MS：災害來臨時的恐慌性拋售：債券和股票市場的證據(jù)

Thisarticlewasdownloadedby:[

]On:12March2024,At:06:22

Publisher:InstituteforOperationsResearchandtheManagementSciences(INFORMS)

INFORMSislocatedinMaryland,USA

ManagementScience

Publicationdetails,includinginstructionsforauthorsandsubscriptioninformation:

PanicSellingWhenDisasterStrikes:EvidenceintheBond

andStockMarkets

ThanhD.Huynh,YingXia

Tocitethisarticle:

ThanhD.Huynh,YingXia(2023)PanicSellingWhenDisasterStrikes:EvidenceintheBondandStockMarkets.ManagementScience69(12):7448-7467.

/10.1287/mnsc.2021.4018

Fulltermsandconditionsofuse:

/Publications/Librarians-Portal/PubsOnLine-Terms-and

-

Conditions

Thisarticlemaybeusedonlyforthepurposesofresearch,teaching,and/orprivatestudy.Commercialuseorsystematicdownloading(byrobotsorotherautomaticprocesses)isprohibitedwithoutexplicitPublisher

approval,unlessotherwisenoted.Formoreinformation,contactpermissions@.

ThePublisherdoesnotwarrantorguaranteethearticle’saccuracy,completeness,merchantability,fitnessforaparticularpurpose,ornon-infringement.Descriptionsof,orreferencesto,productsorpublications,orinclusionofanadvertisementinthisarticle,neitherconstitutesnorimpliesaguarantee,endorsement,or

supportofclaimsmadeofthatproduct,publication,orservice.

Copyright?2021,INFORMS

Pleasescrolldownforarticle—itisonsubsequentpages

>inferms.

With12,500membersfromnearly90countries,INFORMSisthelargestinternationalassociationofoperationsresearch(O.R.)

andanalyticsprofessionalsandstudents.INFORMSprovidesuniquenetworkingandlearningopportunitiesforindividual

professionals,andorganizationsofalltypesandsizes,tobetterunderstandanduseO.R.andanalyticstoolsandmethodsto

transformstrategicvisionsandachievebetteroutcomes.

FormoreinformationonINFORMS,itspublications,membership,ormeetingsvisit

7448

Downloadedfromby[

]on12March2024,at06:22.Forpersonaluseonly,allrightsreserved.

/journal/mnsc

MANAGEMENTSCIENCEVol.69,No.12,December2023,pp.7448–7467ISSN0025-1909(print),ISSN1526-5501(online)

PanicSellingWhenDisasterStrikes:EvidenceintheBondandStockMarkets

ThanhD.Huynh,aYingXiaa

aDepartmentofBankingandFinance,MonashBusinessSchool,MonashUniversity,Victoria3145,Australia

Contact:

thanh.huynh@

,

/0000-0001-7970-1464

(TDH);

ying.xia@

,

/0000-0003-1488-3732

(YX)

Received:March24,2020

Revised:September14,2020

Accepted:January16,2021

PublishedOnlineinArticlesinAdvance:August11,2021

/10.1287/mnsc.2021.4018

Copyright:?2021INFORMS

Abstract.Thisstudyusesdisaggregatedestablishment-leveldatatoidentifyairm’sexpo-suretophysicalclimateriskandexaminesinvestors’reactiontonaturaldisastersinboththeU.S.corporatebondandstockmarkets.Weindthat,whenairmisexposedtodisas-ters,investorsoverreactbydepressingthecurrentbondandstockprices,causingfuturere-turnstobehigher.However,irmswithastrongenvironmentalproileexperiencelowersellingpressureontheirbondsandstocks,althoughtheirfundamentalsweakenedfollow-ingdisasters.Theevidencesuggeststhatcorporateinvestmentinimprovingenvironmen-talproilespaysoffwhenclimatechangeriskismaterialized.

History:ThispaperwasacceptedbyColinMayer,SpecialSectionofManagementScienceonBusiness

andClimateChange.

SupplementalMaterial:Thedatailesandonlineappendixareavailableat

/10.1287/mnsc

.2021.4018

.

Keywords:physicalclimaterisk.naturaldisasters.overreaction.establishment-leveldata

1.Introduction

TheUnitedStatesisamongthetopthreecountrieshitbythegreatestnumberofnaturaldisastersoverthepasttwodecades(TheEconomist

2017

).1

BlackRock(

2019

)predictsthatextremeclimaticeventswillcostatleast1%ofgrossdomesticproduct(GDP)ofmostmetropolitanareasintheUnitedStatesby2060–2080undera“noclimateaction”scenario.HsiangandJina(

2014

)estimatethataheftycyclonecarriesthesameimpactonincomepercapitaasabankingcrisis,anditcanhaveaprolongedin?uenceontheeconomicde-velopmentofacountry.Inthisstudy,welookintothestockandbondmarketsandexaminehowinvestorsreactwhenairmisexposedtonaturaldisasters.

Giventhatstocksandbondsarethetwomajorcom-ponentsofairm’scostofinancing,examiningthemarketreactiontodisasterswilldeepenourunder-standingofthepossiblepenaltythatinvestorsimposeonexposedirms.Thisunderstandingalsoguidesmanagersonthetypeofcorporateinvestmentsthatmighthelpmitigatethesecosts.Forpolicymakers,itisimperativetoexaminewhetherthemarketreactionisconditionaluponirmattributes,becauseitimpliesthatpoliciespromotingeficientpricingofphysicalclimateriskmaynotbehomogenousforallirms.

Despitetheimportanceofthisirm-levelanalysis,thesequestionshavenotbeenfullyexploredinthecli-mateinanceliterature,possiblybecauseoftherestrict-edavailabilityofgranulardetailsonairm’sestablish-mentlocations.Ourstudycontributestothisliterature

byusingcomprehensiveestablishment-leveldataonsubsidiaries,branches,andplantsofU.S.irmsfromtheNationalEstablishmentTime-Series(NETS)data-baseandthecounty-levelinformationondisasterdamagesfromtheSpatialHazardEventsandLossesDatabasefortheUnitedStates(SHELDUS).Airmisdeemedtobeexposedtodisasterswhenitsestablish-mentsarelocatedindisaster-struckcounties.

Usingbothportfolioanalysisandmultivariateregres-sions,weindapositiverelationbetweenairm’sexpo-suretonaturaldisastersanditsfuturestockandbondreturns.Speciically,whenairm’sestablishmentsarelocatedincountieshitbynaturaldisasters,itsfuturemonthlystockreturnsincreaseby14.6basispoints(bps)anditsfuturemonthlybondreturnsincreaseby8.4bps.Themagnitudeiseconomicallymeaningful,becausetheseestimatesareequivalentto14%and16%ofthesamplemeansofstocksandbonds,respectively.Theseresultsarerobusttousingalternativemeasuresofirm-leveleconomicexposureandcontrollingforknownriskcharacteristics,unobservabletime-invariantdifferencesacrossirms,andseasonalityeffects.Moreover,theseef-fectsarestrongerasdisastersaremoresalient.Placebotestsofpseudo-disastersonunexposedirmsshowaninsigniicantrelation,suggestingthatourindingsarenotspuriousresults.

Wefurtherindthat,afteradisasterstrikes,thees-tablishment’ssalesandtheirm’saggregatesalesdecrease.

2

Theseresultsserveasvalidationforouridentiicationofaffectedestablishments,becauseitis

HuynhandXia:PanicSellingWhenDisasterStrikes

ManagementScience,2023,vol.69,no.12,pp.7448–7467,?2021INFORMS7449

Downloadedfromby[

]on12March2024,at06:22.Forpersonaluseonly,allrightsreserved.

successfulincapturingdisasterimpactsonirmfun-damentals.Theyalsoindicatethatbothbehavioraltheory(investors’overreactiontodisasters;DeBondtandThaler

1985

)andrationalexpectation(anincreaseinriskbecauseoflossesinirmfundamentals)couldexplainourheadlineindings.Wethereforepresentseveralindings,which,takentogether,leantowardthebehavioralexplanation.

Acentralpredictionoftheoverreactionhypothesisisthatinvestorsoverreacttodisastersbydepressingthecurrentpricesofexposedirms’stocksandbonds,causingthecontemporaneousrelationbetweenre-turnsanddisasterstobenegative.Aspricesdropbe-lowfundamentalvalues,theywillbounceback.Weindeedindthatcontemporaneousreturnsonbondsandstocksaresigniicantlylowerwhentheirmisex-posedtodisasters.Asaconsequenceofthispricepressure,weobserveanincreaseinreturnsonstocksandbondsofexposedirmsafterward.

Wenextuseinsidertradingactivitiestoexaminewhetherpricechangesarepurelydrivenbythereduc-tionofirmfundamentals.Ifinvestorsoverreacttodisastersbyoversellingexposedirms’stocks,thenin-siderscouldbebetteroffpurchasingtheirownstocks,becausestockpriceswerebelowtheirfundamentalvalues(Seyhun

1990

,RozeffandZaman

1998

).Wein-deedindthatinsiders’purchasesincrease,consistentwiththenotionthattheseinsidersbelievetheirownirms’sharepricesareoverlydepressed.

Totheextentthatattentiontoclimatechangein?uen-cesinvestors’reactiontodisasters,weexpectthatourre-sultsaremorepronouncedinrecentyearswheninvest-ors’concernsaboutclimatechangeriskheightened(Kruegeretal.

2020

).UsingthepublicationoftheSternReviewin2006asaneventthatarousedpublicattentiontoclimatechange(Painter

2020

),weindthattheeffectofdisastersissigniicantonlyafter2006.

Arelatedquestioniswhyinvestorsaresurprisedbynaturaldisasters.Afterall,someregionsintheUnitedStatesaremorevulnerabletocertainpromi-nentdisastersthanotherareas.Traditionalassetpric-ingtheorypositsthatphysicalclimateriskcouldbediversiiedinamarketportfolioandthus,thepricesofexposedirms(and,inturn,theircostofcapital)shouldnotchangefollowingadisaster.Wecontendthatthisnullhypothesisisnotsupportedempirically,becausenaturaldisasterscannotbeperfectlypre-dicted(Rehseetal.

2019

)and,asclimatescientistsshow,climatechangehasfurtherexacerbatedanypre-dictability.

3

Withourdetaileddataonirmgeogra-phy,weshowinSection

3

thatthereisalargedegreeofuncertaintyinthedistributionofbothprominentandnonprominentdisastersinagivenarea(e.g.,un-certaintyaboutthelikelihoodofoccurring,timing,andthepotentialdollardamages).Therandomnessofdisasterdistributionssuggeststhatgeographic

diversiicationofbusinesslocationscouldnothelpairmtoavoiddisasterscompletely,becausethereisachancethattheirm’sestablishmentswillbehitbydisastersatsomepointintimeinthefuture.

Toexaminehowirmscanmitigatethesellingpres-sureontheirstocksandbondswhentheyareexposedtodisasters,weexplorewhetherinvestors’reactioniscondi-tionalonairm’senvironmentalproile.Weindthatthesellingpressureislesspronouncedamongirmswithahighenvironmentalscore(EScore),despitethefactthatbothhigh-andlow-EScoreirmssufferedanequallossinestablishmentsalesafterdisasters.

4

Consistently,irmswithmoresustainability-orientedandlong-terminstitu-tionalinvestorsalsoexperiencelowersellingpressurefol-lowingdisasters.Theseresultsareconsistentwiththeno-tionthat,becauseinvestorshavenonpecuniarymotiveswheninvestinginhigh-EScoreirms(Bakeretal.

2018,

HartzmarkandSussman

2019

,Barberetal.

2021

),theyimposealowerpenaltyontheseirms’stocksandbondswhentheirmsareexposedtodisasters.Theseindingsalsohaveimportantimplicationsforpolicymakers,whowishtoemphasizethebeneitsofcorporateinvestmentinenvironmentalproiles:itpaysoffwhentheconse-quencesofclimatechangearematerialized.

Inthenextsection,weoutlineourcontributionstotheclimateinanceliteratureanddevelopourhypoth-eses.Section

3

describesourdataandsampleselec-tion.InSection

4

,wereportthemainempiricalre-sults,robustnessanalyses,andtestsofalternativeexplanations.WeconcludeinSection

5

.

2.RelatedLiteratureandHypotheses

2.1.ABriefReviewofRelatedLiteratureandContributions

Ourstudycontributestoaburgeoningclimateinanceliteraturethatexaminestheeffectsofclimatechangerisksonirms,andinparticular,theassetpricingimpli-cationsoftheserisks.Astrandofthisliteraturestudiescarbonrisk,socialcostsofcarbonemissions,andthereturnpremiumofcarbon-intensiveirms(Barnettetal.

2020

;BoltonandKacperczyk

2020

,

2021

;Choietal.

2020

;Hsuetal.

2020

;Ilhanetal.

2021

).Otherstudiesexaminetheeffectsofclimateregulatoryriskandtherelationbetweenirmriskandenvironment,social,andgovernance(ESG)(deGreiffetal.

2018

,Dunnetal.

2018

,Hoepneretal.

2020

,Seltzeretal.

2020

).Demandsforhedgingagainstclimateriskcanalsoaffectassetprices(Engleetal.

2020

,HuynhandXia

2021

).

Ourstudyisrelatedtothestrandofclimateinanceliteraturethatexaminesphysicalclimateriskintheformofextremeclimaticevents.Hongetal.(

2019

)indthatinternationalmarketsunderreacttothelong-termtrendsindroughts.Aloketal.(

2020

)showthatmutualfundmanagerslocatednearadisasterzoneunder-weightdisaster-pronestocks.Otherstudiesexamine

HuynhandXia:PanicSellingWhenDisasterStrikes

7450ManagementScience,2023,vol.69,no.12,pp.7448–7467,?2021INFORMS

Downloadedfromby[

]on12March2024,at06:22.Forpersonaluseonly,allrightsreserved.

theeffectsofextremeclimaticeventsonirms’draw-ingofbankcreditlines(Brownetal.

2021

),insuranceirms(MassaandZhang

2021

),corporatecashhold-ings(DessaintandMatray

2017

),manufacturingirms’performance(Hsuetal.

2018

),andsupplychains(Bar-rotandSauvagnat

2016

)

.5

Althoughtheliteraturehasstudiedthepricingofcli-materiskinvariousassetclasses,thereisadearthofresearchthatestimatestheeffectsofairm’sdirectex-posuretodisastersonbothstockandcorporatebondreturns,whicharethetwomajorcomponentsoftheirm’scapitalstructure.

6

Ourstudyaimstoillthisvoidbyusingacomprehensivedatasetonestablish-ment-levellocationstoidentifyairm’sexposuretonaturaldisastersatthecountylevel.Thisidentiicationisimportantbecauseairm’smainbusinessoperationstypicallyhappenatestablishments.Becausestocksandcorporatebondsareissuedbytheexposedirm,ifdis-astersaffecttheirm,thenbothstockandbondpricescouldreacttotheseimpacts.Furthermore,bondsaremainlyheldbyinstitutions(Baietal.

2019

),whoarein-creasinglyconcernedaboutclimaterisk(Ilhanetal.

2020

,Kruegeretal.

2020

).Totheextentthatinstitu-tionsdrivethemarketreactiontodisasters,weshouldexpectthattheeffectisalsopresentinbonds,therebyofferinganotherlaboratorytotestourhypothesis.

Wealsocontributetotheliteraturebyshowingthatexposedirmswithhigh-EScoreorhighsustainability-orientedinstitutionalownershipexperiencelowersellingpressure.Incontrast,thesellingpressureisgreaterforexposedirmswithhightotalinstitutionalownership.Theseresultsdeepenourunderstandingoftheroleofi-nancialinstitutionsinpricingphysicalclimaterisk.

2.2.Hypotheses

Ourstudyexaminestheeffectofdisastersonirms’stockandbondreturns.Becausedisasterscouldabruptlyaffectairm’sbusinessoperations,thead-verseimpactofdisasterscouldcauseinvestorstocom-mandahigherexpectedreturnonexposedirms(therisk-compensationhypothesis).Alternatively,invest-orscouldoverreacttothedisasterimpactbydepress-ingtheexposedirm’sstockandbondprices.Aspricesdropbelowthefundamentalvalue,theyreboundinthefollowingmonth(theoverreactionhypothesis).Bothhypothesesleadtothefollowingprediction.

Hypothesis1.Therelationbetweendisastersandone-month-aheadreturnsonstocksandbondsofexposed?rmsispositive.

Thetwoexplanations(overreactionandrationalrisk-compensationhypotheses)arenotnecessarilymutuallyexclusive.Ournexthypothesesattempttoprovideindicationsastowhichexplanationislikelytoplayalargerrole.Thetrueimpactofdisasters(thelossoffundamentalvaluesbecauseofdisasters)

cannotbeimmediatelyobserved.However,asCovalandStafford(

2007

)andotherssuggest,theexpostde-viationbetweenpricesandfundamentalvaluescanprovideuswithanindicationofoverreaction(i.e.,sig-niicantreturnreversalsafterdisasters).IfHypothesis

1

isdrivenbyinvestors’overreaction,thenweexpectthatthecontemporaneousreturnontheexposedirminthemonthofdisasterswillbelower,asitre?ectstheoversellingactivityofinvestors(DeBondtandThaler

1985

).Wethusstatethesecondhypothesis,asfollows.

Hypothesis2.Ifinvestorsoverreacttodisasters,thenthecontemporaneousrelationbetweenstock(bond)returnsanddisastersisnegative.

Wealsoexaminethelong-runreturndifferentialbe-tweenexposedandunexposedirms.IftherelationinHypothesis

1

continuesinthelongrun,thenitsuggeststhatinvestorsupdatetheirbeliefabouttheclimateriskofexposedirms,causingthepremiumonexposedirmstobehigherthanunexposedirms.Suchacontin-uationinreturndifferentialwouldbemoreconsistentwiththerisk-compensationhypothesis.

Anotherwaytodeterminewhetherpricesarede-pressedbeyondthefundamentalvalueistoexamineinsidertradingactivities.RozeffandZaman(

1998

)andLou(

2014

)positthat,becauseinsidershavesupe-riorknowledgeabouttheirownirms,insidertradingrevealswhethertheseinsidersbelievetheirstockpri-cesareundervaluedwithouttheneedtomeasurethelosstofundamentalsbecauseofdisasters.Speciically,ifstockpricesareoverlydepressed,theninsidersarebetteroffpurchasingtheirownsharesandtakead-vantageofthetemporarymisvaluation(Hongetal.

2008

).Ontheotherhand,ifinsidersbelievethatstockpricesjustiiablyre?ectthelossofirmfundamentalsbecauseofdisasterimpacts,thenwedonotexpecttoseeachangeintheirtradingactivity.

Hypothesis3.Ifinvestorsoverreacttodisasters,theninsiders’sharepurchasesincreasewhentheir?rmisexposedtodisasters.

Theeffectofdisastersonreturnscoulddependoninvestors’attentiontoclimatechange.Priorstudies(Kruegeretal.

2020

,SchlenkerandTaylor

2021

)sug-gestthatinvestorsbegantopayattentiontoclimatechangeriskssincetheearly2000s.Wethereforeexpectthattheeffectofdisasterswouldbemorepronouncedinthelateryearsofoursampleperiod.Painter(

2020

)suggeststhatthepublicationoftheSternReviewin2006isagoodcandidateeventtotestthisattentionef-fect.TheSternReviewwastheirstofitskindtoquantifythecoststoaddressclimatechangeanditsimpactontheglobaleconomy(Stern

2007

).Itattractedextensivemediacoveragearoundtheworldandarousedpublicattentiontoglobalwarming.Indeed,

HuynhandXia:PanicSellingWhenDisasterStrikes

ManagementScience,2023,vol.69,no.12,pp.7448–7467,?2021INFORMS7451

Downloadedfromby[

]on12March2024,at06:22.Forpersonaluseonly,allrightsreserved.

Painter(

2020

)indsthatGooglesearchvolumeforcli-matechangeincreasedsigniicantlyafterthereleaseoftheSternReviewandthatthepricingofclimateriskinthemunicipalbondmarketissigniicantonlyafter2006.Assuch,weconjecturethatinvestors’over-reactiontodisasterswouldbemorepronouncedafter2006.If,however,rationalinvestorsdominatethepricingofdisasters,thentherisk-compensationhypothesiswouldpredicttheeffectofdisastersonre-turnstobesigniicantevenbefore2006.

Hypothesis4.Theeffectofdisastersonstockandbondre-turnsismorepronouncedafter2006.

Takentogether,Hypotheses

2

–

4

,ifsupported,sug-gestthattheeffectofdisastersonstockandbondre-turnsisnotpurelydrivenbytheadverseimpactofdisastersonirms’fundamentals.

Investorshaveheterogenouspreferencestowardsustainability.HongandKacperczyk(

2009

)showthatinvestorsstayawayfrom“sin”stocks.Bakeretal.(

2018

)indthatinvestorsarewillingtoacceptlowerre-turnsongreenbonds.HartzmarkandSussman(

2019

)indthatlow-sustainabilityfundsexperiencesignii-cantout?ows,whereashigh-sustainabilityfundsre-ceivelargein?owseventhoughtheydonotgeneratehigherreturns,suggestingthatinvestorspreferhigh-sustainabilityfundsfornonpecuniaryreasons.Theseindingsarealsoechoedbyanumberofstudies(Linsetal.

2017

,RiedlandSmeets

2017

,Albuquerqueetal.

2019

,Chowdhryetal.

2019

,Barberetal.

2021

).Fur-thermore,institutionalinvestorswhoareconcernedaboutclimaterisktendtobelong-terminvestorsandsustainability-orientedinvestors(GibsonandKrueger

2018

,Kruegeretal.

2020

).Takentogether,priorre-searchsuggeststhat,ifairmhasastrongenviron-mentalproile,moresustainability-orientedinvestors,andmorelong-terminvestors,itsstockandbondpri-cesarelesslikelytobedepressedafteritishitbydis-asters.Wethereforeproposethefollowinghypothesis.

Hypothesis5.Theeffectofdisastersonreturnsisstron-geramong?rmswithlowenvironmentalperformance,?rmswithfewersustainability-orientedinvestors,and?rmswithmoreshort-terminvestors.

3.DataandVariableConstruction3.1.NaturalDisastersandFirms，

EstablishmentData

Wecollectcounty-levelinformationonnaturaldisastereventsintheUnitedStatesfromSHELDUS.Thisdata-basecontainsinformationonthetypeofnaturaldisasterevents,theFederalInformationProcessingStandards(FIPS)codeofaffectedcounties,county-leveldollardamagescausedbyeachtypeofevents,durationofeachtypeofevent,andmonthandyearoftheevent.WefollowBarrotandSauvagnat(

2016

)andselect

disastersthatlastforlessthan30days.Asweknowes-tablishments’counties,wealsoensurethatanestablish-mentisexposedtodisasterbyrequiringthatthetotales-timateddamagetoeachcountyismorethan$10millionin2015constantdollars.Thiscutoffisapproximatelytheminimumaveragedamagecausedbybillion-dollardis-asterstoeachcounty.

7

Weobtaininformationonsubsidiaries,branches,andplantsofmultistateirmsfromtheNETSdatabasebe-tween1990and2015,whichissuppliedbyDunandBradstreet(D&B)andismaintainedbyWallsandAsso-ciates.

8

Foreachestablishment,theNETSdatabasepro-videsauniquepermanentestablishmentidentiier(whichdoesnotchangeevenwhentheestablishmentchangeslocation),historicalinformationonthelocationoftheestablishment(FIPS),sales,andthenumberofemployeesattheestablishmentattheyearlyfrequency.Importantly,theNETSdatabaseisfreefromsurvivor-shipbias.Thatis,ifanestablishmentchangeslocationorwascloseddown,weareabletoobserveitshistoricallocationsthroughoutitslife,theyearofrelocation,anditsirstyearandlastyearofactivebusiness.

WematchestablishmentsintheNETSdatabasewithirmsinCompustatandtheCenterforResearchinSecurityPrices(CRSP)bytheirhistoricallegalnames,supplementedwithcarefulmanualchecking.Asstandardintheliterature,wedonotconsideri-nancialirms(thosewithSICcodes6000-6999)andnoncommonstocks(thosewithCRSPsharecodesdif-ferentfrom10or11).

Ourirstmeasureofairm’sexposuretonaturaldi-sasterisadummyvariable,Disaster,whichisequaltooneifatleastoneoftheirm’sestablishmentsislocatedinacountyhitbynaturaldisastersandzerootherwise.Forthesecondmeasure,weusetheinformationoncounty-leveldisasterdamagesandcomputeairm-levelexposure,Ln(Damage),astheestablishment-weightedaverageoftotaldamagesacrossthecountieswheretheirm’sestablishmentsarelocated,usingestablishments’salesasweight.Ourresultsdonotqualitativelychangewhenweuseestablishmentemploymentasweightorwhenweusea50%/50%combinationofestablishmentsalesandemploymentasweight.

Ourinalsamplecontainsatotalof640,203uniqueestablishments.Onaverage,airminoursamplehas75establishments.Intheonlineappendix,TableIA1,wereporttheyearlyvariationinthepercentageofestablishmentsexposedtonaturaldisastersoutofthetotalnumberofestablishments.

9

Wecanseethatthenumberofexposedestablishmentsvariesoverthesam-pleperiod,reachingatroughof4.8%in1994andapeakof22.9%in1998.

Toobtainasenseofthecross-sectionaldistributionofdisasterimpacts,Figure

1

presentsaheatmapofes-tablishmentsexposedtonaturaldisastersacrossthestates.Itshowsthatdisasterscanaffectany

HuynhandXia:PanicSellingWhenDisasterStrikes

7452ManagementScience,2023,vol.69,no.12,pp.7448–7467,?2021INFORMS

Downloadedfromby[

]on12March2024,at06:22.Forpersonaluseonly,allrightsreserved.

Figure1.(Coloronline)LocationsofEstablishmentsExposedtoNaturalDisastersBetween1990and2015

Notes.This?guredepictstheaveragenumberofestablishmentsthatwereexposedtonaturaldisastersoverthefullsampleperiod.Anestablish-mentisdeemedexposedifitslocalcountyishitbyanaturaldisasterthatcoststhecountyatleast$10millionin2015constantdollars.Ineachyearinagivenstate,wecountthenumberofexposedestablishments.Wethentaketheaveragenumberofexposedestablishmentsacrossyears.Informationonthecounty-levellocationsofestablishmentsisobtainedfromtheNETSdatabase,whereasinformationoncounty-leveldisastersiscollectedfromSHELDUS.

establishmentthroughoutthecontinentalUnitedStatesandnotjustthoseincoastalcounties.Intheon-lineappendix,FigureIA1,weplotsnapshotsofthisheatmap.We?ndthat,althoughCaliforniahasthemostestablishmentsexposedtodisastersoverthefullsampleperiod,disastersdonothitthisstateeveryyear.

10

Giventheunpredictabilitya