The Effect of Employment Protection Legislation and Financial Market Imperfections on Investment: Evidence from a Firm-Level Panel of EU Countries

Transcription

1 DISCUSSION PAPER SERIES IZA DP No The Effet of Employment Protetion Legislation and Finanial Market Imperfetions on Investment: Evidene from a Firm-Level Panel of EU Countries Federio Cingano Maro Leonardi Julian Messina Giovanni Pia April 2009 Forshungsinstitut zur Zukunft der Arbeit Institute for the Study of Labor

2 The Effet of Employment Protetion Legislation and Finanial Market Imperfetions on Investment: Evidene from a Firm-Level Panel of EU Countries Federio Cingano Bank of Italy Maro Leonardi University of Milan and IZA Julian Messina University of Girona, FEDEA and IZA Giovanni Pia University of Salerno and CSEF Disussion Paper No April 2009 IZA P.O. Box Bonn Germany Phone: Fax: iza@iza.org Any opinions expressed here are those of the author(s) and not those of IZA. Researh published in this series may inlude views on poliy, but the institute itself takes no institutional poliy positions. The Institute for the Study of Labor (IZA) in Bonn is a loal and virtual international researh enter and a plae of ommuniation between siene, politis and business. IZA is an independent nonprofit organization supported by Deutshe Post Foundation. The enter is assoiated with the University of Bonn and offers a stimulating researh environment through its international network, workshops and onferenes, data servie, projet support, researh visits and dotoral program. IZA engages in (i) original and internationally ompetitive researh in all fields of labor eonomis, (ii) development of poliy onepts, and (iii) dissemination of researh results and onepts to the interested publi. IZA Disussion Papers often represent preliminary work and are irulated to enourage disussion. Citation of suh a paper should aount for its provisional harater. A revised version may be available diretly from the author.

3 IZA Disussion Paper No April 2009 ABSTRACT The Effet of Employment Protetion Legislation and Finanial Market Imperfetions on Investment: Evidene from a Firm-Level Panel of EU Countries * This paper analyzes the joint effet of EPL and finanial market imperfetions on investment, apital-labour substitution, labour produtivity and job realloation in a ross-ountry framework. In the spirit of Rajan and Zingales (1998) and Cione and (2006), we exploit variation in the need for realloation at the setoral and aggregate level to assess the average effet of EPL on firms poliies. Then, exploiting firm-level information we study if the effet of EPL is stronger in firms with lower levels of internal resoures. We find that, on average, EPL redues investment per worker, apital per worker and value added per worker in high realloation setors relative to low realloation setors. The redution in the apitallabour ratio is less pronouned in firms with higher internal resoures, suggesting that finanial onstraints exaerbate the negative effets of EPL on apital deepening. JEL Classifiation: J21 Keywords: apital-labor substitution, labor market imperfetions, finanial market imperfetions Corresponding author: Maro Leonardi Dipartimento Studi del Lavoro e del Welfare Università Statale di Milano Via Conservatorio Milano Italy maro.leonardi@unimi.it * Paper presented at the 49th panel meeting of Eonomi Poliy in Brussels. The authors would like to thank the editor in harge, Jan van Ours, three anonymous referees, Samuel Bentolila, Marel Jansen, Juan Franiso Jimeno and Ernesto Villanueva, for useful omments and suggestions. This paper was partly prepared while Julián Messina visited the Bank of Spain, whose hospitality is greatly aknowledged. Julian Messina also thanks finanial support from the researh grant SEJ of the Spanish Ministry of Siene and Tehnology. The views expressed here are those of the authors and do not neessarily reflet those of the Bank of Italy or the Bank of Spain. The usual dislaimer applies.

4 Introdution A large literature has established the importane of Employment Protetion Legislation (EPL) in affeting job flows by reduing both workers hiring and firing. Given that EPL represents an obstale to the realloation of resoures, it is plausible that it also has a bearing on firms investment deisions, on the apital-labour ratio and, eventually, on produtivity. 1 However, if EPL affets the pattern of investment and apital aumulation, a natural question left so far unaddressed by the literature onerns the impat of finanial market imperfetions on firms response to more stringent employment protetion provisions. The ability to adjust the apital stok or to adopt new tehnologies in the fae of striter EPL is in fat likely to be different in firms that have aess to redit rather than finanially onstrained firms. Finanially onstrained firms may, for example, be unable to hannel all their internal resoures to produtive investments when an inrease in EPL raises labour osts and workers bargaining power. The purpose of this paper is to provide an empirial analysis of the effets of finaning onstraints on the adjustment of apital and labour in response to a hange in EPL. We analyze the joint effet of EPL and finanial market imperfetions on investment, apital-labour substitution, labour produtivity and job realloation in a ross-ountry framework. In the spirit of Rajan and Zingales (1998) and Cione and (2006), we exploit variation in the need for realloation at the setoral and aggregate level to assess the average effet of EPL on firms poliies. Then, exploiting firm-level information on the availability of firms internal resoures we build different variables measuring firms liquidity (operating ash-flow and net liquid assets) and study if the effet of EPL is stronger in firms with lower levels of internal resoures that are therefore more likely to be finanially onstrained. While there is a growing onsensus in the empirial literature around the idea that employment protetion regulations have important effets on employment adjustment, relatively little is known about the effets of employment protetion on investment, apital deepening and labour produtivity. Only reently these issues have reeived attention. Bassanini et al. (2009) look at the effet of EPL on job realloation and TFP, using industry-level data (EUKLEMS) and find a negative effet of EPL on TFP. Autor et al. study the US ase and, exploiting mirodata, find that after an inrease in EPL apital deepening inreases and TFP delines. 1 See Autor et al., Bassanini et al. (2009) and Cingano et al. (2008), disussed in the next setion. 2

5 This paper fouses on the role of finanial market imperfetions in shaping the response of firms to EPL. In our ase, the use of firm-level data is ruial not only to ontrol for firm-level heterogeneity but also and most importantly to ahieve identifiation of the effet of interest, namely the interation between firm availability of internal finane (finanial market imperfetions) and aggregate EPL (labor market fritions). To this aim, we exploit the availability of firm-level information from Amadeus data whih is, to the best of our knowledge, the only available soure of omparable firm-level information on balane sheets for ountries for whih we also have information on the nationwide level of EPL. One reason for the lak of studies on the effets of EPL on investment and apital deepening is that, while theoretial models offer lear preditions regarding the effets of adjustment osts on job turnover, they provide no guidane on the expeted effets of employment protetion laws on apital investment, the apital-labour ratio and produtivity. Moreover, both the theoretial and empirial literature are virtually silent on the interation between finanial markets and EPL, as disussed in the literature setion. In general the presene of dismissal osts will raise firms adjustments osts. For this reason firms may have inentives to distort their prodution hoies toward the more flexible input, thus substituting labour for apital. On the other hand, EPL may strengthen workers bargaining power and exaerbate the hold up problem typial of investment deisions, resulting in less investment per worker. Hene, for a given tehnology, stringent firing osts might result in less apital per worker. However, when firms an adapt their tehnologies, higher EPL should favour the adoption of more apital-intensive tehnologies in the longer run. The final result on investment (and therefore the apital-labour ratio in the long run) is ambiguous and may therefore depend on workers bargaining power and on the time span of the data. EPL will also typially have an ambiguous effet on labour produtivity: if dismissal protetions indue firms to retain (some) unprodutive workers, this will ause a deline in labour produtivity, eteris paribus. Offsetting this fator, employment protetion not only protets workers, it also indues human apital aumulation and thus improves produtivity (Belot et al. 2007). Furthermore firms may sreen new hires more stringently when firing osts are high, leading to a favourable ompositional shift in the produtivity of the employed workfore. The first part of this paper assesses the average effet of EPL on investment, the apital-labour ratio and labour produtivity using Amadeus data. Following Rajan and Zingales (1998), our es- 3

6 timates exploit both variation in the regulation aross ountries and the different relevane of the onstraints imposed by regulation on firms in different setors. We estimate the role of EPL looking at whether its impat is greater in industries where, in the absene of regulations, job realloation would be higher. Exploiting the possibility to alulate job flows in different ountries and industries from firm-level data, the intrinsi degree of volatility at the industry level is measured omputing industry job realloation in a hypothetial fritionless ountry with no employment regulation that faes world-average realloation shoks (Cione and, 2006). The analysis on firms hoies of apital and labour inputs shows that, on average, EPL redues investment per worker, apital per worker and measured labour produtivity (value added per worker) in high realloation setors relative to low realloation setors. The estimated effets are non-negligible. Reduing employment protetion from the level prevailing in Greee in 1997 to that observed in Ireland in the same year (this shift orrespond to the 90th to the 10th perentile of the ountry-by-year EPL distribution in our sample) inreases apital intensity in industries at the 10th relative to industries at the 90th perentile of the realloation distribution by 12%. The same exerise would imply an inrease in investment per worker and in value added per worker in high relative to low turnover industries by 6.8% and 8.6%, respetively. As most previous literature, we also find that EPL raises adjustment osts, as we find that EPL redues job flows proportionately more in industries that require a higher level of realloation. One the average effets of EPL are established, we put to test the main hypothesis of the paper, namely that finanial market imperfetions affet firms responses to shoks in ountries and setors that are differently affeted by EPL. We build two different firm-level measures of finanial liquidity, one based on flows (ash-flow) and one based on stoks (net liquid assets). As the finanial variables that measure liquidity vary at the firm level, we are able to ontrol for any timeinvariant unobserved firm harateristis that may affet our dependent variables and are orrelated with the level of firms internal resoures by using firms fixed effets, thus fully exploiting the firm-level dimension of the dataset. We first show that the mere presene of an interation term between EPL and firm-level measures of liquidity in the estimating equation does not alter the previous result that EPL has a negative signifiant impat on our dependent variables in high realloation setors relative to low realloation setors. However, if we do not ontrol for unobserved heterogeneity, we also find that the negative effet of EPL on the apital-labour ratio is larger the higher the level of internal resoures of the firm. This ounterintuitive effet is likely driven by omposition effets at the in- 4

7 dustry level, as firms produing larger ash-flow not only have higher ollateral and easier aess to redit, but are also different along other dimensions, suh as growth prospets and exporting propensity. In fat, when we move to a speifiation that inludes firms fixed effets to ontrol for firms time-invariant unobserved harateristis and identify the effets of EPL from ontrasts of within-firm hanges, we find that EPL still redues the apital-labour ratio, but less so in firms with higher internal resoures (as measured either by ash-flow or by net liquid assets). This result seems to favour the interpretation that finanial onstraints exaerbate the negative effets of EPL on apital deepening. However, we take this finding as suggestive but only tentative as, when we turn to investments, we find only weak (if any) evidene of a differential effet of EPL on firms with different levels of liquid resoures, a result possibly due to the lumpiness of the investment proess, whih deserves further exploration. Finally, neither value added per worker nor job realloation are affeted differentially by EPL in firms with different levels of internal resoures. Poliy Impliations Our paper delivers potentially important poliy impliations. The debate on the role of EPL needs to onsider not only the diret effet on employment flows, but also the indiret impat due to distorted investment inentives. Investment subsidies usually do not take into aount the possible distortions indued by EPL and therefore may be exessive or insuffiient. Our findings potentially bear on the observed differenes in the speed of strutural hange aross developed ountries (Rogerson, 2008): the distorted inentives for apital investment and their produtivity effets may slow down the strutural hange from manufaturing industries towards servies. Sine most of the employment growth ours in the servie setor, these distortions may redue employment growth, effiieny and inome growth. Regarding the interation with finanial markets, firms with low levels of liquid assets may have redued ability to adjust their apital stok; the diffiult substitution of apital for labour might lead to a derease in the produtivity of the employed workers. The obvious poliy impliation is that EPL is more harmful for liquidity onstrained firms, or for setors and ountries where aess to external redit is more diffiult, and poliies to alleviate the effets of EPL should be targeted there (or alternatively that finanial onstraints should be first softened for those same firms and setors). However it is also true that EPL provides insurane to workers against firing. 5

8 Hene, from the point of view of overall welfare, employment protetion poliies should be jointly evaluated with finanial market fritions in the lassi effiieny-equity trade-off. The paper is strutured as follows: Setion 2 reviews the literature and the various mehanisms whih link EPL to investment in apital and to labour produtivity yielding ambiguous results for both. Setion 3 illustrates the researh method and disusses the identifiation strategy, while Setion 4 introdues the data used for the study. Setion 5 presents the basi results on the average effet of EPL while setion 6 disusses the differential effets in finanially sound vs. finanially fragile firms. Setion 7 provides some robustness heks and extends the analysis to examine the differential role of employment protetion for temporary and permanent workers. Setion 8 disusses the poliy impliations and onludes. 2. Theoretial Considerations and Previous Empirial Literature In this setion we review part of a very large literature on the eonomi impat of EPL, foussing on the expeted effets on (1) investment and apital-labour substitution and (2) labour produtivity. Models of labour demand in the presene of adjustment osts unambiguously predit a negative impat of firing restritions on firing and hiring deisions (Bertola, 1990). The effets of EPL on job realloation (the sum of firm s job reation and (absolute levels of) job destrution) have been extensively studied in the empirial literature, and onsequently they are not the fous of this paper. It suffies to say that there is quite a onsensus on the negative effets of EPL on job realloation. Among the reent empirial papers, Autor et al. (2006) and Kugler and Pia (2008) study the impat of EPL on employment realloation at the firm level in the US and Italy, respetively. At the ross-ountry level, Gomez-Salvador et al. (2004), Mio and Pages (2004) and Haltiwanger et al. (2006) among others exploit ross-ountry differenes in EPL to establish a negative relationship between job flows and firing restritions. 2.1 EPL, investment and apital-labour substitution As argued in the introdution, there are theoretial reasons to expet an ambiguous effet of EPL on the apital labour ratio. First, notie that firing osts are likely to push up labour osts even though labour may bear part of the burden via lower wages. For instane, severane payments and notie periods are a transfer from the firm to the worker, and it is likely that workers at least partially ompensate firms for this transfer (Leonardi and Pia, 2008). Also the tax omponent of firing osts does not 6

9 neessarily raise labour osts one to one in ountries with an experiene rating sheme, as the reeipts an be used to ompensate firms via lower unemployment insurane premiums (as in Blanhard and Tirole, 2004). While in perfet labour markets an inrease in the ost of labour will imply substitution of labour with more apital, in models with wage bargaining between workers and firms there may be the opposite effet. When there is wage bargaining, workers will use the protetion of EPL to laim higher wages (Bentolila and Dolado 1994, and Garibaldi and Violante 2005). EPL will strengthen the outside option of workers and worsen the outside option of firms in the wage bargain. As a result, EPL may result in a higher bargained wage. When EPL pushes up labour osts, beause it is an atual ost or beause the government or workers do not fully ompensate firms for their transfers, firms will redue their hiring until expeted osts are in line with expeted profits again and/or will redue their investment. In models with apital investment an inrease of EPL and of the bargained wage gives rise to a hold up problem. If workers and employers meet in a random, ostly proess, some investment deisions have to be taken after a worker (of a given skill level) has been loated and hired: and sine replaing that worker would be ostly, the worker an in general try and bargain for higher wages if investment inreases the job s produtivity. The employer is held up by the worker, who lowers the employer s private returns to investment and therefore his/her inentive to invest (Bertola 1994). A different ase arises in the longer run when firms are not held up by irreversible investments and tehnology adoption beomes an issue. More EPL means that labour is more ostly and when adopting new tehnologies firms will hoose more apital intensive tehnologies (see among others Caballero and Hammour, 1998, Alesina and Zeira, 2006 and Koeniger and Leonardi, 2007). 2.2 EPL and labour produtivity The impat of EPL on labour produtivity is also, in priniple, ambiguous. On the one side, EPL hampers the realloation of workers and jobs aross industries and firms. Therefore, when the importane of realloation for produtivity is large, produtivity falls. On the other side, EPL may have a positive effet on produtivity via speifi investments and learning-by-doing. Mixed results are instead found in studies that fous on partial EPL reform via the introdution of tem- 7

10 porary ontrats. A sreening period of temporary ontrats may lead to better mathes, inreasing produtivity, but the inentives for speifi investments and the period for learning-by-doing may fall, reduing produtivity. a) Considerations suggesting a negative effet of EPL on produtivity More stringent EPL may redue produtivity beause of slerosis in the prodution struture (i.e. EPL is an obstale to realloation of ativity aross industries and to risk-taking), beause higher skill losses during longer periods of unemployment, or beause employees, shielded from a possible layoff due to firing osts, tend to shirk on the job more often. As to realloation and risk-taking, aording to Nikell and Layard (1999) there seems to be no evidene that either striter labour standards or employment protetion lowers produtivity growth rates. For their empirial analysis, Nikell and Layard use aggregate data for 20 OECD ountries observed in the period In some speifiations they atually find a positive effet of EPL on the growth rate of labour produtivity but this effet disappears in other speifiations. Other papers emphasize the effets of EPL on realloation via entry and exit of firms. Hopenhayn and Rogerson (1993) show how the distortion indued by firing restritions pushes firms to use resoures less effiiently. As a result, employment levels adjust at a lower speed and produtivity is redued. Poshke emphasises the role of firing osts in the seletion of the most effiient firms and the exit deision of low produtivity firms, if exiting firms annot avoid paying them. Samaniego (2006) laims that firing restritions are more ostly in industries haraterised by rapid tehnologial hange suh as ICT. Countries where regulations are more stringent will therefore tend to speialise in industries with a slow rate of tehnial hange. Some studies emphasize the obstale of EPL to undertake risky ativities. Bartelsman and Hinloopen (2005) find that EPL has a signifiant negative effet on investments in ICT. They run regressions using data for 13 OECD ountries for the period They onlude that EPL redues the inentive for firms to invest in innovative ativities with high returns and a high risk of failure beause firms want to avoid the risk of paying high firing osts. Saint-Paul (2002) argues that high firing osts may indue seondary innovation that improves existing produts rather than introduing riskier ones. 8

11 Ihino and Riphahn (2005) and Riphahn (2005) laim that layoff protetion (or the lak thereof during the probation period) might also affet produtivity by reduing worker effort beause there is less threat of layoff in response to poor work performane or absenteeism. b) Considerations suggesting a positive effet of EPL on produtivity More stringent EPL may also promote speifi investments and result in more learning-by-doing, whih may inrease produtivity. EPL also provides insurane against uninsurable labour inome risk, and this may allow for better searh of jobs. Belot et al. propose a framework where, by providing additional job seurity, protetion against dismissal may inrease workers inentives to invest in firm-speifi human apital, therefore enhaning produtivity. On the other hand, higher firing osts raise separation osts, inrease the bargaining power of the worker, and thereby raise wages. Only at low levels of employment protetion is an inrease in EPL benefiial to produtivity-growth, and the positive effets of employment protetion are larger in setors where firm-speifi skills matter more. Wasmer (2006) suggests that by induing substitution of speifi for general skills, firing restritions may have a negative effet on produtivity when workers need to be realloated aross industries and industry-speifi skills beome useless. Lagos (2006) laims that if stringent EPL raises reservation wages, average produtivity an inrease simply beause firms beome more seletive and less produtive mathes are not realised. Bertola (2004) shows that the additional insurane via severane pay may also result in a produtivity gain in the spirit of Aemoglu and Shimer (1999), making workers more willing to leave their low-produtive job to look for a more produtive one. The empirial part of most of the papers reviewed, if present at all, is based on ross-ountry regressions on aggregate outomes. However, this approah potentially suffers from well-known severe problems. First of all, reverse ausality: the stritness of EPL may depend on labour market onditions. Seond, omitted variables may bias the results: EPL may pik up the effet of other fators unobserved by the eonometriian that drive the ross-ountry differenes in labour market performane. Third, most studies fous on overall EPL, without distinguishing between EPL provisions for fixed-term and permanent ontrats. As far as we know, very few studies go beyond ountry-level data. Sarpetta et al. (2002) analyse the effets of EPL and entralized bargaining on firm produtivity and firm dynamis using harmonized data for 17 manufaturing industries in 18 ountries, over the period They 9

12 find that strit EPL has a signifiant negative impat on produtivity only in ountries with an intermediate degree of entralisation/oordination in wage bargaining. Autor et al. study the impat of adoption of wrongful-disharge protetion norms in the US, using ross-state differenes in the timing of adoption. Exploiting mirodata, they find that apital deepening is inreased while TFP is redued. Quantitatively, they alulate a drop in produtivity, with an average elastiity in the order of 0.03 to Similar findings are provided by Cingano et al. (2008) using Italian data to examine a 1990 reform that raised dismissal osts for firms with fewer than 15 employees only. Mio and Pagés (2004) analyse the differene in the effets of EPL aross setors within a ertain ountry. They argue that EPL is more binding in setors that are more suseptible to tehnologial and demand shoks. They use data for the manufaturing setor for 18 ountries during the 1980s and 1990s, and find a negative relationship between layoff osts and the level of labour produtivity espeially in those setors with higher needs for flexibility. The paper losest to ours is Bassanini et al. (2009), whih uses setoral harmonized data from EUKLEMS for 17 industries in 18 industrial eonomies over the past two deades. They onsider EPL together with other labour market institutions and the extent to whih EPL is binding in partiular industries. They find a negative effet of EPL on total fator produtivity (TFP) and onlude that reforms of overly strit dismissal regulation in many OECD ountries an be justified on the grounds of fostering TFP growth. 3. Empirial Framework In order to desribe the identifiation strategy that allows us to estimate the joint effet of labour and finanial market imperfetions, we proeed in two steps. In setion 3.1 we desribe the identifiation strategy of EPL negleting redit markets, and in setion 3.2 we extend our empirial framework to allow for the presene of (imperfet) apital markets. 3.1 Identifiation of the average effet of EPL on firm-level outomes Our empirial strategy relies on a well-established approah developed in the finane literature by Rajan and Zingales (1998) and reently adopted in labour studies (see Mio et al. 2004, Fonsea and Utrero, 2006, Haltiwanger et al and Bassanini et al., 2009) to estimate the impat of some ountry harateristi (often, measures of regulation) on eonomi performane aounting 10

13 for geographial- and tehnologial-speifi time-invariant unobservables. The basi idea underlying the approah is to exploit the fat that while the amount of regulation is given for all firms within a ountry, its impat ould be different if, due to tehnologial harateristis or to the inidene of aggregate shoks, firms do differ as to the frequeny or amount of required labour realloation. In this ase, the importane of employment protetion legislation an be inferred by looking at whether firms requiring more realloation see a better performane in ountries with less restritive legislation. The main problem with this approah is reovering a plausible measure of employment realloation requirements. Job flows are in fat not ustomarily inluded among offiial statistis and even if they were observable at the firm or industry level, they would likely reflet idiosynrati omponents endogenous to the level of EPL in eah ountry. This implies they would in general not just reflet the amount of realloation of a fritionless environment, where the extent of yearly flows only responds to, say, tehnologial firm or industry differenes. Hene, using atual labour realloation is likely to yield biased estimates of the impat of EPL on performane. Following the influential study of Rajan and Zingales (1998) on finanial development, one popular approah to this problem is to proxy for firms harateristis in the absene of distortions using data from a flexible market eonomy. For example, Mio and Pages (2004), Haltiwanger et al. (2006) and Bassanini et al. (2009) use realloation figures omputed for US industries. Their underlying assumption is that suh baseline should proxy for the tehnologial and market driven employment realloation aross industries in the absene of poliy-indued osts of adjustment. Applying this approah implies estimating the standard differenes-in-differenes speifiation of studies exploiting ross-ountry ross-industry data: Y ijt = ( E BenhFlow ) ε (1) t j δ + Et ϕ + X ijtγ + µ t + µ j + µ + D + ijt where Y ijt is the outome variable of firm i in ountry, industry j at time t; E t is a ountryvarying index of employment protetion legislation; BenhFlow j is the extent of intrinsi job realloation in setor j (below we desribe the different proxies used in this paper). The various speifiations enompass different sets of year-, industry- and ountry- effets (respetively µ, µ, ) and their interations D. The matrix X ijt inludes firm-level ontrol t j µ variables and ε ijt is the residual. We take are of the intra-luster orrelation of standard errors. 11

14 Equation (1) allows to estimate the average effet of EPL exploiting variability at the ountrysetor-time level in the relationship between employment legislation and outomes. At this stage, we do not inlude firms fixed effets beause they would wash away all the industry by ountry variation making the identifiation of the variable of interest (E benhflow) rely only on the (limited) time variation of the EPL index. The oeffiient δ in equation (1) aptures the effet of employment regulation on the variable of interest. One way to interpret δ is thinking of the average differene in the variable of interest Y between two industries haraterized by high and low realloation flows (say, orresponding to the 10 th and 90 th perentile of the observed distribution, respetively). Then estimates of δ in equation (1) an be thought of as the implied hange in suh differential as employment protetion is inreased by an arbitrary amount (say, equivalent to the 10 th -90 th differene observed aross ountries). Following the standard benhmark-ountry approah would require proxying the setoral intrinsi need for job realloation using data from the most flexible market eonomy available (the US or, in our sample, the UK). The appropriateness of the benhmark-ountry approah an however be questioned along two dimensions. First, the validity of the benhmark hinges on the representativeness of the industry in the benhmark ountry, within the set of ountries overed in the sample. Even if US realloation rates in a given industry are a good proxy of the intrinsi needs of realloation in that setor, it might be the ase that within setor heterogeneity aross ountries limits the omparative exerise. An example illustrates well this problem. If the researher is using benhmark flows measures at the 2-digit industry level of aggregation, the realloation in setor 35 Manufature of Transport Equipment in the US, would serve as benhmark realloation for the remaining ountries in the sample. However, going finer in the industry lassifiation one finds that industry 35 is omposed, among others, of sub-setors 3511 Building and repairing ships and boats, 3530, Manufature of airraft and spaeraft and 3542 Manufature of biyles. The benhmarking requires that either intrinsi needs of realloation in the three sub-setors are similar, or that the average within setor industry mix in every ountry in the sample is well proxied by the average industry mix in the US. A finer level of aggregation of the benhmark would limit this problem. Seond, the benhmark-ountry approah has been questioned for representing a measure of short- rather than long-term industry-differenes (Fisman and Love, 2003). This would imply in our ase that the benhmark onstitutes a noisy proxy of fritionless (or tehnologial) industry 12

15 realloation requirements. In a reent paper, Cione and (2006) have formalized this problem by showing that if the benhmark reflets, among other fators, idiosynrati shoks, then the measurement error stemming from ountry-benhmarking an indue both upward and downward biases in the estimates of δ in (1). In our ase, if employment realloation aross industries in the benhmark ountry orrelates more losely with realloation in low-epl ountries than in high-epl ountries, then one might find signifiant effets of regulation even if there were not. To irumvent the problem Cione and (2006, 2007) proposed a methodology to onstrut a world-average benhmark measure not refleting idiosynrati fators speifi to a ountry or regulatory environment. Exploiting the availability of industry (or firm-) speifi figures of job realloation JR j, suh a measure an be obtained in our ase regressing job realloation measured at a detailed industry level on ountry dummies interated with time dummies, industry dummies and industry dummies interated with ountry-level EPL: JR jt = α + λ + θ E + υ (2) j t j t jt where the interation term θ allows to absorb the marginal effet of employment protetion j E t on job realloation in eah industry j, and λ t aounts for time-varying differenes at the ountry level. Hene αˆ j aptures the extent of industry job realloation in a ountry not subjet to firing restritions (we are ontrolling for EPL), whih is subjet to world average supply and demand shoks. This is the measure of fritionless setoral realloation (BenhFlow j = αˆ j ) in equation (1) that will be used in the paper. Hene, we ollapse our firm level data (desribed below) at ountry-industry-year ells and run a regression following equation (2) in order to extrat the αˆ j s. The job realloation rate is defined, following Davis and Haltiwanger (1990), as JR jt = e 2 e ijt e + e jit 1 i j, ijt ijt 1 where subsripts are defined as above. In order to preserve a minimal level of representativeness in eah ell, we drop all ells where job realloation was omputed for less than 10 firms. While the methodology allows avoiding ountry-speifi idiosynrasies, its main limitation is that, sine no ountry in our sample has zero EPL, it omputes trustworthy fritionless rates only under the assumption that out-of-sample preditions are reliable. For this 13

16 reason, we hek the robustness of this approah, by using as an alternative benhmark the setoral job realloation rates (averaged over time) of the ountry with the lowest level of EPL in our sample. 2 Comparing the results obtained using the two alternative measures is interesting to assess to what extent widely used benhmark-ountry proxies reflet idiosynrati shoks. Figure 1 below depits the relationship between atual job realloation in the UK, the ountry with the lowest level of EPL in our sample, measured at the 4-digit industry level (more than 400 setors) with the measure obtained following equation (2). The piture shows that the atual UK job realloation rate and the fritionless job realloation measure are strongly positively related. The slope of the linear fit (dotted line) is positive and signifiant. Although the relationship between both measures is positive and signifiant, it is different from a hypothetial 45% line, suggesting that UK job flows are a mix of world average and idiosynrati needs for realloation. Figure 1. Job realloation in UK and world-average measure. 2 One an argue that the fritionless measure using only within sample ountries has an endogeneity problem and that, insofar as the driving variable appears to be EPL on regular ontrats, benhmarks based on layoffs would be more pertinent than benhmarks based on turnover (for example servies are notoriously high turnover but low layoff industries). To address this problem we also used the setoral layoff rates from the US (a ountry external to the sample) taken from Bassanini et al. (2009) as an alternative benhmark. Speifiations based on this measure give insignifiant but qualitatively similar results. This is likely due to the fat that this measure is available only for 16 setor, rather than for the more than 400 setors implied by the Amadeus 4-digit disaggregation. 14

17 Finally, notie that there might be an issue of endogeneity of regulations. It is likely, for example, that ountries that experiene high turnover rates have a high demand for strit employment protetion legislation. Alternatively, ountries with low employment reation may tend to protet existing jobs. An attempt to address the problem using instrumental variables an be found in Bassanini et al. (2009). Our approah allows us to use ountry (by time) and setor fixed effets to ontrol for all observable and unobservable ountry and setor harateristis. In partiular, it allows to ontrol for differenes in ountry and setor output volatility, thus alleviating the potential problem of endogeneity of regulations present in ross-ountry regressions. In fat to laim endogeneity in our approah, one would have to argue that aross ountries a high level of turnover or low job reation in some setors determines the level of employment protetion in the whole ountry. 3.2 Identifiation of the joint effet of EPL and finanial market imperfetions As explained in the introdution, the main ontribution of our analysis lies in the study of the joint effet of EPL (labour market fritions) and finanial onstraints on the apital-labour ratio, investment and labour produtivity. We therefore relate to the large literature that looks at the determinants of apital investment and finds aess to the redit market to be one of the important fators affeting investment. Most empirial studies of investment and finaning onstraints, in the tradition of Fazzari, Hubbard and Petersen (1988) typially regress a measure of investment on a measure of investment opportunities (Tobin's q) as well as a measure of ash flow, i.e. they estimate the sensitivity of investment to ash flow onditional on q. These empirial speifiations imply that, in the absene of finaning onstraints, investment is likely to be subjet to adjustment osts that prevent the apital stok adjusting ontinuously to maintain equality between the marginal revenue produt and the user ost of apital. In the absene of finanial fritions, Tobin s q is a suffiient statisti for investment opportunities, whih means that nothing but q should matter in investment equations. A positive orrelation between investment and liquidity, onditional on q, is therefore taken as evidene of the presene of finanial market imperfetions that prevent positive net present value projets to be finaned, possibly beause of moral hazard problems. In this paper, differently, we study the joint effet of EPL and finanial onstraints on the outome variables i.e. the differential effet of EPL on all outome variables for finanially on- 15

18 strained firms. The impat of redit and labour market imperfetions on investment has been theoretially analyzed in Rendón (2004) and in Wasmer and Weil (2004), who showed that job reation is limited by finaning onstraints even in the presene of a flexible labour market. There are not many papers that investigate empirially the joint influene of imperfet finanial and labour markets on investment, with the exeption of Calagnini and Giombini (2008). Their results show a negative orrelation between EPL levels and investments. In partiular they find that firms faing negative shoks see their finanial onstraints worsening in ountries with greater labour market rigidities. The interation of finanial fritions and EPL is evaluated in our ross-ountry panel data framework exploiting the interation between labour and finanial market imperfetions at the firm-level. To this purpose we augment our baseline speifiation (1) with the operating ashflow of firm i at any observed year t-1. We take the lagged value of ash-flow in order to make sure that we measure liquidity before investments are made. Otherwise, reverse ausality may be at work sine high investments may generate low levels of liquidity. We interat ash-flow with EPL and Benhflows separately, and with their interation EPL Benhflows, as desribed in the following equation: Y ijt = ( E t + ( E BenhFlow ) δ + ( E t CF ijt 1 j 0 t j CF ijt 1 ) δ + ( CF BenhFlow ) δ + E φ + CF 3 t 0 1 ijt 1 BenhFlow φ + βroa ijt 1 1 ijt 1 j ) δ + + µ + D + ε i 2 ijt (3) where CF = ijt 1 Cash Flow Fixed Assets ijt 1 ijt 2 and D is a vetor of dummy variables inluding ountry by year interations. The oeffiient δ 3 of third level interation term E CF BenhFlow ) aptures the effet of a hange in EPL ( t ijt 1 j on investment and on the other dependent variables in firms with different levels of internal resoures in setors with different volatilities of employment. If higher levels of internal resoures failitate apital deepening (as in presene of redit onstraints) then this interation term should positively enter the investment per worker and K/L equations. Sine omputing Tobin s q requires information on the market value of the firm and the vast majority of firms in our sample are unlisted, in this work we will measure investment opportunities with the rate of Return on Assets (ROA in equation 3). 16

19 Note that our speifiations now inlude firm level fixed effets, sine the variable of interest in this ase is not an aggregate variable as in the previous speifiation, but varies over time within firms. In this ontext, it beomes ruial to ontrol for any unobserved fator that remains onstant within firms and might be orrelated with the ash flow measure. One may in fat argue that firms able to produe a higher ash-flow have (1) easier aess to redit but (2) are also typially larger and (3) behave differently along many (unobservable) dimensions. To the extent that these unobserved fators are time invariant, they are aounted for by firms fixed effets. To test the robustness of our results we use two measures of internal resoures. The first is operating ash-flow. The seond, following Cleary, Povel and Raith (2005), is a stok measure of liquidity alled net liquid assets, desribed in detail in the robustness setion. 4. Data desription Our main data soure is Amadeus, a firm level data set olleted by the Bureau van Dijk (BvD) ontaining balane-sheet information of European firms. 3 The information is gathered by speialized national servie providers and is homogenised applying uniform formats in order to allow aurate ross-ountry omparisons. We used the largest version of Amadeus in its 2006 DVD format, whih overs firms of all sizes for the period However, the overage prior to 1997 is relatively limited. Similarly, EPL data is only available up to Hene, we restrit the analysis to the period , but robustness heks adding these additional years are provided in Setion 7. The 14 ountries under study are: Belgium, the Czek Republi, Denmark, Finland, Frane, Germany, Greee, Italy, the Netherlands, Poland, Portugal, Spain, Sweden and the United Kingdom. 4 The limitations of this firm-level dataset are well-known. First, what the data ompany is able to report depends on how demanding the aounting standards of a ountry are and whih firms indeed report. Therefore, without any doubt, the sample is biased toward ountries with more de- 3 See Messina and Vallanti and Konings et al. (2005) for desriptions of Amadeus in different researh ontexts. Giannangeli and Gómez-Salvador (2008) use Amadeus in an study of the soures of growth in manufaturing produtivity in five European ountries. 4 We tried to over all ountries in Amadeus for whih EPL data from the OECD was available. Austria and Germany onstitute speial ases in Amadeus. Most firms in these two ountries have limited information on their balane sheets, inluding employment and very few finanial items. After data leaning, this results in insuffiient observations in the ase of Austria for most of the speifiations. Hene, Austria is dropped from the analysis. Slovakia, Ireland and Hungary were also dropped due to small samples. There are very few German firms too in the sample, but suffiient to be present in most ountry, year and setor ells. The analysis in the paper inludes Germany, and robustness heks exluding speifi ountries are disussed at the end of the paper. 17

20 manding aounting standards and more transparent firms. If anything, this sample seletion bias should make it harder to find a signifiant impat of finanial market imperfetions on firms response to striter EPL. Moreover, in any given ountry, the sample may not be representative of the underlying population. In priniple this problem ould be addressed by ross-heking our results (at least those not involving firm-speifi measures among the variables of interest) against estimates obtained running the orresponding regressions at a more aggregate level using information from ountry-industry datasets. In pratie, this strategy is largely undermined by the signifiant lak of data on investment and apital stoks at the industry level. For example, the inreasingly used Euklems dataset assembling industry-level aounts for EU members at a 2-digit level of disaggregation laks information on apital stok for ountries as Frane, Spain and Belgium, whih onstitute a relevant share of observations in our sample. Unfortunately, the OECD Stan dataset, a possible alternative soure of data even though with a oarser industry breakdown, also present a signifiant fration of missing values as regards the stok of apital. To get reassured that Amadeus firms does not ompletely misrepresent the population distribution we aggregated our data to the orresponding Euklems industry-level breakdown and omputed orrelations between ountry-industry shares of employment and value added in the two datasets (suh information is available in Euklems for all ountries in our sample). In 2003, the orrelation is 0.44 in the ase of employment and 0.35 in the ase of value added. Despite the above desribed limitations the use of Amadeus is beoming widespread in the eonomi profession for several reasons. First, the relassifiation of the balane sheets appears reliable, sine no attempt is made to reonstrut items that are missing from the original balane sheets or diffiult to reonstrut. In fat, many variables are missing, espeially for firms inorporated in ountries where aounting praties are less transparent. Another important advantage of Amadeus is that most of the firms inluded in the data set are private, allowing to fous on a sample that is more representative than the listed ompanies typially analysed in studies on redit markets (see Rajan and Zingales, 1995 and Boot, et al., 2001). This naturally entails some shortomings given that the information available for private firms is less detailed. Moreover, sine private firms are not traded, only book values are available and it is not possible to evaluate the market values of debt ratios, whih would provide useful additional information. However, this shortoming are not likely to hamper the analysis beause previous studies (Rajan and Zingales, 1995 and Boot, et al., 2001) do not find any signifiant differenes in fators orrelated with debt to book and market apital. 18

21 For the aims of this paper the advantages of looking at a panel of balane sheet data for firms in different ountries largely prevail over the disadvantages. First and foremost, the availability of balane sheet data allows us to study whether and to what extent labour market regulation interats with finanial onstraints when firms reat to aggregate or idiosynrati shoks. This analysis simply annot be performed on setoral data. 5 Seond, even when fousing on the average effets of employment protetion, the use of firm-level data is advisable, as one an aount for industry and ountry speifi unobserved harateristis in ways that studies based on aggregate data are unable to orret for. This makes our study less subjet to miss-speifiation and omitted variable biases. Finally, the firm-level data is lassified at a very detailed industry dimension (4-digit NACE lassifiation). The possibility of onstruting the benhmark fritionless job flow measure at suh a refined level of aggregation helps us limiting possible problems of omparability of industries disussed above. 4.1 Desriptive Statistis Figure 2 reports the average values of our variables of interest by ountry, thus giving a visual representation of the desriptive statistis by ountry presented in Table 1. The upper left panel reports both EPL levels (averages in ) and hanges (from 1997 to 2003) and shows that the United Kingdom displays the lowest level of EPL in our sample period while Portugal displays the highest. It is noteworthy (and also well-known) that EPL varies very little over time. The upper left panel of Figure 2 shows that most ountries display very little or no variation of EPL over time, with the exeption of Italy. Table 2, whih reports desriptive statistis by year, shows that average EPL ranges from 2.44 and 2.56 over our sample period. This is why we do not inlude firms FE in the estimation of equation (1). The remaining panels of Figure 2 show the average levels of apital per worker, value added per worker and investment per worker. Germany and Belgium exhibit the highest values of all of them (exept I/L, where Poland ranks higher than Belgium) while the United Kingdom ranks very low both for investment and for apital. However, the most interesting statistis are perhaps those presented in table 1, where we show the average levels of job realloation, apital per worker, value added per worker and investment per worker in high and low job realloation setors and high and low EPL ountries, where we take as low (high) job realloation setors those below (above) the 25 th (75 th ) perentile of the distri- 5 Few reent papers addressed a similar issue in a totally different framework, i.e. studying the determinants of orporate ontrol (Atanassov and Kim 2008, Pagano and Volpin 2005, Bozaya and Kerr 2008). 19

22 bution of the fritionless job realloation and as low (high) EPL ountries those below (above) the 25 th (75 th ) perentile of the distribution of EPL. The list of the low- and high- realloation setors is in the note to the table. Figure 2. Summary Statistis by Country We perform an unonditional diff-in-diff exerise that mimis our identifiation strategy in equation (1). The results of this exerise, while allowing a better understanding of our identifiation strategy, should be taken with aution beause (1) the regression exerise is based on a ontinuous treatment (i.e. in the regressions we will not arbitrarily divide setors into two groups but rather use the ontinuous measure of intrinsi job realloation) and (2) beause of the presene of aggregate onfounding fators. Notie, for example, that in our data Job Realloation (first row) is higher in high EPL ountries, both in setors lassified as displaying low and high (intrinsi) needs of job realloation. This suggests the existene of a positive orrelation between employment legislation and turnover at the aggregate level, as in the ase in whih ountries tend to have higher levels of protetion in response to exogenously indued inreases in flows (e.g. a deep restruturing of the produtive struture). That is one of the reasons why our analysis will look at within-ountry effets of EPL exploiting a diff-in-diff approah as in Rajan and Zingales (1998). Moreover, it suggests that using ountry-speifi reallo- 20

23 ation rates might indue severe biases beause realloation might responds to ountry-speifi legislation. Our attempt to onstrut a world-average benhmark measure not refleting idiosynrati fators speifi to a ountry or regulatory environment aims at addressing this issue. Overall, table 3 shows a negative unonditional effet of EPL on job realloation and value added per worker, while the industry-level differential in apital and investment intensities seem to inrease. As we will see, the onditional ontinuous-treatment version of this approah points to substantially different results as far as investment and apital per worker are onerned. 5. Results: average effet of EPL In this setion we doument the relevane of employment regulations exploiting differenes in setor harateristis to implement a differene-in-differenes estimation method. We first look at the effets on apital and investment normalized by unit of labour (setion 5.1). Then we look at the effets on labour produtivity and for omparison with previous studies we assess whether employment protetion legislation affets the level of job realloation (setion 5.2). In setion 6 we move to the analysis of the role of internal resoures in firms responses to striter firing restritions Capital and investment per worker The five olumns of Table 2 show that EPL redues the apital-labour ratio in firms operating in the most affeted setors. The first olumn inludes ountry, year and setor dummies (438 four digits dummies) to absorb institutional, tehnologial and time speifi effets. In the seond olumn we inlude industry-by-time dummies to ontrol for differential trends by industry in the outome variable. For example some industries may experiene faster (e.g. the omputer industry) or lower-than-average (e.g. manufaturing) apital adjustment or job realloation or produtivity growth in all ountries. In the same olumn we also inlude ountry-by-time dummies to ontrol for all ountry-speifi time-varying harateristis (for example all national-level institutions) whih have the same effets aross industries. Notie that this set of dummies absorbs the main effet of EPL, as this variable only varies by ountry and time. 6 In table 2 we find strong adverse effets of employment protetion on apital-labour ratios. The oeffiient on the interation is strongly signifiant and in the neighbourhood of In order to 6 Note that while we study firm-level outomes, our variable of interest in equation (1), the interation term EPL benhflows, varies at the four-digit industry level in 14 ountries and 7 years. We take are of the intra-luster orrelation of standard errors likely to arise in all the speifiations disussed. 21

24 get an idea of their magnitude, onsider the ratio of apital-intensity between industry 1561 "Manufature of grain mills produt" and industry 2955 "Manufature of mahinery for paper or paperboard prodution", the two industries we estimate at the 10 th and 90 th perentile of the apital-labour ratios distribution. Then our estimates imply that reduing employment protetion from the level Greee in 1997 to that in Ireland the same year (this shift orrespond to the 90 th to the 10 th perentile of the ountry-by-year EPL distribution in our sample) would inrease suh ratio by 12%. In a first attempt to ontrol for heterogeneity, the third olumn inludes firm-level ontrols (firmsize, exports and number of subsidiaries). 7 In the fourth olumn we aggregate Amadeus data at the industry level and we run regressions at the industry level to ease omparison with results obtained by previous literature using EUKLEMS data (Bassanini et al., 2009). In both olumns 3 and 4 the results are qualitatively unhanged with respet to the first two olumns. The last olumn of tables 2 uses setoral job turnover in the UK as benhmark to test the robustness of our fritionless measure. The results are qualitatively similar and will be further ommented upon in the robustness setion. We then turn to examine the effet of employment protetion legislation on investment. We report the results in Table 3 for investment normalized by units of labour, as this is the relevant variable in models of hold-up. Results in olumns 1 to 5 indiate a signifiant negative effet of EPL on investment over labour units. The oeffiient in olumns 1, 2 and 3 lies around -0.2 and is strongly signifiant. To gauge the magnitude of the effets, we repeat the thought experiment of dereasing employment protetion legislation from the 90 th to the 10 th perentile of the observed distribution. This implies an inrease of 6.8% in the average I/L ratio between two industries at the 10 th and 90 th perentile of the observed distribution. Negative results on the apital-labour ratio are onsistent with results on I/L that show that investment is atually falling relative to the units of labour employed (table 3). The overall piture shows a redution in apital investment (relative to labour) in onsequene of an inrease in EPL as predited from hold up theories (see setion 2). 7 Exports and number of subsidiaries display non missing values only for a relatively limited number of observations. As for these variables the non missing values are always stritly positive, it is plausible to interpret them as zeros. We do so and hek the robustness of this assumption adding one additional ontrol at the time. We find that the results are unhanged. 22

25 5.2. Labour produtivity and job realloation In table 4 we explore the effet of EPL on labour produtivity finding strong and signifiantly negative oeffiients of around (whih an be quantified with the previous exerise in an inrease of 8.6 perent in the average value added per worker ratio in high versus low realloation industries. Regarding job realloation, although the negative effet of EPL on job flows is well established (see referenes in setion 2), most previous studies look at setoral rather than firm level data. Our results are presented in Table 5. The oeffiients estimated in olumns 1 and 2 are similar in magnitude ( and ). In summary we find that the sign of the oeffiient on the interation terms is negative and statistially signifiant i.e. firms in more volatile industries present lower levels of job realloation in ountries with more stringent employment protetion laws. These results are in line with most of the previous literature on EPL and job flows. In order to get an idea of the magnitude of the effets, we onsider the differential in the (average) realloation rate between the two industries at the 10 th and 90 th perentile of the omputed distribution of fritionless realloation. Then our estimates imply that reduing employment protetion from the level of Greee in 1997 to that in Ireland the same year (this shift orrespond to the 90 th to the 10 th perentile of the ountry-by-year EPL distribution in our sample) inreases yearly realloation by 1,4 perentage points, against a median realloation rate of firms in our sample of 5.7%. The negative results on labour produtivity are onsistent with previous empirial literature (e.g. Autor et al, 2007 and Bassanini et al., 2009) and are somewhat expeted as we have assessed that EPL not only redues job flows in table 5 (as in muh of the previous literature) but also redues apital investment per unit of labour (table 3) and the apital-labour ratio (table 2). If realloation of labour is important and EPL hampers job realloation aross and within setors, then produtivity falls. Indeed, finding an effet of EPL on job realloation is a pre-requisite to laim that dismissal restritions hamper the optimization of resoures and alloative effiieny (Bertola, 1990). On top of that, labour produtivity also falls beause apital investment per worker falls. 23

26 6. EPL and the role of Finanial Market Imperfetions We are now in the position to analyse the results on the joint effet of EPL and finanial onstraints on the outome variables i.e. the differential effet of EPL on all outome variables for finanially onstrained firms. Our empirial strategy, outlined in equation (3), amounts to evaluate whether there is a differential effet of EPL in firms with different levels of internal resoures on the variables analysed so far: apital per worker, investment per worker, value added per worker and job realloation rates. We build two different measures of finanial liquidity, one based on the flow of internal resoures potentially available for investment purposes (operating ash-flow) and, for robustness purposes, one based on the stok of internal resoures (net liquid assets, see robustness setion for details). As the finanial variables that measure liquidity vary at the firm level, we are now able to ontrol for any time-invariant unobserved firm harateristi that may affet the dependent variables while being orrelated with the level of firms internal resoures by the use of firms fixed effets, thus fully exploiting the firm-level dimension of the dataset. As before, we first look at the effet on apital and investment normalized by unit of labour and on the apital-labour ratio (setion 6.1). Then we look at the effets on labour produtivity and job realloation (setion 6.2) Capital and investment per worker Table 6 reports results on the ratio of apital to labour. The first olumn simply expands the speifiation reported in the seond olumn of table 2 by adding firm ash-flow (and its interations with respetively EPL, benhflows and EPL benhflows) and the lagged value of ROA to ontrol for the profitability of the firm. This olumn shows that the mere presene of firm-level measures of liquidity (and their interation terms) in the estimating equation does not alter the result that EPL has a negative signifiant impat on K/L in high realloation setors relative to low realloation setors, as shown by the negative and signifiant oeffiient of EPL BF. 8 To assess whether firms with higher levels of internal resoures respond differently to EPL, we need to look at the oeffiient of the triple interation term EPL BF Internal Resoures. The oeffiient is negative and signifiant, meaning that the negative effet of EPL on the apital- 8 Note, however, that this effet is evaluated at zero-ash flow, but simulations at values at the 90 th and 10 th perentile of the ash-flow distribution show that the effet is always negative and statistially signifiant. 24

27 labour ratio is larger the higher the level of internal resoures of the firm. This ounterintuitive results is likely driven by omposition effets at the industry level, as firms produing larger ash-flow not only have higher ollateral and easier aess to redit, but are also different along other dimensions, suh as growth prospets and exporting propensity. For this reason the seond olumn inludes firms fixed effets and therefore ontrols for all firms time-invariant unobserved harateristis that may be orrelated with ash-flow, thus identifying the effets of interest from ontrasts of within-firm hanges. We now find that EPL still redues the apital-labour ratio, but less so in firms with higher internal resoures as the oeffiient on EPL BF Internal Resoures is now positive and signifiant. Having a high ash flow thus redues signifiantly the negative effet on the apital-labour ratio or equivalently, from the point of view of finanially onstrained firms, they have to redue apital more when EPL inreases and they are held up by the request of higher wages. This result is robust to the inlusion of ountry by year dummies (olumn 3); the use of a different measure of liquidity (olumn 4, further disussed in the robustness setion); additional firm-level ontrols (olumn 5) and the use of UK job turnover as an alternative proxy for the fritionless job realloation rate (olumn 6). We have interpreted the negative effet of EPL on apital investment and the apital-labour ratio in the basi speifiation of equation (1) along the lines of the hold up theory. The results of equation (3), whih looks at differential effets depending on the internal finanial struture of firms, are onsistent with the same view: the presene of striter EPL disinentives the use of internal funds for finaning new investments: i.e., if apital is largely sunk and high EPL favours ex-post profit appropriation by workers, firms will use their internal funds to pay higher wages and will invest less. This is all the more true for finanially onstrained firms with low liquidity. Table 7 turns to the results on investment per worker. It shows in all olumns that having a high ash-flow does not make any differene for the negative impat on investment per unit of labour as oeffiient on EPL BF Internal Resoures is always insignifiant, exept for olumn 4 whih uses net liquid assets and shows that in the fae of high EPL more liquid firms tend to invest more. We speulate that the absene of a differential effet on the investment level of firms with different levels of liquidity may be due to the lumpy nature of the investment proess, as investments episodes tend to be rare and of substantial magnitude. Indeed in our data, more than 25

28 6% of observations on investments are zero and the observed distribution is extremely skewed to the left Labour produtivity and job realloation Table 8 onsiders labour produtivity. Overall, the results indiate that the negative impat of EPL does not seem to depend of the level firms internal resoures as the oeffiient on EPL BF Internal Resoures is always insignifiant, exept for olumn 6 whih uses UK job realloation. Finally, we do not find any differential effet of EPL on job realloation depending on the level of internal resoures either (table 9 exept for olumn 4 whih uses net liquid assets). Summing up, the result on apital per worker seems to favour the interpretation that finanial onstraints exaerbate the negative effets of EPL on apital deepening. However, we take this finding as suggestive but only tentative as, when we turn to investments, we find no (or at best only weak) evidene of a differential effet of EPL on firms with different levels of liquid resoures, a result possibly due to the lumpiness of the investment proess whih deserves further exploration. Finally, neither value added per worker nor job realloation are affeted differentially by EPL in firms with different levels of internal resoures. 7. Robustness heks In this setion, we provide a number of robustness heks for our baseline regressions. We test robustness with respet to a) benhmark measure of job realloation; b) balaned unbalaned samples; ) the speifiation of the equation; d) exlusions of speifi setors or ountries; e) the measure of EPL; f) the firm-level measure of internal resoures. a) The last olumn of tables 2 to 9 uses UK turnover as benhmark to test the robustness of our benhmark measure. The results show that in all ases the results are negative and larger than with the measure exept for the ase of table 5 (job realloation) where EPL FJR is now insignifiant and for the ase of table 8 (VA/L) where the oeffiient of interest EPL FJR Internal Resoures beomes signifiant. The differene between the UK realloation benhmark and the world-average measure reflet the idiosynrasies of the UK labour market as explained in the data setion. 26

29 b) The sample is unbalaned, therefore inludes entry of new firms and exit. Thus, the overall effet we measure inludes both the diret impat on inumbent firms and the indiret ompositional effet through entry and exit. However, we are not able to disentangle the two effets primarily beause firms an enter or exit the Amadeus sample for many reasons (e.g. merger, aquisition, hange of name, hange in the obligation to provide/have a balane sheet) that prevent us from reliably measuring the true entry and exit. To try and have an idea of the extent to whih our effets are due to the hurning of firms, we ompare the results obtained on the unbalaned sample (whih inludes entry and exit) with results on a balaned sample of firms that stay in sample every year from 1997 to We have a bit less than of suh firms in the sample. We find that the results on the balaned sample are virtually the same as on the unbalaned sample. If anything, they are quantitatively stronger. ) We also run regressions where the dependent variable is in growth rates rather than levels, to allow for the possibility that EPL might affet the growth rate of variables rather than levels. However, notie that investment per worker (I/L) and Job Realloation are already growth rates (I/L is preisely the rate of hange of K/L). Therefore we regressed the rate of hange of VA/L on EPL (inluding the interation terms with firms internal resoures) and found no differential effet of EPL in firms with larger levels of internal resoures, very muh as in the ase of the variable in levels. Additionally, one may also worry that our variables may be affeted by past EPL. However, these effets are likely to be aptured by the urrent level of EPL given the wellknown limited time variability of EPL, and by ountry and time dummies. d) We assess the impat of the exlusion of speifi setors in the regression. We have used our preferred speifiation, whih inludes setor by year and ountry by year fixed effets (and orresponds to olumns 2 in tables 2 to 5). Hene, identifiation relies in within ountry variation aross setors, in the spirit of the original ontribution of Rajan and Zingales (1998). Dropping one setor at a time never turns the sign of our variable of interest, the interation of EPL with benhmark flows, whih remains negative when JR, I/L, K/L and VA/L are the dependent variables in eah of the 446 regressions. Moreover, the oeffiients are always statistially signifiant at the 10% level, the t-statistis ranging from 3.2 to 5.69 in the ase of JR, from 1.65 to 3 in the ase of I/L, from 2.37 to 3.52 in K/L regressions and from 1.89 to 3.16 (exept one single ase where the t-statisti is 1.38) when the dependent variable is VA/L. 27

30 Figure 3. Effet of EPL on K/L, I/L, VA/L and JR exluding one ountry at the time Our next exerise examines the impat of the presene of speifi ountries in the sample. Figure 3 shows the impat of dropping one ountry at a time in eah of our outome variables, fousing on the speifiation that inludes ountry by year and setor by year fixed effets. The estimates presented in the text are relatively stable when speifi ountries are exluded from the sample. In all the ases the estimated effets retain their negative sign, with one notable exeption; the interation term EPL BenhFlows in the labour produtivity regression beomes positive when the UK is exluded from the sample. The exlusion of Frane from JR regressions (and of Frane and UK for I/L regressions) also somewhat dampens the negative sign, as the oeffiient of the interation term, although retaining its negative sign, beomes non-signifiant at standards levels of testing. e) EPL data is available up to 2003, while our firm level dataset ontains information for 2004 and We have investigated a possible extension of the OECD EPL index. The Fondazione Rodolfo de Benedetti has olleted information on EPL reforms in the period and lassifies them in strutural and marginal, depending on the sope of the regulatory hange. None of the ountries in our sample experiened strutural EPL reforms during , but some did follow marginal reforms. We have repeated our regressions under the assumption that the EPL levels remain onstant in eah ountry after 2003, and results (available upon request) are 28