stockholm unemployment & real exchange rate

EssaysonUnemploymentandRealExchangeRatesHansLindbladiic HansLindblad,Stockholm,2010ISSN1404-3491ISBN978-91-7447-173-1PrintedinSwedenbyPrintCenterUS-AB,Stockholm2010Distributor: DepartmentofEconomics,StockholmUniversityiiiDoctoralDissertationDepartmentofEconomicsStockholmUniversityAbstractThisthesisconsistsoffourself-containedessays.In the rst essay, Persistence in Swedish Unemployment Rates, the objec-tiveistostudythepersistenceinSwedishunemploymentrates. Persistenceexistsifthereisnoorweaktendencyinunemploymentratestorevertbacktopreviouslevels. Persistenceiscausedby: naturalrateshocks,longunem-ploymentcycles, andspill-overfromcyclical topermanentunemployment.Theanalysisiscarriedoutinanunobservedcomponentsmodel, whereun-employment is decomposed into a permanent, natural rate, component and acyclical component. There is evidence of high persistence; an increase in thecyclical componentof unemploymentby1percentagepointspillsoverandincreases the natural rate by 0.5-0.8 percentage points. The results also sug-gest that the quick rise of unemployment rates during 1992-1994 was causedbybothlargepermanentandcyclicalshocksincombinationwithspill-overeects.Inthesecondessay, TheEquilibriumRateof UnemploymentinaSmallOpenEconomy, we challenge the commonassumption, whenempiricallyanalyzingthedeterminantsof unemployment, thattheeconomyisclosed.However,theequilibriumrateofunemploymentisnotidentiedinanopeneconomyunless the foreignsector is modeled. Therefore, we set upandestimateastructural unobservedcomponentsopeneconomymodel fortheunemployment rate and the real exchange rate. This approach enables us tosimultaneouslydeterminechangesincyclicalaswellasinequilibriumcom-ponents of these two variables. Our estimates indicate that the foreign sectorisofsubstantialimportancewhenexplainingmovementsintheNAIRU.Inthethirdessay, ASimultaneousModel of theSwedishKrona, theUSivDollarandtheEuro, theobjectiveistoincreasetheunderstandingof ex-change rate movements. We simultaneously estimate the real exchange ratesbetweentheSwedishKrona, theUSDollar andtheEuro. Theexchangerate movements are showntobe well explainedbypotential output, theoutputgap, termsof trade, thefractionof prime-agedpeopleinthepopu-lation,andstructuralgovernmentbudgetdecits. Sincepotentialoutputisunobservable, cyclicalandpotentialoutputareestimatedinanunobservedcomponents framework together with a Phillips curve. The model works wellin out of sample exercises. The point estimates suggest that the relative bud-getsituationfortheUSversusEuropeisaprimecandidateforexplainingthechangesintheUSD/EURexchangerate.Thelastessay, Wages, Employment, andUnemployment: TheEect ofBenets,TaxesandLaborMobility,studieshowwagesandemploymentareaected by unemployment insurance and labor mobility. I show that the wageeect of higher unemployment benets canbeeither positiveor negative,dependingonthe specicationof unionutilityfunctionandthe taxationschemefornancingthebenets. Thecommonclaimintheliteraturethatwages are lower when a sector bears a higher fraction of unemployment costsdoes not holdingeneral. I alsoshowthat labor mobilityacross sectorsnormally reduces the wage level, and thereby also reduces unemployment. Itisalsoshownthatincreasedcompetitionreduceswagesandunemployment.vToAgnetha,CarolineandMadeleineviviiAcknowledgementsIn writing this thesis I have beneted from comments and support from manypeople. Ithankyoufromthebottomofmyheart. Iwouldparticularlyliketothank:AndersPalmer,forsupport,andhelpingmetofundmyrstyearofPhD-studies.AndersForslund,forvaluablediscussions.PerJansson,forfruitfulandinterestingdiscussionsandalotofhelp.PeterSellin, myco-authorontwoof thechaptersinthisthesis. Withoutoursplendidcooperationandyourskillsthisthesiswouldnotexist.Lars Calmfors, my rst advisor, for being so patient, so supportive, and foralwaysbelievinginme.TorstenPersson,mysecondadvisor,forhelpingmepassthenishline.StureandIngridGustafsson,myfatherinlawandmylatemotherinlaw,foralwayssupportingme.Per, Maja, Mats, LottaandJenny, mybrothersandsisters, foryouren-couragement.Per-Olof and Gunilla, my parents, for all your love and for always believinginme.CombiningPhD-studiesandwritingathesiswithafull timejob, hasleftaminimumof speartimewithmyfamily. Therefore, andbecauseIlove themsomuch, I wouldespeciallylike tothankCarolineandMadeleine, mybeloveddaughters, whosechildhoodshavebeendif-ferentandwhoIowealotoftime. Ihopeitisnottoolatetocatchviiiup. Youhavealwaysencouragedandbelievedinme. Yourloveandsupporthavebeencrucial. Iloveyou.LastandmostIwanttothankAgnetha, mybelovedandwisewife, foryourunconditional love, yourhelpandsupport, forputtingupwithme,andformakingmylifemuchmorefun.TableofContents1 Introduction 1References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 PersistenceinSwedishUnemploymentRates 151 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 Themodel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 Empiricalresults . . . . . . . . . . . . . . . . . . . . . . . . . 294 Summaryandconclusions . . . . . . . . . . . . . . . . . . . . 40References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44A Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473 The EquilibriumRate of Unemployment inaSmall OpenEconomy 551 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . 552 Theoreticaldiscussion . . . . . . . . . . . . . . . . . . . . . . 583 Theempiricalmodel . . . . . . . . . . . . . . . . . . . . . . . 664 Empiricalresults . . . . . . . . . . . . . . . . . . . . . . . . . 725 Conclusionsandsummary . . . . . . . . . . . . . . . . . . . . 82References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86A Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 894 ASimultaneousModeloftheSwedishKrona,theUSDollarandtheEuro 99ixx TABLEOFCONTENTS1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . 992 Theoreticaldiscussion . . . . . . . . . . . . . . . . . . . . . . 1023 Empiricalspecication . . . . . . . . . . . . . . . . . . . . . . 1064 Empiricalresults . . . . . . . . . . . . . . . . . . . . . . . . . 1135 Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125A Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1295 Wages, Employment, andUnemployment: The Eects ofBenets,TaxesandLaborMobility 1391 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . 1392 Amodelofwagesettingwithoutlabormobility . . . . . . . . 1423 Amodelwithendogenousmobility . . . . . . . . . . . . . . . 1534 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169Chapter1IntroductionDuringthebetterpartofthe1980seconomicpolicyinSwedenwasclearlyexpansionary, whichisnotsustainableinthelongrun. Atthesametime,thecredit market was deregulatedandthesavings ratefell rapidly. Oneconsequenceof thederegulationwasthatreal estatepricesroseduringthelate1980sandtheearly1990s. Intheearly1990stheeconomywasseverelyoverheated and Sweden was hit by a dramatic nancial crisis. The overheatedeconomyhadalsoledtoanappreciationoftherealexchangerateandthuslossesincompetitiveness.The situation was clearly not sustainable and eventually the bubble bursted.Expected and actual ination fell. Sveriges Riksbank (Central Bank of Swe-den) andthegovernment initiallytriedtodefendthexedexchange-rateregimethroughextremeinterestrates(500%)andscal tightening. How-ever,keepinginmindSwedenshistoryofdevaluations,thesepolicieslackedcredibility. This in combination with the size of the relative price adjustmentneeded to restore competitiveness lead to a break down of the xed exchangerateregime. Thenominal exchangeratedepreciated. Uncertaintyandthelackof credibilitykeptnominal interestrateshigh. Thusgivenfallingex-pected ination, real interest rates increased enormously while GDP fell andpublicnances deteriorated, thegovernment scal decit reacheddouble-12 INTRODUCTIONdigitgures. Theloweractivityintheeconomyincombinationwithapick-up in productivity triggered a massive rise in the unemployment rate. Openunemployment rose from less than 2% in 1989 to 8% in 1994, and total unem-ployment(thesumofopenunemploymentandparticipationinactivelabormarketprograms)increasedfrom3%to13%duringthesameperiod. SeeJonung(2003)foramoreelaboratedescriptionofthatperiod.A key question was whether the higher unemployment rate would persistasithadinmanyothercountries, orif Swedenwasdierent. Giventhisquestion, themainthemeinthisthesisistoincreasetheunderstandingofthe determinants of the Swedish unemployment rate. Since Sweden is a smallandveryopeneconomy,specialattentionisgiventotherealexchangeratewhichplaysacrucialroleinexplainingtheunemploymentdynamics.Paper 1 TheSwedishunemploymentrecordduringtheearly1990swassimilar tothedevelopment intherest of Europeadecadeearlier. Inmyrstpaper,PersistenceinSwedishUnemploymentRates,themainquestionis whether the increase in Sweden is temporary or if the high unemploymentgureswouldpersist,asithadinmanyotherEuropeancountries.1InSwe-den, activelabormarketprograms(ALMPs)havebeenanimportantpartof thelabormarketduringmuchof thepost-warperiod. If suchprogramsare successfull, persistence would be lower and the unemployment rate woulddecreaserapidly. However,ifALMPspushwageshighertheywouldworsentheproblem.Persistence or hysteresiscould bedue to severalreasons. Layard,NickellandJackman(1991)suggestthattemporaryunemploymentmightleadtosuchadepreciationofhumancapitalthattheprobabilityofre-employmentis considerably reduced for the unemployed. Blanchard and Summers (1986)suggest that reductions in the capital stock associated with reduced employ-mentafter anadverseshockmightreducethedemandfor labor. Finally,1This paper is my Ph-Licentiate thesis, examined in December 1996 and has not beenrevised since then.3BlanchardandSummers (1986), Gottfries andHorn(1986) andLindbeckand Snower (1988) suggest that insiders might be important in the wage for-mationprocess. Iftheinsiders, whennegotiatingwages, donotcareaboutoutsiders, the unemployed, wages will be set too high to increase employment.There was no consensus in the empirical literature in the rst part of the1990sregardingthepersistenceof theSwedishunemploymentrate. Holm-lund (1993) suggested that the natural rate of unemployment (NAIRU)wasslightlyhigherinthebeginningof the1990sthantwodecadesearlier.Forslund(1995)arguedthattheriseoftheopenequilibriumunemploymentratecouldbeof theorderof magnitudeof 1-4percentagepointsbetween1990 and 1993. Assarsson and Jansson (1998), on the other hand, suggestedthattheunemploymentrateswereverypersistentandthattheNAIRUhadincreasedatleastasmuchastheunemploymentrateandthatthecyclicaluctuationsintheunemploymentratewereverysmall.AssarssonandJanssonsresultsweretooprovocativetobeleftunchal-lenged. Inordertoinvestigatetheexistenceandtheextentof persistenceinSwedishopenandtotalunemployment,andtotesttherobustnessoftheresults inAssarssonandJansson, I developedanunobservedcomponents(UC)labor-marketmodel. TheUCmodeliscombinedwiththewellknownLayard,NickellandJackmanmodelforprice-settingandwage-settinginanopeneconomy. Theresultingmodelisidentiedindierentways. Theesti-mated model incorporates the appealing time series features of the UC modeland has a clear economic interpretation. In addition, previous studies on per-sistence had only addressed open unemployment. As mentioned above, activelabormarketprogramshavealwaysbeenanimportantpartoftheSwedishlabormarket. Thus, therstpaperalsoinvestigatesiftotalunemploymentexhibitspersistancepatternssimilar tothoseofopenunemployment. Inthemodel, persistencecanbecausedbythreefactors: permanentshocksmaychange unemployment permanently, unemployment may contain a cycle withlongperiodicity,andcyclicalshocksmayspillovertopermanentunemploy-4 INTRODUCTIONment(hysteresis).Inordertoinvestigatethestabilityoftheestimates,themodelsareesti-mated over two dierent time periods, 1972-1994 and 1972-1991. The reasonforsplittingthesamplein1991wasthatunemploymentincreaseddramati-callyduringtheperiod1992-1994.Theresultssuggest, asinAssarssonandJansson, largespill-overeectsfromcyclical topermanentunemploymentandthattheNAIRUincreasedsubstantially duringthelastyearsofthesample. However,whenshorteningthe sample, the point estimates of the spill-over parameter are reduced. Theestimatedparameterisingeneral notsignicantlydierentfromzero, andsignicantlydierentfromthelong-samplepointestimates.Inthecaseoftotalunemploymentandthelongsample,thereisnoevi-dencethattotal unemploymentreinforcestheunemploymenteectsontherst dierence of ination. However, there is robust evidence of a signicantspill-over eect, andcontrarytothe openunemployment case, the pointestimates of the spill-over eect are not signicantly dierent from the short-sample estimates. The average periodicity of the cyclical component increasessomewhat,comparedtothecasewithopenunemployment. Inotherwords,ALMPsincreasepersistence.It is important to recognize that, up to 1991, both open and total unem-ploymentratesinSwedenwerelowandstable. However, thevariationsintheshortsamplemadeitpossibletotraceoutasignicantspill-overeectsin some cases. In total, a signicant spill-over eect is detected in 7 out of 10cases, and in 6 out of 10 cases for open and total unemployment respectively.Thus,thereisfairlyrobustevidenceforasubstantialspill-overeect.Acommonassumptionwhenanalyzingunemploymentisthattheecon-omyisclosed. Inempiricalresearch,suchmodelsareoftenusedevenwhendealingwitheconomieswhereforeigntradeconstitutesaconsiderablefrac-tion of the economy. However, Layard, Nickell and Jackman (1991) and Bean(1994) show that the NAIRU is not identied in an open economy unless the5foreignsectorismodeled. Omittingtheforeignsectorcouldthereforepo-tentiallybiastheNAIRUestimates. AclosedeconomyalternativemodelisnestedinthemodelthatIsetup. Whentested,theclosedeconomymodelcannotberejected.Amainresult of therst paper is that openandtotal unemploymentexhibitsimilarpatternsof persistence. Moreover, theresultssuggestthatthe favorable Swedish unemployment experience up to 1991 was due to smallshocks, ratherthantothepresenceofspill-overeectsfromcyclical unem-ployment to the NAIRU. The unemployment development during 1992-1994,however, seems tobecausedbylargeshocks as well as thepresenceof asubstantialspill-overeect. Comparingtheseresultswiththeresultsinthenextchaptersuggeststhattheidenticationof themodel over-empathizestheroleofthepermanetcomponent.Onepolicyconclusionfromthemodelwouldbethattoreducethehighunemploymentrateintheearly1990soneoughttostimulatetheeconomyand thereby reduce the rate of cyclical unemployment. The output gap eecton ination would be short lived since the large spill-over eect would reduceNAIRU quite rapidly. In 1996, Sveriges Riksbank indeed lowered the interestratebynearly500basispoints,from8.91percentto4.10percent.Paper 2 In 1997, the Swedish unemployment rate started to decline and bymid 2001, it had reached the lowest level since the early 1990s. Obviously, itis important to know whether this decline in unemployment reects a changein NAIRU or a narrowing of the unemployment gap. Apel and Jansson (1999)suggestthattheNAIRUincreasedduringthersthalfofthe1990s. Turneretal. (2001)foundthatNAIRUincreasedsubstantiallyfrom1980to1995.Mysecondpaper, TheEquilibriumRateof Unemployment inaSmallOpenEconomy, writtentogether withPeter Sellin, is tosome extent anupdate of the rst paper. Central questions are how the NAIRU has evolved,andwhatexplainsthemovements. Moreover, theopeneconomyaspectis6 INTRODUCTIONrevisitedandmodeledandestimatedinamoreelaborateway.Onepolicyconclusionfromtheresultsintherstpaperistostimulatethe economy when unemployment is high. Large spill-over eects from cycli-cal to permanent unemployment implies that an expansionary policy reducestheNAIRUinthenearfuture. Thus,inordertodesignappropriatepoliciestocombatunemploymentitisimportanttoknowtowhatdegreeobservedunemploymentisduetoalackof demandandtowhatdegreeitisduetostructuralfactors. EventhoughtherstpapersuggeststhattheNAIRUisaected by the cyclical unemployment rate, theory tells us that it is also ex-plained by a set of structural factors, e.g. taxes, active labor market policies,thereplacementratioanddemographicfactors.Asdiscussedabove, omittingtheforeignsectorcanbiastheNAIRUes-timates. Itisstraightforwardtoshowthatinanopeneconomytheunem-ploymentratewill, atleastpartly, bedrivenbythereal exchangerate. Adepreciationof thereal exchangerateincreasesconsumerprices. Inordertorestoreconsumerrealwages, realwagecostsforrmsmustincreaseandthis reduces employment. The determinants of the real exchange rate shouldthusbeincludedinthesetofstructuralfactorsdeterminingtheNAIRU.Instead of modelling the NAIRU as a stochastic trend, which I did in therst paper,we now model it as a function of a set of theoretically motivatedvariables. OurapproachtakestheunobservedcomponentsmodelofSalemi(1999)astepforwardbysettingupanopeneconomyframeworkandsimul-taneouslyestimatetherateof unemployment andthereal exchangerate.Toidentifythemodel,wealsoestimateanexpectations-augmentedPhillipscurve.As it turns out,we can safely reject the assumption of a closed economy.Indeed, the foreignsector turns out to be quite important inexplainingthedevelopment of theunemployment rateandtheNAIRU. Ignoringtheexistence of the foreign sector will thus bias the results, and according to thepointestimates,thebiascanbesubstantial.7Our analysis successfully explains the development of the Swedish unem-ploymentrate. Adepreciationof thereal exchangerate, ahigherreplace-mentratioandhighertaxeswill raisetheNAIRU, whereasanincreaseinthe size of labor market programs tends to reduce it. According to our pointestimates, theNAIRUincreasedbyapproximately1-1.5percentagepointsduring 1980-1985, and remained quite stable thereafter. This nding impliesthat thedramaticunemployment hikeinthe1990s was mainlyacyclicalphenomenon, whichisquiteadierentconclusionfromtheonereachedinpaper1. Thedierenceisprobablyduetotheidenticationof themodelandtheassumptionintherstpaperthatNAIRUisarandomwalk.The real exchange rate is successfully modeled as a function of the termsof trade, the evolution of demographics relative to the rest of the world, andthedomesticandforeignstructural budgetdecits. Itisassumedthatnetforeign assets which are often treated as exogenous to the real exchange ratearedeterminedbydemographicvariablesandstructural budgetdecits.Our focus ondemographics relative tothe rest of the worldandthe useof structural decits intheexchangerateequationis anovelty. Alargerfractionofthepopulationaged45-59yearsandalargerdomesticstructuralbudgetdecitdepreciateoftherealexchangerate,whileimprovedtermsoftrade and a larger foreign structural budget decit appreciate it. Our modelexplainsasurprisinglylargefractionof thevariationinthereal exchangerate, especiallykeepinginmindthatthesampleperiodspansnolessthanve devaluations and a shift from a xed to a oating nominal exchange rateevents that wedonot control for. Wealsondthat thereal exchangeratedoes not revert faster totheequilibriumreal exchangerate(REER)duringtheoating-rateperiod,thaninthexedrateperiod. Thusitseemsas adjustments to the real exchange rate during the xed-rate period mainlytook place through devaluations and not through changes in the relative pricelevel.ThePhillipscurveseemstobealiveandwellandchangesintherate8 INTRODUCTIONof ination are well explained by the unemployment gap and dierent supplysideshocks. Aninterestingresultisthatpersistenceintheunemploymentrateinuencetheresults. Thechangeintherateofinationisaectedbythe change in the unemployment gap, indicating that if the gap is closed toorapidly,thismaycauserisingination.Paper3 Giventheresultsinpaper2,theexchangerateappearstobeofgreat importance, boththeoreticallyandempirically, whenexplainingthedevelopment of the unemployment rate. Therefore, it is troubesome that theevolutionof areal exchangerateishardtounderstand, makingithardtoanalyzethemovementsinunemployment. Itisoftenclaimedthatthebestguessabouttomorrowsrealexchangerateistodaysgoingrate.Inmythirdpaper, ASimultaneous Model of the Swedish Krona, theUSDollar andtheEuro, writtentogether withPeter Sellin, wechallengethisconclusionandtrytoexplainthelarge,andoftenpersistent,swingsweobserveinasetofrealexchangerates.Themostcommonapproachwhenestimatingreal exchangeratesistostudyabilateral exchangerateortocompareonecurrencytoabasketofother currencies, adjusting for relative price levels. The challenge and contri-bution of this chapter is to estimate a small system of dierent real exchangerates in order to simultaneously analyze movements in the real exchange ratebetweentheKrona, theEuroandtheDollar, themostimportantreal ex-changeratesforSwedishexports. Atri-polarmodelfortheGermanMark,the US Dollar and the Japanese Yen has previously been estimated by Mac-Donald,R.andI.Marsh(1999).Ourempirical model isbasedontheoretical resultsfromthenewopeneconomymacroeconomics literatureinspiredbytheworkof ObstfeldandRogo(1995). Inourspecicationtherealexchangerateisdeterminedbyterms of trade, relative potential output and a set of exogenous variables thatare assumed to determine the net foreign asset position. Potential output in9Sweden,theEuroareaandtheUSisestimatedseparatelyinanunobservedcomponents framework, including an expectations augmented Phillips curve.Ourresultsarepromising. Weareabletosurprisinglywell explainthevariationinoutput, inationandinthechangeof thereal exchangerates.Our point estimates are in line with theory, of expected magnitude and highlystatisticallysignicant.An increasein thetermsof trade,innet foreign assetsor inrelativepro-duction, in one country raises wealth which, in turn, stimulates consumptionofnon-tradedgoods. Inaddition,higherincomereduceslaborsupplywhichreduces the production of non-traded goods and thus raises the relative priceof non-tradables. Both these eects appreciate the real exchange rate. How-ever, netforeignassetscannotbetreatedasexogenous. Weassumethattherelativedemographicsituationandrelativestructural scal positionofthe government are exogenous determinants of net foreign assets and the realexchangerate.Our results implythat potential output growthis highest inthe US,followedbytheEuroarea. Thisisnotverysurprisinggiventhefactthatactual growthisrankedinthesameway. ThePhillipscurveseemstobesteepestintheUS. InationismoresensitivetotheoutputgapintheUSthaninSwedenandtheEuroarea. Anincreaseinrelativepotentialoutputstrengthens,asexpected,therealexchangerate. TheKronaandtheDollarbenetmorefromthiseectthantheEuro. Anincreaseinexportpricesrelative to import prices i.e. the terms of trade strengthens the currencyof the exporting country, which again is as expected. It seems as if the Euroareabenets themost fromanincreaseinexport prices. Anincreaseintherelativemiddle-agecohortsisfoundtoweakentheexchangerate, butthisresultisnotveryrobust. Thestructural budgetsituationis, however,ofgreatimportanceandanincreaseinthedecithasadepreciatingeect.The point estimates indicate that this eect is larger for the US and the Euroarea than for Sweden. Our results indicate that an increase in the US decit10 INTRODUCTIONby one per cent of GDP could depreciate the USD/EUR by as much as 8 percent.Ourexchangeratemodelsseemtobequitestable. Inanout-of-sampleexercise, stretchingover8years, weareabletocloselytracktheactual de-velopmentofthethreerealexchangerates.Moreover, for countries that havejoinedtheEuropeanUnionbut notyet adoptedthe Euro, it is important tounderstandthe development ofreal exchangerates. Whendeterminingthenominal conversionrate, itisimportantthatitdoesnotinduceaninationshockinordertorestoreaneconomicallymotivatedreal exchangerate. Webelievethatourmodel canbeofsomehelpinthisrespect.Paper 4 The macro labor literature often appeals to a mechanism wherebywagesaresetasamark-upontheunemploymentbenetlevel, suchthathigherbenetswill raisewagesandreduceemployment. Theempirical re-sultsare, however, notclear-cut(e.g. Calmfors, ForslundandHemstr om,2004). One way of getting around the adverse, wage-raising and employment-dampening,eectmightbetonanceunemploymentbenetsinsuchawayas to create incentives for lower wages or not laying o workers. An exampleisanexperience-ratedpay-rolltax, wherethepay-rolltax, atleasttosomeextent, issettoreect individual employers lay-ohistory(e.g. KruegerandMeyer, 2002). However, at thesametimehiringcosts areincreasedwhichreducestheincentivetohireworkers(Hamermesh, 1993). HolmlundandLundborg(1988, 1989)andCalmfors(1995)suggestthatwagesettingandemployment canbeinuencedbydierentiatingthecontributions tounemploymentsupportbyareasofwagebargaining. Theideaistogetthebargaining parties to internalize some of the costs for unemployment, therebyincreasingtheperceivedbenetsofwagemoderation.Inmylastchapter, Wages, Employment, andUnemployment: TheEf-fects of Benets, Taxes and Labor Mobility, I analyze how wages, employment11and unemployment are aected by introducing labor mobility in a model withdierentsectors. Withinthisframework,Ichecksomeoftheestablishedre-sults on how wages and employment respond to changes in the unemploymentbenet level, and to changes in the fraction of unemployment costs borne byeachsectorintheeconomy. Regardingtheanalysisof theunemploymentinsurance, Irevisitsomeearlierresults, whiletheanalysisof mobilityandwagesisanovelty.Thegeneral resultisthatintersectoral labormobilitycanbothreduceand increase the incentive for high wages, depending on the elasticity of labordemand. However, for elasticity values often found in the empirical literature,mobilityreducestheincentiveforhigherwages. Thereasonisthathigherwagesinasectorincreaselaborsupplyandunemployment, whichreducesthegainsfrompushingupwages. Giventhatlabormobilityreduceswages,it is also shown that wages decrease in the number of sectors in the economy.Thus,competitionreduceswagesandincreasesemployment.Moreover, I showthat the wage level is unambiguouslydecreasingintheweightonemploymentintheunionutilityfunctionthisresultholdsregardlessofwhetherlaborismobileornot.Introducingataxschemewherethebargainingpartieshavetoconsiderthat their actions aect thetaxrate, is expectedtomitigatetheadverseeectsonwagesandemploymentfromunemploymentbenets. Thisresultisshowntoberobusttodierentspecicationsof utilityfunctionsonlyaslongaslaborisnotmobile. However,whenlaborismobiletheresulthingesonthespecicationoftheutilityfunction.It is also shown that a standard result in the literature, that an increasedunemployment benet level increases the wage level, does not hold in general.A policy conclusion from the analysis in this last chapter is that lower un-employment can be achieved either by making unions care about employmentorbyfacilitatinglabormobility.12 INTRODUCTIONReferences[1] Apel,M.andP.Jansson(1999).SystemestimatesofpotentialoutputandtheNAIRU,Empirical Economics,Vol24,No.3:373-388[2] Assarsson, B. andP. Jansson(1998). Unemploymentpersistence: thecaseofSweden,AppliedEconomicsLetters,Vol5,Issue1: 25-29[3] Bean, C. (1994). European unemployment: a survey, Journal of Eco-nomicLiterature32:573-619[4] Blanchard, O.J. and L.H. Summers (1986). Hysteresis and the EuropeanUnemploymentProblem. In S. Fischer (ed) NBER Macroeconomics An-nual1986.MITPress,Cambridge,Massachusetts.Vol.1:15-78[5] Calmfors, L. (1995). Labour market policy and unemployment, Euro-peanEconomicReview39:583-592[6] Calmfors, L., A. Forslund, andM. Hemstr om(2004). The eects ofactive labor-market policies in Sweden: What is the evidence?, in J Agell,M Keen and J Weichenrieder (ed), Labor Market Institutions and PublicRegulation,MITPress.[7] Forslund, A. (1995). UnemploymentIs Sweden still Dierent?SwedishEconomicPolicyReview,Vol.2,No.1:15-58[8] Gottfries, N. and H. Horn (1986). Wage Formation and the PersistenceofUnemployment,TheEconomicJournal,Vol.97,No.388:877-884[9] Hamermesh,D.S.(1993).LaborDemand.PrincetonUniversityPress.[10] Holmlund,B.(1993).Arbetsloshetskrisenkonjunkturfenomenellersys-temfel (The Unemployment CrisesABusiness Cycle Phenomena orStructural Imbalance?). Supplement12toNyavillkorf orekonomi ochpolitik,SOU1993:16,NorstedtsTryckeriAB,Stockholm.REFERENCES 13[11] Holmlund,B.andP.Lundborg(1988).UnemploymentInsuranceandUnion Wage Setting, ScandinavianJournalofEconomicsVol.90,No.2:161-172[12] Holmlund, B. and P. Lundborg (1989). Unemployment InsuranceSchemes for Reducingthe Natural Rate of Unemployment, JournalofPublicEconomics38.[13] Krueger, A. B. andB. D. Meyer(2002). Laborsupplyeectsof socialinsurance.NBERWorkingPaperseriesNo.9014.[14] Layard, R., S. Nickell and R. Jackman (1991). Unemployment: Macroe-conomicPerformanceandtheLabourMarket.OxfordUniversityPress,Oxford.[15] Lindbeck, A. and D.J. Snower (1988). The Insider-Outsider The-oryof Employment andUnemployment. MITPress, Cambridge, Mas-sachusetts.[16] Lindblad, H. and P. Sellin (2010). A Simultaneous Model of the SwedishKrona,theUSDollarandtheEuro.Chapter4inthisthesis.[17] MacDonald, R. and I. Marsh (1999). Currency spillovers and tri-polarity: asimultaneousmodel of theU.S. dollar, GermanmarkandJapaneseyen,JournalofInternationalMoneyandFinanceVol.23:99-111.[18] Obstfeld, M. andK. Rogo(1995). Exchangeratedynamicsredux,Journal ofPolitical Economy91,624-660.[19] Salemi, M.K. (1999). Estimating the natural rate of unemployment andtestingthenatural ratehypothesis,Journal of AppliedEconometricsVol.14:1-2514 INTRODUCTIONChapter2PersistenceinSwedishUnemploymentRates1 IntroductionSwedenexperiencedadramaticriseintherateofunemploymentduringtheearly1990s. Openunemploymentrosefrom1.7%in1990to8.0%in1994,andtotalunemployment(thesumofopenunemploymentandparticipationinactivelabormarketprograms)increasedfrom2.9%to13.3%. ThisriseisillustratedinFigure1. During1995theguresmoderatedslightly.TheSwedishunemploymentrecordduringtheearly1990sissimilartothe development in the rest of Europe in the late 1970s and the beginning ofthe1980s. AcrucialquestioniswhethertheSwedishincreaseistemporaryorifthehighunemploymentgureswillpersist,asinmanyotherEuropeancountries.This essay is my Ph-Licentiate thesis, examined in December 1996 and has not beenrevised since then. I would like to thank Lars Calmfors, Anders Forslund, John Hassler, PerJansson, Christina Lonnblad, Stefan Palmqvist, Brian Reinhold, Paul Soderlind and sem-inar participants at the IIES for helpful comments. Financial support from the Ministryof Finance and the Swedish Council for Research in the Humanities and Social Sciences isgratefully acknowledged.1516 PERSISTENCEINSWEDISHUNEMPLOYMENTRATESPrevious empirical studies on the natural rate of unemployment suggestsatleastfourdierentwaystoapproachthisquestion.(1)Holmlund(1993) estimates recursivelyanexpectations-augmentedPhillips curve model. His results suggest that the natural rate of open unem-ploymentwashigherinthebeginningofthe1990sthantwodecadesearlier.However, thechangeissmall anduncertainwhichmightbeduetoseveralfactors. Holmlunds sample ends in 1992 and the method used is not withoutproblems. Thedevelopmentduring1994and1995, withrisingwageina-tion, might, however, beinterpretedasanindicationthatthenatural ratehasincreased.(2)Analternativeapproachistoestimateastructuralmodel. Forslund(1995) estimates a dynamic versionof the Layard, Nickell andJackman(1991)model (henceforthdenotedtheLNJmodel)forpriceandwageset-tingunderimperfectcompetitionandareduced-formunemploymentequa-tion. His results indicate that the rise of the open equilibrium unemploymentratecouldbeof theorderof magnitudeof 1-4percentagepointsbetween1990and1993. Thisspreadislargewhenconsideringthatunemploymentincreasedby6.5percentagepointsduringthatperiod.(3) Univariatetimeseriesmodelshavealsobeenusedforstudyingthepersistenceinopenunemploymentrates(AlogoskousandManning, 1988;Barro, 1988; Elmeskov, 1994). Animportant result is that thedegreeofpersistenceseemstobehighinEuropeandloworlowerintheUS.However, aunivariateanalysis does not recognizethat supplyandde-mandshocksmightaectunemploymentindierentways. Supplyshocks,like productivityshocks andchanges ininstitutional settings, are usuallyexpectedtoaectunemploymentpermanently,whereasdemandshocksandstabilizationpolicyoftenareassumedtohavetransitoryeectsonly. How-ever, ademandshockmightalsopermanentlyaectunemployment. Ifthetransition following a shock is slow, temporary unemployment might be quitelong lived. There might be several reasons for this longevity. One explanation1. INTRODUCTION 17mightbethateventemporaryunemploymentcanleadtosuchadeprecia-tionofhumancapitalthattheprobabilityofre-employmentisconsiderablyreducedfortheunemployed(Layard,NickellandJackman,1991). Anotherexplanationisthatreductionsinthecapital stockassociatedwithreducedemployment after anadverseshock reducethe demandfor labor (BlanchardandSummers, 1986). Finally, insidersmightbeimportantinthewagefor-mationprocess(BlanchardandSummers, 1986; GottfriesandHorn, 1986;LindbeckandSnower,1988).(4)Thetimeseriespropertiesoftheopenunemploymentratehavealsobeen studied in an unobserved components model, allowing for dierent typesof shocksandlettingthecyclical stanceof unemploymenthavepermanenteects(JaegerandParkinson, 1994; AssarssonandJansson, 1995). Inthismodel theunemployment rateis decomposedintoanon-stationaryandastationarycomponent. The results inAssarssonandJanssonconcerningSweden, indicate that the non-stationary part closely follows the actual openunemploymentrate,i.e. thatthecyclicaluctuationsareverysmall. More-over, accordingtotheirestimates, 77%ofthecyclical unemploymentspillsover to the permanent component. This eect is very large in an internationalcomparison.The time series approach is interesting and reveals some important prop-ertiesoftheunemploymentseries, buttosomeextentitlacksatheoreticalunderpinning. Thisdeciencyisnotpresentinthestructuralmodels.This paper has three objectives. First, the time series approach in JaegerandParkinsonisgivenmorestructurebycombiningitwiththeLNJopen-economymodel. Thus, theresultingmodel will incorporatetheappealingtimeseriesfeaturesoftheUCmodelandhaveamorecleareconomicinter-pretation.Second,thiscombinedmodel isusedto test therobustnessofthe resultsinAssarssonandJanssonregardingpersistenceandlargespill-overeectsfromthecyclicaltothepermanentcomponentofunemployment.18 PERSISTENCEINSWEDISHUNEMPLOYMENTRATESThird, previousstudiesonpersistencehaveonlyaddressedopenunem-ployment. InSwedenactivelabormarketprograms(ALMPs)areanimpor-tantpartof thelabormarket. Thispaperalsoinvestigatesif total unem-ployment exhibits patterns similar to those of open unemployment regardingpersistence and the spill-over from the cyclical to the non-stationary compo-nent. Thus, this section provides new evidence regarding persistence in totalunemploymentrates.2 Themodel2.1 TheunobservedcomponentsmodelSinceoneoftheobjectivesofthispaperistotestiftheresultsinAssarssonand Jansson for Sweden are robust, the unobserved components (UC) modeloftheunemploymentrateusedbytheseauthorsisadopted.There is persistence if there is no or weak tendency for unemployment torevertbacktopreviouslevelsafterachange. Persistencemaybecausedbythree factors, either individually or in combination. First, natural rate shockschangeunemploymentpermanently. Second, unemploymentmaycontainacycle with long periodicity, i.e. it takes a long time for cyclical unemploymenttorevertbacktopreviouslevelsafterbeinghitbyashock. Third, cyclicalshocksmayspill overtopermanentunemployment. Thus, spill-overeectsimplypersistencebuttheconverseisnottrue.Unemployment,u,isassumedtoconsistoftwounobservedcomponents,anon-stationarycomponent,un,andastationarycomponent,uc. Hence,ut= unt+uct. (2.1)A non-stationary process can be modelled in many dierent ways. Here, thepermanentcomponentisassumedtofollowarandomwalkplusaspill-overterm.2. THEMODEL 19Therandomwalkassumptionisanimportant,butnotproblem-free,re-strictiononwhichtheresultsareconditioned. Quah(1992)showsthatarandom walk specication is biased towards establishing the permanent com-ponent as important. However, if only employed workers have insider powerin wage setting, employment will follow a random walk (Blanchard and Sum-mers,1986).Unemployedworkers arenot likelytoturnintooutsiders immediately.Nor will adepreciationof humancapital duetounemployment occur di-rectly. Below, itisassumedthatthespill-overfromcyclical topermanentunemploymentoccurswithatwo-periodlag(i.e. aoneyearlagintheem-pirical partbelow). Thislagassumptionisnotcrucial, butittsthedatawell. Anexpressionforthenon-stationarypartofunemploymentisthusunt= unt1 +nt+uct2, (2.2)wheremeasuresthespill-overeectofinterest, [0, 1].Thecyclicalcomponentisdescribedbythefollowingstationaryprocessuct= 1uct1 +2uct2 +ct, (2.3)wheretheparameters1and2provideameasureoftheperiodicityofthecyclical component. Stationarityimpliesthat therootsof thepolynomialequation1 1L 2L2= 0,where L is the lag operator, should lie outside the unit circle. The assumptionof anAR(2)processforthecyclical componentcanberelaxed, butintheempiricalworkitisfoundthatittsthedatawell.The innovations it are assumed to be mutually uncorrelated and normallydistributedwithvariances2i,i = n, c. Forthemodeltomakesense,aunitrootintheunemploymentrateisanecessaryandtestableassumption.20 PERSISTENCEINSWEDISHUNEMPLOYMENTRATES2.2 AlternativeidentifyingequationsTheUCmodel givenbyequations (2.1)-(2.3) is ingeneral not identied,i.e. thesecondmomentsoftheobservedvariableareinsucienttouniquelydetermine all parameters of the model. In order to identify the model, infor-mationonanadditionalvariablethatisrelatedtoeitherthepermanentorthecyclicalcomponentisneeded.Jaeger andParkinsonassume that mean-adjustedreal growthrate ofoutputintheG-7countriesisrelatedonlytochangesinthecyclicalunem-ploymentrateineachcountry. AssarssonandJanssoninsteaduseamean-adjustedgrowthrate of aworldindustrial productionseries. Hence, theidentifyingassumptioninthesepapers is that mean-adjustedforeignpro-ductiongrowthratesareexogenouswithrespecttocountry-specicnaturalrateshocks. AssarssonandJanssonclaimthatalthoughnotproblem-free,wenotethatthisidentifyingassumptionisconsistentwithmosttraditionalmodels of the labour market in that it implies that structural domestic factorsareconsideredtobethekeydeterminantsofacountrysnaturalrate.Inthemodelbelow,dependinguponhowtheequilibriumisdened,thenaturalratemayormaynotdependuponthedemand/outputgapabroad.Moreover, theinterpretationof theidentifyingequationinAssarssonandJansson is unclear. The cyclical unemployment in Sweden can obviously notbe of any great importance in explaining the world output gap. Instead theiridentifyingequationshouldbeseenasrepresentingaco-movementrelation-shipwithoutcausalityimplications.Inthispaper, theLNJopen-economymodel forpriceandwagesettingisusedinordertoderiveanidentifyingequation. Themodel will actuallysuggest two equations, where either can be used to achieve identication. Inaddition, a version of Phelps (Phelps, 1968) expectations-augmented Phillipscurveistestedforcomparison.Asimpliedandlog-linearizedversionof theLNJmodel isbrieypre-sentedbelow. Themodel ishighlysimpliedwithrespecttoinformation,2. THEMODEL 21butitfollowstheusualconventionthatfocusesonlyonpriceuncertainties.Thegeneral ideaisthatreal wagesandemploymentaredeterminedbytworelationships, describingrmandunionbehavior. Itisassumedthatboth product and labor markets are characterized by imperfect competition.Firmsareassumedtohaveperfectforesightaboutpriceandwagelevels.Pricesaresetasamark-uponwagecosts. Changesindemandmayinu-ence price-setting because the marginal productivity of labor depends on theamountoflaborused. Theprice-settingbehaviorcanbemodelledasp w = 0 +1(y yn), (2.4)wherep,w,y,andynrespectivelydenotetheproducerpricelevel,thenom-inal wagecostincludingpayroll taxes, actual demand, andequilibriumde-mand. Output is determinedbyrms supplyingwhatever is demanded.Lower-caselettersdenotenatural logarithmsunlessotherwiseindicated. Itis unclear what is to be expected with respect to 1. On the one hand, a pos-itive value indicates that decreasing returns to labor tend to make the mark-up an increasing function of demand. On the other hand, empirical evidenceindicatesthatthemark-upisinsensitivetochangesindemand, suggestingthat 1is close to zero (Layard, Nickell and Jackman, 1991; Forslund, 1995).The deviation of demand from its equilibrium level is related to the labormarketbyanOkuns-lawequationy yn= (u un), (2.5)whereundenotestheequilibriumunemploymentrate. Theexpectedsignofispositive.Wagesettingisdescribedbyarelationshipthatcanbederivedfrome.g.bargainingmodels (Layard, Nickell andJackman, 1991; Blanchower andOswald, 1994; Forslund, 1995). Bargainingisassumedtotakeplaceatthermorindustrylevel. Theoutcomeisthatwagesaresetasamark-upon22 PERSISTENCEINSWEDISHUNEMPLOYMENTRATESthevalueofthebargainingpartiesalternativepay-os.The value of the alternative to being employed at the current rm dependsontwosetsoffactors: theprobabilityofendingupinotherstatesandtheincomesinthesestates. IntheSwedishcase, itisreasonabletodistinguishbetweenthreedierentstates: employmentelsewhereintheeconomy,openunemployment, and participation in a labor market program. In this section,itwillbeassumedthatparticipationinalabormarketprogramisaperfectsubstituteforemployment(thisassumptionisrelaxedinsection3.2). Theprobabilityof endingupinemploymentelsewhereandreceivingthegoingmarket wage, depends on the employment rate in the economy. Individuals inopen unemployment are assumed to receive a given fraction, the replacementratio,oftheprevailingmarketwagesasanincome.Wagesettersareassumedtobeconcernedaboutthereal consumptionwages,wc,i.e. take-homewagesinrelationtoconsumerprices,wc= w + ln(1 T) ln(1 +S) pc,where Tis the average income tax rate, Sis the payroll tax rate, and pc is theconsumerpricelevel. Therealproductwage,wp,isdenedasw p. Thus,thereisawedgebetweenthereal productwageandthereal consumptionwage,wp wc= ln(1 +S) ln(1 T) +pc p.Thewedgeconsistsof twoparts, ataxpartwhichdependsonpayroll andincometaxes, andapricepartwhichconsistsof theratioof theconsumerpricetotheproducerprice. The(logarithmof the)consumerpricecanbeexpressedasaweightedaverageof productandimportprices, pc=p +(1 )(e +p),whereeisthe(logarithmofthe)nominalexchangerate,pthe(logarithmofthe)foreignpriceleveland 1.74 THEEQUILIBRIUMRATEOFUNEMPLOYMENTinequation(3.16). Theopeneconomymodel assumptionis examinedbytesting the hypothesis that the exchange rate parameter is zero in the unem-ployment equation reported in Table 2, and by examining dierent statisticalproperties. Looking at the exchange rate parameter, we reject the hypothesisof a closed economy framework at the 5 percent level in 3 out of 4 cases. Theclosedeconomyassumptionalsogiveclearsignsofmisspecicationintermsof serial correlation in the residuals. Also, the goodness of t is poorer in theclosedeconomymodel. MoreovertheexclusionoftheforeignsectorresultsinaseverelymisspeciedPhillipscurve. Theresidualsareserialcorrelated,andnot normalydistributed. Themissecicationof thePhillips curveisextratroublesomesinceitisusedforidentifyingthewholemodel. Theseresultsareimportantandshowthatmodelingtheeconomyasclosedmightgivebiasedresults.Inthe estimatedopeneconomymodel, as expected, adepreciationofthereal exchangerateraisesthenatural rateofunemployment/NAIRU. Adepreciation of 10 percent will increase the NAIRU by 0,2 percentage pointsoraround5-6percent. Theintuitionisthatadepreciationtendstopushupconsumer prices. Inorder tocounteract areductionof theconsumersreal wages, wages have to increase. However, higher wages will raise the realproduct wage level. Higher real product wages will reduce labor demand andincreaseunemployment.Anincreaseinlabormarketprogramsreducesthenatural rateof openunemployment. An increase in ALMP by 1 percent of the labor force reducesopenunemploymentbyapproximately0,4percetagepoints. Thisisinlinewiththeresult of other studies (Calmfors et al., 2004). Thereplacementratio has the expected positive sign, but is imprecisely estimated. This couldbe due to the treatment of the replacement ratio as exogenous. Forslund andKolm(2000)foundanempiricalrelationshipbetweenthereplacementratioandtheaccommodationrate. AhighertaxwedgeleadstoanincreaseintheNAIRU, asexpected. Thisindicatesthattheconsumerreal wagedoes4. EMPIRICALRESULTS 75not fully accommodate changes in the wedge. The estimated variance of thecyclical shockislargerthanthatofthenatural rateshocks. ThisisinlinewiththeresultsinLindblad(1997),butcontrarytotheresultsinApelandJansson(1999).Figure1showstheactual, natural andcyclical ratesof unemploymentaccordingtoourestimates. ThelevelsoftheNAIRUandthecyclicalcom-ponentshouldbeinterpretedwithgreatcautionduetotheidentifyingre-strictionthatthesamplemeanoftheactualrateofunemploymentandtheNAIRU are the same. The results suggest that the NAIRU increased by 1-1.5percentage points during 1980-85 and has remained more or less stable sincethen. Thus, during the last decade of the sample the NAIRU increased little,which implies that the dramatic changes in the Swedish unemployment ratesduringthe1990smainlywasacyclicalphenomenon.These results are in contrast to the results in Lindblad (1997), where thepermanentpartoftheunemploymentrateroseatleastasfastastheactualrate during the early 1990s. In Lindblad (1997) we assumed that the NAIRUfollowed a random walk and was aected by a spill over from the cyclical partof unemployment. The random walk specication can be biased establishingthe permanent component as important. Moreover, the cycle and thereby thewhole model is identied by a Phillips curve model without allowing for speedlimits,i.e. howfasttheunemploymentgapcanbeclosedwithoutcausingachangeinination. Thus, notmodelingaspeedlimitcangiveappearancethat the NAIRU is closer to the current unemployment level than is actuallythecase. Giventhedramaticdevelopmentoftheunemploymentrateintheearly 1990s, the random walk assumption and the spill over channel from thecycletothepermanentcomponent,andthespecicationofthecyclemakesitdiculttocapturethefastriseinunemploymentintheearly1990sasacyclicalcomponent.Beside these important dierences, there are some results that are similar.Thevarianceofthecyclicalshockissmallerthanforthepermanentshock,76 THEEQUILIBRIUMRATEOFUNEMPLOYMENTandsomeof thepoint estimates of thesupplysideshocks inthePhillipscurveareruyofthesamemagnitudeinbothpapers.TheimportanceofselectedvariablesWedidsomefurtheranalysisofwhatisdrivingtheNAIRUinmodel1. Wefound that the tax wedge has contributed to an increase in the NAIRU duringtheperiod1972-1990of almostafull percentagepoint. During1990-1991thetaxwedgeresultedina0.5percentagepointsreductionintheNAIRU,coincidingwiththeencompassingtaxreformsof1990and1991,whichweredesignedtobringdownratesandbroadenthetaxbase. After1991,thetaxwedgehashadanegligibleeect. Theweakeningofthereal exchangeratesince 1972 has resulted in an increase of the NAIRU of around 0.6 percentagepoints. AnincreasingaccommodationratiotendedtoreducetheNAIRUduringthe1970s,followedbyasharpincreasein1980-81. Inthe1990s,theaccommodationratiocontributedtomakingtheNAIRUmorevolatile. ThereplacementratiohascontributedtoadecreaseintheNAIRUsince1994.RobustnessModels3and4suggestthattheestimatesintheunemploymentequationcould be unstable across the subperiods. Model 3 especially does not seem tobe well specied. This evidence of instability is not surprising considering thesmall sample size and the evolution of the open unemployment series depictedinFigure1. However, wenotethatthepositiveeectof adepreciationoftherealexchangerateandthenegativeeectoflabormarketprogramsonopenunemploymentneverthelessarestableacrossthesubperiods.4. EMPIRICALRESULTS 774.3 RealexchangeratesMainresultsIn Table 4 we present the estimated real exchange rate equation (3.17). Therst dierence of all non-stationaryvariables are includedas explanatoryvariablestocapturetheshort-rundynamicsalongwiththedomesticunem-ploymentgapandtheforeignoutputgap. Recallthatanegativeparameterindicatesthatanincreaseof thevariablewill appreciatethereal exchangerate.Theunemploymentgapcoecient10ispositiveasexpectedandhighlysignicant inmodel 1andmarginallysignicant inmodel 2. Our resultsalsoshowthattherelativesizeof middleagedcohortsdepreciatestherealexchange rate. This is consistent with a desire among middle-aged to increasesavingsbyincreasingnetexports,butitcanalsobeinterpretedasevidencethattheinvestmenteectdominates. Increasedtermsof tradeappreciatesthecurrencyas expected. Anincreaseddomesticstructural decit andadecreasedforeigndecitdepreciatesthereal exchangerate, indicatingthatthewealtheectdominatesthedemandeect. Adecreasedforeigndecitbyonepercentagepointdepreciatestheexchangerateby5-8percent. Theforeignoutputgapparameterisalwayssignicant, butthesignisnottheexpected one. The error correction component implicit in Table 4 for Models1(TCW)and2(SEK/EUR)are:0.186_et1 _0.16(d d)t1 0.86qt1 + 0.009gt1 0.075gt1_,and0.198_et1 _0.11(d d)t1 0.56qt1 + 0.006gt1 0.045gt1_.The speed of the error correction is about the same for the two exchangerates,withahalf-lifeof3-4quarters. Thelong-runelasticitiesaregenerally78 THEEQUILIBRIUMRATEOFUNEMPLOYMENThigherfortheTCWcomparedtotheSEK/EUR.10The long-run relationship or the equilibrium real exchange rate accordingtomodel 1isshowntogetherwiththeactual rateinFigure2. Bothfollowadepreciatingtrendoverthewholesampleperiod1972-2001. Thenominalexchangeratewasxed(butdevaluedseveraltimes)upuntilNovember19,1992, whentheprevious pegwas givenupandthekronawas allowedtooat. IncontrasttotheNAIRUandcyclicalunemployment,thetransitorycomponent of the exchange rate in the graphs includes the foreign output gapandtherstdierencesof thelong-runexogenousvariablesinadditiontothe domestic unemployment gap. According to the model, the real exchangerate has been undervalued during most of the oating rate period, since 1993.However,duetotheidenticationofthemodel,thisinterpretationrestsontheassumptionthat themeanof theequilibriumexchangeratecoincideswiththesamplemeanoftheactualrate.InFigure3, weshowresults fromthecompletemodel that allows forshort-run dynamics as well as both domestic and foreign cyclical movements.The model does a good job of tracking the movements in the actual exchangerate, except for the mid-1980s. The results are promising, especially keepinginmindthatthesampleperiodspansnolessthanvedevaluationsunderthexedexchangerateregimeandaregimeshifttoafreeoat.Figure4decomposestheestimatedmodel intoacyclical component, acomponentdependingontherst-dierencedvariables,andanerrorcorrec-tion component in each period. We note that the cyclical component exerteda depreciating eect on the exchange rate during most of the 1990s and thatsomeoftheovershootingaftertheoatingofthekronainlate1992canac-tuallybeexplainedbythecyclicalpositionoftheSwedisheconomyrelativetoitsmaintradingpartners.10The long-run elasticities are obtained by dividing the parameter of interest with theparameter in front of the lagged exchange rate.4. EMPIRICALRESULTS 79TheimportanceofselectedvariablesWhen analyzing how dierent factors in model 1 have inuenced the REER,wendthatthedemographicvariablecontributedtoappreciatingtheex-change rate in the 1970s, which was a period when the group of middle-agedwas decliningrelativetotheTCWarea. However, duringthe1980s thisgroupincreasedandcontributedtoaweakeningof theexchangeratebutthistrendwasreversedagaininthe1990s.Duringtheearly1970sthetermsoftradedeterioratedandhadadepre-ciatingeectfollowedbyasharpappreciatingeectin1986, asoil pricesplummeted in the aftermath of the OPEC debacle. In recent years the termsof trade have again deteriorated and contributed to the weakening of the realexchangerate.TheSwedishstructuralbudgetdecithadobviouslyitsmostnoticeableeectonthereal exchangerateduringthecrisisyearsof theearly1990s.Duringthedeeprecessionof1991-93thebudgetstancewentfromasurplusof 2 percent to a decit of 7 percent. This increase in the decit contributedto the depreciation of the exchange rate, which is in line with the wealth eectdiscussedabove. Thereductionoftheforeignbudgetdecithadastronglydepreciating eect on the real exchange rate during the oating exchange rateperiod since 1992,although the trend was broken in 2000. Even though it isnatural, given our point estimates, to expect an appreciation of the TCW orEurowhenthebudgetstanceinthesecurrencyareasimproves,wehadnotexpectedtondsuchastrongeectof theforeigndecitontheREER, areduceddecitby1percentagepointdepreciatesthereal exchangerateby5-8percent. Acloserlookatthetimeseriesoftheforeignstructuraldecitledustosuspectaspuriousregressionresult. However, lookingatModel3fortheshorterperiod1972-1992,excludingtheforeignbudgetconsolidationduring the 1990s, still results in a highly signicant (but smaller) estimatedcoecient. In Model 4 (1982-2001) the coecient is twice as high in absolutetermsandevenmoresignicant. Thus, theresultssuggestthatthestrong80 THEEQUILIBRIUMRATEOFUNEMPLOYMENTappreciationof theTCWandtheEuroagainsttheSwedishKronaduringthe 1990s was primarily driven by the improved foreign scal position. Eventhough the improvement in the Swedish scal balance during the second partofthe1990swasgreater,theeectwassmaller.DierentregimesAkeyresultinLarsson(2002)isthatthereal exchangeratedynamicsareregimedependent. Shendsthatdeviationsfromlongrunequilibriumarecorrected more quickly when the nominal exchange rate has been allowed tooatfreely. Inordertotesttherobustnessof thisresult, weestimatedaversion of Model 1 with an exchange rate regime-shift variable.11We did thisbyspecifyingthecoecientonthelaggedexchangerateas11= b0 +b1Dt,whereDt=1duringtheoatingrateperiodandDt=0duringthexedrateperiod. Inthismodel wecantestwhetherornottheerrorcorrectionisfasterduringtheoatingrateperiod(|b0 +b1|> |b0|). Thisprovestobethecase, withahalf-lifeof lessthanayearcomparedtoahalf-lifeof fouryears during the xed-rate period, when corrections through devaluations areexcludedbyusingdummyvariablestoaccountforthespikesintheseriescaused by the devaluations. Since Larsson (2002) uses devaluation dummies,ourresultindicatethatherndingsarerobust.However, if weinsteadincludethedevaluations, i.e. donotusedeval-uationdummies, thereisnosignicantdierencebetweentheerrorcorrec-tioncoecientintheoatingversusxed-rateperiods. Hence, duringthexed-rate periodadjustments of the real exchange rate alsomainlytookplacethroughchangesinthenominalexchangerate(devaluations)andnotthroughchanges intherelativepricelevel. Our results indicatethat thequestion whether or not the exchange rate dynamics are regime dependent isstillopen,andpartlyamatterofhowthelargedevaluationsofthenominal11Other variables can also behave dierently under dierent regimes. This is, however,not examined in this paper.4. EMPIRICALRESULTS 81exchangeratearehandled.4.4 ThePhillipsCurveMainresultsThePhillipscurveisnaturallyofspecialinterestforstabilizationpolicy. Asseenabove,aunitrootintherateofinationisnotrejected. Aunitrootishowever rejected for the change in the rate of ination. Even though Swedenhas had a credible ination target regime since mid 1990s, the rate of inationshouldstatisticallybetreatedasanonstationaryvariable.All parameters inthePhillips curveequationhavetheexpectedsignsandaresignicant andsurprisinglystable(seeTable4). Thesizeof thedierent parameters seem reasonable. An increase in the unemployment gapby1percentagepointinModel 1will reducetherateofinationbyabout0.6percentagepointsfourquarterslater.12TheOkuncoecientisoftenestimatedtobeintherange1.5-2. Giventhis, ourestimatessuggestthatanincreaseintheoutputgapby1percentagepointwill increaseinationayear later by0.3-0.4percentage points. This is muchinline withtheresultsinApel andJansson(1999)aswell aswithexistingrulesofthumb.Thesumof theimport priceshocks arereasonable, andnot far fromtheimportpenetrationinprivateconsumptionaccordingtoStatisticsSweden.Regardingproductivity,ourpointestimatesindicatealimitedpass-throughintheshortrunasexpected.SpeedlimitDierent kinds of rigidities can give rise to persistence and hysteresis (Layard,NickellandJackman,1991;Lindblad,1997). Persistencecanexertastrong12See equation(3.20) where the point estimates are included. Ashift incyclicalunemployment lasting a year gives an eect on ination that can be calculated as0.51 + 4 0.02 0.6.82 THEEQUILIBRIUMRATEOFUNEMPLOYMENTinuenceonshort-runinationarypressureanditcangivetheappearancethatthe(long-run)NAIRUisclosertothecurrentlevel of unemploymentthanisactuallythecase. Thisisimportantforpolicymakerssinceithasimplicationsforboththesizeandthetimingoftheproperpolicy.The ndingof asignicant eect fromthe laggedunemployment gapsuggests that there maybealimit tothespeedwithwhichthe gapcanbe closedwithout causingachange inination. This canbe illustratedbyrewritingthePhillipscurve(droppingtheimportpriceandproductivityshocks)as:4pt= (20 +21) uct 21uctorusingthepointestimatesfromModel1,4pt= 0.021uct 0.51uct. (3.20)Thus, itisclearthatnotonlythelevel butalsothechange intheunem-ployment gap may be of notable importance for the ination dynamics. Thespeed-limit, i.e. howfastcyclical unemploymentcanfall withoutcausingachangeinination,canbecalculatedbysetting4pt=0andsolvingforuct,whichgivesuct= 0.04uct.Thegapshouldbeclosedbymaximum4percentperquarter. Thusevenif theunemploymentgapispositiveandexertingadownwardpressureonination,inationwillincreaseifcyclicalemploymentisfallingtofast.135 ConclusionsandsummaryThe Swedish unemployment rate rose rapidly during the early 1990s, but hasfallenfrom1997to2008. Thecentral questionintheintroductionof this13The presence of a speed limit, showing that the gap only can be closed slowly, supportsthe idea of a smooth monetary policy.5. CONCLUSIONSANDSUMMARY 83paper is to what degree the dramatic changes in the Swedish unemploymentrate during the 1990s reect changes in the natural rate/NAIRU and changesinthebusinesscycle.To answer this questionwe estimate anopen-economyversionof theLayard,NickellandJackman(1991)modelforpriceandwagesettinginanunobservedcomponentsframework.Instead of modelling the NAIRU as a stochastic trend in a closed-economyframework, we have modeled it in an open economy framework as a functionoftheoreticallymotivatedvariables: taxes,activelabormarketpolicies,thereplacementratio,demographicfactorsand,sincewearemodellinganopeneconomy,therealexchangerate. Allvariablesexcepttherealexchangeratearetreatedasexogenous. Therealexchangerateismodeledandestimatedsimultaneouslywiththerestof themodel. Hence, wehavesimultaneouslyestimatedequilibriumandcyclicalmovementsintherateofunemploymentand in the real exchange rate and to identify the model an expectations-augmentedPhillipscurve.Wereject theclosed-economyframework. Our results showthat it isimportant to model the open-economy aspects. The closed-economy version,nestedbytheopen-economymodel,issafelyrejected. Moreover,theclosed-economyversionmisbehavesinastatisticalmanner,anditisnotidentiedproperly. Ignoringtheexistenceof theforeignsectorwill biastheresults,and according to the point estimates, the bias is of considerable importance.Thedevelopmentof theSwedishunemploymentrateissuccessfullyex-plained. A depreciation of the real exchange rate, a higher replacement ratioandhigher taxes will, as expected, raise the NAIRU. Anincrease inthesize of labor market programs tends to reduce open unemployment, which isalsoexpected. Accordingtoourpointestimates, theNAIRUincreasedbyapproximately1-1.5percentagepointsduring1980-1985. Thereafterithasremained quite stable. This nding implies that the dramatic unemploymentchangesinthe1990sweremainlyacyclicalphenomenon.84 THEEQUILIBRIUMRATEOFUNEMPLOYMENTMosteconometricstudiesof real exchangeratesusethetermsof tradeandsomemeasureof netforeignassetsasexplanatoryvariables. However,netforeignassetscannotbetreatedasexogenous. Intheempiricalanalysiswe therefore assumed that net foreign assets are determined by demographicvariables and structural budget decits. The real exchange rate was success-fullymodeledusingthetermsoftrade, theevolutionofdemographicsrela-tive to the rest of the world, and the domestic and foreign structural budgetdecits. Our focus on the demographics relative to the corresponding foreigngroupandtheuseof structural decitsintheexchangerateequationisanovelty to the literature. A larger fraction of the population aged 45-59 yearsandalargerdomesticstructural budgetdecitcauseadepreciationof therealexchangerate. Improvedtermsoftradeandalargerforeignstructuralbudget decit cause an appreciation of the real exchange rate. Regarding thestrongappreciatingeectfromahigherforeignstructuraldecit,webelievethatitshouldbethoughtofasadepreciationoftheforeigncurrencyratherthanas astrengtheningof theSwedishKrona. Our exchangeratemodelexplainsasurprisinglylargefraction, upto50percent, of thevariationinthe real exchange rate. The results are promising, especially keeping in mindthatthesampleperiodspansnolessthanvedevaluationsandashiftfromapegtoaoatofthenominal exchangerate, whichareeventsthatwedonotcontrolfor.We also nd that the real exchange rate does not revert any faster to theREERduringtheoating-rateperiod, thanduringthexedexchangerateperiod. It appears that adjustments tothereal exchangerateduringthexed-rateperiodmainlytookplacethroughdevaluationsandnotthroughchangesintherelativepricelevel.ThePhillipscurveseemstobealiveandwell. Changesintherateofinationarewellexplainedbytheunemploymentgapandshockstoimportprices andproductivity. Our results alsoindicatethat thechangeintherateofinationisaectedbythechangeintheunemploymentgap, dueto5. CONCLUSIONSANDSUMMARY 85persistence or other rigidities. This suggests that there is a limit to the speedwithwhichthegapcanbeclosedwithoutcausingrisingination,andthusgivesanargumentforasmoothmonetarypolicy.Moreover, the obvious policy conclussion from the results in this paper isthehighunemploymentrateinthe1990swasduetoalackofdemandandthatanexpansionarypolicyhadbeenaproperresponse. Thisisthesamerecommendationas inLindblad(1997), despitethedierent results. Thereasonforthisisthatif theNAIRUisaectedbypersistenceandcyclicalunemployment, thenademandshockwill actuallynot onlydecrease thecyclicalpartofunemploymentbutalsothepermanentcomponent.86 THEEQUILIBRIUMRATEOFUNEMPLOYMENTReferences[1] Apel,M.andP.Jansson(1999).SystemestimatesofpotentialoutputandtheNAIRU,Empirical EconomicsVol.24,No.3:373-388[2] Assarsson, B. and P. Jansson (1998). Some further evidence on hystere-sisinunemployment: Thecaseof Sweden,AppliedEconomicsLettersVol5,Issue1:25-29[3] Ball, L. and N.G. Mankiw (1995). What do budget decits do?in Budgetdecitsanddebt: issuesandoptions, asymposiumsponsoredbyTheFederalReserveBankofKansasCity,JacksonHole,Wyoming.[4] Bean,C.(1994).Europeanunemployment: asurvey,JournalofEco-nomicLiterature32:573-619[5] Blanchower, D.G. and A.J. Oswald (1994). The wage curve. MIT Press,Cambridge,Massachusetts.[6] Calmfors,L.andA.Forslund(1990).WageformationinSweden.InL.Calmfors (ed.). Wage formation and macroeconomic policy in the Nordiccountries.SNSandOxfordUniversityPress,Oxford.[7] Calmfors, L. andH. Lang (1995). Macroeconomic eects of activelabourmarketprogrammesinaunionwage-settingmodel, TheEco-nomicJournal 105:601-619[8] Campos, J., N. Ericsson and D. Hendry (2005). General-to-specic mod-eling: an overview and selected bibliography, International Finance Dis-cussionPaperNo.838,BoardofGovernorsoftheFederalReserveSys-tem[9] Elmeskov, J. (1994). Nordic unemployment in a Nordic perspec-tive,SwedishEconomicPolicyReview1:1-2REFERENCES 87[10] Forslund, A. (1995). Unemployment - is Sweden still dierent? SwedishEconomicPolicyReview2:15-58[11] Forslund, A. andA.-S. Kolm(2004). Activelabormarketpoliciesandreal-wagedetermination-Swedishevidence, inS. Polachek(ed.): Ac-counting for Worker Well-Being, ResearchinLabor Economics, Vol.23.[12] Hakkio, C.S. (1996). Theeect of budget decit reductionontheex-changerate.EconomicReview,FederalReserveBankofKansasCity.[13] Harvey,A.C.(1989).Forecasting,structural timeseriesmodelsandtheKalmanlter,CambridgeUniversityPress.[14] Hendry, D. andK. Juselius(1999). ExplainingCointegrationAnalysis:PartI.Mimeo.[15] Holmlund, B. (1993). Arbetsloshetskrisen - konjunkturfenomen eller sys-temfel? Appendix12toSOU1993:16, Lindbeck, A. etel Nyavillkorf orekonomiochpolitik.Allmannaforlaget,Stockholm.[16] Joyce, M. andS. Wren-Lewis (1991). Theroleof thereal exchangerateandcapacityutilizationinconvergencetotheNAIRU,EconomicJournal Vol.101:497-507[17] King, T.B. andJ. Morley(2007).Insearchofthenatural rateofun-employment,Journal ofMonetaryEconomics54:550-564[18] Lane, P.R. andG.M. Milesi-Ferretti (2004). Thetransferproblemre-visited: net foreign assets and real exchange rates, TheReviewofEco-nomicsandStatistics84:4[19] Larsson, A. (2003). TheSwedishReal ExchangeRateunderDierentCurrencyRegimes.WorkingpaperNo.180,FIEF.88 THEEQUILIBRIUMRATEOFUNEMPLOYMENT[20] Layard,R.,S.NickelandR.Jackman(1991).Unemployment: macroe-conomicperformanceandthelabourmarket. OxfordUniversityPress,Oxford[21] Lindblad, H. (1997). Persistence in Swedish unemployment rates. Work-ingPaper 1997:3, Dept. of Economics, Universityof Stockholm, andChapter2inthisthesis.[22] Lindblad, H. and P. Sellin (2010). ASimultaneousModeloftheSwedishKrona,theUSDollarandtheEuro.Chapter4inthisthesis.[23] Lindh, T. andB. Malmberg(1999). Agestructureandthecurrentac-count - achangingrelation? WorkingPaper 1999:21, Dept. of Eco-nomics,UppsalaUniversity.[24] Mankiw,N.G.(2006).Macroeconomics,WHFreeman&Co.[25] Nickell, SandR. Layard(1999). Labour market institutions andeco-nomicperformance.HandbookofLaborEconomics,3:3029-3084.[26] Phillips, P.C.B. and B.E. Hansen (1990). Statistical inference in instru-mental variables regressionwithI(1) processes,Reviewof EconomicStudies57:99-125[27] Salemi, M.K. (1999). Estimating the natural rate of unemployment andtestingthenatural ratehypothesis,Journal of AppliedEconometrics14:1-25[28] Turner,D.,L.Boone,C.Giorno,M.Meacci,D.RaeandP.Richardson(2001). Estimating the structural rate of unemployment for then OECDcountries.OECDEconomicStudiesNo.33,2001/2.A. APPENDIX 89A AppendixA.1 DataappendixForeign variables have been computed from the weights used in constructingtheTCWindexfor theSwedishkrona. Whencomputingtheweights fortheEMU,theTCWweightshavebeenrescaledtosumtoone. PopulationstatisticswereobtainedfromStatisticsSwedenandtherespectivestatisticsbureausoftheothercountriesinvolved.a = ln(ALMP/(ALMP+OU)istheaccommodationrate.ALMP=Thesumofalldierentactivelabormarketprogramsthathave existed during the period 1972-2001. Source: Swedish Public Em-ploymentService(formerlyAMS).Monthlydatahavebeenconvertedtoquarterlybytakingthemean.d=Netsaversratio,computedasthesumof45-59yearsoldrelativetothetotalpopulation.d=TCW(orEMU)weightednetsaversratio.e = ln(TCWP/Pc) or e = ln(SEK/EURP/Pc) is the real exchangerateforModel 1and2respectively(thereal TCWindexissetequalto100at18November1992, whiletherealSEK/EURissetequaltoitsnominalequivalentat2001:1). Source: SverigesRiksbank.g= structural government decit (as percent of potential GDP). Source:OECD,MainEconomicIndicators.g=TCW(orEMU)weightedstructuralgovernmentdecit(asper-centofpotentialGDP).Source: OECD,MainEconomicIndicators.Y =GDP=Real grossdomesticproduct(s.a.). Source: StatisticsSweden,NationalAccounts.90 THEEQUILIBRIUMRATEOFUNEMPLOYMENTY= GDP= TCW (or EMU) weighted gross domestic product (s.a.).Source: OECDMainEconomicIndicatorsandowncomputations.P=Swedishconsumerpriceindex. Source: StatisticsSweden.P=ATCW(or EMU) weightedconsumer price index. Source:OECDMainEconomicIndicatorsandowncomputations.pr = log of labor productivity (GDP per hours worked). Source: Statis-ticsSweden.Pc= PUND= TheUND1Xcoreinationpriceindex isour measureofinationinthePhillipscurve. Source: SwedishRiksbank.q= PX/PM, the terms of trade computed as export deator over importdeator. Source: StatisticsSweden.r =replacementratio: thelogofmaximumdailyunemploymentcom-pensationdividedbyeighttimestheaveragehourlywage. Thelatteris computed as the total wage bill divided by the total number of hoursworked. Source: SwedishPublicEmploymentService(unemploymentcompensation)andStatisticsSweden.TCW=Acompetitivenessweightednominal eectiveexchangerateindex(18November1992=100). Source: SverigesRiksbank. =ln_(1 +V AT)_1+S1T_, whereV AT is direct eects ontheCPIfromindirecttaxesandsubsidies, Sispay-roll taxandT isincometax.u=OU/TLFis therateof unemployment, whereOU=openun-employmentandTLF=totallaborforce. Source: StatisticsSweden,LabourForceSurvey(AKU).(y yn)=Foreigncyclicaldemand, wherepotentialoutputiscom-putedastheHP-ltered(1600)seriesofGDP.A. APPENDIX 91Figure 1. The actual, natural and cyclical rates of unemployment NAIRU _ _ _and Unemployment rate _____1972 1975 1978 1981 1984 1987 1990 1993 1996 1999123456789Cyclical Component1972 1975 1978 1981 1984 1987 1990 1993 1996 1999-3-2-1012345 92 THEEQUILIBRIUMRATEOFUNEMPLOYMENTFigure 2. The actual and equilibrium real exchange rates REER _ _ _and Real TCW exchange rate _____1972 1975 1978 1981 1984 1987 1990 1993 1996 19998896104112120128136144Transitory Component1972 1975 1978 1981 1984 1987 1990 1993 1996 1999-15-10-505101520 A. APPENDIX 93Figure 3. The actual and predicted values for the real exchange rates Model _ _ _and actual real TCW exchange rate _____1972 1975 1978 1981 1984 1987 1990 1993 1996 19998896104112120128136144Difference between actual and model1972 1975 1978 1981 1984 1987 1990 1993 1996 1999-15-10-5051015 94 THEEQUILIBRIUMRATEOFUNEMPLOYMENTFigure 4. A decomposition of the real exchange rate Cyclical component1972 1975 1978 1981 1984 1987 1990 1993 1996 1999-3-2-1012345Dynamics component1972 1975 1978 1981 1984 1987 1990 1993 1996 1999-4-3-2-1012345Error correction component1972 1975 1978 1981 1984 1987 1990 1993 1996 1999-4-3-2-10123 A. APPENDIX 95Table1: Unitroottests***, **, *denotesignicanceatthe1, 5, and10percentlevel respectively, usingMacKinnon critical values for rejection of the hypothesis of a unit root. Two sets of statisticsare reported for foreign variables: Model 1 (TCW)/Model 2 (SEK/EUR).Variable Teststatistic Lagsu -1.36 1e -1.68/-1.97 1/1e -7.28/-7.181/14p -0.24 44p -5.742a -2.831r -2.881 -2.15 4d d-1.50/-1.63 6/5(d d) -3.12/-2.55 5/5q -2.04 1q -7.501g -2.08 3g -3.423g-1.58/-1.38 2/4g-5.11/-4.731/4y yn-5.24/-5.324/44pM-8.9834pr -8.46496 THEEQUILIBRIUMRATEOFUNEMPLOYMENTTable2: UnemploymentequationDependentvariable: openunemployment, percentoftotal laborforce. Whitestandarderrorsarereportedwithinparenthesesbelowtheestimatedcoecients. Serial corr. isaBox-LjungQ(k)testagainstserialcorrelationbasedonk=11autocorrelations. Crosscorr. isaQ(k)testof thecorrelationof residualsacrossequationswithk=5lags/5leads. Heterosk. isEnglesLMtestagainstrstorderautoregressiveconditional het-eroskedasticity (Chi-squared with 1 d.f.). R2d is the coecient of determination suggestedfor non-stationary time series by Harvey (1989). A positiveR2d implies a better t than arandom walk.Variable Model1 Model2 Model3 Model4uct11.83(0.04)1.83(0.02)1.94(0.10)1.83(0.10)uct20.85(0.04)0.84(0.02)0.99(0.10)0.85(0.10)at1.82(0.27)1.81(0.25)1.56(0.19)1.71(0.30)at10.50(0.33)0.48(0.37)0.05(0.19)0.60(0.42)rt0.04(0.14)0.09(0.13)0.16(0.15)0.45(0.42)rt10.23(0.25)0.25(0.26)0.12(0.15)1.28(0.49)t0.68(0.30)0.64(0.38)0.43(0.39)0.96(0.91)t10.51(0.41)0.41(0.45)0.37(0.43)0.14(1.02)t30.84(0.33)0.77(0.44)0.13(0.34)0.18(1.04)et11.81(0.70)1.55(0.71)0.18(0.48)2.31(1.17)cu0.13(0.02)0.13(0.02)0.08(0.02)0.16(0.03)nu0.08(0.02)0.08(0.02)0.04(0.01)0.09(0.02)R20.99 0.99 0.98 0.99R2d0.57 0.57 0.72 0.58Serial corr. (p-value) 0.630 0.576 0.501 0.466Cross corr. ne(p-value) 0.538/0.739 0.751/0.654 0.106/0.014 0.325/0.361Cross corr. p (p-value) 0.515/0.488 0.566/0.536 0.051/0.053 0.946/0.689Heterosk. (p-value) 0.005 0.001 0.597 0.326Exchange rate TCW SEK/EUR TCW TCWSample 1972-2001 1972-2001 1972-1992 1982-2001A. APPENDIX 97Table3: ExchangerateequationDependentvariable: thepercentagechangeinthereal exchangerate. Whitestandarderrors are reported within parentheses belowthe estimated coecients. Serial cor.is a Box-Ljung Q(k) test against serial correlation based on k=11 autocorrelations.Crosscorr. isaQ(k)testof thecorrelationof residualsacrossequationswithk=5lags/5leads. Heterosk. isEnglesLMtestagainstrstorderautoregressiveconditionalheteroskedasticity (Chi-squared with 1 d.f.).Variable Model1 Model2 Model3 Model4uct4.7 103(1.6103)3.6 103(2.2103)3.7 103(4.0103)6.1 103(2.8103)et10.186(0.048)0.198(0.055)0.238(0.065)0.217(0.057)(d d)t10.030(0.010)0.021(0.006)0.027(0.008)0.082(0.029)qt10.160(0.058)0.111(0.067)0.203(0.064)0.152(0.074)gt11.6 103(1.5103)1.2 103(1.7103)1.2 104(1.6103)2.4 103(1.6103)gt10.014(0.004)0.009(0.003)0.012(0.004)0.023(0.005)(d d)t0.014(0.040)0.033(0.048)0.004(0.047)0.100(0.071)qt0.601(0.204)0.437(0.141)0.577(0.146)0.608(0.208)gt2.9 104(3.7103)1.0 103(4.2103)9.5 103(5.0103)2.1 103(3.8103)gt3.2 103(6.4103)2.4 103(6.9103)5.9 104(7.9103)1.6 103(11.3103)(y yn)t5.3 103(1.6103)6.3 103(2.2103)2.1 103(2.3103)15.3 103(4.6103)e0.022(0.002)0.024(0.001)0.019(0.002)0.022(0.002)R20.34 0.27 0.39 0.53Serial corr. (p-value) 0.439 0.091 0.005 0.169Cross corr. nu(p-value) 0.739/0.539 0.654/0.751 0.014/0.106 0.361/0.325Cross corr.p (p-value) 0.177/0.486 0.562/0.779 0.105/0.160 0.056/0.140Heterosk. (p-value) 0.951 0.775 0.876 0.833Exchange rate SEK/TCW SEK/EUR SEK/TCW SEK/TCWSample 1972-2001 1972-2001 1972-1992 1982-200198 THEEQUILIBRIUMRATEOFUNEMPLOYMENTTable4: Theexpectations-augmentedPhillipscurveDependent variable: the quarterly change in the annual rate of ination. White standarderrors are reported within parentheses below the estimated coecients. Serial corr. is aBox-Ljung Q(k) test against serial correlation based on k=11 autocorrelations. Cross corr.is a Q(k) test of the correlation of residuals across equations with K = 5 lags/5 leads. Het-erosk. is Engles LM test against rst order autoregressive conditional heteroskedasticity(Chi-squared with 1 d.f.).Variable Model1 Model2 Model3 Model4uct0.531(0.118)0.609(0.124)0.610(0.067)0.455(0.057)uct10.510(0.112)0.586(0.122)0.364(0.060)0.462(0.053)yct0.060(0.036)0.045(0.037)0.058(0.041)0.151(0.095)2pm,t0.062(0.018)0.065(0.017)0.060(0.020)0.064(0.017)2pm,t30.056(0.016)0.056(0.017)0.057(0.017)0.066(0.015)2prt20.100(0.030)0.100(0.026)0.106(0.031)0.107(0.056)2prt30.086(0.034)0.086(0.025)0.098(0.032)0.062(0.045)p0.527(0.044)0.527(0.045)0.546(0.046)0.460(0.057) 0.554(0.081)0.555(0.077)0.616(0.095)0.301(0.114)R20.56 0.56 0.64 0.57Serial corr. (p-value) 0.612 0.614 0.214 0.145Cross corr. nu(p-value) 0.488/0.515 0.536/0.566 0.053/0.051 0.689/0.946Cross corr.ne(p-value) 0.486/0.177 0.779/0.566 0.160/0.105 0.140/0.056Heterosk. (p-value) 0.998 0.812 0.765 0.812Exchange rate TCW SEK/EUR TCW TCWSample 1972-2001 1972-2001 1972-1992 1982-2001Chapter4ASimultaneousModeloftheSwedishKrona,theUSDollarandtheEuro1 IntroductionSwedenis asmall openeconomywithexports andimports amountingtoabout 54and47per cent of GDPrespectivelyin2008. Thevalueof theKrona (SEK) against other currencies thus becomes an essential determinantof thestateof theSwedisheconomy. Whenanalyzingexchangerates, themostcommonapproachintheempiricalinternationalmacroeconomicsistolookat either abilateral exchangerateor at aweightedindexof foreigncurrencies. Swedens largest trading partner is the Euro area, which absorbsabout40percentofSwedensexports. Thustheexchangeratevis-a-vistheThis paper is written together with Peter Sellin. We would like to thank Malin Adolf-son,Lars Calmfors,Torsten Persson and seminar participants at Sveriges Riksbank,theNational Instituteof EconomicResearchandatAEAInternational ConferenceonEx-change rate econometrics spring 2005 for helpful comments. The views expressed in thispaper are solely the responsibility of the authors and should not be interpreted as reectingthe views of the Executive Board of Sveriges Riksbank.99100 ASIMULTANEOUSCURRENCYMODELEuro(EUR)isofnatural interest. Butsinceabout60percentofSwedishexports are destined elsewhere, the SEK and the EUR exchange rate againstothercurrenciessuchastheUSDollar(USD)arealsoofgreatinterest.Themaincontributionof this paper is tosimultaneouslyestimatetheshort-andlong-rundynamicsofthereal exchangeratesbetweentheSEK,theEURandtheUSD.Usually, and this paper is no exception, the real exchange rate is expressedinterms of consumer goods. Then, the real exchange rate betweentwocurrencies shows howmuchconsumptionthe home countryhas tospendinordertoobtainoneunitof foreignconsumption. Forinstance, therealexchange rate between the SEK and the USD measures how many units of theSwedish consumption basket it takes to buy one unit of the US consumptionbasket. Fromthisdenition, thereal exchangeratecanbewrittenastheproductofthenominalexchangerateandtheratiobetweentheforeignandthedomesticconsumerpricelevels.According to international macroeconomic theory, the real exchange rateis determinedbyaset of institutional andeconomicfactors suchas rela-tiveproductivity, termsof trade(therelativepricebetweenexportedandimported goods and services), net foreign assets and openness (Masson, Kre-mers, andHorne, 1994; Bergvall, 2002andLaneandMilesi-Ferretti, 2004).AvoluminousliteraturetakesitsstartingpointinsomesimpleaccountingrelationshipsandisreviewedinMacDonald(2000). Empiricaltestingofthemore rigorous theoretical models in the new open macroeconomics literatureissurveyedinGhironi(2000)andLane(2001).It is usually postulated that there exists a long-run empirical relationshipbetween the real exchange rate and a set of theoretically motivated explana-tory variables. Bergvall (2002), Nilsson (2002), and Lane and Milesi-Ferretti(2004)estimatereal exchangerateequationsderivedfromnewopenecon-omy models and nd that the real exchange rate is cointegrated with relativeGDP,thetermsoftrade,andnetforeignassets. Alexius(2001)estimatesa1. INTRODUCTION 101commontrendsmodel fortheNordiccountriesandndsthatrelativepro-ductivityisthemostimportantdeterminantof long-runmovementsintherealexchangerateinallthecountriesinvestigated. Hjelm(2001)alsondsthat productivity shocks are the main force driving the US real exchange raterelative to the UK, Germany and Japan. MacDonald and Marsh (1999) use apurchasing-power-paritymodel augmentedwithaninterest-ratedierentialtosimultaneouslyestimatetheexchangeratesbetweentheUSDollar, theGermanD-MarkandtheJapaneseYen. Theirstudyhighlightsthecomplexinteractionsamongthevariablesinatri-polarexchangeratesystem.Aprime determinant of real exchange rate dynamics is relative trendproductivity or relative trend output. If there exists a stable reaction functionbetweenthe interest rate andthe output gap, the dierence incapacityutilization should also be of importance when explaining the dynamics of theexchange rate. Since neither potential output nor output gaps are observable,actual output is often used as a proxy for the former and output minus an adhoc trend as a proxy for the latter. An alternative is to use economic theorytodeneandestimatetheoutputgapandpotentialoutput.Inthispaper,wewilldenecyclicaloutputinastandardPhillips-curveframework and estimate cyclical and potential output for each country in anunobservedcomponents(UC)model. Wesubsequentlyusetheseestimatedvariablesinourexchangeratemodel. Otherexamplesof UC-model appli-cations arefoundinLindblad(1997), Apel andJansson(1999), Laubach(2001),Turneretal. (2001)andHornetal. (2007).Our results indicate that a surprisingly large part, up to two thirds, of thevariation in the percentage change in the real exchange rates is explained byvariationsinpotentialoutput,termsoftradeandstructuralbudgetdecits.Moreover, intermsof explanatorypower, itappearspreferabletouseesti-matedpotentialoutputinsteadofactualoutput.This paper is organized as follows. First, we discuss the theoretical back-groundforthedeterminationofrealexchangerates. Second,wespecifythe102 ASIMULTANEOUSCURRENCYMODELempirical model. Third, wediscusstheoutputfromestimatingthemodelsandbrieycommentonforecastingresults. Finally,wesumup.2 TheoreticaldiscussionA theoretical background to our empirical exchange rate specication is foundinthenewopenmacroeconomics literature, e.g. LaneandMilesi-Ferretti(2004).1These authors set up a small open-economy model, where consumersmaximizelifetimeutilitysubjecttoanintertemporalbudgetconstraintandrmsmaximizeprots. Instantaneoushouseholdutilityisafunctionofcon-sumption of tradable and non-tradable goods and labor input. For simplicity,rmsareassumedtoonlyproduceanon-tradedgoodandsetpricesinor-dertomaximizeprots. Productionoftradedgoodsisanendowmentthatsells inthe worldmarket at agivenexport price. Solvingthe Lane andMilesi-Ferretti model, givesthefollowinglong-rundeterminantsof therealexchange rate: exogenous terms of trade, net foreign assets, and relative pro-ductioninthetradeablesector. Theintuitionisquitestraightforward: anincreaseinthetermsoftrade,netforeignassetsorproductioninonecoun-tryraisesthatcountrysincomeandstimulatesconsumptionofnon-tradedgoods,whichtendstoappreciateitsrealexchangerate. Higherincomealsoreduces labor supply, which reduces the production of non-traded goods andthusraisestherelativepriceofnon-tradeables, alsocausingareal appreci-ation. Thus, twoeects, higherdemandforgoodsandlowerlaborsupply,pushtherealexchangerateinthesamedirection.Inthemodel describedabove, thetermsof tradeeectisunambiguousthroughitswealtheectontherelativepriceofnon-tradeables: anincrease1The (log) real exchange rate between SEK and USD can be written asesu= csu+ pups,wherecsuis the (log) nominal SEK/USD exchange rate andp is the (log) consumer pricelevel. Superscriptu denotes the US and superscripts denotes Sweden.2. THEORETICALDISCUSSION 103will strengthen the real exchange rate.2In other models, the eect of a termsof trade shock on the real exchange rate can be theoretically ambiguous, de-pending e.g., on whether there is a home bias within the category of tradeablegoodsasinObstfeldandRogo(2004). Ahomebiasimpliesthathome-producedtradeableshaveagreaterweightthanforeign-producedtradeablesin the consumption basket. A relative increase in the price of home-producedtradeables(improvedtermsoftrade)willthenhaveagreaterimpactontheconsumerpriceindexinthehomecountrycomparedtotheforeigncountry,leadingtoanappreciationoftherealexchangerate. Empiricallyapositivetermsof tradeshockhasinmostcasesbeenfoundtoleadtoasignicantappreciationof thereal exchangerate(Dungey, 2002; Nilsson, 2002; andLaneandMilesi-Ferretti, 2004), whichis consistent withboththewealtheectandahomebias, althoughthereareexceptions(seee.g. AlexiusandNilsson,2000).Higher net foreignassets induce households toincrease their demandfor non-tradedgoods. At the same time higher net foreignassets causeareductionof laborsupplies, whichreducestheproductionof non-tradedgoods. Both the goods demand and the labor supply eects raise the relativepriceof non-tradedgoods, implyinganappreciationof thereal exchangerate. Hence, higher(lower)netforeignassetsarerelatedtoanappreciated(depreciated)realexchangerate.Acountrysnetforeignassetpositionisdeterminedbyitsaccumulatedprivate and public net savings.3Since the private savings ratios dier amongcohorts of the population, an important and plausibly exogenous determinant2It is important to understand that there is a fundamental dierence between terms oftrade and the real exchange rate. Terms of trade is the relative price of tradeable goods andservices. An increase in terms of trade increases wealth. Wealth increases the demand fornon-tradeables and lowers labor supply. This increases the relative price of non-tradeables,and thereby the real exchange rate, since the real exchange rate is measured as the relativeprice of consumer goods. Non-tradeables are not a part of terms of trade.3A similar decomposition of net foreign assets into demographic and scal variables isemployed in Masson, Kremers and Horne (1994) and Lane and Milesi-Ferretti (2001).104 ASIMULTANEOUSCURRENCYMODELofaccumulatednetsavingsistheevolutionofdemographyovertime. Sincethe demographic composition is quite persistent, it also is an indicator of thenetforeignassetposition.Usually, a higher proportion of prime- or middle-aged people in the popu-lation is assumed to stimulate savings as these cohorts prepare for retirement.Higgins(1998)ndssubstantialdemographiceectsonthecurrentaccountbalanceforanumberof countries. However, healsoshowsthatthedemo-graphicstructurehaslittleeectonnetsavinginmoreclosedeconomies.Lindh and Malmberg (1999) present evidence that the proportion of middle-agedpeopleinthepopulationhas apositiveeect oninvestment as wellasonsaving, makingtheeectonnetsavingambiguous. Thereasonforapositiveeectoninvestmentisnotobvious. Twopossibleexplanationsaregiven. First, themiddle-agedtendtotransferwealthfromreal tonancialassetswhich,givenahomebiasinnancialinvestment,woulddecreasethelocalcostofcapital. Second,thegroupofmiddle-agedpeopleisarelativelyproductiveagegroup, whichmeans alower eectivecapital intensityandthusahigherreturntocapital.4However even if the net cohort eect is to increase the countrys nancialsaving, the onlywaytodosois viaanimprovedcurrent account. Thetextbookwayof achievinghigher net exports is bydepreciatingthe realexchange rate (Ball and Mankiw, 1995). Thus, net eect of the demographiccompositiononnetforeignassetsandtherealexchangerateisanempiricalquestion.Another important determinant of net savingis the scal positionofthegovernment sector (totheextent that Ricardianequivalencedoes notholdfully). Thus, highlevelsof publicdebtmaybeassociatedwithalowlevelofnetforeignassets. Ahigherpublicdebtthatisnotosetbyhigherprivate assets will reduce the net foreign assets. In this case, a higher public4Lindh and Malmberg (1999) dene middle-aged as the age group 50-64 years old, whilewe use 45-59 years old in the empirical part of the paper. We use this denition becausethe countries in our study have a lower actual retirement age than 65.2. THEORETICALDISCUSSION 105debt woulddepreciatethereal exchangeratethroughthewealthchanneldiscussedaboveawealtheect. However, other eects couldgointhereversedirection. Hakkio(1996)discussesseveralpossiblechannelsthroughwhichthescal positioncouldinuencetheexchangerate. Oneof thesechannelsisthetextbookeectthatanincreaseinthegovernmentbudgetdecit, and thus also an increase in debt, should raise demand and thus pricesof non-tradeable goods, which is synonymous with an appreciation of the realexchangerateademandeect. However, thisrequiresthatscal policyiscredible. Thus,whetheranincreaseddebtwillleadtoanappreciationoradepreciationisanempiricalquestiondependingontherelativesizeofthewealthanddemandeects.Thus, we regard the net foreign asset position as a function of the demo-graphiccompositionandthescalpositionofthegovernmentsector. Thesevariablesaretreatedasexogenousinthereducedformexchangerateequa-tions below. Lindh and Malmberg (1999) and Hakkio (1996) also argue thatthesevariablesaecttherealexchangeratedirectly.Yet, anothermechanismisthathigherrelativeproductivitystrengthensthe real exchange rate through the Balassa-Samuelson eect. If productivitygrowthisfasterinthedomestictradeablesector, comparedtotheforeigntradeable sector, this triggers higher real wage increases whichspill overtorelativelyhigherpriceincreasesfornon-tradeablesrelativetotradeables.Thisimpliesanappreciationoftherealexchangerate.Since data on relative productivity are often unreliable, a standard short-cutistouserelativeGDP, c.f. LaneandMilesi-Ferretti (2004). However,this short-cut isnot without problems. RelativeGDPreects bothtrendandbusinesscycledierences. Whatshouldmatterinthelongrunisreallyrelativetrendoutput,implyingthatamoreappropriatemeasureisrelativepotentialGDP.Theinterestdierentialisoftenarguedtobeanimportantdeterminantoftherealexchangerate(e.g. inMacDonaldandMarsh1999). Intheshort106 ASIMULTANEOUSCURRENCYMODELrun, weknowthatreal interestratesvaryoverthebusinesscycle. If thereexistsastablereactionfunctionbetweentheinterest rateandtheoutputgap,wecanuseouroutputgapvariablestocapturethesedeterminants.3 EmpiricalspecicationBased on the discussion above, we arrive at the following empirical equationsforthe(log)realexchangeratesoftheKrona:esu= 0+1qs+2qu+3ds+4du+5gs+6gu+7yn,s+8yn,u+9yc,s+10yc,u,(4.1)ese= 0+1qs+2qe+3ds+4de+5gs+6ge+7yn,s+8yn,e+9yc,s+10yc,e,(4.2)and,rulingoutarbitragepossibilities,wehaveeue= eseesu(4.3)where q is the terms of trade, d is the share of middle-aged people in the pop-ulation, and gis the structural government budget decit as a share of GDP,and ynand ycare trend output and output gap, respectively. Superscript su,se and ue denote SEK/USD, SEK/EUR and USD/EUR and superscript u, eand s denote the US, the Euro area and Sweden respectively. The discussionaboveindicatedthatboththepublicdebtanddecitcouldbeincludedintheequationsabove. However,aswillbediscussedshortly,wewillonlyusethedecitvariable.SinceSwedenisasmall economy, andtheSEKunlikeUSD, EURandYenis not areservcurrency, we impose aset of restrictions whichwetestedandcouldnot reject ensuringthat Swedishvariables donot af-fecttheUSD/EURexchangerate. Thus, weassumethatSwedenstermsof trade, demographics, structural budgetdecitandoutputinuencethe3. EMPIRICALSPECIFICATION 107SEK/USDandtheSEK/EURexchangeratesbythesameamount, leavingtheUSD/EURunaected. Withtheserestrictions,equation(4.2)becomes:ese= 0+1qs+2qe+3ds+4de+5gs+6ge+7yn,s+8yn,e+9yc,s+10yc,e.(4.4)All movements intheexchangerates at business cyclefrequencies areassumedtobecapturedbytheoutput-gapvariablesinequations(4.1)and(4.4). Thespecicationsaboveareexibleandallowfordistinctcoecientsone.g.,domesticandforeignpotentialoutput. Theresultsfromestimatingamorerestrictedmodelarereportedbelow.Basedonndingsinthepreviousempiricalliterature,whichhavefoundarobustpositiverelationbetweenthetermsoftradeandtherealexchangerate, weexpectthat10. TowhatextenttermsoftradeshocksintheEuroareaandtheUnitedStatescouldtrulybetreatedas exogenous is perhaps debatable, but here we follow the current practice intheempirical literature, togetcomparableresults. Examplesof exogenoustermsoftradeshocksareoilpriceshocksandshockstooth