LABOR CONTRACTS AND FLEXIBILITY: EVIDENCE FROM A LABORMARKET REFORM IN SPAIN

VICTOR AGUIRREGABIRIA AND CESAR ALONSO-BORREGO∗

This paper evaluates the effects of a labor market reform in Spain that removedrestrictions on fixed-term or temporary contracts. Our empirical results are based onlongitudinal firm-level data that cover observations before and after the reform. Weposit and estimate a dynamic labor demand model with indefinite and fixed-term laborcontracts, and a general structure of labor adjustment costs. Experiments using theestimated model show important positive effects of the reform on total employment(i.e., a 3.5% increase) and job turnover. There is a strong substitution of permanentby temporary workers (i.e., a 10% decline in permanent employment). The effects onlabor productivity and the value of firms are very small. In contrast, a counterfactualreform that halved all firing costs would produce the same employment increase as theactual reform, but much larger improvements in productivity and in the value of firms.(JEL J23, J32, J41)

I. INTRODUCTION

Regulation of workers’ dismissal is one ofthe labor market institutions most commonlyinvoked to explain the large and persistent dif-ferences between European and North Ameri-can unemployment rates. The consequences ofjob security provisions on labor market per-formance have been broadly analyzed both atthe theoretical and at the empirical level. Froma theoretical point of view, the effect of fir-ing costs on employment is ambiguous. Fir-ing costs reduce both hiring during expansions,and dismissals during downturns. The net effectdepends on different factors, including the sizeof hiring and firing costs and the persistence

*We would like to thank the comments received fromJeff Campbell, Andres Erosa, Jose E. Galdon, James Heck-man, Andreas Hornstein, and Ariel Pakes, as well as fromseminar participants at the University of Chicago, Universityof Minnesota, Queen’s University, University of Waterloo,University of Western Ontario, and Yale University. We alsoacknowledge Ricardo Mestre and the staff of the Central deBalances del Banco de Espana for providing the raw data.The second author acknowledges research funding from theSpanish Ministry of Finance and Competitiveness, Grant no.ECO2012-31358.Aguirregabiria: Professor, Department of Economics, Uni-

versity of Toronto, Toronto, Ontario M5S 3G7, Canada.Phone 1 416 978 4358, Fax 1 416 978 6713, E-mailvictor.aguirregabiria@utoronto.ca

Alonso-Borrego: Associate Professor, Department of Eco-nomics, Universidad Carlos III de Madrid, Madrid28229, Spain. Phone 34 916249749, Fax 34 916249875,E-mail cesar.alonso@uc3m.es

of demand and supply shocks. The empiricalstudies differ in the data used (aggregate data,industry-level data, household- and firm-leveldata), the scope of the analysis (from the studyof a particular country to cross-country compar-isons), and on the methodological approach. Theresults are not conclusive, particularly regardingthe effects on the level of employment. There-fore, the employment effect of firing costs is anempirical question that should be evaluated caseby case. The labor market reforms implementedby several European countries since the 1980sprovide unique information to identify the effectof firing costs on labor market outcomes.

In this article, we study the effects onemployment, job turnover, and firms’ produc-tivity of a Spanish labor market reform thattook place in 1984, which liberalized the use oftemporary contracts and reduced the redundancypayments at termination of these contracts. Afterthe reform, temporary contracts could be applied

ABBREVIATIONS

CBBE: Balance Sheets of the Bank of SpainET: Workers’ StatuteGDP: Gross Domestic ProductMLE: Maximum Likelihood EstimatorNPL: Nested Pseudo LikelihoodOECD: Organization for Economic Co-operation and

Development

1

Economic Inquiry(ISSN 0095-2583)

doi:10.1111/ecin.12077© 2014 Western Economic Association International

2 ECONOMIC INQUIRY

by any firm, irrespective of their size, industry,or performance, to any type of worker, irre-spective of their occupation, age, or gender.Nevertheless, the stringent dismissal regulationsfor indefinite-duration or permanent contractsremained unchanged after the reform.

Our approach is based on the estimation of amicro-econometric dynamic structural model oflabor demand, with permanent and temporaryemployment, and on the comparison betweenthe pre-reform and post-reform periods in theestimates of the structural parameters and in thesteady-state distributions of employment. Ourprimary source of data consists of a longitudinalsample of 2,356 Spanish manufacturing firms,during the period 1982–1993, with informationon employment by type of contract, capital,output, and wages. In our model, we considerthat the reform may introduce changes in firingand hiring costs of temporary workers, and intheir productivity relative to permanent workers.

The estimated hiring and firing costs for tem-porary workers are significantly lower after thereform. Experiments using the estimated modelshow important positive effects of the reform ontotal employment (i.e., a 3.5% increase) and jobturnover in steady state. Further, the reform ledto a strong substitution of permanent by tempo-rary workers, with a 10% decline in permanentemployment. The effect on labor productivity isnegative, and the effect on the value of firms isnegligible.

We also compare the actual reform with acounterfactual reform consisting of halving fir-ing costs for both types of contracts. Althoughthe employment increases under both reformsare alike, the employment composition by con-tract is very different. As a consequence, suchcounterfactual reform had led to much largerimprovements in the productivity and the valueof firms. Compared with this counterfactualreform, the factual introduction of temporarycontracts leads to excess turnover and a largeproportion of employment of workers with lowfirm-specific experience.

The structural approach is justified on sev-eral grounds. First, the fact that the reformunder study was applicable to any type of firmand to any type of worker, makes implausi-ble a differences-in-differences approach basedon comparing the outcomes before and afterthe reform between agents affected differentlyby such reform. Second, given that some otherinstitutional changes took place in Spain after1984 (for instance, the Spanish entry in the

European Economic Community in 1986), areduced form approach does not ensure that weare controlling for the sort of structural changesthat we want to consider, that is, those whichaffected firing costs of temporary workers. Andthird, we are also interested in the evaluation ofcounterfactual policies.

There is a large literature on the structuralestimation of dynamic structural models of labordemand that goes back to the seminal paperby Sargent (1978).1 Our model builds on andextends papers on dynamic structural modelsof labor demand with nonconvex adjustmentcosts, such as Rota (2004), and Cooper andWillis (2004). The most relevant extensions arethe following: (a) we consider two types oflabor contracts, temporary and permanent; (b)our specification of labor adjustment costs isvery general and allows for fixed, linear, andquadratic, asymmetric adjustment costs whichcan be different for the two types of contracts;and (c) the specification of the unobservedvariables in the econometric model is quiteflexible and it includes unobservables in theproduction function, in the marginal costs, andin the fixed costs of the two types of labor.

As our dataset only provides informationon employment stocks, but not on employ-ment flows, our specification for labor adjust-ment costs is defined in terms of net employ-ment changes. The lack of information aboutgross employment changes prevents to disen-tangle among quantitatively similar net employ-ment changes which result from simultane-ous hires, layoffs, and voluntary quits. Sev-eral papers using datasets that directly measuregross employment flows, like Abowd, Corbel,and Kramarz (1999) and Goux, Maurin, andPauchet (2001), have shown how net employ-ment changes can result from large simulta-neous flows of hiring, firing, and voluntaryquits. In particular, distinguishing between lay-offs and voluntary quits can make a substan-tial difference in the case of permanent work-ers, with high severance payments. In an ear-lier version of this article, Aguirregabiria andAlonso-Borrego (1999) exploited complemen-tary information on severance payments to dis-tinguish between costly and noncostly reduc-tions in net employment, obtaining firing costestimates that are larger than the ones that ignore

1. See Pfann and Palm (1993), Hamermesh (1993), andthe survey paper by Bond and Van Reenen (2007) forreferences.

AGUIRREGABIRIA & ALONSO-BORREGO: LABOR MARKET REFORM 3

this distinction.2 The employment effects of thelabor market reform using those estimates offiring costs are stronger than the ones that wereport in this paper.

The database that we exploit in this paperover-samples large firms. However, we believethat our main results on the effects of thelabor market reform are even stronger for thewhole population of Spanish firms. In partic-ular, the estimation of the model shows thatthe effects of the reform on the level ofemployment, proportion of temporary labor, andjob turnover is significantly larger for smallerfirms.

The rest of the paper is organized as fol-lows. In Section II, we provide an overviewof the previous literature on the effects of jobsecurity provisions, and describe the institutionalfeatures of the Spanish labor market. SectionIII describes our dataset. In Section IV, weexplain the theoretical model as well as ourkey identification assumptions to evaluate theeffects of the policy change. The estimationresults of the structural model are provided inSection V. We present experiments that eval-uate the effects of the reform in Section VI.Section VII summarizes our main findings andconcludes.

II. THE ROLE OF JOB SECURITY PROVISIONS

A. Previous Evidence

There is a broad and growing literature onthe consequences of job security provisionson the labor market. A first line of researchuses longitudinal data of countries in orderto evaluate the effects of severance pay onseveral labor market outcomes exploiting thedifferences across countries. Using a panel of

2. When exploiting information on severance payments,we defined negative employment changes as layoffs ifseverance payments were strictly positive; otherwise, weimputed them as separations due to voluntary quits. Undersuch definitions, a large fraction of negative employmentchanges were imputed as voluntary quits, and therefore wemeasured a frequency and an amount of layoffs which weresubstantially smaller. Still, we were not able to identifysimultaneous hires, layoffs, and voluntary quits. However,we can reasonably argue that the frequency and amount ofsimultaneous hires and layoffs is lower than the frequencyof simultaneous layoffs and voluntary quits. Therefore, webelieve that firing cost estimates based on net employmentchanges that ignore voluntary quits, as well as simultaneoushires and layoffs, provide a lower bound to the firing costparameter that we would have obtained if we observed thedifferent gross employment flows.

Organization for Economic Co-operation andDevelopment (OECD) countries and construct-ing two alternative measures of severance pay,Lazear (1990) found that severance pay has neg-ative effects on employment and activity rates,and a positive effect on unemployment. Addi-son and Grosso (1996) corroborate the posi-tive influence of severance pay on unemploy-ment, but they find very little evidence “tosuggest that its contribution to rising unem-ployment is material.” Burgess, Knetter, andMichelacci (2000) evaluated the effects of jobsecurity provisions on the adjustment speed ofemployment and output, using longitudinal dataon the seven largest OECD countries disag-gregated by 2-digit industries, which allowsto control for differences in adjustment speedamong industries. Their results point out thatless regulated countries show a faster adjust-ment. Using a sample of OECD and LatinAmerican countries, Heckman and Pages (2004)found a strongly negative effect of job secu-rity provisions on employment rates, such effectvarying substantially among different types ofworkers.

A similar line of research has evaluated theconsequences of job security regulations, bycomparing a small number of countries. Abra-ham and Houseman (1993, 1994) compare theadjustment speed of employment and hours inmanufacturing industries in response to demandshocks in several European countries (Germany,France, and Belgium) and in the United States.Their main finding is that the higher costs ofadjusting employment levels in European coun-tries are compensated by the lower costs ofadjusting average hours, and therefore there areno substantial differences in the adjustment oftotal labor input. Bover, Garcıa-Perea, and Por-tugal (2000) try to explain why unemploymentrates in Spain and Portugal are so different eventhough their labor market institutions appearto share many similarities. The primary factorexplaining the much higher unemployment ratein Spain appears to be its lower level of wageflexibility, combined with a much more gener-ous system of unemployment insurance.

A second line of research has addressed theeffects of severance payments on employmentby means of calibration of theoretical mod-els. Bentolila and Bertola (1990) calibrate apartial equilibrium labor demand model usingaggregate data of several European countries,obtaining negligible effects. In a similar setting,Bertola (1992) finds that job security provisions

4 ECONOMIC INQUIRY

do not necessarily lower average employmentunless further restrictions on wage flexibility,such as minimum wage legislation, operate.In contrast, Hopenhayn and Rogerson (1993)calibrate a general equilibrium model using U.S.firm-level data and considering entry and exitof firms, obtaining that an introduction of fir-ing costs would reduce employment substan-tially. Cabrales and Hopenhayn (1997) calibratea similar model using firm-level evidence onjob matches before and after the 1984 reformin Spain which allowed the widespread use oftemporary contracts, finding that the reform hasinduced a large increase in the turnover rate buta moderate effect on employment. Blanchardand Landier (2002) and Cahuc and Postel-Vinay(2002) consider matching models to assess theeffect of partial labor market reforms like theone undertaken in Spain in 1984, finding thatalthough fixed-term contracts tend to foster jobcreation, the impact on employment might benegative. Their results also suggest that narrow-ing firing costs differences between permanentand fixed term would increase employment. Ina similar strand, Costain, Jimeno, and Thomas(2010) study the cyclical properties of the Span-ish dual labor market, looking both at unemploy-ment level and unemployment volatility. Inter-estingly, they find that to eliminate duality with-out raising the level of unemployment, firingcosts should be substantially cut. Bentolila etal. (2012) consider France and Spain, whichshare similar labor market institutions exceptfor a larger gap in firing costs between perma-nent and temporary workers, and laxer rules onthe use of temporary contracts in Spain thanin France. Their calibration results show thatthese differences account for the strikingly largerincrease in unemployment during the currentGreat Recession in Spain with respect to France.Guell and Rodrıguez-Mora (2010) analyze theeffects of introducing temporary contracts in thecontext of an efficiency wage model, concludingthat temporary contracts may increase unem-ployment if rigidities on wage setting, such asminimum wages, exist. Alvarez and Veracierto(2012) extend an Islands model with undirectedsearch and complete markets to deal with sever-ance taxes conditional on tenure. They interpretthis dependence as a form of temporary con-tracts. In a similar framework, Veracierto (2001)assesses the short-run consequences of introduc-ing labor market flexibility. Both papers findthat fixed-term contracts may increase unem-ployment.

A third line of research has exploited databefore and after specific reforms in the labormarket in order to evaluate how changes injob security provisions have affected labor mar-ket outcomes using a differences-in-differencesapproach. Kugler (2004) exploits the variationin employment protection between workers cov-ered and not covered by the legislation, to studythe effect of firing costs on unemployment haz-ard, using Colombian individual data before andafter a reform which reduced firing costs. In asimilar fashion, Kugler and Pica (2008) exploitthe differences in employment protection amongsmall and larger firms to study the effect onworker and job flows of firing costs. The resultsprovide evidence about the negative effect ofseverance payments on both employment leveland employment changes. Using establishment-level data for the United States and the vari-ation across states in dismissal costs, Autor,Kerr, and Kugler (2007) find that dismissalcosts reduce employment flows and total fac-tor productivity. Hunt (2000) exploited industry-level German data to conclude that the Ger-man reform in 1985 which facilitated the use oftemporary contracts did not affect employmentadjustment. In a similar line but with a differ-ent approach, Bentolila and Saint-Paul (1992)use firm-level data to evaluate the effect ofa Spanish reform which introduced temporarycontracts and find a rise in the speed of adjust-ment, although they do not use data before thereform.

Our approach combines the estimation of adynamic structural model of labor demand witha comparison of estimated structural param-eters before and after the policy change in1984. There is a large literature on the struc-tural estimation of dynamic structural mod-els of labor demand, which dates back to theseminal paper by Sargent (1978). As men-tioned in Section I, our model builds on andextends the recent literature on dynamic struc-tural models of labor demand with noncon-vex adjustment costs (Rota 2004; Cooper andWillis 2004). As far as we know, this is theonly dynamic structural model of labor demandthat has been used to evaluate a labor marketreform.

B. Labor Contract Regulations in Spain

According to the OECD, the Spanish labormarket is among the most regulated in Europe.Job security rules and, in particular, strong

AGUIRREGABIRIA & ALONSO-BORREGO: LABOR MARKET REFORM 5

mandatory severance payments, contribute im-portantly to the rigidity of such regulations.3 The1984 reform, which eliminated most of the pre-vious restrictions to the use of temporary con-tracts, has been one of the major legal changesof the Spanish labor market. To understand themotivation of this reform and the context inwhich it took place, we provide a description ofthe institutional background before the reform,and the subsequent changes occurred.

Until the late 1970s, the Spanish labor mar-ket was characterized by a hyper-regulated sys-tem of industrial relations under the monitor-ing of a single “union” to which both employ-ers and employees had to belong. The prohibi-tion of trade unions and the practical absenceof collective bargaining were “compensated”with regulations that guaranteed full employ-ment stability: in practice, most jobs were full-time jobs of indefinite duration. This institu-tional framework was transformed progressivelyafter Franco’s death in 1975. The first impor-tant change came in 1977 with the Royal Decreeof Industrial Relations. The official single unionwas dismantled and free trade unions were legal-ized. Although the decree also recognized newgrounds for fair dismissals based on economicreasons and simplified the legal procedures forcollective redundancies, job security rules werebasically unchanged.

In 1980, the Estatuto de los Trabajadores(Workers’ Statute, ET hereinafter) establishedthe conditions for a modern system of collectivebargaining comparable to the ones prevailing inother democratic European countries. However,it maintained many of the legal and adminis-trative restrictions on dismissals. For permanentworkers (those with an indefinite-term contract),mandatory severance payments were 20 days ofsalary per year of job tenure (up to a maximumof 1 year’s wages) if the dismissal is consid-ered “fair,” and 45 days (up to a maximum of42 months’ wages) if it is considered “unfair.”There are two reasons for fair dismissals: thoseattributable to the worker (incompetence or neg-ligence in his job), and other objective reasonsthat cannot be attributed to the worker (eco-nomic or technological reasons). However, the

3. Bentolila and Dolado (1994), using OECD data forselected European countries, find striking differences inregulations about authorization procedures for dismissalsand mandatory severance payments for fair and unfairdismissals, with Denmark and the UK having the less severe,and France, Greece, Portugal, and Spain being the countrieswith the most stringent regulations.

scope of the second reason was very limited.Furthermore, the burden of proof for a fair dis-missal must be assumed by the firm (see Bento-lila 1997). If the worker does not acknowledgethe dismissal—as it is usually the case—hemay appeal to the labor court. This obliges thefirm to bear a legal process, during which itshould pay interim wages while labor courtsreach their decisions (up to 2 months). Giventhat labor courts are in many cases favorableto the workers, the agreed severance paymentscan even exceed the statutory amounts for unfairdismissals. Besides, any dismissal (except fordisciplinary reasons) requires written advancenotice of 30 days. These job security rules forpermanent workers remained unchanged until1997.4

The ET allowed temporary or fixed-termcontracts, which could be cancelled at termina-tion with a much smaller severance payment andwithout court or regulatory intervention. How-ever, their use was mainly limited to jobs thatwere temporary in nature because of the sea-sonal nature of the production activity, the needto cover absent posts, or the start-up of a newfirm. These contracts had been used earlier toa certain extent in agriculture, construction, andservices industries; however, their use in manu-facturing had been very scarce.

Figure 1 shows the evolution of gross domes-tic product (GDP) growth and unemployment inSpain. Despite the fact that GDP was growingat the beginning of the 1980s, the unemploy-ment rate kept rising, and by the end of 1984,it was close to its peak (about 21%), and theSpanish economy was suffering the dismantlingprocess of obsolete plants in the heavy indus-tries. This fact, together with the complaintsof entrepreneurs about the rigid employmentlegislation, forced the government to broadenthe scope of temporary contracts in an attemptto boost employment. The ET was reformedin 1984, introducing the most important legalchange of the Spanish labor market in the pre-vious two decades, removing most of the restric-tions on noncausal fixed-term contracts. Themain feature of the reform was that the use

4. In 1997, trade unions and employer organizationssigned the Acuerdo Interconfederal para la Estabilidad delEmpleo (National Agreement on Employment Stability).This agreement led to a new permanent contract whichmaintained the severance payments for fair dismissals butlowered those for unfair dismissals to 33 days of salary (upto 42 months’ wages), yet its scope was limited to certaintypes of workers.

6 ECONOMIC INQUIRY

FIGURE 1Unemployment Rate and GDP Growth in Spain

0

5

10

15

20

25P

erce

ntag

e ra

te

1980 1985 1990 1995 2000 2005

Year

Unemployment rate Rate of change of GDP

Source: Spanish Labor Force Survey and National Accounts.

of temporary contracts is no longer linked tothe principle of causality, so that they could beapplied to any activity, temporary or not, andto any type of firm or worker. Furthermore,they might be signed for short periods (3, 6,or 12 months), firing costs at termination werelow (12 days of wages per year of tenure) oreven zero in some cases, and their extinctioncould not be appealed to labor courts. Never-theless, the reform established that temporarycontracts could only be renewed up to 3 years,when the firm should decide whether to offerthe worker a permanent contract or to dismisshim.5 Importantly, the reform did not alter thestringent dismissal regulations for permanent orindefinite-duration contracts.

After this reform, the number and the pro-portion of temporary jobs in the Spanish econ-omy soared. Spain became by far the Europeancountry with the highest percentage of tempo-rary employment, which amounted to 80% ofhires in the period 1986–1990 (Bentolila andDolado 1994). This important increase in tem-porary employment points out that firms havefound these contracts attractive to reduce fir-ing costs. Nevertheless, this behavior is consis-tent with either positive or negative employmenteffects of the reform. Evaluating the effects ofthe reform on employment and output requires

5. Furthermore, if a firm lays off a temporary worker, itmust wait for a year in order to hire him again.

to analyze how individual firms’ hiring and fir-ing decisions have changed after the reform.

III. DATA AND PRELIMINARY EVIDENCE

The main dataset consists of an unbalancedpanel of 2, 356 nonenergy manufacturing firmscollected from the database of Central de Bal-ances del Banco de Espana (Balance Sheets ofthe Bank of Spain, CBBE hereinafter) between1982 and 1993. The firms included in the rawdatabase represent almost 40% of the total valueadded in Spanish manufacturing. This datasetprovides firm-level annual information on thebalance sheets and other complementary infor-mation on economic variables, such as employ-ment by type of contract, output, physical cap-ital, and the total wage bill. Due to the factthat response is completely voluntary, the largestfirms are overrepresented in the sample. There-fore, care must be taken in extrapolating to thewhole population of Spanish firms the resultsbased on our sample. The criteria for selectionof the sample and construction of the variablesused in the empirical analysis (market value ofthe capital stocks, wages, etc.) are described inAppendix A.

Wage information by type of contract is avail-able from the Spanish wage distribution survey(Distribucion Salarial ). However, that surveyis only available for the years 1988 and 1992,

AGUIRREGABIRIA & ALONSO-BORREGO: LABOR MARKET REFORM 7

and it provides only aggregate information. Wedescribe in Section IV.A our approach to obtainestimates of wages of permanent and temporaryworkers for the whole period 1982–1993.

Figure 2 presents the evolution of the propor-tion of temporary workers in the Spanish econ-omy and in our sample.6 The share of tempo-rary contracts, which was estimated to be about10% of total employment and 3% of manufac-turing employment in 1984, rose to 35% and31%, respectively, in 1995, remaining at highlevels since then. It is not surprising that theshare of temporary employment increased grad-ually since the reform. Due to the substantialfiring costs for permanent workers, firms couldwait until some permanent workers retire or quitvoluntarily to circumvent severance payments.Figure 2 shows also a large disparity betweenthe proportion of temporary workers from theLabor Force Survey and the proportion from theCBBE. The main factor to explain this discrep-ancy is that CBBE overrepresents large manu-facturing firms, and this type of firm tends tohave a smaller proportion of temporary workers(see Figure 4 below).

In Figure 3, we report the evolution of thegrowth rates of real output and employment forour sample. The sample period covers an expan-sion, 1986–1989, and a recession, 1990–1993.However, both the number and the proportionof permanent employees exhibit a monotonousdecrease. After the introduction of the reform,in November 1984, temporary employment rosesignificantly from 1986 to 1990 and decreasedfrom 1990 to 1993, and its share in total employ-ment rose from 2.89% in 1985 to 9.72% in1993. The evolution of temporary employmentin our sample keeps coherency with the aggre-gate series for the overall economy, and partic-ularly with the aggregate series for manufactur-ing (in fact, the correlation coefficient betweenboth series is above 90%). However, the figurefor our sample is substantially below the aggre-gate figures, which were well above 20% at thebeginning of the 1990s. This discrepancy is dueto the fact that larger companies, which are over-represented in our sample, are more promptedto use permanent employment than small ormedium ones. In Figure 4, we can see that in oursample the proportion of temporary employmentdecreases with size.

6. Unfortunately, the Spanish Labor Force Survey didnot report any information about the type of contract before1987.

Figure 5 presents the job creation and jobdestruction rates for permanent and tempo-rary employment using the statistics proposedby Davis and Haltiwanger (1992).7 The smalljob turnover rates for permanent employmentcontrasts with the very high rates for tempo-rary employment. Furthermore, the creation anddestruction rates for temporary employment aremuch more correlated with the cycle than thosefor permanent employment. This is evidence ofhow firing costs can have very important effectson job turnover rates. It also reflects the fact thatalthough the reform introduced larger flexibil-ity for new hires, it kept the core of permanentemployees unaffected.

Figure 6 presents the evolution in the pro-portion of our firms with positive, negative,and zero annual change in permanent employ-ment. We observe a remarkable frequency ofno adjustments in permanent employment (about19%), which is fairly stable over the cycle, sug-gesting an important persistence in permanentemployment. This evidence is consistent withthe existence of lump-sum or kinked adjustmentcosts.

Table 1 presents descriptive statistics onemployment and productivity for a pre-reformperiod (1982–1984) and a post-reform period(1989–1992). For the sake of comparability, weconsider a common subsample of 389 firms inthe two periods. The post-reform period hasbeen also selected for comparability reasons,such that the pattern of output growth is verysimilar to the pre-reform period. From this com-parison, we can establish some interesting facts.The median employment growth is similar inthe two periods, but it is significantly more dis-persed after the reform. The shrink in perma-nent employment is compensated by temporaryemployment, so that total employment exhibitsan increase of 5.4% for the median firm interms of number of workers, 8.3% for smallerfirms, and 0.3% for larger firms. We can seethat firm productivity, as measured by the salesto wage bill ratio, goes down after the reform.Interestingly, in Section VI, our counterfactualexperiments based on the estimated model showsimilar magnitudes for the steady-state effects ofthe labor market reform on total employment,

7. Our measures are based on firm-level data insteadof plant-level data as in Davis and Haltiwanger (1992) forthe United States. This can be a factor, in addition to thedifferent labor market institutions in Spain and the UnitedStates, that contributes to the smaller job turnover rates thatwe find in our data.

8 ECONOMIC INQUIRY

FIGURE 2Share of Temporary Employment in Total Employment

0

10

20

30

40P

erce

ntag

e ra

te

1983 1988 1993 1998 2003

Year

All industries Manufacturing CBBE sample

Note: Spanish Labor Force Survey did not report information about the type of contract before 1987.Source: Spanish Labor Force Survey and CBBE sample of Spanish manufacturing firms.

proportion of temporary workers, and produc-tivity.

IV. MODEL

A. Basic Assumptions

Consider an economy with two types of laborcontracts: fixed-term and indefinite-durationcontracts. We denote employees as temporaryor permanent depending on whether they enjoya fixed-term or an indefinite-duration contract,respectively. In principle, the only exogenousfeature that distinguishes a permanent and atemporary contract lies in the dismissal costs.Firms are enforced by law to pay a severanceto each dismissed permanent worker, but tem-porary workers are not entitled to any com-pensation upon dismissal. Although dismissalcosts appear as the only exogenous differencebetween these two contract types, they can gen-erate, endogenously, further differences betweenworkers. Particularly, two major differences areexpected to appear. On the one hand, incen-tives to invest in firm-specific human capitalare stronger for workers with indefinite-termcontracts than for those with fixed-term con-tracts. This fact might create a productivity gapbetween permanent and temporary workers. Onthe other hand, the higher costs of dismissalswill place permanent workers in a better bar-gaining position within the firm. This fact might

induce a wage gap between permanent and tem-porary workers. We incorporate these differen-tial features in our model, yet we take them asexogenous for the sake of simplicity.

Firms produce a homogeneous good usinglabor as the only variable input, and sell theiroutput in a competitive market.8 Every periodt , the firm chooses the amounts of permanentand temporary labor that maximize its expectedintertemporal profit, Et

(∑∞j=0 βj�t+j

), where

Et is the conditional expectation given the infor-mation up to period t , �t denotes profits atperiod t , and β ∈ (0, 1) is the discount fac-tor. Current profits, measured in output units,are:

�t = Yt − WBt − ACt + ξt ,(1)

where Yt is real output, WBt is the wage bill,ACt represents labor adjustment costs, and theterm ξt contains other components of currentprofit which are observable to the firm but unob-servable to the econometrician. Physical capitalis treated as a component of the firm idiosyn-cratic shock and it is assumed to follow anexogenous process.

8. Alternatively, we may consider that firms competein monopolistic product markets with isoelastic demandcurves. In that setting, our production function should bereinterpreted as a revenue function, and its parameters as acombination of technological parameters and the elasticityof demand.

AGUIRREGABIRIA & ALONSO-BORREGO: LABOR MARKET REFORM 9

FIGURE 3Rates of Growth of Output and Employment

-10

0

10

20

30

Per

cent

age

rate

1983 1985 1987 1989 1991 1993

Year

Real output Total emp.Permanent emp. Temporary emp.

Source: CBBE sample of Spanish manufacturing firms.

The production technology is described bythe production function

Yt = (LPt + λ LT

t )αL exp(ηt ),(2)

where LPt and LT

t represent the correspond-ing amounts of firm’s permanent and temporaryworkers; αL ∈ (0, 1) and λ ∈ (0, 1) are param-eters; and ηt is an exogenous and idiosyncraticproductivity shock, assumed to follow a first-order Markov process with transition probabilityfunction fη(ηt+1|ηt ).

As mentioned above, dismissal cost is theonly exogenous difference between temporaryand permanent workers. Therefore, if we hadthe same type of workers under each con-tract, they would be perfect substitutes withone-to-one rate of substitution. However, as wehave mentioned earlier, differences in dismissalcosts can endogenously generate differences inthe relative productivity of permanent and tem-porary workers. We account for such differencesthrough the parameter λ, which measures theproductivity of temporary workers with respectto permanent workers. We could have otherwiseconsidered a constant elasticity of substitutiontechnology of production in permanent and tem-porary workers to allow certain complementaritybetween worker types. However, in this arti-cle we do not see permanent and temporaryworkers as different labor inputs. Besides, the

existence of labor adjustment costs and, impor-tantly, uncertainty about future productivity andwages, implies that, despite the perfect substitu-tion between worker types, the optimal propor-tion of temporary employment is not either zeroor one in most of the cases.

The wage bill is WBt = WTt LT

t + W Pt LP

t ,where WT

t and W Pt are the wages of tempo-

rary and permanent workers, respectively. Thewage of temporary workers is determined atthe market level, and it is the same for allfirms operating in the market. However, thewage of permanent workers is firm-specific (e.g.,internal labor market, rent-sharing). The pairof wages Wt = (WT

t ,W Pt ) follows a first-order

Markov process with transition probability func-tion fW(Wt+1|Wt).

Our sample information on the firm’s totalwage bill is not broken down by type of con-tract. The assumption that the wage of tem-porary workers is the same across firms is arestriction that we use to identify the evolutionover time in the average wage ratio betweentemporary and permanent workers. We describethis identification approach below. Temporaryworkers do not have bargaining power or firm-specific experience. Empirical evidence fromother data sources with information on wages atthe worker level (Wage Structure Survey 1995)shows that, though the wages of temporary

10 ECONOMIC INQUIRY

FIGURE 4Share of Temporary Employment in Total Employment by Firm Average Size

0

5

10

15

1983 1985 1987 1989 1991 1993

Year

Up to 25 workers 25 to 75 workers75 to 200 workers More than 200 workers

Source: CBBE sample of Spanish manufacturing firms.

workers varies across firms, they have a muchweaker correlation with measures of firm sizethan the wages of permanent workers. An impli-cation of our assumption is that we interpretfirms with high wage-bill-to-permanent-workersratio as firms with costly permanent workers,such that we expect that, ceteris paribus, thesefirms are more willing to substitute permanentworkers with temporary workers.

The wage bill of firm i at year t is WBit =WT

it LTit + W P

itLPit . Given the assumption that the

wage of temporary workers is the same for everyfirm, we have that:

WBit/LPit = WT

t

(LT

it /LPit

) + W Pit .(3)

We observe WBit/LPit and LT

it /LPit but we do

not have data on WTt and W P

it . If the wageof permanent workers were mean independentof the temporary-to-permanent ratio, LT

it /LPit ,

we could estimate the value WTt by running

an OLS regression for WBit/LPit on LT

it /LPit at

year t . Moreover, the residual of that regres-sion would be a consistent estimator of the wageof permanent workers at the firm level. How-ever, such estimate of WT

t will be affected byan upward endogeneity bias if, as we expect,the temporary-to-permanent ratio is positivelycorrelated with the wage of permanent work-ers. To control for this bias, we consider afixed-effect or within-firms estimator. That is,

we assume that the wage of permanent work-ers is:

W Pit = μi + γt + uit(4)

where μi is a firm fixed-effect; γt is an aggregateeffect; and uit is a shock assumed to be uncor-related with the temporary-to-permanent ratio.Under this assumption, the fixed-effects estima-tor provides consistent estimates of WT

t .The specification of labor adjustment costs,

defined in terms of net employment changes, foreach type of employment, �L

jt ≡ L

jt − L

j

t−1(j = T,P), includes both lump-sum and linearcomponents:

(5)

ACt = 1(�LTt > 0)

×[θTH0 +θT

H1�LTt +θT

H2

(�LT

t

)2]

+1(�LTt < 0)

[θTF0 − θT

F1�LTt +θT

F2

(�LT

t

)2]

+1(�LPt > 0)

[θPH0 + θP

H1�LPt +θP

H2

(�LP

t

)2]

+1(�LPt < 0)

[θPF0 − θP

F1�LPt + θP

F2

(�LP

t

)2]

where 1(.) is the binary indicator function; {θj

H0,

θj

H1, θj

H2, θj

F0, θj

F1, θj

F2 : j = T,P } are (nonneg-ative) parameters. The first two summands refer

AGUIRREGABIRIA & ALONSO-BORREGO: LABOR MARKET REFORM 11

FIGURE 5Rates of Job Creation and Job Destruction by Contract Type

-40

-20

0

20

40

1983 1985 1987 1989 1991 1993 1983 1985 1987 1989 1991 1993

Permanent Temporary

Real output Job creation Job destruction

Per

cent

age

Year

Note: The rates for job destruction appear with negative sign.Source: CBBE sample of Spanish manufacturing firms.

to hiring and firing costs of temporary workers.The third and fourth terms are the correspondingadjustment costs for hiring and firing permanentworkers.

There are different sources associated withadjustment of labor inputs, the most obviousbeing those associated to technological disrup-tion and, very especially, the legal costs imposedby the regulations on hiring and firing. But thereare further sources that can magnify or mitigatethe total costs of adjustment. In particular, in thecase of firing, the firm can bear additional costsif the workers go to the labor court adducingunfair dismissal: litigation costs, and the costsassociated with the firms obligation to pay thewage of the fired worker until the case is set-tled by the court (see Galdon-Sanchez and Guell2000). On the contrary, firms and dismissedworkers may eventually settle an agreement toeither avoid litigation by which dismissal com-pensation may be below the legal redundancypayment for unfair dismissal, or pay the workeroutplacement services, or negotiate payment forearly retirement. In fact, we are not imposingthat adjustment costs are necessarily above orbelow the administrative ones, using a revealedpreference approach.

For the sake of simplicity, and based on theempirical evidence, the model does not accountexplicitly for the firm’s decision on conversionof temporary contracts into permanent contracts.Dolado, Garcıa-Serrano, and Jimeno (2002) and

Guell and Petrongolo (2007) indicate that theconversion rate of temporary into permanentcontracts in Spain is low, reflecting the factthat “employers use those contracts more asa flexible device to adjust employment in theface of adverse shocks than as a screeningdevice.”

The firm chooses employment changes so asto maximize its expected intertemporal profit.We consider a discrete choice model such thatthe set of possible values of (�LP

t , �LTt ) is

discrete and finite. The main reason why weconsider a discrete model is that there is muchlumpiness in these employment decisions. Inour data, the frequency of zeroes in annualemployment changes is 18.8% for permanentemployment and 49.1% for temporary employ-ment. Furthermore, the frequency of employ-ment changes within −5 and +5 workers is65.5% for permanent labor and 80.4% for tem-porary labor. Let D be the finite set of possiblediscrete values for (�LP

t , �LTt ).

The component of current profits that isunobservable from the point of view of theeconometrician is specified as follows:

ξt ≡ ξ(�LPt , �LT

t , εt ) = σP εPt �LP

t(6)

+ σT εTt �LT

t + σ0 ε0t (�LP

t , �LTt ).

σP, σT, and σ0 are parameters, and we use σε torepresent the vector (σP, σT, σ0)

′. εPt and εT

t aremutually independent standard normal random

12 ECONOMIC INQUIRY

FIGURE 6Net Changes in Permanent Employment

0

10

20

30

40

50

Pro

port

ion

of fi

rms

(%)

1983 1985 1987 1989 1991 1993

Year

Real output growth Positive changesZero changes Negative changes

Source: CBBE sample of Spanish manufacturing firms.

TABLE 1Descriptive Statistics. Balanced Panel 1982–1992 (389 Firms)

Period 1982–1984 Period 1989–1992

Variable 1st Quartile Median 3rd Quartile 1st Quartile Median 3rd Quartile

Growth real output −6.2% 2.4% 10.5% −7.1% 2.3% 11.5%Growth total employment −3.6% −0.6% 2.6% −5.6% −0.6% 4.4%Number of workers 60 131 297 65 137 298Permanent workers 55 128 276 56 121 272Temporary workers 0 0 3 0 6 22% Temp workers 0.0% 0.0% 2.2% 0.0% 4.9% 13.6%Ratio (sales / wage bill) 4.2 5.7 8.5 4.3 5.6 7.8Number of observations 1, 167 1, 556

variables which are independently distributedover time and across firms. These variables tryto capture unobserved heterogeneity in hiringand firing costs across firms and over time.For instance, the entitled dismissal cost of themarginal worker may vary across firms becauseof differences in workers’ years of tenure. Thevariables ε0

t try to capture further sources ofunobserved heterogeneity. For every possiblepair of discrete values (j, k) ∈ D, the variableε0t (j, k) is a logit error which is independently

and identically distributed over time and acrossfirms with type I extreme value distribution. Weuse εt to denote the vector of unobservables{εP

t , εTt , ε0

t (k, j) : (k, j) ∈ D}. This combinationof normal errors and logit errors resembles the

specification in the mixed (or random coeffi-cients) multinomial logit model in McFaddenand Train (2000). The existence of the termsεPt �LP

t and εTt �LT

t imply that unobservables arecorrelated across choice alternatives. Hence, themodel does not hold the property of indepen-dence of irrelevant alternatives, which is a well-known limitation of the standard multinomiallogit model. Nevertheless, the model retainssome features of the multinomial logit, whichfacilitates estimation and ensures good proper-ties of the maximum likelihood estimator.9

9. When σP = σT = 0 and σ0 > 0, our specificationbecomes the one in the standard conditional logit model.When σ0 = 0, σP > 0, and σT > 0, we have a bivariateordered probit.

AGUIRREGABIRIA & ALONSO-BORREGO: LABOR MARKET REFORM 13

FIGURE 7Share of Fixed-Term Contracts in Total Hiring

in Spain

Source: Bover et al. (2002), Figure 1.

Every period, the firm has perfect knowledgeabout its stocks of permanent and temporarylabor, wages, and the realized values of pro-ductivity and cost shocks, but it has uncertaintyabout the future values of these shocks. Letxt ≡ (LP

t−1, LTt−1,Wt ,ηt ) be the vector of state

variables at period t , excluding εt. And let dt bea vector of two categorical variables represent-ing the corresponding decisions (�LP

t , �LTt ).

The profit function can be written as:

�t = z (dt , xt ) θ + ξ(dt , εt )(7)

where θ is the 13 × 1 vector (1, θTH0, θ

TH1, θ

TH2,

θTF0, θ

TF1, θ

TF2, θ

PH0, θ

PH1, θ

PH2, θ

PF0, θ

PF1, θ

PF2), and

z (dt , xt ) is a 1 × 13 vector with correspondingelements

⎛⎜⎜⎜⎜⎜⎜⎜⎝

Yt − WBt , 1(�LTt >0), 1(�LT

t >0)�LTt , 1(�LT

t >0)(�LT

t

)2,

1(�LTt < 0), −1(�LT

t < 0)�LTt , 1(�LT

t < 0)(�LT

t

)2,

1(�LPt > 0) , 1(�LP

t > 0)�LPt , 1(�LP

t > 0)(�LP

t

)2,

1(�LPt < 0) , − 1(�LP

t < 0)�LPt , 1(�LP

t < 0)(�LP

t

)2

⎞⎟⎟⎟⎟⎟⎟⎟⎠

(8)

We can represent the firm’s decision problemusing the Bellman equation:

V (xt , εt ) = maxdt∈D

[z (dt , xt ) θ + ξ(dt , εt )(9)

+ β∑xt+1

∫V (xt+1, εt+1) fε(dεt+1)

× fx(xt+1|xt , dt )

]

where fε is the density function of εt, and fx isthe transition of the vector of state variables xt ,that has the following structure:

fx(xt+1|xt , dt ) = 1{LP

t = LPt−1 + �LP

t

}(10)

× 1{LT

t = LTt−1 + �LT

t

}× fη(ηt+1|ηt ) fW (Wt+1|Wt)

The optimal solution of this dynamic labordemand model can be represented, followingAguirregabiria and Mira (2002), as the uniquefixed point of a mapping in the space ofconditional choice probabilities Pr(dt |xt ) (seeAppendix B).

B. Assumptions for Policy Evaluation

Our main interest is to evaluate the effectsof the 1984 labor market reform on employ-ment and productivity in the Spanish manufac-turing industry, using our longitudinal sample ofSpanish firms. The reform extended the use oftemporary contracts to any activity, temporary ornot, and reduced firing costs for these contractsfrom 45 days to 12 days of wages per year oftenure. The new regulation applied to every typeof firms and workers, regardless of size, indus-try, occupation, etc. Our approach for evaluatingthe effects of this reform is based on the estima-tion of our structural model and on the construc-tion of the counterfactual steady-state distribu-tions of employment with and without the labormarket reform. Some identification assumptionsare necessary to establish that we can use thepre-reform and the post-reform sample periodsto construct consistent estimates of the steady-state distributions of employment with the oldand the new policies. This subsection discussesthese assumptions.a) Nonanticipation of the reform. If agents wouldhave anticipated the policy change, their behav-ior before the reform would not represent theiroptimal decisions if the reform would not havetaken place. For instance, some firms willing tohire in 1983 or 1984 had preferred to postponehiring and firing decisions until 1985 in order touse the new type of labor contract. Departuresfrom this assumption might bias our estimates oflabor adjustment costs for the pre-reform period.Since our sample covers only 3 years before thepolicy change, there is not very much we cando to control for this potential source of bias.b) Instantaneous learning about the features ofthe new policy. Looking at our data, there is clearevidence that a long transition period to the new

14 ECONOMIC INQUIRY

FIGURE 8Time Series of the Estimated Average Wages by Contract

1

1.5

2

2.5

3W

ages

in m

illio

n pe

seta

s 19

90

1983 1985 1987 1989 1991 1993

Year

Permanent Temporary

Note: Dotted lines denote the 95% confidence intervals.Source: Own calculations from CBBE sample of Spanish manufacturing firms.

steady state took place after the reform. In par-ticular, the proportion of temporary employmentincreased almost every year between 1984 and1993, and was kept at high levels since then(see Figure 2). Such transition period could beexplained by the existence of large firing costsfor permanent workers. However, another rea-son that might have contributed to this long tran-sition, and which is not considered in our model,is that the firm learning process about the fea-tures of the new policy rule were slow. Since ourmodel assumes instantaneous learning, it rulesout this alternative explanation. The fact that thenumber of new temporary contracts exerted alarge increase shortly after the reform, in 1985,seems to support our assumption. This is shownin Figure 7, taken from Bover, Arellano, andBentolila (2002), that shows how the share offixed-term contracts in total hiring jumped fromlevels below 20%, just before the reform, tomore than 90% just after the reform, and itremained at that level at least until 1994.

It is important to emphasize that our assump-tion of instantaneous learning does not implythat the new steady state was reached instanta-neously after the reform. While firms’ optimaldecision rules on hiring and firing may havejumped to their new post-reform values soonafter the implementation of the new policy, itmay take several years for the distributions ofemployment by type of contracts to reach theirnew steady state. Actually, both Figures 2 and7 are consistent with this assumption.

c) Policy-related and policy-invariant parame-ters. The reform entailed a change in the dis-missal costs of temporary workers, θT

F0, θTF1,

θTF2. Hiring costs of temporary workers, θT

H0 ,θTH1, θT

H2, their relative productivity, λ, and hir-ing costs of permanent workers may have beenaffected by the reform as well. Therefore, θT

F0 ,θTF1, θT

F2, θTH0, θT

H1, θTH2, θP

H0, θPH1, θ

PH2 and λ

are policy related parameters. It seems plausiblethat the technological parameter αL, firing costsof permanent workers, θP

F0, θPF1, θP

F2, and thestochastic process of the productivity shock arepolicy invariant parameters.

It must be recalled that temporary contractswere linked to the principle of causality beforethe reform, so that they must be aimed at jobsthat were temporary in nature. However, thiswas no longer the case after the reform, whichallowed temporary contracts for any kind of job.For this reason, we must expect that the type ofworkers to which temporary contracts are aimedwere different before and after the reform. Ourempirical model acknowledges this fact lettingwages, relative productivity and firing costs todiffer before and after the reform. Therefore,neither our results nor our counterfactual analy-sis are limited by this fact.

In an equilibrium framework, the stochasticprocess of wages may be affected by this reform.Our estimation of the econometric model takesthis into account. However, our model is ofpartial equilibrium and our policy evaluation

AGUIRREGABIRIA & ALONSO-BORREGO: LABOR MARKET REFORM 15

assessments provide partial equilibrium effects.We provide more discussion on this point whenwe present our counterfactual experiments inSection VI.

V. ESTIMATION OF THE MODEL

We have a panel dataset with firm-level,annual-frequency information on output, em-ployment by type of contract, physical capital,investment, and wage bill: {Yit , L

Tit , L

Pit , Kit , Iit ,

WBit : i = 1, ..., N; t = 1, ..., Ti}. Our econo-metric model consists of the production func-tion, the stochastic processes for the productiv-ity shock and wages, and the dynamic modelfor the demand of permanent and temporarylabor. The vector of structural parameters is{αL,λ, θ, σε,π,β}, where θ is (as defined in pre-vious section) the vector of adjustment costsparameters, and π is the vector of parameters inthe transition probabilities of the state variables.We estimate these parameters in two steps. In afirst step, we estimate {αL,λ,π} from the pro-duction function and transition data. In a sec-ond step, we estimate θ and σε by maximumlikelihood in the dynamic labor demand model.The maximum likelihood estimator (MLE) inthe second step is a partial MLE, because theparameters in the production function have beenestimated separately. Therefore, our estimationapproach does not provide fully efficient esti-mates. However, as we show in Section V.C,our estimates of the parameters in adjustmentcosts are quite precise. Before describing ourestimation methods and results, we undertakethe estimation of wages by type of contract.

A. Estimation of Wages

The combination of Equations (3) and (4)implies the following regression equation in firstdifferences:(

(WBit/LPit ) − (WBit−1/L

Pit−1)

)(11)

= γt + WTt

(LT

it /LPit

)+WT

t−1

(LT

it−1/LPit−1

) + uit

where γt ≡ γt − γt−1, and uit ≡ uit − uit−1.We estimate the “parameters” {γt ,W

Tt : t =

1983, . . . , 1993} by using OLS in Equation (11).Given the estimate WT

t , we can get an esti-mate of the wage of permanent workers in firmi as W P

it = WBit/LPit − WT

t

(LT

it /LPit

). Figure 8

presents the time series of our estimates forthe average wages of permanent and temporary

workers. According to our estimates, the wagedifferential between contracts was small beforethe reform, but widened very importantly after1984. This result is consistent with the evidenceprovided by Bentolila and Dolado (1994), bywhich if unions set wages for all workers and aredominated by permanent workers, then the exis-tence of temporary contracts increases the jobsecurity and the bargaining power of permanentworkers. Using longitudinal data, they find thatan increase of 1 percentage point in the shareof temporary contracts raises the growth rate ofthe wages of permanent workers by about 0.3%.Besides, Dolado, Garcıa-Serrano, and Jimeno(2002) argue that, given the weaker bargainingposition of temporary workers and the higherwage pressure on permanent contracts, employ-ers tend to under-classify temporary workerswhen assigning them to lower occupational cate-gories, which determine their wage level. Also,unlike permanent workers, experience is typi-cally not accounted in the wages of temporaryworkers.

B. Estimation of the Production Function

The specification of the production functionin Equation (2) treats physical capital as a com-ponent of the productivity shock ηit . This isa convenient assumption to reduce the dimen-sionality of decision and state spaces. Althoughwe maintain this assumption throughout the arti-cle, in the estimation of the production function,we incorporate explicitly physical capital andestimate the technological parameter associatedwith this input. Looking at this estimate is a wayof checking for the validity of the specificationand for the economic sense of the estimationresults. We consider a Cobb–Douglas produc-tion function in terms of physical capital andproduction-equivalent units of labor. The pro-duction function in logarithms is:

ln Yit = αK ln Kit + αL ln(LPit + λ LT

it ) + ωit

(12)

where Kit is the installed capital at the begin-ning of year t ; ωit is the “pure” productivityshock such that ηit = αK ln Kit + ωit .

It is well known that the OLS estimation ofthis equation may suffer of endogeneity biasbecause of the correlation between the inputsand the unobservable productivity shock (seeGriliches and Mairesse 1998). Furthermore, ifthe productivity shock is serially correlated,lagged values of inputs and output are also

16 ECONOMIC INQUIRY

correlated with the unobservables, and there-fore they cannot be used as instruments. Usinginput prices (e.g., wages) as instruments is alsoproblematic. Some input prices do not havevariability at the firm level (e.g., the wage oftemporary workers, or the price of capital),and those prices that do have that variabilityare very suspicious of being correlated withfirm’s productivity (e.g., the wage of permanentworkers).

Our identification of the parameters in theproduction function is based on the control func-tion approach proposed by Olley and Pakes(1996). Our application of this method is inthe spirit of the extension proposed by Acker-berg, Caves, and Frazer (2006). Let Iit =g(Kit , L

Pi,t−1, L

Ti,t−1, Ct ,ωit ,χit ) be the opti-

mal decision rule for investment in physicalcapital, Iit , where Ct represents input prices,and χit represents other possible shocks affect-ing investment which are unobserved to theresearcher and not serially correlated and inde-pendent of the other state variables in the model.Provided that function g is strictly increasing inthe productivity shock ωit , there is an inversefunction such that ωit = g−1(Iit , Kit , L

Pi,t−1,

LTi,t−1, Ct ,χit ). Based on this expression, we

can decompose ωit in two additive terms: ωit =ωe

it + χ∗it , where ωe

it ≡ E(ωit |Iit , Kit , L

Pi,t−1,

LTi,t−1, Ct

)and χ∗

it is the remaining part of ωit .This decomposition has two important features.First, ωe

it only depends on observable variables.And second, χ∗

it is, by construction, mean inde-pendent of

(Iit , Kit , L

Pi,t−1, L

Ti,t−1, Ct

), and also

of LPit and LT

it . Therefore, we can write the pro-duction function as,

ln Yit = αL ln(LPit + λ LT

it ) + ηeit + χ∗

it(13)

where ηeit = αK ln Kit + ωe

it . Note that ηeit is a

smooth function of (Iit , Kit , LPi,t−1, L

Ti,t−1, Ct ).

We can control for this term by includinga high order polynomial in these observablevariables. The key identification assumption isthat there are i.i.d. shocks εit and χit affect-ing current employment and investment, respec-tively, which are mutually independent. Underthis assumption, we can use current invest-ment to control for the endogenous part ofthe productivity shock ωit , and still we havesome variability left in the current employ-ment variables LP

it and LTit , to identify αL and

λ.Once we have estimated αL and λ, we can

exploit the assumption on the Markov process

of ωit to estimate αK. First, we obtain estimatesof ηit as the residuals ln Yit − αL ln(LP

it + λLTit ).

According to the model, ηit = αK ln Kit + ωit .Assuming that ωit follows an AR(1) process:ωit = ρω ωi,t−1 + ait with ait ∼ iid (0, σ2

a), wehave that(

ηit − ρω ηi,t−1)

(14)

= αK(ln Kit − ρω ln Ki,t−1

) + ait .

Since the innovation ait is independent of ηi,t−1,ln Kit and ln Ki,t−1, we can estimate αK and ρω

using nonlinear least squares.In Table 2, we present our estimates of the

production function parameters. For the sakeof comparison, we report estimates using boththe Olley–Pakes method and the (inconsistent)nonlinear least squares estimator. All the esti-mations include time dummies and 20 indus-try dummies. The control function ηe

it includesall the terms of a second order polynomial in(Iit , Kit , L

Pi,t−1, L

Ti,t−1) and interactions of these

terms with time dummies, what entails a totalof 164 regressors. The parameters λ, ρω, and σa

are allowed to change between the pre-reformand the post-reform period. However, whereasa change in λ might be attributed to the reform,changes in ρω or in σa might not. Comparingthe two reported estimates, both the magnitudesand the qualitative results are fairly similar, themajor differences concerning the λ parameterbefore the reform. The estimates of the param-eters ρω and σa before and after the reformsuggest small reductions in the persistence ofthe productivity shock and in the variability ofthe innovation.

The point estimates imply some decreasingreturns to scale, though the hypothesis of con-stant returns to scale cannot be rejected undertypical significance levels. The most interestingresult in this table is the post-reform increase inthe relative efficiency of temporary labor. Whilethis input was just half as efficient as permanentlabor before 1984, it has become almost as effi-cient after the reform. A possible explanationfor this result is that adverse selection was amore serious problem for temporary labor in thepre-reform period. However, we should be cau-tious to attribute this parameter change entirelyto the reform. For instance, young workers inSpain during this period were significantly moreeducated than older cohorts, and they have alsoaccounted for a large proportion of temporarycontracts.

AGUIRREGABIRIA & ALONSO-BORREGO: LABOR MARKET REFORM 17

TABLE 2Estimation of Production Function Parameters. Unbalanced Panel 1982–1993 (2,356 Firms)a

Least Squares Olley–Pakes

Parameters Estimate (SE)b Estimate (SE)(2)

αK 0.260 (0.006) 0.294 (0.028)

αL 0.690 (0.008) 0.680 (0.036)

Pre-Reform λ 0.666 (0.093) 0.549 (0.150)

Post-Reform λ 0.895 (0.035) 0.913 (0.054)

Pre-Reform ρω 0.955 (0.010) 0.957 (0.011)

Post-Reform ρω 0.931 (0.003) 0.943 (0.003)

Pre-Reform σa 0.174 (0.0025) 0.172 (0.0025)

Post-Reform σa 0.207 (0.0030) 0.204 (0.0030)

# Observationsc 16,640 15,985

aAll the estimations include time dummies and 20 industry dummies.bStandard errors are robust of heteroscedasticity and autocorrelation.cIn Olley–Pakes estimation we can use only those observations with investment different than zero. This explains the

smaller number of observations.

The fact that the efficiency of temporaryworkers has increased after the reform is consis-tent with the increase in the wage gap betweenpermanent and temporary workers. We are argu-ing that the positive effect of productivity onthe relative wage of temporary workers is out-weighed by a weakening in the bargaining posi-tion of temporary workers. See at the end ofSection V.A above our discussion on wage bar-gaining and the empirical evidence in Bentolilaand Dolado (1994) and Dolado, Garcıa-Serrano,and Jimeno (2002).

C. Estimation of the Dynamic Labor DemandModel

We estimate the dynamic labor demandmodel using the nested pseudo likelihood (NPL)algorithm proposed by Aguirregabiria and Mira(2002). The NPL is a procedure to estimate dis-crete choice dynamic programming models that,in the context of single-agent models, providesthe maximum likelihood estimator of the struc-tural parameters. We provide here a descriptionof this procedure in the context of our model. Inthis section, we treat the variables WT

t , W Pit , and

ηit as observable to the researcher. These vari-ables, in fact, have been consistently estimatedin a first step, and therefore we actually observethe estimated values WT

t , W Pit , and ηit . For nota-

tional convenience, we will omit the “hats.” Thefact that the estimated values include estima-tion error does not affect the consistency of ourestimator of θ, though it affects its asymptoticvariance.

Let P0(dit |xit ) be the true distribution ofemployment changes, dit ≡ {�LP

it , �LTit }, con-

ditional on the state variables, xit ≡ (LPit−1,

LTit−1,Wt ,ηit ), in the population under study.

Define the vector P0 ≡ {P0(d|x) : (d, x) ∈ D ×X}. And define the (pseudo) log-likelihood func-tion:

(15)

Q(θ, σε,π, P0)

= ∑Ni=1

∑Ti

t=1 ln �(dit | xit ; θ, σε,π, P0)

where �(d | x; θ, σε,π, P0) is a function thatrepresents the probability that d is the optimalemployment choice for the firm given that thestate is x, the structural parameters are (θ, σε,π),and the firm believes that it will behave inthe future according to the choice probabilitiesP0. We describe the structure of the probabilityfunction � in Appendix B.

Let P0 be a nonparametric estimator of theset of conditional choice probabilities in P0.And let π be an estimator of the param-eters in the transition probability functionsof wages and the productivity shock. Giventhese estimates, we can calculate the presentvalues that enter into the probability func-tion � (see Appendix B) to obtain a crite-rion function Q(θ, σε, π, P0) that is the log-likelihood of a random-coefficients multino-mial logit model, where the random coeffi-cients come from the term εP�LP(d)(σP/σ0) +εT�LT(d)(σT/σ0). Given this likelihood, we canestimate the parameters θ, σ0, σP, and σT. Notethat these parameters are separately identified

18 ECONOMIC INQUIRY

TABLE 3Distribution of Employment Changes. Unbalanced Panel

Pre-Reform Period: 1982–1984

Change in Temporary Employment

% ≤−3 −2 −1 0 +1 +2 ≥+3 Total

Change in Permanent Employment ≤−3 3.2 0.5 2.2 22.8 2.2 1.1 6.0 37.9−2 0.3 0.3 0.4 3.8 0.4 0.2 0.5 5.9−1 0.4 0.1 0.3 6.5 0.5 0.4 0.5 8.80 1.1 0.2 0.6 9.5 1.3 0.6 1.8 15.1

+1 0.6 0.1 0.5 4.6 0.8 0.4 0.6 7.4+2 0.5 0.1 0.2 2.5 0.3 0.2 0.8 4.4

≥+3 2.3 0.5 0.7 11.7 0.9 0.4 3.8 20.4Total 8.5 1.7 4.9 61.3 6.3 3.4 13.9 100.0

Post-Reform Period: 1989–1992

Change in Temporary Employment

% ≤−3 −2 −1 0 +1 +2 ≥+3 Total

Change in Permanent Employment ≤−3 5.5 1.1 1.5 10.1 1.7 1.2 7.9 28.9−2 0.8 0.2 0.3 2.6 0.7 0.3 1.1 5.9−1 0.9 0.3 0.7 4.1 0.7 0.4 1.3 8.40 1.5 0.5 1.4 11.3 1.6 1.0 2.3 19.6

+1 1.0 0.3 0.8 3.2 0.9 0.4 1.2 7.7+2 0.5 0.4 0.5 2.6 0.5 0.3 0.9 5.8

≥+3 6.3 0.7 0.8 7.7 1.0 1.1 6.0 23.7Total 16.5 3.5 6.0 41.5 7.0 4.7 20.7 100.0

from θ/σ0, σT/σ0, and σP/σ0 because the firstelement of θ, which is associated with thevalue of output minus the wage bill, is equalto 1. The estimator of (θ, σε) that maximizesQ(θ, σε, π, P0) is consistent and asymptoticallyequivalent to the maximum likelihood estimator(see Proposition 4 in Aguirregabiria and Mira2002). A recursive extension of this two-stepmethod returns the (conditional) maximum like-lihood estimator of (θ, σε). The time-discountfactor is fixed at β = 0.95.(a) Discretization of employment changes (deci-sion variable) and of employment levels (endoge-nous state variables). The main reason why weconsider a discrete model of labor demand isthe significant lumpiness in observed employ-ment changes. Table 3 presents the empiricaldistributions of annual changes in temporaryand permanent employment, i.e., in the num-ber of workers. These variables have a discretedistribution, and a small number of discrete val-ues account for a large proportion of observa-tions. For instance, 57% of the observations ofemployment changes are between −5 and +5workers (11 discrete values), and 72% of theobservations are between −10 and +10 (21 val-ues). We believe that a continuous-choice modelof labor demand is not appropriate for this type

of data. Furthermore, the solution and estimationof the dynamic decision model requires dis-cretization of state variables.

Although the distributions of employmentchanges and employment levels are discrete andlumpy, they have also long tails. We wouldneed a discrete grid with too many values toaccount for more than 90% of the sample val-ues of these variables. A large dimension ofthe state space implies a significant increasein computational cost in the solution and esti-mation of the model. To deal with this issue,we consider a discretization that is firm spe-cific. The main reason why the range of sam-ple variation of employment variables is quitelarge is the existence of between-firm hetero-geneity in the level of employment. Actually,the range of values in the variation of employ-ment over time for a firm is much narrower thanthe range of values in the cross-sectional distri-bution of employment across firms. Therefore,using a firm-specific discretization, we can cap-ture most between-firms and within-firms sam-ple variation in employment with a relativelysmall and manageable state space. For j = T, P,�L

j

it and Lj

it be the discretized values of �Lj

it

and Lj

it , respectively. Let Li be the “long-term”total employment of firm i as measured by

AGUIRREGABIRIA & ALONSO-BORREGO: LABOR MARKET REFORM 19

the firm-specific sample mean.10 And definethe discrete variables d

j

it ≡ 100(�Lj

it /Li), and�j

it ≡ 100(Lj

it /Li). The range of values for thediscretized employment changes, d

j

it , is definedas percentages of Li between −20% and +20%with steps of 2 percentage points, i.e., d

j

it ∈{−20, −18, . . . , −4, −2, 0, +2, +4, . . . , +18,+20}. Similarly, �

j

it , is defined as percentagesof Li between 40% and 120%, with steps of 2percentage points, i.e., �

j

it ∈ {40, 42, . . . , 118,120}. Given the discrete variables d

j

it and �j

it , wecan obtain the values for employment changesand employment levels as �L

j

it = (dj

it /100)Li

and Lj

it = (�j

it /100)Li , respectively. In Figure 9,we present the histograms of the discrete deci-sion variables dP

it and dTit , and of the endogenous

state variables �Pit and �T

it .(b) Discretization of exogenous state variables.We follow Tauchen (1986) and Tauchen andHussey (1991) to choose the discretization gridof the exogenous state variables (WT

t ,W Pit ,ηit ).

For each of these variables, we estimate anAR(1) process and follow Tauchen-Hussey pro-cedure. However, for the state variables (W P

it ,ηit ) we apply a different discretization foreach individual firm. That is, the discretiza-tion applies to the variables in deviations withrespect to their firm-specific means: Wit − Wi

and ηit − ηi . By using firm-specific discretiza-tions, we can capture most of the time-seriesvariability of the state variables without havingto consider too many grid points. The total num-ber of cells in the discretized state space X is6,888 (i.e., 41 for permanent employment, 21for temporary employment, 2 for wage of tem-poraries, 2 for wage of permanents, and 2 forthe productivity shock).

To estimate our dynamic labor demandmodel, we have considered alternative spec-ifications of the unobservables, including thepure conditional logit without random coeffi-cients and different random-coefficient models,and alternative assumptions on the variancesof the ε’s, including homoscedasticity andheteroscedasticity. The choice of our most-preferred specification has been based on twocriteria. First, the sign and magnitude of the esti-mated parameters should have economic sense.

10. We could have used other measure of firm size forthe discretization, such as average capital stock, Ki , averageoutput, Y i , or average TFP, ηi . All these variables are highlycorrelated with average employment.

Second, the model should provide a reasonablefit for aggregate statistics such as the aggre-gate time path of the proportion of temporaryworkers, the percentage of zeroes in the distri-bution of employment changes, the average jobturnover rates, and the cross-sectional varianceof employment levels.

Following these criteria, our favorite spec-ification is a model where labor adjustmentcosts (fixed, linear, and quadratic) and thestandard deviation of the unobservable ε’s areproportional to the firm-specific mean of thesalary-per-worker. For instance, the linear costof firing permanent workers for firm i is θP

F1i =θPF1 Wi , where θP

F1 is the same parameter forevery firm and Wi is firm i’s mean salary perworker, i.e., Wi = (1/Ti)

∑Ti

t=1 Wit . The samespecification applies to the other θ parametersin labor adjustment costs. Similarly, the vari-ances of the unobservables are var(εP

it ) = σ2P

W2i , var(εT

it ) = σ2T W

2i , and var(ε0

it ) = σ20 W

2i .

It is important to note that the model with ran-dom coefficients provides both more sensibleresults and better fit that the pure conditionallogit model. For instance, under the conditionallogit model, the estimates of some lump-sumadjustment costs are negative and significant,and most quadratic adjustment costs are unre-alistically large. Besides, such model fails to fitthe thick tails in the empirical distribution ofemployment changes.

Table 4 presents the estimates of the dynamiclabor demand model for our preferred specifi-cation. We have estimated the model for threesubperiods: the pre-reform period 1983–1984,and two post-reform periods, 1985–1988 and1989–1992. Table 5 provides goodness-of-fitmeasures of the estimated model. The fit of theestimated model, for the three sample periods,to the different aggregate statistics, is very good,except for the proportion of zeros in the changeof temporary employment. In this latter case, themodel under estimates such proportion, yet thedegree of underestimation is similar for the threesample periods.

Regarding adjustment costs, the linear com-ponents reveal as the most important ones forboth contract types, either for hiring or for fir-ing. The quadratic components are very smalland nonsignificant. The fixed components aregenerally small, except the fixed cost of firingpermanent workers in the three periods, and thefixed cost of hiring temporary workers in thepre-reform period.

20 ECONOMIC INQUIRY

FIGURE 9Histograms of Discretized Decision and State Variables

0

0.02

0.04

0.06

0.08

0.1

Den

sity

-20 -10 0 10 200

0.05

0.1

0.15

0.2

-20 -10 0 10 20

0

0.01

0.02

0.03

0.04

Den

sity

Den

sity

Den

sity

40 60 80 100 1200

0.05

0.1

0.15

0.2

0.25

0 10 20 30 40

In the case of linear hiring components, thehiring costs per worker are fairly similar for bothcontract types, ranging between 10% and 18%of a worker’s annual salary. Interestingly, lin-ear hiring costs for both contract types seemto experience a decline after the reform. Thisfact may be pointing out that the screening costsof workers are lowered after the reform. In thecase of temporary workers, most new contractsbecome temporary contract after the reform, sothat the pool of candidates for a job under atemporary contracts is hugely widened. In thecase of permanent workers, hiring a permanentworker after the reform means, typically, pro-moting a temporary worker who was already inthe firm to a permanent position. Such promo-tion may be less expensive than the recruitmentof an outsider.

The linear firing costs for permanent workersamount between 46% and 53% of a worker’sannual salary, and are fairly similar betweenthe pre-reform and the post-reform periods.

Taking into account that the regulated severancepayments for permanent contracts established45 days of salary for each year of tenure, ourestimate of the linear firing cost corresponds tothe dismissal compensation for a worker with 4years of tenure. This seems a reasonable tenurefor the typical dismissed permanent worker.Linear firing costs for temporary workers, onthe other hand, are relatively small (between 4%and 10%), showing a decline between the pre-reform and the post-reform periods, when theybecome nonsignificant.

Comparing the estimated adjustment costcomponents before and after the reform, we findseveral significant reductions in the post-reformperiod. Many of them affected the structure ofadjustment cost for temporary employment: inparticular, its fixed and linear components ofboth hiring and firing costs were reduced. Themost outstanding drop corresponded to fixed hir-ing costs for temporary employment. This dra-matic change in such costs indicates clearly the

AGUIRREGABIRIA & ALONSO-BORREGO: LABOR MARKET REFORM 21

TABLE 4Estimation of of the Dynamic Labor Demand Model. Unbalanced Panel 1982–1993 (2,356 Firms)a

Period 1982–1984 Period 1985–1988 Period 1989–1992

AC Parameters Permanent Temporary Permanent Temporary Permanent Temporary

Fixed hiring cost: θjH0,i/Wi 0.012 1.417∗∗ 0.018 0.097∗∗ 0.028 0.049

(0.061) (0.060) (0.035) (0.039) (0.041) (0.046)

Linear hiring cost: θjH1,i/Wi 0.183∗∗ 0.181∗∗ 0.117∗∗ 0.107∗∗ 0.101∗∗ 0.089∗∗

(0.058) (0.049) (0.038) (0.041) (0.043) (0.045)

Quad hiring cost: θjH2,i/Wi 0.0003 0.0001 0.0006 −0.0001 0.0005 −0.0008

(0.0006) (0.0004) (0.0008) (0.0006) (0.0005) (0.0009)

Fixed firing cost: θj

F0,i/Wi 0.083∗∗ 0.067 0.136∗∗ 0.061 0.080∗∗ 0.058(0.038) (0.094) (0.024) (0.084) (0.036) (0.113)

Linear firing cost: θj

F1,i/Wi 0.514∗∗ 0.098∗∗ 0.464∗∗ 0.060 0.528∗∗ 0.051(0.098) (0.045) (0.035) (0.037) (0.080) (0.048)

Quad firing cost: θjF2,i/Wi −0.00043∗ −0.00006 0.00006 0.00037 −0.0006 0.0005

(0.00022) (0.0011) (0.0007) (0.0008) (0.0008) (0.0009)

Other Parameters Period 1982–1984 Period 1985–1988 Period 1989–1992

σP,i/Wi 0.611 0.515 0.540(0.080) (0.058) (0.079)

σT ,i/W i 0.170 0.172 0.149(0.025) (0.019) (0.026)

σ0,i /W i 0.871 0.710 0.761(0.069) (0.054) (0.064)

Period 1982–1984 Period 1985–1988 Period 1989–1992

# Observations 2, 274 7, 219 6, 257LR indexb 0.232 0.220 0.267

aAll the parameters are unit-free because all firing and hiring costs are proportional to the firm-specific average wage Wi .Standard errors are reported between parentheses.

bThe LR (Likelihood Ratio) Index is a measure of goodness of fit defined as 1 − (log L/ log L0), where log L is thelog-likelihood of the estimated model, and log L0 is the log-likelihood under the the hypothesis that all parameters exceptσ0,i are equal to zero.

difference in the aim of temporary contracts,and therefore in the characteristics of work-ers hired under such contracts, before and afterthe reform. Namely, the link to the causalityprinciple (cover an absent post, seasonal work,production campaign, etc.) can justify high hir-ing costs before the reform. Under the lim-ited reasons that justify the use of temporarycontracts before the reform, firms may havestrong incentives to screen candidates, whichcan explain substantial fixed costs of hiring (e.g.,when hiring was aimed to cover unexpectedabsent posts). Together with the significant dropin the fixed cost of hiring temporary workers,we find an increase in the proportion of firmsusing temporary workers from 45% in 1984 to75% in 1993, accompanied by a more inten-sive use by these firms of temporary contracts.Concerning permanent employment, the linear

hiring and firing costs were reduced after thereform.

Finally, the parameters related with the dis-persion of the unobservable shocks are all sig-nificantly positive, and keep stable over time.

From our estimation of the structural eq-uations, the overall picture that appears onthe effects of the reform is the following:(1) it has made it cheaper to hire and firetemporary workers, both at the intensive andat the extensive margin; (2) it has reducedthe cost of hiring permanent workers, probablybecause promoting an insider (temporary) toa permanent position is less expensive thanrecruiting an outsider; (3) the productivity oftemporary workers has become closer to the oneof permanents; and (4) the wage-gap betweenpermanent and temporary workers has widenedafter the reform.

22 ECONOMIC INQUIRY

TABLE 5Goodness of Fit Measures of the Estimated Model. Unbalanced Panel 1982–1993 (2,356 Firms)

Period 1983–1984 Period 1985–1988 Period 1989–1992Model Model Model

Statistics (Empirical) (Empirical) (Empirical)

Median permanent employment 98.0 64.0 59.0per firm (95.0) (66.0) (56.0)

Mean proportion of temporary 4.4% 6.6% 11.8%workers (4.3%) (6.9%) (11.3)

Percentage of 14.8% 18.1% 19.7%zeroes in �LP (15.1%) (18.8%) (19.6%)

Percentage of 46.9% 39.8% 28.1%zeroes in �LT (52.8%) (43.9%) (32.5%)

Median Value of dP 4.0% 4.8% 5.6%conditional on dP > 0 (3.9%) (5.2%) (5.7%)

Median Value of dP −3.9% −4.0% −5.1%conditional on dP < 0 (−3.8%) (−4.3%) (−5.2%)

Median Value of dT 1.7% 2.9% 4.0%conditional on dT > 0 (1.7%) (2.7%) (4.2%)

Median Value of dT −1.3% −2.3% −3.7%conditional on dT < 0 (−1.4%) (−2.0%) (−3.8%)

Cross-sectional Variance 1.64 1.66 1.56log Perm. Employment (1.66) (1.72) (1.59)

Cross-sectional Variance 1.54 1.60 1.49log Total Employment (1.59) (1.64) (1.51)

VI. POLICY EVALUATION

We use the estimated model to evaluate theeffects, on employment, job turnover, productiv-ity and firms’ value, of the introduction of tem-porary contracts. We also compare such effectswith those associated with a counterfactual pol-icy that halves linear-firing costs for all typeof workers. To implement these policy evalu-ations, we select the firms active in the samplein 1984. For this group of firms, we solve forthe value function and the optimal decision rulein three dynamic programming models: a “pre-reform” model, a “post-reform” model, and acounterfactual model. For the three models, realwages are assumed to be constant at their 1984levels (i.e., the policy evaluation considers par-tial equilibrium effects), and the stochastic pro-cess of the productivity shock is the one for theperiod 1983–1984. In the pre-reform model, thevalue of the other structural parameters are theones estimated for the period 1983–1984. Forthe post-reform model, the structural parame-ters are the estimates for the period 1989–1992.Finally, for the counterfactual model, we fix thevalues of the parameters at their 1983–1984level, except for the linear firing costs θP

F1and θT

F1 which are reduced by half: i.e., thecounterfactual values of θP

F1,i/W i and θTF1,i/W i

are 0.257 and 0.049, respectively. For eachmodel, we calculate the steady-state distributionof the state variables and use this distribution toobtain the mean values of employment, output,etc.

Table 6 presents the results of these experi-ments. The introduction of temporary contractshad important positive effects on total employ-ment (a 3.5% increase) and job turnover. Theincrease in total employment is associated witha strong substitution of permanent by temporaryworkers: the proportion of temporary workersrises from 3.8% to 16.2%. Permanent employ-ment declines by 10%. The positive effectson productivity (0.7%) and the value of firms(1.2%) are small. These effects contrast sub-stantially with the ones of the counterfactualreform. While the effects on total employmentare alike (a 4.1% increase), the counterfac-tual reform had improved permanent employ-ment (6.6% increase), labor productivity (1.9%increase), and the value of firms (4.8%). Further-more, the proportion of temporary employmentbecomes almost null (1.3%).

Recently, several Spanish economists (seeAndres et al. 2009) have advocated a new labormarket reform in Spain to establish a single(permanent) contract with lower firing costs.Actually, in our experiment, the proportion of

AGUIRREGABIRIA & ALONSO-BORREGO: LABOR MARKET REFORM 23

TABLE 6Evaluation of the Labor Market Reform

Pre-Reform Post-Reform CounterfactualStatistics Economya Economya Reforma

Median permanent employment per firm 99.0 89.1 (−10.0%) 105.5 (+6.6%)Median total employment per firm 102.7 106.3 (+3.5%) 106.9 (+4.1)Mean proportion of temporary workers 3.8% 16.2% 1.3%Median absolute value of �LP 2.9% 2.5% 4.2%Median absolute value of �LT 0.0% 3.1% 0.9%Median output per firmb 100 100.7 101.9Median value of a firmb 100 101.2 104.8

aThe values of the structural parameters are: for the pre-reform model, the ones estimated for the period 1983–1984;for the post-reform model, the ones estimated for the period 1989–1992; for the counterfactual model, we consider the1983–1984 parameters except for the linear firing costs θP

F1 and θTF1 which are reduced by half.

bWe normalize at 100 output-per-firm and value of a firm in the pre-reform model.

temporary workers under the counterfactual isvery small (1.3%). Our counterfactual policyis very similar to the advocated labor marketreform if firing costs were reduced by 50%.More generally, our estimates show that a suffi-ciently large reduction in firing costs of perma-nent workers would make temporary contractssuboptimal, being equivalent to the legal elimi-nation of this type of contract.

VII. CONCLUDING REMARKS

Using panel data of Spanish manufactur-ing firms, we have estimated a dynamic labordemand model and evaluated the effects ofa reform that introduced temporary contractsin 1984. The structural model allows for arich specification of labor adjustment costs,including fixed, linear, and quadratic compo-nents, and unobserved firm-heterogeneity (i.e.,random coefficients). The model with random-coefficients provides a better fit and more sensi-ble results that a simpler conditional logit model.Our estimation results show significant changesin structural parameters after the reform. Hiringand firing temporary workers has become lessexpensive, both at the intensive and at the exten-sive margins, and the cost of hiring permanentworkers has declined. Based on the estimatedmodel, we present counterfactual experiments toevaluate the effects of the reform. The reformhad important effects on employment and jobturnover, but modest effects on productivity andvalue of firms. However, we also find that acounterfactual policy that halved firing costs forboth contracts implies similar effects on totalemployment, but the positive effects on output,

value of firms, and permanent employment aremuch stronger.

Between 1984 and 1997, the labor mar-ket regulations remained essentially unchanged.Since 1997, several reforms of labor market reg-ulations have been undertaken. Most of themhave been aimed at limiting the widespread useof temporary contracts, and boost the use ofpermanent contracts through lower social secu-rity and termination costs. The rules to limittemporary contracts have proved ineffective,to the extent that the proportion of temporaryworkers has remained stable around 30% forthe total economy (with a slightly lower inci-dence in manufacturing). With regard to thereduction in severance payments for permanentcontracts, they have been limited to new con-tracts of certain collectives of workers, whilethe severance payments for existing contractsand for many new permanent contracts haveremained unchanged. The behavior of the Span-ish labor market, therefore, has not lived sub-stantial changes since the 1984 reform, and fea-tures Spain nowadays as the OECD country withthe highest proportion of temporary employ-ment.

The duality of the Spanish labor market hasbeen strengthened in the last two decades, sothat 30% of the working people, those withtemporary contracts, bear most of employmentturnover, to the extent that all the flexibility ofthe labor market is provided by them. The inci-dence of these contracts differs very much byworker characteristics, affecting more female,young, and less educated people, as well as peo-ple with long unemployment episodes. We haveseen that the introduction of these contracts ledto an increase in the employment level, but at

24 ECONOMIC INQUIRY

the expense of a lower productivity per worker.In fact, unlike other countries (see Goux, Mau-rin, and Pauchet 2001), it seems that the roleof temporary contracts as a screening deviceunder asymmetric information is very limited.In the presence of such low conversion rates,firm-specific human capital investment seemsto be negatively affected by temporary con-tracts. According to Dolado, Garcıa-Serrano,and Jimeno (2002), during the cyclical upturnof 1986–1990, where employment growth wasmostly due to temporary contracts, labor produc-tivity hardly reached an annual average growthof 1%. And labor productivity growth has beensimilarly low during the 1997–2000 upturn. Thelack of firm-specific human capital investmentmay be behind this fact. Precisely, the coun-terfactual reform that we have posed illustratesthis negative aspect of temporary contracts. Wehave seen that an alternative reform reducingall firing costs by 50% would provide a similaremployment level, yet with a low proportion oftemporary employment, and as a consequence,a higher productivity per worker than under thecurrent setting. The current regulations of theSpanish labor market fail to provide incentivesfor firm-specific human capital investment. Sev-eral recent initiatives have advocated for theelimination of the firing cost gap between tem-porary and permanent contracts in Europe, intro-ducing a single labor contract with a severancepay that would be low at the beginning of thecontract and increase with workers’ tenure. Thelast labor market reform, implemented by theSpanish Government at the beginning of 2012,has reduced dismissal costs for permanent work-ers who will be hired thereafter and widenedthe causes for fair dismissals, but use of tem-porary contracts has not been limited, so thedual labor market still prevails. In contrast withthe existing dual schedule, the single contractproposal does not discourage job creation andfirm-specific human capital investment, removesthe incentives to the inefficient turnover of tem-porary workers and keeps protection to tenuredworkers.

Finally, we must acknowledge that the afore-mentioned data limitations may affect our esti-mates, and therefore our results should be qual-ified. First, our dataset lacks measures of grossemployment flows, so that our results are basedon measures net employment changes. Giventhe relative importance of voluntary quits andsimultaneous hires and layoffs in the presenceof sizeable redundancy payments, this may lead

to underestimate the importance of firing costsfor permanent workers. Second, our dataset cov-ers only manufacturing firms, and overrepresentslarge firms. The incidence of temporary con-tracts is lower in manufacturing than in othersectors of the economy, and is also lower forlarger firms. Regarding this, we believe that theconsequences of the two-tier labor market thatwe have analyzed, as well as the potential effectsof the counterfactual reform, are expected to belarger for the whole economy.

APPENDIX A. DATA DESCRIPTION

Construction of the CBBE Dataset

The sample consists of an unbalanced panel of nonen-ergy manufacturing firms, recorded in the database of theBank of Spain’s Central Balance Sheet Office. This datasetwas started in 1982 collecting firm data on a voluntary basis,which implied that largest firms were overrepresented in thesample. However, the tendency in subsequent years has beencharacterized by the addition of firms of smaller relativesize. The firms included in this database represent almost40% of the total value added in Spanish manufacturing, so itaccounts for a very large proportion of output in the Spanisheconomy. In Table A1, we present the sample distributionof firms by industry and size (measured as the time averageof firm’s employees).

To obtain the final sample of 2,356 firms, we haveeliminated those for which some of the following variableswere negative or took implausible values: book valueof capital stock, sales, gross output, total labor costs,permanent employment, and temporary employment. Also,we have disregarded those firms with a public share above50%. Finally, for the purposes of estimation, we haverequired firms to have at least 5 years of consecutiveobservations. Under these criteria, we have discarded 12%of the observations. In terms of total number of employees,the average employment is very similar in the final sampleand in the raw data (248 vs. 238 employees, respectively).

Construction of Variables

Employment. Employees are disaggregated by type of con-tract in permanent (open-ended) and temporary (fixed-term)employees. Information on the the average duration (inweeks) of temporary contracts is also available. To maintainmeasurement consistency, the number of temporary employ-ees is normalized in annual terms by multiplying the numberof temporary employees along the year times the aver-age number of weeks worked by temporary employees anddivided by 52. It is worth to notice that, as happens in mostfirm-level datasets, there is not information on employmentflows along the year, and therefore only net employmentchanges are observed.Wages. The information on the firm’s total wage bill (whichallows to calculate the average wage rate for total employeesat the firm-level) is not broken down by type of contract.Output. Gross output at retail prices is calculated as totalsales, plus the change in finished product inventories andother income from the production process, minus taxesderived on the production (net of subsidies). Real outputhas been obtained using as deflator the Retail Price Index at3-digit industry level.

AGUIRREGABIRIA & ALONSO-BORREGO: LABOR MARKET REFORM 25

TABLE A1Distribution of Firms by 2-digit Industry and by Size. Unbalanced Panel 1982–1993 (2,356 Firms)

Small Med1 Med 2 Large Total

Iron, steel, and metal (22) Abs. freq. 5 8 10 22 45% by ind. 11.11 17.78 22.22 48.89 100.00% by size 1.29 0.94 1.73 4.10 1.91

Bldg. materials, glass, and ceramics (24) Abs. freq. 27 88 34 33 182% by ind. 14.84 48.35 18.68 18.13 100.00% by size 6.98 10.29 5.89 6.15 7.72

Chemicals (25) Abs. freq. 39 99 76 92 306% by ind. 12.75 32.35 24.84 32.07 100.00% by size 10.08 11.58 13.17 17.13 12.99

Nonferrous metal (31) Abs. freq. 38 103 53 31 225% by ind. 16.89 45.78 23.56 13.78 100.00% by size 9.82 12.05 9.19 5.77 9.55

Basic machinery (32) Abs. freq. 29 52 47 33 161% by ind. 18.01 32.30 29.19 20.50 100.00% by size 7.49 6.08 8.15 6.15 6.83

Office machinery (33) Abs. freq. 0 1 0 3 4% by ind. 0.00 25.00 0.00 75.00 100.00% by size 0.00 0.12 0.00 0.56 0.17

Electric materials (34) Abs. freq. 11 29 24 35 99% by ind. 11.11 29.29 24.24 35.35 100.00% by size 2.84 3.39 4.16 6.52 4.20

Electronic (35) Abs. freq. 3 8 10 14 35% by ind. 8.57 22.86 28.57 40.00 100.00% by size 0.78 0.94 1.73 2.61 1.49

Motor vehicles (36) Abs. freq. 8 21 25 36 90% by ind. 8.89 23.33 27.78 40.00 100.00% by size 2.07 2.46 4.33 6.70 3.82

Ship building (37) Abs. freq. 3 2 2 6 13% by ind. 23.08 15.38 15.38 46.15 100.00% by size 0.78 0.23 0.35 1.12 0.55

Other motor vehicles (38) Abs. freq. 2 5 5 6 18% by ind. 11.11 27.78 27.78 33.33 100.00% by size 0.52 0.58 0.87 1.12 0.76

Precision instruments (39) Abs. freq. 2 8 3 4 17% by ind. 11.76 47.06 17.65 23.53 100.00% by size 0.52 0.94 0.52 0.74 0.72

Nonelaborated food (41) Abs. freq. 53 83 46 48 230% by ind. 23.04 36.09 20.00 20.87 100.00% by size 13.70 9.71 7.97 8.94 9.76

Food, tobacco, and drinks (42) Abs. freq. 53 51 31 45 180% by ind. 29.44 28.33 17.22 25.00 100.00% by size 13.70 5.96 5.37 8.38 7.64

Basic textile (43) Abs. freq. 20 57 53 37 167% by ind. 11.98 34.13 31.74 22.16 100.00% by size 5.17 6.67 9.19 6.89 7.09

Leather (44) Abs. freq. 4 16 12 4 36% by ind. 11.11 44.44 33.33 11.11 100.00% by size 1.03 1.87 2.08 0.74 1.53

Garment (45) Abs. freq. 11 48 34 22 115% by ind. 9.57 41.74 29.57 19.13 100.00% by size 2.84 5.61 5.89 4.10 4.88

Wood and furniture (46) Abs. freq. 21 45 26 8 100% by ind. 21.00 45.00 26.00 8.00 100.00% by size 5.43 5.26 4.51 1.49 4.24

26 ECONOMIC INQUIRY

TABLE A1Continued

Small Med1 Med 2 Large Total

Cellulose and paper edition (47) Abs. freq. 29 63 42 33 167% by ind. 17.37 37.72 25.15 19.76 100.00% by size 7.49 7.37 7.28 6.15 7.09

Plastic materials (48) Abs. freq. 22 46 33 17 118% by ind. 18.64 38.98 27.97 14.41 100.00% by size 5.68 5.38 5.72 3.17 5.01

Other nonbasic (49) Abs. freq. 7 22 11 8 48% by ind. 14.58 45.83 22.92 16.67 100.00% by size 1.81 2.57 1.91 1.49 2.04

Total Abs. freq. 387 855 577 537 2,356% by ind. 16.43 36.29 24.49 22.79 100.00% by size 100.00 100.00 100.00 100.00 100.00

Note: Size is defined in accordance with the firm’s average number of employees, which is up to 25 for Small, between25 and 75 for Med1, between 75 and 200 for Med2, and greater than 200 for Large.

Physical capital. We are interested in depreciable physicalcapital which is already productive, so Land and Capitalstock in course of construction are excluded from thedefinition of the stock of physical capital. Since the CBBEdoes not have independent estimates of investment available,gross nominal investment Iit must be imputed from changesin the book value of physical capital with a correction fordepreciation, that is Iit = KNBit − KNBi,t−1 + Depit +Revit where, KNBit = KGBit − ADepit is the book valueof the net stock of physical (book value of the gross stockof physical capital KGBit minus accumulated depreciationADepit ); Depit is the accounting depreciation during theyear; and Revit is the net variation in the book value ofphysical capital and in its accumulated depreciation dueto positive and/or negative revaluations. To calculate thereplacement value of capital, we use a perpetual inventorymethod which takes account for depreciation and inflation.To do this, an initial value for the first year that datais available for a given firm is calculated as q1Ki1 =(q1/q1−AAi

) × KGBit (1 − δi )AAi where qt is the price

deflator of the stock of physical capital at year t ; δi is theaverage depreciation rate of the stock of physical capital;and AAi is the average age of the stock of physicalcapital, which is approximated by the ratio ADepi1/Depi1for the first year in which data for the firm are available.Furthermore, the average depreciation rate is computed atthe firm level as the ratio of firm’s average accountingdepreciation to the firm’s average accumulated depreciation.As regards price indices, the corresponding GDP implicitdeflator of investment goods is used (Source: INE). Therecursive method to compute the replacement value ofthe stock of physical capital from the second year thatdata is available is qtKit = (qt /qt−1) × Ki,t−1(1 − δi ) +Iit ,which assumes that investment occurs at the end ofthe year.

APPENDIX B. PROBABILISTIC REPRESENTATION OFFIRMS’ EMPLOYMENT DECISIONS

Remember that the specification of the one-profit func-tion is �t = z (dt , xt ) θ + ξ(dt , εt ), where θ is the 13 × 1vector

(1, θT

H0, θTH1, θ

TH2, θ

TF0, θ

TF1, θ

TF2, θ

PH0, θ

PH1, θ

PH2, θ

PF0,

θPF1, θ

PF2

), and z

(dt , xt

)is a 1 × 13 vector as described in

Equation (8). The unobservable term ξ(dt , εt ) is:

ξ(dt , εt ) = σP εPt �LP(dt ) + σT εT

t �LP(dt )(A1)

+σ0 ε0t (�LP(dt ), �LT(dt ))

where �LP(d) and �LTT(d) represents the values of �LPt

and �LTt associated to the discrete choice alternative dt = d.

If we had a “myopic” model (i.e., β = 0), the choice proba-bilities in the likelihood function would have the followingform:

(A2)

Pr(d | x)

=∫ exp

{z (d, x)

θ

σ0+ εP�LP(d)

σP

σ0+ εT�LT(d)

σT

σ0

}∑j∈D

exp

{z (j, x)

θ

σ0+ εP�LP(j)

σP

σ0+εT�LT(j)

σT

σ0

}

×φ(dεP)φ(dεT)

where φ is the pdf of the standard normal. We cancalculate numerically the double integral in this proba-bility function. In this article, we have calculated thisintegral using the Gauss-Legendre quadrature method pro-vided by the command intquad2 in the GAUSS softwarepackage.

In our dynamic, forward-looking, model, the choiceprobabilities have a similar structure as in (A2) but wehave to replace the current profit z (d, x) θ/σ0 by thepresent value (expected and discounted intertemporal profits)of choosing alternative d when the state is x. Let X

be the space of possible values of xt . A firm’s optimalbehavior can be represented using a vector of optimalchoice probabilities P = {P(d|x) : (d, x) ∈ D × X}, whereP(d|x) is the probability that the optimal decision is d

conditional on the value of xt being x. Define also π asthe vector of parameters that characterize the transitionprobabilities fη and fW . We can obtain the vector P asthe unique fixed point of a contraction mapping in thespace of conditional choice probabilities: P = �(P). Underour assumptions on the probability distribution of εt , this

AGUIRREGABIRIA & ALONSO-BORREGO: LABOR MARKET REFORM 27

mapping is:

(A3)

�(d | x; P)

=∫ exp

⎧⎪⎨⎪⎩

ZP (d, x)θ

σ0+ eP (d, x)

+εP�LP(d)σP

σ0+ εT�LT(d)

σT

σ0

⎫⎪⎬⎪⎭

∑j∈D

exp

⎧⎪⎨⎪⎩

ZP (j, x)θ

σ0+ eP (j, x)

+εP�LP(j)σP

σ0+ εT�LT(j)

σT

σ0

⎫⎪⎬⎪⎭

× φ(dεP)φ(dεT).

ZP (d, x)θ

σ0+ eP (d, x) is the present value of choosing

alternative d when the state is x. More specifically, ZP (d, x)

is a 1 × 13 vector with the present values of the streamof current and future values of the vector z (., .) condi-tional on the current value of (dt , xt ) being (d, x) andunder the assumption that the firm will behave in thefuture according to the choice probabilities in P. Forinstance, the first element of the vector ZP (d, x) is thepresent value of output minus wage bill. In a similar man-ner, eP (d, x) is a scalar containing the present value ofthe stream of future realizations of ξ(dt+j , εt+j ) associ-ated with optimal future choices. More formally, we havethat:

ZP (d, x) = z (d, x) +∑∞

s=1βj

E(z(d∗

t+s , xt+s

)(A4)

| dt = d, xt = x; P)

and

eP (d, x) =∑∞

s=1βj(A5)

× E(ξ(d∗

t+s , εt+s ) | dt = d, xt = x; P)

where d∗t+s represents the optimal employment decision

s periods ahead under the assumption that the probabili-ties in P are the ones associated with the optimal deci-sion rule. It is important to emphasize that to obtain thevalues ZP (d, x) and eP (d, x) we only need to know theprobabilities P, the parameters π, and the discount factorβ.

In what follows, we outline the procedure to calculate thevalues of ZP (d, x) and eP (d, x), which is explained in depthby Aguirregabiria and Mira (2002, 2010). Let W P

z (x)θ +W P

e (x) be the expected discounted utility of behavingaccording to choice probabilities P from current period t andinto the infinite future when xt = x. Consider that the spaceof state variables X is discrete. Define the matrix WP

z ≡{W P

z (x) : x ∈ X} and the vector WPe ≡ {W P

e (x) : x ∈ X}. Itcan be shown that WP

z is the unique solution of the recursiveequation WP

z = ∑d∈D P(d) ∗ {z(d)+β Fx(d) WP

z }, whereP(d) is the column vector of choice probabilities {P(d|x) :x ∈ X}; z(d) is the matrix {z(d, x) : x ∈ X}; Fx(d) is atransition probability matrix with elements fx(xt+1|xt , d);and ∗ is the element-by-element product. Likewise, WP

e

is the unique solution of the recursive equation WPe =∑

d∈D P(d) ∗ {eP(d)+β Fx(d) WPz } where eP(d) is the

vector {eP(d, x) : x ∈ X} and eP(d, x) ≡ E(ξ(d∗t , εt )|d∗

t =d, xt = x), which is a known function of the choice prob-abilities P(d|x). The objects WP

z and WPe can be com-

puted by successive approximations, iterating on the con-traction mappings that implicitly define them. Note that

the mappings that implicitly define WPz and WP

e are lin-ear in these objects. Therefore, there is a closed formexpression for WP

z and for WPe . For instance, WP

z = (I −β

∑Ja=0 P(a) ∗ Fx(a))−1 ∑J

a=0 P(a) ∗ z(a). When the num-ber of cells in X is small enough, matrix inversion algo-rithms may be preferable to successive approximations. Thematrix (I − βF)−1 can also be approximated using the seriesI + βF + β2F2 + · · · + βKFK , with K large enough. Thiscan be easier than matrix inversion. More generally, thisinverse matrix can be obtained iterating in A (successiveapproximations) in the mapping A = I + βF A.

Given our definition of the state variables in deviationswith respect to firm-specific means, the matrix WP

z and thevector WP

e are firm-specific, and they should be calculatedon a firm-by-firm basis. This means that we have to solve2, 356 systems of linear equations (one for each firm), eachwith a dimension of 6, 888 discrete values in the state space.For the calculation of the values eP(d, x) we have used thesimplifying assumption that the future values of εP and εT areequal to the expected values. Therefore, the only componentin eP(d, x) is the one that comes from the expectation of thefuture extreme value error ε0.

REFERENCES

Abowd, J., P. Corbel, and F. Kramarz. “The Entry andExit of Workers and the Growth of Employment:An Analysis of French Establishments.” Review ofEconomics and Statistics 81(2), 1999, 170–87.

Abraham, K., and S. Houseman. Job Security in America:Lessons from Germany. Washington, DC: BrookingsInstitution Press, 1993.

. “Does Employment Protection Inhibit Labor MarketFlexibility? Lessons from Germany, France, and Bel-gium,” in Social Protection vs. Economic Flexibility.Is There a Tradeoff? edited by R. M. Blank. Chicago:University of Chicago Press, 1994.

Ackerberg, D., K. Caves, and G. Frazer. “Structural Identifi-cation of Production Functions.” MPRA Paper 38349,University Library of Munich, 2006.

Addison, J., and J.-L. Grosso. “Job Security ProvisionsEmployment: Revised Estimates.” Industrial Rela-tions: A Journal of Economy and Society, 35, 1996,585–603.

Aguirregabiria, V., and C. Alonso-Borrego. “Labor Con-tracts and Flexibility: Evidence from a Labor MarketReform in Spain.” Working Paper 99-27, UniversidadCarlos III de Madrid, 1999.

Aguirregabiria, V., and P. Mira. “Swapping the Nested FixedPoint Algorithm: A Class of Estimators for DiscreteMarkov Decision Models.” Econometrica, 70, 2002,1519–43.

. “Dynamic Discrete Choice Structural Models: ASurvey.” Journal of Econometrics, 156, 2010, 38–67.

Alvarez, F., and M. Veracierto. “Fixed-Term EmploymentContracts in an Equilibrium Search Model.” Journalof Economic Theory, 147(5), 2012, 1725–53.

Andres, J., S. Bentolila, J. Dolado, and F. Felgueroso.“Propuesta para la reactivacion laboral en Espana.”El Paıs, 2009. Accessed April 26, 2013. http://elpais.com/diario/2009/04/26/negocio/1240751679_850215.html. English version published online in http://www.crisis09.es/PDF/restart-the-labor-market.pdf.

Autor, D., W. Kerr, and A. Kugler. “Does EmploymentProtection Reduce Productivity? Evidence from USStates.” The Economic Journal, 117, 2007, F189–217.

Bentolila, S. “Las Decisiones Judiciales sobre Despidos enEspana: Una Perspectiva Economica.” Cuadernos deInformacion Economica 128/129, 1997, 170–79.

28 ECONOMIC INQUIRY

Bentolila, S., and G. Bertola. “Firing Costs and LaborDemand: How Bad Is Eurosclerosis?” Review of Eco-nomic Studies, 57, 1990, 381–402.

Bentolila, S., and Dolado, J. “Labour Flexibility and Wages:Lessons from Spain.” Economic Policy, 18, 1994,53–99.

Bentolila, S., and G. Saint-Paul. “The MacroeconomicImpact of Flexible Labor Contracts, with an Applica-tion to Spain.” European Economic Review, 36, 1992,1013–53.

Bentolila, S., P. Cahuc, J. J. Dolado, and T. Le Barbanchon.“Two-tier Labour Markets in the Great Recession:France versus Spain.” The Economic Journal, 122,2012, F155–187.

Bertola, G. “Labor Turnover Costs and Average LaborDemand.” Journal of Labor Economics, 10, 1992,389–411.

Blanchard, O., and A. Landier. “The Perverse Effects ofPartial Labor Market Reforms: Fixed-Term Contractsin France.” Economic Journal, 112, 2002, F214–244.

Bond, S., and J. Van Reenen. “Microeconometric Modelsof Investment and Employment,” in Handbook ofEconometrics, Vol. 6A, edited by J. J. Heckman andE. Leamer. Amsterdam: North Holland, 2007.

Bover, O., M. Arellano, and S. Bentolila. “UnemploymentDuration, Benefit Duration and the Business Cycle.”The Economic Journal, 112, 2002, 223–65.

Bover, O., P. Garcia-Perea, and P. Portugal. “A ComparativeStudy of the Portuguese and Spanish Labour Markets.”Economic Policy, 15, 2000, 381–428.

Burgess, S., M. Knetter, and C. Michelacci. “Employmentand Output Adjustment in the OECD: A DisaggregateAnalysis of the Role of Job Security Provisions.”Economica, 67, 2000, 419–35.

Cabrales, A., and H. Hopenhayn. “Labor Market Flexibil-ity and Aggregate Employment Volatility.” Carnegie-Rochester Conference Series on Public Policy, 46,1997, 189–228.

Cahuc, P., and F. Postel-Vinay. “Temporary Jobs, Employ-ment Protection and Labor Market Performance.”Labour Economics, 9, 2002, 63–91.

Cooper, R., and J. Willis. “A Comment on the Economicsof Labor Adjustment: Mind the Gap.” American Eco-nomic Review, 94, 2004, 1223–37.

Costain, J., J. F. Jimeno, and C. Thomas. “EmploymentFluctuations in a Dual Labor Market.” Working Paper1013, Bank of Spain, 2010.

Davis, S., and J. Haltiwanger. “Gross Job Creation,Gross Job Destruction, and Employment Realloca-tion.” Quarterly Journal of Economics, 107, 1992,819–63.

Dolado, J., C. Garcıa-Serrano, and J. F. Jimeno. “Draw-ing Lessons from the Boom of Temporary Jobs inSpain.” Economic Journal 112(Special Supplement,2002, F270–95.

Galdon-Sanchez, J. E., and Guell, M. “Let’s Go toCourt! Firing Costs and Dismissal Conflicts.” Indus-trial Relations Section Working Paper No. 444, Prince-ton University, 2000.

Goux, D., E. Maurin, and M. Pauchet. “Fixed-Term Con-tracts and the Dynamics of Labour Demand.” Euro-pean Economic Review 45, 2001, 532–52.

Griliches, Z., and J. Mairesse. “Production Functions: TheSearch for Identification,” in Econometrics and Eco-nomic Theory in the Twentieth Century: The RagnarFrisch Centennial Symposium edited by S. Strøm.Cambridge, UK: Cambridge University Press, 1998.

Guell, M., and B. Petrongolo. “How Binding Are LegalLimits? Transitions from Temporary to PermanentWork in Spain.” Labour Economics, 14, 2007, 153–83.

Guell, M., and J. V. Rodrıguez-Mora. “Temporary Con-tracts, Incentives and Unemployment.” Centre for Eco-nomic Policy Research, Discussion paper No. 8116,2010.

Hamermesh, D. Labor Demand. Princeton: Academic Press,1993.

Heckman, J. J., and C. Pages. “Law and Employment:Lessons from Latin America and the Caribbean,” Intro-duction in Law and Employment: Lessons from LatinAmerica and the Caribbean, edited by J. J. Heckmanand C. Pages. Chicago: University of Chicago Press,2004.

Hopenhayn, H., and R. Rogerson. “Job Turnover and PolicyEvaluation: A General Equilibrium Analysis.” Journalof Political Economy, 101, 1993, 915–38.

Hunt, J. “Firing Costs, Employment Fluctuations and Aver-age Employment: An Examination of Germany.” Eco-nomica, 67, 2000, 177–202.

Kugler, A. “The Effect of Job Security Regulations on LaborMarket Flexibility: Evidence from the ColombianLabor Market Reform,” in Law and Employment:Lessons from Latin America and the Caribbean, editedby J. J. Heckman and C. Pages. Chicago: Universityof Chicago Press, 2004.

Kugler, A., and G. Pica. “Effects of Employment Protectionon Worker and Job Flows: Evidence from the 1990Italian Reform.” Labour Economics, 15, 2008, 78–95.

Lazear, E. “Job Security Provisions and Employment.”Quarterly Journal of Economics, 105, 1990, 699–726.

McFadden, D., and K. Train. “Mixed MNL Models forDiscrete Response.” Journal of Applied Econometrics,15(5), 2000, 447–70.

Olley, S., and A. Pakes. “The Dynamics of Productivity inthe Telecommunications Equipment Industry.” Econo-metrica, 64, 1996, 1263–97.

Pfann, G., and F. Palm. “Asymmetric Adjustment Costs inNon-Linear Labour Demand Models for the Nether-lands and the UK Manufacturing Sectors.” Review ofEconomic Studies, 60, 1993, 397–412.

Rota, P. “Estimating Labor Demand with Fixed Costs.”International Economic Review, 45, 2004, 25–48.

Sargent, T. “Estimation of Dynamic Labor Demand Sched-ules under Rational Expectations.” Journal of PoliticalEconomy, 86, 1978, 1009–44.

Tauchen, G. “Finite State Markov-Chain Approximationsto Univariate and Vector Autoregressions.” EconomicsLetters, 20, 1986, 177–81.

Tauchen, G., and R. Hussey. “Quadrature-Based Meth-ods for Obtaining Approximate Solutions to Nonlin-ear Asset Pricing Models.” Econometrica, 59, 1991,371–96.

Veracierto, M. “Employment Flows, Capital Mobility, andPolicy Analysis,” International Economic Review,42(3), 2001, 571–95.