RESULTADOS Y DISCUSION
3.3. RENDIMIENTOS Y COSTOS
5.1. Alternativebid-askspreadmeasures
Toconfirmthatourfindingisrobusttootherliquiditymeasures,wereplicatetheregressionsbyreplacingtherelative bid-askspreadwiththeeffectivespreadandtherelativeeffectivespreadasthedependentvariables.FollowingGrullonetal. (2004),theeffectivespreadisdefinedastwicethedifferencebetweenthetransactionpriceandthespreadmidpoint,andthe relativeeffectivespreadisdefinedastheeffectivespreaddividedbythemidpointofthespread.Table5providestheresults oftheanalysis.ThechangeintheSVIissignificantlynegativeinbothmodelsIandIIofPanelA.Thisisconsistentwithour mainfindingthatactiveattentionhelpstoimprovestockliquidity.Bothonlinenewscoverageandadvertisingexpenditure arefoundtoreducethebid-askspreadaswell.Next,weusetheturnoverrateasanalternativeliquiditymeasureandrepeat theanalysis(seeDataretal.,1998)15.InPanelB,weshowthatboththeactiveandpassiveattentionmeasuresarepositively
associatedwiththeshareturnover.Thisconfirmsthatourresultsarerobusttovariousspecificationsofliquidity.
5.2. Propensityscorematching
Ourresearchdesignmaybesubjecttoendogeneityconcern.Theconcernstemsfrompossiblereversescausalityor selectioneffectsinthatretailinvestorattentionisnotrandomlyassignedtofirms.Forexample,liquidityshocksmayaffect retailinvestorattention.Inaddition,corporateannouncements,eventsorfirmcharacteristicsmayaffectbothliquiditylevel andretailinvestorattention.Tocontrolfortheendogeneityissueandtodrawcausalinferences,andtoexploreattributors ofsubstantialincreaseinretailinvestorattention,weapplypropensityscorematching(PSM)(RosenbaumandRubin,1983) toreplicateourmaintests.Toimplementtheapproach,wefirstdefinesubstantialincreaseinSVI(SInSVI)isequalto1 ifthechangeinSVIisabovethe10%percentileand0otherwise.Wethenestimatealogitregressiontotheincidenceof substantialincreaseinSVI(SInSVI)basedontraditionalattentionmeasuresandfirmcharacteristicsasshowninTable5D. Then,weconstructaone-to-onenearest-neighbormatchedsamplebasedontheclosestpredictedvalue(propensityscores) fromthelogitregression.Inthisway,eachfirmwithsubstantialincreaseinattention(SInSVI=1)ismatchedwithanother counterpartfirmwithsimilarcharacteristics,butwithoutsubstantialincreaseinattention(SInSVI=0).Itisworthtonote thattheresultsofmodelIthatfirmswithmoreadvertisingcosts,largersize,higheroperatingperformance,higherturnover, andhighervolatility,andfirmslistedinNASDAQaremorelikelytoexperiencesubstantialincreaseinSVI,whereasfirms whichalsoattracthighmediacoveragearelesslikelytoexperiencesuchchange.Suchrandomizedexperimentsamplewith 3002observationsislesssubjecttotheendogeneityissue.AsshowninTable5C,wereplicatethemaintestsbyusingthe matchingsampletogetthemorereliableandrigorousresults.Theresultsremainconsistentinthatsubstantialincreasein SVIimprovesbreadthofownershipandstockliquidity.
Tocontrolforpossibleconfoundingeffectsofcorporateevents,wereplicatethePSMtestbyincorporatingdummy variablescapturingindexinclusion,corporatesplitsordividendpayments.Itisalsoworthtonotethatthesecorporate eventsdonotsignificantlyaffectthechangeinattentionasdocumentedinthefirststepofthePSMapproach.Theresults fromthesecondstepremainconsistent.
5.3. Testforcausality
Tofurtherstrengthenthecausalinferencesoftheresults,wealsoemploytheGrangertestasadditionalrobustness checksbyfollowingthestudiesofthedeterminantsofstockliquidity(e.g.,Roulstone,2003;Chordiaetal.,2005;Goyenko andUkhov,2009)16.TheGrangercausalitytestisusedtodeterminewhetheronetimeseriesisusefulinforecastinganother.
Thelogicisasfollows:supposethatwehavethreetimeseries:thechangeininvestorattention(SVI),theliquidity(RBAS)
15Tradingactivityreflectedbyturnoverrateisaweakmeasureofliquiditybecausetradingvolumecouldbehighwhenliquidityislow(Pastorand Stambaugh,2003).Grullonetal.(2004)findtheturnoverratetobeweaklycorrelatedwithadvertisingexpenditure.
16Forexample,Roulstone(2003)usesGrangercausalitytesttoidentifythecausalrelationshipbetweenanalystfollowingandmarketliquidity.Hisresults
22 R.Ding,W.Hou/Int.Fin.Markets,Inst.andMoney37(2015)12–26
Table5
Robustnesschecks.
PanelA.SVIchangeandrelativespread
Dependentvariables
Effectivespread Relativeeffectivespread
ModelI ModelII ChangeinSVI −0.11**(−2.02) −0.01***(−2.47) Ln(numberofNews) −0.01***(−3.07) −0.001***(−3.87) Ln(advertising) −0.02***(−4.97) −0.001***(−5.46) Ln(firmage) 0.05***(3.87) 0.001***(4.74) ROA 0.11***(4.75) 0.01***(5.22) Ln(firmsize) 0.13***(5.28) 0.01***(5.67) 1/shareprice 0.62***(3.19) 0.03***(3.91) Ln(turnover) 1.00***(5.32) 0.05***(5.8) Ln(returnvolatility) −0.68***(−5.39) −0.03***(−5.79) NASDAQ −0.25***(−5.21) −0.01***(−5.66) Year Y Y Industry Y Y Observation 14,690 14,690 Adj.R2 0.79 0.79
PanelB.SVIchangeandturnover
[1,0] Dependentvariable:turnover
ModelI ModelII ModelIII
ChangeinSVI 0.12***(3.8) 0.12***(3.8) 0.12***(3.8) Ln(numberofnews) 0.01*** 0.01***(3.59) (2.99) Ln(advertising) 0.01*** (10.44) Ln(firmage) −0.06***(−9.47) −0.05***(−9.41) −0.05***(−9.37) ROA −0.06***(−6.71) −0.06***(−6.60) −0.06***(−6.31) Ln(firmsize) −0.08***(−16.87) −0.08***(−15.79) −0.09***(−17.37) 1/shareprice −0.65***(−5.87) −0.68***(−6.00) −0.71***(−6.21) Ln(returnvolatility) 0.65***(62.04) 0.65***(62.01) 0.65***(62.25) NASDAQ 0.27***(20.62) 0.28***(21.13) 0.27***(21.03) Year Y Y Y Industry Y Y Y Observation 14,690 14,690 14,690 Adj.R2 0.483 0.484 0.487
PanelC.SubstantialincreaseinSVIandpropensityscorematching Dependentvariable
SubstantialincreaseinSVI Ln(numberofshareholdersinthousand) Relativespreadin%
ModelI ModelII ModelIII
SInSVI 0.14***(2.59) −0.08**(−2.52) Ln(numberofnews) −0.02***(−2.79) 0.04***(2.64) 0(−0.25) Ln(advertising) 0.01*(1.87) 0.02***(2.76) −0.01*(−1.92) Ln(firmage) 0.03*(1.7) 0.36***(7.51) −0.03*(−1.66) ROA 0.06**(2.06) 0.60***(9.71) −0.01(−0.16) Ln(Firmsize) 0.11***(7.13) 0.92***(31.16) −0.03(−1.54) 1/shareprice −0.87**(−2.01) 15.30***(10.70) 5.09***(5.96) Ln(turnover) 0.06**(2.05) −0.12(−1.56) 0.08**(2.2) Ln(returnvolatility) 0.15***(4.21) −0.36***(−4.53) 2.47***(38.33) NASDAQ 0.30***(7.63) −0.71***(−7.96) −0.13***(−2.65) Year Y Y Y Industry Y Y Y Observation 14690 3002 3002 AdjR2 0.0214 0.4739 0.7873
R.Ding,W.Hou/Int.Fin.Markets,Inst.andMoney37(2015)12–26 23 Table5(Continued)
PanelD.Corporateeventsandtransparency Dependentvariable
Ln(numberofshareholdersinthousand) Relativespreadin%
ModelI ModelII ModelIII ModelIV ModelV
ChangeinSVI 0.21*(1.7) 0.27**(2.07) −0.24***(−3.43) −0.23***(−3.01) −0.23***(−3.28) Ln(numberofnews) 0.04(0.88) 0.03(0.78) 0.01(1.02) 0.01(1.02) 0(0.27) Ln(advertising) 0.01(0.24) 0(0.04) 0(−0.74) 0(−0.94) 0(−0.65) Ln(firmage) 0.27**(2.01) 0.25*(1.96) −0.01(−0.44) 0(−0.21) 0(−0.04) Return −0.20*(−1.75) −0.27**(−2.31) ROA 0.31(1.52) 0.35*(1.75) −0.01(−0.21) 0(−0.12) 0(0.01) Ln(firmsize) 0.87***(10.4) 0.88***(10.28) 0(0.04) 0(−0.04) 0.01(0.45) 1/shareprice 8.13***(3.35) 8.28***(3.42) 2.97***(4.76) 2.87***(4.62) 2.92***(4.6) Ln(turnover) −0.08(−0.47) −0.11(−0.64) 0.12***(2.78) 0.13***(2.77) 0.14***(3.2) Ln(returnvolatility) −0.13(−1.09) −0.15(−1.11) 2.48***(36.71) 2.50***(35.36) 2.44***(34.07) NASDAQ −0.64**(−2.11) −0.64**(−2.05) −0.02(−0.55) −0.02(−0.54) −0.02(−0.4) Event dividend 0.08(0.77) 0.04*(1.89) Eventsplit −0.17(−0.59) −0.14(−1.4) Big4 −0.24(−1.3) Year Y Y Y Y Y Industry Y Y Y Y Y
Adjustclustererrors Y Y Y Y Y
Excludeeventmonths N Y N Y N
Obs 14690 12591 14690 12591 14690
Adj.R2 0.3981 0.3906 0.7937 0.7918 0.7923
PanelAandBofthistablereporttheestimatesfrompanelregressionsrelatingthealternativeliquiditymeasurestoactiveinvestorattention(SVI).The sampleincludestheconstituentsoftheS&P500overaperiodofsixyearsfromJanuary2004toDecember2009.Theeffectivespreadisdefinedastwicethe differencebetweenthetransactionpriceandthespreadmidpoint.Therelativeeffectivespreadisdefinedastheeffectivespreaddividedbythemidpoint ofthespread.ShareturnoverisconstructedfromCRSPasthemonthlyaverageofthesharevolumedividedbythesharesoutstanding.TheSVIisthesearch frequencyofastockticker,obtainedfromGoogleTrends.ThechangeinSVIisthedifferencebetweentheSVIinweektandthemedianvalueoftheSVIduring thepreviouseightweeks.The“numberofnews”istheonlinemediacoveragefromGoogleNews.AdvertisingexpenditureisobtainedfromCompustat.The relativebid-askspreadisthemonthlyaverageoftheratioofthedailyinsidespreadtothemidpointofthedailyinsidespread,obtainedfromCRSP.Firm ageisthenumberofyearsforwhichthefirmhasbeenincludedinCRSP.ReturnonassetsisconstructedfromCompustatastheannualoperatingincome beforedepreciation,scaledbytotalassets.Firmsizeisthemarketcapitalization,calculatedastheproductofthetotalnumberofoutstandingsharesand theannualclosingprice.Returnvolatilityisthemonthlyaverageofthestandarddeviationofdailyreturns,obtainedfromCRSP.NASDAQequals1forfirms listedontheNASDAQ,and0otherwise.
PanelBpresentstheresultsbasedonthepropensityscorematching.SubstantialincreaseinSVI(SInSVI)issetto1ifthechangeinSVIisabovethetop 10percentileand0otherwise.Matchingsampleof3002observationisconstructedbyModel1,with1501firmsastreatmentgroup(SInSVI=1)and1501 firmsascontrolgroup(SInSVI=0).
PanelCpresentstheresultsaftercontrollingforcorporateeventsorfinancialreportingtransparency.Eventsplitthatequals1ifsharesplitshappenedin themonthand0otherwise;similarly,Eventdividendequals1ifthedividenddistributionwasannouncedinthemonthand0otherwise.
*,**and***representsignificanceatthe10%,5%and1%levels.Standarderrorsareadjustedforheteroscedasticity.
andavectorofcontrolvariablesthathavepredictivepowerforliquidity(Control).WefirstusepastvaluesofRBAS(upto previous10weeks),SVIandControltoforecastRBAS(Eq.(4)).Then,weusepastvaluesofSVI,RBASandControlto predictSVI.IftheresultsrejectthehypothesisthatpastvaluesofRBAScanpredictSVI(Eq.(5))butfailtorejectthe hypothesisthatpastvaluesofSVIcanpredictRBAS(Eq.(4)),thisindicatesthatthepastvaluesofSVIprovidestatistically significantinformationaboutRBAS.Thatis,SVIisabletoGranger-causeRBAS.
First,inordertocheckwhetherchangesintheSVIleadtochangesinliquidity,weuseastandardFtesttotestthejoint hypothesisthatı1=ı2=...ı10=0forthefollowingregression:
RBASi,t=˛+ 10 j=1 ˇjRBASi,t−j+ 10 k=1 ıkSVIi,t−k+ Control (4)
Then,wetestthejointhypothesisthat1=2=...10=0forthefollowingregression:
SVIi,t=+ 10 j=1 jRBASi,t−j+ 10 k=1 kSVIi,t−k+ Control (5)
TheuntabulatedresultsshowthehypothesisthatachangeintheSVIdoesnotcauseachangeinliquidityisrejectedat the10%confidencelevel.However,wefailtorejectthehypothesisthatliquiditydoesnotcauseachangeintheSVIata conventionallevel.Thefindingsstayrobusttotheinclusionofhigher-orderlagsofliquidity.Therefore,ourinferenceisthat anincreaseininvestors’attentionpaidtoastock(reflectedbyachangeintheSVI)causesliquiditytoimprove.
24 R.Ding,W.Hou/Int.Fin.Markets,Inst.andMoney37(2015)12–26
5.4. Otherconfoundingfactors
Toshowthatourresultsarenotdrivenbyconfoundingcorporateeventssuchasstocksplitanddividends,weincorporate themasadditionalcontrolvariablesinourtests.ThefirstdummyvariableisEventsplitthatequals1ifsharesplitshappened inthemonthand0otherwise;similarly,Eventdividendequals1ifthedividenddistributionwasannouncedinthemonth and0otherwise.Alternatively,wealsoreplicatethetestsbyexcludingthefirm-monthobservationswiththesecorporate events,andtheresultsarerobust.TheresultsreportedinthecolumnsItoIVinPanelDofTable5remainconsistentinthat investorattentionmeasuredbyChangeinSVIincreasesshareholderbaseandimprovestockliquidity.Theresultsalsohold whenweincorporateothereventssuchasIndex,whichequals1ifthefirmisnewlyincludedinS&P500indexinthemonth and0otherwise.
Aspreviousliteratureshowsthatfinancialreportingtransparencyisassociatedwithstockliquidity(i.e.,LangandMaffett, 2011),wecreateadummyvariabletoproxyforfinancialreportingtransparency.Specifically,Big4issetto1forfirmsaudited byBigfourauditingfirms(namelyKPMG,Ernst&Young,DeloitteToucheTohmatsu,andPricewaterhouseCoopers),and0 otherwise.WeaddthemasadditionalcontrolvariablesintheanalysisasreportedincolumnVofPanelD,andfindthat theresultsremainconsistent.Wealsoconsideralternativemeasuresforfinancialreportingtransparencysuchasgoingon concernauditopinionsfromauditingfirmbutfindthatS&P500firmsinoursampledonotreceivegoingonconcernaudit opinions.
6. Conclusion
The“homebias”literaturesuggeststhatinvestorsareinclinedtoinvestinfirmsthattheyarefamiliarwith.Inorder togetfamiliarwithafirm,investorsneedtoacquirerelevantinformation.Individualinvestors,whoaregenerallyunable toaccessprofessionalinformationvendorssuchasReutersorBloomberg,mayincreasinglyrelyonGoogle,thedominant Internetsearchengine,tosearchforrelevantinformationbeforemakinginvestmentdecisions.Inthispaper,weusesearch frequencydataonS&P500stocksbetweenJanuary2004andDecember2009,providedbyGoogleTrends,asadirectmeasure ofactiveinvestorattention,andexaminetheimpactofthisactiveattentionfromretailinvestorsontheshareholderbaseand stockliquidity.Wefindthisactiveattentionmeasuretobedistinctfrompassivemeasuressuchasthenumberofnewsitems availableontheInternet(basedonGoogleNews;news.google.com)17andadvertisingexpenditure.Despitethepositive
correlationbetweentheactiveandpassiveattentionmeasures,almost95%ofthecross-sectionalvariationintheformer cannotbeexplainedbythelatter.Wefurthershowthattheincreasedinvestorattentionindicatedbythesearchvolume index(SVI)andGoogleNewscontributestoabroadershareholderbase.Furthermore,increasedinvestorattentionleadsto areducedrelativebid-askspreadandahigherturnoverrate.Ourfindingsarerobusttothecontroloffirmcharacteristics suggestedinGrullonetal.(2004),andtoalternativemeasuresofstockliquidity.
Ourstudycontributestotheburgeoningliteratureontheroleofinvestorattentioninthedynamicsofassetpricing. StudiesinthisstreamofliteratureincludeBarberandOdean(2008)oninvestorattentionandindividualinvestors’trading behavior,Yuan(2009)onrecording-breakingeventsrelatedtotheDowindexandfront-pagecoverageinnewspapersas proxiesforinvestorattention,and itsimpactontradingbehaviorandmarketreturns,and Daetal.(2011)oninvestor attentionmeasuredbyGooglesearchfrequency,anditseffectonIPOreturnsandthepricepressurehypothesisproposed byBarberandOdean(2008).Thisstudyalsoextendstheliteratureonthe“investorrecognitionhypothesis”(e.g.,Grullon etal.,2004;FangandPeress,2009).Inmarketswithinformationasymmetry,investorsarelesslikelytopossesstherequired information.Consequently,securitieswithlowerinvestorrecognitionbecomelessliquidandhavetoofferahigherreturn tocompensatefortheir“illiquidity”.Thefactthatasecurityisattractingmoreattentionfrominvestorscanenableittobe “betterrecognized”.Asaresult,stockswithincreasedinvestorattentionbecomemoreliquid.Ourresultsgenerallylend credencetothe“investorrecognition”hypothesis.
Thefindingsofourstudyhaveimplicationsforcompaniesthatwishtopromoteinvestorrecognition.Companiesmay intentionallymakethemselvesmorevisibleontheInternetandespeciallyintheGooglesearchengineinordertoattract theattentionofpotentialinvestors.Ourresultsmightbeofinteresttoparticipantsinthefinancialmarkets(e.g.,liquidity traders),inthattheymaybenefitfromsophisticatedmodelsthatincorporateindividuals’informationacquisitionbehavior intopredictionsofstockliquidity.Finally,ourfindingsmayincentivizeGoogleandothersearchenginecompaniestofurther improvetheirserviceintermsofprovidingmoretimelyandaccuratedataonpublicsearchbehavior,andtoprofitbyselling suchservices.
Ourstudyissubjecttothefollowinglimitations.First,althoughweapplypropensityscorematchingandGrangertest toaddresstheconcernofendogeneityproblem,futureresearchmayattempttoidentitysomeexogenousshiftininvestor’s attentionandfurtherexplorethecasualrelationbetweeninvestors’attentionandstockliquidity.Wenoticethatcorporate eventscouldnottriggersubstantialchangeinretailinvestors’attentioninthatnon-professionalinvestorsdonottendto
17 Yuan(2009),Tetlock(2010)andFangandPeress(2009)useeithertheLexisNexisdatabaseortheDowJonesnewsarchivetodeterminethenumber
ofnewspaperarticlesrelatedtoastock.TheGooglenewschannelincludesnewsfromthemostpopularEnglish-languagenewssites,suchTheNewYork Times,Bloomberg,Reuters,theGuardian,CBSNews,BBCNews,TimesOnline,andCNN,andthusoffersbroadernewscoverage.Wearguethatonlinemedia coverageismoreaccessiblethannewspapercoverageforretailinvestors,whoaremorelikelytosearchforinformationontheInternet.
R.Ding,W.Hou/Int.Fin.Markets,Inst.andMoney37(2015)12–26 25
closelyfollowcorporateannouncements.Weconjecturethatretailinvestorattentioncouldbeaffectedbyproxiesrelated tocoverageonsocialmedia.Second,ouranalysisisexclusivelybuiltonthesampleofS&P500stocks,whicharelargeand well-establishedfirmswithhighvisibility.Itisexpectedthatsmallerandlessrecognizedfirmswouldbenefitmorefrom increasedinvestorattention,thereforeaddingsmallfirmstofuturestudiesmightstrengthentheanalysis.Notethatifthe tickersofsomesmallfirmsarerarelysearchedfor,Googlewillreturnavalueofzero.Third,assuggestedbyDaetal.(2011), SVIlargelyreflectstheattentionofretailinvestorsbecauseinstitutionalinvestorshaveaccesstoprofessionalinformation vendorssuchastheBloombergorReuters.Ingeneralthetradingbehaviorofanaverageretailinvestorislesslikelytohave aremarkableeffectonthestockprice,whichmightplausiblyexplainwhyactiveattentionmeasureofretailinvestorsis insignificantlyrelatedtotheAmihud(2002)illiquiditymeasure.Furtherresearchmayexamineawiderangeofliquidity measures.Finally,themajorityofresearchinthisareaisbasedonUSdata18.AsGooglebecomesanincreasinglyimportant
sourceofinformationforinvestorsaroundtheworld,itmightbeinterestingtoexplorethecapitalmarketconsequencesof investors’activedemandforinformationinothermarkets(e.g.,theUK,otherEuropeancountriesandAsia).Weleavethis forfutureresearch.
AppendixA.
PanelA.Activeattentionandbreadthofownership:univariateanalysis
SVI Difference t-statistics
Low High
All 46.96 94.27 47.31*** 14.38
Small 22.45 24.72 2.27** 2.06
Large 78.86 153.24 74.38*** 11.54
PanelB.Activeattentionandstockliquidity:univariateanalysis
SVI Difference t-statistics
Low high
Allstock 3.08 2.89 −0.19*** 6.38
Small 3.44 3.27 −0.17*** 3.43
Large 2.62 2.56 −0.06* 1.72
PanelC.Firmfixedeffects
Dependentvariable
Ln(numberofshareholdersinthousand) Relativespreadin%
ModelI ModelII ModelIII ModelIV
ChangeinSVI 0.01**(2.11) 0.02(0.64) −0.03***(−3.57) −0.24***(−4.44) Ln(numberofNews) 0.01(0.83) 0.01(0.74) 0.04**(2.36) 0.04**(2.53) Ln(advertising) −0.06***(−12.28) −0.06***(−12.2) −0.01(−0.96) −0.01(−1.13) Ln(firmage) −0.01(−0.22) −0.01(−0.27) 0.27***(4.17) 0.27***(4.24) Return 0.05(1.49) 0.05(1.47) ROA 0.08***(4.45) 0.08***(4.4) 0.03(0.74) 0.03(0.8) Ln(Firmsize) 0.02(1.15) 0.02(1.2) −0.14***(−4.88) −0.14***(−4.89) 1/Shareprice 0.95***(6.57) 0.94***(6.53) 2.08***(7.53) 2.11***(7.65) Ln(turnover) 0.10***(7.73) 0.10***(7.75) 0.14***(5.74) 0.14***(5.87) Ln(returnvolatility) −0.04***(−3.91) −0.04***(−3.9) 2.48***(121.89) 2.48***(121.97)
NASDAQ Omitted Omitted Omitted Omitted
Year Y Y Y Y
Industry Y Y Y Y
Firmfixedeffects Y Y Y Y
Obs. 14690 14690 14690 14690
OverallR2 0.0198 0.0209 0.7674 0.7673
PanelAreportsthenumberofshareholders(inthousand)forstockswithlowandhighattention.ThesampleincludestheconstituentsofS&P500over aperiodof6yearsfromJanuary2004toDecember2009.TheattentionismeasuredbySVI,thesearchfrequencyofstocktickerfromGoogleTrends.The firmsarefurtherclassifiedintosmallandlargeonesbasedonthemeanofmarketcapitalization.*,**and***representsignificanceatthe10%,5%and1%
level.
PanelBreportspresetstherelativebid-askspread(in%)forstockswithlowandhighattention.ThesampleincludestheconstituentsofS&P500overa periodof6yearsfromJanuary2004toDecember2009.TheattentionismeasuredbySVI,thesearchfrequencyofstocktickerfromGoogleTrends.The firmsarefurtherclassifiedintosmallandlargeonesbasedonthemeanofmarketcapitalization.*,**and***representsignificanceatthe10%,5%and1%
level.
PanelCreportstheestimatesfrompanelregressionswithfirmfixedeffectsapplied.
18OneexceptionisAouadietal.(2013),whichconstructattentionmeasurewithFrenchinvestor’sonlinesearchbehavior,providedbyGoogle,andshow