Macroeconomic implications of oil price fluctuations: a regime-switching framework for the euro area

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Macroeconomic implications of oil price fluctuations: a regime-switching framework for the euro area F´ed´eric Holm-Hadulla and Kirstin Hubrich 2017-063 Please cite this paper as: Holm-Hadulla, F´ed´eric and Kirstin Hubrich (2017). “Macroeconomic implications of oil pricefluctuations: aregime-switchingframeworkfortheeuroarea,”FinanceandEconomics DiscussionSeries2017-063. Washington: BoardofGovernorsoftheFederalReserveSystem, https://doi.org/10.17016/FEDS.2017.063. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Macroeconomic implications of oil price fluctuations: a regime-switching framework for the euro area FédéricHolm-Hadulla KirstinHubrich EuropeanCentralBank FederalReserveBoard May4,2017 Abstract Weinvestigatewhethertheresponseofthemacro-economytooilpriceshocksundergoes episodic changes. Employing a regime-switching vector autoregressive model we identifytworegimesthatarecharacterizedbyqualitativelydifferentpatternsineconomic activity and inflation following oil price shocks in the euro area. In the normal regime, oil price shocks trigger only limited and short-lived adjustments in these variables. In theadverseregime, bycontrast, oilpriceshocksarefollowedbysizeableandsustained macroeconomicfluctuations,withinflationandeconomicactivitymovinginthesamedirection as the oil price. The responses of inflation expectations and wage growth point to second-round effects as a potential driver of the dynamics characterising the adverse regime.Thesystematicresponseofmonetarypolicyworksagainstsuchsecond-roundeffectsintheadverseregimebutisinsufficienttofullyoffsetthem.Themodelalsodelivers (conditional)probabilitiesforbeing(staying)ineitherregime,whichmayhelpinterpret oil price fluctuations – and inform deliberations on the adequate policy response – in real-time. JELClassification: E31,E52,C32 Keywords: Regime Switching models, time-varying transition probabilities, oil prices, inflationexpectations,inflation 1We thank Lutz Kilian, Hans-Joachim Klöckers, Wolfgang Lemke, Simone Manganelli, Roberto Motto,UlrichMüller,MassimoRostagno,ChrisSims,RobertTetlow,RobertVigfusson,DanielWaggoner, Mark Watson, Tao Zha, and conference participants at the International Association for Applied Econometrics Conference 2016, the Oxmetrics Conference for Macroeconomics and Finance 2016, and the 10th International Conference on Computational and Financial Econometrics, as well as seminar participants at Princeton University, the European Central Bank and the Federal Reserve Boardforusefulcomments. Theauthorscanbecontactedatfederic.holm-hadulla@ecb.europa.euand kirstin.hubrich@frb.gov. We also thank Nikolaj Broberg, Mattia Colombo, Nick Ligthart and Kevin Starnesforexcellentresearchassistance. Theviewsexpressedherearethoseoftheauthorsanddonot necessarily reflect those of the European Central Bank or the Federal Reserve System or the staff of eitheroftheseinstitutions.

Cheaperoilisararepieceofgoodnewsfor(...) theeurocurrencyarea,since[it] shouldboostthespendingpowerofEurope’sconsumers(...) amidtheeurozone’s longslump. (TheWallStreetJournal,November14,2014) [A]danger[oftheoil-priceslump]isthatanevendeeperdipininflation(...) may haveanunwelcomesecond-roundeffectbydraggingdowninflationexpectations. (TheEconomist,December4,2014) 1 Introduction Howhasthesharpoilpricedeclinesincemid-2014affectedmacroeconomicprospectsinthe euro area? Like in previous episodes of major oil price fluctuations, this question has generated substantial debate – and strongly divergent assessments – in the economics profession. Adopting a benign view, several observers have argued that the lower oil price will support theeconomicrecoverybyraisingrealdisposableincomesandprofitsofeuroareahouseholds and firms. Others, instead, have cautioned that the oil price slump may become entrenched in inflation expectations, thus leading to second-round effects that reinforce the prevailing disinflationarypressuresandpotentiallydampentherecovery.1 From a monetary policy perspective, it is crucial to establish the relative merit of these differentassessments. Judgingbypublicstatementsbypolicy-makersinvariousjurisdictions, centralbankswouldtypicallyconsiderchangingthemonetarypolicystanceonlyinthelatter scenario, in which oil price fluctuations feed through to inflation expectations, hence risking to exert durable effects on actual inflation dynamics. By contrast, absent such second-round 1See, for instance, Mohaddes and Pesaran (2016) for a benign interpretation and Obstfeld et al. (2016) for an adverse interpretation of this recent collapse in oil prices. An analogous debate took place in the context ofthesteepupwardtrendinoilpricesstartinginend-2003andacceleratingfromearly-2007tomid-2008,when someobserversexpressedconcernsthatsecond-roundeffectsmayleadtosustainedinflationarypressures,whereas otherscontestedthisclaim.Forasummaryofthedebateatthattime,seeHannon(2008). 1

effects,centralbankswouldtendtopreservetheprevailingmonetarypolicystance,thus‘lookingthrough’theoilpricefluctuations.2 Tooperationalisethisnuancedreactionfunction,centralbanksinturnhavetotakeastand on which scenario they consider more likely to prevail; and, since the occurrence of secondround effects may be an episodic phenomenon, they may have to update this assessment on a regular basis. At the same time, a large body of literature has shown that policy mistakes in either direction (i.e. overly activist monetary policy responses to oil price fluctuations that provetransient,aswellasoverlyinertialmonetarypolicythatallowsinflationexpectationsto becomeunanchored)mayseverelyhampermacroeconomicperformance.3 The aim of the current paper is to examine whether the notion of episodic changes in the macroeconomic implications of oil price fluctuations finds empirical support in the euro areacontext. Tothisend,weusearegime-switchingvectorautoregression(RS-VAR)model, estimated with Bayesian methods, that allows for time-variation in model coefficients, thus helpingtodetectregime-dependentdifferencesinhoweconomicactivityandinflationevolve in the aftermath of oil price shocks (henceforth referred to as coefficient switching). The model also allows for time-variation in shock variances (variance switching) since regimedependent differences in macroeconomic dynamics could also result from sequences of unusual shocks in certain episodes. Accounting for variance switching helps avoid that these differencesarewronglyattributedtocoefficientswitching,whichisthekeyhypothesisinthe debateonsecond-roundeffects.4 In terms of methodology, we employ the regime-switching VAR model developed in Hubrich, Waggoner and Zha (2016). In contrast to earlier regime-switching VAR models, 2Forstatementsthatcouldbeinterpretedinthisdirection,seeforinstanceECBPresidentMarioDraghibefore theEuropeanParliamentonJune23,2011:Inprinciple,ifcommoditypricechangesareofatemporarynature,one canlookthroughthevolatilityininflationtriggeredbytheirfirst-roundeffects. However,theriskofsecondround effectsmustbecontrasted(...) topreventthattheyhavealastingimpactonmedium-terminflationexpectations. (...) Insuchcases,anadjustmentofthemonetarypolicystancewouldberequiredtopreservepricestabilityand keepinflationexpectationswell-anchored. AsimilardistinctionemergesfromChairYellen’sassessmentinthe December12,2015pressconference:Foranumberofyearsbetween2004and2008,wehadaseriesofincreases in oil prices that (...) raised inflation (...) and we judged those increases to be transitory as well and looked throughthem. Wedomonitorinflationexpectationsverycarefully. Ifwesawinameaningfulwaythatinflation expectationswereeithermovingupinawaythatmadethemseemunanchoredordown,thatwouldbeofconcern. 3See, for instance, Bernanke et al. (1997); Barsky and Kilian (2001); Bodenstein et al. (2008); Nakov and Pescatori(2010);Kormilitsina(2011);Bodensteinetal.(2012);andNatal(2012). 4Theimportanceoftime-variationinthevarianceofkeymacroeconomicaggregateshasmostclearlyemerged fromtheanalysisofcompetingexplanationsforthesteepdeclineininflationandoutputvolatilitysincethemid- 1980s(oftenreferredtoastheGreatModeration);see,forinstance,Primiceri(2005);SimsandZha(2006). 2

as proposed in Sims et al. (2008) and employed in Hubrich and Tetlow (2015), this method includestime-varyingtransitionprobabilitiesthatdependonthestateoftheeconomy. Moreover, it is, to our knowledge, the first to incorporate time-varying transition probabilities in a VAR framework (rather than in single equation regression models; see section 2 for further discussion). Theanalysisprovidesevidenceoftworegimesthatarecharacterizedbyqualitativelydifferent patterns in economic activity and inflation following oil price shocks. In the normal regime,downwardoilpriceshocks,forinstance,arefollowedbyonlytransitoryandmoderate declinesininflationandasmallincreaseineconomicactivity,whichmayattenuatetheinitial disinflationaryeffectoftheoil-pricedeclines. Intheadverseregime,bycontrast,theresponses of economic activity and inflation are much more sizeable and sustained than in the normal regime and both variables move in the same direction as the oil price shock. Market-based measures of inflation expectations, which may act as a symptom for the presence of secondroundeffects,alsoshowstrikingdifferencesacrossregimes. While,inbothregimes,inflation expectations initially adjust in the same direction as the oil price shock, this impact is again moderate and transitory in the normal regime whereas it is pronounced and persistent in the adverse regime. Overall, the dynamics characterising the adverse regime are consistent with thepatternstobeexpectedinthepresenceofsecond-roundeffectsthatmayamplifyandprolongtheimpactofanoilpriceshockoninflation. Thesystematicresponseofmonetarypolicy generallyworksagainstsuchsecond-roundeffectsintheadverseregimebutisinsufficientto fullyoffsetthem. Model extensions point to wage-price spirals, of the type described in Hofmann et al. (2012),asanimportantchannelbywhichsecond-roundeffectsmayoperate. Bycontrast,we donotfindsupportforainterest-ratechannel(proposed,e.g.,byObstfeldetal.(2016)),which wouldoperateiffallinginflationexpectationsputupwardpressureonlong-termrealratesand hence dampened real activity in the adverse regime (with analogous dynamics in response to risinginflationexpectations;seealsoBodensteinetal.(2013)). The analysis also delivers time-varying probabilities that allow us to form a view on whethertheeuroareaeconomyismorelikelytobeinoneregimeortheotheratagivenpoint intime,aswellastheconditionalprobabilityofstayinginthatregime,providedithasprevi- 3

ouslyprevailed. Sincethisinformationisavailableatregular(monthly)frequency,themodel may provide information supporting the deliberations on the adequate response to observed oilpricefluctuationsinreal-time. Thesefeaturesalsohelpuscommentonpastepisodesand, in particular, the recent debate on the macroeconomic implications of the oil-price declines observed since mid-2014 until early-2015. Our model indicates that the euro area economy wasinanadverseregimeatthattimeandhencefavoursapessimisticassessmentoftheirimplicationsforthestrengthoftheeuroareaeconomy. Indeed,counterfactualexperimentsshow that economic activity, inflation and inflation expectations over the second half of 2014 were much more sluggish than we would have expected, had the euro area not been mired in the adverse regime. These findings are robust to several modifications in the model set-up and sampleperiodsconsideredintheanalysis. Relation to the literature. Our paper contributes to an active literature aiming to shed light on how the implications of oil price shocks may differ depending on macroeconomic circumstances. Onestrandofthisliteraturefocusesonthesourcesoftheunderlyingoilprice shocks as a determinant of its macroeconomic implications. To this end, several papers have usedstructuralVARstomodeltheglobalcrudeoilmarketandfoundthatthemacroeconomic consequences of oil price fluctuations differ depending on whether they originate from an oil supply shock, an increase in aggregate demand, or an increase in the precautionary demand for oil (see, for instance, Kilian (2008, 2009), Peersmann and Van Robays (2009), Lippi and Nobili (2012), Baumeister and Peersman (2013), Kilian and Murphy (2014), Baumeister and Hamilton (2015), and Caldara et al. (2016)). A second strand of this literature focuses on differencesinthetransmissionofoilpriceshocks,eitherbycomparingpre-definedhistorical episodes (as in Blanchard and Galí (2007) and Nakov and Pescatori (2010)); or by developingmodelsthatexplicitlyallowfornon-linearities(asinHamilton(2003))andtime-variation in the impact of oil shocks (as in Van Robays (2012), Baumeister and Peersman (2013), and Leduc et al. (2016)). A third strand relevant to our analysis evaluates how central banks respond to oil price shocks, inter alia to explore whether changes in the systematic monetary policy response may explain the decline in the reduced-form relationship between oil prices andothermacroeconomicaggregatesduringtheGreatModeration.5 5Inadditiontosomeofthepapersquotedinfootnote3see,forinstance,Bernankeetal.(2004),Hamiltonand Herrera(2004)HerreraandPesavento(2009),andBjornlandetal.(2016). 4

Our paper is closest to the second strand of the literature in that it also explicitly models time-variationintheeffectofoilpriceshocks. But,toourknowledge,itisthefirstpaperto: (i) account for such time variation in the euro area context; (ii) grant an explicit role to inflation expectations in the propagation of oil price shocks; and (iii) use a regime-switching model withendogenousswitchingtoanalyseoilpriceshocks. Moreover,ourmodelprovidesanewperspectiveonhowoilpricefluctuationsentermonetary policy reaction functions – an issue that has been subject to a long-standing debate in the related literature. Initial contributions to this literature, focusing on the US context, point toapronouncedsystematicmonetarypolicyresponsetooilpriceshocksthatcounteractstheir impact on inflation (see Bernanke et al. (1997)). Subsequent analysis instead has favoured a more nuanced reaction function that is conditional on the source of the shock; in particular, bothVAR-andDSGE-basedanalyses,indicatethatUSmonetarypolicytypicallycounteracts oil price shocks if they originate from global demand pressures, while adopting the opposite response to oil price shocks driven by supply disruptions (see, e.g., Kilian and Lewis (2011) andBodensteinetal.(2012)). Ourpaperprovidesfurtherevidenceofsuchnuancedresponse, inthateuroareamonetarypolicyhastendedtocounteractoilpriceshocksonlywhentheyrisk inducingsecond-roundeffects. The remainder of the paper is organized as follows. Section 2 presents the methodology underlying the regime-switching model. Section 3 shows how the estimated economic adjustments in response to oil price shocks differ depending on the prevailing regime; it also reports how the estimated (conditional) probability of being (staying) in a certain regime has evolved since the start of the sample period; and it zooms in on the episode of collapsing oil prices since mid-2014, using counterfactual experiments to trace out how the euro area economy would have been expected to behave, if had it not switched to the adverse regime. Section4modifiesandextendsthebaselinespecificationtoexploredifferentchannelsgiving risetodifferencesacrossregimesandtoexpandthesamplebacktothe1970sforsomeofthe estimations. Section5concludes. 5

2 Methodology We employ the Regime-Switching Vectorautoregressive (RS-VAR) model with time-varying transitionprobabilitiesdevelopedinHubrich,WaggonerandZha(2016). Most of the methodological literature focusses on models with constant probabilities of Markovswitching,includingtheseminalpaperbyHamilton(1989),aswellasthesubsequent contributions by Chauvet (1998), Kim and Nelson (1999), Frühwirth-Schnatter (2004), Sims andZha(2006),Simsetal.(2008)andHubrichandTetlow(2015). In this paper, instead, we use a model that makes the transition probability dependent on the state of the economy, and therefore time-varying. This provides insights about how one or more variables in the system may affect the probability of the economy staying in one particularregime. A parallel strand of the literature allows for time-variation in the probability of regimeswitching but resorts to univariate or multivariate regression set-ups; see: Filardo (1994), Dieboldetal.(1993),Kim(2004),Kimetal.(2008),AmisanoandFagan(2013),Bazzietal. (2014)aswellasChangetal.(2017). Inthesepapers,theprobabilityofregimeswitchingdependsoncertainvariablesofinterest,buttheregressionset-upsdonotpermitfeedbackeffects amongtheendogenousvariables. Building on and extending the framework presented in Sims et al. (2008) (SWZ08), the RS-VARmodelbyHubrichetal.(2016)usedinthispapercombinesthesedifferentstrandsof the literature. It allows for time-varying transition probabilities which depend on the state of theeconomywhilemodellingtheinterdependenciesoftheendogenousvariablesandimposing structuralidentifyingassumptionsinaregimeswitchingstructuralvectorautoregressivemodel withtime-varyingtransitionprobabilities. In this paper we employ this methodology to investigate the transmission of oil price shocks. Given our focus, we choose oil price changes as the driver of switching between different regimes. This facilitates relating the interpretation of the resulting regimes to oil pricechangesandtheirtransmissionthroughthemacroeconomy. 6

2.1 Model For 1 ≤t ≤ T, let y be an n-dimensional vector of endogenous variables, let z be an mt t dimensional vector of exogenous variables, and let s be a discrete latent variable taking h t distinctvalues. LetΘbeavectorofparameterscontrollingthedistributionofy andletqbea t vectorofparameterscontrollingtheprocesss . Wewilldenote{y ,···,y }byY,{z ,···,z } t 1 t t 1 t byZ ,and{s ,···,s }byS . Weassumethedistributionoftheexogenousvariablez satisfies t 1 t t t p(z |s ,Y ,Z ,Θ,q)= p(z |Z ). (1) t t t−1 t−1 t t−1 In this paper, we consider the time varying structural vector autoregression (RS-SVAR) definedby A (sc)y =A (sc)x +Ξ−1(sv)ε , 0 t t + t t t t where x(cid:48) = (cid:2) y(cid:48) ,···,y(cid:48) ,1 (cid:3) , t t−1 t−p ε is a standard normal vector of shocks, and sc and sv take on values in {1,···,hc} and t t t {1,···,hv}, respectively. In the notation of equation 1, s =(sc;sv); the number of regimes t t t for each process ish = hc,hv and the regime changes in the coefficients are driven by the variable(s) influencing the probability of staying in a certain regime, while for the variance switching we assume a Markov process. The only exogenous variable is a constant, z =1. t In this paper, we allow for two coefficient regimes and for two shock-variance regimes. The vectorΘconsistsoftheelementsofA (sc),A (sc),andΞ(sv). 0 t + t t 2.2 TheTransitionMatrix We denote the time-varying probability for switching from regime j to i, p(s = i|s = t+1 t j,Y,Z ,Θ,q), by p and assume that the diagonal elements of the transition matrix, which t t i,j,t givethetime-varyingpersistenceofthe jth regime,areoftheform: 1 p = j,j,t 1+e−uj,t 7

where u =c +γ y . j,t j j t,t−k+1 and y(cid:48) = [y(cid:48),···y(cid:48) ]. With two regimes the off-diagonal elements will be of the t,t−k+1 t t−k+1 form (1−p ,j,t). For more than two regimes we assume that the off-diagonal elements j are Dirichlet-scaled. To achieve parsimony in the context of this complex specification, in our model we choose k = 1 and most of the elements of γ will be restricted to zero. The j off-diagonalwillbeoftheform: p =(1−p )pˆ i,j,t j,j,t i,j where∑ i(cid:54)=j pˆ i,j =1. Thepriorontheparametersc j andγ j willbenormal. Inthegeneralnotationpresentedhere,theprobabilityofstayinginaparticularcoefficientregimecandependon oneormore(andpossiblyall)oftheendogenousvariables. Inourmodel,weareinterestedin identifyingdifferentregimesforthetransmissionofoilpricefluctuations. Therefore,wemake theprobabilityofstayinginaparticularcoefficientregimedependentonoilpricechanges. 2.3 Estimationprocedure,dataandidentification We present our empirical results in terms of probabilities and impulse responses as estimates at the posterior mode. The posterior mode is computed via a blockwise BFGS optimization algorithmfollowingSWZ08. Itiswellknownthatthequalityoftheestimationoftheposterior mode depends on choosing good initial values for the algorithm. We used the Dynamic StriatedMetropolisHastingsSampler(Waggoner,WongandZha,2015)tosimulatetheposterior distribution. Weusedanumberofrandomdrawsasinitialvaluesfortheoptimizationroutine and chose the posterior model estimate that corresponded to the highest peak of all resulting local peaks. A by-product of this sampler is the computation of the marginal data density (MDD)thatweusetocomparethestatisticalfitofdifferentalternativespecifications. The RS-VAR, in the baseline specification, includes: euro area industrial production as a measureofeconomicactivity(which, incontrasttodataonrealGDP,isavailableatmonthly frequency);theeuroareaHarmonizedIndexofConsumerPrices(HICP)asameasureofinflation;theBrentcrudeoilprice(incurrentUS-Dollars);thebilateralUS-Dollar/Euroexchange 8

rate;6 5-year/5-year break-even inflation rates as a market-based measure of long-term inflationexpectations;andthe3-monthEURIBORasameasureforshort-terminterestrates.7 For identification, we apply a Cholesky decomposition. In our baseline specification, the variables are ordered as in the previous paragraph, which implies that we impose a zero restrictiononthecontemporaneouseffectofoilpriceshocksoneconomicactivityandinflation, while allowing the oil price to react to these variables immediately. This assumption is consistentwiththeapproachadopted,forinstance,inBernankeetal.(1997)andChristianoetal. (1996)andmaybemotivatedbytheroleofoilasaglobally-tradedfinancialassetwhoseprice could, in principle, reflect any relevant information at high frequency, whereas adjustments in industrial production and overall price dynamics in the economy are likely to proceed at a moresluggishpace.8 Atthesametime,substantialpartsoftherelatedliteraturehaveadopted an alternative approach, treating oil prices as predetermined with respect to innovations in economic activity and inflation – an assumption that received empirical support from Kilian andVega(2011).9 Against this background, we conducted two exercises to inform our modeling approach. First, in line with the related literature, we assessed the robustness of our model to an alternative ordering (with oil prices ranking first) and found our key findings to remain broadly unchanged.10 Second, we replicated the analysis in Kilian and Vega (2011) for the euro area sampleconsideredinthecurrentpaper(seeappendixA.1). Basedonthisanalysis,wecannot rule out a contemporaneous effect of euro area macroeconomic news on oil prices over the sampleperiod,thusarguingfortheorderingadoptedinthebaselinespecificationasthemore prudentapproachinthiscontext. All models with regime switching either in the variances or the coefficients, or both, outperform the constant parameter model in terms of their marginal data densities; accordingly, we statistically reject the hypothesis that there are no non-linearities. The model with two 6OneofthemodificationsofourbaselinespecificationdirectlyincludeoilpricesinEuroandthereforedrop theexchangeratefromthesystemofequations. 7Industrialproduction,HICPandoilpricesareincludedasyear-on-yearlog-changesandtheremainingvariablesaredefinedinpercent. 8SeeFratzscheretal.(2014). 9See,e.g.,DavisandHaltiwanger(2001),KiseokandNi(2002),LeducandSill(2004),BlanchardandGalí (2007),KilianandLewis(2011)andStockandWatson(2016). 10Forexample,BlanchardandGalí(2007)exploredifferentorderingsforthepriceofoilwhereasRotemberg andWoodford(1996)approachsimilarendogeneityconcernsthroughsamplesplits. 9

variance and two coefficient regimes and constant transition probabilities is not significantly different from that same model with a time-varying transition matrix. This, in turn, implies that estimating additional parameters to allow for a time-varying probability of staying in a regime conditional on the state of the economy does not deteriorate the statistical fit, while providingfurtherinsightsontheroleofoilpricesindeterminingeconomicdynamics. ThebasicsampleincludeseuroareadataatmonthlyfrequencyovertheperiodfromFebruary 2004 to January 2015 and, for some of the model extensions, the period from February 2004 to December 2015. The starting point of the sample is dictated by the availability of data on break-even inflation rates, which were not recorded in a consistent manner prior to February2004(norwereanyothermarket-basedmeasuresoflong-terminflationexpectations and survey-based measures of inflation expectations are generally not available at monthly frequency). Theuseofdifferentsampleend-datesismotivatedbytheevolutionofthemacroeconomic environmentoverrecentyears: tokeepthebaselinemodeltractable,itincludesoneshort-term interest rate variable to capture the prevailing monetary policy stance; however, since mid- 2014,andevenmoreforcefullysinceend-January2015,theECBhasstartedrelyingonasset purchaseprogrammestoinjectadditionalmonetarypolicyaccommodation,alsoreflectingan increasinglylimitedscopeforfurtherreductionsinmonetarypolicyinterestrates. Sincesuch centralbankassetpurchasesdirectlyinterveneatthelongerendoftheyieldcurve,itappears moresuitabletomovetoabroaderspecificationinthisenvironment,includingalsoalongerterm interest rate to provide a more comprehensive characterization of the monetary policy stance. For this reason, we first estimate the baseline model over a shorter sample period until end January 2015, before the ECB engaged in its expanded asset purchase programme (see section3); second, weextendthemodeltoalsoincludedatafortheentireyear2015andadd ameasureforthelong-terminterestrateasafurthervariabletothemodel(seesection4). 10

3 Regime-dependent effects of oil price shocks – baseline results 3.1 Differencesinthegrowth-andinflation-responseacrossregimes The results from our baseline specification, shown in Figure 1, provide evidence of two regimes that are characterized by very different patterns in economic activity and inflation inresponsetotheoilpriceshock. Inthenormalregime(slashedblueline)oilpriceshocksare followedbyonlysmallmacroeconomicfluctuations. Inflationbrieflydeclinesaftertheshock, but only by a few basis points and the decline is fully reversed over the 24-months horizon forwhichimpulseresponsesareshown. Inflationexpectationsfollowasimilarpathasactual inflation,decliningslightlyaftertheshockbutthenrecoveringafterafewmonths. Economic activity,inturn,increasesslightlyaftertheshock(inlinewiththebenigninterpretationofthe oilpriceshockassupportingdomesticdemand). In the adverse regime (solid red line), the oil price shock is slightly larger on impact (in bothregimes,theshockiscalculatedasonestandarddeviationoftheobservedoilpricemoves over the time periods when the respective regime prevailed; accordingly, the difference in themagnitudesoftherespectiveshocksindicatesthatoilpricestendedtobesomewhatmore volatileintheadverseregime). Buttheoilpricethenrecoversmorequicklythaninthenormal regimeand,overall,doesnotdisplaystrikinglydifferentpatternsacrossregimes. The impact of the shock on the other variables in the system does, however, differ in relevant ways. Instead of rising, as in the normal regime, economic activity declines, hitting a trough after two quarters that is almost a percentage point below steady state and, despite some recovery, remains below steady state levels until the end of the horizon. In contrast to the benign regime, where economic activity acts as a stabiliser, this downward effect of the oil price shock on economic activity thus tends to amplify the disinflationary forces. As a consequence, actual HICP inflation undergoes a pronounced and persistent decline, settling almost0.2percentagepointsbelowsteadystatebytheendofthehorizon. Thisdeclineisalso reflected in inflation expectations, which show a similarly modest drop on impact as in the normalregime,butthencontinuetodriftdownovertheentirehorizon.11 11One could argue that, under certain circumstances, an adjustment in inflation expectations as observed in theadverseregimemaysupportthecentralbankincounteractingdeviationsofinflationratesfromitspreferred levels. Forinstance,ifoilpricessuddenlyriseinconditionsofbelow-targetinflationrates,anupwardadjustment ininflationexpectationsmayacceleratethereturntotarget.Still,weconsistentlyrefertothisregimeas“adverse” 11

Overall, the dynamics observed in the adverse regime are consistent with second round effectsexertingaprotractedadverseimpactontheeconomy,translatingintodeclinesinactual aswellasexpectedinflation(furtherintuitionandanassessmentofthechannelsthatmaygive risetosecond-roundeffectsareprovidedinthebelowDiscussionandsection4).12 Failing to account for these regime-dependent dynamics may lead observers to miss important characteristics of the inflation process and economic activity in the aftermath of an oil price shock. This becomes clear when comparing the impulse response functions from a constant parameter VAR (also plotted in Figure 1 via the dotted black line) with those from theregime-switchingmodel. Inparticular,restrictingcoefficientsandvariancestobeconstant mayleadobserverstounderestimatethepronouncedandpersistenteffectsthatoilpriceshocks mayexertoneconomicactivityandinflationintheadverseregime,whileprovidingthewrong signforoutputandinflationresponseinthenormalregime. Bothtypesofmisjudgementmay, inturn,contributetopolicymistakesintheresponsetooilpriceshocks. Theregime-switchingframeworkalsoprovidesforarichercharacterizationofthemonetarypolicyresponsetooilpriceshocks. TheconstantparameterVARwouldpointtoarather mechanical response by which monetary policy would always tend to accommodate the oil price shocks (via a monetary tightening following upward shocks and a loosening following downward shocks to the price of oil). Consistent with recent literature and with central bank communication, however, the RS-VAR instead favours a more nuanced assessment. In fact, monetary policy accommodates the oil price shock only in the adverse regime, e.g. leading todecliningshort-terminterestratesinresponsetoadropinoilprices(althoughtheresultant monetary loosening is not sufficient to offset the disinflationary forces). By contrast, in the normalregime,theshort-terminterestrateinsteadwouldtendtorisefollowingthedownward oilpriceshock,thuscounteractingtheensuingboostineconomicactivity. Thesepatternsecho thefindingsfromrecentliteratureaimingtodistinguishmonetarypolicyresponsestodifferent typesofshocks(seesection3.3forfurtherdiscussion). basedonthewell-establishednotionthatade-anchoringofinflationexpectationsgenerallyrendersacentralbank’s macroeconomicstabilizationobjectiveshardertoachieve;seeWoodford(2007). 12Exchangeratesdonotplayamajorroleintheadjustmentdynamics, possiblyreflectingthecountervailing effectsofthedecliningoilprice,whichwouldtendtoinduceanappreciationoftheEuroversustheUS-Dollar, andtheweakeningeuroareaeconomy,whichwouldinduceadepreciation. 12

Figure 1: Impulse responses to oil price shock (one standard deviation shock), slashed blue: normalregime,solidred: adverseregime 13

3.2 Episodesofsecond-roundeffects Figure2displayssmoothedprobabilitiesofbeinginanormalregime(withgrey-shadedareas showing periods in which this probability exceeded 50%) and the time-varying conditional probabilitiesofstayinginanormalregime(depictedbythesolidblackline). The smoothed probabilities show that the euro area economy entered the adverse regime at various occasions since the start of the sample. Typically, regime-switches occurred after a sequence of pronounced, unidirectional oil price changes – for instance in the episodes of strong oil price declines and subsequent increases over the period 2008-2011 (for reference, seeFigure9plottingtheevolutionofoilpricesoverthesampleperiod). InAugust-2014, the euroareaeconomyagainswitchedtotheadverseregimeand,afterashortperiodinthenormal regimeinthelastquarterof2014,returnedtotheadverseregimebytheturnoftheyear. Meanwhile,theconditionalprobabilityofstayinginanormalregime(giventheeconomy wasinthatregime),whichisafunctionoftheoilpriceevolution,declinedsteeplyjustbefore theeconomytransitionedtotheadverseregimeinlate-2008. Afteraprolongedperiodwitha highconditionalprobabilityofstayinginthenormalregime(whichstartsinend-2009–before the smoothed probabilities actually indicate a return to the normal regime), this probability againfellsteeplyinthesecondhalfof2014andapproachedzerobyJanuary2015. Accordingly, our model clearly assigns the period from August 2014 to January 2015, which had triggered intense debate among the policy observers (see section 1), to the adverse regime. Hence, the regime-switching framework indicates that the oil price declines observedoverthatperiodarelikelytohavereinforcedtheprevailingdisinflationarypressures viasecond-roundeffects,thusfavoringthemorepessimisticassessmentofthatepisode. 3.3 Regime-dependenttransmissionversusdifferentsourcesoilpriceshocks Itisinterestingtorelatethesefindingstothestrandoftheliteraturethatemphasisesdifferent sourcesof oilpriceshocks asakey determinantoftheir macroeconomiceffects. Whencomparingtheimpulseresponsefunctionsbetweenthenormalandtheadverseregimes,important parallels emerge with that literature. In the normal regime, the macroeconomic adjustments tooilpricefluctuationsresemblethedynamicsonemayexpectafteragenuinesupply-driven shock: in response to falling oil prices, for instance, euro area economic activity temporarily 14

strengthens,consistentwithincreaseddisposableincomeforeuroareahouseholdsandreduced energy input costs for euro area producers; meanwhile, the oil price decline initially exerts some moderate direct downward pressure on inflation, probably reflecting the role of energy intheconsumptionbasketandsomepass-throughoflowerinputcoststoconsumerprices,but thedeclineisquicklyoffsetbytheuptickinactivity. Consistentwiththeshort-livedinflation response,inflationexpectationsquicklyreturntosteadystate. Intheadverseregime,bycontrast,theco-movementpatternsareclosertowhatmaybeexpectedinresponsetoaglobaldemand-drivenoilpriceshock. Againresortingtotheexample of falling oil prices, the concomitant protracted slump in economic activity, which materialises despite an initial beneficial effect on euro area disposable incomes and input costs, is consistent with a situation in which the oil price decline signals a downgrade in broader economic prospects, inducing firms and households to revise down their expectations for future inflationand,viathischannel,providinganincentivetopostponenominalspendingdecisions. Thisinter-temporalsubstitution,inturn,furtherdepressesdomesticactivityandreinforcesthe downwardpressureoninflation,thusgivingrisetosecond-roundeffects. Thedifferentmonetarypolicyresponsesacrossregimesfurtherstrengthentheseparallels. For instance, using a structural VAR, Kilian and Lewis (2011) find that US monetary policy tended to counteract only demand-driven oil price shocks, which the authors rationalise with theintentiontopreemptwage-pricespirals,whereassupply-inducedoilpriceshockstriggera monetarypolicyresponseintheoppositedirection. Thispattern,whichhasalsofoundsupport in DSGE-frameworks (see, e.g., Natal (2012) and Bodenstein et al. (2012)), again conforms withtheshort-terminterestratedynamicswedetectinthedifferentregimes,withthenormal regimemappingintoasupply-shockandtheadverseregimeintoademand-shockscenario. Whiletheseparallelsareintuitivelyplausible,theoverlapbetweenthedifferentapproaches remains partial. First, the regime-switching framework produces a sharper distinction in the estimatedimpulseresponsefunctionsbetweenthedifferentregimesthanwhattherelatedliteraturefindsfordifferenttypesofshocksintheeuroareacontext. Forinstance,Peersmannand VanRobays(2009)findaquantitativelysimilar,positive,impactofoilpriceincreasesoneuro areainflation,independentlyofwhetheritisdrivenbyasupplyordemandshock(incontrast tothequalitativelydifferentpatternsobservedinFigure1). Second,theperiodsassignedtothe 15

adverseregimeinourmodelhavereceiveddifferentappraisalswhenviewedthroughthelens of the nature of shocks. For instance, the shock decompositions in Baumeister and Hamilton (2015)indeedattributeanimportantpartoftheoilpricecollapseattheonsetofthe‘GreatRecession’in2008toglobaleconomicactivityandoildemandshocks.13 Whileourmodelalso assignstheeuroareaeconomytotheadverseregimeinthisepisode,thisregimepersistswell beyond the period over which Baumeister and Hamilton (2015) see global economic activity anddemandshockstodominate;also,accordingtoourmodel,theeuroareaeconomyre-enters the adverse regime around end-2010 to early-2011 – a period over which global activity and demandshockshavebeenfairlymuteaccordingtoBaumeisterandHamilton(2015). Accordingly, the regime-switching perspective constitutes an interesting complement to theexistingliteraturethatdisentanglessupply-anddemand-factorscausingoilpricechanges. In particular, by allowing for different propagation patterns of oil price shocks at different points in time and determining the probability of being in a specific regime, the model may add to policy-makers’ information sets when deciding on the appropriate response to an observedoilpricefluctuation. Figure2: Time-varyingprobabilityforbeinginanormalregime(grey-shadedarea)andconditionalprobabilityofstayinginthatregime(blackline) 13BaumeisterandHamilton(2015)measureglobaleconomicactivitybyindustrialproductionintheOECDand sixlargenon-OECDeconomies. 16

3.4 Theroleofinflationexpectations Asafinalexercisewithourbaselinemodel,Figure3presentsacounterfactualexperimentthat allowsustogainfurtherinsightintothecontentiousepisodestartinginthesecondhalfof2014 andtoillustratehowthedynamicsofthesystemdifferacrossregimes. Thebasicset-upofthis counterfactualexperimentistotraceouthowthedifferentvariableswouldhaveevolvedifthe euroareaeconomyhadnotswitchedtotheadverseregimebyAugust2014. The counterfactual points to striking differences in economic outcomes across regimes. Economic activity and inflation would have been more than 1.5 percentage point higher than observed by the end of 2014, had the normal rather than the adverse regime prevailed at that time. Moreover,inflationexpectationswouldhavecontinuedtohoveraround2%,ratherthan followingamarkeddownwardpathtoanewhistoricalminimumof1.6%bytheendof2014 – in contrast to the early years of the crisis, when inflation expectations faced upward pressure from rising oil prices thus contributing to the “missing disinflation” in 2009-2011 (see CoibionandGorodnichenko(2015)). Interestingly,alsotheoilpricewouldhavestagedadynamic recovery instead of the actual steep decline. This indicates that the weakness in euro areaeconomicactivityandinflationfrommid-2014toearly-2015hasbeenoneofthedrivers oftheglobaloilpricecollapseoverthesecondhalfof2014. 17

Figure 3: Counterfactual Experiment: Normal regime instead of volatile oil price regime (greendashedline: actualdata,redsolidline: counterfactualpath) 18

4 Extensions Whiletheadverse-regimedynamicsinthebaselinespecificationareconsistentwiththepresence of second-round effects, further analysis is necessary to identify potential channels by which these effects may operate. To explore this issue and to test the robustness of our key findings, we modify and extend our model in several directions. First, we estimate a more parsimonious specification that directly expresses oil prices in euro and omits the exchange ratefromthemodel;theresultantreductioninthesizeoftheVARisconvenientinviewofthe additionalvariablesthatarelateraddedtothemodeltostudydifferentchannelsand,froman economic perspective, may be motivated by the fact that ultimately it is the oil price in euro thatmattersforeuroareaconsumersandfirms. Second,weaugmentthemodelwithmeasures for euro area wage growth and the long-term real interest rate, so as to study two alternative channels that the related literature has proposed as giving rise to second-round effects. Third, we re-run the regime-switching model on a different euro area data set that extends significantly further back in time (ranging back to the 1970s), while requiring us to move to quarterlyfrequencyandtoexcludeinflationexpectationsfromthesetofvariables. 4.1 Oilpriceineuro–amoreparsimoniousspecification As apparent from Figures 4 and 5, the specification with oil prices in euro confirms our key results. In particular, the impulse response functions show nearly identical patterns for economic activity, inflation, and inflation expectations as in the baseline, and the differences across regimes are similarly pronounced. Also, the assignment of periods to the two regimes is broadly unchanged, albeit with somewhat more frequent switches to the adverse regime over the period 2010-2012 than in the baseline specification. The period around the turn of theyear2014to2015isagainassignedtotheadverseregimeandthedropintheconditional probabilityofstayinginthenormalregimeinthesecondhalfof2014isalsoconfirmedbythis alternative specification. Accordingly, this more parsimonious model is a suitable substitute to the baseline when adding further variables, and in particular wages, to study the channels givingrisetosecond-roundeffects. 19

4.2 Wage-pricespiralsandtherealinterestratechannel As explained by Hofmann et al. (2012), a wage-channel for second-round effects may operate if nominal wage growth rises (falls) in response to observed increases (declines) in the inflation rate, thereby amplifying the initial (dis-)inflationary push via a mutually reinforcing wage-pricespiral(seep. 769). Asasourceforsuchwage-pricespirals,theauthorsemphasise theexistenceofexplicitorimplicitwageindexation. Sincewageindexationisdeterminedby (typically slow-moving) changes in labour market institutions, it does not appear as a likely candidate for the relatively frequent regime-switches reported in Figures 2 and 5 of the currentpaper,whichweinsteadrationalizedwithadjustmentsininflationexpectations. However, followingastandardhybridPhillipscurvespecification(seeGalíandGertler(1999)andBlanchardandRiggi(2013)),wageindexationmightoperateviaanalogouseconomicmechanisms as a de-anchoring of inflation expectations: like wage indexation, a de-anchoring of inflationexpectationsraisestheweightonthebackward-lookingcomponentofthehybridPhillips curve,thusfacilitatingasimilartypeofwage-pricespiral.14 The real interest rate channel, emphasised for instance by Obstfeld et al. (2016), would operate if the fall in inflation expectations triggered by the oil price decline were to induce an effective tightening in financial conditions which in turn dampens aggregate demand – a scenario that is particularly likely if monetary policy is constrained in its ability to inject additional accommodation to counteract the upward pressure on long-term real interest rates and arrest the decline in inflation expectations. As a consequence, even an initially benign oil-pricedeclinemayexertanegativeeffectoneconomicactivity. The evidence from our model confirms wages as a channel for second-round effects, whereasrealinterestratesdonotemergeasarelevantfactor. Chart6addstheannualgrowth rate of nominal negotiated wages in the euro area to the previous specification. Consistent withourearlierfindings,economicactivity,inflationandinflationexpectations,intheadverse regime,followoilpricesontheirwaydown. Inthenormalregime,bycontrast,thesevariables showonlyaverysmallresponseandanoppositesignforeconomicactivity(comparedtoearlierspecificationsthereisamoredynamicrecoveryintheoilprice,whichalsopullsupactual 14Formally,thisbecomesapparentfromacomparisonoftheγ-parametersinequation7inHofmannetal.(2012) andinequation24inGalíandGertler(1999). 20

and expected inflation). As would be expected in the presence of wage-based second-round effects,wages–afteraninitialuptick–enteronalastingdownwardpathintheadverseregime whereasnosuchdynamicsmaterialiseinthenormalregime. Chart7extendsthebaselinemodelwithameasureofthereallong-terminterestrate, definedasthe10-yearsovereignbondyieldsofeuroareacountries,weightedbytheirrespective shares in euro-area GDP and deflated with market based inflation expectations over the same maturity. It also extends the sample period until end-2015 since the inclusion of a long-term interest rate variable renders the model better suited to account for the effects of the ECB’s expandedassetpurchaseprogrammeadoptedinJanuary2015(seesection2.3).15 Theimpulse responsefunctionsagainconfirmthebroadfindingsfromourbaselinespecification. But,contrarytothehypothesisofareal-interestratechannel, whichwouldrequireanincreaseinreal rates in the adverse regime, the real rate falls. A possible interpretation is that monetary policy,throughforwardguidanceandnon-standardmeasuresthatdirectlyinterveneatthelonger end of the yield curve, has been able to overcompensate the upward pressure on real rates triggeredbythedeclineininflationexpectations,evenasthescopeforshort-terminterestrate cutsbecamelimited. 4.3 Expandingthesamplebacktothe1970’s Theuseofinflationexpectations,forwhichthedatastartonlyin2004,hasimpliedanexclusionofseveralinterestingoil-priceepisodesfromthescopeoftheanalysis. Asafinalexercise, weaddressthislimitationbyre-runningtheRS-VARontheArea-wideModeldatabase,which providesquarterlydatabacktothe1970s(seeFaganetal.(2005)). Inanalogytothebaseline, the variables in the model include real GDP growth, HICP inflation, the oil price change in USD,thebilateralUS-Dollar/Euroexchangerate,nominalpercapitawagegrowthandanominal short-term interest rate. The sample begins in 1974, when crude oil prices started being determinedonanintegratedglobaloilmarket(seeKilian(2014)). Basedonthislongersample,weagainfindopposingsignsfortheresponseofeconomicactivityinthenormalversustheadverseregime,withrealGDPfallinginresponsetoadeclinein thepriceofoil. Atthesametime,therealGDPresponseisweakerthanthatofindustrialpro- 15Forsakeofcomparability,weusedthesamesamplealsoforthespecificationincludingwagesshowninFigure 6. 21

ductioninthepreviousspecifications,consistentwiththeinclusionofseveralless-responsive expenditure components, such as government consumption, in the real GDP variable. Meanwhile,theresponseininflationalsodiffersacrossregimes,butinamorenuancedmannerthan overtheshortersample: thedynamicsintheadverseregimearecharacterisedbyanadjustment inthesamedirectionastheoilpricethatissimilarto,albeitsomewhatflatterthan,inthebaseline. Following our previous reasoning, the positive co-movement of economic activity and inflation with the oil price points to the presence of second-round effects – an interpretation thatisalsosupportedbythecorrespondingdeclineinwagegrowth. Inthenormalregime,by contrast,theinflationresponseisinitiallyverysubdued,thusechoingthecross-regimedifferencesfoundfortheshortersample. However,inflationthencontinuestoproceedonasecular downward path that even crosses that from the adverse-regime after around one and a half years. Accordingly, the regimes differ in the pace with which inflation adjusts to the shock, rather than in the direction of the response (in contrast to the baseline where the responses differ in sign). Given qualitatively similar patterns in wage growth, the results indicate that, for the longer sample, second-round effects constitute a more pervasive phenomenon than in the post-2003 period considered in our earlier specifications; the latter, in turn, are likely to bemorerepresentativeforcurrentconditionsgiventhesampleisconfinedtorelativelyrecent observations. Itisnoteworthy,thattheassignmentofdifferenttimeperiodstothetworegimesdisplays substantial overlap with our previous specifications for the parts of the sample that coincide. Again, the post-Lehman crisis episode, as well as the period of collapsing oil prices starting in second half of 2014 are assigned to the adverse regime. The respective regime episodes aremorepersistent,however,inthatsomeofthebriefswitchesafter2009backtothenormal regimefromtheshortersamplearenotpickedupbythisspecification. Over the earlier (non-overlapping) sample, three additional episodes are assigned to the adverseregime,whichmayexplainsomeofthedifferencesintheimpulseresponsescompared to the baseline (even though the important broad features carry over): the first in the mid- 1970s, the second from the mid-1980s to the early 1990s, and the third from the late 1990s tothemid-2000s. Allthreeepisodeswerecharacterisedbyapronouncedco-movementinoil andbroaderconsumerpriceinflation. 22

Inparticular,thelattertwoepisodescorrespond,respectively,to: (i) theperiodofdeclining and converging euro area inflation rates in the run-up to the Maastricht Treaty; and (ii) the periodofrisinginflationintheyearsrightaftertheintroductionoftheeuro. Accordingly,our model would indicate that the pre-euro area disinflation process may have been supported by downwardsecond-roundeffectsinducedbytheoilpricedeclinesoverthatperiod;andthatthe concerns about upward second-round effects that motivated a sequence of ECB rate hikes in theearlyyearsoftheeuroareamayhavebeenwarrantedatthattime.16 The first adverse-regime episode, in turn, coincides with the oil price hikes around the 1973YomKippurWarandtheensuingAraboilembargoin1973-1974. Bycontrast,thelate- 1970s to early-1980s – another period characterised by pronounced oil price spikes – are not assigned to the adverse regime, consistent with the relatively short-lived increase in inflation aroundthattime,whichwasswiftlyreversedwiththeonsetofaglobaldis-inflationaryphase overthe1980s. 5 Conclusions In this paper, we have analysed whether the dynamics of euro area economic activity and inflationfollowingoilpriceshocksundergoesepisodicchanges. Tothisend,weusedaregimeswitchingVARwithendogenousswitchingfollowinganovelapproachdevelopedbyHubrich etal.(2016). Wefindthatoilpricefluctuationsaretypicallyfollowedbylimitedadjustments in inflation and economic activity, a situation we refer to as a normal regime. Occasionally, however, the economy enters into an adverse regime where oil price shocks herald sizeable and sustained macroeconomic fluctuations, with inflation (actual and expected), as well as economicactivity,movinginthesamedirectionastheoilpriceshock. Overall,thedynamics observedintheadverseregimeareconsistentwiththepresenceofsecond-roundeffectsofoil price shocks on growth and inflation, with wage-price spirals acting as a key channel for the positiveco-movementbetweenthesevariables. Zooming in on the episode of collapsing oil prices from mid-2014 to early-2015 – which generatedalivelydebateamongcentralbankobserversonwhetherornotitwarrantedamon- 16AlreadyinJanuary2000, theintroductorystatementtotheECBpressconferencereferredtosecond-round effectsasarisktopricestability. 23

etarypolicyresponse–ourmodelindicatesthattheadverseregimeislikelytohaveprevailed atthattime. Accordingly,concernsofnegativesecond-effectsreinforcingthethen-prevailing disinflationary pressures appeared warranted. In fact, according to our baseline estimates, economic activity and inflation would have been more than 1.5 percentage point higher than observed,hadtheeuroareaeconomynotbeenmiredinanadverseregimeatthattime. Moreover,inflationexpectationswouldhavecontinuedtohoveraround2%,ratherthanfollowinga markeddownwardpathtoanewhistoricalminimumof1.6%bytheendof2014. 24

A Appendices A.1 Responseofoilpricestoeuroareamacroeconomicnews The related literature is split on whether oil prices may be treated as predetermined with respecttootherkeymacroeconomicaggregates(seesection2.3). Thisquestioniscentraltothe appropriateorderingofvariablesinrecursiveidentificationschemesofthetypeconsideredin the current paper. Kilian and Vega (2011) propose a test of the hypothesis of predetermined energy prices which relies on regressions of WTI crude oil and US gasoline prices on the surprise component of a wide range of US macroeconomic indicators (defined as the differencebetweendatareleasesandexantesurveyexpectationsfortherespectivevariable). When accounting for potential distortions in test statistics due to “data mining”,17 the estimates in Kilian and Vega (2011) fail to reject the null hypothesis that energy prices are contemporaneously unaffected by macroeconomic news, regardless of which indicator is included as explanatoryvariable; whethertheregressionsarebasedonadailyormonthlyfrequency; and whethertheyincludeeachmacroeconomicnewsindicatorseparatelyorjointlyinamultivariate set-up. Accordingly, the results lend support to an identification strategy that orders oil beforethevariablesmeasuringeconomicactivityandinflation. Here, we replicate this exercise for the euro area sample considered in the current paper (focusing on the impact of news regarding inflation and industrial production, which are the two macroeconomic variables included in the VAR model). An assessment of whether the evidence in support of predetermined oil prices carries over to this sample appears warrantedespeciallyinviewofthemorerecenttimeperioditcoverscomparedtoKilianandVega (2011). AsdocumentedbyFratzscheretal.(2014),thebehaviourofoilpriceshasundergone a marked shift in the early 2000s, becoming more sensitive to changes in other asset prices sincethen. Oursample,whichstartsin2004,fallsentirelyintothe‘higher-sensitivity’period, whereastheKilian-Vegasamplefortheoil-priceregressionsstretchesfrom1983to2008,thus predominantlyconsistingofobservationsfromthe‘low-sensitivity’period.18 17Formally,thesuiteofregressionsinKilianandVega(2011)isusedtotestonehypothesis,namelythatmacroeconomicnewshaveanon-zeroeffectonenergyprices,usingmany differentspecifications. Thistypeofspecificationsearchrunscountertostandardnotionsofstatisticalinference(seeLeamer(1978)),andinparticularmay leadtoanover-rejectionofthenullhypothesisofnoeffect. 18LikeKilianandVega(2011),Fratzscheretal.(2014)donotfindsignificanteffectsformostmacroeconomic news indicators but this does not provide sufficient support for the assumption of predetermined oil prices in 25

The results of this exercise speak against ordering oil prices before the macroeconomic variablesinouranalysis. Toseethis,table1presentsOLSestimatesoftheequation: R =α+βS p +γSi+ε t t t t where R is the log-change in the price of oil from the end of day t−1 to the end of day t; t S p and Si, respectively, are news released on day t regarding euro area industrial production t t and headline HICP inflation (i.e. the two macro variables ordered before the oil prices in the baseline specification introduced in section 2.3); β and γ are the slope coefficients; α is an intercept and ε an error term. Following the approach in Kilian and Vega (2011), the news t variablesarecalculatedasthedifferencebetweenthedatareleaseandthemedianexpectation from Bloomberg surveys for the respective variable and normalised by the sample standard deviationofthatdifferencetofacilitatetheinterpretationofcoefficients. Thedatarangefrom February2004tomid-September2016(fordescriptivestatisticsseetable2). Table1: Regressionestimates(oilpricechangesasdependentvariable) inUSD inEUR Industrialproduction -0.013 -0.073 (0.089) (0.091) HICPinflation 0.142∗ 0.145∗ (0.076) (0.084) Constant 0.019 -0.065 (0.112) (0.122) Observations 292 292 R2 0.006 0.006 *p<0.10**p<0.05and***p<0.01.Hetroskedasticityrobuststandarderrorsinparentheses. While surprises in industrial production do not affect oil prices, the estimates point to a positivecoefficientonHICPsurprisesthatissignificantata10%level(againstatwo-sidedalternative),independentofwhetheroilpricesareexpressedinUSDandEURterms.19 Notwiththecontextofouranalysis: first,Fratzscheretal.(2014)includethenewsindicatorsalongsidearangeofother financialassetpricesthatmayimmediatelyreflectanymacroeconomicnewsandmayhenceconcealsignificant effectsduetomulti-collinearity;ourmodel,bycontrast,doesnotincludesomeofthesefinancialassetprices(in particularequitieswhicharelikelytobeparticularlysensitivetomacroeconomicnews)andmayhencepointto differentconclusions. Second, Fratzscheretal.(2014)includeawiderrangeofmacroeconomicnewsvariables beyondthoserelevanttoourmodelwhichmayreinforcemulti-collinearityissues. 19TestingH 0 againstaone-sidedalternative,asinKilianandVega(2011),wouldrenderγsignificantata5% level, whileleavingtheconclusionsunalteredregardingβ. UnlikeKilianandVega(2011), wedonotcompute ‘data-miningrobuststandarderrors’sincetheregressionsarebasedononlyonespecificationtotestthesignificance 26

standingtherelativelylimitedquantitativeeffectandanoveralllowexplanatorypowerofthe regression,thesignificanteffectofHICPsurprisesindicatesthatitismoreprudenttonottreat oilpricesaspredeterminedintheMS-VARintroducedinsection2.3. Table2: Summarystatistics Variable Mean Std. Dev. Min. Max. OilpriceinUSD 0.018 1.888 -7.572 8.017 OilpriceinEUR -0.06 2.045 -11.507 9.164 Industrialproduction -0.105 1 -4.482 4.98 HICPinflation -0.019 1 -2.768 2.768 Notes: Oilpricevariablesdefinedas100timeslogchangeintheoilpricelevelfromendofdayt−1toendof dayt. Industrial production and HICP inflation defined as difference between release and ex ante expectations normalisedbystandarddeviation;seetextforadditionaldetail. ofthetwomacrovariablesrelevantfortheMS-VAR.ThisisincontrasttoKilianandVega(2011)whoestimate multiple specifications to test the hypothesis that any element of a set of 30 macroeconomic news variables is significant;itisthisrepeatedapplicationofthesamestatisticaltestthatmotivatestheiruseofdata-miningrobust standarderrors. 27

References Amisano, G. and Fagan, G. (2013). Money growth and inflation: A regime switching approach,JournalofInternationalMoneyandFinance33:118–145. Barsky,R.B.andKilian,L.(2001). DowereallyknowthatoilcausedtheGreatStagflation? amonetaryalternative,NBERMacroeconomicsAnnual200116:137–183. Baumeister, C. and Hamilton, J. D. (2015). Structural interpretation of vector autoregressions with incomplete identification: Revisiting the role of oil supply and demand shocks, WorkingPaper. Baumeister,C.andPeersman,G.(2013). Time-varyingeffectsofoilsupplyshocksontheus economy,AmericanEconomicJournal: Macroeconomics5(4):1–28. Bazzi, M., Blasques, F., Koopman, S. J. and Lucas, A. (2014). Time Varying Transition ProbabilitiesforMarkovRegimeSwitchingModels,TinbergenInstituteDiscussionPapers 14-072/III,TinbergenInstitute. Bernanke, B. S., Gertler, M. and Watson, M. (1997). Systematic monetary policy and the effectsofoilpriceshocks,BrookingsPapersonEconomicActivity1. Bernanke, B. S., Gertler, M. and Watson, M. W. (2004). Reply: Oil shocks and aggregate macroeconomicbehavior: Theroleofmonetarypolicy,JournalofMoney,CreditandBanking36(2):287–291. Bjornland, H. C., Larsen, V. H. and Maih, J. (2016). Oil and macroeconomic (in)stability, WorkingPaper2016/12,NorgesBank. Blanchard, O. J. and Galí, J. (2007). The macroeconomic effects of oil shocks. why are the 2000ssodifferentfromthe1970s?,NBERWorkingPaper(13368). Blanchard, O. J. and Riggi, M. (2013). Why are the 2000s so different from the 1970s? a structural interpretation of changes in the macroeconomic effects of oil prices, Journal of theEuropeanEconomicAssociation11(5):1032–1052. 28

Bodenstein, M., Erceg, C.J. andGuerrieri, L.(2008). Optimalmonetary policywithdistinct coreandheadlineinflationrates,JournalofMonetaryEconomics55:S18–33. Bodenstein,M.,Guerrieri,L.andGust,C.J.(2013).Oilshocksandthezeroboundonnominal interestrates,JournalofInternationalMoneyandFinance32(C):941–967. Bodenstein, M., Guerrieri, L. and Kilian, L. (2012). Monetary policy responses to oil price fluctuations,IMFEconomicReview60(4):470–504. Caldara, D., Cavallo, M. and Iacoviello, M. (2016). Oil Price Elasticities and Oil PriceFluctuations,InternationalFinanceDiscussionPapers1173,BoardofGovernorsoftheFederal ReserveSystem(U.S.). Chang,Y.,Choi,Y.andPark,J.Y.(2017).Anewapproachtomodelregimeswitching,Journal ofEconometrics196(1):127–143. Chauvet,M.(1998). Aneconometriccharacterizationofbusinesscycledynamicswithfactor structureandregimeswitching,InternationalEconomicReview39(4):969–996. Christiano, L. J., Eichenbaum, M. and Evans, C. (1996). The effects of monetary policy shocks: Evidence from the flow of funds, The Review of Economics and Statistics 78(1):16–34. Coibion, O. and Gorodnichenko, Y. (2015). Is the phillips curve alive and well after all? inflation expectations and the missing disinflation, American Economic Journal: Macroeconomics7(1):197–232. Davis,S.J.andHaltiwanger,J.(2001). Sectoraljobcreationanddestructionresponsestooil pricechanges,JournalofMonetaryEconomics48:465–512. Diebold, F. X., Lee, J.-H. and Weinbach, G. C. (1993). Regime switching with time-varying transitionprobabilities,Technicalreport. Fagan,G.,Henry,J.andMestre,R.(2005). Anarea-widemodelfortheeuroarea,Economic Modelling22(1):39–59. 29

Filardo, A. J. (1994). Business-Cycle Phases and Their Transitional Dynamics, Journal of Business&EconomicStatistics12(3):299–308. Fratzscher,M.,Schneider,D.andVanRobays,I.(2014). Oilprices,exchangeratesandasset prices,ECBWorkingPaper(1689). Frühwirth-Schnatter, S. (2004). Estimating marginal likelihoods for mixture and markov switching models using bridge sampling techniques, The Econometrics Journal 7(1): 143– 167. Galí,J.andGertler,M.(1999). Inflationdynamics: Astructuraleconometricanalysis,Journal ofMonetaryEconomics44(2):195–222. Hamilton,J.D.(1989). Anewapproachtotheeconomicanalysisofnonstationarytimeseries andthebusinesscycle,Econometrica57(2):357–384. Hamilton,J.D.(2003). Whatisanoilshock?,JournalofEconometrics113:363–398. Hamilton, J. D. and Herrera, A. M. (2004). Oil Shocks and Aggregate Macroeconomic Behavior: The Role of Monetary Policy: Comment, Journal of Money, Credit and Banking 36(2):265–86. Hannon,P.(2008). Bis: Noevidenceofsecond-roundeffectsfromhighoilprices. Herrera, A. M. and Pesavento, E. (2009). Oil price shocks, systematic monetary policy, and theGreatModeration,MacroeconomicDynamics13(1):107–137. Hofmann, B., Peersman, G. and Straub, R. (2012). Time variation in u.s. wage dynamics, JournalofMonetaryEconomics59(8):769–783. Hubrich,K.andTetlow,R.T.(2015). Financialstressandeconomicdynamics: Thetransmissionofcrises,JournalofMonetaryEconomics70(1):100–115. Hubrich, K., Waggoner, D. and Zha, T. (2016). Monetary policy and financial stability: Unconventional monetary policy, leverage and financial stress. Manuscript, Federal Reserve Board. 30

Kilian, L. (2008). Theeconomic effects ofenergy price shocks, Journal ofEconomic Literature46(4):871–909. Kilian,L.(2009). Notalloilpriceshocksarealike: Disentanglingdemandandsupplyshocks inthecrudeoilmarket,AmericanEconomicReview99(3):1053–69. Kilian, L. (2014). Oil price shocks: Causes and consequences, Annual Review of Resource Economics6(1):133–154. Kilian, L. and Lewis, L. T. (2011). Does the fed respond to oil price shocks, The Economic Journal121:1047–1072. Kilian, L. and Murphy, D. P. (2014). The role of inventories and speculative trading in the globalmarketforcrudeoil,JournalofAppliedEconometrics29(3):454–478. Kilian,L.andVega,C.(2011). Doenergypricesrespondtou.s.macroeconomicnews? atest of the hypothesis of predetermined energy prices, The Review of Economic and Statistics 93(2):660–671. Kim,C.-J.-J.(2004). Markov-switchingmodelswithendogenousexplanatoryvariables,JournalofEconometrics122(1):127–136. Kim, C.-J. and Nelson, C. R. (1999). Has The U.S. Economy Become More Stable? A Bayesian Approach Based On A Markov-Switching Model Of The Business Cycle, The ReviewofEconomicsandStatistics81(4):608–616. Kim,C.-J.,Piger,J.andStartz,R.(2008). EstimationofMarkovregime-switchingregression modelswithendogenousswitching,JournalofEconometrics143(2):263–273. Kiseok, L. and Ni, S. (2002). On the dynamic effects of oil price shocks: A study using industryleveldata,JournalofMonetaryEconomics49:823–852. Kormilitsina, A. (2011). Oil price shocks and the optimality of monetary policy, Review of EconomicDynamics14(1):199–223. Leamer, E. E. (1978). Specification Searches: Ad Hoc Inference With Non-Experimental Data,NewYork: JohnWileyandSons,Inc. 31

Leduc, S., Moran, K. and Vigfusson, R. J. (2016). Learning in the Oil Futures Markets: EvidenceandMacroeconomicImplications,InternationalFinanceDiscussionPapers1179, BoardofGovernorsoftheFederalReserveSystem(U.S.). Leduc,S.andSill,K.(2004). Aquantitativeanalysisofoil-priceshocks,systematicmonetary policy,andeconomicdownturns,JournalofMonetaryEconomics51:781–808. Lippi, F. and Nobili, A. (2012). Oil And The Macroeconomy: A Quantitative Structural Analysis,JournaloftheEuropeanEconomicAssociation10(5):1059–1083. Mohaddes, K. and Pesaran, M. H. (2016). Oil prices and the global economy: is it different thistimearound?,CambridgeWorkingPaperEconomics(1640). Nakov, A. and Pescatori, A. (2010). Oil and the Great Moderation, The Economic Journal 120(543):137–183. Natal, J.-M. (2012). Monetary policy response to oil price shocks, Journal of Money, Credit, andBanking44(1):53–101. Obstfeld,M.,Milesi-Ferretti,G.M.andArezki,R.(2016). Oilpricesandtheglobaleconomy: It’scomplicated,iMFdirectblogpostedonMarch24. Peersmann, G. and Van Robays, I. (2009). Oil and the euro area economy, Economic Policy pp.603–640. Rotemberg, J. J. and Woodford, M. (1996). Imperfect competition and the effects of energy priceincreasesoneconomicactivity,JournalofMoney,CreditandBanking28(4):549–577. Sims, C. A., Waggoner, D. F. and Zha, T. (2008). Methods for inference in large multipleequationmarkov-switchingmodels,JournalofEconometrics146(2):255–274. Sims,C.A.andZha,T.(2006).Werethereregimeswitchesinu.s.monetarypolicy?,American EconomicReview96(1):54–81. Stock,J.andWatson,M.(2016). Dynamicfactormodels,factor-augmentedvectorautoregressions,andstructuralvectorautoregressionsinmacroeconomics,inJ.B.TaylorandH.Uhlig (eds),HandbookofMacroeconomics,Vol.2,Elsevier,pp.415–525. 32

VanRobays,I.(2012). Macroeconomicuncertaintyandtheimpactofoilshocks,ECBWorkingPaper(1479). Woodford, M. (2007). The case for forecast targeting as a monetary policy strategy, Journal ofEconomicPerspectives21(4):3–24. 33

Figure 4: Impulse responses to oil price shock (one standard deviation shock), slashed blue: normalregime,solidred: adverseregime Output growth 0.5 0 −0.5 −1 0 5 10 15 20 months stniop egatnecrep Inflation 0.05 0 −0.05 −0.1 −0.15 0 5 10 15 20 months stniop egatnecrep Oil price changes 2 0 −2 −4 −6 −8 0 5 10 15 20 months stniop egatnecrep Inflation Expect. (5Y5YBEIR) 0.04 0.02 0 −0.02 −0.04 −0.06 0 5 10 15 20 months stniop egatnecrep EURIBOR 0.08 0.06 0.04 0.02 0 0 5 10 15 20 months stniop egatnecrep NNNNNooooorrrrrmmmmmaaaaalllll AAAAAdddddvvvvveeeeerrrrrssssseeeee 34

Figure5: Time-varyingprobabilityforbeinginanormalregime(grey-shadedarea)andconditionalprobabilityofstayinginthatregime(blackline)–oilpriceineuro 35

Figure 6: Impulse responses to oil price shock in model with nominal negotiated wages (one standarddeviationshock),slashedblue: normalregime,solidred: adverseregime 36

Figure7: Impulseresponsestooilpriceshockinmodelwithlong-termrealinterestrate(one standarddeviationshock),slashedblue: normalregime,solidred: adverseregime 37

Figure 8: Impulse responses to oil price shock for 1974-2015 sample, slashed blue: normal regime,solidred: adverseregime 0.08 0.06 0.04 0.02 0 −0.02 −0.04 −0.06 0 5 10 15 20 months stniop egatnecrep Output growth 0 −0.02 −0.04 −0.06 −0.08 −0.1 −0.12 0 5 10 15 20 months stniop egatnecrep Inflation −8 −10 −12 −14 −16 −18 −20 −22 0 5 10 15 20 months stniop egatnecrep Oil price changes 0.015 0.01 0.005 0 −0.005 −0.01 0 5 10 15 20 months stniop egatnecrep USD/EUR exchange rate 0.02 0 −0.02 −0.04 −0.06 −0.08 −0.1 0 5 10 15 20 months stniop egatnecrep Wage changes 0 −0.02 −0.04 −0.06 −0.08 −0.1 0 5 10 15 20 months stniop egatnecrep EURIBOR 38

Figure9: EvolutionofBrentcrudeoilprices(indifferentcurrencies) Source: Bloomberg 39