dealing with dealers: cds comovements in europe · sergio mayordomo was at the spanish securities...
TRANSCRIPT
DealingwithDealers:SovereignCDSComovementsinEurope*
MiguelAntón
IESEBusinessSchool
SergioMayordomo+UniversidaddeNavarra
MaríaRodríguez‐MorenoUniversidaddeNavarra
FirstDraft:November2013
*Antón:DepartmentofFinance,IESEBusinessSchool,Av.Pearson21,08034Barcelona,Spain.Email:[email protected]. Mayordomo and Rodríguez‐Moreno: Department of Economics and BusinessAdministration, University of Navarra, Edificio Amigos, 31009 Pamplona, Spain. Email: smayor‐[email protected] and [email protected] . This paper was partially drafted and circulated whileSergioMayordomowasattheSpanishSecuritiesandExchangeCommissionunderthetitle“IntradaycreditriskspilloversintheEuropeansovereignCDSmarket”.WeareverygratefulforcommentsfromAnaBabus,XavierFreixas,MireiaGiné,AntonioMoreno,FulvioPegoraro,TanoSantos,NealStoughton,CarlesVergara,XavierVives,Pierre‐OlivierWeill,HaoxiangZhu,aswellasconferenceparticipantsatthe 7th Financial Risks International Forum in Paris 2014, the Arne Ryde Workshop in FinancialEconomics in Lund 2014, and seminar participants at IESE Lunchtime workshop. Miguel AntónacknowledgesthefinancialsupportoftheEuropeanCommission,MarieCurieCIG(GAno.303990),andthefinancialsupportoftheSpanishMinistryofEconomyandCompetitiveness(Projectref:ECO2011‐29533)atPublic‐PrivateSectorResearchCenter‐ IESEBusinessSchool,UniversityofNavarra,Spain.Sergio Mayordomo acknowledges financial support from PIUNA (University of Navarra) and theSpanishMinistryofEconomyandCompetitiveness(grantECO2012‐32554).
+Correspondingauthor.
1
DealingwithDealers:SovereignCDSComovementsinEurope
ABSTRACT
AsimplemeasureofcommonalityinthequotesthatdealersgiveforSovereignEuro‐
pean CDS is a powerful predictor of cross‐sectional variation in CDS return
correlation,controllingforliquidityanddefaultrisks,andothercountry‐paircharac‐
teristicsandmacrovariables.Weshowthattheseresultsareconsistentwithanon‐
fundamental price pressuremechanism in two different ways. First, the predicting
effectofthecommonalityisstrongerfordealersthatexperiencesellingpressure,and
secondly, we show that the effect comes primarily from uninformed dealers. An
instrumentalvariableanalysisandtheuseof the“CDSnaked”ban fromtheGerman
BaFin as exogenous shock confirm that our findings reflect indeed a causal relation
betweencommonalityinquotesandCDSreturncomovement.
JELClassification:G12,G14.
Keywords:SovereignCDS,comovements,commonalities,dealers.
2
I. Introduction
CreditDefaultSwap(CDS)spreadshavebeenwidelyusedasameasureofcreditrisk,
becausetheyprovideinsuranceagainstthedefaultofanunderlyingsecurity.Regula‐
torsandacademicscontinuetousethemasakeytooltoassessthecreditworthiness
ofabondoranentity.Anincreasingbodyofliteraturehasshown,however,thatCDS
spreads not only convey information about credit risk, but also about liquidity and
counterparty risk.1Much less attention has been given, however, to the sources of
comovement in CDS spreads. Understanding the drivers of these correlations is
central to regulatory and policy work, as they are at the heart of systemic risk by
definition–strongerinterlinksbetweendifferententitiesrevealahighersystemicrisk.
Ifthisisimportantamongcorporatecontracts,itiscrucialwithsovereignCDS,espe‐
ciallyinEurope,asacrediteventinonecountrycanhaveastrongimpactinthewhole
Euroarea.
WefocusonconnectingcountriesthroughtheCDSdealerstheyhave incom‐
mon.Specifically,weshowthatasimplemeasureofcommonalityinthequotesthat
dealers give for Sovereign European CDS is a powerful predictor of cross‐sectional
variation in CDS return correlation, controlling for liquidity and default risks, and
othercountry‐paircharacteristicsandmacrovariables.
Theanalysisalsoexplorescross‐sectionalvariationinthestrengthoftheeffect.
In particular we show that the commonality in the quotes given by dealers has a
stronger effect on subsequent correlation when the common dealers have excess
inventoryriskandfacesellingpressure.Dealersaimingtosellprotectionforthetwo
countriesatthesametimeleadstoa largercomovement,duetotheeffectof forced
sellingat firesaleprices.Wealsofindthatcommonality inquotesfromdealersthat
are uninformed on the CDS prices leads to stronger comovements, given that they
tendtoreviewpricessimilarlyforthetwocountriesinthepairwhentheylackinfor‐
mation.
1SeeLongstaff,Pan,PedersenandSingleton(2005)forliquidityriskinCDS,andArora,Gandhi,andLongstaff(2012)forcounterpartyrisk.
3
CDSmarketdealersplayacrucialroleinprovidingliquiditytothemarketby
disseminatingbidsandofferstopotentialclients,seekingtotradecreditprotection.2
AlthoughafewpapersanalyzetheeffectoftheliquidityprovisiononCDSprices,not
muchisknownabouttheeffectofdealers’liquidityprovisionsonCDSpricecomove‐
ments.Inthispaper,weoffernewinsightsonthatissuebyexploitingavailabledata
providedbyCMA that consistsof intradayquotes for11EuropeanMonetaryUnion
countries.Thereason for focusing in theEMUis threefold.First, the levelsofconta‐
gionamong these countriesduring the currentEuropeansovereigndebt crisishave
beenverystrongandhavepersistedforalongperiodoftime.Second,theactivityin
thesovereignCDShasincreasedsignificantly.Thegrossnotionalamountoutstanding
bytheendof2008was$405billion,whilebytheendofoursample(October2011)
the gross notional amount outstanding for the 11 European countries was $1,047
billion.Infact,accordingtodataprovidedbyDTCC,France,Italy,andGermanywhere
thetop3referenceentitiesintermsofthenetnotionalamountoutstandingbySep‐
tember 2011, including sovereign and corporate references, being Spain the 5th
referenceentity.Belgium,AustriaandPortugalwereinthe11th,12th,and14thplace,
respectively. Third, all the CDS have similar characteristics in terms of currency,
restructuringclause,andtiming.
CDSdatavendorsemploytheirmethodologiestoofferdailyquotesthatareob‐
tainedaftercombiningthequotesreceivedbydifferentdealers.Wetestwhetherthe
commonquotes reportedby thesamedealer foragivenpairof countriesaffect the
correlation of CDS spreads, and find that they do. The effect of the commonality in
quotesissignificantatanystandardsignificancelevelandhasaverystrongforecast‐
ing power on the future comovements among sovereign CDS spreads. In fact, the
economicimpactofthecommonalityvariableisstrongerthantheoneattributableto
theremainingexplanatoryvariables,includingthetraditionalfundamentalvariables.
2The importance of dealers’ activity in the CDS market is remarkable. At the end of 2011, dealersaccounted for 58% of notional amounts outstanding, and 64% of gross market values in the CDSmarket.AdditionalevidenceisprovidedbyRobertPickel,CEOofISDA,inhistestimonytoCongressinMarch10,2009,statingthat86%oftheDepositoryTrust&ClearingCorporation(DTCC)tradesweredealer‐to‐dealertrades.Finally,TangandYan(2010)alsosustainthatmostCDScontractsaretradedthroughdealerswhoeithertradewithotherdealersdirectlyortradethroughaninterdealerbroker.
4
The relationshipbetween commonality inquotes andCDS return correlation
could also go the otherway, so endogeneity is a key concern. For example, dealers
couldchoosetogivemorequotestocountrieswhoseCDSpricesaremorecorrelated.
To address this concernwe provide two pieces of evidence.We first implement an
instrumentalvariableapproach,andusetheleverageofbrokersanddealersasproxy
for the commonality inquotes givenbydealers. Themechanism throughwhich the
excess correlation is caused is the selling of CDS at fire sale prices, because some
dealersreachtheirrisk‐bearingcapacity.Inthissense,brokersanddealers’leverage
proxies well for the risk bearing appetite of dealers, as stated recently by Adrian,
Etula,andMuir(2013).Also,astandardtestshowsthat the instrument isvalid.Sec‐
ondly, we use an unexpected ban of naked CDS by the German Securities and
ExchangeCommission(BaFin)onMay18,2010.Thisunexpectedeventresultedinthe
willingnessofdealerstoreducetheirinventories,whichinturnresultsinexogenous
variation in commonality in quotes. These two pieces of evidence confirm that our
findingcomesindeedfromacausalrelationbetweencommonalityinquotesandCDS
returncomovement.
Finallywe show that our finding is robust to the exclusion of Greece in the
analysis,totheuseofpre‐crisisandcrisisperiods,totheuseoffilteredCDSreturnsto
amarketmodel,andtodifferentdefinitionsofcommonalityinquotes.Therestofthe
paperisorganizedasfollows.InSection2,wereviewtheliterature.InSection3,we
describeourmethodologyanddatasources.Section4presentsourresults.InSection
5wepresentourconclusions.
II. LiteratureReview
Ourwork links topapersstudying thecomovementsorcorrelationsbetween
the CDS spreads of the EMU countries during the current European sovereign debt
crisis. Independently of the econometricmethodology employed tomeasure conta‐
gion, thesepapers document an increasing trend in the comovements of credit risk
indicators in the EMU countries since the beginning of the crisis (see Andenmatten
andBrill (2011), Zhang, Schwaab, andLucas (2011),Alter and Schüler (2012),Kal‐
5
baskaandGatkowski(2012),GündüzandKaya(2013),ManasseandZavalloni(2013),
amongothers). Inthisprocessofcontagion,peripheralcountrieshavebecomemore
vulnerable to the euro zone contagion and have exhibited stronger comovements
betweenthem.
InspiteofthegrowingliteratureonthedriversofcreditriskandCDSreturns
in European countries in the context of the European sovereign debt crisis; little is
knownaboutthedeterminantsoftheobservedcomovementsandinterlinkagesofthe
countries’levelsofcreditriskandinCDSreturns.Thepotentialsetofdriverscouldbe
relativelysimilartothedriversthatexplainandpredictcorporateCDSprices:default,
liquidity,riskpremium,andcounterpartyriskfactors(seeAnderson(2012)orPuand
Zhao(2010)).3Forexample,theroleofthedefaultfactorsisdocumentedbyManasse
and Zavalloni (2013), who find that the countrymacro fundamentals during crisis
periodexplain80%ofthevulnerabilityofagivencountrytocontagion.Besidesmacro
fundamentals, Alter and Schüler (2012) document that financial sector shocks also
affectsovereignCDSspreadsintheshort‐run.Inthesameline,DieckmannandPlank
(2012)showthatthehighcorrelationsobservedsincethebeginningofthefinancial
crisisareexplainedbythestateofacountry’sdomesticfinancialsystem(itaffectsthe
expectationsaboutbailoutsandtheburdenofgovernmentintervention).
Wesharesomeoftheobjectivespursuedbythesepreviouspapersbyanalyz‐
ingtheeffectofcountryspecificandglobalvariablesreferringtodefaultrisk,liquidity,
and risk appetite on the correlations among EMU sovereign CDS spread changes.
Given that previous literature has left a significant part of these correlations unex‐
plained,ouraim is tobuildonprevious literatureand testwhether theCDSmarket
structure and their dealers’ activity helps improve the explanatory power on such
correlations.TheimportanceofdealersintheCDSmarketdescribedabovehighlights
theroleofthedealers’activityintheCDSpricescomovements.
3ThesefactorshavealsobeenfoundtobesignificantdeterminantsoftheEMUsovereignbondandCDSspreads by Geyer, Kossmeier, and Pichler (2004), Beber, Brandt and Kavajecz (2009), Mayordomo,Peña, and Schwartz (2012), Favero, Pagano, and Von Thadden (2010), Bernoth, von Hagen, andSchuknecht(2012),orBadaoui,Cathcart,andEl‐Jahel(2013),amongothers.
6
TheinabilityoffundamentalstofullyexplainthesecomovementslinkstoBar‐
beris and Schleifer’s (2003) theory. They develop a theory where assets commove
beyond fundamentals, because of market frictions or noise‐trader sentiment. They
provideevidenceinBarberis,Schleifer,andWurgler(2005)insupportofthealterna‐
tivefriction‐basedview,usingdataoninclusionsintotheS&P500.
Ouranalysis focusesontheeffectof twospecificmarket frictionsontheCDS
prices comovements. On the one hand,we consider the effect of asymmetric infor‐
mationfromthepriorthatnotalldealersareequallyinformedateverydate.Onthe
other hand,we consider the effect of the inventory risk that could cause that some
dealerssuffersellingpressureintheCDScontactsofagivensetofcountries.
Market liquidity can be provided by both informed and uninformed agents.
Empirical literature has documented that market players in the stock market are
asymmetrically informed about asset values and that financial market transactions
may conveyprivate informationabout fundamental values (seeHasbrouck (1991a),
Hasbrouck(1991b),orDufourandEngle (2000),amongothers).Thephenomenaof
asymmetric informationanditseffectontheassetpricesshouldbeevenmoreobvi‐
ousinOTCmarkets.DufourandNguyen(2012)analyzetheinformationalcontentof
tradingactivityintheeuroareasovereignbondsandfindastrongevidenceofinfor‐
mationasymmetry inthismarket thatexplainsthecross‐sectionalvariationofbond
yields.
Recent studies referred to the CDS market such as Acharya and Johnson
(2007), Berndt and Ostrovnaya (2008), and Angelopoulos and Giamouridis (2012)
document the existence of informed trading in the CDSmarket on the basis of the
equitymarketasabenchmarkforpublicinformation.Thebetterinformeddealersin
theCDSmarketcouldbemajorbanksthatmayhaveaccesstonon‐publicinformation
on CDS obligors, through their lending and investment banking activities, and can
potentiallytradeonthisinformation(AcharyaandJohnson(2007)).
The informationheterogeneity affects to theCDSprices (Gündüz,Nasev, and
Trapp (2013), and Tang and Yan (2010)) and their liquidity (Qiu and Yu (2012)).
7
Gündüz, Nasev, and Trapp (2013) use a proprietary dataset of CDS transactions to
showthatCDSpremiacontainasignificantnon‐defaultrelatedcomponentwhichCDS
traders charge to protect themselves against informational and real frictions. Tang
andYan(2010)showthatCDSnetbuyinginteresthasinformationcontentforfuture
corporateCDSpricechanges.QiuandYu(2012)studythedeterminationofliquidity
provision in the single‐name CDS market as measured by the number of distinct
dealersprovidingquotesandsupporttheexistenceofendogenousliquidityprovision
by informed financial institutions. They conclude that the degree of information
heterogeneityplaysanimportantroleinhowliquidityisreflectedintheCDSpremi‐
um.TheymeasureinformationflowfromtheCDSmarkettothestockmarket.Inour
approach, however, we only use information exclusively from the CDSmarket, be‐
causethereisalackof“sovereignequityprices”.
Another friction affecting financial market is the existence of inventory risk
thatcouldderive insellingpressure.Theprobabilitythat intermediariesfaceselling
pressure increases remarkablyduring liquidity crisis as theongoing financial crisis.
AsexploredinPedersen(2009),inAugust2007therewasasignificantliquidityevent
in which some quants were forced to unwind and others also reduced positions
leadingtoastrongpricepressureandliquidityspirals,inwhichsellingleadstomore
sellingandeven forcedselling.AsstatedbyMitchell,Pedersen,andPulvino (2007),
thecontinuedsellingpressurefromtheproprietarytradingdesksiscausedbyinter‐
nalcapitalconstraintsthatwerelikelyimposedasaresultofthelargelosses.Inthis
vein, Brunnermeier and Pedersen (2009) provide a model that links a security’s
market liquidity and traders’ funding liquidity. When dealers hit their capital con‐
straintsthentheyareforcedtoreducetheirpositions;thisleadstoanexcesssupply
andsothepricedeclines,whichinturnleadstohighermargins,furthertighteningthe
dealers’fundingconstraint,andsoon.
Shachar (2013)documents a significant effectof inventory riskon corporate
CDS prices. Using a proprietary dataset on transactions of CDS contracts on North
Americanfinancialfirms,sheexaminestheroleofliquidityprovisionbydealersinthe
8
CDS market, and finds that order imbalances of end‐users cause significant price
impact.
Shockstothefundingconstraintofthedealersectoraffectallsecuritiestraded
byagivendealerandso,theycouldpropitiatecross‐stockpricepressure.HoandStoll
(1983)showthatwhenthedealertradesmorethanonestock,shenotonlylowersthe
bidandaskquotesinthestockherecentlyboughttoreduceinventorylevelsbutalso
adjusts quotes in other stocks to reduce his total inventory risk. Themagnitude of
quote adjustments in the other stocks depends on underlying correlations. In the
same vein, Andrade, Chang, and Seasholes (2008) use data from the Taiwan Stock
Exchangetodocument thatan imbalance inonestockalsoaffects thepriceofother
stocks. They show that this cross‐stock price pressure is higher among stockswith
morecorrelatedcashflowsthanamongstockswithlesscorrelatedcashflows.
The effect of liquidity shocks affecting financially constrained intermediaries
on the comovementsof fixed income instruments’ returnshasbeendocumentedby
Acharya,Schaefer,andZhang(2008).Theydocumentanincreaseofliquidityriskfor
market‐makersaroundtheepisodeofGMandForddowngradesinMay2005duetoa
wide‐spreadsell‐offintheircorporatebonds.Thisisshownbyasignificantimbalance
inthedealerquotestowardssales.Theyshowthatthisimbalancetowardssalesinthe
volumeandfrequencyofquotesonGMandFordbondsexplainsasignificantportion
of the excess comovement in the bond and CDS prices of all industries, not just in
those of auto firms. The effectwas stronger for firmswith a sub‐investment grade
comparedtoinvestment‐gradefirms,consistentwiththehypothesisthatintermediar‐
ieswithdrew liquiditymore sharply frombondswithgreater inventory risk.4In the
same vein, Micu, Remolona, andWooldridge (2006) show that the price impact of
rating downgrades on CDS not only from their information content but also from
sellingpressurebyrestrictedinvestors.Thepricepressurefromrestrictedinvestors
4The stronger effect on sub‐investment grade firms suggests that regulation could also cause somesellingpressureondowngradedcountries.Ellul, Jotikasthira,andLundblad(2011)paperinvestigatesfire sales of downgraded corporate bonds induced by regulatory constraints imposed on insurancecompanies. Thus, insurance companies that are relatively more constrained by regulation are, onaverage,more likely toselldowngradedbondthatgeneratessellingpressurearound thedowngradewhichcausespricepressuressuchthatpricescoulddeviatefromfundamentalvalues.
9
shouldhaveatemporaryimpactgiventhatinformedtraderwoulddrivepricebackto
the fundamental value. Nevertheless, this mean reversal could not occur in poor
liquidityconditions.
Asdiscussedabove,previousstudieshaveemployedtransactionpricestodoc‐
ument theeffectofmarket frictionson theCDSprices.Weprovideevidenceon the
effectofthesefrictionsonthecomovementsinCDSreturns.CDSpricesareconstruct‐
edthroughacompilationofquotescollectedfromtheparticipantsintheCDSmarket.
Nevertheless,thelackofinformationonquotesatthedealerlevelleavesagapinthe
literaturedocumentingtheroleofdealers’activityonCDScontracts.Theaccesstothe
quotesreportedforeachspecificdealerenablesustofillthisgapintheliteratureand
providenewevidence about the role of non‐fundamental factors in explaining such
comovementsinperiodsinwhichmarketfrictionsintheCDSemerge.Inafirststage,
we finda significant roleof thedealer commonalities inquotesacross countrieson
thecomovementsofthechangesintheirCDSspreads.Inasecondstage,wedocument
a significant effect of the information asymmetries and the inventory risk, which
derives insellingpressure, facedbytheCDSmarketdealersonthecomovementsof
CDSspreads.Theseresultsarerobusttoawidevarietyofalternativespecifications.
III. DataandMethodology
A. DataandSample
Intraday CDS quotes disaggregated at the contributor or dealer level come
froma dataset provided by CMA, for 11European countries (Austria, Belgium, Fin‐
land, France,Germany,Greece,Netherlands, Ireland, Italy,Portugal, and Spain), and
spanningfromJanuary2008toOctober2011.TheintradayCDSdealerquotes(both
executableandindicative)comefromover‐the‐countercommunicationbetweenCDS
dealers and buy‐side institutions, including hedge funds and investment banks’
proprietary tradingdesks.5ThedailydatareportedbyCMAcomes fromthese intra‐
5As explained in Qiu and Yu (2012), the process of trading in the CDSmarket usually begins withclients receiving indicative quotes from dealers through information providers such as Bloomberg.They then initiate a request‐for‐quote with a single dealer or multiple dealers by phone, email, orthroughanelectronictradingplatform.Dealerscanrespondwithcompetitivebindingquotesthatoften
10
dayquotes.CMAcollects thebuy‐sidedata foreverycontractandaggregates it toa
daily frequency (requiring at least three distinct dealers). CMA implements then
severalprocedurestomitigatetheinfluenceofoutliers.
CDSquotesemployedinthisstudyare5‐yearmaturitycontracts(themostliq‐
uidone)denominated inUSDollars.For thoseobservations forwhichweonlyhave
informationontheCDSup‐frontpricesbutnotfortheCDSspreads,wecalculatethe
spread following the ISDA CDS Standard Model to convert upfront payments into
spreads.6Toguaranteeaminimumlevelofsynchronicity,weexcludequotesoutside
themainworking hours (7am to 8pmGMT+1) and quotes given on Saturdays and
Sundays.7Information related to control variables comes from other sources, ex‐
plainedinsubsequentsubsections.
TableIreportsthesummarystatisticsofthefinalsampleofCDSquotesandthe
shareofquotesbydealers.PanelAdisaggregatesatcountrylevelthetotalnumberof
quotesanddealers,aswellasthedailyaverage.Wehavemorethanhalfamillionof
quotesforeveryperipheralcountryoverthewholesampleperiod(i.e.,morethan572
quotes on daily average). The total number of quotes in the core countries ranges
from469,751(France)to331,887(Finland).Regardingthenumberofdealersgiving
quotes toacertaincountrywedonotobservesizeabledifferencesacrosscountries.
We observe that there are around 90 dealers providing quotes to the 11 European
countries. Nevertheless, Panel B shows that not all dealers are equally active. Con‐
cretely,the10mostactivedealersprovide45.9%ofthetotalnumberofquotesinour
sampleandthe30mostactivedealerscoverthe90.8%ofthetotalnumberofquotes.
<InsertTableIhere>
ComovementsinthesovereignCDSarecomputedasthemonthlycorrelationof
dailysovereignCDSreturns forcountries iand jinmonth, , , forthesampleof11
result in actual transactions. They can also respondwith non‐competitive quoteswithwide bid‐askspreadsorchoosenottoprovidequotesiftheydonotwishtotrade.6http://www.cdsmodel.com/7Quotesoutsidethesehours,andinweekends,arescarce.Infact,theyrepresent2.25%ofallobserva‐tions.Duetothelowpercentageofexcludedquotes,wefindsimilarresultswhenweincludetheminouranalysis.
11
Europeancountries(55differentcountry‐pairs)andfortheperiodofJanuary2008to
October2011.Figure1showsthemedianof , ,jointlywiththe5thand95thpercen‐
tiles.FromthebeginningofthesampletotheLehmanBrotherscollapseweobservea
widedispersionacrosscorrelationrangingfrom‐0.45to0.96.SinceSeptember2008,
themedianofthecorrelationsfluctuatessteadilybetween0.5and0.9.The5thand95th
bands show a small dispersion in March 2009, due to the implementation of the
economic stimulus package in the US, and in May 2010, due to the Greek bailout
request.However,weobserveagreaterdispersionsinceMay2010,wherethereisa
sizeabledecreaseofcorrelationsinthe5thpercentile.Thiscomesasaconsequenceof
thedisproportionatelylargeincreaseoftheGreekandotherperipheralCDSpremiain
comparisontothecorecountriesCDSpremia.TableIIreportsthedescriptivestatis‐
ticsofthemonthlycorrelationofdailysovereignCDSreturnsforcountries iand jin
month jointlywithalternativespecificationsofthiscorrelation.
<InsertFigure1andTableIIhere>
B. MeasuringCommonalityinQuotes
i.Commonquotesreportedbythesamedealerforagivenpairofcountries.
Ourmainvariableofinterestmeasurestheamountofcommonquotesgivenby
different dealers to each pair of countries eachmonth.We label this variableCom‐
monalityinQuotes,anddefineitas:
∑ min ,
∈ 0,0.5 (1)
where and are the number of quotes given to country a and country b
respectivelybydealerdinagivenmonthtwhile and arethetotalnumberof
quotesgivenbyalldealerstocountriesaandbat timet, respectively.Thisvariable
capturestheconnectivityofcountriesaandbduetocommonalityinquotesgivenby
CDSdealers. Ifadealergives1quotetoFrance,and10quotestoSpain,wesaythat
FranceandSpainonlyshare“1commonquote”fromthatdealer,theminimumofthe
12
two.Wethenaggregatethatvalueforalldealersgivingquotestobothcountries,and
normalizeitdividingitbythesumoftotalquotesgiventoFranceandSpain.
Figure 2 shows themedian of the variable commonality in quotes, together
withits5thand95thpercentilebands.Byconstruction,thevariablerangesfrom0(no
commonalityinquotes)to0.5(strongcommonalityinquotes).Weobservethatfrom
the beginning of the sample to May 2009 the median performs an upward trend
increasing from0.22 to 0.47while the 5th and95th percentiles tighten reaching the
tightest point in May 2009. There is a clear and significant time‐series, and cross‐
sectionalvariationinthisvariable.
<InsertFigure2here>
ii.Commonality inquotesandsellingpressurereportedbythesamedealer foragiven
pairofcountriesdependingonhersellingpressure
Wenow turn toamoredisaggregatedversionof the commonality inquotes.
Dealers giving low ask prices aremore willing to sell than dealers giving high ask
prices.Therearedifferentreasonswhyadealercouldgivealowaskprice,andthus
bewillingtosell.Ifthedealergivesquotesbasedoninformation,thenshecouldgive
lowaskprices foronecountryandhighaskprices foranothercountry. If,however,
thedealerisgivinglowaskpricesinbothcountries,shemightbefacingsomeinven‐
tory‐relatedproblem,andmightbewillingorforcedtosellpartofit(Shachar(2013)).
Tocapturetheeffectofthissellingpressure,wedecomposethevariableCom‐
monalityinQuotesintwovariables.CommonalityfromSellingPressureisdefinedasin
Equation (1),butweonlyuse thequoteswhoseaskprice is in the firstquartile for
countriesaandbrelativetothetotalnumberofquotesofdealerd.Alowaskpricein
the two countries at the same timewould indicate that the dealer is aiming to sell
protectionforthetwocountriesandso,itcouldleadtoalargercomovement,dueto
the effect of forced selling at fire sale prices (see Antón and Polk (2013), and Lou
(2013)).Thesecondvariablecapturesthecombinationsthatdonotcomefromselling
pressure,i.e.,wheretheaskpriceisnotinthefirstquartileinanyofthetwocountries,
andwelabelitasCommonalitynotfromSellingPressure.
13
iii.Commonalityinquotesandmarketinformation
To test the informationchannel inouranalysis,weestimate theeffectof the
commonquotes reportedbydealers to two countries dependingon their degreeof
informationontheCDSprice.
To define the dealers as informed/uninformed we employ the Gonzalo and
Granger’s (1995) model which is based on the following Vector Error Correction
Model (VECM) specification and is used to study the effectiveness of the different
dealersintermsofpricediscovery:
Δ Γ Δ (2)
where Equation (2) is a system of two equations constructed from a vector auto‐
regressive(VAR)andanerrorcorrectionterm(ECT).ThevectorXtincludesapairof
CDS quotes or prices of the same underlying country from a given dealer and the
averageCDSquoteoftheremainingdealersinthemarket.TheECTisdefinedbythe
product 1 tX where 1, , are estimated in an auxiliary cointegration
regression.TheseriesforthepairofCDSpricesincludedin 1tX mustbecointegrated
to develop this analysis and the cointegrating relation is defined by
)( 1,2321,11 tDEALERtDEALERt CDSCDSX which can be interpreted as the long‐
run equilibrium. The parameter vector ’ , contains the error correction
coefficientsmeasuringeachprice’sexpectedspeedineliminatingthepricedifference
anditisthebaseofthepricediscoverymetrics.Theparametervector i for 1, . . ,
withpindicatingthetotalnumberoflags,containsthecoefficientsoftheVARsystem
measuringtheeffectofthelaggedfirstdifferenceinthepairofCDSquotesonthefirst
difference of such quotes at time t.8Finally, denotes a white noise vector. The
percentagesofpricediscoveryoftheCDSquotei(where 1, 2)canbedefinedfrom
thefollowingmetrics , 1,2whicharebasedontheelementsofthevectorα’:
; (3)
8TheoptimalnumberoflagsisdeterminedbymeansoftheSchwarzinformationcriteria.
14
Thevectorα’containsthecoefficientsthatdeterminecontributiontopricedis‐
coverybytheindividualdealerandtheaveragedealer.Thus,giventhatGG1+GG2=1we
conclude that dealer 1 leads theprocess of price discoverywith respect to average
contributionsoftheremainingdealerswheneverdealer1pricediscoverymetricGG1
ishigherthan0.5.Thecloserto1(0)thehigher(lower)theabilityofthedealeris.We
estimatethepricediscoverymetricforeachcountry,month,anddealerusingallthe
possiblepairsofCDSspreads.Thepricediscoverymetricsareestimatedusingintra‐
dayinformationovertheperiodJanuary2008–October2011.Inordertoguaranteea
minimum of synchronicity in quotes we group the quotes in ranges of two hours.
Thus,fortherange7.00AMto9.00AM(GMT+1)theCDSpricetobeemployedinthe
pricediscoveryanalysisistheaveragepricewithinsuchrange.Weestimatetheprice
discoverymetricsonamonthlybasisusingrollingwindowsofintradaypricesinthe
lastquarter.That is,weuseamaximumofaround460observationsperestimation.
Weestimatethepricediscoverymetricforagivendealerinagivenmonthwhenthe
dealerhasreportedquotesinatleast22workingdaysinthelastquarter.9Themiss‐
ing observations are replaced by the last available intraday quote in order to
guarantee an adequate number of observations for the estimation. We use infor‐
mation fora totalof33dealers thataccording toPanelBofTable I representmore
than91%ofthetotalquotesintheCDSmarket.
Commonalityinquotesdependingoninformationisdefinedthenas:
∑ min ,
, , (4)
whereidenotesthelevelofinformationofthedealerdintheCDSpricesofcountries
aandb.Weconsiderthreepossibilitiesaboutthedegreeofinformationofdealerd:(i)
heisinformedaboutbothcountriesaandb(I),(ii)heisnotinformedaboutanyofthe
twocountries(NI), (iii)he is informedaboutonecountrybut isnotabouttheother
country(R).Thus,wehaveacommonalityvariableforeachoneofthethreeprevious
9We coincidewith Longstaff (2010) andArce, Mayordomo, andPeña (2013) on the affirmation thatprice‐discovery process in financial markets may be state‐dependent and perform a dynamic pricediscoveryanalysisonamonthlybasis.
15
possibilities.Thenotationissimilartotheoneemployedinthebaselinecommonality
variablebutnow and are thenumberofquotesgiventocountrya and
countryb respectively by dealerdwith a level of information iin a givenmonth t.
Regardingthedenominator, and arethetotalnumberofquotesgivenbyall
dealersinthecategoryofinformationitocountriesaandbattimet,respectively.
denotesthenumberofdealersineachofthethreecategoriesdenotingthedegreeof
informationthatwillbedeterminedonthebasisofthepricediscoverymethodology
fromGonzaloandGranger(1995).Thisvariablewillenableustoknowwhetherthe
comovementsareaffectedbytheactivityofthemoreinformeddealersorwhetherthe
non‐informeddealersarepushingthelevelsofcontagionwiththeirquotedprices.
C. ModelingCross‐SectionalVariationinSovereignCDSComovement
The following equation represents the panel estimate of monthly cross‐
sectional regressions forecasting the monthly correlation of daily sovereign CDS
returnsforcountriesiandjinmontht , forthesampleof11Europeancountries
(55differentcountry‐pairs)andfortheperiodofJanuary2008toOctober2011:
, ∗ ,∗ ∗ , ,
(5)
where ,∗ referstoaourmeasureofcommonalityinthequotesthatdealersgive
to both countries in the pair at month 1as defined in Section III.B and
, containsthesetofkcontrolsthatincludethedependentvariablelagged
onemonth, a group ofmacro controls, and a group of pair‐level controls. All these
controlsareexplainedinthenextsubsection.Theparameter referstothecountry‐
pairfixedeffectsthatareincludedintheestimation.Thestandarderrorsaredouble‐
clusteredatthecountry‐pairandmonthlevel.Inthreedifferentspecificationsweuse
thethreedifferentversionsofCC:thebaseline, thedisaggregated insellingpressure
facedbydealers,andthedisaggregatedinthedegreeofinformationofthedealers.
16
D. Controls
D.1 Globalvariables
TheuseofglobalfactorsismotivatedbyLongstaff,Pan,Pedersen,andSingle‐
ton’s(2012)studyofthenatureofsovereigncreditriskonthebasisofCDSspreads.
Their results showthat themajorityof sovereigncredit risk canbe linked toglobal
factors(asingleprincipalcomponentaccountsfor64%ofthevariationinsovereign
credit spreads).We focuson threevariables thataim toproxy for funding liquidity,
counterpartyrisk,andtheEuropeanCentralBank(ECB)bondpurchasesthroughthe
SecuritiesMarketProgram(SMP).
Financingcostsoffinancialinstitutions:Weusethespreadbetweentheuncol‐
lateralized interbank3‐monthsborrowingrate (Euribor)and theOvernight Interest
Swap(OIS).TheEuriborrepresentstheunsecuredinterestrateatwhichbanks lend
money tootherbankswhichsatisfycertaincreditworthinesscriteria. Euriborrates
comprisetheexpectedriskfreeoveraspecificpremium:thetermpremium,thecredit
riskpremiumofunsecuredtrading,andtheliquidityriskpremiuminlieuofovernight
(McAndrews,Sarkar,andWang (2008)).On theotherhand,OIS isequivalent to the
average of the overnight interest rates expecteduntilmaturity. It is almost riskless
andhence it isnotsubject topressuresassociatedwith thoserisks.Therefore,Euri‐
bor‐OIS spread over the same term quantifies the premium that banks pay when
borrowing funds for a pre‐determined period relative to the expected interest cost
fromarepeatedlyrollingoverfundingintheovernightmarket(Aït‐Sahalia,Andritzky,
Jobst, Nowak, and Tamirisa (2012)). An adverse liquidity shock would reduce the
amountofcapitalavailableforfinancialintermediarieswhichinturnlowertheability
oftheirtradingdesktoprovideliquidity.Asliquidityinthemarketworsens,trading
fallsandtheshort‐termcashinflowsoftheseinstitutionsdroptoo,sincemostoftheir
profitarisefrommarket‐makingrevenues.TheincreaseintheEuribor‐OISandso,in
the cost of capital for financially constrained intermediaries is the common factor
which influences both liquidity risk and correlation risk and due to this common
factorthecorrelationbetweenreturnsincreases.Thisrelationbetweenliquidityand
correlationriskisjointlyaddressedinAcharyaandSchaefer(2006).Forthisreason,
17
weexpectthathigherfinancingcostswouldgohandinhandwithstrongercomove‐
mentsinCDSspreads.
Counterparty risk: followingArora,Gandhi, andLongstaff (2012),weuse the
dealers’CDSspreadstoconstructthecounterpartyriskvariable,andwefollowArce,
Mayordomo,andPeña’s(2013)methodology.WeproxycounterpartyriskintheCDS
market through the firstprincipalcomponentobtained fromtheCDSspreadsof the
main14banksthatactasdealersinthatmarket.10Thehigherthecounterpartyrisk,
thelowertheconfidenceamonginstitutionalinvestorsintheCDSmarketandso,the
moredifficult to findacounterparty tobuyor sellprotection, the lower themarket
activity, and the higher should be the correlation risk. . Thus,we expect a positive
effectofthisvariableonthedependentvariable.
ECB bond purchases: Finally,we control by the purchases conducted by the
ECB in theopenmarket in thecontextof theSecuritiesMarketProgram(SMP) that
was launched inMay2010. Itwasdesignedtoconductoutright interventions in the
euroareapublicandprivatedebtsecuritiesmarketwiththeobjectiveof(a)address‐
ing the malfunctioning of securities markets; and (b) restoring an appropriate
monetarypolicy transmissionmechanism.Weconsider this interventionasaglobal
variableofinterestduetotheparticulardisclosureoftheprograminwhichtheECB
didnotdisclosethesetoftargetedsecurities/countries,thetimespanoftheprogram
or the amount to be spent. Preliminary studies suggest that SMP purchases had a
positivebutshort‐livedeffectonmarketfunctioningbyreducingliquiditypremiaand
lowering level and volatility of yields (Manganelli (2012)). Therefore, we expect a
commondecreaseinthestrainofthesovereigndebtcountriesandhenceanincrease,
atleasttemporary,inthecomovementsassociatedtothosepurchases.
Interactiontermswithperipheralcountriespairs:Totakeintoaccountpossi‐
ble asymmetries (i.e., differential increasing/decreasing effects in the comovements
10The14maindealersare:BankofAmerica,Barclays,BNPParibas,Citigroup,CreditSuisse,DeutscheBank,GoldmanSachs,HSBC,JPMorgan,MorganStanley,RoyalBankofScotland,SocietéGenerale,UBS,andWachovia/WellsFargo.Thesedealersarethemostactiveglobalderivativesdealersandareknownas the G14 (see, for instance, ISDA ResearchNotes (2010) on the Concentration of OTCDerivativesamongMajorDealers).
18
betweentheCDSpricesofcountriesindistress)weinteractthethreeglobalvariables
with a dummy variable that takes value 1 when the two countries in the pair are
peripheralcountries(Greece,Ireland,Italy,Portugal,andSpain).
D.2 Pair/CountrySpecificVariables
This set of variables accounts fordifferences/similaritiesbetween two coun‐
trieswhichpotentiallyaffecttothecomovementsintheCDSspreads.Wecontrolby
four groups of county specific variables: credit risk of financial institutions, risk
premium,CDSliquidity,andmacroeconomicvariables.Foreverypairofcountrieswe
introduce thesevariablesas themonthly correlationcomputedusingdailyobserva‐
tions.However,giventhelowerfrequencyofthemacroeconomicinformation,weuse
theabsolutevalueoftheirdifferencetoproxyforthedivergenceofthecountrieswith
respecttothecorrespondingvariables.
Credit risk of financial institutions: Acharya, Drechsler, and Schnabl (2011)
study the relationship between the financial and sovereign CDSs in the Eurozone
countriesfor2007‐2011andshowthattheannouncementoffinancialsectorbailouts
wasassociatedwithanimmediate,unprecedentedwideningofsovereignCDSspreads
andnarrowingofbankCDSspreads.However,post‐bailoutsthereemergedasignifi‐
cant comovement between bank CDS and sovereign CDS. Thus, the stronger the
relationshipbetween the financial sectorsof twogivencountries, theeasiest is that
theshockstofinancialinstitutionsinagivencountryaffectthesovereignsectorofthe
othercountry.Tocontrolforthiscomovementsweconsiderthecorrelationbetween
thelogreturnoftheCDSspreadsofthebankingsectorofthecorrespondingcountries.
Weexpectapositiveeffectforthiscontrolvariable.
Countryspecificriskpremium:Tocontrolforthesimilaritiesofthecountries
intermsoftheirriskpremium,weusethecorrelationbetweenthesquaredreturnsof
the country stock indexes. The country risk premiums have been found to have a
positiveeffectincreditriskbythepreviousliterature(DieckmannandPlank,2012).
We therefore expect that the higher the correlation between the stockmarkets log
returns,thehigherthecorrelationbetweenthesovereignCDSlogreturns.
19
CDS liquidity:Toproxy for theeffectof liquidity in thecomovementsweuse
the correlationbetween the sovereignCDS liquidity,proxiedby the relativebid‐ask
spread(i.e.,bid‐askspreadrelativetothemid‐spread).Previousliteraturehasdocu‐
mentedtheexistenceofaliquiditypremiuminsovereignCDSprices.Wethusexpect
that thehigher the correlationbetween the liquiditypremiumof two countries, the
largerwouldbethecorrelationintheprices.
Macro variables: We consider two macro fundamentals in our analysis: the
government debt and the government net deficit/surplus relative to GDP. These
variablesenableustoproxyforthestockofdebtinthecountriesandtheaccumulated
deficitandhavebeenfoundtohavesignificanteffectsonthesovereigncreditspreads
inBernoth,vonHagen,andSchuknecht(2012)andMayordomo,Peña,andSchwartz
(2012)amongothers.Tomeasurethesimilaritiesofagivenpairofcountriesinterms
of their macro fundamentals, we take the absolute difference in the value of the
previous two variables. We expect a negative effect of the absolute differences in
relative debt anddeficit,meaning thathigher differences are associatedwith lower
correlationsbetweenthesovereignCDSlogreturns.
IV. Results
A. ForecastingComovements
PanelAofTableIIIreportsourbaselineresults.Weemploythreealternative
specifications forwhichwe report the resultswithandwithout theCommonality in
Quotes variable to emphasize its power in forecasting comovements. Both columns
(1) and (2) contain the dependent variable lagged one month to control for any
possibleautocorrelationorpersistenceinthatvariable.Theresultssuggestamoder‐
ate persistence that discards any type of unit root in the dependent variable.
Regardingthenewvariableincludedincolumn(2),weobserveapositiveandsignifi‐
cant effect of the Commonality inQuotes variable: increases in the commonquotes
significantly increase the correlation between the sovereign CDS log returns. Apart
frombeingstatisticallysignificant,theforecastingpowerofthisvariableseemstobe
sizeable because theR‐squared increases by a 31% from24.2% to 31.6% after the
20
inclusion of this variable in the regression. In columns (3) and (4) we add global
controlvariablesandchecktheforecastingpowerofthecommonalitiesinquotes.
Anincreaseinthefinancingcostsoffinancialinstitutionsleadstoasignificant
increase in the comovements of CDS log returns in the pairs formed by peripheral
countries with respect to the core countries. As said before, and increase in this
variablewouldlimittheabilityofthefinancialintermediariestradingdesktoprovide
liquidityandthiseffectcouldbemoresevereintheCDSofperipheralcountries.
Asexpected,wefindanoverallpositiveandsignificanteffectoftheECBsover‐
eignbondpurchasesdue its effectiveness todiminish the levels of credit risk, their
volatility,andtheliquiditypremiumdocumentedbyManganelli(2012).Nevertheless,
we findanegative significantdifferentialeffect in thegroupofperipheral countries
thatwere targeted in theSMPwith respect to thenon‐peripheral countriesbut still
positive in aggregate terms (0.003‐0.002 = 0.001).11The lower magnitude of this
positiveeffectoftheECBSMPonthecomovementsamongperipheralcountriescould
beduetothepre‐existenthighlevelsbeforetheinterventionoftheECB.Thus,theECB
bondpurchasesmaketheperipheralcountriesmoresimilartothecorecountries in
termsofthevariationoftheircreditriskandweakenstheflight‐to‐qualityobserved
duringtheepisodeswiththehighestlevelsoffinancialstress.
Weobserve that the coefficient for theCommonalities inQuotes variable re‐
mains positive and strongly significant after controlling by global effects thatwere
simultaneouslyfacedbythetwocountriesineachpair.Theforecastingpowerofthis
variablewhenconsideringglobalvariablesdiminishesbutitisalsoremarkablegiven
thattheR‐squaredincreasesbymorethan21%from27.6%to33.5%.
Incolumns(5)and(6)weincludethepair/countryspecificvariables.Consist‐
entlywiththeexistenceofasignificantliquiditypremiuminCDSspreads;wefindthat
thestrongertherelationbetweentheliquidityoftheCDScontractsofagivenpairof
11The Governing Council of the ECB decided the 21st of February 2013 to publish the Eurosystem’sholdingsofsecuritiesacquiredundertheSecuritiesMarketsProgramwiththereportingthefollowingcountryby countrybreakdown: Italy (103billioneuros), Spain (44billion euros),Greece (34billioneuros),Portugal(23billioneuros)andIreland(14billioneuros).
21
countries,thestrongerthecomovementsintheirprices.Resultsalsoshowthatamong
themacro fundamentals, the similarity in thedegreeof the countries’ indebtedness
play a significant role in comovements. Thus, if two given countries exhibit a high
ratioofdebtrelativetoGDPthemarkettendstopushtheirCDSinthesamedirection.
ThecoefficientfortheCommonalitiesinQuotesvariableremainspositiveandstrong‐
ly significant after adding both global and local control variables. The forecasting
power of commonalities also diminishes when we consider pair/country specific
variablesbutitstillexhibitsasizeableincrease(18%from29.2%to34.5%).
Finally,weanalyzetheeconomicsignificanceofthevariablesaccordingtothe
baselineresultsobtainedincolumn(6).Theeconomicsignificanceofthestatistically
significant variables is obtained from the ratio thatmeasures the change in the de‐
pendent variable relative to its average value given a change in an explanatory
variable equal to its standard deviation, ceteris paribus. The coefficient with the
largest economic significance is the one for the commonality in quotes such that a
change equal to its standard deviation would lead to a change in the dependent
variable equal to 12.4% of its average value. The economic effect of the remaining
significantvariablesinabsoluteterms,byorderofrelevance,is7.91%forthelagged
dependent variable, 5.16% theoverall effect of theECBbondpurchasesbeing such
effectlowerforthecaseoftheperipheralcountries(4.54%),3.62%forthecorrelation
intheCDSliquidityofeachpairofcountries,1.1%fortheabsolutedifferenceofdebt
toGDP,and1.1%thedifferentialeffectofEuribor‐OISontheperipheralcountries.All
theseeffectsarewellbelowtheeffectofthecommonalityinquotes.
<InsertTableIIIhere>
TherevisioninNovember2009ofthemisleadingstatisticsoffiscaldeficitsby
theGreek authoritieswas the immediate triggerof the current European sovereign
debtcrisis.Weanalyzethepotentialeffectofthemostinfluentialeventinthesample
by excluding Greece from our analysis and splitting the sample in two sub‐periods
using as the break point such event. These results are reported in Table IV. Inde‐
pendentlyontheexclusionofGreeceandthesampleperiodemployedinouranalysis
we find that the commonality in quotes has a positive and significant effect on the
22
comovement of sovereign CDS spreads, being the economic impact of this variable
almost15%.Someresultsarealsoworthmentioning.Comparingcolumns(1)and(2)
ofthistable,weobservethattheexclusionofGreecefromouranalysisweakensthe
effectoftheECBbondpurchases,whichisnotsignificantat5%,suggestingthatthese
purchases were very effective in weakening the potential contagion originated in
Greece. In fact theprogram started to be effective inMay2010 coincidingwith the
rescueofGreeceandthefirstremarkableincreaseintheEMUsovereignCDSspreads.
TheeffectoftheEuribor‐OISvariableinteractedwiththeperipheralcountriesdummy
andthecorrelationintheCDSrelativebidaskspreadsarenotsignificantat5%after
excluding Greece from the estimation analysis. This result is consistentwith a pre‐
dominant role of Greece in the liquidity risk channel leading to higher levels of
contagion.
We find someworthmentioning differences between the pre‐crisis (column
(3)) and crisis (column (4)) periods. Although three of the country/pair specific
variableshadasignificanteffectontheCDScomovementsinthepre‐crisisperiod,no
significanteffectremainsduringthecrisisperiod.Theoppositeeffectisfoundforthe
globalvariables.Thesechanges in theroleofpairspecificandglobalvariablesafter
thebeginningofthecrisissuggeststhatthecountryfundamentalswerelessrelevant
toexplaincomovementsthanglobalriskormarketsentimentfactors.12
<InsertTableIVhere>
B. ComovementsandDealersInformation
Informedtraderscanbedefinedasthoseagentswhohavenon‐public,market‐
sensitiveinformation.Previousliteratureemploysthestockmarketasabenchmarkof
publicinformationtoanalysetheexistenceofinformedtraders.Nevertheless,ouraim
istoclassifytheagentsasbeingmoreorlessinformedthanthe“average”dealer(i.e.
theaverageCDSpricesobtainedthroughthecontributionofalldealers)onthebasis
ofpricediscoverymetricsthatareobtainedusingtheGonzaloandGranger’s(1995)
methodology.Giventhedatarequirementsforestimatingthedealer’spricediscovery 12ThevariablesrelatedtotheECBbondpurchasescannotbeemployedinthefirstpartofthesample(column(3)).
23
metrics, we can estimate suchmetrics in 67% of themonths in our sample. These
metricsareestimatedforatotalof33dealers.Theaveragepricediscoverymetricis
0.28anditindicatesthat,onaverage,eachindividualdealerislessinformedthanthe
averagedealer.TheaveragepricediscoverymetricbeforeNovember2009was0.17.
Afterthatdatetheaveragemetricincreasedto0.33.Althoughthedealers’information
quality could have improved due to the higher liquidity in the sovereign CDS con‐
tracts, dealers were on average less informed than the “average” dealer. The
thresholdof0.5,whichdetermineswhetheradealerisinformedornot,coincideswith
thethirdquartileofthedistributionofthepricediscoverymetrics(i.e.theprobability
thatagivendealerisinformedinbothcountriesinagivenmonthis25%).
The effects of the commonalties in quotes dependingon thedegree of infor‐
mationofthedealersarereportedinTableV.Column(1)reportstheresultsobtained
when we consider the commonality in quotes coming from dealers who are less
informed than the averagedealer (“uninformed”hereafter). Column (2) reports the
results forthecase inwhichtheexplanatoryvariablefor thecommonality inquotes
only considers information from dealers who aremore informed than the average
dealer(“informed”hereafter).Thecommonalityinquotesemployedincolumn(3)is
obtainedusinginformationfromdealerswhoareinformedaboutonecountrybutare
notabouttheothercountry.Weobservethatthethreecommonalityvariablesexhibit
asignificanteffectondependentvariable.Nevertheless,theexplanatorypowerofthe
commonality of uniformed dealers (34.2%) is higher than the one observed in col‐
umns(2)and(3)(31.2and31.5%,respectively).
Theindividualuseofthethreetypesofcommonalitiesexhibitsalwayssignifi‐
canteffectsonthecomovements.Infact,thecorrelationacrossthesevariablesishigh
(e.g., the correlation between the commonality in quotes for the “informed” and
“uninformed” dealers is 0.65; 0.44 between “informed” and “others”; and 0.74 be‐
tween “uninformed”and “others”) suggesting thatdealers tend to followa common
pattern in termsofprovidingquotes for specificpairsof countries.Nevertheless, to
compare the effect of the commonalities given by the dealers with different infor‐
mationweestimatetheireffectsjointlyincolumn(4).Wefindthattheonlysignificant
24
effect at 5 and 1% levels comes for the uninformed dealers while the role of the
informed dealers is not significant and the effect of the remaining dealers is only
significant at 10%.Weperforma test to checkwhether the coefficient for thenon‐
informeddealers issignificantlyhigher thanthecoefficient for the informeddealers
and find that it is at 5% significance level. According to Dornbusch, Park, and
Claessens(2000),intheabsenceofbetterinformationtothecontrary,investorsmay
believe that a financial crisis in one country could lead to similar crises in other
countries.This channelpresumes that investors are imperfectly informedabout the
fair price of credit risk in a given country and thusmake decisions on the basis of
otherreasons,whichmayormaynotreflectthetruecreditriskofthecountry,suchas
the belonging to the same monetary union. This theory based on the existence of
informational asymmetries could explain the stronger comovements when the less
informeddealersgivequotestoagivenpairofcountriesatthesametime.
Thecoefficientforthenon‐informeddealersissignificantlyhigherthantheco‐
efficient for thedealers thatare informedonacountrybutnotontheotheronlyat
10%significancelevel.Infact,thislastresultconfirmstheprevioustheoryaboutwhy
thecommonalitiesinthenon‐informeddealerstendtoincreasecomovements.Thus,
whendealershaveinformationonagivencountrybutnotontheother,theytendto
inferthepriceof thecountryonwhichtheir information ispoorerfromthecountry
onwhichtheyhavebetterinformation.13
<InsertTableVhere>
C. ComovementsandDealersSellingPressures
Inthissection,weperformatesttocheckwhetherthecoefficientforthedeal‐
erswithsellingpressureissignificantlylargerthantheonefortherestofthedealers.
We consider common low ask prices for a given pair of countries to define selling
pressure because dealerswill attempt to hedge their exposure to inventory risk by
adjustingtheaskspread.Theeffectsofthecommonaltiesinquotesdependingonthe
degree of dealers selling pressure are reported in TableVI. Column (1) reports the 13Similarresultsarefoundwhenweusedailyinformationinsteadofintradayinformationforestimat‐ingthepricediscoverymetrics.
25
resultsobtainedwhenweconsider the commonality inquotes coming fromdealers
whofacesellingpressurewhilecolumn(2)containstheresultsobtainedusingcom‐
monalities from dealers who do not face such pressure. We observe that the two
commonalityvariablesexhibitasignificanteffectat5%levelondependentvariable.
Tocomparetheeffectofthecommonalitiesgivenbythedealersfacingdiffer‐
entdegreesofsellingpressureweestimatetheirjointeffectsincolumn(3).Wefind
thatwhenconsideredjointlybotheffectsaresignificantat1%level.Weperformatest
tocheckwhetherthecoefficientforthedealersfacingsellingpressureisstatistically
significantlyhigherat5%levelthanthecoefficientforthedealersthatdonotsuffer
sellingpressureandfindthatitis.14Thisresultsupportsthetheorythatshockstothe
fundingconstraintofagivendealerwouldforcethisdealertoreduceherpositionsin
certain securities. This forced selling of securities would affect to their prices and
propitiatecross‐CDScontractspricepressurethatwouldbereflectedinthecomove‐
mentsoftheirprices.
<InsertTableVIhere>
D. DealingwithEndogeneity
InouranalysiswehaveregressedmonthlyCDScomovementsonthedealers’
quotescommonalitieslaggedonemonth.Nevertheless,endogeneitymaybeaconcern
here because it is plausible thatCDS comovements’ innovationsmay also affect the
dealers’ commonquotesat thesametime, throughsomebehaviorobserved insuch
correlations that may persist for a given number of months. To conclude that the
commonality in quotes is indeed causing CDS comovements to increase; we re‐
estimate the regressions reported in Equation (5) using two differentmethods: an
instrumentalvariableapproachandanexogenousshocktocommonalityinquotes.
Wefirstlyconsidertheuseofinstrumentalvariables.Weneedaninstrumental
variable that affects exclusively all the participants in the CDSmarket. The channel
throughwhichweexplainthiseffectofcommonalityincorrelationsisthevariationin 14Thesellingpressureisobtainedasthenumberofquotesbyeachdealerd inwhichtheaskpriceisthefirstquartileforcountriesaandbrelativetothetotalnumberofquotesofdealerd.Wehaveusedothercuttingpointssuchasthepercentile33%andresultsaresimilartoonesreportedinTableVI.
26
therisk‐bearingcapacityofthedealers(aswehaveseenintheprevioussection).We
proxy this risk‐bearing capacity with the log changes in the leverage of securities
brokers‐dealers in the US (half of the most active participants in the CDS market,
whicharecommonlyknownastheG14,areAmericanbanks).FollowingAdrian,Etula,
andMuir(2013),theleverageforbrokersanddealers(BD)isdefinedastheratioof
thesecuritiesbroker‐dealerstotalfinancialassetstonetworthdefinedasthediffer‐
ence between total financial assets minus total liabilities. Thus, our instrument
capturesthetime‐varyingbalancesheetcapacityoffinancialintermediariesandso,as
funding costs tighten, balance sheet capacity falls and intermediaries are forced to
deleverage by selling assets. In fact, Adrian, Etula, and Muir (2013) find that the
exposures to thebroker‐dealer leverage factorcanaloneexplain theaverageexcess
returns on awide variety of assets including Treasury bonds. Although our instru‐
ment isalsorelatedtofinancingcostsandglobal liquidityrisk, itmainlyreflectsthe
securitiesbroker‐dealersactivityandimbalances.Thereasonforfocusingindealersis
obviousgiventheirroleintheCDSmarket(i.e.bytheendof2011dealersaccounted
for64%ofgrossmarketvaluesintheCDSmarket). Additionally,Acharya,Schaefer,
and Zhang (2008) document that a significant imbalance in the dealers quotes to‐
wardssalesexplainasignificantportionoftheexcesscomovementsinbondandCDS
pricesaround theGMandForddowngrades.Theseauthorsshowthat “theeffectof
bondimbalanceonCDSwasstrongerforfirmswithasub‐investmentgradecompared
to investment‐grade firms, consistentwith the hypothesis that intermediarieswith‐
drewliquiditymoresharplyfrombondswithgreaterinventoryrisk”.Thus,aliquidity
shockwouldmakeliquiditytodryupmoreseverelyforsub‐investment‐gradesecuri‐
tiesthanforinvestment‐gradesecurities.
We run an instrumental variable regressionwith fixed effects and robust to
heteroskedasticity in which the common quotes are instrumented through the log
change inthe leverageofmarketparticipants.Resultsarereported incolumn(1)of
Table VII.We perform theKleibergen‐PaapRank LM statistic to checkwhether the
equation is identified that is, whether the excluded instrument (the securities bro‐
kers‐dealers leverage) is “relevant” (correlated with the endogenous regressor).
Accordingto thisunder‐identificationtestwereject thenullhypothesis(equation is
27
under‐identified) and so, the instrument is relevant and themodel is identified.We
alsoperformaweakidentificationtesttoanalyzewhetherthebrokers‐dealerslever‐
ageiscorrelatedwiththecommonquotesbutonlyweakly.Forthistestandgiventhat
theestimationisrobusttoheteroskedasticity,weusetheKleibergen‐PaapWaldRank
Fstatistic.Thestatisticobtainedincomparisonwiththecorrespondingcriticalvalues
enablesustorejectthehypothesesthattheequationisweaklyidentified.Asitcanbe
inferred from the significant coefficient for the dealers’ commonalitieswe conclude
thatthepotentialendogeneityofthesecommonalitiesdoesnotbiasourresults.
The previous instrumental regression is exactly identified and so,we cannot
checkthevalidityoftheinstrument.Weconsideranewinstrumentalvariableregres‐
sion using the interaction of the log change in broker‐dealer’s leverage and the
financing costs (Euribor‐OIS) as an additional instrument. This new instrument
captures thedealers’balancesheetcapacitywhen fundingcosts tighten.Theresults
arereportedincolumn(2)ofTableVIIandconfirmthesignificanteffectofthedeal‐
ers’commonalityinquotesonthesovereignCDSreturnscomovements.Inviewofthe
Kleibergen‐PaapRankLMstatisticandp‐valueweconcludethattheinstrumentsare
relevant relevant. The use of the second instrument also enables us to employ the
overindentification test based on Hansen J statistic and to confirm that the instru‐
mentsarevalidgiventhattheyareuncorrelatedwiththeerrorterm
Inoursecondattempttoexamineifthepreviousresultsaredrivenbyendoge‐
neity concerns, we consider the ban of naked transactions in CDS on Eurozone
sovereignbondsimposedbytheGermanSecuritiesandExchangeCommission(BaF‐
in)onMay18,2010asanexogenousshock.AccordingtotheEuropeanCommission‐
MEMO/11/713,thebuyeroftheCDSis“naked”ifhedoesnothaveanexposurewhich
he is seeking to hedge either to the sovereign debt itself, or to assets or liabilities
whose value is correlated to the sovereign debt.15This constraint is of special rele‐
vancefortheCDSmarketgiventhatalthoughthereisnodataonwhatportionofthe
market constitutes “naked” positions, someestimates put thenumber at around80
15SeeEuropeanCommission‐MEMO/11/713“RegulationonShortSellingandCreditDefaultSwaps‐Frequentlyaskedquestions”.
28
percentof themarket.16In fact, inmanycases thecontractsareused tohedgerisks
associatedwithacountrythatarenotdirectlyassociatedtoitsdebt.Additionally,the
use of this shock is motivated by the effect that it would have on the CDSmarket
participantsandactivitygiven that itwouldreduce the liquidityavailable for inves‐
tors lookinghedge theirpositions, aswell asmake itharder forparticipants toexit
existingtrades.AccordingtoDarrenFox,aregulatorlawyerwhoadviseshedgefunds
at Simmons&Simmons inLondon “theway it’sbeenannounced […] it’s sentmany
marketparticipantsintopanicmode”.17Insimilarterms,JohanKindermann,acapital
marketslawyeratSimmons&SimmonsinFrankfurtsaidinaninterview:“youcannot
imagine what broke lose here after BaFin’s announcement [..] this will lead to an
uproar in the markets tomorrow. Short‐sellers will now, even tonight, try to close
theirpositionsatmarketswheretheycanstilldoso‐‐iftheyfindanypossibilitiesleft
at all now”. For all these reasons the relation to the common quotes variables is
obvioussince thebancould limit thepotentialnumberofcounterpartiesandso the
frequencyofcommonquoteswoulddependona lowernumberof institutions.This
interventionalsocontributestofearinvestorsaboutfurtherregulatoryinterventions
inthemarket.
<InsertTableVIIhere>
We employ a difference‐in‐differences approach to analyzewhether pairs of
countrieswithdifferent levelsofcommonalities inquotessufferdifferentdegreesof
comovementswhen they face the unexpected shock of the BaFin’s ban onMay 18,
2010.Accordingly,pairsofcountrieswithhighlevelsofcommonalitiesinquotesare
definedasthetreatmentgroup,andpairswithlowercommonalitiesarethecontrolor
non‐treatedgroup.Werankallpairsofcountriesbasedontheiraveragecommonali‐
tiesinquotesfrom1Mayto12May,2010.ThedummyvariableofTop‐quintileisset
tounityiftheaveragelevelofcommonquotesisinthetop‐quintile(80‐percentileand
above),andzerofortheremainingpairsbelowthetop‐quintile(below80‐percentile).
Wethenobtaintheaveragecorrelation(dependentvariable)forthethreedaysafter
16Seehttp://www.reuters.com/article/2010/05/19/germany‐swaps‐idUSN196324520100519.17 See http://www.bloomberg.com/news/2010‐05‐18/germany‐to‐start‐temporary‐ban‐on‐naked‐short‐selling‐of‐euro‐bonds‐banks.html.
29
theestablishmentoftheban(19‐21May,2010)andthesameforthethreebanking
daysbeforetheban(13,14,and17May)toperformthediff‐in‐diffinwhichwewould
haveapre‐banobservationforthe55pairsandapost‐banobservationforthesame
pairs.18ThedummyvariableofPost‐BaFinBanissettounityifthedateis19‐21May,
2010,andzeroduringthethreebankingdaysbeforetheban.Athirddummyvariable
Post‐BaFinBanxTop‐quintile is the cross‐product of the previous two dummy varia‐
bles.
Thediff‐in‐diff regressionresultsare reported inTableVIII.Column(1)con‐
tainstheresultsobtainedwhenwedonotincludeotherexplanatoryvariablesbutthe
threepreviousdummieswhile column (2) contains the control variables at country
levelwithadailyfrequencygiventhattheeffectofthosevariablesthatdonotexhibit
anychangeinMay2010wouldbecollectedintheconstanttermwhiletheeffectofthe
global variables is collected in the BaFin dummy. The coefficient estimates of the
cross‐product dummy (Post‐BaFinBanxTop‐quintile) in the twodiff‐in‐diff specifica‐
tionsaresignificantlypositive,suggestingthatagivenpairofcountrieswithstronger
commonalitiesinquotesbeforethebanexhibitedhigherlevelsofcomovementsafter
theban.Notsurprisingly,wefindthattheBaFinbanincreasedthecomovementsby
itself.
<InsertTableVIIIhere>
E. ExtensionsandRobustnesstests
a. OthermeasuresofsovereignCDSreturncorrelation
Inthissectionwefirstanalyzetheeffectofthecommonalitiesinquotesonthe
correlations among sovereign CDS filtered returns. These correlations can be inter‐
preted as contagion according to Bekaert, Harvey, and Ng (2005) who define
contagionas“excesscorrelation,that is,correlationoverandabovewhatonewould
expect fromeconomic fundamentals.”LikeBekaert,Harvey,andNg (2005),we take
anassetpricingperspectivetomeasuringeconomicfundamentalsandidentifyconta‐
18Thereweremanyevents inMay2010relatedtotheEuropeansovereigndebtcrisis.To isolate theeffectofthebanwefocusintheabovementionedthreedaysaroundtheban.
30
gion by the correlation of an asset pricing model’s residuals. These residuals are
obtainedfromaregressioninwhichthedependentvariableistheCDSlogreturnofa
given country and the explanatory variable is a market variable. We employ two
differentspecificationsofthemarketvariabletoestimatethefilteredCDSreturnsfor
agivencountryi.Inthefirstspecification(column(2)ofTableIX),weusetheaverage
dailyCDSreturnofallcountriesinthesampleexceptcountryi.19Inthesecondspeci‐
fication(column(3)ofTableIX),weusethedailyreturnoftheiTraxxIndex,whichis
anindexthatconsistsofCDSonEuropeancorporations.Wealsocomputetheexcess
correlation in the sovereignCDS log‐returns as thedifferencebetween themonthly
correlationofthedailysovereignCDSlog‐returnsandthemonthlycorrelationofthe
dailypercentagechangesofthesovereignbondyields.Resultsarereportedincolumn
(4)ofTableIX.
WeobservethatwhenwefiltertheCDSreturnsofagivencountrywiththeav‐
erageCDSreturnoftheremainingEMUcountriesthere isa largedecreaseintheR‐
squaredfrom34.5%to25.2%.Thecoefficientofthecommonalityinquotesismuch
lowerthaninthebaselineanalysisbutstillstatisticallysignificantat5%level.Infact,
the commonality in liquidity is the only significant regressor, besides the lagged
dependentvariable.Thisresult isworthmentioninggiventhatweareeliminatinga
highportionof the comovementsacross countrieswhen filtering theCDS return. In
fact,theR‐squaredobtainedwiththemarketmodelregressioninthefirststageison
average 0.64.Whenwe filter the returns using the iTraxx returns, the explanatory
powerof theregressorsalsodiminishessignificantly(theR‐squareddecreases from
34.5% to 27%). Nevertheless, the coefficient for the commonalities in liquidity is
similartotheoneobtainedinthebaselineanalysisandalsosignificantat1%signifi‐
cancelevel.
Wenextmodel the excess correlation as the difference between the correla‐
tions in the CDS and bond markets. If the hypothesis that CDS and bond spreads
19The only traded index on European sovereign CDS (SovXWestern Europe) started trading on 28September2009.Thisindexconsistsof15countriesbutitsinitialdateisfarawayfromthebeginningofoursampleandforthisreasonwedecidedtocreateanequalityweightedindexthatisavailableforthewholeperiod.
31
represent twomeasuresof thesamecredit risk is true, then thedifferencebetween
bothcorrelationsshouldcanceloutthecomovementsdrivenbycountryfundamentals
thataresignificantdeterminantsofcreditrisk.Badaoui,CathcartandEl‐Jahel’s(2013)
find that sovereign bond spreads are less subject to liquidity frictions than CDS
spreadsandso,weexpectthattheCDSmarketfrictionshaveastrongereffectonsuch
comovements. We find that the commonalities in quotes significantly explain the
excesscorrelationwhile,asexpected,countryandglobal risk factorsarenotsignifi‐
cantnowgiventhattheyarealsocontainedinthebondyields.
<InsertTableIXhere>
b. Othermeasuresofcommonalityinquotes
Alternatively to the baseline commonality measure we consider three other
variationsofsuchmeasure.Thefirstvariationisdefinedas:
2∑ min NQ , NQ
∑ sum NQ ,NQ∈ 0,0.5 (6)
whereNQ andNQ are the number of quotes given to country a and country b
respectivelybydealerd,whoisamongthedealersgivingcommonquotestocountries
aandbinagivenmonth.Thenoveltyofthismeasureisthatitignoresthequotesthat
aregivenbydealersonlyreportingCDSpricesforasinglecountry.Theresultsforthis
measure are reported in column (2) of Table X and are qualitatively similar to the
onesobtained for thebaselinespecification(column(1))andso, theresultsarenot
biasedbyrestrictingouranalysistothedealersgivingpricestoagivenpairofcoun‐
tries.
Thesecondvariationforthemeasureofcommonalitiesisdefinedasfollows:
3min NQ , NQsum NQ , NQ
∈ 0,0.5 ∗ NCD (7)
32
whereNCDisthenumberofcommondealersgivingquotestoapairofcountries.The
differenceof thismeasurewith respect to 2is that it aggre‐
gatesthecommonalityratioatdealerlevel.Contraryto 2, in
whichdealersreportingahighernumberofquoteshaveahigherweightinthecom‐
putationofthemeasure,thisnewmeasureconsidersalldealersequally.Thevalueof
this measure depends on the number of dealers giving common quotes in a given
monthanditsvaluesspanfromzeroto0.5*NCD.Ontheonehand,thedependenceon
thenumberofdealersenablesustoassignhighervaluestothecommonalitymeasure
asthemarketparticipantsactivityincreases.Ontheotherhand,thisprocedurecould
makethismeasuredependentonmarketactivity.Thelowermagnitudeforthecoeffi‐
cient of this variable as reported in column (3) is due to the different scale that
depends on NCD. Nevertheless, the commonality measure is still significant at 1%
levelconfirmingthattheresultsarenotdrivenbutthemostactivedealersintheCDS
market. Regarding the coefficients for the remaindervariables, themaindifference
with respect to the baseline specification is that the correlation between the CDS
relativebid‐askspreads isnotsignificantnow.Thiseffect isdue to the fact that the
commonalitydefinedinthiswayisalsoameasureofliquidityasitdependsonmarket
activity(i.e.NCD).
Toavoidthedependenceofthecommonalitymeasureonmarketactivity,werescale
ittoconstructanewmeasurewithpotentialoutcomesspanningfrom0to1:
43
0.5 ∗∈ 0,1 (8)
Resultsaresimilartotheonesreportedincolumns(1)and(2)ofTableXsup‐
portingthestatementthattheresultsarenotdrivenbutthemostactivedealersinthe
CDSmarket.
<InsertTableXhere>
33
c. OtherfrequenciesfordefiningsovereignCDSreturncorrelation
Toshowthat theeffectof thecommonalitydoesnotdependon thedata fre‐
quency employed we repeat our analysis using daily data (964 trading days). To
implementthedailyanalysisweneedtodefinethecorrelationbetweensovereignCDS
onadailybasis.Forsuchaim,wetakeadvantageoftheintradayquotestocalculate
hourlyCDSreturns.Weaggregatethequotesperhourusingthosereportedfrom7.00
to19.00(GMT+1)suchthatweuse13observationstocomputethedailycorrelation.20
Weconsider fourdifferentspecifications todefine thedailycorrelationbetweenthe
hourlyCDSreturnsreportedinTableXI.First,weestimatethedailycorrelationusing
CDSquotessuchthatwhenthereisamissingvalueforagivendealerinagivenhour
we impute the previous quote available for this dealer (column (2)). The second
specification(column(3))issimilartothepreviousonebutweexcludetheobserva‐
tionsforwhichthehourlyCDSreturnis0todiscardanyarbitraryimputationbeyond
ourscope.Toavoidanybiasduetotheeffectoftheimputedobservationswecompute
thedailycorrelationleavingthemissingvalueswithoutreplacement(column(4)).In
the lastspecification(column(5))we leave themissingvalueswithoutreplacement
andadditionallyexcludeobservations forwhich thehourly return is0 for thesame
reason explained in specification contained in column (3). Regarding the remaining
independent variables, the absolute differences for a given pair of countries in the
ratiosofdebtanddeficit toGDParetheonesused inTable IIIandhaveaquarterly
frequency.ThevariablesreferredtotheECBBondPurchasesareupdatedweeklydue
totheinformationavailability.Theremainingexplanatoryvariablesareupdatedona
dailybasis.
<InsertTableXIhere>
V. Conclusion
CDSdatavendorsemploytheirmethodologiestoofferdailyquotesthatareob‐
tainedaftercombiningthequotesreceivedbydifferentdealers.Wetestwhetherthe
20The lower frequency of quotes for the years 2008 and 2009 impedes to increase the number ofintradayobservationsemployed toobtain theobservationsgiven thatweneed touse thesame timespanforalltheyearsinthesample.
34
commonquotesforagivenpairofcountries,reportedbythesamedealer,affectthe
correlationbetweenCDSspreadsandfindthattheydo.
Wefindthattheeffectofthecommonalityinquotesissignificantatanystand‐
ard significance level and has very strong forecasting power on the future
comovements between sovereign CDS spread. In fact, the economic impact of the
commonality variable is much stronger than the one attributable to the remaining
explanatoryvariablesconsideredinouranalysis,includingthetraditionalfundamen‐
talvariables.ThisresultisrobusttotheinclusionornotofGreeceintheanalysis,the
pre‐crisisandcrisisperiods,thedatafrequency,theuseofabnormalCDSreturnson
the basis of amarketmodel to compute the comovements, or considering different
specificationsforliquiditycommonalityinquotes.
The strong effect of this commonality in quotes is explained by the strategy
adoptedby thedealers todealwith twomarket frictions: inventory risk and asym‐
metric information.The comovements among sovereignCDSare strongerwhen the
dealers have excess inventory risk and face selling pressure. Dealers aiming to sell
protectionforthetwocountriesatthesametimecouldleadtoalargercomovement,
duetotheeffectofforcedsellingatfiresaleprices.Finally,wefindthatthosedealers
are not better informed than the “average dealer” on the CDS prices also lead to
strongercomovements,giventhatthelackofinformationleadthemtoreviewprices
similarlyforthetwocountriesformingthepair.
Besidescontributingtotheacademic literatureonCDSmarketcomovements,
our findingswill be informative to regulators and investors and present a series of
relevantpolicyimplications.
Theintensityofthetwomainchannelsincreasingthecomovementsoverand
abovetheeffectoffundamentals(sellingpressureandlackofinformationbydealers)
couldbeweakenedbyimprovingthetransparencyandinitiatingthecentralclearing.
Aproper level of transparencywould lead to a lower adverse selection, to improve
dealerrisksharing,andtobetterinformationforinvestors.Thus,highertransparency
wouldcontributetoabetterfunctioningof theCDSmarketandtoabetter informa‐
35
tional content inCDSprices about the real credit riskof a given country.Themore
realistic and fairer prices of credit risk would propitiate a proper measuring and
monitoringofcomovementsandcontagionacrosscountriesthroughtheCDSmarket.
On the other hand, higher level of transparency could diminish trading profits for
marketparticipantsandcoulddeter themore informeddealers fromplacingquotes
thatcouldreducetheir informationaladvantageandso, theiropportunities forarbi‐
trage.Asaconsequence,itcouldleadtoalowerliquidityortoachangeinthequoting
practicesofmarketparticipants.
Finally,thenewevidenceonthedeterminantsofthecomovementamongsov‐
ereignCDSspreadshasimportantimplicationsforriskdiversificationoftheeurozone
debtportfoliosgiventhatinvestorsshouldunderstandthatan importantpartofthe
comovements intheirportfolios isnotdueto fundamentalsbut tocommonalities in
the dealers’ quotes. A proper understanding of comovements across fair prices of
creditriskwouldcontributetoimprovethehedginganddiversificationstrategies.
36
References
AcharyaV.V., I.DrechslerandSchnabl,P. (2011). “APyrrhicVictory?BankBailoutsandSovereignCreditRisk”,NBERWorkingPaperNo.17136.
Acharya,V.V.,andJohnson,T.C.(2007).“Insidertradingincreditderivatives”,JournalofFinancialEconomics84,110‐141.
Acharya,V.andSchaefer,S.(2006).“LiquidityRiskandCorrelationRisk:ImplicationsforRiskManagement”,LondonBusinessSchool,WorkingPaper.
Acharya, V., S. Schaefer, S. and Zhang, Y. (2007). “Liquidity risk and correlation risk: AclinicalstudyoftheGeneralMotorsandFordDowngradeofMay2005”,WorkingPaper.
Adrian,T.,Etula,E.,andMuir,T.(2013).“FinancialIntermediariesandtheCross‐SectionofAssetReturns”.FederalReserveBankofNewYork.StaffReportNo.464
Aït‐Sahalia,Y.,Andritzky,J.,Jobst,A.,Nowak,S.,andTamirisa,N.(2012).“Marketresponsetopolicyinitiativesduringtheglobalfinancialcrisis”,JournalofInternationalEconomics87,162‐177.
Alter,A.,andSchüler,Y.S.(2012).“CreditSpreadInterdependenciesofEuropeanStatesandBanksduringtheFinancialCrisis”,JournalofBankingandFinance,36,3444‐3468.
Andenmatten,S.,andBrill,F.(2011).“MeasuringComovementsofCDSPremiaduringtheGreekDebtCrisis”,UniversityofBern,DepartmentofEconomicsDiscussionPaper11‐04.
Anderson,M.(2012).“ContagionandExcessCorrelationinCreditDefaultSwaps”,Work‐ingPaperSSRN2023043.
Andrade,S.C.,Chang,C.,Seasholes,M.S.(2008).“Tradingimbalances,predictablerever‐sals,andcross‐stockpricepressure”,JournalofFinancialEconomics88,406–423.
Angelopoulos,G.,andGiamouridis,D.(2012).“Dissectinginsidertradingincreditderiva‐tives”,WorkingPaper.
Antón,M.andPolkC.(2013).“ConnectedStocks”,JournalofFinance,forthcoming.
Arce,O.,Mayordomo,S.,andPeña,J.I.(2013).“Credit‐riskvaluationinthesovereignCDSand bondmarkets: evidence from the euro area crisis”, Journal of InternationalMoneyandFinance,35,124‐145.
Arora,N.,Gandhi, P., andLongstaff, F. (2012). “CounterpartyCreditRiskand theCreditDefaultSwapMarket”,JournalofFinancialEconomics103,280‐293.
BadaouiS.,CathcartL.,andEl‐JahelL.(2013).“Dosovereigncreditdefaultswapsrepre‐sent a clean measure of sovereign default risk? A factor model approach”, Journal ofBanking&Finance,37,2392‐2407.
Barberis,N.andShleifer,A.(2003).“Style investing”, JournalofFinancialEconomics68,161‐199.
37
Barberis, N., Shleifer, A. and Wurgler, J. (2005). “Comovement”, Journal of FinancialEconomics75,283–317.
Beber,A.,Brandt,M.W.andKavajecz,K.A.(2009).“Flight‐to‐qualityorflight‐to‐liquidity?EvidencefromtheEuro‐areabondmarket”,ReviewofFinancialStudies,22,925‐957.
BekaertG.,Harvey,C.R.,andNg,A.(2005).“Marketintegrationandcontagion”,JournalofBusiness,78,39‐70.
Berndt,A.,andOstrovnaya,A.(2008).“DoEquityMarketsFavorCreditMarketNewsOverOptionsMarketNews?”,WorkingPaperSSRN972806.
Bernoth, K., von Hagen J. and Schuknecht, L. (2012). “Sovereign Risk Premiums in theEuropean Government Bond Market”, Journal of International Money and Finance, 31,975–995.
Brunnermeier,M.K.,andPedersen,L.H.(2009).“MarketLiquidityandFundingLiquidi‐ty”,ReviewofFinancialStudies,22,2201‐2238.
Cathcart,L,El‐Jahel,L.,andEvans,L.(2010).“TheCorrelationStructureoftheCDSMar‐ket:AnEmpiricalInvestigation”,TheJournalofFixedIncome,22,53‐74.
Dieckmann,S.andPlank,T. (2012). “DefaultRiskofAdvancedEconomies:AnEmpiricalAnalysisofCreditDefaultSwapsduringtheFinancialCrisis”,ReviewofFinance,16,903‐934.
Dornbusch, R., Park, Y. C., and Claessens, S. (2000). “Contagion: Understanding how itspreads”,TheWorldBankresearchobserver,15(2),177‐198.
Dufour, A., and Engle, R. F. (2000). “Time and the price impact of a trade”, Journal ofFinance55,2467‐2498.
Dufour,A.,andNguyen,M.(2012).“PermanentTradingImpactsandBondYields”,Euro‐peanJournalofFinance,18,841‐864.
Ellul,A.,Jotikasthira,C.,andLundblad,C.T.(2011).“RegulatoryPressureandFireSalesintheCorporateBondMarket”,JournalofFinancialEconomics,101,596–620.
Favero,C.,Pagano,M.andVonThaddenE.‐L.(2010)“HowDoesLiquidityAffectGovern‐mentBondYields?”,JournalofFinancialandQuantitativeAnalysis,45,107‐134
Forbes, K., and Rigobon, R. (2001). “Contagion in Latin America: Definitions, measure‐ment,andpolicyimplications”,Economia,1,1‐46.
Forbes,K.,andRigobon,R.(2002)“Nocontagion,onlyinterdependence:Measuringstockmarketcomovements”,JournalofFinance,57,2223‐2261.
Geyer,A.Kossmeier, S. andPichler, S. (2004). “Measuring SystematicRisk inEMUGov‐ernmentYieldSpreads”,ReviewofFinance,8,171–197.
Gündüz, Y., andKaya,O. (2013). “LongMemory andCorrelationDynamics of EurozoneSovereignCDSs”,WorkingPaperSSRN2248559.
38
Gündüz,Y.,Lüdecke,T.,andUhrig‐Homburg,M.(2007).“TradingCreditDefaultSwapsviaInterdealerBrokers”,JournalofFinancialServicesResearch,32,141–159.
Gündüz,Y.,Nasev, J.,andTrapp,M.(2013).“ThePriceImpactofCDSTrading”,WorkingPaper.
Hasbrouck, J. (1991a). “Measuring the information content of stock trades”, Journal ofFinance46,179‐207.
Hasbrouck, J. (1991b). “The summary informativeness of stock trades: An econometricanalysis”,ReviewofFinancialStudies4,571‐595.
Ho,T.,Stoll,H., (1983). “Thedynamicsofdealermarketsundercompetition”, JournalofFinance,38,1053–1074.
Kalbaska, A. and Gatkowski,M. (2012). “Eurozone Sovereign Contagion: Evidence fromtheCDSMarket(2005‐2010)”, JournalofEconomicBehaviorandOrganization,83,657‐673.
King,M.A.,andWadhwani,S.(1990).“Transmissionofvolatilitybetweenstockmarkets”,TheReviewofFinancialStudies,3,5‐33.
Levene,D.B.(2012).“CreditDefaultSwapsandInsiderTrading”,WorkingPaper.
Longstaff, F. A., Pan, J., Pedersen, J. P., and Singleton, K. J. (2005). “How Sovereign IsSovereignCreditRisk?”,AmericanEconomicJournal:Macroeconomics,3,75‐103.
Longstaff,F.A(2010).“TheSubprimeCreditCrisisandContagioninFinancialMarkets”,JournalofFinancialEconomics,97,436–450.
Manasse,P.andZavalloni,L.(2013).“SovereignContagioninEurope:EvidencefromtheCDSMarket”,TheRiminiCentreforEconomicAnalysis.WP13‐08.
Manganelli,S.(2012).“Theimpactofthesecuritiesmarketprogramme”,ResearchBulle‐tin17,2‐5.
Mayordomo,S.,Peña,J.I.,andSchwartz,E.(2012).“TowardsaCommonEuropeanMone‐taryUnionRiskFreeRate”,WorkingPaper.
McAndrews,J.,A.SarkarandZ.Wang(2008).“TheeffectofthetermauctionfacilityontheLondonInter‐BankOfferedrate”.FederalReserveBankofNewYorkStaffReportsNo.335
Micu, M., Remolona, E., Wooldridge, P. (2006). “The price impact of rating announce‐ments:whichannouncementsmatter?”,BISWorkingPapersNo207.
Mitchell,M.,Pedersen,L.H.,Pulvino,T.(2007).“SlowMovingCapital”,AmericanEconom‐icReview,97,215‐220,
Pedersen, L. H. (2009). “When Everyone Runs for the Exit”, International Journal ofCentralBanking,5,177‐199.
Pu,X., andZhao,X. (2010). “Correlation inCreditRisk”,Officeof theComptrollerof theCurrencyEconomicsWorkingPaper2009‐5.
39
Qiu,J.,andYu,F.(2012).“EndogenousLiquidityinCreditDerivatives”,JournalofFinancialEconomics,103,611–631.
Rigobon,R.(2002).“Contagion:Howtomeasureit?”,InS.Edwards&J.A.Frankel(Eds.),Preventingcurrencycrisesinemergingmarkets(269‐334).UniversityofChicagoPress.
Shachar,O. (2013). “Exposingtheexposed: Intermediationcapacity in thecreditdefaultswapmarket”.WorkingPaper.
Tang, D. Y., and Yan, H. (2010). “Does the TailWag the Dog? The Price Impact of CDSTrading”,WorkingPaper.
Zhang, X., Schwaab, B., and Lucas, A. (2011). “Conditional Probabilities and ContagionMeasuresforEuroAreaSovereignDefaultRisk”,TinbergenInstituteDiscussionPaperNo.11‐176/2/DSF29.
40
TableI:SummaryStatistics
This table reports summarystatisticson thequotesanddealers reportingquotesper country.PanelAcontainsinformationonthetotalnumberofquotesandthedailyaveragenumberofquotespercountryaswellasthetotalandthedailyaveragenumberofdealersgivingquotesforeachcountry.PanelBsummarizesthetotalnumberofquotesperdealerandthedealer’smarketshare.
PANELA:Descriptivestatistics.Sample:January2008‐Oct2011
Country NumberofQuotes NumberofDealers
Aggregate DailyAverage Aggregate DailyAverage
Austria 448773 458 90 23.5
Greece 598638 593 93 24.5
Italy 581538 572 91 24.4
Portugal 624875 621 92 24.2
Spain 682032 673 93 24.6
Finland 331887 355 90 22.1
Germany 386181 391 89 22.5
Belgium 449147 446 90 23.3
Ireland 606506 604 92 24.1
France 469751 472 91 23.0
Netherlands 347764 365 89 22.6
PANELB:Shareofquotesbydealers
Dealer NumberofQuotes
Top15 Total Share
1 303851 5.5%
2 298052 5.4%
3 274772 5.0%
4 268195 4.9%
5 264496 4.8%
6 262095 4.7%
7 240076 4.3%
8 221383 4.0%
9 204477 3.7%
10 200736 3.6%
11 183192 3.3%12 183031 3.3%
13 182606 3.3%
14 166659 3.0%
15 164469 3.0%
1‐10 2538133 45.9%
11‐20 1609411 29.1%
21‐30 873379 15.8%
31‐40 335378 6.1%
41‐50 110119 2.0%
51‐95 60672 1.1%
41
TableII:SummaryStatistics(cont’d)
This table reports summarystatisticson theCDS spread level (CDS), thehourly anddailyCDS spread log return(cCDS), thedaily(CORR_d)andmonthly(CORR_hmandCORR_dm)correlationbetweentheCDSlogreturnforallthe pairs of EMU countries, and the daily (CC_d) andmonthly (CC_m) commonality inquotes for such countries.CORR_d,CORR_hm,andCC_dareobtainedusinghour‐leveldatawhileCORR_dmiscalculatedfromdailydata.PanelAreportstheinformationforthe11EMUcountrieslistedinTable1whilePanelBexcludesGreece.PanelsCandDbreakdownthemeanandstd.dev.peryear.PanelCreferstothe11EMUcountriesandPanelDexcludesGreece.
PANELA:ALLCOUNTRIESVariable Freq mean Std pctl_0 pctl_1 pctl_5 pctl_50 pctl_95 pctl_99 pctl_100CDS Hour 203 421 5 9 23 85 800 1631 8533cCDS Hour 0.00 0.02 ‐0.48 ‐0.04 ‐0.02 0.00 0.02 0.04 0.40cCDS Daily 0.00 0.06 ‐0.63 ‐0.14 ‐0.08 0.00 0.09 0.18 0.64CORR_d Daily 0.34 0.47 ‐1.00 ‐1.00 ‐0.61 0.44 0.93 1.00 1.00CORR_hm Monthly 0.38 0.27 ‐1.00 ‐0.39 ‐0.10 0.41 0.77 0.88 1.00CORR_dm Monthly 0.65 0.26 ‐0.58 ‐0.29 0.10 0.72 0.91 0.95 0.98CC_d Daily 0.38 0.09 0.01 0.05 0.17 0.41 0.48 0.48 0.50CC_m Monthly 0.37 0.09 0.04 0.11 0.18 0.40 0.47 0.48 0.48
PANELB:ALLCOUNTRIESbutGREECEVariable Freq mean Std pctl_0 pctl_1 pctl_5 pctl_50 pctl_95 pctl_99 pctl_100CDS Hour 143 179 5 8 22 78 547 927 1251cCDS Hour 0.00 0.02 ‐0.48 ‐0.04 ‐0.02 0.00 0.02 0.04 0.39cCDS Daily 0.00 0.06 ‐0.63 ‐0.14 ‐0.08 0.00 0.09 0.18 0.64CORR_d Daily 0.35 0.47 ‐1.00 ‐1.00 ‐0.60 0.45 0.93 1.00 1.00CORR_hm Monthly 0.39 0.28 ‐1.00 ‐0.41 ‐0.09 0.41 0.78 0.88 1.00CORR_dm Monthly 0.66 0.26 ‐0.58 ‐0.29 0.11 0.73 0.91 0.95 0.98CC_d Daily 0.38 0.09 0.01 0.05 0.17 0.41 0.48 0.48 0.50CC_m Monthly 0.37 0.09 0.04 0.11 0.18 0.40 0.46 0.48 0.48
PANELC:SUBPERIODS 2008 2009 2010 2011Variable Freq Mean std mean std mean std mean stdCDS Hour 45 42 93 65 181 201 414 727cCDS Hour 0.00 0.03 0.00 0.01 0.00 0.01 0.00 0.01cCDS Daily 0.01 0.08 0.00 0.05 0.00 0.05 0.00 0.04CORR_d Daily 0.12 0.72 0.22 0.44 0.47 0.34 0.46 0.35CORR_hm Monthly 0.14 0.31 0.32 0.16 0.55 0.16 0.52 0.20CORR_dm Monthly 0.41 0.34 0.76 0.12 0.71 0.17 0.70 0.18CC_d Daily 0.28 0.12 0.41 0.05 0.41 0.06 0.40 0.07CC_m Monthly 0.27 0.09 0.41 0.03 0.41 0.05 0.39 0.06
PANELD:SUBPERIODS(AllbGr) 2008 2009 2010 2011Variable Freq Mean std mean std mean std mean stdCDS Hour 41 37 85 61 132 117 260 271cCDS Hour 0.00 0.04 0.00 0.01 0.00 0.01 0.00 0.01cCDS Daily 0.01 0.08 0.00 0.05 0.00 0.05 0.00 0.04CORR_d Daily 0.12 0.73 0.22 0.44 0.46 0.34 0.50 0.34CORR_hm Monthly 0.14 0.31 0.32 0.17 0.55 0.16 0.56 0.17CORR_dm Monthly 0.41 0.35 0.76 0.12 0.73 0.15 0.73 0.15CC_d Daily 0.28 0.12 0.41 0.05 0.42 0.06 0.41 0.06CC_m Monthly 0.27 0.08 0.41 0.02 0.41 0.04 0.40 0.06
42
TableIII:CommonalityinQuotesPredictsCDSReturnCorrelation
ThistablereportstheestimatesofmonthlypanelregressionsforecastingthecorrelationofdailySovereignCDSlogReturns inmonth t for thesampleof11Europeancountries listed inTable I,PanelA.Theregressionsareat thecountry‐pairlevel(55differentcountry‐pairs),andfortheperiodofJanuary2008toNovember2011(47months).TheindependentvariablesincludetheCommonalityinQuotes,whichreferstothequotesgivenbydealerstobothcountriesinthepair,andasetofcontrols,allofthemint‐1. ∑ min , /
,whereNQadtandNQbdtarethenumberofquotesgiventocountryaandcountrybrespectivelybydealerdinagivenmonthtwhileTQatandTQbtarethetotalnumberofquotesgivenbyalldealerstocountriesaandbattimet,respectively.Thesetofcontrolsincludethedependentvariablelaggedonemonth,agroupofglobalcontrols,andagroupofpair‐levelcontrols.Theglobalcontrolsarethechangeinthelogofcounterpartyrisk,thechange in the log of Euribor‐OIS, and ECB Bond Purchases. These three variables are also interacted with thevariablePeripheral,adummythattakesvale1whenthetwocountriesinthepairareperipheralcountries(Greece,Ireland,Italy,Portugal,andSpain).Thefivepair‐levelcontrolsareself‐explanatoryinthewaytheyarelabeledinthetable,andinthebodyofthetext.Country‐pairfixedeffectsareincludedinthesixspecifications.Timeeffectsarenotincludedgiventhatthemajorityofspecifications(3to6)containmacrocontrols.Standarderrorsinbracketsaredouble‐clusteredatthecountry‐pairandmonthlevel.*,**,and***indicatestatisticalsignificanceatthe10%,5%,and1%levels,respectively.
PANELA DependentVariable:CorrelationofdailyCDSlog‐returnst (1) (2) (3) (4) (5) (6)
CommonalitiesinQuotest‐1 1.228*** 1.140*** 1.087***[0.197] [0.178] [0.180]
CorrelationofCDSLogRet.t‐1 0.400*** 0.227** 0.375*** 0.225** 0.343*** 0.205*[0.116] [0.108] [0.118] [0.110] [0.118] [0.109]
ΔLogCounterpartyRiskt‐1 ‐0.144 ‐0.077 ‐0.132 ‐0.071[0.133] [0.109] [0.137] [0.113]
ΔLogCounterpartyRiskt‐1xPeripheral ‐0.056 ‐0.109 ‐0.068 ‐0.114[0.070] [0.072] [0.071] [0.071]
ΔLogEuribor‐OISt‐1 ‐0.125 ‐0.084 ‐0.108 ‐0.072[0.085] [0.075] [0.084] [0.075]
ΔLogEuribor‐OISt‐1xPeripheral 0.171** 0.175** 0.156** 0.163**[0.070] [0.070] [0.074] [0.072]
ECBBondPurchasest‐1 0.004*** 0.003*** 0.004*** 0.003**[0.001] [0.001] [0.001] [0.001]
ECBBondPurchasest‐1xPeripheral ‐0.003*** ‐0.002** ‐0.003*** ‐0.002**[0.001] [0.001] [0.001] [0.001]
Corr.CountryBanksCDSLogRet.t‐1 ‐0.016 0.005[0.021] [0.016]
Corr.CountryStockIndexesVola.t‐1 ‐0.008 ‐0.012[0.030] [0.023]
Corr.CDSRelativeBid‐Askt‐1 0.086** 0.066**[0.035] [0.033]
Abs|DeficittoGDPi‐DeficittoGDPj|t‐1 ‐0.001 ‐0.001[0.001] [0.001]
Abs|DebttoGDPi‐DebttoGDPj|t‐1 ‐0.001*** ‐0.001**[0.001] [0.001]
Constant 0.369*** ‐0.021 0.373*** 0.006 0.475*** 0.113 [0.085] [0.106] [0.085] [0.102] [0.102] [0.116]
Observations 2,370 2,370 2,370 2,370 2,370 2,370R‐squared 0.242 0.316 0.276 0.335 0.292 0.345
43
TableIV:CommonalityinQuotesPredictsCDSReturnCorrelation(cont’d)
PanelBof thisTableshowstheestimatesof theregressionssimilartocolumn(6)inPanelA,butwhenGreeceisexcludedfromthesample(column(2)),andtwodifferentsubperiodsareanalyzed,correspondingtothefirstandsecondpartofthesample:January2008toNovember2009incolumn(3),andDecember2009toOctober2011incolumn (4). Everything else remains as in Panel A. Column (1) replicates column (6) of Panel A for comparisonpurposes.
PANELB DependentVariable:CorrelationofdailyCDSlog‐returns
(1) (2) (3) (4) All NoGreece Jan08‐Nov09 Dec09‐Nov11
CommonalitiesinQuotest‐1 1.087*** 1.296*** 1.400*** 0.866***[0.180] [0.175] [0.199] [0.237]
CorrelationofCDSLogRet.t‐1 0.205* 0.179* 0.103 ‐0.069[0.109] [0.107] [0.125] [0.060]
ΔLogCounterpartyRiskt‐1 ‐0.071 ‐0.051 ‐0.071 0.487***[0.113] [0.107] [0.119] [0.101]
ΔLogCounterpartyRiskt‐1xPeripheral ‐0.114 ‐0.059 ‐0.075 ‐0.138[0.071] [0.065] [0.069] [0.090]
ΔLogEuribor‐OISt‐1 ‐0.072 ‐0.067 ‐0.025 ‐0.100[0.075] [0.068] [0.073] [0.084]
ΔLogEuribor‐OISt‐1xPeripheral 0.163** 0.165* 0.148 0.125**[0.072] [0.089] [0.131] [0.051]
ECBBondPurchasest‐1 0.003** 0.004*** ‐0.001[0.001] [0.001] [0.001]
ECBBondPurchasest‐1xPeripheral ‐0.002** ‐0.002* ‐0.001[0.001] [0.001] [0.001]
Corr.CountryBanksCDSLogRet.t‐1 0.005 0.001 ‐0.028 ‐0.004[0.016] [0.014] [0.023] [0.027]
Corr.CountryStockIndexesVola.t‐1 ‐0.012 ‐0.020 ‐0.040 ‐0.018[0.023] [0.035] [0.081] [0.031]
Corr.CDSRelativeBid‐Askt‐1 0.066** 0.057* 0.161*** 0.021[0.033] [0.033] [0.062] [0.034]
Abs|DeficittoGDPi‐DeficittoGDPj|t‐1 ‐0.001 ‐0.001 ‐0.007** ‐0.001[0.001] [0.001] [0.003] [0.001]
Abs|DebttoGDPi‐DebttoGDPj|t‐1 ‐0.001** ‐0.001** ‐0.003 ‐0.001[0.001] [0.001] [0.002] [0.001]
Constant 0.113 0.140 ‐0.028 0.436*** [0.116] [0.114] [0.135] [0.106]
Observations 2,370 1,935 864 1,071R‐squared 0.345 0.370 0.420 0.345
44
TableV:CommonalityinQuotesfromUninformedTraders
This table reports panel estimates of monthly cross‐sectional regressions forecasting the correlation of dailySovereignCDSReturnsinmonth forthesampleof11EuropeancountrieslistedinTableI,PanelA.Theregressionsareatthecountry‐pair level(55differentcountry‐pairs),andfortheperiodofJanuary2008toOctober2011(47months).TheindependentvariablesinthistablearesimilartothoseinTableIII.ThenoveltyhereisthatwenowlookindependentlyatthecommonquotesfordealersdependingontheirdegreeofinformationthatisdeterminedonthebasisoftheGonzaloandGranger’s(1995)pricediscoverymethodology.Thus,CommonalityfromUninformedTradersmeasuresthevariableCommonalityinQuotesfordealerswhoarelessinformedthantheaverageontheCDSpricesof the twocountries forming thepair. Similarly,CommonalityfromInformedTradersmeasures thevariableCommonalityinQuotes for dealerswith price discoverymetrics indicating that they aremore informed than theaveragedealerontheCDSpricesofthetwocountriesinthepair.Finally,CommonalityfromOtherTradersmeasuresthecommonquotes fordealers informedonacountrybutnoton theother.Allothercontrolsareas inTable III.Country‐pair fixed effects are included in the three specifications. Time effects are not included given that allspecifications contain macro controls. Standard errors in brackets are double‐clustered at the country‐pair andmonthlevel.*,**,and***indicatestatisticalsignificanceatthe10%,5%,and1%levels,respectively.
DepVar:CorrofdailyCDSlog‐returnst
(1) (2) (3) (4)
CommonalityfromUninformedTraderst‐1 1.188*** 1.009***[0.278] [0.255]
CommonalityfromInformedTraderst‐1 1.000*** 0.279[0.297] [0.234]
CommonalityfromOtherTraderst‐1 0.788** 0.448*[0.314] [0.239]
CorrelationofCDSLogRet.t‐1 0.205* 0.290*** 0.254** 0.160[0.111] [0.111] [0.103] [0.103]
ΔLogCounterpartyRiskt‐1 ‐0.076 ‐0.104 ‐0.078 ‐0.046[0.111] [0.126] [0.120] [0.106]
ΔLogCounterpartyRiskt‐1xPeripheral ‐0.059 ‐0.062 ‐0.065 ‐0.057[0.068] [0.073] [0.069] [0.067]
ΔLogEuribor‐OISt‐1 ‐0.108 ‐0.109 ‐0.081 ‐0.093[0.074] [0.079] [0.077] [0.071]
ΔLogEuribor‐OISt‐1xPeripheral 0.146** 0.170** 0.147** 0.146**[0.064] [0.080] [0.070] [0.063]
ECBBondPurchasest‐1 0.004*** 0.003** 0.003** 0.003**[0.001] [0.001] [0.001] [0.001]
ECBBondPurchasest‐1xPeripheral ‐0.003** ‐0.004*** ‐0.003*** ‐0.003***[0.001] [0.001] [0.001] [0.001]
Corr.CountryBanksCDSLogRet.t‐1 ‐0.025 ‐0.016 ‐0.016 ‐0.024[0.021] [0.020] [0.021] [0.020]
Corr.CountryStockIndexesVola.t‐1 0.001 ‐0.006 0.000 0.005[0.019] [0.025] [0.024] [0.016]
Corr.CDSRelativeBid‐Askt‐1 0.085*** 0.097*** 0.094*** 0.093***[0.033] [0.035] [0.034] [0.033]
Abs|DeficittoGDPi‐DeficittoGDPj|t‐1 ‐0.002 ‐0.002 ‐0.002 ‐0.002[0.001] [0.001] [0.001] [0.001]
Abs|DebttoGDPi‐DebttoGDPj|t‐1 ‐0.001*** ‐0.001*** ‐0.001*** ‐0.001***[0.000] [0.001] [0.001] [0.000]
Constant 0.363*** 0.455*** 0.417*** 0.342*** [0.107] [0.104] [0.114] [0.115]
Observations 2,370 2,370 2,370 2,370R‐squared 0.342 0.312 0.315 0.353
45
TableVI:CommonalityinQuotesfromSellingPressure
This table reports panel estimates of monthly cross‐sectional regressions forecasting the correlation of dailySovereignCDSReturnsinmonth forthesampleof11EuropeancountrieslistedinTableI,PanelA.Theregressionsareatthecountry‐pair level(55differentcountry‐pairs),andfortheperiodofJanuary2008toOctober2011(47months). The independent variables in this table are similar to those in Table III. The novelty here is that thevariableCommonalityinQuotesisdefined in adifferentway. Instead ofCommonalityinQuotes for all dealers,wenowlookindependentlyatthecommonquotesfordealersthatexperiencesellingpressureinbothcountriesofthepair,andthosethatdonotexperiencesellingpressure.Tocapturetheeffectofthissellingpressure,wedecomposethevariableCommonality inQuotes in twovariables.Toconstruct theCommonalityfromSellingPressureweonlyuse the quotes by a given dealerwhose ask price is in the first quartile for the two countries forming the pairrelative to the total number of quotes. Similarly, Commonality not from SellingPressuremeasures the variableCommonalityinQuotesfordealerswhoseofferoraskquotesarenotinthefirstquartileinanyofthetwocountries.Allothercontrols(macro,andcountry‐pair)areasinTableIII.Country‐pairfixedeffectsareincludedinthethreespecifications.Timeeffectsarenotincludedgiventhatallspecificationscontainmacrocontrols.Standarderrorsinbracketsaredouble‐clusteredat thecountry‐pairandmonth level.*, **,and*** indicatestatisticalsignificanceatthe10%,5%,and1%levels,respectively.
DepVar:CorrofdailyCDSlog‐returnst(1) (2) (3)
CommonalityfromSellingPressuret‐1 1.113** 1.724***[0.432] [0.415]
CommonalitynotfromSellingPressuret‐1 0.801*** 1.024***[0.202] [0.186]
CorrelationofCDSLogRet.t‐1 0.323*** 0.256** 0.201*[0.115] [0.116] [0.108]
ΔLogCounterpartyRiskt‐1 ‐0.115 ‐0.100 ‐0.064[0.131] [0.121] [0.111]
ΔLogCounterpartyRiskt‐1xPeripheral ‐0.063 ‐0.105 ‐0.108[0.073] [0.069] [0.071]
ΔLogEuribor‐OISt‐1 ‐0.087 ‐0.097 ‐0.061[0.085] [0.079] [0.076]
ΔLogEuribor‐OISt‐1xPeripheral 0.162** 0.157** 0.166**[0.073] [0.074] [0.072]
ECBBondPurchasest‐1 0.004*** 0.004*** 0.003**[0.001] [0.001] [0.001]
ECBBondPurchasest‐1xPeripheral ‐0.003*** ‐0.003** ‐0.002***[0.001] [0.001] [0.001]
Corr.CountryBanksCDSLogRet.t‐1 ‐0.016 ‐0.001 0.004[0.019] [0.019] [0.015]
Corr.CountryStockIndexesVola.t‐1 ‐0.021 ‐0.002 ‐0.019[0.028] [0.025] [0.022]
Corr.CDSRelativeBid‐Askt‐1 0.078** 0.076** 0.062*[0.034] [0.034] [0.032]
Abs|DeficittoGDPi‐DeficittoGDPj| t‐1 ‐0.001 ‐0.001 ‐0.001[0.001] [0.001] [0.001]
Abs|DebttoGDPi‐DebttoGDPj|t‐1 ‐0.001** ‐0.001*** ‐0.001**[0.001] [0.001] [0.001]
Constant 0.448*** 0.228** 0.117 [0.106] [0.105] [0.117]Observations 2,370 2,370 2,370R‐squared 0.305 0.322 0.350
46
TableVII:CommonalityInstrumentedbyBroker‐Dealers’Leverage
This table reports panel estimates of monthly cross‐sectional regressions forecasting the correlation of dailySovereignCDSReturnsinmonth forthesampleof11EuropeancountrieslistedinTableI,PanelA.ThevariableCommonalityinQuotesis instrumentedwith the log changeofbroker‐dealer’s leverage in column (1) and the logchangeofbroker‐dealer’sleveragejointwiththeinteractionofthisleverageandthefinancingcostsproxy(Euribor‐OIS)incolumn(2).Theregressionsareatthecountry‐pairlevel(55differentcountry‐pairs),andfortheperiodofJanuary2008 toOctober2011(47months).Standarderrors inbracketsaredouble‐clusteredat thecountry‐pairandmonthlevel.*,**,and***indicatestatisticalsignificanceatthe10%,5%,and1%levels,respectively.
PanelA Dep.Variable:CorrelationofdailyCDSlog‐returns (1) (2)CommonalitiesinQuotest‐1 1.839*** 1.876*** [0.336] [0.222]
CorrelationCDSLogRet.t‐1 0.110*** 0.105*** [0.042] [0.033]
Corr.CountryBanksCDSLogRet.t‐1 0.023 0.024 [0.021] [0.019]
Corr.CountryStockIndexesVola.t‐1 ‐0.013 ‐0.013 [0.017] [0.017]
Corr.CDSRelativeBid‐Askt‐1 0.050*** 0.050*** [0.014] [0.013]
ΔLogCounterpartyRiskt‐1 ‐0.030 ‐0.028 [0.026] [0.027]
ΔLogCounterpartyRiskt‐1xPeripheral ‐0.146*** ‐0.147*** [0.048] [0.048]
ΔLogEuribor‐OISt‐1 ‐0.050** ‐0.049** [0.025] [0.022]
ΔLogEuribor‐OISt‐1xPeripheral 0.165*** 0.165*** [0.040] [0.041]
ECBBondPurchasest‐1 0.003*** 0.003*** [0.000] [0.000]
ECBBondPurchasest‐1xPeripheral ‐0.002** ‐0.002** [0.001] [0.001]
Abs|DeficittoGDPi‐DeficittoGDPj|t‐1 ‐0.000 ‐0.000 [0.001] [0.001]
Abs|DebttoGDPi‐DebttoGDPj|t‐1 ‐0.001* ‐0.001* [0.001] [0.001]
Observations 2,380 2,380Numberofcountry‐pair 55 55R‐squared 0.261 0.258Underidentificationtest(Kleibergen‐PaaprkLMstatistic) 30.595 31.601
Chi‐sq(1)P‐val 0.006 0.000Overidentificationtest(HansenJstatistic) Equationexactlyidentified 0.030Chi‐sq(1)P‐val 0.863Instrumented: CommonalitiesinQuotest‐1Excludedinstruments: ΔLogLeverage ΔLogLeverage;
ΔLogLeveragexFinancingCost
47
TableVIII:ExogenousShock:“NakedCDS”banned,BaFin
Thistablereportsadifference‐in‐differenceanalysisusingtheBaFin’sbanonMay18,2010asanexogenousshock.Weusetheinformationforthethreedaysbeforeandthethreedaysaftertheannouncementandanalysetheeffectofcommonalitiesontheaveragecorrelationsconsideringasthetreatmentgroupthosepairsofcountrieswithhighlevelsofcommonalities inquotesaccordingtotheaveragecommonalities inquotesfrom1Mayto12May,2010.Thedependentvariableconsistsofoneobservationperpairbeforeandaftertheestablishmentoftheman.Thus,wecalculatetheaveragecorrelationforthethreedaysaftertheban(19‐21May,2010)andforthethreebankingdaysbeforetheban(13,14,and17May).Thesampleconsistsof55observationsforthepre‐banperiodand55observa‐tions for the post‐ban period. The variableTop‐quintile is a dummy variable that takes 1 if the average level ofcommonquotes is in the top‐quintile (80‐percentileandabove),andzero for the remainingpairsbelow the top‐quintile(below80‐percentile).ThevariablePost‐BaFinBanisadummythattakes1ifthedateis19‐21May,2010,andzeroduringthethreebankingdaysbeforetheban.AthirddummyvariablePost‐BaFinBanxTop‐quintileisthecross‐productoftheprevioustwodummyvariables.Column(2)alsocontainsthecontrolvariablesatcountrylevelwithadailyfrequency.Standarderrorsinbracketsaredouble‐clusteredatthecountry‐pairandmonthlevel.*,**,and***indicatestatisticalsignificanceatthe10%,5%,and1%levels,respectively.
DependentVariable:AverageCorrelationofdailyCDS log‐returnst
(1) (2)Post‐Bafin'sBan 0.043 0.069* [0.026] [0.040]Post‐Bafin'sBanxTop‐quintile 0.254*** 0.292*** [0.026] [0.037]Top‐quintile ‐0.001 ‐0.022 [0.025] [0.023]Corr.CountryBanksCDSLogRet.t‐1 0.403*** [0.110]Corr.CountryStockIndexesVola.t‐1 ‐0.201 [0.135]Corr.CDSRelativeBid‐Askt‐1 0.005 [0.217]Constant 0.598*** 0.442 [0.025] [0.272]Observations 110 110R‐squared 0.031 0.178
48
TableIX:OthermeasuresofCDSReturnCorrelation
This table reports panel estimates of monthly cross‐sectional regressions forecasting the correlation of dailySovereignCDSReturns (in column (1)), aswell as correlationof residuals of Sovereign CDSReturns in differentspecifications, in month for the sample of 11 European countries listed in Table I, Panel A. Column (1) justreplicatescolumn(6)ofTableIIIforcomparisonpurposes.Column(2)showsthespecificationwherethedepend‐ent variable is the correlation of residuals from the regression: , whereAvgCDSRetistheaveragedailyCDSReturnofallcountriesinthesampleexceptiatt‐1.Column(3)showsasimilarspecification,butreplacingtheAvgCDSRetwithiTraxxRet,whichisthedailyreturnontheiTraxxIndex.Column(4)showsthespecificationinwhichthedependentvariableisthedifferencebetweenthemonthlycorrelationofdailysovereignCDS log‐returnsand themonthlycorrelationofdailypercentagechangesof sovereignbondyields.Theregressionsareatthecountry‐pairlevel(55differentcountry‐pairs),andfortheperiodofJanuary2008toOctober2011(47months).TheindependentvariablesinthistablearesimilartothoseinTableIII.Country‐pairfixedeffectsare included in all four specifications. Time effects are not included given that all specifications contain macrocontrols.Standarderrorsinbracketsaredouble‐clusteredatthecountry‐pairandmonthlevel.*,**,and***indicatestatisticalsignificanceatthe10%,5%,and1%levels,respectively.
DependentVariablet:Corrof
CDSLog‐returnsCorrof
ResidualMACorrof
ResidualMI
ExcessofCorrbtw.CDSLog‐returns
andBond
(1) (2) (3) (4)
CommonalitiesinQuotest‐1 1.089*** 0.180** 0.965*** 0.945***[0.179] [0.078] [0.185] [0.325]
DependentVariablet‐1 0.205* 0.122*** 0.128 0.567***[0.109] [0.033] [0.084] [0.080]
ΔLogCounterpartyRiskt‐1 ‐0.072 0.023 ‐0.099 ‐0.117[0.113] [0.039] [0.106] [0.155]
ΔLogCounterpartyRiskt‐1xPeripheral ‐0.114 ‐0.112 ‐0.119 0.069[0.071] [0.079] [0.081] [0.137]
ΔLogEuribor‐OISt‐1 ‐0.075 0.004 ‐0.115 ‐0.170[0.075] [0.031] [0.085] [0.152]
ΔLogEuribor‐OISt‐1xPeripheral 0.160** 0.119 0.211** 0.220[0.070] [0.074] [0.082] [0.134]
ECBBondPurchasest‐1 0.003** 0.000 0.002 0.008***[0.001] [0.001] [0.001] [0.003]
ECBBondPurchasest‐1xPeripheral ‐0.002** ‐0.002 ‐0.003** ‐0.002[0.001] [0.002] [0.002] [0.004]
Corr.CountryBanksCDSLogRet.t‐1 0.009 ‐0.018 0.010 ‐0.026[0.015] [0.027] [0.025] [0.058]
Corr.CountryStockIndexesVola.t‐1 ‐0.011 0.028 0.050 ‐0.021[0.023] [0.020] [0.032] [0.057]
Corr.CDSRelativeBid‐Askt‐1 0.064* ‐0.011 0.044 ‐0.070[0.033] [0.013] [0.037] [0.068]
Abs|DeficittoGDPi‐DeficittoGDPj|t‐1 ‐0.001 ‐0.000 ‐0.001 ‐0.001[0.001] [0.001] [0.001] [0.002]
Abs|DebttoGDPi‐DebttoGDPj|t‐1 ‐0.001** ‐0.002 ‐0.002*** 0.001[0.000] [0.001] [0.001] [0.001]
Constant 0.106 ‐0.056 0.155 ‐0.250 [0.114] [0.074] [0.108] [0.167]
Observations 2,380 2,390 2,349 2,380R‐squared 0.345 0.252 0.270 0.512
49
TableX:RobustnesstoothermeasuresofCommonalityinQuotes
This table reports panel estimates of monthly cross‐sectional regressions forecasting the correlation of dailySovereignCDSReturnsinmonth forthesampleof11EuropeancountrieslistedinTableI,PanelA.Theregressionsareatthecountry‐pair level(55differentcountry‐pairs),andfortheperiodofJanuary2008toOctober2011(47months).The independentvariables in this tablearesimilar to those inTable III,buteachof the4specificationsusesadifferentdefinitionofthevariableCommonalityinQuotes.Thefirstcolumnissimilartocolumn(6)ofTableIII,ourbenchmarkdefinitionofCommonalityinQuotes.Columns(2)to(4)correspondtothefollowingdefinitionsofCommonalityinQuotes:Column(1), ∑ min , / ∈[0,0.5]Column(2), 2 ∑ min , /∑ sum , ∈ 0,0.5 Column(3), 3 ∑ min , / sum , ∈ 0,0.5 ∗ Column(4), 4 3/ 0.5 ∗ ∈ 0,1 where and are thenumber of quotes given to countryaand countryb respectivelybydealerd in agivenmonthtwhile and arethetotalnumberofquotesgivenbyalldealerstocountriesaandbattimet,respectively, andNCD is the number of common dealers giving quotes to a pair of countries. Country‐pair fixedeffects are included in the three specifications. Timeeffects arenot includedgiven that all specifications containmacrocontrols.Standarderrorsinbracketsaredouble‐clusteredatthecountry‐pairandmonthlevel.*,**,and***indicatestatisticalsignificanceatthe10%,5%,and1%levels,respectively.
DepVar:CorrofdailyCDSlog‐returnst
(1) (2) (3) (4)
CommonalityinQuotest‐1 1.087*** 1.032*** 0.031*** 0.586***(Differentvariableforeachcolumn) [0.180] [0.181] [0.005] [0.104]
CorrelationofCDSLogRet.t‐1 0.205* 0.220** 0.144 0.214**[0.109] [0.110] [0.107] [0.105]
ΔLogCounterpartyRiskt‐1 ‐0.071 ‐0.077 ‐0.052 ‐0.078[0.113] [0.117] [0.103] [0.114]
ΔLogCounterpartyRiskt‐1xPeripheral ‐0.114 ‐0.109 ‐0.122* ‐0.106[0.071] [0.072] [0.065] [0.074]
ΔLogEuribor‐OISt‐1 ‐0.072 ‐0.078 ‐0.012 ‐0.074[0.075] [0.077] [0.073] [0.076]
ΔLogEuribor‐OISt‐1xPeripheral 0.163** 0.165** 0.165** 0.165**[0.072] [0.074] [0.066] [0.078]
ECBBondPurchasest‐1 0.003** 0.004** 0.003* 0.003**[0.001] [0.001] [0.001] [0.001]
ECBBondPurchasest‐1xPeripheral ‐0.002** ‐0.002** ‐0.002*** ‐0.003**[0.001] [0.001] [0.001] [0.001]
Corr.CountryBanksCDSLogRet.t‐1 0.005 0.002 ‐0.025 0.002[0.016] [0.017] [0.018] [0.018]
Corr.CountryStockIndexesVola.t‐1 ‐0.012 ‐0.012 ‐0.002 ‐0.014[0.023] [0.024] [0.022] [0.023]
Corr.CDSRelativeBid‐Askt‐1 0.066** 0.067** 0.040 0.067**[0.033] [0.033] [0.031] [0.033]
Abs|DeficittoGDPi‐DeficittoGDPj|t‐1 ‐0.001 ‐0.001 ‐0.001 ‐0.001[0.001] [0.001] [0.001] [0.001]
Abs|DebttoGDPi‐DebttoGDPj|t‐1 ‐0.001** ‐0.001** ‐0.001** ‐0.001**[0.001] [0.001] [0.001] [0.001]
Constant 0.113 0.126 0.183 0.097 [0.116] [0.119] [0.112] [0.132]
Observations 2,370 2,370 2,370 2,370R‐squared 0.345 0.338 0.380 0.345
50
TableXI:CommonalityinQuotesandCDSCorrelationusingDailyData
This table reports panel estimates of daily cross‐sectional regressions forecasting the correlation of intra‐dailySovereignCDSReturns. The specification is similar to column (6) inTable III (replicated here in column (1) forcomparisonpurposes).Theregressionsareatthecountry‐pairlevel(55differentcountry‐pairs),andfortheperiodofJanuary2008toOctober2011,butusingdailydata(964tradingdays).TheindependentvariablesinthistablearesimilartothoseinTableIII,butupdateddaily,exceptECBBondPurchases,Abs|DeficittoGDPi‐DeficittoGDPj|andAbs|DebttoGDPi‐DebttoGDPj|,thatareupdatedlessfrequently(ECBPurchasesweekly,andthoserelatedtoGDP,quarterly).The four specifications fordaily analysis correspond to fourdifferentwaysofdefining thedependentvariable(thedailycorrelationofSovereignCDSReturn)usinghourlychangesinCDS.Column(2)correspondstothespecificationwhere if there is amissing CDS quote for a dealer in a given hour,we impute the previous quote.Column (3) is as column (2), butwe exclude the observations forwhich the hourly CDS return is 0. Column (4)leavesthemissingvalueswithoutreplacementforthepreviousvalue,andcolumn(5)leavesthemissingvalues,andexcludes theobservations forwhich thehourlyCDSreturn is0.Country‐pair fixedeffectsare included inall fivespecifications.Timeeffectsarenotincludedgiventhatallspecificationscontainmacrocontrols.Standarderrorsinbracketsaredouble‐clusteredatthecountry‐pairanddaylevel.*,**,and***indicatestatisticalsignificanceatthe10%,5%,and1%levels,respectively.
Dep.Var:CorrofCDSLogReturnst Monthly DailyFill Fill,nozeros Nofill Nofill,nozeros
(1) (2) (3) (4) (5)
CommonalityinQuotest‐1 1.087*** 0.829*** 0.886*** 0.759*** 0.752***[0.180] [0.080] [0.096] [0.124] [0.129]
CorrelationofCDSLogReturnst‐1 0.205* 0.181*** 0.155*** 0.145*** 0.143***[0.109] [0.016] [0.014] [0.015] [0.015]
ΔLogCounterpartyRiskt‐1 ‐0.071 0.085 0.217* 0.138 0.158[0.113] [0.114] [0.117] [0.155] [0.159]
ΔLogCounterpartyRiskt‐1xPeripheral ‐0.114 ‐0.073 ‐0.142 ‐0.241*** ‐0.248***[0.071] [0.063] [0.100] [0.064] [0.087]
ΔLogEuribor‐OISt‐1 ‐0.072 0.231** 0.253*** 0.233** 0.235**[0.075] [0.090] [0.092] [0.093] [0.094]
ΔLogEuribor‐OISt‐1xPeripheral 0.163** 0.013 ‐0.002 ‐0.021 ‐0.044[0.072] [0.077] [0.080] [0.092] [0.096]
ECBBondPurchasest‐1 0.003** 0.012*** 0.012*** 0.008*** 0.008***[0.001] [0.002] [0.002] [0.002] [0.002]
ECBBondPurchasest‐1xPeripheral ‐0.002** ‐0.007*** ‐0.007*** ‐0.006** ‐0.006**[0.001] [0.002] [0.002] [0.003] [0.003]
Corr.CountryBanksCDSLogRet.t‐1 0.005 0.037*** 0.044*** 0.049*** 0.049***[0.016] [0.008] [0.009] [0.009] [0.009]
Corr.CountryStockIndexesVola.t‐1 ‐0.012 ‐0.005 ‐0.009 ‐0.013 ‐0.013[0.023] [0.014] [0.014] [0.014] [0.014]
Corr.CDSRelativeBid‐Askt‐1 0.066** ‐0.000 ‐0.003 ‐0.009 ‐0.009[0.033] [0.007] [0.007] [0.007] [0.007]
Abs|DeficittoGDPi‐DeficittoGDPj|t‐1 ‐0.001 ‐0.001 ‐0.001 ‐0.001 ‐0.001[0.001] [0.001] [0.001] [0.001] [0.001]
Abs|DebttoGDPi‐DebttoGDPj|t‐1 ‐0.001** ‐0.001 ‐0.001 ‐0.001 ‐0.001[0.001] [0.001] [0.001] [0.002] [0.002]
Constant 0.113 ‐0.013 ‐0.025 0.013 0.023 [0.116] [0.099] [0.111] [0.127] [0.131]
Observations 2,370 35,220 34,101 32,798 32,440R‐squared 0.345 0.134 0.108 0.094 0.091
51
Figure1:ComovementsinSovereignCDS
ThisfiguredepictsthecomovementsinsovereignCDSlog‐returns .Itrepresentsthemonthlycorrelationof
dailysovereignCDSlogreturnsforthe11EMUcountriesconsidered(i.e.,55differentcountry‐pairs)fortheperiodofJanuary2008toNovember2011.Thechartshowsthemediancorrelation(dashedline),togetherwiththeir5thand95thpercentile(shadedarea).
-.5
0.5
1
Jan2008 Jan2009 Jan2010 Jan2011 Jan2012
Co-Movements in Sovereign CDS
52
Figure2:CommonalitiesinQuotes
Thisfiguredepictsthecommonalitiesinquotes definedonthebasisofEquation(1).Itreferstothequotes
given by dealers to both countries in the pair. ∑ min , /,where and arethenumberofquotesgiventocountryaandcountrybrespectivelybydealerd
inagivenmonthtwhile and arethetotalnumberofquotesgivenbyalldealerstocountriesaandbattime t, respectively. The chart shows the median correlation (dashed line), together with their 5th and 95thpercentile(shadedarea).
0.1
.2.3
.4.5
Jan2008 Jan2009 Jan2010 Jan2011 Jan2012
Commonalities in Quotes