Do-It-Yourself medicine? The impact of light cannabis liberalization on prescription drugs

(1)

ContentslistsavailableatScienceDirect

Journal

of

Health

Economics

j o ur na l ho me p a g e:ww w . e l s e v i e r . c o m / l o c a t e / e c o n b a s e

Do-It-Yourself

medicine?

The

impact

of

light

cannabis

liberalization

on

prescription

drugs

Vincenzo

Carrieri

a,b,c

_,

_Leonardo

_Madio

d,e

_,

_Francesco

_Principe

f,∗

a_Department_of_Law,_Economics_and_Sociology,_“Magna_Graecia”_University,_Catanzaro,_Italy b_Institute_of_Labor_Economics_(IZA),_Bonn,_Germany

c_RWI-Research_Network,_Essen,_Germany

d_Toulouse_School_of_Economics,_Universitè_Toulouse₁_Capitole,_Toulouse,_France e_CESifo,_Munich,_Germany

f_Erasmus_School_of_Economics,_Erasmus_University_Rotterdam,_The_Netherlands

a

r

t

i

c

l

e

i

n

f

o

Articlehistory:

Received4February2020

Receivedinrevisedform3June2020 Accepted10August2020

Availableonline6September2020 JELclassiﬁcation: H51 H75 I18 K32 K42 Keywords: Lightcannabis Self-Medication Marijuana Difference-In-Difference Prescriptiondrugs CBD

a

b

s

t

r

a

c

t

Governments worldwide are increasingly concerned about the booming use of CBD (cannabidiol)products.However,weknowlittleabouttheimpactoftheirliberalization. WestudyauniquecaseofunintendedliberalizationofaCBD-basedproduct(lightcannabis) thatoccurredinItalyin2017.Usinguniqueandhigh-frequencydataonprescriptiondrug salesandbyexploitingthestaggeredlocalavailabilityofthenewproductineachItalian province,wedocumentasigniﬁcantsubstitutioneffectbetweenlightcannabisand anx-iolytics,sedatives,opioids,anti-depressantsandanti-psychotics.Resultsareinformative forregulatorsandsuggestthatbansonlightcannabisusewoulddisregardtheneedsof patientstoseekeffectivereliefsoftheirsymptoms.

Introduction

The cannabis market has gained momentum world-wide.Rangingfromrecreationalandmedicalmarijuanato cannabis-derivedproducts-suchasCBD(cannabidiol),a relaxantcompoundofcannabis-itsconsumptionis

boom-∗ Correspondingauthorat:ErasmusSchoolofEconomics,Erasmus Uni-versityRotterdam,BurgemeesterOudlaan50,3062PARotterdam,The Netherlands.

E-mail addresses: vincenzo.carrieri@unicz.it (V. Carrieri), leonardo.madio@tse-fr.eu(L.Madio),principe@ese.eur.nl(F.Principe).

ingaroundtheWorld. InEurope,marijuana isthemost usedillicit drug,withapproximately 20%of individuals aged15–24havinguseditin2018(ECCMDA,2019).1_More

recently,low-strengthversionsofcannabissuchashemp (orindustrialhemp),richinCBD,havealsobecomewidely

1 _In_the_US,_ten_states_and_the_District_of_Columbia_allowed_for

recre-ationaluseofcannabisandmanyothersapprovedmedicalmarijuana laws.Asof2018,Canadapassedalawtoremoverestrictionson produc-tion,possession,andconsumptionofmarijuana.InEurope,instead,most stateshaveastrongstanceagainstbothversionsofcannabis(ECCMDA, 2019).

https://doi.org/10.1016/j.jhealeco.2020.102371

(2)

popularandsoldintheformofherbalcannabis(referredto aslightcannabis,hereafter),lotions,extractsandcandies. AmajorcaseofinterestisItaly,acountrywitha conserva-tiveviewoncannabis,butinwhich“lightcannabisshops” unexpectedlyblossomedduetoaloopholeinthe legisla-tionandreceivedmassivemediaandpoliticalattention.2

Withthe2018FarmBill,hempwasremovedfromthe def-initionofmarijuana oftheControlledSubstancesActand similarproductshavebecomeavailableintheUSaswell, wherethisindustryisexpectedtobeworth$20billionby 2024(BDSAnalytics,2019).

Despitetheconsiderableinterest thattheseproducts havereceived,littleisknownabouttheimpactofthisCBD marketandseveralconcernshavebeenraisedregarding thehealtheffectsoftheiruse.While,aspointedoutbythe NationalAcademiesofSciences,Engineering,andMedicine (2017),CBDmightnotpresentitselfasariskforhuman health,thisiscertainlythecaseofitsmisusetotreat seri-ousdisorders.InMay2019,justbeforelaunchingapublic hearingwithexpertsonCBD,theFoodandDrug Admin-istration(FDA)issuedawarningthat“misleadingandfalse claimsassociatedwithCBDproductsmayleadconsumersto putoffgettingimportantmedicalcare,suchasproper diagno-sis,treatment,andsupportivecare”(FDA,2019).3_Likewise,

duetohighuncertainty, in2019NewYorkCitybanned CBD-derivedfoodanddrinks,whereasseveralstates(i.e., Kansas,Indiana,Louisiana,Texas,andmorerecentlyNorth Carolina)announcedorenforcedabanonlightcannabis (TheNewYorkTimes,2019).Asimilarhardstanceonlight cannabiswasproposedbytheItalianNationalHealth Coun-cil(CSS)in2018.TheCSSexpresseditsconcernaboutthe safetyofproductsbasedoncannabisinﬂorescenceand sug-gestedabanonthecommercializationoftheproduct.In theirstatement,theCSSraisedconcernsaboutthe impos-sibilityof monitoringindividual users,its effects in the short-andlong-term,especiallyforsomevulnerable con-sumers(e.g.,minors,pregnantwomen,patientswithother pathologies).

Thispaperaimstoreachamorein-depth understand-ingofthe“hidden”useoflightcannabisandattemptsto investigatehowpeoplehavereactedtotheintroductionof thisnewandpotentiallyriskyproductinthemarket.On theonehand,asalreadypointedoutbypreviousstudies (e.g.Carrierietal.,2019),lightcannabiscanbeasubstitute forstreetmarijuanaandthusgeneratespilloversonthe ille-galmarket.Ontheotherhand,foritsrelaxanteffects,this productcanberegardedasasubstituteforexistingdrugs andinduceself-medication.

2 _In_May_2019,_the_industry_received_massive_media_and_political

atten-tionwhentheItalianMinistryofInterior,MatteoSalvini,startedacrusade againstlocalretailers.Localretailerswereaccusedofbeing“placesofmass miseducation”andhelpingthetransitiontorealdrugs(Reuters,2019).

3 _In_2019,_the_US_FDA_issued_FAQs_regarding_hemp_and_the_recently

approvedtheFarmBill(2018).Respondingtothequestionn.4“Aside fromEpidiolex,arethereotherCBDdrugproductsthatareFDA-approved? WhatabouttheproductsI’veseeninstoresoronline?”,FDAstatedthat“We areawarethatsomeﬁrmsaremarketingCBDproductstotreatdiseasesor forothertherapeuticuses,andwehaveissuedseveralwarningletterstosuch ﬁrms.”

Toexplorethissecondchannel,weexploitauniquecase ofunintendedliberalizationthatoccurredinItalyin2017 whenlightcannabiswasunexpectedlyallowedto circu-late inthemarket.4 _Moreover,_we _use_a_unique_dataset

thatcombinesthelocalavailabilityofthislightcannabis acrossItalianprovincesandmonthlysalesofalargesetof prescriptiondrugsforwhichCBDisoftenadvertisedtobe effectivebyproducers.Theseincludeopioids,anxiolytics, sedatives,anti-migraines, anti-epileptics,anti-psychotics andanti-depressants.

SeveralfeaturesoftheItaliancaseofliberalizationmake itanidealsettingtoexplorethecausalsubstitutioneffect betweenlightcannabis andprescriptiondrug consump-tion.InDecember2016,theItaliangovernmentpasseda law(Law242/16)tofacilitatethecultivationofindustrial hemp in Italy. Due toa loophole in thelegislation, the lawrendered“notillegal”thelarge-scale commercializa-tionofthecannabisﬂower(lightcannabis)intheabsence of psychotropic effects(0.2−0.6% THC). Giventhe unin-tendedscopeoftheliberalization,theproductwassoldas acollector’sitemandnotasaproductsuitableforhuman consumption,sinceMay2017.Critically,duetoalackof anticipation effects,this policyshock providesa plausi-blyexogenousvariationinthepolicysetting.Indeed,this allowsustotestthesubstitutionbetweenexistingdrugs andlightcannabisintheabsenceofinstitutional adapta-tionandpotentialconfoundingfactors(suchaschangesin nationalorlocalhealthpolicies).

Asecondcritical aspectis thattheliberalizationwas accompanied by territorial heterogeneityin the market availabilityoftheproduct.Thepresenceoflightcannabis shops was primarily driven by the morphological and geographicalconditionsoftheterritorythatmadethe cul-tivationofcannabiscropsmoresuitable(seee.g.,Carrieri et al., 2019). The ﬁrst retailers of light cannabis were thealreadyexistinggrowshopsthat,beforethelawwas enacted,weresellingproductsrelatedtothecultivation ofindustrialhempand,afterthechangeinthelegislation, began sellinglight cannabis exploiting large economies of scope. Subsequently, the commercialization of light cannabis ﬂourishedata differentpace acrossprovinces. Asthemarketavailabilityoftheproductwasnotlinked tothedemandforillegalormedicalmarijuanaaswellas ofpharmaceuticals,themarketentryoftheseretailerscan beregardedasplausiblyexogenousinoursetting.

Toidentifytheeffectofinterest,weexploitthe idiosyn-cratic availability of light cannabis in the 106 NUTS-3 Italianprovincesinastaggereddifference-in-differences (DiD)framework in theperiod oftime surroundingthe approvalofthelaw.Weﬁndthatthelocalavailabilityof lightcannabisledtoasigniﬁcantdecreaseinthenumber ofdispensedboxesofanxiolyticsbyapproximately11.5%,a reductionofdispensedsedativesby10%andareductionof dispensedanti-psychoticsby4.8%.Morenuancedbutstill

4_Another_interesting_case,_which_dealt_with_indoor_{prostitution,}_is

pro-videdbyCunninghamandShah(2018).Theauthorsstudytheeffectsof aRhodeIslandDistrictCourtruling,whichunexpectedlydecriminalized indoorprostitution,onseveraloutcomessuchassexualinfectionsand violence,amongstothers.

(3)

significanteffectsarefoundforanti-epileptics(-1.5%), anti-depressants(-1.2%),opioids(-1.2%),anti-migraines(-1%). Aneventstudyspecificationshowsthatthesubstitution betweenthesepharmaceuticalshadalargereffectstarting fromthethirdmonthaftertheintroductionoftheproduct inthelocalmarketandremainsstatisticallysignificantalso aftersixmonthspost-liberalization.Thisisconsistentwith a learningprocess andthedynamicsofself-medication, whichmaystemfromtheneedtobeawareofthelocal mar-ketavailabilityoftheproduct,experimentation,andfinally, moresubstantialpartialorfullself-medication.This inter-pretation is furthercorroboratedby anecdotalevidence fromGoogleTrends,whichshowsanincreasingnumber ofqueriesonthepotentialclinicaleffectsoflightcannabis afterthepolicy.Thefigureindicatesthatpatientsmight have searchedfor information regarding thenew prod-uctandtheirpotentialrelaxanteffectswhenthisbecame available.

Finally,wefindthatdrugprescriptionsacrossprovinces didnottrenddifferentlyforuptosixmonthspriortothe lightcannabissellerentryintheprovince,thus reinforc-ingouridentificationstrategy.Theseresultsarerobustin anumberofchecks,includingalternativemodel specifica-tions,randomizationtestsbasedonfaketreatmentsand placeboregressionsusingdrugsforwhichanaprioricase ofsubstitutionwithlightcannabiscannotbemade.In addi-tion,wealsoperformadecompositionbasedontherecent contribution ondifference-in-differencemethods in the presenceofastaggeredrolloutofthetreatment,as pro-videdbyGoodman-Bacon(2018).

Our findings contribute toseveral streams of litera-ture.Specifically,theyrelatetostudiesonthesubstitution effectsinducedbymarijuanalegalization.Dinardo(2001) and Crost (2012) used minimum drinking age regula-tionstoshowclearsubstitutionpatternsbetweenalcohol and marijuana. Powell et al. (2018) show that medi-cal marijuana laws, and in particular, the number of marijuanadispensaries,areassociatedwithfeweropioid overdoses.Similarpatternswerealreadydocumentedby Shy(2017),Liangetal.(2018);Chanetal.(2020);Smith (2020)andMcMichaeletal.(2020).BradfordandBradford (2016;2017;2018)studiedhowtheavailabilityofmedical marijuanaimpactedonMedicareandMedicaiddrug pre-scriptionsintheUS,withasignificantreductionforthose relatedtopainrelief,anxiety,nausea,depression,psychosis andsleepdisorders.

Ourcontributiontothisstreamofliteratureis three-fold. First, this represents the ﬁrst-ever evidence from Europeonthesubstitutioneffectsinducedbyacompound ofcannabis,theCBD.Astheproductisincreasingly avail-ableinotherEuropeancountries(e.g.,Belgium,Germany), ourresultsproviderelevantimplicationsforpolicymakers. Second,whilethereisahighdegreeofvariationinmedical andrecreationalmarijuanalawsintheUS-wheresome statesadopteda moreliberal(almostrecreational mari-juana)approach,andothersastricterone(wherepeople needaterminaldiagnosistogetaprescription)-the Ital-ianexperiencehasbeenbothliberalandstrictatthesame time.Moreimportantly,theloopholeinthelegislationwas homogenousintheentirecountry,althoughthelocal avail-abilityoftheproductwasinitiallyheterogeneousbetween

localareas.Arguably,thisrepresentsasuitablesettingto exploremedicalsubstitutionbetweenlightcannabisand severaldifferenttypesof prescriptiondrugs by exploit-ingterritorialheterogeneityinproductavailability.Third, substitutioneffectsarisingfrommedicalmarijuanalaws intheUSrequireadoctor’sprescription.Despitethefact thatthephysician’sinvolvementmightbequitelimited, i.e.,becauseoftheirroleasgatekeeper,thisentails,inany case,akindofsupply-sidedrivensubstitution5_._In_our_case,

suchasubstitutionispurelyconsumer-drivenasitcannot involvetherecommendationofamedicalprofessional.This isbecauselightcannabisissoldasatechnicaland collec-tor’sitemand,hence,virtuallyconsiderednotsuitablefor humanconsumption.

Moreover, our paper adds to the literature on the economics of these (potentially) risky behaviors (for a review, see e.g.,Cawley and Ruhm, 2012).Traditionally, thesebehaviors,andinparticularaddictiveones(suchas heavydrinking, smoking,and drugabuse),have mostly beenconsideredasaconsequenceofimperfectlyrational individual’schoices,time-inconsistentpreferencesor trig-geredbyincompleteinformationregardingpossiblerisks.6

Morerecently,riskybehaviorshavebeenassociatedwith theso-called“deathofdespair”(CaseandDeaton,2015). However, thesemay also emerge as a form of rational self-medicationforsymptoms noteffectively treated by currentlyavailabledrugs.Thisisthenewperspective high-lightedbyDardenand Papageorge(2018),who recently proposedanovelrationalinterpretationthatwas consis-tent witha utility-maximizingframework and tested it empirically.Theyﬁndthattheavailabilityofnewand bet-terdrugsfordepressionhasledtoareductionofformsof self-medication,suchasalcohol,intheUS.Ourstudy con-tributestothisliteraturefromanovelanglebytestingthe effectoftheavailabilityofanewproductsuitablefor self-medicationonexistingdrugconsumption.Theunintended natureofthisliberalizationofferstheuniqueopportunity totestforthepresenceofself-medicationbyexploitingan exogenousvariationintheavailabilityoftheproduct.

Lastly,ouranalysisalsocomplementstheliterature con-cerned with the effect of marijuana liberalization. This hasalreadyshowneffectsonotheroutcomes,e.g.,crime (Carrierietal.,2019;ChangandJacobson,2017;Gavrilova etal.,2017;ChuandTownsend,2018;Dragoneetal.,2019, inter-alia),trafﬁcfatalities(Andersonetal.,2013;Hansen et al., 2020a), teenager use (Wen et al., 2015, Cerven ´yˇ etal.,2017),fertility(Baggioetal.,2020),bodyweightand obesity(Sabiaetal.,2017)butitlacksstudiesondrug con-sumptionandself-medication.

5 _In_the_US,_physicians_in_states_where_there_is_cannabis_{liberalization}

cansignastateformcertifyingthatthepatienthasaqualifyingmedical conditiontobuycannabisatdispensaries.However,theirinvolvementin suggestingacannabis-basedtherapymaybeverylimitedandonlyrelated totheirroleasgatekeepers.

6 _Starting_with_seminal_{contributions}_of_Becker_and_Murphy_(1988),_the

literatureonrationaladdictionhasbeenbroadlytestedondifferent out-comes,e.g.,alcohol(BaltagiandGrifﬁn,2002;Dragone,2009),cigarettes (Chaloupka,1991;Beckeretal.,1994),cocaine(GrossmanandChaloupka, 1998),andopium(VanOurs,1995).

(4)

Thepaper is structured asfollows: In Section2, we presentthepolicyreformthatoccurredinItaly.InSection 3,wediscussourdataandtheidentiﬁcationstrategy.In Section4,wepresentourmainresults,followedbysome sensitivityanalysesandrobustnesschecksoftheseresults inSection5.Intheﬁnalsection,wepresentsome conclud-ingremarks.

Institutionalsetting

Italyhasalonghistoricaltraditionofthecultivationof cannabis.One oftheearliest largecannabis cultivations datesbacktothe1stcenturyA.D.(Mercurietal.,2002). In the 1940s, Italy was the second-largest producer of industrialcannabisworldwide.Despitethisfact,marijuana cultivation,possession,andsaleremainillegal,exceptfor itsindustrialandmedicaluse.

However, in 2016, the government passed Law 242/2016,whichwasintendedtoremovesomerestrictions oncultivation,transformation,and commercializationof industrialhemp,a strainofCannabissativa,widelyused fortextiles, clothing,and food.For itsuse, hempis not classiﬁedasadrugasitis(almost)freeofthe psychoac-tivecannabiscompound(THC).Thelattershouldbekept below0.6,anditisrichinCBD,arelaxantcompound. Fail-ingtoaccountforthislimitwouldimplynarcoticeffects andhencebesubjecttoconﬁscation,closure,andriskof convictionfordrugdealing.In2018,asdemandforCBD productsgrew,Italyhadapproximately4,000haofland dedicatedtohempcultivation,thatis,tentimesmorethan in2013.

The law, however, did not explicitly intervene on thecommercializationof thecannabis ﬂower,leaving a loopholeinthelegislation.Asaresult,asnotexplicitly for-bidden,herbalcannabiswasessentiallyliberalized.From May 2017 onwards, once the inﬂorescence phase was over,severalstartupsexploitedthisgreyandcompletely unregulatedmarketandstartedsellinglightcannabisas a “technical product”, that is, as a collector’s item not meanttobesmokedorconsumed.Thesameretailersalso startedsellingotherCBD-basedproductssuchasoil,leaves, extracts,foodandbeverage.

Interestingly, the local market availability of light cannabisdidnotarisesimultaneouslyinallgeographical areasofthenationalterritory.Duetotheunannounced liberalization,localavailability in theﬁrst months after thepolicymainlyaffectedthoseareaspreviouslyserved bygrowshops,thatis,shopssellingseedsand cannabis-relatedproducts.Asdocumentedelsewhere(e.g.,Carrieri etal.,2019),thesegrowshopsweremainlyconcentrated inthoseareasinwhichindustrialcannabiscultivationwas morelikelyduetothegeographicalandmorphological con-ditionsoftheterritory.Asaresult,theﬁrstretailersselling theproductwerethosegrowshopsthat,beforethepolicy shock,werealreadysupplyingindustrialhempanditems foritscultivationand thatcouldexploitexistingsupply chainsandnetworks.

Fig. 1 depicts this staggered entry into the market during theperiod covered by this study.In May 2017, 22 out of 106 provinces were served by at least one light cannabis retailer. Aftera ﬁrst introductory phase,

light cannabis shops blossomed, thereby becoming a “socialphenomenon”.Tobaccoandherbalistshops, para-pharmacists and automatic machinesbegan selling this product as well and expanded their local coverage. As shownbyFig.1,byFebruary2018,thenumberofprovinces reached87andtheentirecountrywasgraduallycoveredby retailersthroughout2018.In2019,morethan1000shops wereopen.

Interestingly,inMay2019,thesituationinthemarket changed dramatically. The former Minister of the Inte-rior,MatteoSalvini,announcedacrackdownonthisherbal cannabissalesandseveralretailershadtheirproducts con-ﬁscatedasa precaution.Thereason wasthat,inseveral cases,theﬂowerscontainedmoreTHCthanthatallowed by thelawtoavoid being consideredasan intoxicating substance. In turn, this triggered intervention from the police.7_In_the_same_month,_the_Court_of_Cassation,_Italy’s

SupremeCourt,aimedtoclosetheloopholeinthe legisla-tionbydecidingtoimposeabanonthesalesofcannabis derivativessuchas“oil,leaves,inﬂorescences,andresin”. However, the Court’s ruling left a backdoor openfor a mildinterpretationofthelawforthoseCBD-derived prod-uctsfreeofnarcoticeffects,creatingmoreuncertaintyin themarket.Akinprecautions,manycompaniesandshops shut down in June 2019. This paper thus exploits this phenomenonduringtheperiodinwhichthemarketwas entirelyleftunregulated.Fig.2providesatimeframework regardingourstudy,whichcoversmarketentrydatafrom May2017toFebruary2018.

Dataandmethods

Weuseauniquelongitudinaldatasetrecordingmonthly drugsales and mappingthelocalmarket availability of retailerssellinglightcannabisattheprovincelevel(106 NUTS-3provinces)overtheperiodfromJanuary2016to February2018.8_Data_on_these_dispensaries_were_collected

usingtheArchiveInternetWaybackMachineonthewebsites ofthefourmainproducersoflightcannabisin2017.These werethenmatchedwithdataonprescriptiondrugsales obtainedbyFederfarma,theItalianassociationofpharmacy owners.Ourdatatracksthepharmaceuticalexpenditure oftheItalianLocal HealthAuthorities(ASLs)andcovers morethan95%ofItalianpharmacies.AsoneormoreASLs canservea givenprovince,datawereaggregatedatthe provincelevel(106provinces).

Ourdatasettracksall“ClassA”drugsdispensedbythe ItalianNHScontainingactiveingredientsoftenusedtotreat symptomsforwhichmarijuanacanalsoprovidereliefor, at least,relaxanteffects. Speciﬁcally,theserefer to

opi-7_This_was_not_the_case_of_CBD,_for_which_no_limitation_in_its_quantity

permilligramexisted.

8_We_collected _monthly _information_on_the _entry _in _a_market _in

each Italian province by having accessto archived copies of their earlypagesofthefourmainproducersin2017(Easyjoint, Marymoon-light,RealHemp,XXXJoint)usingtheInternetArchiveWaybackMachine https://archive.org/web/.DatawerecollectedmonthlyfromMay2017 (ﬁrstentryinthemarket)andusingthelastaccessiblepageforeach month.Whendatawerenotavailable,theaveragenumberofretailers betweenthemonthswithmissinginformationwasimputed.

(5)

Fig.1.TimingoflocalavailabilityoflightCannabis.

Themapreportsthedifferenttimingoflocalavailabilityoflightcannabisinthe106provincesweconsiderstartingfromMay2017.Dataareretrieved fromtheInternetWaybackMachineArchive.

Fig.2.Timeline.

OurstudycoversmonthlydatafromJanuary2016toFebruary2018.TheLaw(242/2016)wasapprovedonDecember2016.InMay2017,theﬁrstentryin themarketoccurred(seeFig.2).InMay2019,Italy’sHighCourtdecidedaboutthepossibilitytocommercializetheproduct.

(6)

Table1

Descriptivestatistics.

Variable ATC/DDD Description Mean Std.Dev.

Sedatives N05C Monthlynumberofdispensedsedativesandhypnoticdrugsperprovince(boxes) 28.20 109.73

Anxiolytics N05B Monthlynumberofdispensedanxiolyticsperprovince(boxes) 72.52 359.45

Anti-epileptics N05A Monthlynumberofdispensedanti-epilepticsperprovince(boxes) 18,459.67 21,474.39

Opioids N02A Monthlynumberofdispensedopioidsperprovince(boxes) 12,610.91 13,065.20

Anti-migraines N02C Monthlynumberofdispensedanti-migrainesperprovince(boxes) 2,504.26 2,626.09

Anti-psychotics N05A Monthlynumberofdispensedanti-psychoticsperprovince(boxes) 4,802.44 5,903.55

Anti-depressives N06C Monthlynumberofdispensedanti-depressivesperprovince(boxes) 27,198.25 26,290.68

Nr.Observations 106provincesX26months 2756

Note:dataaremadeavailablebyFederfarmaforall“ClasseA”drugsdispensedbytheItalianNHS.SalesareregisteredbasedontheWHOATC/DDD,which classiﬁesdrugsbasedonatechnicalunitofmeasurementcalledtheDeﬁnedDailyDose(DDD).

oids,anxiolytics,hypnoticsandsedatives,anti-migraines,

anti-epileptics, anti-psychotics and anti-depressives, as

standardizedfortheATC/DDDclassiﬁcationoftheWHO

CentreforDrugStatisticsMethodology.Thisclassiﬁcation

standardizesdrugsbasedontheactiveprincipleandorgans

theytreat(AnatomicalTherapeuticChemicalClassiﬁcation

SystemATC)andoftheDeﬁnedDailyDose(DDD).The

lat-teraccountsfor“theaveragemaintenancedoseperdayfora

drugusedforitsmainindicationinadults”.

Importantly,thesedrugs requirea doctor’s

prescrip-tion,whichistypicallyprovidedin-personbythephysician,

andisavailable eitherfreeofcharge orwithaminimal

patientco-payment.9_Moreover,_there_are_two_important

capswhicharedesignedtoregulateindividualdrug con-sumptionand make it strictlyadherent tostandardized therapies,asdefinedbytheATC/DDDsystem.First,thereis acaponthenumberofboxesthatadoctorcanprescribe, whichistypically2forthedrugsconsidered.Second,there isacaponthelengthoftheperiodaprescriptioncanbe refilled,whichistypically30days,andthisensures com-pliancewiththedosinginterval.Inourframework,these capsallowustomeasurethesubstitutioneffectsinterms ofDDDsforeachATCand,thus,ruleoutanyintertemporal changesinthenumberofmonthlypackagessold.Finally, ourdatasettracksallprescriptionsthattranslatedintoa finalsaleineachprovince.

Thepharmaceuticalsweconsiderhaveshownpatterns ofsubstitutabilitywithmedicalmarijuana(Bradfordand Bradford, 2016; 2017; 2018), which however presents somedifferenceswithrespecttothelightone.Forinstance, medical marijuana, richin THC, islargely used totreat chronic pain, glaucoma, insomnia and anxiety. Instead, foritsclinicaleffects, CBDis oftenassociatedwith anti-psychotic,analgesic,anti-inﬂammatory,anti-arthritic,and anti-neoplasticpropertiesandisusedtotreat inﬂamma-tions,migraines,depression,andanxiety(Blessingetal., 2015; CIBG, 2018). However, CBD-based products have comeunderscrutinyintheUSformisleadingclaimsmade by some CBD producers relative to its effectiveness in

9 _According_to_Federfarma_(2018),_“Classe_A_“drugs_are_the_most

con-sumeddrugsinItaly,accountingforapproximately52%oftheentire marketforpharmaceuticalproducts(whichalsoincludedietaryproducts, herbs,para-pharmaceuticals,andproductsforhealthandbeautycare). Co-payments,whichdependsontheregionalco-paymentsettingsand individuals’equivalentincome,rangefrom1europerboxto4europer theentireprescription.

treating theabove pathologies. Apartfrom Epidiolex, a drug recentlyapproved by the FDA to treat rare forms of epilepsy, noother drug contains CBD. Opioids were alsoincludedas,accordingtorecentanecdotalevidence, Canadianveteransstartedsubstitutingopioids(alongwith benzodiazepines)withmarijuanaaftertheirintroduction ofmedicalmarijuana inCanadatotreatanxiety, insom-nia and pain relief. Several scholars also documented similar patterns of substitutability with medical mari-juana(e.g.,BradfordandBradford,2016,2018;Shy,2017; Bachhuberetal.,2014).

Table1describesthemainvariablesweuseandshows alargeheterogeneityacrossdrugcategoriesinthe num-ber of dispensed drug packets. On a monthly average, theItalianNHSprovidesreimbursementfor28packetsof sedativesand72boxesofanxiolyticsperprovince.Much highernumbersaredocumentedforanti-epileptics(18,460 packets),antipsychotics(4802packets)andanti-migraines (2504boxes).With27,198packetssoldonaverage, anti-depressantsrepresentthedrugwiththelargestsalesinour sample.ISTAT, theItalianNationalInstituteofStatistics, estimatesthatmorethan2.4millionItalianssufferedfrom mental healthproblemsin 2015,and approximately1.3 millionpresentedsymptomsofdepression(ISTAT,2018).

Interestingly,opioids,whicharewidelyusedintheUS andCanadaforacuteandchronicpainreliefbutwhichhave severesideeffectssuchasdependencyandsedation,show lowernumbers,accountingforanaveragemonthlysaleat theprovincelevelof12,610packets.Althoughthenumber ofprescriptionsincreasedinrecentyears,thesenumbers depictthegenerallevelofskepticismamongItalians con-cerningthisdrug.

Finally,bymatchingthisinformationwiththose avail-able from light cannabis local retailers, we obtained a balancedpanelwithatotalof2756province-month obser-vations.

Identiﬁcationstrategy

Inordertoidentifythecausaleffectoflightcannabis on the prescriptions of drugs, we employ a staggered DiD,which exploitstheidiosyncraticavailabilityoflight cannabis in a given province.10 _Thus, _our_{identiﬁcation}

10_In_text_context_of_substances,_a_similar_{identiﬁcation}_strategy_based

onastaggeredrolloutoftheopiumpolicyintheJavaduringtheperiod 1875–1904wasprovidedbyVanLuijkandVanOurs(2001).Similarly,

(7)

Table2

Difference-in-differencesregression.

(1) (2) (3) (4) (5) (6) (7) (8)

All Sedatives Anxiolytics Anti-epileptics Opioids Anti-migraines Anti-psychotics Anti-depressives

DiD −0.016*** −0.095* −0.114** −0.015*** −0.012** −0.009* −0.048** −0.012***

0.004 0.053 0.055 0.004 0.005 0.005 0.018 0.003

Controls yes yes yes yes yes yes yes yes

YearFE yes yes yes yes yes yes yes yes

MonthFE yes yes yes yes yes yes yes yes

ProvinceFE yes yes yes yes yes yes yes yes

N 2756 2756 2756 2756 2756 2756 2756 2756

Log-transformationofthedependentvariable.S.E.clusteredattheprovince-levelinitalics.***,**,*indicatestatisticalsigniﬁcanceat1%,5%,and10% respectively.

reliesbothonthestaggeredtimingoftheproduct

avail-abilityandtheprovinceswithoutanyretailerasthecontrol

group.11_More_formally,_we_estimate_the_following

equa-tion: Yit=˛+

k

ˇkEntryik+ıXit+i+t+εit

where Y_it is thenumber of packets of dispensed drugs (i.e.,opioids, anxiolytics,sedatives, anti-migraines, anti-epileptics, anti-depressants and anti-psychotics) reim-bursedbyNHSatthetimetintheprovincei,Entryisan indicatorthattakesvalue1ifatleastacannabisretailerhas enteredinallperiodsk<tinprovincei.␥and␮areprovince andtime(monthandyear)ﬁxedeffects,␧representsthe errorterm.Xitisavectorofcontrolsforprovince popula-tionsizeanddensityandadummyforthepost-Mayperiod totakeintoaccounteventualchangesthatoccurredatthe nationallevelaftertheunintendedliberalization.Our coef-ﬁcientofinterestis␤,whichcapturesthemonthlychange inthesalesofdispenseddrugsduetothelocalavailability oflightcannabis.

Thecredibilityofouridentiﬁcationstrategyreliesonthe naturalexperimentwhichcharacterizedthepolicy.As dis-cussed,thepolicywasunannouncedandconcernedhemp, a product that,when soldas light herbalcannabis, dif-fersinitscompositionfrombothrecreationalandmedical marijuana.Thus,theentryoflightcannabisretailersina localmarketcanberegardedasplausiblyexogenousasit wasnotlinkedtothedemandforillegalormedical mari-juanaaswellaspharmaceuticals.Moreover,thepossibility of endogenous entryis essentiallyruled out forseveral reasons.First, theprecisenature oftheunintended lib-eralizationprocessrenderedanyanticipatoryeffecttobe implausible.ThelawwasapprovedinDecember2016and, afteranearlycultivationphase,theproductwasﬁrst mar-ketedinMay2017.s,thelocalavailabilityduringtheperiod underinvestigationmainlydependedonthe

geographi-Dragoneetal.(2018)exploitthestaggeredadoptionoftherecreational marijuanalawsinthestatesofWashingtonandOregontostudyitseffects oncrime.

11_In_our_{identiﬁcation} _strategy, _we_cannot_rule _out_{cross-province}

movementstobuylightcannabis,asrecentlyshownbyHansenetal. (2020b).However,intheperiodweconsider,itisimplausiblethatpeople managedtoknowaboutthelocalavailabilityoflightcannabisina differ-entprovinceandsystematicallymovetothatprovinceaccordinglytobuy theproduct.Asaresult,inoursetting,cross-provinceshoppingdoesnot representaconcreteissue.

calpresenceofgrow-shopsbeforethepolicychange.The latterareretailersspecializedinindustrialhemp,which weretheﬁrsttoselllightcannabisaftertheliberalization byexploiting largeeconomies of scopeand their exist-ingsupplychains(see,forexample,Carrierietal.,2019). Finally,wemakeuseofhigh-frequencydatatofocusona narrowtimewindowsurroundingtheimplementationof the(unintended)policy(May2017–February2018).This allowsustoruleoutpotentialchangesinnationalandlocal healthpoliciesconcerningprescribeddrugsand system-aticchangesintheprescriptionchoicesofmedicaldoctors. Thisisbecauseanychangerequirestimetobefully opera-tionalandextensivelyincludedinmedicalguidelinesand protocols.

Moreformally,ourempiricalanalysisreliesonthe clas-sicalDiD’sidentifyingassumptionsinthepre-liberalization periods,thatis,theexistenceofacommontrendindrug prescriptions. In our setting, it implies assuming that thoseprovincesexperiencingdifferenttimingsinthelocal availabilityoflightcannabis(treatmentgroup)andthose provincesneverservedbycannabisretailersintheperiod weconsider(controlgroup)shouldhaveobservedthesame pre-policytrendsforalldrugcategories.InSection5,we ﬁndstrongsupportforthis hypothesis byperforminga visualinspectionofcommontrendsandmoreformaltests suchasplaceboregressionsandfalsiﬁcationtestsbasedon alternativeapproachestostatisticalinference.

Results

OurresultsarereportedinTable2forall pharmaceu-ticalsforwhichmedicalmarijuanacanbeconsideredasa substituteoradjuvanttherapy.Foreaseofinterpretation, thedependentvariableisexpressedinlogs.Thisallowsus tointerprettheDID-coefﬁcientastheaveragepercentage changeinthemonthlynumberofdispenseddrugsresulting fromthelocalavailabilityoflightcannabis.

Foralldrugcategories,wedocumentasignificantand negativeeffect.Specifically,asthemarketavailabilityof lightcannabisbecamepossibleduetotheentryofatleast oneretailerinagivenprovince,thenumberofdispensed drug sales fell by approximately 1.6%, on average. The extentofthisreductionrevealsa considerabledegreeof heterogeneity.Theboxesofanxiolyticsprescribedby doc-tors and soldby pharmacies significantly decreased by 11.4%,thesedativesconsumptiondecreased by approxi-mately10%,whilethenumberofanti-psychoticsdecreased

(8)

by4.8%.These drugsaccount forthe mostconsiderable reductions.Interestingly,thesearealsothetypeofdrugs forwhichCBD–butnotlightcannabisitself-isrecognized oradvertisedas havinga clinicaleffect,that is,totreat anxietyandpsychosis(Blessingetal.,2015;CIBG,2018). Thisisintuitivelyexplainedbytherelaxantpropertiesof CBD,whichisoftenusedtotreatsleepdisorders. More-over,thelargecoefﬁcientsthatweobserveforsedatives andanxiolyticscanbeexplainedbythemarketing strate-giesforcannabisandCBD-relatedproductssold,whichare typicallyadvertisedforitsrelaxanteffects.These market-ingstrategiesfueledagovernmentinvestigationintheUS. Thisresultmaysuggestthatpatientsdecidedto(partially) abandonaconventionalmedicaltreatmentandswitched tolightcannabistotreatsymptomssuchasanxietyand sleepdisorders.

Morerelativelynuancedyet,signiﬁcantreductionsare foundforotherpharmaceuticals,whichappeartooffera morechronictherapy.Theseﬁndingsareconsistentwitha self-medicationinterpretationalsoforthesedrugs.Thereis amildaveragemonthlyreductioninthenumberofpackets foranti-epileptics(-1.5%),anti-depressants(-1.2%),opioids (-1.2%)andanti-migraines(approximately-1%).Theseare alldrugsrequiringaconstantandconsistenttherapy,often prescribedbyspecialists,andforwhichtheswitchingto an“alternativetherapy”basedonself-medicationmaybe moreproblematic, especially forrisk-averse individuals. Still,despitetheshorttimewindow,localmarket availabil-ityoflightcannabisandothercannabis-derivedproducts ledtoareductioninprescriptiondrugsbutwithalower intensity.

Moreover, it is noteworthy that opioids, anti-depressants and anti-epileptics are all pharmaceuticals thatshowseveresideeffectsandcanbeassociatedwitha socialstigma.InItaly,opioidsarealsogenerallyless pre-scribedthanintheUS,whereunregulatedprescriptionled toanopioidcrisis(CaseandDeaton,2015).Indeed,some patientsmayhaveseeninthisnewproductthepossibility toseek(partial)relieforimprovetheirqualityoflife,for instance,by makinguse ofa mostly accessibleproduct which does not require any medical prescriptions and whichisperceivedtobeofregularuse(e.g.,oils,essences, orevensimilartorecreationalyetillegalcannabis).

Aninterpretationconsistentwiththeself-medication hypothesis is also anecdotally supported by looking at onlineinformationseeking.Fig.3providesfurthersupport fortheuseoflightcannabisasaformofself-medication usingGoogleTrendsstatistics.Itshowsthatboththe gen-eralinterest intheproduct and itsusetotreat anxiety disorderspeakedinthemonthoftheintroductionofthe productintotheItalianmarket(May2017)andremained atahigherlevelinthesubsequentperiod.Thissupports theideathatpeoplewereseekinginformationonlinefor treatingtheirsymptomsbeforeactuallyswitchingtothe newproduct.

Onecannotethatourresultssharesomesimilarities withthosefoundbyBradfordandBradford(2016;2017). However,therearerelevant differences.Theyshoweda largerreductionindrugprescriptions(upto10–20%)than ours. These differences can be attributed to a number ofcauses. First,theyfocusonsomepopulationsamples

restrictedtothoseeligibleforeitherMedicarePartDor Medicaid,whereasourdatareferstotheoverallnumber ofdrugsales(andindeed)prescriptionswithoutage lim-itationandeligibilityconstraints.Hence,ourresultsmay indicatethatsubstitutioneffectsarelowerwhenestimated onageneralpopulation.Second,wefocusonanon-medical treatment(i.e.lightcannabis)thatcanbeboughtwithout a physician’sindication.Indeed,weestimatea demand-drivensubstitution.Withtheselenses,ourresultssuggest thatasubstitutiondrivenbyself-medicationmaybelower thantheonealsoinducedbythephysician.

Robustnesschecks

To assess the robustness of our results, we present resultsofseveralchecks.First,werestrictthetimebefore thepolicy.Thisrendersthetimewindowsmore symmetri-calbeforeandafterthepolicy(May2016-February2018). Despitereducingthenumberofobservations(2332),our main resultsand intuitionsremain unaltered.Estimates of the DID coefficients are reported in the first row of Table3 foralldrugcategories weconsider.Specifically, thelocalavailabilityoflightcannabisleadstoareduction inthenumberofdispensedboxesofsedativesby11.5%, anxiolyticsby12.3%andanti-psychoticsby4.3%. Consis-tentlywiththebaselinespecification,wefindamoresubtle butsignificanteffectsonotherprescribeddrugs:the sell-ingofanti-epilepticsdecreasesby1.5%,whereasthoseof anti-depressants,opioids,andanti-migrainesdecreaseby approximately1%.

Second,weincludealineartimetrendtocaptureany time-varyingconfoundingfactor whichmightaffectour estimates. Estimatesof theDIDcoefficient arereported in the second row of Table 3 and are very similar to those reported in the main model specification. More-over,weincludeprovince-specifictimetrends.Thisallows us toallay anyremainingconcerns regarding province-specificchangesinprescriptionsanddrugssale.Resultsare reportedinthelastrowofTable3andarequalitatively sim-ilartothosepresentedinTable2.However,weobservea reductioninthecoefficientofanti-psychotics,whichloses significancebutremainsnegative.Wethussuggestamore cautiousinterpretationofthiseffect.

Third,wetestwhetherthecommontrendassumption canbecrediblymaintained.Atypicalconcernwhichmay arisewhenadoptingaDIDapproachisthepresenceof pre-policytrends,whichmaydrivethemainresults.Toallay thisconcern,wemakeavisualinspectionofthetrendsfor provincesexperiencingearly(May–September2017)and late(October2017–February2018)localavailabilityinthe marketandforthoseprovincesneverexposedtoaccessto lightcannabisduringtheperiodweconsider.Thisallows ustoverifywhetherprovincesexperiencingdifferent tim-ingsintheentryoflightcannabisretailersfollowedsimilar trendsindrugsalesbeforetheactualliberalization(May 2017).AshighlightedinFig.4,pre-policytrendsare paral-lelandthepost-policydropindispenseddrugsisconsistent withthetimingofthelocalavailabilityintreatedprovinces, comparedtothecontrols.Thissupportsthecredibilityof thecommontrendhypothesisinoursetting.

(9)

Fig.3.Googletrendsquerieson“cannabislight”and“cannabislight+anxiolytics”inItaly.

TheﬁgurepresentsthenumberofGoogleSearchquerieson“cannabislight”and“cannabislight+ansiolitico(anxiolytics)”inItalyduringtheperiodwe considered.InMay2017,whentheproductwasannounced,thenumberofquerieshadaspikeandthenumberofqueriesremainedatahigherlevelinthe subsequentperiod.

Table3

Robustnesschecks:parameterestimates.

(1) (2) (3) (4) (5) (6) (7) (8)

All Sedatives Anxiolytics Anti-epileptics Opioids Anti-migraines Anti-psychotics Anti-depressives

Shortertime −0.016*** −0.115** −0.123** −0.015*** −0.011** −0.010** −0.043** −0.012*** window 0.003 0.054 0.058 0.004 0.004 0.005 0.018 0.003 Lineartrend −0.016*** −0.100* −0.115** −0.014*** −0.011** −0.010* −0.047** −0.012*** 0.004 0.052 0.055 0.004 0.005 0.005 0.018 0.003 Province-trend −0.017*** −0.186** −0.157* −0.018*** −0.019*** −0.012** −0.020 −0.011*** 0.003 0.076 0.093 0.003 0.004 0.005 0.013 0.003

Log-transformationofthedependentvariable.S.E.clusteredattheprovince-levelinitalics.***,**,*indicatestatisticalsignificanceat1%,5%and10%, respectively.ThefirstrowreportstheestimatesoftheDIDcoefficientconsideringashortertimewindow(May2016–February2018).Thesecondrow reportstheestimatesoftheDIDcoefficientinthepresenceofalineartrend.ThethirdrowreportstheestimatesoftheDIDcoefficientinthepresenceofa province-specifictrend.

Fig.4.Commontrend.

Thefigurepresentsthetrendsinalldispenseddrugsinourdataset (opioids, anxiolytics, sedatives, anti-migraines, anti-epileptics, anti-psychoticsandanti-depressives)forprovincesexposedtoearly,lateand nolocalaccessibilitytolightcannabis.Earlyentryreferstolocal avail-abilityduringthefirst5monthsafterthefirstentry(betweenMayand September2017).Lateentryreferstolocalavailabilityduringthelast5 monthsweobserve,whichisbetweenOctober2017andFebruary2018.

Fourth,weperformaplacebotestusingasdependent

variablesprescriptiondrugsforwhichthereisno

medi-calevidenceof possiblesubstitutionpatternswithlight

cannabis.Indeed,weconsiderinsulin,anti-hypertensives

andgenito-urinarysystemdrugs(sexhormonesand

uro-logical),whichdonotbelongtothecategory“N–Nervous

System”oftheWHOCollaboratingCentreforDrug

Statis-ticsMethodology.Forthesedrugs,weusetheATC/DDD

classiﬁcationwhich standardizestheaveragedrugdoses

foradults.Astreatingotherpathologies,theseprescription

drugsshouldnotbeaffectedatallbythelightcannabis

availability.EstimatesarereportedinTable4.AlltheDID

coefﬁcientsarenotstatisticallysigniﬁcantandveryclose tozeroinmagnitude.Thisprovidesfurthersupporttoour mainresultsintermsofcausality.

Toreduceanyresidualconcernaboutpossible viola-tionsofcommontrendassumptions, wealsoperforma permutationtestbasedonaMonteCarlosimulation.The permutationtestalsoallowsustoexploretherobustness oftheresultstoassumptions aboutthestructure ofthe errordistribution.Thisisastrategyusedinmanyempirical applications(e.g.,WingandMarier,2014).Hence,we sim-ulatetheeffectoflocalaccessibilitytolightcannabis by randomlyassigningthetreatmenttoprovincesat differ-entpointsintime,inplaceoftherealone.Werepeatthis

(10)

Fig.5.Randomizationinference.

Thefiguresshowthedistributionsoftheplaceboestimatesbasedon5000permutationsforalloutoutcomesseparately.Theverticalredlinesrepresent theestimatedcoefficientsinourbaselinespecificationinTable2.

Table4

Robustnesschecks:placeboregressions.

(1) (2) (3)

Insulin Anti-hypertensives SexHormonesand

Urologicals

DiD 0.015 −0.002 0.006

0.034 0.005 0.006

Controls yes yes yes

YearFE yes yes yes

MonthFE yes yes yes

ProvinceFE yes yes yes

N 2756 2756 2756

ThedrugsconsideredrefertothefollowingATC/DDDclassiﬁcationof theWHOCollaboratingCentreforDrugStatisticsMethodology:A10A (Insulin),C02(Anti-hypertensive),G03andG04(Sexhormones, modu-latorsofthegenitalsystemandUrologicals).Logtransformationofthe dependentvariable.S.E.clusteredattheprovince-levelinitalics.

procedure5000timestogenerateadistributionofplacebo

treatmenteffects.Fig.5presentsthenon-parametric

distri-butionsoftheseplaceboestimatesforallprescribeddrugs includedinourstudy,separately.

Fig.5showsthattheaverageoftheplacebotreatments iszeroandtheactualcoefﬁcient,whichisdepictedbythe redverticalline,fallsfarfromthelefttailofthedistribution. Asaresult,thisindicatesthatthenegativeandsigniﬁcant

effectweﬁndondrugprescriptionsisveryimplausibleto haveoccurredbychance.

Finally,toaccountfortherecentdevelopmentsinthe econometric literature in the presence of a staggered rollout of thetreatment(e.g., Atheyand Imbens, 2018; Goodman-Bacon,2018),westudythetime-varyingnature ofthepolicyandtheweightsassociatedwithits compo-nent.FollowingGoodman-Bacon(2018),weplotallthe2 ×2pairwiseDIDestimates(ontheverticalaxis)against theirweights(onthehorizontalaxis).Fig.6indicatesthat theopeningof a lightcannabis shopledtoa reduction inthesalesofprescriptiondrugs.Theredlinerepresents theweightedDIDcoefﬁcientonceaccountingforthe dif-ferentweightsassociatedtoitscomponent.Notably,the coefﬁcientsareallnegativeandqualitativelysimilartoour benchmarkresults.12

Onecannotethefollowingtworesults. First,treated vs.nevertreatedaretheoneswiththelargestweight.For

12_Estimates_shown_in_Fig.₆_are_based_on_a_{seasonally-adjusted}

out-comevariableinordertotakeintoaccounttheseasonalpatternindrug prescription.TheoriginalDiDaccountsforthepresenceofmonth dum-mies,whicharenotimplementableyetusingthedecompositionroutines availableinstatisticalsoftware.Consequently,whilethemainsourcesof heterogeneityintheestimatescanbeassessed,pointestimatesarenot directlycomparable

(11)

Fig.6.2×2DiDEstimates(Goodman-BaconDecomposition).

Theﬁgureplotseach2×2DiDestimateagainsttheirweightscalculatedfollowingtheGoodman-Bacon(2018)’sdecompositionforallprescriptiondrugs consideredinourmodel.Thetrianglereferstoestimateswhereonetiminggroupactsastreatmentgroupandthenevertreatedcountryascontrolgroup. Thegreyx’srefertotiming-onlyestimatesinwhichtheearlytreatedgroupactsastreatmentgroupandthelatertreatedgroupascontrolgroup.Theblack x’srefertotiming-onlyestimatesinwhichthelatertreatedgroupactsastreatmentgroupandtheearliertreatedgroupascontrolgroup.Theredhorizontal linereferstotheweightedaverageDiDcoefﬁcientresultingfromtheGoodman-Bacon’sdecomposition.

instance,theweightassociatedwiththisgroupisalways 52%.Thissuggeststhatthelatteristhemainsourceof iden-tificationinoursetting.Second,wenotethatwhilesome 2×2pairwisecomparisonsarepositive,theirweightis veryclosetozero.Thus,theaverageweightedDID coeffi-cientisalwaysnegativeforallouroutcomesand,insome cases,thisisevenrelativelylarger.Forexample,inthecase ofsedatives,theaveragecoefficientindicatesareduction insalesof-0.16(-0.095inourbaselineDiD).Inthecase of anxiolytics,theaveragecoefficientis-0.22(-0.114 in thebaselineDiD).However,wetakeaconservativestance, interpretingtheresultsfromourmainDIDasalowerbound estimate.

Eventstudy

Toshedsomefurtherlightontheobservedsubstitution, wealsoexplorethedynamicofpatients’responsestolocal marketavailabilityofthenewcannabis-lightproduct.By exploitingaratherrarepolicyset-upwithheterogeneous andstaggeringeffectsatalocallevel,oursettingisideal forstudyingpatients’responsesandforverifyingthe self-medicationhypothesisdiscussedabove.Wethusperform aneventstudyanalysisconceivedasfollows.Weincludein equation(1)aseriesofdummiescodingthemonthinwhich entryinthemarketoccurredandonetosixmonths pre-entryandpost-entry.One-monthpre-entryistheexcluded

dummyforeachdimensionandissetequaltozeroinour presentationoftheresults.

Resultsof theevent studyspecification are reported graphicallyinFig.7.Wefindthatreductionsinthenumber ofdispenseddrugsoccursincethesecondmonthafterthe entryoflocalretailersanditismorepronouncedfromthe thirdmonthonwards.Thefigurealsoshowsnon-linearity inthesubstitutionpatternbutprovestobeastatistically significanteffectalsointhesixmonthsfollowingthelocal product availability.Importantly, Fig.7 alsoshows that there is little evidence of systematic pre-trends affect-ingtheresultsandthisprovidesfurthersupportforthe hypothesisofexogeneityinthecannabislightsellerentry ineachprovince.Lastly,lookingatthemagnitudeofthese effects,wefindanaveragepost-treatment coefficientof -0.015.Thisis very close toourmain DiDspecification (column1,Table2).Thissuggestsonceagainthattiming heterogeneityisnotdrivingourmainresults.

Discussion

Manycountriesworldwidehavelegalizedor decrimi-nalizedmarijuanaforrecreationalandmedicalpurposes. Othershaverecentlylegalizedthecultivationfor commer-cializationofhemp,anindustrialstrainofcannabisrichin CBDandalmostfreeofpsychoactivecomponents.While ontheonehand,thismarketisgainingmomentumandis oftendescribedasthenew“greenoil”,ontheotherhand,

(12)

Fig.7.Eventstudy.

Theﬁgurepresentsaneventstudyoftheeffectsoflocalavailability(representedbytheredverticalline)onalldispenseddrugs.One-monthpre-entryis theexcludeddummyforeachdimensionandissetequaltozero.

uncertaintyis increasingamongregulators and authori-tiesaboutpotentialmisuse.Giventheuncertaintyonhow togovernthis phenomenon and exploiting the difﬁcul-tiesindistinguishinglightcannabisfromillegalmarijuana, severalstatessuchasKansas,Louisiana,TexasandNorth Carolinaopted for atough stancebybanningsmokable hemp.Yet,thismostlyunregulatedmarketkeptbooming. Thispaperprovidesaself-medicationargumentforsuch popularitybyexploitingauniqueopportunityofferedbya loopholeintheItalianlegislationregardingthe commer-cializationofanewCBD-basedproduct.Indeed,welook atwhethertheavailabilityofanewproductsuitablefor self-medicationimpactedondispenseddrugsalestotreat anxiety,psychosis,chronicpain,insomnia,migraineand epilepsy.

UsingastaggeredDiDmodelandconsideringmonthly datafor106Italianprovincesduringtheperiodbetween January2016and February2018,wefindthat thelocal availabilityoflightcannabisledtoasignificantandlarge reductions of dispensed boxes of anxiolytics and seda-tives,whichamountedtoapproximately10%.Thesedrugs usuallytreatsymptomsforwhich CBDis ofteneffective and for which a non-specialistcan easilydetect symp-toms. We also find that the entry of a light cannabis retailer in a given province led a 1–1.5% reduction in thenumberofanti-epileptic,anti-depressant,opioidand anti-migraineprescriptions.Incontrast,prescriptionsfor psychotic patients decreased by approximately 4–5%. Muchhighercoefficientsaremorelikelytoemergewhen consideringamore extended timewindow.Indeed, our coefficientindicatesalowerboundfor theeffectoflight cannabisonprescriptiondrugs.Incontrast,nosubstitution effectwasfoundforprescriptiondrugswithoutrelaxant effects,suchasinsulin,hypertensives,aswellassex hor-monesandurological.

Our results are compatible with a self-medication hypothesis, that is, the adoption of risky behaviors to seeka quality of life improvements and relief. Indeed, the large-scale accessibilityto the new product, which

wasadvertisedasarelaxantone,inducedsomepatients to abandon traditional medicine to seek relief. Self-medicationarisesastheproductwasneithersuitablefor medicalpurposesnorwasallowedtobeconsumed.This renders less likely that the substitutional patterns we observedweredrivenbymedicaladvice.Asdiscussed,light cannabisliberalizationwasduetoalegislativevoidand, forthisreason,unannouncedandnotcapableofcreating ananticipatoryeffect,thereforereducingthelikelihoodof “offtherecord”medicaladvice.Ourestimatesassumemore relevancewhenconsideringtherelativelyshorttime win-dowweconsider,thatis10monthsafterthepolicywas implemented,thelackofclinicalsupport,andtheunusual wayinwhichlightcannabiswasmadeavailableaswellas ourfocusonashortperiodaftertheintroductionofthe product.Patientresponsetonewdrugsandtherapies is usuallyheterogeneousandsluggishbecauseoftypicalrisk aversion.Nonetheless,weﬁndasigniﬁcantandnegative effectonseveraldrugstreatingpathologiesforwhich med-icalcannabis (andnotlightcannabis)hasdemonstrated somedegreeofeffectiveness.

Theeventstudyalsoprovidesadditionalinsightsinto patients’ responses.We observethat substitutional pat-ternsaremoreaccentuatedstartingfromthethirdmonth aftertheentryoftheﬁrstlightcannabisretailerinthelocal areaandsigniﬁcantalsoaftersixmonthspostentry.This result mayindicatethat individualsstarted substituting therapyaftersomeweeksofexperimentationorjustafter realizingthatthenewproducthadbecomelocally avail-able,therebyreinforcingthe“rational”characterizationof self-medication.Inthissense,thepaperprovidesa self-medicationexplanation,differentfrom“hype-seeking”for theexplosivesuccessofcannabis-derivedproducts,even intheabsenceofthepsychoactivecompound(THC),inthe USandEurope.

Finally,ourestimatespresentaseriesofimportant pub-licpolicyimplicationsthatgobeyondItaly’scase.First,the substitutionweidentifymaystemfromthepatients’needs ofmoreeffectiverelief,whichisnotcurrentlyaddressed

(13)

bytraditionalmedicine.Thisconfiguresaformofmarket failurethatlawmakersshouldaccountfor.Onewaycould betoimprovethecurrentdistributionchannelofmedical marijuana.Absentanyfurtherpolicychangeinthis direc-tion,ourstudysuggeststhatpoliciessuchasabanonlight cannabis,aspromotedbyseveralUSstatesandbyItalian policymakersin2019,mighteventuallydecreasepatients’ welfareasdisregardingtheirneedsformoreeffectiverelief. Second, forms of self-medication may, at first, ring the alarm bellsof policymakersas individuals may not follow expert advice even when taking care of their health. Our results suggest that regulatory authorities shouldbecautious andvigilant asthelarge-scale avail-abilityoflightcannabismayinducesubstitutionpatterns not clinically validated. Moreover,our study highlights thatregulationoflightcannabisisrequired.Inasimilar vein,amoregeneralpublichealthcampaignshould com-plement sucha policy.For instance,instructingdoctors and providing labels and certification alongside infor-mation regarding doses for these products may be a first attempt to make light cannabis, and in general CBD-derived product consumption, much more con-trolled.

Third,fromapublicpolicyperspective,weobservethat theshiftin consumptionfromtraditional drugs tolight cannabiscamefromanon-negligiblecostforapatient.This isbecause,unlikethedrugsweconsider,whichareeither fullyreimbursedbytheItalianNHSorsubjecttoasmall co-paymentfromapatient,lightcannabisisoftensoldat 8–10eurospergram.Thissuggeststhattheunintended policyliberalizationledtoanunexpectedcontractionof pharmaceutical expenditure,ceteris paribus.As pharma-ceuticalexpenditureworldwidehasrapidlyincreasedin recentyearsandthisisequallytrueinItaly(Federfarma, 2018),thepresenceofsuchshort-termﬁnancialbeneﬁts shouldbeconsideredamongtheotherwelfareeffectsof thisunintendedpolicyexperiment.

Acknowledgments

We thank the Editor Maarten Lindeboom and two anonymousrefereesforinsightfulanddetailedcomments. We are gratefultoAnna-KatharinaBöhm,Davide Drag-one,SandyTubeuf,andJanC.vanOurs,participantsatthe NHESG2019Conference(Iceland),theItalianAssociation forHealthEconomics(AIES,Pisa2019),theFirstBolzano AppliedMicroeconomicsWorkshop(2019),alongside sem-inarparticipantsattheUniversity ofYork,University of Pavia,andattheErasmusSchoolofEconomicsfor help-ful comments on previous drafts of the paper. Finally, we also thank Andrea Agnello for assistance over the use of data and Walter Vigna. Usual disclaimers apply. Leonardo Madio acknowledges funding from the Euro-peanResearchCouncil(ERC)undertheEuropeanUnion’s Horizon2020researchandinnovationprogramme(grant agreement No 670494) and from the French National Research Agency (ANR)under the Investments for the Futureprogramme(Investissementsd’Avenir,grant ANR-17-EURE-0010).

AppendixA. Supplementarydata

Supplementary material related to this arti-cle can be found, in the online version, at doi:https://doi.org/10.1016/j.jhealeco.2020.102371.

References

Anderson,M.D.,Hansen,B.,Rees,D.I.,2013.Medicalmarijuanalaws, trafﬁcfatalities,andalcoholconsumption.J.LawEcon.56(2), 333–369.

Athey,S.,Imbens,G.W.,2018.Design-BasedAnalysisin

Difference-in-DifferencesSet-TingsWithStaggeredAdoption.NBER WorkingPaper24963.NationalBureauofEconomicResearch. Bachhuber,M.A.,Saloner,B.,Cunningham,C.O.,Barry,C.L.,2014.Medical

cannabislawsandopioidanalgesicoverdosemortalityintheUnited States,1999-2010.JAMAIntern.Med.174(10),1668–1673. Baggio,M.,Chong,A.,Simon,D.,2020.Sex,drugs,andbabybooms:can

behaviorovercomebiology?J.HealthEcon.,Forthcoming. Baltagi,B.H.,Grifﬁn,J.M.,2002.Rationaladdictiontoalcohol:paneldata

analysisofliquorconsumption.HealthEcon.11(6),485–491. BDSAnalytics,Availableat:2019.U.S.CBDMarketAnticipatedtoReach

$20BillioninSalesby2024.

https://bdsanalytics.com/u-s-cbd-market-anticipated-to-reach-20 -billion-in-sales-by-2024/.

Becker,G.S.,Murphy,K.M.,1988.Atheoryofrationaladdiction.J.Polit. Econ.96(4),675–700.

Becker,G.S.,Grossman,M.,Murphy,K.M.,1994.Anempiricalanalysisof cigaretteaddiction.Am.Econ.Rev.84(3),396–418.

Blessing,E.M.,Steenkamp,M.M.,Manzanares,J.,Marmar,C.R.,2015. Cannabidiolasapotentialtreatmentforanxietydisorders. Neurotherapeutics12(4),825–836.

Bradford,A.C.,Bradford,W.D.,2017.Medicalmarijuanalawsmaybe associatedwithadeclineinthenumberofprescriptionsfor Medicaidenrollees.HealthAff.36(5),945–951.

Bradford,A.C.,Bradford,W.D.,2018.TheimpactofmedicalCannabis legalizationonprescriptionmedicationuseandcostsunder medicarepartd.J.LawEcon.61(3),461–487.

Bradford,A.C.,Bradford,W.D.,2016.Medicalmarijuanalawsreduce prescriptionmedicationuseinmedicarepartd.HealthAff.35, 1230–1236.

Carrieri,V.,Madio,L.,Principe,F.,2019.Lightcannabisandorganized crime:evidencefrom(unintended)liberalizationinItaly.Eur.Econ. Rev.113,63–79.

Case,A.,Deaton,A.,2015.Risingmorbidityandmortalityinmidlife amongwhitenon-hispanicamericansinthe21stcentury.Proc.Natl. Acad.Sci.112(49),15078–15083.

Cawley,J.,Ruhm,C.J.,2012.Theeconomicsofriskyhealthbehaviors. Chapter3.In:McGuire,ThomasG.,Pauly,MarkV.,Barros,PedroPita (Eds.),HandbookofHealthEconomics,Vol.2.Elsevier,NewYork,pp. 95–199.

ˇ

Cerven ´y,J.,Chomynová,P.,Mravˇcík,V.,vanOurs,J.C.,2017.Cannabis decriminalizationandtheageofonsetofcannabisuse.Int.J.Drug Policy43,122–129.

Chaloupka,F.,1991.Rationaladdictivebehaviorandcigarettesmoking.J. Polit.Econ.99(4),722–742.

Chan,N.W.,Burkhardt,J.,Flyr,M.,2020.Theeffectsofrecreational marijuanalegalizationanddispensingonopioidmortality.Econ.Inq. 58(2),589–606.

Chang,T.Y.,Jacobson,M.,2017.Goingtopot?Theimpactofdispensary closuresoncrime.J.UrbanEcon.100,120–136.

CIBG,Availableat:2018.ReportonMedicalCannabisintheNetherlands. OfﬁceforMedicalCannabis.

https://english.cannabisbureau.nl/medicinal-cannabis/documents/ circulars/2018/02/20/patients-guide-medicinal-cannabis. Crost,B.,2012.Theeffectofalcoholavailabilityonmarijuanause:

evidencefromtheminimumlegaldrinkingage.J.HealthEcon.31 (1),112–121.

Cunningham,S.,Shah,M.,2018.Decriminalizingindoorprostitution: implicationsforsexualviolenceandpublichealth.Rev.Econ.Stud. 85(3),1683–1715.

Darden,M.E.,Papageorge,N.W.,2018.RationalSelf-Medication(No. w25371).NationalBureauofEconomicResearch.

Dinardo,J.,2001.Alcohol,marijuana,andamericanyouth:the unintendedconsequencesofgovernmentregulation.J.HealthEcon. 20(6),991–1010.

(14)

Dragone,D.,2009.Arationaleatingmodelofbinges,dietsandobesity.J. HealthEcon.28(4),799–804.

Dragone,D.,Prarolo,G.,Vanin,P.,Zanella,G.,2019.Crimeandthe legalizationofrecreationalmarijuana.J.Econ.Behav.Organ.159, 488–501.

FDA,Availableat:2019.FDARegulationofCannabisand Cannabis-derivedProducts:QuestionsandAnswers.

https://www.fda.gov/news-events/public-health-focus/fda-regulation

-cannabis-and-cannabis-derived-products-questions-and-answers#farmbill.

Federfarma,2018.LaSpesaFarmaceuticaNel2017.Analisi

Dell’andamentoDellaSpesaFarmaceuticaConvenzionataaLivello NazionaleERegionale.

Gavrilova,E.,Kamada,T.,Zoutman,F.,2017.Islegalpotcrippling Mexicandrugtrafﬁckingorganisations?Theeffectofmedical marijuanalawsonUScrime.Econ.J.129(617),375–407. Goodman-Bacon,A.,2018.Difference-in-DifferencesWithVariationin

TreatmentTiming.NBERWorkingPaper25018.NationalBureauof EconomicResearch.

Grossman,M.,Chaloupka,F.J.,1998.Thedemandforcocainebyyoung adults:arationaladdictionapproach.J.HealthEcon.17(4),427–474. Hansen,B.,Miller,K.,Weber,C.,2020a.Earlyevidenceonrecreational

marijuanalegalizationandtrafﬁcfatalities.Econ.Inq.58(2), 547–568.

Hansen,B.,Miller,K.,Weber,C.,2020b.Federalism,partialprohibition, andcross-bordersales:evidencefromrecreationalmarijuana.J. PublicEcon.187,104159.

ISTAT,Availableat:2018.LaSaluteMentaleNelleVarieFasiDellaVita. ISTAT2018–Anni2015-2017.

https://www.istat.it/it/archivio/219807.

Liang,D.,Bao,Y.,Wallace,M.,Grant,I.,Shi,Y.,2018.Medicalcannabis legalizationandopioidprescriptions:evidenceonUSMedicaid enrolleesduring1993–2014.Addiction113(11),2060–2070. McMichael,B.J.,VanHorn,R.L.,Viscusi,W.K.,2020.Theimpactof cannabisaccesslawsonopioidprescribing.J.HealthEcon.69, 102273.

Mercuri,A.M.,Accorsi,C.A.,Mazzanti,M.B.,2002.Thelonghistoryof CannabisanditscultivationbytheRomansincentralItaly,shownby pollenrecordsfromLagoAlbanoandLagodiNemi.Veg.Hist. Archaeobot.11(4),263–276.

NationalAcademiesofSciences,Engineering,andMedicine,Report releasedonJanuary12,20172017.Thehealtheffectsofcannabisand cannabinoids:theCurrentStateofEvidenceandRecommendations forResearch.

Powell,D.,Pacula,R.L.,Jacobson,M.,2018.Domedicalmarijuanalaws reduceaddictionsanddeathsrelatedtopainkillers?J.HealthEcon. 58,29–42.

Sabia,J.J.,Swigert,J.,Young,T.,2017.Theeffectofmedicalmarijuana lawsonbodyweight.HealthEcon.26(1),6–34.

Shy,Y.,2017.Medicalmarijuanapoliciesandhospitalizationsrelatedto marijuanaandopioidpainreliever.DrugAlcoholDepend.173, 144–150.

Smith,R.A.,2020.Theeffectsofmedicalmarijuanadispensarieson adverseopioidoutcomes.Econ.Inq.58(2),569–588.

TheNewYorkTimes,July4,2019.Availableat:2019.NorthCarolina ProposesSmokableHempBanAsDemandGrows.

https://www.nytimes.com/aponline/2019/07/04/us/ap-us-smokable-hemp-bans.html.

VanLuijk,E.W.,VanOurs,J.C.,2001.Theeffectsofgovernmentpolicyon druguse:java,1875–1904.J.Econ.Hist.61(1),1–18.

VanOurs,J.C.,1995.Thepriceelasticityofharddrugs:thecaseofopium intheDutchEastIndies,1923-1938.J.Polit.Econ.103(2),261–279. Wen,H.,Hockenberry,J.M.,Cummings,J.R.,2015.Theeffectofmedical

marijuanalawsonadolescenceandadultuseofmarijuana,alcohol, andothersubstances.J.HealthEcon.42,64–80.

Wing,C.,Marier,A.,2014.Effectsofoccupationalregulationsonthecost ofdentalservices:evidencefromdentalinsuranceclaims.J.Health Econ.34,131–143.