Labor Court Inputs, Judicial Cases Outcomes and Labor Flows: Identifying Real EPL

Transcription

1 Labor Court Inuts, Judiial Cases Outomes and Labor Flos: Identifying Real EPL Henri Fraisse, Banque de Frane 1 Franis Kramarz, CREST-INSEE, CEPR, and IZA Corinne Prost, CREST-INSEE This Version: 11/05/2010 Abstrat Using a data set of individual labor disutes brought to ourt over the years 1990 to 2003 in Frane, e examine the imat of the enforement of Emloyment Protetion Legislation on labor market outomes. First, e resent a simle theoretial model shoing that judiial ase outomes annot be diretly interreted in terms of EPL. When a large fration of ases go to trial, it may ell be a sign of lo firing osts ith firms faing lo litigation osts and being therefore illing to go to ourt or, onversely, a sign of high firing osts of orkers faing lo litigation osts and being therefore illing to sue the firm. Seond, e exloit our model as ell as the Frenh institutional setting to generate instruments for these endogenous outomes. Using these instruments, e sho that labor ourts deisions have a ausal effet on labor flos. More droed ases and more trials ause more job destrutions: more trials indeed are a sign of loer searation osts. More settlements, higher filing rates, a larger fration of orkers reresented at trial, large layer density damen job destrution. JEL lassifiation: J32, J53, J63, K31 Keyords: emloyment rotetion legislation, labor flos, labor judges, unfair dismissal, Frane 1 Introdution The effets of emloyment rotetion legislation (EPL, hereafter) on labor markets have mostly been examined through hanges in the legislation (see for instane Autor, Donohue, and Shab, 2006 for the imat of exetions to the emloyment-at-ill in the US). Hoever, enforement of the la may matter more than its ontent (see for examle Bhattaharya and Daouk, 2002 ho find that insider trading las derease the ost of equity only hen a ase has been roseuted). Moreover, a legislative text never exists in isolation but ithin a eb of texts. Therefore, any alteration of the la in one dimension is likely to have an imat in other dimensions and eah la must be onsidered as embedded in the legal system (see for examle Autor, 2003 ho shos ho hanges in the Unjust Dismissal Dotrine imated the temorary hel industry in the United States). 1 Corresonding author: hf42@ornell.edu. +33 (0) Banque de Frane-31 Rue Croix-des-Petits Chams Paris, Frane. We are grateful to John Abod, David Autor, Jed DeVaro, Mar Ferrai, Larry Kahn, and Iona Marinesu for useful omments. We also thank seminar artiiants at the Banque de Frane, Crest, MIT, Ne York University, University of Chiago, Yale University as ell as artiiants at the EDHEC- GAIN-ADRES onferene, IZA summer shool, EALE onferene, and ESSLE onferene. The oinions exressed here are the authors on and do not neessarily reflet the vies of the Banque de Frane and Insee. 1

2 The Frenh EPL system haraterized by large searation osts, high overage by olletive bargaining agreements, oerful unions is usually onsidered by international organizations as one of the stritest. It also rodues every year a large amount of legal roedures related to individual labor disutes (roughly 160,000 ne ases every year, as e ill see). These roedures are omlex, ostly, and an last for years. It is these legal roedures that e analyze in this aer. We examine their imat on emloyment flos. Hoever, in ontrast ith virtually all of the literature, e do not fous on legislation hanges but on the judiial roess itself. We measure labor ourts outomes orkers or firms vitories, oniliations -- diretly using all ases that took lae in Frane beteen 1990 and We relate these outomes to the legal environment of eah Frenh labor ourt. We then try to understand ho these legal outomes affet job reation and job destrution ithin the jurisdition of eah ourt, using flos measured at those establishments that are ithin the jurisdition of eah ourt. This strategy is, e believe, better suited to measuring and identifying real EPL sine the mere ording of a ne legislative text often tells little about its real imat. We build a model relating firing osts to judiial ativities of the ourts. The model as ell as data ill hel us sho ho hanges in the legislative environment that diretly affet the legal osts of going to ourt transform the hole struture of otential labor ourts outomes. Beause e ant to relate legal ase outomes ith EPL, our ontribution to the literature has to be both methodologial and emirial. On the methodology side, e define here inreasing EPL uniquely by inreasing searations osts. Endoed ith this definition, and using Bentolila and Bertola (1990) s very general result that an inrease in searation osts dereases labor flos (both at entry and at exit), e are in osition to identify those legal outomes of labor ourts that onstitute real measures of EPL as those that derease labor flos. 2 Indeed, e sho theoretially and emirially that some features of legislation that are, aarently, heling orkers have the exat oosite effet and derease searations osts. The main reason for suh a surrising result omes from omosition effets. These omosition effets have diret onsequenes on final searation osts: for instane, if only exensive ases ome to ourt, on one side the firm eonomizes on the extensive margin (less ases) but ays more on the intensive margin (more exensive). Therefore, measured ase outomes do not have a simle and univoal interretation in terms of searation osts: more trials may sometimes mean larger searation osts, but it may also mean loer searation osts. This result does not only aly to labor ourts but also to divore or more generally to any legislation that alters the deisions of orkers, oules, firms hen they ontrat, sue, or indeed go to ourt. 3 2 Under onditions that e disuss later. 3 This has not esaed some analysts; see for instane Stevenson (2007) on legislation and divore rates. 2

3 On the emirial side, our ontribution is fourfold. First, e onsider measures of EPL varying aross sae and time based on the various measures of judiial ases outomes diretly oming from legislation enforement. In Frane, orkers an ontest the onditions of a firing by filing a ase to one of the 264 loal labor ourts. We use information olleted by the Frenh Ministry of Justie on all ases that ere filed over the eriod (2 millions of ases) to omute, at the loal level of the ourt and for eah year, various EPL indiators haraterizing the enforement of the labor las: fration of ases leading to a oniliation beteen arties, to a trial, resulting in a orker s vitory, to a ase dismissed by the ourt, or the fration of ases in hih orkers or firms ere legally reresented. We math these loal indiators ith loal measures of the legal environment (judges, administrative staff, layers ) as ell as loal measures of job flos à la Davis and Haltianger (distinguishing in artiular the extensive and the intensive margins, 1992). Seond, as e ork at the level of Frane, a ountry in hih many institutions are entralized and do not vary aross the Frenh territory (minimum age, unemloyment benefits, age bargaining ) e are able to ontrol for most of the Frenh labor market institutions, even though there is enough variation beteen as ell as ithin loal labor ourts. Third, thanks to the reise Frenh institutional setting and those loal measures of legal environment, e use an instrumental variables strategy to orret for the endogeneity from hih estimation of the relation beteen eonomi onditions, inluding labor flos, and aliation of the labor las might suffer. These instruments rely on administrative features residing over the alloation of judges and judiial lerks aross Frenh territory, on the loation of universities training layers, and on union traditions aross Frane. All these features are shon to be disonneted from loal business onditions. Then, e measure the effet of various ourt outomes on emloyment flos, identifying in so doing those that are valid measures of inreasing EPL and those that are not. Fourth, beause labor las in Frane do not vary aross loal areas (see hoever Chemin and Wasmer, 2009, on the notieable exetion of the orking time redution las in one Frenh region, Alsae-Moselle), a differene in differene aroah is essentially imossible to imlement in a Frenh setting. Our loal enforement indiators aroah (aired ith the aroriate instruments) offers one of the fe redible substitutes to this lassi identifiation strategy, in addition one that has never been used, to analyze Frane. As in all of the emirial aers e are aare of, our aer fouses on the imat of labor regulations on labor market harateristis and leave aside the elfare gains from job stability hih must be taken into aount for oliy reommendations. 4 Hoever, and in ontrast ith the existing emirial literature, our labor ourt outomes ature some dimensions of the quality of labor relations hih aording to Blanhard and Philion (2004) or Algan and Cahu (2009) are related to the evolution of labor market onditions. 4 See Bertola (2003) for a theoretial model onsidering risk-averse orkers and otential ositive effet of EPL on elfare. 3

4 Related literature: In a seminal aer, Lazear (1990), ho used unemloyment benefits and severane ayments given to a blue ollar ith 10 years of servie as a roxy for labor market flexibility, aknoledges that high overall labor market flexibility an revail ithout being atured by any of these to variables. Hene, one might onlude ith Freeman (2004) that ross-ountry analyses are hardly onvining sine ith only 30 or so advaned ountries, highly orrelated outomes, and infrequent hanges in institutions, the number of onfigurations an easily exeed the number of indeendent data oints. Moreover, in these ross-ountry analyses, the tyial EPL indies are deemed exogenous and imerfetly ature ho the behaviors of unions, emloyer federations, or government regulators hange over time. Taking stok of these ritiques, a reent strand of literature has assessed the imat of EPL ithin ountries. This strategy tyially involves measuring the imat of a hange in legislation targeted to a seifi ategory ithin a hole ountry or -- in the ase of the US -- the imat of the differential timing in the introdution of a ne EPL aross different states. Wrongful-disharge rotetions ere adoted by US state ourts during the last three deades. Autor, Donohue, and Shab (2004) take advantage of the beteen-state variation in the timing of the introdution of these labor las and estimate their imat using differene-in-differene estimators. The imlied-ontrat exetion la, meaning that the emloyer imliitly romised not to terminate a orker ithout good ause, is found to have redued state emloyment rate by 0.8 to 1.6%. Boeri and Jimeno (2005), using the 1990s tightening of the Italian regulation for firms ith less than 15 emloyees, find that the threshold does matter in onditioning layoff and hiring robabilities but find no signifiant imat on emloyment groth. Bauer et al. (2007) find no effet of the hange in the German EPL exemtion for small firms on orker turnover. Martins (2009) reahes the same onlusion using Portuguese mathed emloyeremloyee data. Marinesu (2009) uses the hange in emloyment seniority before being able to go to ourt it ent from 200 days to 100 to identify the rotetion effets of EPL. Unfortunately, this last strand of researh suffers from several donsides as ell. First, these studies do not rovide information onerning the degree of enforement of labor regulations. To hih extent these regulations ere used by orkers to defend their on interests? Are these regulations atually binding for emloyers? For instane, in the ase of the US, even if judiial breahes to the emloyment-at-ill dotrine have been judged by some state ourts, e have little evidene on the extent to hih they are used or even knon by the orkers. Therefore, e do not kno if they at as a redible threat to the emloyment-at-ill oliy. The state of California reognized the aliation of the ovenant of good faith and fair dealing to emloyment relationshis in In Marh and Aril 1986, about 100 ases ere filed in Los Angeles hih imlies an aroximate number of 1,000 4

5 for the entire year in the entire state 5 (hene, about 80 ases er million orkers). 6 In omarison, for Frane, ith a oulation and GDP lose to those of California, 160,000 ases take lae in any given year. Of ourse, one ould argue that the la an at on the emloyer in a re-emtive ay but, to ature any effet in the data, this imat should be very strong (or onversely the imat on emloyment of labor ourts should be extreme in Frane). Seond, labor las are subjet to ourt interretation and tend to vary over time and sae. As ointed out by The OECD 2004 Emloyment Outlook, even if an emloyer may be santioned in ase of non-reset of EPL, these rovisions are subjet to ourt interretation and this may onstitute a major (but often hidden) soure of variation in EPL stritness both aross ountries and over time. In addition, methodologially, the timing of introdution of a ne EPL an substantially alter the results. Indeed, Miles (2000) - using a different lassifiation of ases hen identifying the adotion dates - finds no signifiant effets. Third, roblems of endogeneity abound: ourt interretation ith the ensuing imat might not be exogenous as market onditions are likely to have an imat on the lenieny of the ourts, the introdution of ne las, or orkers roensity to go to ourt. Ihino et al. (2003), using miro data on labor ourt ases, fous on this institutional endogeneity of EPL enforement. Studying the ase of an Italian bank ith roughly 20,000 emloyees among hih 409 orkers ere fired and 86 of them ent to trial over more than 20 years, they sho that a higher unemloyment rate inreases orker s robability of inning her ase. In ontrast, Marinesu (2008) - using data from a 1992 survey of Emloyment Tribunal Aliations in Great Britain - finds that a higher unemloyment rate leads to more deisions against the orkers, in artiular hen they ere already re-emloyed. To summarize our results, e sho that labor ourts deisions have a ausal effet on labor flos. More droed ases and more trials ause more job destrutions: more trials indeed are a sign of loer searation osts. More settlements, higher filing rates, a larger fration of orkers reresented at trial damen job destrution. Setion 2 desribes the Frenh labor ourts institutional setting. Setion 3 resents a simle theoretial model relating the judiial ases outomes that result from the enforement of labor las to firing osts. Setion 4 desribes our data sets and rovides some desritive statistis. Setion 5 exlains our emirial methodology to ature EPL ausal effets and resents our regression results on labor flos. We briefly onlude. 5 In 1986, ivil ase filings in Los Angeles reresented about 60% of all ivil ativity in the state of California. 6 These figures are taken from Dertouzos (1988). 5

6 2 Labor Courts in Frane: the Institutional Setting 2.1 Frenh Firing Las Three tyes of events may trigger the firm s deision to fire a orker: a grave misondut, a lay-off due to a slodon in the business ativity, or an insuffiient level of ometene. Under the urrent Frenh la, the searation should be delared as a redundany (or eonomi dismissal). Hoever, in Frane as in many Euroean ountries an eonomi dismissal may entail a more omliated and time onsuming roess as ell as the ayment of large severane fees. On the ontrary, a dismissal for misondut is a faster roess - if not hallenged by the orker or if onfirmed by the labor ourt. Thus the dismissal for just ause imlies a loer firing ost than a redundany. When fired, a Frenh orker might sue the firm. Sine a bill assed in 1973, every individual dismissal must be justified by a real and serious ause and the firm has the burden of roof. Without delving dee into 30 years of jurisrudene that have made this onet simultaneously blurred and reise, real means that the rongdoing justifying the dismissal must be objetively defined, aurate, and in line ith the mandatory firing notifiation letter. For examle, being ten minutes late does not mean being seventy minutes late; a lak of erformane or a lak of trust is not onsidered real if it is not objetively measured. The ause is onsidered as serious only if it is related to the rofessional ativity of the orker and if it makes the labor relation imossible to ontinue. There are various degrees of seriousness. Some lead to grave misondut (for examle bral or thievery) hih allos the emloyer to fully derive the orker of severane ayment (in this ase, the emloyee may lose her unemloyment benefits). In addition to the ause of the dismissal, the emloyee an sue the emloyer if the latter did not follo the mandatory legal stes of the firing roess (for examle the emloyer must notify one eek in advane that the emloyer intends to meet the orker in order to disuss his firing). As ointed out by Galdon-Sanhez and Guell (2003), EPL legislation in Euroean ountries may give rise to a double moral hazard roblem: a orker fired for misondut has an inentive to sue for unfair dismissal and a firm has an inentive to label misondut a searation hih, in reality, is a redundany. Thus, even if this henomenon is obviously hard to detet in the data, the roortion in Frane of dismissals for eonomi reasons dereased from 61% in 1993 to 24% in Frenh Labor Courts The Frenh labor justie is mainly disensed by the Prud hommes hih is the relevant jurisdition to every labor disute arising at the individual level in Frane. There are several labor ourts in eah Prud hommes. As the legislators anted to take into aount industry harateristis of the ases 6

7 brought to ourt, eah Prud homme is divided into 4 setions aording to the main ativity of the firm: Agriulture, Retail Trade, Manufaturing, and Other Ativities (mainly Servies). A fifth setion is dediated to ases involving managers, engineers, and rofessionals irresetive of the ativity of the firm. The judges in the Prud hommes are not rofessional judges and are seen by some as erforming a ubli duty. Eah labor ourt omrises judges reresenting emloyers and judges reresenting emloyees in equal number. These judges are eleted every five years ithin lists established by orkers unions and emloyer federations. All emloyees are entitled to vote. They selet judges in the union lists. Similarly, emloyers vote and selet judges ithin the federation lists. Hene, in every setion of every Prud homme, the number of judges is even, at least four on eah side. All Frenh establishments are alloated to one Prud hommes. On the emloyee side, the eletoral body inludes all rivate setor orkers ith a labor ontrat. They are enrolled on the eletoral list based on a mandatory administrative reort from their emloyer. Unemloyed individuals an also vote but have to enroll on the list by themselves. On the emloyer side, in addition to emloyers and business oners, emloyees entitled to take firing or hiring deisions an also vote for emloyer reresentatives. Prud hommes are suosedly not very formal and should be seen as oniliation boards. Prud hommes ere designed to foster agreements rather than trials. Therefore a first and mandatory ste in eah trial is a oniliation audiene here laintiffs and defenders exlain their grievane and judges try to ush for an agreement. If they do not, the ase is judged. If, in the end, an equal number of judges deides in favor of a orker and against her, there is a tie ( solution de déartage ). In this ase, a single rofessional judge deides the outome of the trial. In the 90 s, 264 Prud hommes ere sread all over metroolitan Frane, a labor ourt being at most ithin a radius of 30 miles from any establishment. A majority of laintiffs ere reresented by a layer. The laintiff or the defender an aeal the deision of the ourt if the stake is larger than a given threshold (about 5,000 euros in 2006). It is orth noting that 60% of the deisions ere aealed in Among them, 55% of these aeals did not overrule the Prud hommes deision, 30% onfirmed it artially. 7 In ase of an emergeny, a summary judgment an be made. Hoever, suh judgments are only temorary and might be overruled afterards. In this aer, e do not onsider these summary judgments. 7 Munoz-Perrez and Serverin (2006). 7

8 For any given ase filed in a labor ourt, the range of outomes is ide. A ase an lead to a full tribunal hearing and be lost or on. It an be lassified as null and void if the laintiff has not shon due diligene in the ondut of her ase. The ase an also be rossed out. This rossing out is less severe than a null and void lassifiation. The orker an reinstate her ase at the oint it has been rossed out and does not have to restart the hole roess. This rossing out an be deided by judges but it an also be the outome of the laintiff s initiative. 8 A ase an either be oniliated during the oniliation ste or outside the tribunal ith a formal agreement sent to the ourt. The motives for suing are multile. The nullifiation of a dismissal is asked in the majority of ases (58%). 9 21% of laintiffs ask for some omensation that as not aid by their former emloyer hereas 9% of laintiffs do not agree ith the level of their severane ayment. In this aer, e do not distinguish beteen these different motives. 2.3 Reent Changes The legal environment did not hange substantially during our samle eriod ( ). 10 In the relatively reent ast of an institution offiially founded in , a 1979 bill radially hanged the institutional settings of the Prud hommes. First, it extended the number of Prud hommes aross Frane in order to guarantee an equal aess among orkers. Seond, it ended the majority rule for eleting reresentatives hih resulted in a more diverse omosition of eah Prud homme. Third, it funded the Prud hommes by the entral administration (Ministry of Justie), an imortant feature in our analysis sine national riniles ruled funding hih entailed muh less sensitivity to loal hanges in the eonomi environment. 3 Litigation and Firing Cost: a Simle Theoretial Model We do not study here the theoretial imat of firing osts on labor market variables. This has been extensively examined elsehere (see Bentolila and Bertola, 1990 and Bertola, 1992). We rely on these authors to say that larger firing osts entail sloer and smaller adjustments than searations induing no osts. Here, e try to illustrate ho firing osts are related to labor ourt inuts and outomes. To do so, e deart from the traditional model of litigation roosed by Priest and Klein (1984) or 8 In 2004, only 27% of rossed out ases ere reinstated. 9 In the very vast majority of the ases on by the orkers, they are not reinstated but reeive a omensatory aard. 10 Minor hanges took lae in relation ith the aliation of the orking time redution and the 35 hours orkeek. 11 Prud hommes an be traed bak to the Middle Ages. 8

9 Bebhuk (1984) or more reently Card and MCall (2009) to run a ost-benefit analysis similar to the one roosed by Flanagan (1989) for disutes related to the omliane to the National Labor Relations At in the US. For simliity, the model that e desribe belo has no unertainty, no asymmetri information that ould exlain hy trials take lae; everything is knon and reditable; e ill ome bak later on this toi and disuss ho our results are affeted by asymmetri information. In our analysis, the emloyer an deliberately hoose to ay a minimal firing ost ith the risk to be sued by the orker; or to ay a larger amount, hih orresonds to the ayment a laintiff ould aet in order to give u any further ossibility of lasuit. Imortant to note here that this last sum is not negotiated beteen the firm and the orker, but is diretly oming from legal reedents (jurisrudene). In Frane, it amounts to one to to years of earnings. Another ay of understanding the model is as follos: a firm hooses to dismiss the orker either for a ersonal motive, aying a small or even zero severane ayment in ase of a disilinary ase, or to dismiss the orker for an eonomi motive (redundany) ith larger severane ayments. Our hyothesis, then, is that hen firms ay the severane ayment orresonding to a redundany, the orkers never hoose to sue the firm (indeed, 97.5% ases in our data ome from dismissals rather than redundanies). When the orker goes to ourt after a dismissal, the firm has to rove that the ase is a legitimate dismissal for ersonal motive rather than a redundany. In the ase of a dismissal for ersonal motive, the firm inurs a minimum ost ( m ) if the dismissal remains unhallenged by the orker. This ost m is loer than the maximum ost M, hih leads the orker not to sue the firm. Yet the firm has to take into aount the robability that the orker files a suit, f, the robability that the ase ends ith a formal agreement in front of the judge, and, hen the oniliation ste fails, the robability that the orker ins,. We assume that during the oniliation ste, the judge tries to reah an agreement using an intermediary ost, given by the jurisrudene, alays loer than M. Note that in order to simly introdue the o-existene of a oniliation stage and a trial stage e onsider to be onstant. The firm annot inrease in order to avoid the trial. Unertainty of the entire roess is summarized through. The firm and the emloyee share this value. In this ost-benefit analysis, e assume that the quality of eah ase is knon by both arties and is related to observed harateristis of the orkers and of the firms. 12 For instane, union or ersonnel delegates or regnant omen are very ell roteted by the la, and the judges are very strit against dismissals of suh individuals. Several ast statements of judgments also sho that 12 As mentioned above, this assumtion is disussed belo. 9

10 judges demand more stringent evidene hen a firm has had large ositive rofits in the years reeding the trial. 13 At this oint, the model has no unertainty, no asymmetri information that exlains hy trials take lae. Theoretially, firms and orkers should agree on a ayment in order to avoid the litigation osts. To features ould be added to the model in order to exlain hy firms and orkers go first to the Prud hommes and then, if ever, to trial. First, osts for reahing an agreement ithout any mediator like the judge ould be larger than the osts at the oniliation stage. This seems lausible sine the Prud hommes institution is seen as a ubli good and the oniliation stage is essentially free. Seond, in line ith the literature in hih trial is an equilibrium outome, e an assume that the orker and the firm have different and irreonilable exetations on the outome of the trial. This assumtion ould lead to a ontrat zone here a settlement amount an be found (see Bebhuk, 1984). When the exetations are not in the ontrat zone, the trial takes lae; else an agreement an be found at the oniliation stage. Beause there is a need to model exetations, omutations beome muh less tratable. The model loses its simliity ithout gaining muh insight for our urose. In addition, as underlined by Sier (2007) suh a model does not fully solve the litigation uzzle sine the oniliation stage should hel the exetations to narro. Again, it is ossible to assume to stages of narroing exetations, one leading to oniliation, one to the trial. 14 From this disussion, it is hoever interesting to note that orkers emloyed in large firms are virtually absent from the Prud hommes. In line ith the above disussion, the various robabilities should be better knon by the human resoures management and union delegates that are alays resent in the larger firms. Hene, they should esae trials and easily agree on searation ayments, as is observed. In small firms, onflits beome often ersonal and diffiult to solve ithout the hel of a neutral third arty, a role aarently layed by the Prud hommes. No, let us go bak to our model. If hoosing m the exeted firing ost for the firm is: E () ( + l ) + (1 )[ ( + F ) + ( 1 ) + l ] f { m m t } + ( f ) m = 1 here F is a omensatory aard for the orker, l is the firm s litigation ost hen the arties reah an agreement at the oniliation stage, and M Otherise, the firing ost is. l t is the firm s litigation ost hen the arties go to trial. 13 Unfortunately, the data do not ontain a firm identifier. Hene, it is not ossible to diretly relate firm and orker behavior. 14 Comutations not inluded in the Aendix are available from the authors. 10

11 As for the orker, if the firm hooses m, the exeted severane ayment is: E () s ( k ) + (1 )[ ( + F ) + ( 1 ) k ] f { m m t } + ( f ) m = 1 k being the ost of litigation for the orker at the oniliation stage, k t being the ost at the trial M stage. Otherise, the reeived ayment is. The arameters f and result from the otimization from the firm and the orker and equal one or zero. The orker hooses to go to ourt or not; and in that ase to aet the oniliation or not. The firm hooses to ay m or M ; and, if at the ourt, to aet the oniliation or not. k We define: = t F, m + kt k =, F Assumtions: * m lt + l =, and F ** = * Condition (1): k + l l : the ost of trial is suffiiently large ( < Condition (2): k > M t > m t m F m l : the gain for the orker at the oniliation stage is larger than the severane ayment she reeives in ase of firing for a ersonal motive ( Condition (3): + l < M < ). : the ost for the firm at the oniliation stage is smaller than the severane ayment reeived by the orker in ase of firing for an eonomi motive. Conditions (1), (2) and (3) taken together allo for the ossibility of a oniliation stage. Condition (4): The omensatory aard is large enough so that hen the firm is ertain to lose at trial, it is less ostly to ay the maximum severane ayment. That is: F < + F + l M m ) t t. It imlies ** < 1 and exludes an equilibrium in hih the la has no deterrent effet, every orker being fired for a ersonal motive. Condition (5): k + k + l < better off at trial than aying M. t t M : there is a robability range for a trial to exist. The firm is Result: Under these assumtions, there are four equilibria, deending on the value of (see Aendix for a roof): Equilibrium (1): for small robabilities <, the firm ays m and the orker does not go to ourt ( = 0 and = 0 ) sine the firm ould refuse the oniliation and the gain at trial ould be f negative for the orker. Equilibrium (2): hen < <, the exeted gain of the orker at trial is ositive. She an redibly threat the firm to go to a full hearing. Sine * < the firm aets to settle ith the 11

12 orker. The settlement amount is loer than the exeted loss of the firm at trial but larger than the exeted gain of the orker. ( = 1 and = 1). f ** Equilibrium (3): hen < <, the orker is better off at the trial stage and refuses to oniliate anymore ( = 1 and = 0 ). f ** Equilibrium (4): hen >, the firm ays M sine it ould be too ostly to go to ourt. The firing ost, deending on the robability of inning of the ase, is shon on Figure 1. m ( ) m lt ( + F) M + l m ** no judiial ase oniliation trial no judiial ase Fig. 1: Firing ost Disussion: We onsider that firms are faing an invariant distribution of ase quality. This model shos that hanges in the inuts have intuitive imats on the firing osts. For instane, an inrease l t in the firm s litigation ost at trial imlies a derease in (Figure 2). The firm ill be more likely to fire high robability orkers ith an eonomi motive to avoid lasuits. It ill also inrease the relative ost of a trial ith reset to oniliation. The exeted firm s ost at trial ill rise. All these hanges lead to an inrease in the firing ost, a derease in the trial and filing rates. Hene, a ** 12

13 smaller number of trials and filing ases aear to be assoiated (in some ases at least) ith more EPL, in ontrast to the usual vie romoted by OECD that equates trials ith EPL. m ( ) m lt ( + F) M + l m ** no judiial ase oniliation trial no judiial ase Fig. 2: Firing ost, ase outomes and an inrease in the litigation osts for the firm Let us study the folloing ase that stands in stark ontrast ith the revious one (Figure 3). An inrease in orkers litigation osts k t ill indue a dereased robability for the orkers to file a ase (through a higher ) as ell as more orkers that refer to oniliate (through a higher ). In this situation (ontrasting ith the revious one), a smaller number of trials is assoiated ith loer total firing osts. In fat, our model shos that firing osts diretly deend on variation in inut osts; yet the link ith ase outomes is ambiguous. 13

14 m ( ) m lt ( + F) M + l m ** no judiial ase oniliation trial no judiial ase Fig. 3: Firing ost, ase outomes and an inrease in the litigation ost for the orker The model an also be used to think about the instruments and disuss the endogeneity roblems that e ill fae hen estimating the relations beteen labor ourt outomes and labor market harateristis. For instane, an adverse shok on the labor market onditions an affet labor ourt outomes. First, aording to the legislator, F omensates the orker for ast and future otential age losses, in artiular by taking into aount the diffiulty of finding a ne and omarable job. The magnitude of F is therefore likely to be ounterylial. An eonomi donturn ushes,, and ** donards hih results, other things being equal, in higher firing osts. 15 Moreover, if orkers exert more effort on the job during a donturn (ith an effiieny age story in mind), the overall distribution of ill shift uards. Third, eonomi onditions might also alter the overall distribution of through judges behavior. Judges shoing a ro-orker bias hen labor market onditions deteriorate inrease the firing osts faed by the firms (see Ihino et alii, 2003). 15 Emirially, Siegelman and Donohue (1995) find that ases of emloyment disrimination rise in donturns and are more likely to be lost. In our model, it means that the derease in ** hen omuting the average orker s vitory rate. more than offsets the derease in 14

15 Our identifiation strategy is based on variations of the legal environment aross years and labor ourts as ell as ithin ourts. These variations might imat ase outomes and the ost of roedure ithout being related to urrent loal eonomi develoments. First, labor judges and judiial lerks involved in labor disutes are unequally distributed over Frenh territory, leading to differentially ongested labor ourts. This imlies an inreasing marginal ost of hallenging the dismissal folloing Buhanan s lub theory of ubli goods and thus loers the firing osts through higher. Seond, one might lausibly assume that a higher layer density indues a stronger ometition amongst layers ushing don the ost of legal reresentation ( k and k ). Third, judges orking at the Prud hommes are union members (emloyees or emloyers) and their behavior is likely to be shaed by a long tradition of industrial relations adoted by their union at the national level and loosely related to loal eonomi develoments. For instane, some unions tend to favor agreements over onfrontation. This should also imat ase outomes. We ill disuss more in deth the exogeneity of these indiators in Setion 5. 4 Data Set and Desritive Statistis 4.1 Individual Cases Data Set Firms and Workers Charateristis 16 t k t Our data soure on individual ases omes from administrative reords made at the level of eah Prud homme and olleted by the statistial deartment of the Frenh Ministry of Justie. Their rimary goal is to monitor labor ourts ativities ith an emhasis on seed of treatment. The data soure is exhaustive for the eriod 1990 to It inludes aroximately 2 millions individual ases. 17 Aart from years 1993, 1994 and 1995, the number of ases treated by labor ourts aears to be stable over the eriod, in stark ontrast ith hat haened in some ountries suh as the UK here a shar inrease took lae (see Figure 4 and Burgess, 2001). For eah ase, the sex and age of the emloyee-laintiff is reorded. There is no reise information on her skill-level in the firm. Nevertheless, the managers setion of the Prud hommes only deals 16 In our aroah, the ongestion of the labor ourts inreases the ost of litigation hih deters orkers either to file or to refuse the settlement and therefore dereases the firing ost. To reent aers by Besanenot and Vraneanu (2008) and Stahler (2008) relate labor ourts to labor markets through the existene of judiial mistakes. The ongestion of the ourt ushes uard the number of judiial mistakes. Bad tye orkers that is orkers that have rightly been fired for a ersonal motive have an inentive to go to ourt gambling on a judiial mistake. This is found under some assumtions and ontrasting ith our vie to have an inreasing imat on the firing osts. 17 We ill not onsider the 2% of ases involving emloyers as laintiffs. 15

16 ith high-skill emloyees and managers. Similarly, sine lo-inome orkers are eligible for finanial hel (13% of the ases) eligibility an be used as a lo-inome indiator. Aroximately one half of these ases are susetible of aeal, 18 hih imlies that the sums at stake are larger than 5,000 euros (in 2005). 53% of the emloyees are reresented by a layer. As for firms harateristis, e kno the industry, the size, and the Prud hommes jurisdition of the emloying firm (i.e. the industry broadly defined). Hoever, e an only differentiate beteen firms ith more and firms ith less than 10 orkers. The size of the firm has to be knon by labor ourt judges beause labor las differ for small firms; more seifially, they are less stringent and try to ease the finanial osts of firing that ould hurt them irreversibly. Small firms are overreresented ith 56% of the filed ases hereas they omrise 25% of the labor fore. For eah ase, the starting date, the ending date, the motives for dismissal, and the ourt deision are reorded. An average ase takes aroximately one year (343 days) ith a standard deviation of 9 months Cases Outomes Indiators Using the individual ases data set, e are able to omute various ase outomes indiators for eah Prud hommes and year. A ase an follo various aths and every grouing is somehat arbitrary but e try to follo elements of our model as muh as ossible. We start by omuting the folloing ases outomes indiators using the most disaggregated lassifiation: inning (res. losing, null and void, rossed out, oniliated, agreement and tied ) is omuted as the ratio of the ases lassified as orker s vitory (res. defeat at trial, null and void, rossed out, oniliated, having led to an agreement, having been judged by a rofessional judge) in year t over the number of ases disosed in year t. We then grou ases in oniliated (ases oniliated or having led to an agreement), droed ( null and void or rossed out ) and trial (ases having reahed trial) and omute rates aordingly, see Table 1. We also omute the orker s vitory rate at trial ( vitory ). About 60% of ases ended by a trial, among hih 75% led to a orker s vitory (see Table 2). Desite the mandatory ste of oniliation, only 11% of the ases ended at this stage. Taking into aount ases that led to an agreement notified to the ourt or to a ithdraal on the orker s side, at least 20% of the filed ases led to an agreement. 20% is also the roortion of ases having been 18 Unfortunately, urrent available data sets do not allo us to trak the ases aross the levels of jurisditions. Whether the deision is aealed by the orker or the firm is also unknon. In 2004, aording to Serverin and Munoz-Perez (2006), among the oulation of aealed ases, only 15% ere overruled. 19 Beause e use ourt-level information for our analysis, rather than ase-level information, our Tables ill reort ourt-year statistis. All ase-level statistis are available from the authors on request. 16

17 droed. All ase outomes or judiial environment indiators dislay a very strong variane over time and aross Prud hommes. In omarison ith hat is observed in a ountry suh as the UK, a large fration of orkers and firms are reresented by a layer desite other available means of reresentation. Desite the oniliation ste hih romotes a quik and ostless resolution of the ases, labor disutes seem to indue imortant litigation osts. Our model hel us understand the relative frequenies of these ase outomes, in artiular the often lo value of k, the orker s litigation ost and l, the firm s litigation ost. Admittedly starting from a high base, e do not observe in Frane a strong inrease in the number of ases brought to the labor ourts. In absolute terms, the number of filed ases inreased by 10% over the 1990 to 2004 eriod. The number of filed ases by unemloyed orkers hovered around five erent over the same eriod. The ylial behaviors of olletive onflits have been thoroughly studied in the literature ithout reahing a onsensus: strike inidene has been found roylial (see Harrison and Steart, 1994) in the Canadian ontext and strike outomes tend to be more favorable to unions hen the national unemloyment rate is loer in the British ontext (see Devereux and Hart, 2010). Evidene of any studies on the ylial behaviors of individual onflits is sare. A regression of the different indiators of outomes on loal measures of the business yle shos that the ase outomes are strongly orrelated ith the yle (see Table A.1) and that traditional labor regulation indies are highly imerfet in this reset. As the mean duration of a ase is around one year, e onsider both the relationshi beteen the outome of the ase and the unemloyment rate at the time of the deision and the relationshi beteen the outome of the ase and the unemloyment rate at the time of filing. In both ases, a high unemloyment rate is assoiated ith a high trial rate, a high inning rate and a small number of agreed, oniliated, or droed ases. Therefore a deteriorating labor market hardens the judiial roess leaving less room to informal deisions. In the same time there are less ases relatively to the number of firings, sine the filing rate dislays a strong ounterylial attern. 4.2 Loal Emloyment Data Set Labor Court Level Loal emloyment flos at the establishment level are omuted from the SIRENE files, maintained at the Frenh statistial institute (INSEE). These files give the reise loation (ity ithin a déartement ) for eah establishment. We omute a set of Davis and Haltianger (1992) indiators over the eriod: job reation (both at the extensive and the intensive margin), job 17

18 destrution (both at the extensive and the intensive margin), and net job reation variables over the eriod. These measures are aggregated by industry (servie, trade, and manufaturing) and size of the establishments (more or less than 10 emloyees) at the ity level as ell as at the Prud homme level, using a 1999 orresondene beteen ities and Prud hommes rovided by the Ministry of Justie. In omarison ith ross-ountry analyses, these indiators also sho a high heterogeneity aross eriods and aross the 264 areas. To measure loal unemloyment, e use the number of unemloyed as registered at the National Labor Ageny (ANPE) for eah ity as ell as the ity labor fore as measured at the 1999 Census. Finally, from 1997 on, e are able to distinguish the reasons for losing one s job (eonomi or ersonal dismissal, entry into the labor fore, end of temorary ontrat...). Unfortunately, there is no data set giving us, at the loal level of the ity, the size of the temorary hel servie industry. Hene, e annot erform an Autor s tye analysis. Hoever, in ontrast ith other Euroean ountries (suh as Sain), the fration of temorary orkers in Frenh total rivate emloyment is lo (about 2.5% in 2009). 4.3 Eletion Data Set The eletions for the Prud hommes are ruial in Frane - at least for the trade unions - as they are the only ay to assess unions reresentative harater at the national level. Over our samle eriod, 4 rounds of eletions took lae, in 1987, 1992, 1997, and For eah round, e olleted the share of judges affiliated ith eah union as ell as the number of judges by setion at the Prud homme level. Union shares of votes are rather stable over time but dislay a great deal of heterogeneity aross Prud hommes (see Table 3). The number of judges did not hange from 1993 to 2002 (see Table 4). Changes took lae in 1992 and after 2002 (see Table 5). For the 1992, 1997 and 2002 rounds, e olleted the turnout rates and the number of orkers ho ere enrolled in the eletoral lists for eah Prud homme. 4.4 Additional Judiial Data In Frane, eah layer has to get liensed and registered at the Bar ( barreau ) in order to be entitled to ratie. We kno the number of layers registered at eah barreau from 1996 to It allos us to have a loal estimate of the number of layers by emloyed orker. As there are feer bars in Frane than Prud hommes (181 versus 264), e math eah Prud homme to the losest bar using shortest route distane and omute the number of layers available to emloyees deending on one single Prud homme. Using the 1999 Census, the national average is 77 layers er 10,000 emloyees, 18

19 going from a minimum of 14 (Creuse) to a maximum of 868 (Paris). From our miro data set on Prud hommes ases, e are able to omute the number of orkers ho ere reresented by a layer by labor ourt and year. We observe a very high orrelation (0.68) beteen the layers densities omuted from these to different soures. Layer data over a shorter eriod than other variables ( ). In addition, e obtained the number of greffiers (lerks) emloyed by the Ministry of Justie attahed to tribunals in the area of eah Tribunal d instane 20, losest to the labor ourt ( Staff hereafter) over the eriod 21. Greffiers are ivil servants in harge of all the administrative tasks, hih inlude assisting the orkers in filing their ases as ell as riting the judgment terms. Their alloation is entrally set (by the Ministry of Justie). 5 Identifiation Strategy and Results 5.1 An Instrumental Aroah We resented in Table 1 our measures of ourt deisions, roerly aggregated to have a simle enough vie of a omlex sequene of outomes. As learly aeared through the model analysis, these deisions droing ases, orkers vitories or losses, settling ases, going to trial, or the duration of a ase are ambiguously related to firing osts. In hat follos, e try to examine the ausal imat of these measures of judiial ases outomes on emloyment flos. We rely on instruments, derived from our model, that are likely to affet the various osts of litigation for the orkers and the firms. They are desribed in the next subsetions after having resented the simle estimating frameork that e adot Estimating Equations We intend to estimate the folloing eonometri model: Flos, t = α 1BC, t + α 2BC, t 1 + βoutome, t + δ + γ t + ε, t (3) 20 As there is more tribunal d instane than Prud hommes (460 versus 264), e use again orthodromi distane for the mathing. 21 Data linearly interolated for 1993 and

20 here Outome, t is a measure of judiial ases outomes at the unit of observation and for year t. BC,t is a business yle indiator. The unit of observation is the Prud homme. Our labor market variables Flos,t are the labor flos at the Prud homme level. δ is a Prud homme fixed effet; γ t is the year effet and ε,t is the statistial residual. In eah regression, observations are lustered at the loal labor market area level. The labor market areas defined by the jurisdition of the Prud hommes dislay a large heterogeneity in size (measured by ative oulation or emloyment). Half of the Prud hommes aount for about 80% of the 1991 total emloyment. We eight our regressions by the 1999 ative oulation of the Prud hommes area (results are unhanged hen using total emloyment at the start of the eriod under revie, 1991). Table 6 resents results of regression (3) ith and ithout fixed effets using OLS. No lear attern is aarent in this Table and, again, it is imossible to kno if an inreased number of oniliations auses job destrutions or job destrutions ause more oniliations in ourt. Therefore, to estimate the arameter β measuring the ausal imat of the judiial ativity on the labor market flos, e adot an instrumental aroah by rojeting our outome indiators on instruments Z, business yle indiators, year dummies and loal labor market fixed effets: Outome, t = μ 1BC, t + μ 2BC, t 1 + λz, t + δ + γ t + υ, t (4) Clearly, the business yle BC is endogenous and also needs to be instrumented: unobserved eonomi shoks might simultaneously imat the quality of the ases brought to labor ourt, bias the judges in their deisions, and affet the labor flos. To do this, e instrument the measure of loal business yles (number of unemloyed registered at the loal emloyment ageny on the 1999 loal labor fore) by the national unemloyment rate (in the sirit of Bartik, 1991 or Blanhard and Katz, 1992) using the folloing relation: U, t δ + γ t + μ U aggregate + η, t = (5) Then, e use the redited value Û of U by (3) to omute our exogenous measure of yle BC as ( U Uˆ )/ U here U is the average of the redited loal unemloyment rate Û Soures of exogenous variation Suitable instruments for estimating the arameter β must exlain the average outomes observed at the level of the labor ourt and be exogenous to urrent labor market develoments. We laim that the institutional settings of the Prud homme itself and the loal legal environment rovide onvining instruments beause, as our model disussed, they are related to ase outomes by affeting various osts of litigation and beause their variations ithin eah Prud homme are essentially random. 20

21 a) Layers One of our instruments is the number of layers enrolled at the loal bar layers of all seialties, not only those seializing in labor disutes, a small fration of the total -- saled by total emloyment of the Prud homme area or the déartement in 1991 ( layer density hereafter). A high layer density is likely to redue legal fees thanks to a higher ometition (see Siegelman and Donohue, 1995 for a similar argument). It also hels to disseminate legal exertise and judiial knoledge of labor disutes among the oulation of orkers. It should orresond to a loer ost of litigation for the orker and hene influenes the outome of the ase. 22 One ould argue that the layer s hoie of loation deends on loal eonomi onditions. First, labor disutes are only a small amount of the total number of ivil ases (11% at the national level 23 ). Seond, in order to get a liense to ratie, a layer must enroll the loal bar hih jurisdition the Prud homme belongs to. This requirement and the building of a reutation and a lientele indue a lo mobility of layers from one region to another. Moreover, a layer tyially enrolled the bar the ity here she studied and her loation referene is likely to be unrelated to the inidene of labor disutes litigation. To see this, first note that there are only telve la shools sread over Frenh territory (see Figure 5). Then, observe the strong overla beteen these areas here layers are trained and those that see the strongest inrease in layer density over time (see Figure 6). Further suorting the identifying assumtion that loal labor market onditions are disonneted from the inrease in layer density, lagged labor flos are found to have no rediting oer on layer density inluding rud hommes fixed effets and year dummies (see Aendix Table A.2). Thus the layers density influenes judiial outomes through the ost and the effiieny of the litigation roess but are likely to be random ithin a Prud homme ith reset to urrent labor market develoments, therefore making it a lausible instrument. b) Clerks and Judges We also onsider as instruments the number of judges and staff in harge of dealing ith judiial ases (saled by the loal 1991 emloyment or by the loal 1999 ative oulation). Both ategories obviously have an imat on judiial deisions as ell. Judiial ativity an be modeled as a rodution funtion for the ase disosition. Beenstok and Haitovsky (2004) using a anel data on Israeli ourts find that judges omlete more ases as their aseloads gro and omlete feer ases hen ne judges are aointed to their ourt. In the ase of Prud homme, the soiologial literature 24 suorts this result and states that faing an inreasing number of ases and having to meet some 22 Logit regressions using our data set of individual ases reahing the trial stage shos that hiring a layer against a unreresented firm inreases the robability of orker vitory by about 4%. 23 See available on line Info Stat justie (2005) «Une évaluation de l ativité des juriditions en 2004» n See Bonaffé-Shmidt (1987). 21