Unemployment and the Social Safety Net During Transitions to a Market Economy Evidence From the...

8/3/2019 Unemployment and the Social Safety Net During Transitions to a Market Economy Evidence From the Czech and Sl

1/27

American Economic Association

Unemployment and the Social Safety Net during Transitions to a Market Economy: Evidencefrom the Czech and Slovak RepublicsAuthor(s): John C. Ham, Jan Svejnar, Katherine TerrellReviewed work(s):Source: The American Economic Review, Vol. 88, No. 5 (Dec., 1998), pp. 1117-1142Published by: American Economic AssociationStable URL: http://www.jstor.org/stable/116863 .

Accessed: 22/12/2011 05:58

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of

content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms

of scholarship. For more information about JSTOR, please contact [email protected].

American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The

American Economic Review.

http://www.jstor.org
http://www.jstor.org/action/showPublisher?publisherCode=aeahttp://www.jstor.org/stable/116863?origin=JSTOR-pdfhttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/stable/116863?origin=JSTOR-pdfhttp://www.jstor.org/action/showPublisher?publisherCode=aea


2/27

Unemploymentand the Social Safety Net DuringTransitionsto a MarketEconomy:Evidence from the Czechand Slovak RepublicsBy JOHN C. HAM, JAN SVEJNAR, AND KATHERINETERRELL

We investigate the remarkablyshort unemployment pells in the Czech Republiccompared to Slovakia and other Central and East European economies. We es-timate hazard functions and find that 40 to 50 percent of the difference in un-employmentdurations between the two republics is accountedfor by differencesin demographics and demand conditions. The remainderis explained by differ-ences in coefficients,proxying the behavioroffirms, individuals,and institutions.In both republics the unemployment ompensationsystem has a moderately neg-ative effect on the exit ratefrom unemployment.Policy makershence have lati-tude in providing adequate social safety nets without jeopardizing efficiency.(JEL C41, H53, J23, J64, 015, P2)The Central and East European (CEE)countries are completing the first decade of a dramatic transition from a centrally plannedeconomy to a market system. Although eco-nomic outcomes have been diverse, all CEEcountries (except for the Czech Republic)have experienced rapidly rising and persist-ently high unemployment rates, which havebeen accompanied by long spells of unem-ployment. By contrast, in the Czech Republicthe unemployment rate has remainedlow andunemployment spells have been short.The unemployment crisis in the CEE coun-tries has contributedto a political backlash asdisenchantedvoters often ousted the first re-form governments after a few years.1This ex-perience underscores the importance of twoseparate questions: (1) Why has the unem-ployment problem associated with the transi-tion to market economies been much lesssevere in the Czech Republic?; and (2) Howcan economies in transition strike a balancebetween (i) reducing governmentinterventionand introducing market incentives, and (ii)

* Ham: Departmentof Economics, 4S01 Forbes Quad,University of Pittsburgh,Pittsburgh,PA 15260; Svejnar:William DavidsonInstitute,Universityof Michigan Busi-ness School, 9th Floor Davidson Hall, University of Mich-igan, 701 Tappan, Ann Arbor, MI 48109, and CERGE-EI,Prague;Terrell: WilliamDavidson Institute,UniversityofMichigan Business School, 7th Floor Davidson Hall, Uni-versity of Michigan, 701 Tappan, Ann Arbor, MI 48109.The paper benefited from the comments of two anony-mous referees, and from presentationsat Brigham YoungUniversity, CERGE-EI,CornellUniversity,Duke Univer-sity, McGill University, McMaster University, the MilkenInstitute,the University of Michigan, New York Univer-sity, Pittsburgh/Carnegie Mellon Universities, PrincetonUniversity,theUniversityof California-Irvine,UCLA, theUniversity of Southern California, the University ofTexas-Austin, the 1993 NEUDC conference at Yale Uni-versity, the 1995 AEA and EEA meetings, and the 1993World Bank Workshopon LaborMarketsin EasternEu-rope and Russia. The authors would like to thank StepanJurajda, Eileen Kopchik, and Hana Pessrova for excep-tional research assistance, and Carolyn Maguire, JanetNightingale, and Debbie Ziolkowski for assistance in pre-paring the manuscript.PatriciaBeeson, Michael Cragg,CurtisEberwein,JohnEngberg, ChristopherFlinn,DavidGreen, George Jakubson,Julia Lane, Robert Miller, andViktor Steiner provided many helpful suggestions. The au-thors would also like to acknowledge support from theNational Science Foundation(GrantsSBR-951-2001 andSES-921-3310), the National Council for Soviet and EastEuropeanResearch (Contract No. 806-34), and the GrantAgency of the Czech Republicto CERGE-EI.Jan Svejnarwould also like to acknowledge sunnort from ACE Grant

No. P96-6095-R. Finally, partof this researchwas carriedout while Ham was a visitor in the EconomicsDepartment,NorthwesternUniversity and in the Institute for PolicyAnalysis, University of Toronto. ile would like to thankboth institutions for theirsupportandhospitality.Any re-mainingerrors are our resnonsihilitv.'See Olivier J. Blanchard (1997) for a theoreticalmodel of the effect of worsening economic conditions onworkeropposition to enterpriserestructuring.1117


3/27

1118 THEAMERICANECONOMICREVIEW DECEMBER 1998providing an adequatesocial safety net that en-sures public support for the transition?In ad-dition to being of academic interest, answersto these two questions are essential for policymakersin the CEE countries, in Western gov-ernments,andat international nstitutionssuchas the World Bank and the InternationalMon-etary Fund.When addressing the first question, andcomparingthe Czech experience to thatof theother CEE countries, policy makers and re-searchers are hamperedby the difficulty in ac-counting for differences in relevant laws andinstitutions, andin the definitions of economicand demographicvariables. To minimize thisdifficulty,we collected parallelmicro data setsfrom the Czech and Slovak Republics. TheSlovak Republic (SR) is a natural "compari-son" countryfor the Czech Republic (CR) foraddressing this question because the two re-publics were one countryfrom 1918 until Jan-uary 1993 (except duringWorld War II). Asa result, the republics shared the same lawsand regulations, institutions, currency, andgovernment programs.Despite this common history, the two re-publics' labor markets have performed sub-stantially differently since the "VelvetRevolution" that overthrew the communistgovernment in November 1989. In January1990, the unemployment rates in both theCzech and Slovak regions of Czechoslovakiastood at 0.1 percent.However, as may be seenfrom Table 1, in 1991 the average unemploy-ment rate was 11.8 percent in the SR and by1996 it had increased to 12.8 percent of thelabor force. By contrast, in the CR, the un-employment rate rose to 3.7 percent in 1991,and by 1996 it had decreasedto 3.2 percentofthe labor force.2

As may also be seen from Table 1, all theCEE economies experienced similar declinesin GDP in the early 1990's. (Note that Po-land's transition,andhence its decline in GDPand rise in unemployment, started one yearearlier than in the other CEEs.) The statisticsin Table 1 also show that while the CR had alower inflow rate to unemployment than the

otherCEEs, the most importantreason for theCR's lower unemploymentrates was its con-siderably higher rate of outflow from unem-ployment.3 Hence, the Czech economy wasundergoing similar restructuringas the otherCEE countriesin terms of GDP decline, but itwas able to reemploy its unemployed at afaster rate than others. On the otherhand, Slo-vakia's unemployment rates and its rates ofinflow to, and outflow from, unemploymentwere quite similar to the other CEE countries.An analysis of the different transition ratesfrom unemploymentin the Czech and SlovakRepublics is therefore key to understandingthe unemployment problemin the CEE regionin general. From a longer-term perspective,this analysis is also useful vis-a-vis Russia andother newly independent states. These coun-tries launchedtheirtransitionsmuch later thanthe CEEcountries,but in recentyears they en-countered similarproblemsof high unemploy-ment ratescoupled with relatively low rates ofoutflow from unemployment.In orderto provide an understandingof thedifferences in labor-marketperformance be-tween the CR and SR (i.e., to answer the firstquestion), we estimate the relative effects ofvarious variables, including demographiccharacteristics, ocal labor demand conditions,and features of the unemployment compensa-tion system (UCS), on the probabilitythatanindividual leaves unemployment (the hazardfunction) in each republic.We derive and im-plement an Oaxaca-type decomposition of thedifference in the (nonlinear) expected unem-ployment durationsbetween the CR and SR inorder to determine which factors account forthe differences between the two countries.Fi--nally, we discuss additional factors that mayunderlie the decomposition and thus may beimportant or explaining the Czech andSlovakdifferences in unemployment.To answer the second question above andassess the disincentive effects of the UCS, we

2 See Ham et al. (1995) for a detailed discussion of theCzech and Slovak labor markets during the early part ofthe transition.

3 During the period of our study, 1991-1993, the av-erage CR and SR inflow rates were 0.8 and 1.3, respec-tively, while the corresponding outflow rates were 21.3and 7.6. Note that it is during this period thatthe unem-ployment rates diverged between the Czech Republic onone hand and Slovakia and the otherCEEcountrieson theotherhand.


4/27

VOL.88 NO. 5 HAM ET AL.: UNEMPLOYMENTN THE CZECHAND SLOVAKREPUBLICS 1119TABLE 1-MACROECONOMIC STATISTICS FOR SELECTED CENTRAL AND EAST EUROPEAN COUNTRIES

Unemployment Inflationrate GDP growth rate (CPI) Inflow rate Outflow rateBulgaria1990 1.5 -9.1 70b -1991 11.5 --11.7 339 - 7.01992 15.6 -7.3 79 1.7 9.21993 16.4 -2.4 64 1.4 6.81994 12.8 1.8 122 1.5 10.21995 10.5 2.1 33 1.5 11.61996 12.5 -9.0 123C 1.7 11.3Czech Republic1990 0.8 -1.2 lo-1991 4.1 --11.5 52 0.9 17.11992 2.6 -3.3 13 0.9 26.61993 3.5 0.6 18 0.7 22.01994 3.2 2.7 10 0.6 21.31995 2.9 5.9 8 0.6 21.31996 3.3 4.2 9 0.6 19.3Hungary1990 1.9 -3.5 29'1991 7.5 -11.9 32 -1992 12.3 -3.1 22 0.9 6.61993 12.1 -0.6 21 1.3 7.71994 10.4 2.9 21 1.1 9.11995 10.4 1.5 28 1.0 7.91996 10.5 1.0 24 1.3 9.4Poland1990 6. la--11.6 585' --1991 11.8a -7.0 60 --1992 13.6a 2.6 44 0.9 4.31993 15.7a 3.8 38 1.1 4.81994 16.Oa 5.2 29 1.2 6.21995 14.9a 7.0 22 1.3 8.01996 13.6a 5.5 20 1.2 8.2Slovak Republic1990 1.5 -2.5 10 --1991 11.8 -14.6 58 1.3 4.81992 10.3 -6.5 9 1.1 10.21993 14.4 -3.7 25 1.5 7.81994 14.8 4.9 12 1.3 7.41995 13.1 6.8 7 1.4 9.51996 12.8 7.0 6 1.4 10.0Notes: Inflow rates areaverageannualrates of the numberflowing into unemploymentin a month dividedby the numberemployed in a month, multiplied by 100;outflow rates are averageannualratesof the numberflowing out of unemploy-ment in a month divided by the numberunemployedin a month, multiplied by 100.Sources: Columns 1,4, and 5: OECD-CETLabourMarketDataBase;columns2 and 3:EuropeanBankforReconstructionTransitionReport 1997, except where noted below.a EuropeanBank for Reconstruction TransitionReport 1997 and TransitionReport Update, April 1997.'Retail tradeprice.EconomistIntelligence Unit, Bulgaria CountryReport (1st Quarter,1992 p. 5).cEconomistIntelligence Unit, Bulgaria CountryReport (4th Quarter,1997 p. 9). Percentchangein average consumerprices.'Economist Intelligence Unit, CzechoslovakiaCountryReport (1st Quarter,1992 p. 3).

e Economist Intelligence Unit, Hungary CountryReport (lst Quarter,1992 p. 3).f Economist Intelligence Unit, Poland CountryReport (lst Quarter,1992 p. 3).


5/27

1120 THEAMERICANECONOMICREVIEW DECEMBER1998use estimates of the hazard function for UCSrecipients to calculate the effect of marginalchanges in the UCS. Withineach republic,wealso compare the experience of recipientsandnonrecipientsof unemploymentbenefitsto ob-tain an alternative (inframarginal)measure ofthe impact of the UCS on the duration of un-employment in each country.4The average unemployment spell lasts threeto four times longer in Slovakia than in theCzech lands. Our firstprincipalfinding,basedon an Oaxaca-type decomposition, is that forrecipients nearly one-half of this difference isexplained by different values of the explana-tory variables in the two republics. The re-maining one-half is accounted for by thedifferentbehavior of firms, ndividuals,and in-stitutions in the labor market, as reflected bydifferences in the coefficients of the hazardfunctions. For nonrecipients,nearly 40 percentof the difference in expected durationbetweenthe two republics is due to differences in ex-planatoryvariables,and the remaining 60 per-cent is due to differences in coefficients. Animportant inding in this context is thattheCR,unlike the SR and perhapsother CEE econo-mies, was able to absorb the low-skilled un-employed into employment at a rate similar tothe rate it absorbed the skilled unemployed.Below we argue that the principal factors un-derlyingthe different coefficients of the hazardfunction are the faster growth of the servicesector,morerapidprivatization,greater nflowof foreign investment, lesser impactof the de-cline in military production, and stricter en-forcementof laborregulationsin the CR thanin the SR. Further,differences are also likelydue to the ages and locations of factoriesin thetwo republics,and the greateropportunities orthe Czechs thanthe Slovaks to work in neigh-boring Western economies. We find that

among the recipients of unemployment bene-fits, the contribution of the explanatory vari-ables comes almost entirely from differencesin demand conditions between the two repub-lics; relatively little of the difference in ex-pected unemployment duration comes fromdifferences in the demographic variables.However, for nonrecipients slightly more thanone-half of the contributionof the explanatoryvariables arises from differences in demo-graphic characteristics of this group betweenthe two republics.Withrespectto oursecond question, we findthatin both republicsthe unemployment com-pensation system has only a moderately neg-ative effect in terms of lengthening anunemployment spell.5 Thus policy makers inboth the low and high unemployment transi-tion economies have considerable latitude inproviding an adequate social safety net with-outjeopardizing efficiency. This finding is im-portant because the negative sociopoliticalbacklash to the transitionmeasures has beensignificant and an adequate social safety netmay be a prerequisite for rallying sufficientpopular supportto complete the transition.The paper is organized as follows. In Sec-tion I we describe our data, while in SectionII we outline the principalfeatures of the UCSin the CEE countries, and in the CR and SRin particular. n Section III we presentour es-timation strategy. In Section IV we first dis-cuss the determinants of the probability ofleaving unemployment in each republic, fo-cusing on the effects of the UCS, demographiccharacteristics, and demand conditions. Wethen comparethe expected durationsof recip-ients to nonrecipientsto obtain an alternativeestimate of the effects of the UCS. In SectionV we decompose the difference in the ex-pected duration of unemployment spells be-tween the CR and SR for both recipients andfor nonrecipients. Since the econometric re-sults point to differences in individual, firm,and institutionalbehavior,we then discuss theWe propose a new identificationstrategy hatprovidesrelatively precise estimates of the effects of the UCS onthe recipients in the two republics. Since the compensationschemes are similar across the CEE countries (see SectionIII), our approachshould be of general interest to thosestudying the UCS in the other CEE countries.For exam-ples of studies that examine the unemploymentdurationeffects of the UCSs in CEE countries, see JohnMicklewright and Gyula Nagy (1994), Jennifer Hunt(1995), Patrick Puhani (1996), and Joachim Wolfe

(1997).

5 Note that although the republics share the same UCS,individuals may respond differently to the system in thetwo republics. We find that the response to a change inunemploymentbenefits is lower in the SR thanin the CR,while the effect of an increase in entitlement is similaracross the two republics.


6/27

VOL.88 NO. 5 HAM ET AL.: UNEMPLOYMENTN THE CZECHAND SLOVAKREPUBLICS 1121factors underlyingthese differences. We con-clude the paper in Section VI.

I. TheDataFor this study, we collected dataon a strat-ified randomsample of 3,000 Czech and3,000Slovak men andwomen who registeredat theirdistrict labor offices as unemployed betweenOctober 1, 1991, and March 31, 1992.6 Wefollowed these individuals from the time oftheir registration o the end of theirunemploy-ment spell or the end of July 1993, whichevercame first. Since the labor-marketexperienceof men and women is likely to differ, in thispaper we focus on men's experience. More-

over, we selected individuals who were not inretraining,did not suffer a prolonged illness,and had no missing values.7This yielded dataon 780 men in the CR and 1,063 men in theSR who received unemploymentbenefits (re-cipients), and482 men in the CR and 229 menin the SR who did not receive benefits(nonrecipients).The basic sample statistics for each groupare given in Table B 1 in Appendix B. As maybe seen from the table, in the CR a recipienthad a 0.052 averageprobabilityof leaving un-employment for a job in a given week, whilethe average transitionrate for a nonrecipientwas 0.063. In the SR the weekly transitionrates from unemployment to employment forrecipients (0.020) and for nonrecipients(0.019) were much lower than those in theCR. The differences in the two labor marketsare also illustratedby the fact that in the CRonly 11.3 percent of the recipients, and 16.1percentof the nonrecipients,did not exit for ajob duringour sample period, while in the SRthe corresponding figures were 34.2 percentand 38.2 percent, respectively.8Moreover, a

much smaller proportion of recipients ex-hausted their benefits in the CR (0.135) thanin the SR (0.455). Finally, it should be notedthat the mean previous wages and unemploy-ment benefits differ only slightly, reflecting theinstitutional similarities across the CR andSR.(We discuss the remaining explanatoryvari-ables in the second half of the tablein SectionIII,when we describethe econometric model.)

II. Characteristicsf theUnemploymentCompensation ystemWith the advent of unemployment at thestart of the transition, all CEE countries de-signed and implemented unemploymentcom-

pensation systems. As they struggledto strikea balance between providing an adequateso-cial safety net and reducing governmentinter-vention while controllingtheirbudget deficits,these governments decided within one or twoyears to reduce the level of protection in un-employment. In this section we first show thatthe principal features of the UCSs in the CEEcountries have been quite similar,and thenwedescribe in more detail the UCS systems in theCzech and Slovak Republics. The material inthis section is important or understandingandmodeling the disincentive effects of the UCSand for assessing the applicabilityof our meth-odology and findings to other CEE countries.

A. The UnemploymentCompensationSystems in CEE CountriesBy the end of 1991, UCSs had been estab-lished in all CEE countries. We briefly high-light the principal features of these systemsfrom late 1991 to 1993, the period of ourstudy.9In terms of who was eligible to collect

unemployment compensation, all the CEEcountries requireda minimum period of pre-vious employment that ranged from sixmonths during the preceding year (Bulgaria,Poland, and Romania) to one year in the pre-ceding three years (CR, SR, and Hungary).The minimumworking period was waived for6During thisperiod therewere 78 districts n theCzechRepublicand 38 districts in the Slovak Republic. We firstrandomly selected 20 districtsin each of the two republicsand then randomlyselected 150 individuals n eachdistrictlaboroffice.7Within our sample, 42 individuals in the CR and 43individuals in the SR entered training.This group is toodifferent from the rest of the sample to justify inclusionand too small to warrant eparateestimation.8 Note that the probability of leaving unemploymentrefers to a given week, while the percentageof recipients

that did not exit for a job refersto the sample period,whichcould be as long as 1.75 years.9 See Organization or Economic Cooperationand De-

velopment (OECD) (1995) for furtherdetails.


7/27

1122 THE AMERICANECONOMICREVIEW DECEMBER 1998new graduates. A second similarity is that thelength of time over which unemploymentcompensation could be collected did not varyacross workers. This was true in the CR andSR, as well as in Albania, Poland, and Ro-mania.10Third, in all of the CEE countries ex-cept Albania, the levels of unemploymentbenefits were based on fixed replacementratesof previous wages that did not vary withworkercharacteristics.Fourth, except for Bul-garia and Poland, these replacementrates fellover the entitlement period for all workers.Fifth, all the CEE countries (except Poland)imposed a similarly low maximum level ofbenefits (between 1.4 and 2.0 times the mini-mum wage). Finally, therewas no indexationof benefits for inflation in any of the CEEs.

B. The Czech and Slovak UnemploymentCompensation SystemIn January 1990, Czechoslovakia intro-ducedUCS thatwas liberal when comparedtothe UCS in the United States, but not whencomparedto West Europeansystems. As in theother CEEs, the governmentsoon began to re-duce benefits and tighten entitlement provi-sions. In this section we focus our discussion

on the three main featuresof the system-el-igibility, entitlement,andbenefits-during theperiod covered by our study. Except for onefeature noted below, the system was essen-tially identical in the CR and SR throughtheend of 1993 (i.e., even after the "Velvet Di-vorce" between the two republics in January1993).Eligibility. -Anyone who worked for atleast 12 months in the preceding three yearswas immediately eligible for unemployment

benefits, unless the person was firedfor causeor quit repeatedly.11Students at the time oftheirgraduation romhigh school oruniversitywere also eligible. Until January 1992, indi-viduals out of the lahor force were eigihble if

they had cared for a young child or a sick/disabledrelative,orif they were in themilitaryor imprisoned, for 12 months in the previousthree years. After January 1992, individualswho were out of the laborforce and not takingcare of children were no longer eligible.

Entitlement. In 1991, all eligible unem-ployed individuals were entitled to 12 monthsof benefits. On January 1, 1992, entitlementwas reduced to six months. Since there was no"grandfatherclause," those who became un-employed after July 1, 1991, received a max-imum of only six months of benefits.Benefits. For most of those who worked

before entering unemployment, the level ofbenefits was set in 1991 at 60 percent of theperson's previous net wage for the first sixmonths of unemployment. However, individ-uals who were laid off because of major or-ganizational changes (redundancy) hadbenefits set at 65 percent of their previouswage.'2 For both groups, the replacementratefell to 50 percent in the second six months ofthe entitlementperiod.OnJanuary1, 1992, thereplacementrate in the first half of the entitle-ment period became 60 percent for workerslaid off for redundancy.'3Throughoutthe pe-riod, those who had never worked before re-ceived benefits equal to 60 percent of theminimum wage in the first half of the entitle-ment period and 50 percentin the second halfof this period.In 1991, a minimum benefit level was set at60 percentof the minimumwage buttherewasno maximum level. In January 1992, a maxi-mum level equal to 150 percent of the mini-mum wage was imposed. Throughout theperiod,a family could receive social assistance

'0 Since then, Poland has had two entitlementperiods,with a higher one for those employed for 25-30 years orin certain regions of the country." Individuals who were firedfor cause or quit repeat-edly were eligible for unemployment benefits after a six-month waiting period.

2 Those dismissed for redundancy also received sev-erance pay. In principle, severance pay was treateddiffer-ently in the two republics until January 1992. In the CRan individual was not eligible for unemployment benefitswhile collecting severance pay. From January1990to Jan-uary 1992, the Slovak authoritiesallowed theunemployedto receive severance pay concurrentlywithunemploymentbenefits. However, most Slovaks who received severancepay did not collect unemployment benefits untilthey hadexhausted their severance pay.'3Throughout 1991-1993, anyone undertaking train-ing received a benefit of 70 percent of his/her previouswage duringthe training period.


8/27

VOL. 88 NO. 5 HAM ET AL.: UNEMPLOYMENTN THE CZECHAND SLOVAKREPUBLICS 1123(welfare) in addition to unemployment com-pensation if the sum of the unemploymentbenefits and the income of other householdmembers was less than the household mini-mum living standard MLS).14Once benefits expired, the unemployedwere eligible for social assistance if theirhousehold income was below the MLS. Allsingle individuals were eligible for welfarebenefits, while a marriedperson was only el-igible if the combined income of the otherhousehold members was below the householdMLS.

Registered Unemployed WhoAre Ineligiblefor Benefits. A significant number of indi-viduals who were ineligible for unemploymentbenefits registered as unemployed. Some reg-istered in orderto obtain the assistance of thedistrict labor office in finding a job. Registra-tion was also a prerequisite or receiving wel-fare. As noted above, those who did not havethe necessaryworkexperience in the previousthreeyears (or its equivalent) were ineligible,as were those who were fired for cause or quitrepeatedly. Further, if a graduating studentstarteda job and lost it before acquiring 12months of experience, he was not eligible forbenefits. 5As will be seen below, we have taken theseprincipal UCS features into accountwhen de-veloping an econometric approach for esti-mating the effects of changes in the UCScharacteristicson unemployment duration.

III. The Estimation trategyA. EconometricModel

We analyze the durationsof Czech andSlo-vak unemployment spells, and their differ-ences, using a duration model. This model ispreferable to a regression model because fac-tors such as demand conditions and unem-

ployment benefits change over an individual'sunemployment spell and this nonstationaritycannot be captured in a regression frame-work."6 or each country we denote the hazardfunction (the probabilityof leaving unemploy-ment to employment) in week r of an unem-ployment spell as 17(1) X(r19) = (1 + exp(-y(rlO)))-Iwhere(2) y(rlO) = Z(r)y + aoB(r) + ajW

+ g(E(r)) + h(r) + 0.In equation (2), the term Z(r) containsvariables measuring demographic character-istics anddemandconditions in week r, whilethe vector y contains the correspondingset ofparameters.Differences in both the variablesand coefficients will determinethe differencesin the probability of leaving unemploymentinthe two republics. The means of these vari-ables aregiven in Appendix Table B 1. Exceptfor age, the demographic variables are in adummy variable form;the only ones requiringexplanationarethe "recentgraduate" andthe

education variables. The recent graduatevari-able is coded 1 if an individual graduatedwithin the last year from a university or highschool.18We use three dummy variables for

4 The MLS is equivalent to the household poverty linein the United States. See Terrell and Daniel Munich(1996) for a detailed description of the MLS in the CR,and Terrellet al. ( 1996) for an equivalentdescriptionforthe SR.'5For a discussion of those covered andnot covered byunemployment compensation in the United States, seeRebecca M. Blank and David E. Card ( 1991).

16 See e.g., Christopher Flinn and James J. Heckman(1982), Heckman and Burton Singer (1984a), NicholasM. Kiefer (1988), and Tony Lancaster (1990). See alsoTheresa J. Devine and Kiefer ( 1991 ) for a comprehensivesurveyof previous empiricalstudies.17 All variables in equations (1) and (2) are individualspecific. We have omitted the individualsubscript or ex-positional ease. In an earlier version of thepaper, we alsoconsidered a multiple exit version of the model since asubstantial number of individuals in the CR (but not theSR) leave unemploymentfor self-employment instead offor a new job. We estimated a separate transitionrateforfindinga new job and forbecoming self-employed. How-ever, the data were not rich enough to estimate thesetransitionrates separately.Thus we treatexits to self-em-ployment as exits to employment in calculatingthe abovehazard.8 Since by definition previouswages do not exist fornew graduates,we set theirwages to 0. Thus thenew grad-uate coefficient also picks up the wage effect. This isequivalent to imputing a common real wage for these

individuals.


9/27

1124 THEAMERICANECONOMICREVIEW DECEMBER 1998educational achievement: (i) graduates of avocational high school or an apprenticeship,(ii) graduatesof an academic high school, and(iii) those with some post-high-school (uni-versity) education. The control group consistsof those with only a junior-high-school edu-cation, the minimum level of education re-quired by law.We use threevariables o account for differ-ences in demand conditions. The first two-quarterlydata on district unemploymentratesand districtvacancyrates or the individual's d-ucation group -take on values that changequarterly ver the durationof a spell and acrossindividuals."9 he thirdvariable s the real valueof per capita ndustrialproduction n the districtin a given year.20 his variable akeson differentvalues acrosscalendaryears and acrossdistricts.In addition, or each districtwe use the ratio ofthe 1991 employment n agriculture o employ-mentin industry o capture he differences n theeconomic structure cross districtsat the begin-ning of the transition.2"As may be seen from Appendix Table B1,thereare substantialdifferences in the averagevalues of the demand variables between thetwo republics.In contrast,thereareonly smalldifferences in the demographiccharacteristicsbetween (1) the Czech and Slovak recipients,and (2) the Czech recipients and nonreci-pients.22 However, there appear to be more

substantialdifferences between recipients andnonrecipients in Slovakia, as well as betweenthe nonrecipients in the two republics (seee.g., education,maritalstatus,andliving in thecapital city).In equation (2), B(r) denotes unemploy-ment benefits in week r, W is the individual'sprevious weekly wage, and g( ) is a functionof remainingentitlementE(r) in week r. Foreach republic, we parameterizeg ( ) as a linearfunction of: (i) remaining weeks of entitle-ment, (ii) a dummy for the last week of enti-tlement before benefits have been exhausted,and (iii) an exhaustion dummy equal to 1 forall weeks after entitlementhasbeen exhausted.We allow for separatecoefficients on these lastthree entitlementvariablesfor marriedand sin-gle men, since single men will most likely beeligible for welfare afterexhaustion of benefitsand this is not necessarily true for the marriedmen.23The term h(r) represents the effect of du-ration dependence on thehazardand0 denotesan unobserved heterogeneity component. Inthe case of time-constant explanatory vari-ables, it is well known that ignoring unob-served heterogeneity or duration dependencebiases the coefficients of the explanatoryvari-ables. In ourcase, both benefitsandremainingentitlement are linked to duration, and wewould expect the potentialbias from ignoringheterogeneity or duration dependence to bemore serious than in the case when they arenot linked.24As a result, we use a polynomial9The denominator n both of these ratesis the relevantpopulation group, rather han labor-force group, since thenumber employed by education group anddistrict are notavailable for this period. We were concerned that thedistrict-level data might be too noisy, particularly n theSR where regional differences are smaller than in the CR.To address this issue more aggregate measureswere con-structed:the unemployment and vacancy rates by educa-tion in the individual's districtplus the congruentdistricts.Since the use of these more aggregate measuresdid notaffect the results, we focus on the results for the districtvariables.

20 The industrial production variable is available onlyat an annual frequency. It is a price-weightedcompositeof total per capita industrial productionin the district in1991 prices.21 We would also like to controlfor differencesin em-ployment in services across and within the two republics.Unfortunately,data on employment in services are notavailable at the district level for our sample period.22The main exception is the proportionof Romany(gypsies), which is higher for nonrecipients than recipi-ents in both countries and higher in Slovakia thanin the

Czech Republic.

23We do not have sufficient information o impute wel-fare benefits for single or marriedmen; therefore we setthe value of benefits to zero once an individual exhaustshis unemployment compensation. Thus the exhaustdummy (interacted with marital status) implicitly picksup the level of welfare benefits after exhaustion.24 Forexample, assume that there is no durationdepen-dence but that there s unobservedheterogeneity.Considermeasuring the effect of lowering benefits at 13 weeks,which we would expect to raise the hazard.However, themeasured hazard in week one could be higher than theaverage hazard at 13 weeks because the unobserved char-acteristics of workers who stay unemployed make themless likely to leave unemployment. If we ignore unob-served heterogeneity, the benefits coefficient will be bi-asedupwards.Alternatively, assume we havenegativedu-ration dependence but no unobserved heterogeneity, andconsider estimating the effect of changes in entitlement.Since entitlementgenerallyfalls as duration ncreases, ig-noring duration dependence also will bias the estimated

entitlement coefficient upwards.


10/27

VOL.88 NO. 5 HAMET AL.: UNEMPLOYMENTN THE CZECHAND SLOVAKREPUBLICS 1125in log duration to measure duration depen-dence and we account for unobservedhetero-geneity using the nonparametric approachdeveloped by Heckmanand Singer (1984b).25We estimate the model by maximum likeli-hood (see part 1 of Appendix A). Since it isdifficultto interpret he parameter stimatesofthe hazardfunction, we use these estimatestocalculate the effect of changing any given ex-planatoryvariableon the expected durationofunemployment.

B. Identificationof the Czech and SlovakUnemploymentCompensationParametersWhen estimating the impact of unemploy-

ment benefits on unemployment durations,itis necessary that the benefit levels vary in-dependently from other determinants of theduration of recipients' spells, particularlyprevious wages. We use five sources of in-dependent variation in benefit levels. First,in 1991 those who were laid off for redun-dancy had a replacement ratio of 65 percentof their previous wage in the first 13 weeks,while the replacement ratio was 60 percentfor other workers. Second, the replacementratio dropped to 50 percent from the thir-teenth to twenty-sixth week of compensatedunemployment for all individuals. Third, amaximum benefit level was imposed in1992. Fourth, a number of individuals hadtheir benefits raised to the minimum level ofbenefits. Fifth, unemployment benefits werenot indexed to inflation and hence we dis-count benefits by monthly movements in theconsumer price index in orderto capturetheerosion of the real value of benefits overtime. On the other hand, we assume that theappropriate proxy for the mean of theworker's wage offer distribution is his realwage (in October 1991 prices) at the time hebegan his spell. Prices and nominal wagesrose by approximately 30 percent from thelast quarterof 1991 to the second quarterof1993, the period covered by our data (KarelDyba and Svejnar, 1995).

Similarly,when estimatingthe impactof thelength of remaining entitlement on unemploy-ment durations, it is also necessary that theweeks of remaining unemployment compen-sation to which a recipient is entitled be in-dependentof other determinantsof the hazardfunction, particularly urrentduration.26n ourempirical work, the principal source of thisvariation n remainingentitlement comes fromthe significant numberof individuals who donot register for unemployment benefits at thetime of their job loss. For such individuals,remaining entitlement is not a simple linearfunction of currentdurationand initial weeksof entitlement (which is constant across thesample).27 One reason that ndividuals registerlate for unemployment benefits is that theyusually exhausttheirseverance paybefore col-lecting benefits. Other ndividuals simply waitto collect benefits;this phenomenonis similarto the phenomenon of less thanfull take-upofunemployment benefits in the United States(see e.g., Patricia M. Anderson and Bruce D.Meyer, 1997).It is worth noting thatthe variation nbenefitlevels and remaining entitlement is larger inmagnitudethan the variationused in studiesofunemployment durations n Canada and WestEuropeancountries. It is probablynot as largeas the variation found among unemploymentinsurance recipients in the United States. Inour empirical work we find that we have suf-ficient variation in benefit levels and entitle-ment to estimate precisely the impact ofbenefits and entitlementon the durationof un-employment spells.

25Following the results of Michael Bakerand AngeloMelino (1997), we are conservative in choosing the de-gree of the polynomial and the number of support pointsof the heterogeneitydistribution. See part1 of AppendixA for more detail.)

26 We cannot exploit the fact that individualsin 1991received a year of entitlement while individualsin 1992received only six months of entitlement. Thechange in thesystem was known as early as October 1991 and, as notedabove, individuals beginning a spell in 1991 were notgrandfathered.Because the CR and SR systems were be-ing computerized n 1991, we were unable to obtain microdataon unemployment spells priorto October 1991.27 Using late registrantsdoes complicate the econo-metric framework.See part2 of Appendix A for a discus-sion of the relevant ssues and the results of our addressing

these complicationsin our empiricalwork.


11/27

1126 THE AMERICANECONOMICREVIEW DECEMBER 1998C. An Alternative Measure of theDifferential Impact of the UnemploymentCompensation System in the CR and the SR

When addressing he second question raisedin the introduction,we can use the estimatedhazard function for recipients to calculate theeffect of marginal changes in the UCS on av-erage unemployment duration,e.g., the effectof a one-week increasein entitlement.We canalso address hisquestionfroma differentangleby exploiting data on recipients and nonreci-pients to obtain an estimateof the impactof aninframarginal hange in the UCS.28First,con-sider the hypothetical scenario where we makea recipient ineligible for unemploymentcom-pensation.Note that a simple comparisonof theexpected durations of recipients and nonreci-pients does not yield the correct magnitudesince it does not control for differencesin thecharacteristicsof the recipients and nonreci-pients. Instead, for each republic we calculatethe effect on unemploymentdurationof lettingrecipients keep the values of theirexplanatoryvariablesbut assigningthem coefficient valuesof the nonrecipients.29 ince the variablesre-lating to unemployment compensation are notavailablefor thenonrecipients,we use a smallerset of explanatory(demographicanddemand)variables,X*, to estimate the hazardrate forthe nonrecipients.Denote the same smaller setof variablesfor recipientsas X *, and the cor-responding parameterestimates for nonreci-pients and recipients as / *r and /3*respectively. Formally,we calculate

where ED (/3, X) denotes the expected dura-tion of unemployment at the mean values ofthe X s.3O In order to streamline notation, inwhat follows we simply use X to denote meanvalues in the expected duration calculations.)We call this "moving someone from being arecipient of unemployment insurance to beinga nonrecipient."We also calculate the effect of movingsomeone fromthe nonrecipientcategoryto therecipient category. To do this we let nonreci-pients keep their mean values of the explana-tory variables but take on the recipients'parametervalues. Fortnallywe calculate(4) Diff 2 = ED(/P*, X*r)

-D((*r n r)-ED(f3*, X*).

D. Decomposing the Differencein Czech and Slovak ExpectedUnemploymentDurationTo address our first research question, re-gardingthe factorsdetermining he differencesin the expected durations of unemploymentinthe two republics,we derive anduse a nonlin-

ear Oaxaca-type decomposition. Let fj denotethe entire vector of parameterestimates fromequation (2) for republic j and Xj denote thevector of mean values of the explanatoryvari-ables in republicj, wherej = c representstheCR andj = s representsthe SR. The differencein the expected durations between the Slovakand Czech Republics is given by(5) EDs - EDC = ED(Ps, Xs)

- ED(PC,Xc).In part 3 of Appendix A, we show how wecalculate the extent to which the difference inthe expected durations is due to: (i) differ-ences in the average characteristics of theCzechs and Slovaks (Xc and Xs), and (ii) dif-ferences in the coefficients (js and c). More-over, we can decompose the contribution ofthe explanatoryvariablesinto the contribution

28See also StevenMarston 1975) and BruceC.Fallick(1991) for a comparisonof recipients and nonrecipients.Phillip B. Levine (1993) allows increases in unemploy-ment benefits to affectthe durationof nonrecipients'spellsbecause they may increase recipients' spells. We do nothave enough data to implement his approach.29 Ideally we would like to control for unobserved dif-ferences between the recipients and nonrecipients.To dothis we wouldneed to estimatea durationmodel withsam-ple selection (see HamandRobert J. LaLonde, 1996). Toidentify and estimate such a model, we need a variablethat determines whether someone was a recipient or not,but does not affect their unemployment duration. Wecould not find a credibleexclusion restriction withwhich

to identify the selection model. 3 See part3 of Appendix A.


12/27

VOL.88 NO. S HAMET AL.: UNEMPLOYMENTN THE CZECHAND SLOVAKREPUBLICS 1127of: (i) demand variables, and (ii) other vari-ables, including the demographic ones. It isnot possible to carry out a similar decompo-sition of the coefficients. (See e.g., RonaldOaxaca and Michael R. Ransom, 1995.) Wecarry out the decompositions for both recipi-ents and nonrecipientsin the two republics.Since the CR and SR have the same UCS,we would not expect the UCS variablesto af-fect the contributionof the differences in theexplanatory variables (especially given thesimilarity of mean previous wages across therepublics). However,differentresponsesto theUCS in the two republicswill affect the con-tributionof the coefficients. Further,by exam-ining the impact of the UCS in each republic,we are able to ascertainthe direction of thiseffect, even if we cannot obtain a quantitativeestimatebecauseof the Oaxaca-Ransom esult.IV. FactorsAffectinghe Probability f LeavingUnemploymentn the CRandthe SR

A. TheDuration Effects of theUnemploymentCompensationVariablesThe estimatedeffects of changes in the UCSon the probabilitythatan individualwill leave

unemployment in the CR versus SR are pre-sented in Table 2. In Panel A of the table wepresentthe estimatesof the hazardcoefficientsfor selected variables for the two republics.31(A negative coefficient indicates that an in-crease in the variable reduces the exit rate.)InPanel B we use the estimated coefficients ofthe Czech and Slovak hazardfunctions to cal-culate the effect of a given change in eachUCS variableon the expected durationof un-employment (in weeks) in each republic.32 i-nally, in Panel C we report he elasticity valuesimplied by the expected duration effects of achange in benefits,entitlement,or the previouswage in each republic.Startingwiththe results or the CRin column1 of Table 2, we find that the level of uliem-

ployment benefits has a statisticallysignificantcoefficient and that a 10-percent ncrease n thelevel of benefitscauses the expected durationofanunemployment pelltoriseby 0.61 of aweek.Sincetheaverageexpectedduration f recipientsis estimatedat approximately17.75 weeks, ourresults mply a moderateelasticityof unemploy-ment durationwith respect to benefitsof 0.34.The estimated coefficients for the remainingweeks of entitlement orsingle and manied menare both significantlynegativeand almostiden-tical. However, the coefficient for the last weekof entitlement s significantlypositive for mar-ried men while it is negative and insignificantfor single men. The differencein these coeffi-cients may reflect the fact thatsingle men willqualifyforwelfarewhen theyexhaust heirben-efits while manymarriedmen will not. The ex-haustion dummy is significantly negative forboth single and marriedmen.33The entitlementcoefficients mplythatan additionalweek of en-titlement results in 0.30 of a week increaseinexpected duration or all men. (The respectivefigures orsingleand marriedmen arequitesim-ilarand thuswe focus on theoveralleffect.)Theestimatedelasticityof expecteddurationwithre-spect to entitlement s 0.44. The coefficient onthe previous wage (which proxiesthe opportu-nity cost of staying unemployed) is not signifi-cantlydifferent rom zero atstandard onfidencelevels.For Slovakia, the relevant results are con-tained in column 2 of Table 2. The benefitcoefficient is negative as expected andimpliesan elasticity of only 0.06. 'Thiscoefficient isnot significantly different from zero, but it isinformative because the 95-percent confidenceinterval is relatively small. We can obtain anupper-boundestimate of the elasticity of ex-pected duration with respect to benefits byusing the lower bound of the confidence inter-val for the benefits' coefficient (which willhave the greatest disincentive effect) .The resulting estimated upper bound for thiselasticity equals 0.13, indicating that the

"' The estimated effects of all variables are reported nAppendix Tables B2 and B3. Ham et al. (1993) provideestimates of the empiricalhazardfunctions.32 Since we do not have a long time series,we calculatea truncatedexpected duration(at four years). See part3

of Appendix A.

3 Recall that this dummy also picks up the effect ofwelfare benefits.3 The SR hazard coefficients are statistically differentfrom the Czech coefficients using a likelihood ratio test." The lower bound of the confidence interval is -7.77- (1.96 x 6.37) = -20.26.


13/27

1128 THE AMERICANECONOMICREVIEW DECEMBER 1998TABLE 2-EFFECTS OF THE UNEMPLOYMENT COMPENSATION SYSTEM

Czech Republic Slovak RepublicA. Coefficients of the hazard function

Weekly unemploymentbenefits - 13.622 -7.766(6.914) (6.370)Previous weekly wage 0.241 6.609(3.036) (2.882)Weeks of remaining entitlement*married -0.383 -1.095(0.140) (0.163)Weeks of remaining entitlement*single -0.391 -0.744(0.134) (0.348)Last week of entitlement*marTied 1.253 0.626(0.363) (0.317)Last week of entitlement*single -0.869 0.744(0.762) (0.348)Benefits exhausted*married -0.722 -0.618(0.349) (0.302)Benefits exhausted*single -1.686 -0.515(0.398) (0.346)

B. Expected durationeffects (in weeks):Base expected duration 17.75 60.71Benefits raised by 10 percent 0.61 0.39Previous wage raised by 10 percent -0.02 -1.49Entitlement raised by 1 week 0.30 0.93Entitlement raisedby 1 week-single men 0.24 0.83Entitlementraisedby 1 week-married men 0.34 1.00

C. Elasticity of expected duration with respectto:Increasein benefits 0.34 0.06Increasein entitlement 0.44 0.41Increasein previous wage --0.01 -0.25

Note: Standarderrors n parentheses.Source: Columns 1 and 4 of AppendixTables B2 and B3.

disincentive effect of increasing benefits isquite small in the SR.36 The coefficients on remaining entitlementfor both single and mar-

36 The effect of raisingbenefits by 10 percent using thelower-bound estimate of the benefits' coefficient is to in-crease duration by 0.772 weeks. Thus the absolute mag-

nitude of the changes of raising benefits is similar in thetwo republics. However, because the base durationis somuch largerin Slovakia, the estimated benefits' elasticityis much smallerin the SR.


14/27

VOL. 88 NO. 5 HAM ET AL.: UNEMPLOYMENTN THE CZECHAND SLOVAKREPUBLICS 1129riedmen are bothnegative andsignificant.Thecoefficients for the last week of entitlementforsingle and marriedmen are positive, statisti-cally significant, and quite similar. Hence, incontrast to the CR, there is a spike in the prob-ability of exiting unemployment in the lastweek of entitlementfor both marriedand sin-gle men. The coefficients on the benefits' ex-hausted variables are statistically significant,have the expected negative sign, andagainarenearly identical for marriedand single individ-uals. Raising entitlement by one weekincreases the expected durationof unemploy-ment by 0.93 of a week. (Again, the effects forsingle and marriedmen are similar.)The esti-mated elasticity of expected durationwith re-spectto raisingentitlements 0.41, which againrepresentsa moderate disincentive effect. Theprevious wage has a positive coefficientand isstatistically ignificant.A 10-percent ncrease nthe previousweekly wage leads to a 1.49-weekdecrease n expectedduration, mplyingan elas-ticityof about -0.25. Thusin theSR, those witha higher opportunitycost of time leave unem-ployment earlier.A comparisonof our benefitandentitlementelasticities for the CR and SR with those fromdifferenthazard unction studiesfor theUnitedStates and Canada -as summarized byDevine and Kiefer (1991 Table 5.2) -indi-cates that our benefit elasticities are on thelower end while ourentitlementelasticitiesareclose to the average values reportedby Devineand Kiefer. Hence, there is little about the be-havioral responses to the UCS system thatlooks different from what we have seen in theUnited Statesand Canada.Czechs and Slovaksrespond to the incentives createdby the UCSsystem in a similarway to their Western coun-terpartsandnot too dissimilarfromeach other.Indeed, since the relevant elasticities arelowerin the SR than in the CR, tightening of theUCS would not reduce the (proportionate)difference in unemployment durations be-tween the two republics. We must thereforeturnto differences in otherfactorsfor possibleexplanations of the Czech-Slovak difference.B. TheDuration Effects of the Demographicand Demand Variables in Each Republic

The full set of coefficient estimates for re-cipients and nonrecipients in the CR and the

SR are contained in Appendix B, Table B2,while the corresponding expected durationcalculations are reported in Table B3. Inthese tables the estimates for Czech and Slo-vak recipients using the full model are con-tained in columns 1 and 4, respectively. Wereport in columns 2 and 5 of the tables theresults from the smaller model for recipientsin the CR and SR, respectively. Finally, col-umns 3 and 6 report the results (from thesmaller model) for nonrecipients in eachrepublic."Our estimation yields a number of inter-esting results for the demographic and de-mand variables. Education has a strikinglydifferent effect in the two republics, raisingthe exit rates from unemployment in Slo-vakia but not much in the Czech Republic.Our results indicate that at least among re-cipients, the CR was able to absorb the dis-placed low-skilled workers (those withouthigh-school or higher education) into em-ployment at about the same rate as the moreeducated ones, but that the SR was unable todo so. This result suggests that men with lit-tle education are a natural group to target inthe SR and other CEEs. We also find thatunemployment duration increases with ageat a similar rate in both the CR and the SR.Single and handicapped rnen are also simi-larly disadvantaged (relative to base dura-tion) in the two republics. Interestingly,Romanies are at a proportionate disadvan-tage in the CR, although they have very longdurations in both republics. The demandvariables have the expected signs, althoughonly the unemployment rate is generallystatistically significant. Differences in theagricultural-industry structure across

37 The likelihood functions for the nonrecipients werepoorly behaved, indicating that the smaller model is stilltoo rich for the relatively small numberof individuals inthese groups.After eliminatingthe vacancy rate from thespecification,the problem with the likelihood function dis-appeared n the SR. In the CR we obtained two optima forthe nonrecipients.The optima were quiteclose in terms ofthe value of the likelihood function, and the problemre-mained when we changed the specification. To err on theside of being conservative, here we reportthe results forthe optimum with the largest discentive effects of non-marginal changes in the UCS. (The other optimumshowed no disincentive effect.) This choice of optimumdid not affect the CR-SR comparisons.


15/27

1130 THE AMERICANECONOMICREVIEW DECEMBER1998TABLE 3-AN ALTERNATIVE MEASURE OF THE IMPACT OF THE UCS:

RECIPIENTS V. NONRECIPIENTS(USING THE SMALLER MODEL)

Czech Republic Slovak RepublicExpected durations(in weeks) of:1. Recipientfl's and X's 18.6 55.5

2. Nonrecipientfl's and X's 15.6 54.83. Recipient fl's, nonrecipientX's 17.3 65.74. Nonrecipientfl's, recipientX's 13.7 46.3

Differences in expected durations:5. Recipient to nonrecipient(row 4-row 1) -4.9 -9.26. Nonrecipientto recipient(row 3-row 2) 1.7 10.9

Sources: Rows 1 and 2 are taken from columns 2, 3, 5, and 6 of Tables B3. The calculationsinrows 3 and 4 are based on the coefficients in columns 2, 3, 5, and 6 of Table B2.

districts do not play much of a role in deter-mining the probability of leaving unemploy-ment in either republic.C. UnemploymentDuration of Recipientsv. NonrecipientsIn this section we report the effect on theexpected duration of unemployment of mov-ing an individual: (i) from the recipient cat-egory to the nonrecipient category, and (ii)from the nonrecipient to the recipient cate-gory. These calculations, presented in Table3, provide estimates of the effects of infra-marginal changes in the UCS.38For the CRwe see that the base expected duration of anunemployment spell for recipients (using thesmaller model) is 18.6 weeks (row 1) andfor nonrecipients it is 15.6 weeks (row 2).

This implies that if we ignore the differencesin characteristics between the recipients andnonrecipients in the CR, we find that movinga recipient to the nonrecipient categorywould reduce his average spell by 16.1 per-cent. In contrast, when we allow for the re-cipients to behave like nonrecipients usingequation (3), we find (in row 5) that moving

an individual from the recipient to the non-recipient category reduces his unemploy-ment duration by 4.9 weeks or 26.3 percent.Alternatively, when we use equation (4) wefind (in row 6) that moving someone fromthe nonrecipient to the recipient category in-creases his unemployment duration by 1.7weeks or 10.9 percent.In the case of the SR, a comparison of thetop two rows of Table 3 indicates that thebase expected duration for nonrecipients is0.7 weeks ( 1.3 percent) shorter than that forrecipients. However, controlling for differ-ences in characteristics using equation (3),one finds that moving a recipient to the non-recipient category decreases unemploymentduration by 9.2 weeks or 16.6 percent (row5). Alternatively, using equation (4) we es-timate in row 6 that moving a nonrecipientto the recipient category increases unem-ployment duration by 10.9 weeks or 19.9percent.39

38 The calculations are based on the smaller model forrecipients and nonrecipients (columns 2, 3, 5, and 6 ofAppendix Tables B2 and B3).

"As noted above, we dropped vacancies from thespecification for the SR nonrecipients. We calculatedequations(3) and(4) with and withoutvacancies droppedfrom the specificationfor SR recipients.This change hadno effect on the estimateddifferences.We also calculatedthe decomposition in equation (5) with and without va-cancies in the CR nonrecipient specification, and againthere was no differencein the estimates.


16/27

VOL.88 NO. S HAM ET AL.: UNEMPLOYMENTN THE CZECHAND SLOVAKREPUBLICS 1131TABLE -DECOMPOSINGHEDIFFERENCENEXPECTEDURATIONETWEENHECZECH NDSLOVAK EPUBLICS

Recipients NonrecipientsNumber Percent Number Percent

ED, - ED, (weeks) 43.0 100.0 39.2 100.0Differences due to:Coefficients 22.7 52.9 24.1 61.5Explanatoryvariables 20.2 47.1 15.1 38.5

Differences from explanatory variables due to:Demographics (-1.2) (-2.8) (8.7) (22.1)Demand conditions (21.4) (49.9) (6.4) (16.3)Sources: Row 1 is obtainedfrom the base expected durations n firstrow, columns 1, 3, 4, and 6 of AppendixTable B3.The decomposition in rows 2 and 3 use the estimationspresentedin the same columns of Appendix Table B2.

Since these experiments correspond todrastic changes in public policy, we considerthese effects to be moderate.40A related, andfrom our standpoint the most important,finding in Table 3 is that the difference inexpected durations between the CR and SRis much greater than the difference betweenrecipients and nonrecipients within either re-public. As noted above, our comparisons ofrecipients and nonrecipients in Table 3 sug-gest that if one were to eliminate the UCSbenefits, one could expect to reduce unem-ployment duration (averaging the two ef-fects) by similar amounts of about 18percent in each republic. In contrast, the re-sults in Table 3 also indicate that if one couldtransplant an unemployed Slovak into theCzech system, one would reduce his ex-

pected unemployment duration by 67 per-cent for a recipient and 71 percent for anonrecipient.4"Hence, our analytical com-parisons of recipients and nonrecipients in-dicate that eliminating the UCS in Slovakiawould have a visible effect on unemploy-ment duration, but that this radical policychange would not go very far in loweringunemployment duration in the SR relative tothe CR.V. ExplainingheDifferenceBetweentheAverageCzechandSlovakUnemployment urationsA. Oaxaca-TypeDecompositions

A primarygoal of this paper s to investigatethe differences in unemploymentdurations nthe CR and SR using the Oaxaca-typedecom-positions. FromTable 4 we see thatduring theperiod of our study, the average recipient inSlovakia was unemployed 43.0 weeks longerthan the typical recipient in the Czech Repub-lic. The decomposition for recipientsin Table4 indicatesthat 20.2 weeks (47 percent) of thedifference in expected duration between thetwo republics is due to differences in the ex-planatoryvariables, and 22.7 weeks (53 per-cent) is accounted for by differences in theestimated coefficients. Interestingly,virtually

40An alternativeapproach o this question would be tocalculate the expected duration of unemployment usingthe estimates of the recipient hazard with entitlementandbenefits set equalto zero. One could comparethismeasurewith the base expected duration or recipients inAppendixTable B3, acknowledging that this is extrapolating wellbeyond the experience of recipients in the sample. Oneproblem in this approachis how to handle the benefitsexhausted variable, since this will pick up: (i) the differ-ence in unemployment compensation and social assis-tance, and (ii) duration dependence. To avoid thisproblem, we set the benefits exhausted dummy equal tozero and set benefits equal to social assistance (but keepentitlementequal to zero) when making this calculation.Using this approach,we find that moving someone fromthe recipient to nonrecipientcategory reduces unemploy-ment durationby 45.7 percentin the CR and by 40.6 per-cent in the SR.

41 This ignores differences in mean characteristics.Weaddress this issue in Section V, subsection A.


17/27

1132 THE AMERICANECONOMICREVIEW DECEMBER 1998none of the contribution of the explanatoryvariablesis driven by differences in the valuesof the demographicvariables between the tworepublics. Instead, this contribution comes al-most entirely from the demand and industrystructurevariables. The finding that over one-half of the difference in unemployment dura-tion between the two republics comes fromdifferencesin the coefficients suggests that thefunctioning of labor-market institutions andthe behavior of firms and agents in the labormarketis very different in the two republics.For example, the difference in the coefficientsreflects in part the fact that the CR can placemost marginalworkers nto employment muchfaster than the SR can. We address this issuepresently.From column 3 of Table 4 we see that theaverage Slovak nonrecipientwas unemployed39.2 weeks longer than the typical Czech non-recipient. Of this difference, 15.1 weeks (or38.5 percent) is accounted for by differencesin the explanatory variables and 29.1 weeks(61.5 percent) is accountedfor by differencesin the estimated coefficients. About three-fifths of the contributionof the difference inthe explanatory variables is coming from thedifference in demographic variables. Thus,unlike the case of recipients, among nonreci-pients the differences in demographic vari-ables play a role in explaining the differencein the expected durationsbetween the CR andSR. As pointed out in Section III, subsectionA, there are important differences in the de-mographic compositions of the nonrecipientsin the two republicsin terms of the proportionsof junior-high-school education, Romany,marital status, and living in the capital city.However, again the large portion explained bythe differences in the estimated coefficientssuggests that there areimportantdifferencesinthe functioning of labor-marketinstitutions,firms and individuals in the two republics.

B. Factors Underlyingthe Differences inEstimatedCoefficientsin the TwoRepublicsThe differences in the estimatedcoefficientspresumably reflect factors such as additionalobserved and unobserved differences in thestructure of the two economies and differentresponses of firms, individuals, and institu-

tions in the labor market. On the basis of in-

terviews that we carried out with policymakers in the two countries, as well as ourreading of other studies, we have identifiedseveral factors that may be important in ac-counting for the role played by the differencesin coefficients in the decompositions:421. In the early 1990's the growth of the ser-vice sector was much faster in the CR thanthe SR, as was the growth of small firms.43This growth could not be accurately mea-sured (at any level of aggregation) sinceofficial statistics were only gathered syste-matically for firms with 25 or moreemployees.2. Privatizationwas carriedout more quickly

in the CR. By 1993, 53.5 percent of allworkerswere in private firms in CR, com-paredto only 32.0 percent in the SR. Thisphenomenon is impossible to measure atthe district evel, butit is importantbecausemuch of the new hiring is likely to havebeen done by private firms.'3. The age and location of factories differsacross the two republics. The CR has fac-tories that tend to be older thanthose in theSR, but their location was chosen by mar-ket forces before communism. Slovakiawas industrialized after the communisttakeover and planners determined the lo-cation of Slovak factories with little regardto the proximityof raw materials or markettransportation osts. It is possible that thelocation factor dominated the capital vin-tage effect and resultedin lower reemploy-ment probabilitiesin the SR.4. The CR attractedover ten times as muchforeign direct investment as the SR in theearly to mid-1990's.455. Slovakia was much harderhit than the CRby the decline in military production.46n-

42 Some of these factors may also explain the differencein observed local demand conditions.4 See e.g., Anton Vavro (1992) and Milan Horalek(1993).4 Czech Statistical Yearbook Czech Statistical Office,1994) and Slovak Statistical Yearbook Slovak StatisticalOffice, 1994).45 EuropeanBank for Reconstructionand DevelopmentTransitionReports 1996 and 1997.46 In the late 1980's, the Visegrad countries were thefifth largest arms exporters in the world. In 1989, their

exports equaled $600 million (Marko Milivojevic, 1995).


18/27

VOL. 88 NO. S HAM ET AL: UNEMPLOYMENTN THE CZECHAND SLOVAKREPUBLICS 1133terestingly, much of the relative decline oc-curred before, rather than during, thetransition. However, the artificially main-tained full employment under central plan-ning meant that this labor redundancymanifested itself when market forcesstarted to operate in the early 1990's. In-deed, the peak in Czechoslovak militaryproductionoccurred n 1987, when militaryproduction is estimated to have accountedfor 3 percent of total industrial productionand generated 73,000 direct jobs and50,000 to 70,000 indirectjobs. At this time,the value of Slovakia's military productionwas 50 percent above thatof the Czech Re-public.47 In 1988 military productionstartedto decline in both republics, but thedecline was much greater in Slovakia. Inthe 1988-1990 period, the value of mili-tary production (in constant prices)dropped by 48 percent in the SR and 35percent in the CR. The real value of mili-tary production fell by an additional two-thirds between 1990 and 1992, and the rateof decline may have been slightly faster inthe SR than in the CR (Ladislav Ivanek,1994) 486. It has been easier for the Czechs than forthe Slovaks to work in neighboring West-ern countries. The CR has a border with(former West) Germany, whereas the SRdoes not, andGermanypursued enientpol-icies toward guest workersfromthe formerCzechoslovakia.49Moreover, the CR has alonger borderwith Austriathanthe SR.7. Finally, from discussions with policy mak-ers anddistrict abor officers in bothrepub-lics, it appearsthat in these early years the

Czech officials were stricter in enforcingthe eligibility andentitlement requirementsof the UCS than the Slovak officials. Thisresult is consistent with the largerratio ofnonrecipients to recipients in the CR sam-ple as comparedto the SR sample.50

VI. ConcludingRemarksOne of the most important issues encoun-tered in CEE economies as they abandonedcentralplanning has been the emergence of alow (3-5 percent) unemployment rate in theCzech Republic, togetherwith a high (double-digit) unemployment rate in Slovakia and allthe otherCEE countries. Since the differential

rise in unemployment in the early phases ofthe transitionwas brought about primarily bymuch higherrates of exit fromunemploymentto employment in the CR than in Slovakia andother CEE economies, a principal goal of thispaper was to investigate the nature andcausesof the difference between these exit rates.Since the Czech and Slovak Republics sharedmany institutional and legal features, we fo-cused our analysis on these two economies,noting that the similarities in the outcomes inSlovakia andthe otherCEEs gave ourfindingsrelatively wide applicability.Our first principal finding comes from thedecomposition of the determinantsof the ex-pected durations of unemployment in theCzech and Slovak Republics. We find thatabout one-half (more than one-third) of thisdifferencefor recipients(nonrecipients) is ex-plained by differences in the values of the ex-planatory variables. For those who receiveunemployment benefits (recipients), we findthat almost all of the contribution of the dif-ference in the explanatory variables arisesfrom differences in the levels of local demandvariables and a variable measuring structuraldifferences at the district level. However,among nonrecipients, differences in demo-graphiccharacteristicsplay a somewhat moreimportantrole thandifferences in demand fac-tors between the CR and SR.

47 Since the Slovak labor force is almost exactly one-half of the Czech labor force, the extent of military pro-duction relative to the size of the economy was clearlymuch largerin Slovakia than in the Czech Republic.48 An offsetting factor was the fact that the Czecho-slovak federal government contributedsignificantly to theconversion efforts of the military producers-about $50million in 1991 and $35 million in 1992. These subsidiestended to mitigate the relative unemployment evels in thetwo republics,as three-fourthsof the subsidy in each yearwent to producers n Slovakia.4 In the southern German State of Bavaria, workersfrom the former Czechoslovakia were even allowed to

work without a work permit.

50 Note that the differencein enforcement cannot be theonly factor determiningthe difference in expected dura-tions between the two republics, given thelargedifferencein the expected durationsof nonrecipientsbetweenthe CRand SR.


19/27

1134 THE AMERICANECONOMICREVIEW DECEMBER 1998The remaining difference in the durationsof the unemployment spells in the two repub-lics is explained by differences in the esti-mated coefficients of the hazard function for

leaving unemployment. In this context, animportant difference between the two repub-lics lies in the relative inability of the SlovakRepublic, and probably also the other CEEcountries, to absorb low-skilled unemployedworkers. Likely explanations for the differ-ences in the estimated coefficients include themore rapid growth of the small-scale, privatefirm (mainly in the service sector) in the CRas comparedto the SR, the relatively strongerimpact of the decline in military productionin the SR than in the CR, the differences inthe response of the manufacturing sector tomarket forces (arising in part from differ-ences in the age and location of factories be-tween the two republics), and the muchhigher level of foreign investment in the CRthan the SR. Differences in the enforcementof the unemployment compensation system inthe two republics also are likely to have con-tributed to differences in the hazard coeffi-cients between the CR and SR.Our second principal finding concerns theeffect of the unemployment compensationscheme, which was identical in the Czech andSlovak Republics. Using two different meth-ods, we estimate that this system has moderateeffects on the durationof unemploymentspellsin each of these republics, comparedwith theestimatedeffects in other studies in the UnitedStates, Canada, and Europe. This result sug-gests that policy makers in governments andinternationalagencies have considerable lati-tude in providing a safety net without endan-gering efficiency. In view of the similaritybetween Slovakia and the other Central andEast European countries in their unemploy-ment situationsand the features of their UCSs,this result is also relevant for policy in othertransition economies.

APPENDIX A1. Standard Contribution to the LikelihoodFunction or Completeand CensoredSpells.Here we show the contribution to the likeli-hood for those who registeratthe district aboroffice immediately upon becoming unem-ployed. The survivor function S(r) -the

probability of a spell lasting longer than rweeks-is given by(Al) S(rIO) = (1 - X(vlO)).The contribution of a spell that ends in weekt is given by(A2) f(tI 0) = X(t|O)S(t - 10).Let b(0) representthe density function for theunobserved heterogeneity. The unconditionalcontributionto the likelihood for the spell thatends in week t is given by(A3) L(t) = fX(t| O)S(t -1 |0)4(0) dO.The contributionof a censored spell is calcu-lated in an analogous manner. We followHeckman andSinger (1984b) andassume that0 is drawnfrom a discrete distribution with Jsupport points 0S, .. v 0-, 1, 0Jand associatedprobabilities PI, ..., - 1. The number ofpoints of support J is determinedby the data.51Recent Monte Carlo evidence by Baker andMelino ( 1997) suggests that care must betaken in choosing the number of supportpoints. They find thatparameterestimatescanbecome unstable when one estimates too manysupportpoints, which suggests that we shouldbe relatively conservative in selecting thenumberof supportpoints.52Given theirresults,we use the Schwartz criterion to choose thenumber of points, since it will lead to a moreparsimonious specification than would a like-lihood ratio test.53In a previous specification(very similar to the one adoptedhere),5 this

' We ignore any complications in the asymptotic dis-tribution theory arising from the fact that J is alsoestimated.52 Note that we follow the literatureand ignore the factthat choosing the numberof supportpoints involves a non-standard testing problem since some parameters will beunidentifiedunderthe null hypothesis.53 See, e.g., George G. Judgeet al. (1980 pp. 425-26).Note that allowing for even two mass points involves es-timating more parameters than would be the case if weassumed a normal distribution or the heterogeneity.54 The earlierspecificationin Hamet al. ( 1996) didnot

use the ratio of employment in agricultural o that in in-


20/27

VOL.88 NO. 5 HAM ET AL.: UNEMPLOYMENTN THE CZECHAND SLOVAKREPUBLICS 1135always led us to two points of support,andweuse two points of support in our currentspec-ification. Following Ham and Samuel A. Rea,Jr. (1987), we choose the degree of the logdurationdependencepolynomial from the no-heterogeneity specification. We again use theSchwartz criterion, assuming that the maxi-mum degree of the polynomial is five.

2. Contribution to the Likelihood for LateRegistrants. -Consider an individual whoregisters after T weeks of uncompensatedun-employment and experiences t additionalweeks of unemployment.The conditional con-tribution to the likelihood of the t weeks ofunemployment afterregisteringis(A4) L(t 0) = X(t + TI0)

t- Ix n (1 -X(T+ vO)).v = I

As noted above, these spells eliminate the col-linearity between remaining entitlement andduration.They also break the identitybetweenthe drop in benefits and duration at 13 weeksand thus help us to identify the benefit effect.However, using these spells raises an econo-metric issue. Workers who register late aredrawn from a differentheterogeneitydistribu-tion than those who register mmediately.Herewe discuss two possible solutions to this prob-lem. As noted by Heckman and Singer(1984a), we should integrate(A4) againstthedistribution of 0 among those who have un-employment spells longer thanT, 4(9 I > T).In ourempiricalwork we considertwo approx-imate solutionsto this problem.First,we allowfor a separate heterogeneity distribution forthose who registerlate. Second, we separatelyallow for a differentheterogeneitydistributionfor those who registerfor unemploymentcom-pensation at least three months after they be-came unemployed.Neither of these approachesaffectedthe results.

These are only approximate solutions tothe problem since they ignore the fact thatthe heterogeneity distribution for those whoregister late depends both on T and on thelagged values of the explanatory variables.In future work we intend to investigate thefollowing approach for addressing these is-sues.55 After explicitly conditioning on thevector of explanatory variables over the en-tire unemployment spell prior to registrationXai, we have(AS) 0(0| t > T, Xai)

Pr(t > T, O1Xai)Pr(t > TIXai)Pr(t > T I0, Xai)4O(O)Pr(t> TIXai)

= S(T IO,Xai) ]4()L[ S(TI r, Xai)4o(r) drj

- gi(TfI, Xai)O(M).Thus we should weight 4(O) the for thosewho registerlate by gi (-). Now

(A6) S(T IO, Xai)T

-I (1I- X.(v I0, Xi (r)),v= 1

where A, is the hazardout of unemploymentfor someone who waits to register (and thusisnot receiving benefits) andX'i = (X (1),Xi' (T)). We cannot estimate A, because weonly see those who survive among those whowait to register; we do not see those who waitto register and leave unemployment beforeregistering at the labor office for benefits. Touse this approach and calculate the weights,we need to approximate w. We are currentlydustry in 1991 as an explanatory variable. With this pre-vious specification, we found that the likelihoodratio testled us to a three-point distribution n the SR and the en-titlement elasticity rose substantially. However, as notedabove, the Schwartz criterion led us to two points of

support.T'Ihis pproachoriginated from a comment by George

Jakubson.


21/27

1136 THEAMERICANECONOMICREVIEW DECEMBER 1998exploring an approximationwhere we use thenonrecipient hazard to proxy XA.

3. Expected Duration Calculations andOaxaca Decompositions. -Write the ex-pected durationof unemployment in each re-public as(A7) EDj = ED(1j, Xj), jc, s,where 13j s the vector of parameterestimatesfor republic and Xj is the vector of the meanvalues of the explanatory variablesin republicj.56 Since we do not have a long time series,we calculate a truncatedexpected duration atfour years)57

-K(A8) tf(t)t=1+ (1 Pr(t < 4 yrs) ) *4 yrs

where K 4 yrs - 1 week.In calculating this expectation, we freeze thehazard after 52 weeks at its value at 52 weeks,since we do not have much data on spellslonger than this.The differencein the expected durationsbe-tween the republics is given by(A9) EDS - EDC = ED (Ps, Xs)

- ED(PC XC).We can decompose this difference into a con-tribution due to the difference in the coeffi-cients and a contributiondue to a difference inthe explanatoryvariables

(AIO) EDS -ED,- (ED(ps, X)ED(PC, Xs))

+ (ED(PC5 S)- ED(fe, Xc) ).

Of course, we could use the alternativedecomposition(Al l) EDS-EEDc (ED(3s.,X c)

ED(PC Xe))+ (ED(Ps5XS)- ED(PS Xc))

Within this framework,an averagemeasureof the contributionof the difference in the co-efficients is(A 12) Diff (,B)- ED PBS Xc)

ED(IC XM))+ (ED(Psf5XS)- ED(Pc3 Xs))) /2.

By the same token, we measure the contribu-tion of the difference in the explanatoryvari-ables as58(A13) Diff (X )-((-ED(Ps (v))

+ (ED(fi5 Xs)- ED(Pc, XC))) /2.

We can also decompose the overall effect ofthe difference in the mean explanatory vari-ables into the respective contributions of dif-ferent subsets of the explanatoryvariables.Forexample, divide the explanatory variables inthe CR into two groups, i.e., Xc = (Xc 1 Xc2)

56 We have droppedthe bar on the X 's to simplify thenotation. One can calculate the expected duration at themean values or calculate it for each individual and thentake the mean, see e.g., Ham and LaLonde (1996). How-ever, the only way we can carry out the decompositionbelow (when we measure the contributionsof the subsetsof explanatory variables) is to calculate the expected du-rations using the mean values, and thus we adopt this ap-proach when calculating all expected durations.57 We experimented with three- and five-year horizons

and there was no qualitativedifference in the results. 58 Unobserved differences are captured n (A12).


22/27

VOL.88 NO. 5 HAM ET AL.: UNEMPLOYMENTN THE CZECH AND SLOVAKREPUBLICS 1137and do the same for the SR. Then it is possible The first term on the right-handside of (A14)to decompose the contributionof the differ- representsthe contribution of the differencesence in the X variables (conditional on the in the XI variablesacross the republicsand theCzech parameterestimates) using second termrepresentsthe contributionof the

differences in the X2 variables.59(A14) ED(P, X) - ED(P, Xc)- (ED(pc, XsIc Xs2)

-ED(pc, XcI, Xs2))+ ( ED(fc,- XcI, Xs2) 5 The decomposition is nonlinear and therefore can besensitive to the order in which one switches the variables.To address this problemwe calculate all possible permu-- ED(I3e, Xc I, Xc2) ) tations and average the estimates.

APPENDIX BTABLE B1-MEAN VALUES OF VARIABLES AT FIRST WEEK OF REGISTERED UNEMPLOYMENT

Czech Republic Slovak RepublicRecipients Nonrecipients Recipients Nonrecipients

Total number of men 780 482 1063 229Average weekly exit rate to jobs oversample period 0.052 0.063 0.020 0.018Proportionthat do not exit to jobs insample period 0.113 0.161 0.342 0.382Proportionof men who exhaust benefitsin sample period 0.135 0.455Weekly benefits (in Kcs)ain 1991prices*10-4 0.044 0.046Previous weekly wage (in Kcs)ain 1991prices* 0-4 0.059 0.062Demographicvariables:Age*10' 3.321 3.148 3.135 3.061Vocationalhigh school 0.573 0.579 0.509 0.467Academic high school 0.179 0.137 0.232 0.162Post high school 0.059 0.056 0.055 0.035Romany 0.030 0.064 0.053 0.140Handicapped 0.104 0.071 0.061 0.087Married 0.506 0.556 0.527 0.489


23/27

1138 THE AMERICANECONOMICREVIEW DECEMBER 1998TABLE B1-Continued.

Czech Republic Slovak RepublicRecipients Nonrecipients Recipients Nonrecipients

Recent graduate 0.141 0.145 0.130 0.153Lives in Prague/B3ratislava 0.090 0.089 0.058 0.009District demand variables:Quarterlyunemploymentratebyeducationgroup 3.093 2.65 9.777 9.39Quarterlyvacancy rateby educationgroup 0.745 0.786 0.344 0.227Annual per capitaindustrialproductionin1991 prices*10-6 0.075 0.068 0.093 0.088Agricultural/industrial mployment ratioin 1991 0.370 0.376 0.572 0.607

a The exchange rateduringthis periodwas about 29 Kcs (Czechoslovak crowns)/$US1.

TABLE B2-ESTIMATED COEFFICIENTS FROM THE HAZARD MODEL

Czech Republic Slovak RepublicRecipients Recipients Nonrecipients Recipients Recipients Nonrecipients

Weekly benefits -13.622 -7.766(6.914) (6.370)Weeks of remaining -0.383 -1.095entitlement*married (0.140) (0.163)Weeks of remaining -0.391 -0.754entitlement*single (0.134) (0.152)Last week of entitlement*married 1.253 0.626(0.363) (0.317)Last week of entitlement*single -0.869 0.744(0.762) (0.348)Previousweekly wage 0.241 6.609(3.036) (2.882)Benefits exhausted*married -0.722 -0.618(-0.349) (0.302)Benefits exhausted*single -1.686 -0.515(0.398) (0.346)Age* 10-' 0.564 -0.232 -0.163 0.125 -0.067 -0.359(0.3 17) (0.046) (0.092) (0.436) (0.054) (0.126)Age squared* 0I-' -11.128 -3.737(4.274) (5.754)Vocationalhigh school 0.308 0.356 0.511 1.298 0.816 1.453(0.160) (0.145) (0.252) (0.224) (0.170) (0.345)Academic high school 0.013 0.104 0.590 1.263 0.701 0.820(0.168) (0.161) (0.292) (0.216) (0.159) (0.346)


24/27

VOL.88 NO. 5 HAMET AL.: UNEMPLOYMENTN THE CZECHAND SLOVAKREPUBLICS 1139TABLE B2-Continued.


Post high school 0.021 0.038 0.803 1.144 0.818 1.372(0.236) (0.232) (0.465) (0.293) (0.242) (0.849)Romany -1.492 -1.484 -2.708 -1.931 --1.486 -1.016(0.330) (0.296) (0.448) (0.337) (0.310) (0.460)Handicapped -0.562 -0.523 -0.837 -1.059 --0.644 0.154(0.156) (0.141) (0.344) (0.304) (0.217) (0.509)Married 0.124 0.322 0.636 1.042 0.555 0.790(0.263) (0.100) (0.181) (0.307) (0.115) (0.250)Recent graduate 0.320 0.213 0.108 0.070 --0.260 -1.432(0.230) (0.141) (0.251) (0.285) (0.155) (0.454)Lives in Prague/Bratislava 0.379 0.172 1.242 - 1.154 --0.906 1.572(0.231) (0.213) (0.437) (0.315) (0.241) (2.110)Districtunemployment rateby -0.138 -0.170 -0.019 -0.076 -0.039 -0.099educationgroup (0.045) (0.044) (0.072) (0.023) (0.018) (0.049)Districtvacancy rateby education 0.086 0.080 0.312 0.080 0.102group (0.115) (0.107) (0.196) (0.080) (0.069)Industrialproduction*10-6 0.634 1.210 13.356 1.021 0.739 -2.077(1.786) (1.767) (2.116) (1.328) (1.050) (2.723)Agricultural/industrialmployment -0.338 -0.171 0.912 -0.049 -0.155 -0.395ratio (0.241) (0.233) (0.479) (0.170) (0.139) (0.322)Log duration 6.536 0.827 -3.099 0.192 3.680 -0.269(1.515) (0.225) (0.748) (0.244) (1.495) (0.084)(Log duration)2 -4.131 -0.154 4.155 -0.098 -5.157(1.013) (0.040) (0.772) (0.045) (1.791)(Log duration)3*10-' 10.449 - 14.777 27.726(2.769) (2.595) (8.904)(Log duration)4*10-2 -9.295 15.729 --62.169(2.640) (2.783) (19.428)(Log duration)5*10-3 48.809(15.451)Log-likelihood -2492.1 -2531.0 -1261.5 -3241.7 --3268.9 -590.8Notes: Standarderrors in parentheses.All equationsinclude a constant(not reported).


25/27

1140 THEAMERICANECONOMICREVIEW DECEMBER 1998TABLE B3-EXPECTED DURATION EXPERIMENTSa


Base expected duration (weeks) 17.75 18.58 15.61 60.71 55.53 54.78Benefits raised by 10 percent 0.61 0.39Wage raisedby 10 percent -0.02 -1.49Entitlement raised by 1 week 0.30 0.93Entitlement raised by 1 week-single man 0.24 0.83Entitlement raised by 1 weekmarriedman 0.34 1.00Aged 25 years v. 35 years -1.32 -3.28 -2.93 -3.57 -2.76 -11.00Aged 45 years v. 35 years 5.06 4.09 3.58 6.31 2.87 14.39Aged 55 years v. 35 years 17.78 9.23 7.82 15.26 5.85 31.93Vocational high school v. juniorhigh school -5.73 -6.71 -16.75 -48.20 -40.68 -51.04Academic high school v. juniorhigh school -0.16 -1.82 -13.65 -44.37 -32.89 -35.07Post high school v. junior highschool -1.07 -1.41 -18.46 -41.66 -39.24 -49.39Recent graduatev. nonrecentgraduate -4.24 -3.00 -1.91 -2.53 11.40 61.54Romany v. non-Romany 43.37 46.40 84.32 66.36 76.95 41.24Handicappedv.nonhandicapped 10.21 9.76 22.70 37.23 30.83 -4.56Prague (Bratislava)v. other -4.86 -2.44 -12.38 40.36 45.15 -30.93Married v. single -6.30 -4.91 -12.46 -30.23 -23.21 -24.70Unemploymentrate increasedby 10 percent 0.64 0.81 0.09 2.70 1.60 2.99Vacancy rate increasedby 10percent -0.09 -0.09 -0.45 -0.10 -0.15Industrialproductionincreasedby 10 percent -0.07 -0.14 -1.59 -0.35 -0.29 0.57Agriculture/industrialatioraisedby 10 percent 0.19 0.10 -0.62 0.10 0.37 0.75

a Based on estimates in Table B2.

8/3/2019 Unemployment and the Social Safety Net

Date post:	06-Apr-2018
Category:	Documents
Upload:	chupakambra
View:	219 times
Download:	0 times

Unemployment and the Social Safety Net During Transitions to a Market Economy Evidence From the...

Documents