (996). Multnatonal Enterprse and Economc Analyss. Cambrdge Unversty Press, Cambrdge. Cunat, A. and Meltz, M. (007). Volatlty, Labor Market Rlexblty, and the Pattern of Comparatve Advantage, CEP Dscusson Paper, No Devereux, M. and Grffth, R. (998). Taxes and the locaton of producton: evdence from a panel of US multnatonals, Journal of Publc Economcs, vol. 68, pp Dunnng, J., (977). Trade, locaton of economc actvty and MNE: a search for an eclectc approach. In: Ohln, B., Hesselborn, P.O., Wjkman, P.M. (Eds.), The Internatonal Allocaton of Economc Actvty. McMllan, London, pp Etrhem, O., Gerdup, K. and Klovland, J. (003). Credt Bankng and Monetary Developments n Norway , n Hstorcal Monetary Statstcs for Norway, Norwegan Central Bank. Ekholm, K. and Hakkala, K. (007). Locaton of R&D and Hgh-tech Producton by Vertcally Integrated Multnatonals, The Economc Journal, vol. 7, pp Flanagan, R. J. (999). Macroeconomc Performance and Collectve Barganng: An Internatonal Perspectve, Journal of Economc Lterature, vol. XXXVII (September), pp Grffth, R., Harrson, R. and Macartney, G. (007). Product Market Reforms, Labour Market Insttutons and Unemployment The Economc Journal, 7 (March), pp. C4 C66. Haaland, J. and Wooton, I. (003). Domestc Labour Markets and Foregn Drect Investment, CEPR Dscusson Paper, no Hall, P. and Soskce, D. (00). An Introducton to Varetes of Captalsm, n P.A.Hall and D. Soskce (eds.) Varetes of Captalsm; The Insttutonal Foundatons of Comparatve Advantage, Oxford Unversty Press, pp Hausman, J., Hall, B., Grlches, G. (984) Econometrc models for count data and an applcaton to the patents-r&d relatonshp Econometrca, vol. 5, pp Klepper, S. (996). Entry, Ext, Growth and Innovaton over the Product Lfe- Cycle, The Amercan Economc Revew, vol. 86, no. 3, pp Macartney, G. (009), Matchng Patents to Frms Accounts, Chapter, PhD Thess, UCL. Ncolett G., Scarpetta, S. and Boylaud, O. (000) Summary ndcators of product market regulaton wth an extenson to employment protecton legslaton, OECD Economcs Department Workng Papers, no

36 Nckell S., Nunzata L. and Ochel W. (005). Unemployment n the OECD snce the 960s. What do we know?, The Economc Journal, vol. 5, pp. -7. Lazear, E. P. (990), Job securty provsons and employment. Quarterly Journal of Economcs, vol. 55, pp Nunn, N. (007). Relatonshp-Specfcty, Incomplete Contracts, and the Pattern of Trade, Quarterly Journal of Economcs, vol., no., pp Pakes, Arel S (986) Patents as Optons: Some Estmates of the Value of Holdng European Patent Stocks, Econometrca, vol. 54, no. 4, pp Pssardes, C. A. (00), Employment protecton. Labour Economcs, vol. 8, pp Puga, D. and Trefler, D. (005). Wake Up and Smell the Gnseng: The Rse of Incremental Innovaton n Low-Wage Countres. CEPR Dscusson Paper, no Sant-Paul, G. (00). Employment protecton, nternatonal specalzaton, and nnovaton, European Economc Revew, vol. 46, pp Sant-Paul, G. (997). Is labour rgdty harmng Europe s compettveness? The effect of job protecton on the pattern of trade and welfare, European Economc Revew, vol. 4, pp Sant-Paul, G. (996). Dual Labor Markets: A Macroeconomc Perspectve, Cambrdge: MIT Press. Samanego, R. (006). Employment protecton and hgh-tech averson, Revew of Economc Dynamcs, vol. 9, pp Scarpetta, S. and Tressel, T. (004). Boostng Productvty va Innovaton and Adopton of New Technologes: Any Role for Labour Market Insttutons?, World Bank Polcy Research Workng Paper, no Shapro, C. and J. Stgltz (984) "Equlbrum Unemployment as a Worker Dscplne Devce," Amercan Economc Revew, vol. 74, no. 3, pp , Storm, S. and Naastepad, C. (007). Labour Market Regulaton and Productvty Growth: Evdence for 0 OECD Countres ( ), ILO, Economc and Labour Market Analyss Department, Workng Paper. 35

37 Fgure : EPL and Radcal Innovaton (all frms) Notes: The graphs shows the relatonshp between radcal nnovaton and EPL for the 37,350 prvate sector frms regstered n the countres presented that fled patents n the perod 997 to 003. These frms were responsble gor the flng of 30,3 patents n ths perod. The x-axs shows the average ndex for Employment Protecton Legslaton (Regular Contracts from OECD (XXXX) over the perod The y-axs shows the average proporton of ctatons made to the scentfc lterature by subsdares located n each country over the perod

38 Fgure : EPL and Radcal Innovaton by Multnatonal Frms Notes: Based on a sample of,378 subsdares of multnatonal frms, see Secton 4 for detals of the data used Ftted lne weghted by number of subsdares. The confdence nterval uses standard errors clustered at the country level. Fgure 3: EPL and Innovaton by Multnatonal Frms Notes: Based on a sample of,378 subsdares of multnatonal frms, see Secton 4 for detals of the data used Ftted lne weghted by number of subsdares. The confdence nterval uses standard errors clustered at the country level. 37