Ricardo J. Caballero Kevin N. Cowan Eduardo MRA Engel Alejandro Micco January 2005

31 Pages·2005·0.16 MB·English

EFFECTIVE LABOR REGULATION
AND MICROECONOMIC FLEXIBILITY∗
Ricardo J. Caballero
Kevin N. Cowan
Eduardo M.R.A. Engel
Alejandro Micco
January 2005
Abstract
Microeconomicflexibilityisatthecoreofeconomicgrowthinmodernmarketeconomiesbecauseit
facilitatestheprocessofcreative-destruction,Themainreasonwhythisprocessisnotinfinitelyfast,is
thepresenceofadjustmentcosts,someofthemtechnological,othersinstitutional.Chiefamongthelatter
islabormarketregulation. Whilefeweconomistsobjecttothehypothesisthatlabormarketregulation
hinderstheprocessofcreative-destruction,itsempiricalsupportislimited. Inthispaperwerevisitthis
hypothesis,usinganewsectoralpanelfor60countriesandamethodologysuitableforsuchapanel. We
findthatjobsecurityregulationclearlyhampersthecreative-destructionprocess,especiallyincountries
whereregulationsarelikelytobeenforced.Movingfromthe20thtothe80thpercentileinjobsecurity,in
countrieswithstrongruleoflaw,cutstheannualspeedofadjustmenttoshocksbyathirdwhileshaving
off about one percent from annual productivity growth. The same movement has negligible effects in
countrieswithweakruleoflaw.
JELCodes: E24,J23,J63,J64,K00.
Keywords: Microeconomicrigidities,creative-destruction,jobsecurityregulation,adjustmentcosts,
ruleoflaw,productivitygrowth.
∗Respectively:MITandNBER;Inter-AmericanDevelopmentBank;YaleUniversityandNBER;Inter-AmericanDevelopment
Bank.WethankJosephAltonji,JohnHaltiwanger,MichaelKeaneandNormanLoayzaforusefulcomments.Caballerothanksthe
NSFforfinancialsupport.
1 Introduction
Microeconomicflexibility, byfacilitatingtheongoingprocessofcreative-destruction, isatthecoreofeco-
nomic growth in modern market economies. This basic idea has been with economists for centuries, was
broughttotheforebySchumpeterfiftyyearsago,andhasrecentlybeenquantifiedinawidevarietyofcon-
texts.1 InUSManufacturing, forexample, morethanhalfofaggregateproductivitygrowthcanbedirectly
linkedtothisprocess.2
The main obstacle faced by microeconomic flexibility is adjustment costs. Some of these costs are
purely technological, others are institutional. Chief among the latter is labor market regulation, in partic-
ular job security provisions. The literature on the impact of labor market regulation on the many different
economic,politicalandsociologicalvariablesassociatedtolabormarketsandtheirparticipantsisextensive
and contentious. However, the proposition that job security provisions reduce restructuring is a point of
agreement.
Despitethisconsensus,theempiricalevidencesupportingthenegativeimpactoflabormarketregulation
on microeconomic flexibility has been scant at best. This is not too surprising, as the obstacles to empiri-
cal success are legions, including poor measurement of restructuring activity and labor market institutions
variables, both within a country and more so across countries.3 In this paper we make a new attempt. We
develop a methodology that allows us to bring together the extensive new data set on labor market regula-
tion constructed by Botero et al. (2004) with comparable cross-country cross-sectoral data on employment
and output from the UNIDO (2002) data-set. We also emphasize the key distinction betweeneffective and
officiallabormarketregulation.
Themethodologybuildsonthesimplepartial-adjustmentideathatlargeradjustmentcostsarereflected
in slower employment adjustment to shocks.4 The accumulation of limited adjustment to these shocks
builds a wedge between frictionless and actual employment, which is the main right hand side variable in
thisapproach. Weproposeanewwayofestimatingthiswedge,whichallowsustopooldataonlabormarket
legislationwithcomparableemploymentandoutputdataforabroadrangeofcountries. Asaresult,weare
able to enlarge the effective sample to 60 economies, more than double the country coverage of previous
studies in this literature.5 Our attempt to measure effective labor regulation interacts existing measures of
jobsecurityprovisionwithmeasuresofruleoflawandgovernmentefficiency.6
1See,e.g.,thereviewinCaballeroandHammour(2000).
2See,e.g.,Foster,HaltiwangerandKrizan(1998).
3Onacloselyrelatedliterature,thereisanextensivebodyofempiricalwork,pioneeredbyLazear(1990),thathasputtogether
dataonjobsecurityprovisionsacrosscountriesandovertime,andmeasuredtheeffectoftheseprovisionsonaggregateemployment.
ArecentsurveyofthisliteraturecanbefoundinHeckmanandPages(2003). Resultsaremixed. Ontheonehand,Lazear(1990),
Grubb and Wells (1993), Nickell (1997) and Heckman and Pages (2000) find a negative relationship between job security and
employment levels. On the other hand Garibaldi and Mauro (1999), OECD (1999), Addison, Texeira and Grosso (2000), and
Freeman(2001)failtofindevidenceofsucharelationship.
4Forsurveysoftheempiricalliteratureonpartial-adjustmentseeNickell(1986)andHammermesh(1993).
5Toourknowledge,thebroadestcross-countrystudytodate–NickellandNuziata(2000)–included20highincomeOECD
countries. Other recent studies, such as Burgess and Knetter (1998) and Burgess et al. (2000), pool industry-level data from 7
OECDeconomies.
6SeeLoboguerreroandPanizza(2003)forasimilarinteractiontermintheirstudyoftherelationbetweenlabormarketinstitu-
1
Ourresultsareclearandrobust: countrieswithlesseffectivejobsecuritylegislationadjustmorequickly
to imbalances between frictionless and actual employment. In countries with strong rule of law, moving
fromthe20thtothe80thpercentileofjobsecuritylowersthespeedofadjustmenttoshocksby35percent
andcutsannualproductivitygrowthby0.85percent. Thesamemovementforcountrieswithlowruleoflaw
onlyreducesthespeedofadjustmentbyapproximately1percentandproductivitygrowthby0.02percent.
The paper proceeds as follows. Section 2 presents the methodology and describes the new data set.
Section3discussesthemainresultsandexplorestheirrobustness. Section4gaugestheimpactofeffective
laborprotectiononproductivitygrowth. Section5concludes.
2 Methodology and Data
2.1 Methodology
2.1.1 Overview
The starting point for our methodology is a simple adjustment hazard model, where the change in the
number of (filled) jobs in sector j in country c between time t−1 and t is a probabilistic (at least to the
econometrician)functionofthegapbetweendesiredandactualemployment:
D e =y Gap Gap ≡e∗ −e , (1)
jct jct jct jct jct jc,t−1
wheree ande∗ denotethelogarithmofemploymentanddesiredemployment,respectively. Therandom
jct jct
variabley ,whichisassumedi.i.d.bothacrosssectorsandovertime,takesvaluesintheinterval[0,1]and
jct
hascountry-specificmeanl andvariancez l (1−l ),with0≤z ≤1. Thismodelcanbeobtainedfroma
c c c c c
generalizationofSargent(1978)andCalvo(1983)(seebelow). Thecasez =0correspondstothestandard
c
quadraticadjustmentmodelasinSargent(1978),thecasez =1totheCalvo(1983)model. Theparameter
c
l captures microeconomic flexibility. As l goes to one, all gaps are closed quickly and microeconomic
c c
flexibilityismaximum. Asl decreases,microeconomicflexibilitydeclines.
c
Equation (1) hints at two important components of our methodology: We need to find a measure of
the employment gap and a strategy to estimate the average (over j and t) speeds of adjustment (the l ).
c
We describe both ingredients in detail in what follows. In a nutshell, we construct estimates of e∗ , the
jct
only unobserved element of the gap, by solving the optimization problem of a sector’s representative firm,
as a function of observables such as labor productivity and a suitable proxy for the average market wage.
We estimate l from (1), based upon the large cross-sectional size of our sample and the well documented
c
heterogeneity in the realizations of the gaps and the y ’s (see, e.g., Caballero, Engel and Haltiwanger
jct
(1997)forUSevidence).
tionsandinflation.
2
2.1.2 Details
Asector’srepresentativefirmhasoutputanddemand:
y = a+a e+b h, (2)
1
p = d− y, (3)
h
wherey, p,e,a,h,d denoteoutput,price,employment,productivity,hoursworkedanddemandshocks,and
h isthepriceelasticityofdemand. Weletg ≡(h −1)/h ,withh >1,0<a <1and0<b <1. Allvariables
areinlogs.
Firmsarecompetitiveinthelabormarketbutpaywagesthatincreasewiththenumberofhoursworked,
H:
w=ko+log(Hµ+W ).
Thiscanbeapproximatedby:
w=wo+µ(h−h), (4)
withwo determinedbyko andW ,andhconstantovertimeandinterpretedbelow. Inordertoensureinterior
solutions,weassumea µ>b andµ>bg .
Akeyassumptionisthattherepresentativefirmwithineachsectoronlyfacesadjustmentcostswhenit
changesemploymentlevels,notwhenitchangesthenumberofhoursworked(beyondovertimepayments).7
It follows that the sector’s choice of hours in every period can be expressed in terms of its current level of
employment,bysolvingthecorrespondingfirstorderconditionforhours.
Inafrictionlesslabormarketthefirm’semploymentlevelalsosatisfiesasimplestaticfirstordercondi-
tionforemployment. Ourfunctionalformsthenimplythattheoptimalchoiceofhours,h,doesnotdepend
ontheemploymentlevel. Apatientcalculationshowsthat
(cid:181) (cid:182)
1 b W
h= log .
µ a µ−b
Wedenotethecorrespondingemploymentlevelbye(cid:98)andrefertoitasthestaticemploymenttarget:
1
e(cid:98)=C+ [d+g a−wo],
1−ag
withCaconstantthatdependsonµ,a ,b andg .
In the absence of adjustment costs the firm’s cash flow, R, is maximized at e(cid:98), taking the value R(cid:98). A
second order Taylor approximation of the firm’s revenue function, net of adjustment costs, around e(cid:98)then
yields
R∼=R(cid:98)−C(cid:48)(e−e(cid:98))2, (5)
7ForevidenceonthisseeSargent(1978)andShapiro(1986).
3
withR(cid:98)unaffectedbythefirms’choicevariables. WithoutlossofgeneralitywesetC(cid:48)=1inwhatfollows.
Firms’ labor adjustment costs are assumed quadratic, with a stochastic proportionality factor k. The
k’s are independent (over time and across sectors within a country), identically distributed, and take both
the value zero and infinity with positive probabilities, thereby allowing for both smooth and lumpy labor
adjustments. Moreprecisely:

 0 withprob.p
 0
k = K withprob.p (6)
t  k

¥ withprob.p ¥
whereK afixednumber,0<K <¥ ,p i≥0,i=0,k,¥ ,andp 0+p k+p ¥ =1.8
Thefirm’sprofitmaximizationproblemattimet thenisequivalentto:
(cid:34) (cid:35)
(cid:169) (cid:170)
minE (cid:229) r j (e −e(cid:98) )2+k (e −e )2 , (7)
t t+j t+j t+j t+j t+j−1
et j≥0
withr denotingthefirm’sdiscountfactor. InAppendixAwesolvethecorrespondingBellmanequationand
showthatthefirm’soptimalemploymentchoicesatisfies
D e =y (e∗−e ), (8)
t t t t−1
withthedynamicemploymenttarget,e∗,definedvia
t
e∗=(1−t )(cid:229) t jE [e(cid:98) ],
t t t+j
j≥0
forsomeconstantt ∈(0,1),and

 0 ifk =¥
 t
y ≡y (k )= n ifk =K (9)
t t  t

1 ifk =0
t
withn ∈(0,1)anexplicitfunctionofK,r ,p 0,p k andp ¥ .
Itfollowsthatafractionp ¥ ofthetimetherepresentativefirmdoesnotadjustitsemployment,afraction
p itadjustsfullytoe∗ andtheremainingperiods,itclosespartofthegapbetweenitsdynamicemployment
0
targetandactualemployment.
Denotingl ≡p +np wehavethat:
0 k
E[y ] = l ,
Var[y ] = l (1−l )−p n (1−n ),
k
with Var[y ] taking values between 0 (quadratic adjustment: p = 1, n = l ) and l (1−l ) (Calvo model:
k
8TheresultsthatfollowcanbeextendedtothecasewhereKisdrawnfromadistributionthattakespositivevalues.
4
p = 0, p = l ). Furthermore, as p decreases from one to zero, Var[y ] covers the full range of values
k 0 k
betweenl (1−l )andzero.
Having derived our estimating equation from first principles, we next turn to deriving a proxy for the
dynamic employment target e∗. For this, note that the relation between the employment gap and the hours
gapfollowsfromtheexpressionsobtainedabovefore(cid:98),h¯ andthefirstorderconditionsatisfiedbyh:
µ−bg
e(cid:98)−e = (h−h). (10)
1−ag
This is the expression used by Caballero and Engel (1993). It is not useful in our case, since we do
not have information on worked hours. Yet the argument leading to (10) also can be used to express the
employmentgapintermsofthemarginallaborproductivitygap:
f
e(cid:98)−e = (v−wo),
1−ag
where v denotes marginal productivity, f ≡µ/(µ−bg ) is decreasing in the elasticity of the marginal wage
schedule with respect to average hours worked, µ−1, and wo was defined in (4). Note that e(cid:98)−e is the
differencebetweenthestatictargete(cid:98)andrealizedemployment,notthedynamicemploymentgape∗ −e
jct jct
relatedtothetermontherighthandsideof(1). However, ifweassumethatd+g a−wo followsarandom
walk (possibly with an exogenously time varying drift) —an assumption consistent with the data9 — we
havethate∗ isequaltoe(cid:98) plusaconstantd . Itfollowsthat
jct jct ct
f (cid:161) (cid:162)
e∗ −e = v −wo +D e +d , (11)
jct jct−1 1−ag jct jct jct ct
j
wherewehaveallowedforsector-specificdifferencesinag . Notethatbothmarginalproductandwagesare
in nominal terms. However, since these expressions are in logs, their difference eliminates the aggregate
pricelevelcomponent.
Weestimatethemarginalproductivityoflabor,v ,usingoutputperworkermultipliedbyanindustry-
jct
levellaborshare,assumedconstantwithinincomegroupsandovertime.
Two natural candidates to proxy for wo are the average (across sectors within a country, at a given
jct
pointintime)ofeitherobservedwagesorobservedmarginalproductivities. Theformerisconsistentwitha
competitivelabormarket,thelattermaybeexpectedtobemorerobustinsettingswithlong-termcontracts
andmultipleformsofcompensation,wherethesalarymaynotrepresenttheactualmarginalcostoflabor.10
We performed estimations using both alternatives and found no discernible differences (see below). This
suggeststhatstatisticalpowercomesmainlyfromthecross-sectiondimension,thatis,fromthewelldocu-
9Poolingallcountriesandsectorstogether,thefirstorderautocorrelationofthemeasureofD e∗ constructedbelowis−0.018.
jct
Computingthiscorrelationbycountrythemeanvalueis0.011withastandarddeviationof0.179.
10While we have assumed a simple competitive market for the base salary (salary for normal hours) within each sector, our
procedurecouldeasilyaccommodateother, morerent-sharinglike, wagesettingmechanisms(withasuitablereinterpretationof
someparameters,butnotl c).
5
mentedandlargemagnitudeofsector-specificshocks. Inwhatfollowswereportthemorerobustalternative
andapproximatewo bytheaveragemarginalproductivity,whichleadsto:
f
e∗ −e = (v −v )+D e +d ≡ Gap +d , (12)
jct jct−1 1−ag jct ·ct jct ct jct ct
j
where v denotes the average, over j, of v , and we use this convention for other variables as well. The
·ct jct
expression above ignores systematic variations in labor productivity across sectors within a country, for
example, because (unobserved) labor quality may differ systematically across sectors. The presence of
such heterogeneity would tend to bias estimates of the speed of adjustment downward. To incorporate this
possibility we subtract from (v −v ) in (12) a moving average of relative sectoral productivity, (cid:98)q ,
jct ·ct jct
where
1
(cid:98)q ≡ [(v −v ) + (v −v )].
jct jct−1 ·ct−1 jct−2 ·ct−2
2
As a robustness check, for our main specifications we also computed (cid:98)q using a three and four periods
jct
moving average, without significant changes in our results (more on this when we check robustness in
Section3.2). Theresultingexpressionfortheestimatedemployment-gapis:
f
e∗ −e = (v −(cid:98)q −v )+D e +d ≡ Gap +d , (13)
jct jct−1 1−ag jct jct ·ct jct ct jct ct
j
where ag is constructed using the sample median of the labor share for sector j across year and income
j
groups.
Rearranging(13),weestimatef from
f
D e = − (D v −D v )+k +u +D e∗ ≡ −f z +k +e , (14)
jct 1−ag jct ·ct ct it jct jct ct jct
j
where k is a country-year dummy, D e∗ is the change in the desired level of employment and z ≡
jct jct
(D v −D v )/(1−ag ). We assume that changes in sectoral labor composition are negligible between
jct ·ct j
two consecutive years. In order to avoid the simultaneity bias present in this equation (D v and D e∗ are
clearlycorrelated)weestimate(14)using(D w −D w )asaninstrumentfor(D v −D v ).11
jct−1 ·ct−1 jct ·ct
Table1reportstheestimationresultsof(14)forthefullsampleofcountriesandacrossincomeandjob
securitygroups. Thefirsttwocolumnsusethefullsample,withandwithouttwopercentofextremevalues
for the independent variable, respectively. The remaining columns report the estimation results for each of
ourthreeincomegroupsandjobsecuritygroups(moreonbothofthesemeasuresinSection2.2). Basedon
our results for the baseline case, we set the value of f at its full sample estimate of 0.4 for all countries in
oursample.
Itisimportanttopointoutthatourmethodologyhassomeadvantagesoverstandardpartialadjustment
11Welagtheinstrumenttodealwiththesimultaneityproblemandusethewageratherthanproductivitytoreducethe(potential)
impactofmeasurementerrorbias.
6
Table1: ESTIMATING f
Specification: (1) (2) (3) (4) (5) (6) (7) (8)
ChangeinEmployment(ln)
z −0.280 −0.394 −0.558 −0.355 −0.387 −0.363 −1.168 −0.352
jct
(0.044) (0.068) (0.135) (0.119) (0.116) (0.091) (.357) (0.103)
Observations 22,810 22,008 8,311 6,378 7,319 7,730 6,883 7,036
IncomeGroup All All 1 2 3 All All All
JobSec.Group All All All All All 1 2 3
Extremeobs.ofinstrument Yes No No No No No No No
Standard errors reported in parentheses. All estimates are significant at the 1% level. All regressions use lagged D wict−D w·ct
as instrumental variable. As described in the main text, zjct represents the log-change of the nominal marginal productivity of
laborineachsector,minusthecountryaverage,dividedbyoneminustheestimatedlaborshare. Allregressionsdisregardthe2%
observationswithmostextremechangeinemploymentvaluesandincludeacountry-yearfixedeffect(k ct in(14)).Incomegroups
are1:HighIncomeOECD,2:HighIncomeNonOECDandUpperMiddleIncome,and3:LowerMiddleIncomeandLowIncome.
JobSecurityGroupscorrespondtothehighest,middleanlowestthirdofthemeasureinBoteroetal.(2004).
estimations. First, itsummarizesina singlevariableallshocks facedbyasector. Thisfeature allowsusto
increaseprecisionandtostudythedeterminantsofthespeedofadjustmentusinginteractionterms. Second,
and related, it only requires data on nominal output and employment, two standard and well-measured
variables in most industrial surveys. Most previous studies on adjustment costs required measures of real
outputoranexogenousmeasureofsectordemand.12
2.1.3 Regressions
Thecentralempiricalquestionofthepresentstudyishowcross-countrydifferencesinjobsecurityregulation
affectthespeedofadjustment. Accordingly,from(1)and(13)itfollowsthatthebasicequationweestimate
is:
D e = l (Gap +d ), (15)
jct ct jct ct
where D e is the log change in employment andl denotes the speed of adjustment. We assume that the
jct ct
lattertakestheform:
l = l˜ +l˜ JSeff, (16)
ct 1 2 ct
where JSeff is a measure of effective job security regulation. In practice we observe job security regulation
ct
(imperfectly),butnottherigorwithwhichitisenforced. Weproxythelatterwitha“ruleoflaw”variable,
12AbrahamandHouseman(1994),Hammermesh(1993),andNickelandNunziata(2000))evaluatethedifferentialresponseof
employmenttoobservedrealoutput.Asecondoptionistoconstructexogenousdemandshocks.Althoughthisapproachovercomes
therealoutputconcerns, itrequiresconstructinganadequatesectorialdemandshockforeverycountry. Acaseinpointarethe
papers by Burgessand Knetter (1998) and Burgesset al. (2000), which use the real exchangerate as their demand shock. The
estimatedeffectsoftherealexchangerateonemploymentareusuallymarginallysignificant,andoftenoftheoppositesignthan
expected.
7
sothat
JSeff=aJS +b(JS RL ), (17)
ct ct ct ct
whereaandbareconstantsandRL isastandardmeasureofruleoflaw(seebelow). Whenb=0thereis
ct
nodifferencebetweendejureanddefactoregulation. Substitutingthisexpressionin(16)andtheresulting
expressionforl in(15),yieldsourmainestimatingequation:
ct
(cid:161) (cid:162) (cid:161) (cid:162)
D e = l Gap +l Gap ×JS +l Gap ×JS ×RL +(cid:101)d +e , (18)
jct 1 jct 2 jct ct 3 jct ct ct ct jct
withl =l˜ ,l =al˜ ,l =bl˜ ,and(cid:101)d denotescountry×timefixedeffects(proportionaltothed defined
1 1 2 2 3 3 ct ct
above).
Themaincoefficientsofinterestarel andl ,whichmeasurehowthespeedofadjustmentvariesacross
2 3
countriesdependingontheirlabormarketregulation(bothdejureanddefacto).
2.2 TheData
Thissectiondescribesoursampleandmainvariables. Additionalvariablesaredefinedasweintroducethem
laterinthetext.
2.2.1 JobSecurityandRuleofLaw
Weusetwomeasuresofjobsecurity,orlegalprotectionagainstdismissal: thejobsecurityindexconstructed
byBoteroetal.(2004)for60countriesworld-wide(henceforthJS )andthejobsecurityindexconstructed
c
by Heckman and Pages (2000) for 24 countries in OECD and Latin America (henceforth HP ). The JS
ct c
measureisavailableforalargersampleofcountriesandincludesabroaderrangeofjobsecurityvariables.
TheHP measurehastheadvantageofhavingtimevariation.
ct
Our main job security index, JS , is the sum of four variables, measured in 1997, each of which takes
c
on values between 0 and 1: (i) grounds for dismissal protection PG , (ii) protection regarding dismissal
c
proceduresPP,(iii)noticeandseverancepaymentsPS ,and(iv)protectionofemploymentintheconstitu-
c c
tionPC . Therulesongroundsofdismissalrangefromallowingtheemploymentrelationtobeterminated
c
byeitherpartyatanytime(employmentatwill)toallowingtheterminationofcontractsonlyunderavery
narrow list of “fair” causes. Protective dismissal procedures require employers to obtain the authorization
of third parties (such as unions and judges) before terminating the employment contract. The third vari-
able, notice and severance payment, is the one closest to the HP measure, and is the normalized sum of
ct
twocomponents: mandatoryseverancepaymentsafter20yearsofemployment(inmonths)andmonthsof
advance notice for dismissals after 20 years of employment(NS =b +SP , t =1997). The four
tc ct+20 ct+20
componentsofJS describedaboveincreasewiththelevelofjobsecurity.
c
TheHeckmanandPagesmeasureisnarrower,includingonlythoseprovisionsthathaveadirectimpact
on the costs of dismissal. To quantify the effects of this legislation, they construct an index that computes
8
the expected (at hiring) cost of a future dismissal. The index includes both the costs of advanced notice
legislationandfiringcosts,andismeasuredinunitsofmonthlywages.
Our estimations also adjust for the level of enforcement of labor legislation. We do this by including
measuresofruleoflawRL andgovernmentefficiencyGE fromKaufmannatal.(1999),andinteractthem
c c
with JS and HP .13 We expect labor market legislation to have a larger impact on adjustment costs in
c ct
countrieswithastrongerruleoflaw(higherRL )andmoreefficientgovernments(higherGE ).
c c
Theinstitutionalvariablesaswellasthecountriesinoursampleandtheircorrespondingincomegroup
are reported in Table 2. Table 3 reports the sample correlations between our main cross-country variables
andsummarystatisticsforeachofthesemeasuresforthreeincomegroups(basedonWorldBankpercapita
incomecategories).14 Asexpected,thecorrelationbetweenthetwomeasuresofjobsecurityispositiveand
significant. Differences can be explained mainly by the broader scope of theJS index. Also as expected,
ct
ruleoflawandgovernmentefficiencyincreasewithincomelevels. Note, however, thatneithermeasureof
job security is positively correlated with income per capita, since bothJS and HP are highest for middle
ct c
incomecountries.
2.2.2 IndustrialStatistics
Ouroutput,employmentandwagedatacomefromthe20023-digitUNIDOIndustrialStatisticsDatabase.
The UNIDO database contains data for the period 1963-2000 for the 28 manufacturing sectors that corre-
spond to the 3 digit ISIC code (revision 2). Because our measures of job security and rule of law are time
invariant and measured in recent years, however, we restrict our sample to the period 1980-2000. Data on
output and labor compensation are in current US dollars (inflation is removed through time effects in our
regressions). ThroughoutthepaperourmaindependentvariableisD e ,thelogchangeintotalemployment
jct
insector jofcountrycinperiodt.
A large number of countries are included in the original dataset — however our sample is constrained
by the cross-country availability of the independent variables measuring job security. In addition, we drop
twopercentofextremeemploymentchangesineachofthethreeincomegroups. Forourmainspecification
theresultingsampleincludes60economies. Table3showsdescriptivestatisticsforthedependentvariable
byincomegroup.
3 Results
Thissectionpresentsourmainresult,showingthateffectivejobsecurityhasasignificantnegativeeffecton
thespeedofadjustmentofemploymenttoshocksintheemployment-gap. Italsopresentsseveralrobustness
exercises.
13ForruleoflawandgovernmentefficiencyweusetheearliestvalueavailableintheKaufmannetal.(1999)database: 1996,
sincethisisclosesttotheBoteroetal.(2004)measure,whichisfor1997.
14Incomegroupsare:1=HighIncomeOECD,2=HighIncomeNonOECDandUpperMiddleIncome,3=LowerMiddleIncome
andLowIncome.
9