[PDF] Smart (Phone) Investing? A within Investor-time Analysis of New

PDF document for free

1. PDF document for free

956_2w28363.pdf

NBER WORKING PAPER SERIES

SMART(PHONE) INVESTING? A WITHIN INVESTOR-TIME ANALYSIS OF NEW

TECHNOLOGIES AND TRADING BEHAVIOR.

Ankit Kalda

Benjamin Loos

Alessandro Previtero

Andreas Hackethal

Working Paper 28363

http://www.nber.org/papers/w28363

NATIONAL BUREAU OF ECONOMIC RESEARCH

1050 Massachusetts Avenue

Cambridge, MA 02138

January 2021

We thank Shlomo Benartzi, Juhani Linnainmaa, Ulrike Malmendier, Brian Melzer, and seminar participantsat Indiana University and BI Norwegian Business School for helpful comments and discussions. The views expressed herein are those of the authors and do not necessarily reflect the views of the NationalBureau of Economic Research. NBER working papers are circulated for discussion and comment purposes. They have not been peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies official

NBER publications.

© 2021 by Ankit Kalda, Benjamin Loos, Alessandro Previtero, and Andreas Hackethal. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source . Smart(Phone) Investing? A within Investor-time Analysis of New Technol ogies and Trading

Behavior.

Ankit Kalda, Benjamin Loos, Alessandro Previtero, and Andreas Hackethal

NBER Working Paper No. 28363

January 2021

JEL No. G11,G40,G50

ABSTRACTUsing transaction-level data from two German banks, we study the effects of smartphones on investorbehavior. Comparing trades by the same investor in the same month across different

platforms, we find that smartphones increase purchasing of riskier and lottery-type assets and chasing past returns. After the adoption of smartphones, investors do not substitute trades across platforms and buy also riskier, lottery-type, and hot investments on other platforms. Using smartphones to trade specific assets or during specific hours contributes to explain our results.

Digital nudges and the device screen size

do not mechanically drive our results. Smartphone

effects are not transitory.Ankit KaldaKelley School of Business1309 E 10th StIndiana UniversityBloomington, IN 47405akalda@iu.edu

Benjamin Loos

TUM School of Management

Arcissstrasse 21

Munich 80333

Germany

beni.loos@gmail.comAlessandro PreviteroKelley School of BusinessIndiana University1309 E. 10th StreetBloomington, IN 47405and NBERaleprevi@indiana.edu

Andreas Hackethal

Goethe University Frankfurt

Grüneburgplatz 1 House of Finance

60323 Frankfurt a.M. Germany

hackethal@gbs.uni-frankfurt.de

1 Introduction

Technologyhas dramaticall ychangedhowretail invest orstrade,from placingor ders using directdial-up connectionsin the 1980sor Internet-basedtrading int he1990s to themore recentrise ofrobo-advisers. Wit hf ew exceptions,theintroduction ofthese newtechnologies isg enerall yassociatedwithadeclineinin ves torportfolio efficiency. 1 Whethergood orbad for inv estors,it isacceptedthatnewtechnologiesinfluencein ves tor behavior.Theem piricale videnceinthese studiescomesfrom somecom parisonsofin- vestorbehaviorbeforeandaftertheadoptionofthenewtechnology,potentiallycontrasted witht hebehavioro vertimeof anothergroupt hatdid not adoptthetechnology. Under theassum ptionthat,absentt heinnovation, inv estorsw ouldhavebehav edintheexact same way,acommoninterpretation oft hise videnceisthat new technologiesinfluence investorsandchanget heirbeha vior.Analter nativeexplanationis thatinvest ors,ins tead, adoptt henewtechnology becausethey arewilling to changetheirtrading behavior int he firstplace. Even ifwecould randoml yassignthe new technologytoinves tors, 2it would stillno tbestr aightfor wardtoconcludethatthenewtechnologychang estheo ver allin- vestorportfolio.If invest orsmanag einvestmentsacrossdifferentaccounts orplatforms, theycoulddecide to substitute acrosstechnologies.Therefore, observingtrades onone platformmightno tbe informative oft heoverallinves tor tradingbehavior . While previousstudies lackthedata to distinguishbetween these alternativeinter pre- tations, theirimplications are,howe ver ,starklydifferent.Ifthenew technologyinfluences investorpreferencesandbeliefs, absentthetechnology inv est orsw ouldhave not changed

theirtradingbehavior.If,instead,itfulfillsuntapped inv estordemand, thenthenewtech-1Forexam ple,whenmoving to onlinetrading,in vestors increasedtur noverand reducedperformance

(Barber andOdean, 2002).More recents tudiesdocument, instead, thatrobo-adviserscould reducein vest- ment mistakes(seeD"A cunto, PrabhalaandRossi,2019;Looset al.,2020).

2D"Acunto,Prabhalaand Rossi(2019)use the randomnessinin ves tors answeringtheir phonetot he

marketingenrollmentcalls asa plausibly exog enousshock tothe probabilityofjoiningtherobo-advisor . 1 nology atbes tacceleratesor makeslesscos tly achangein inv estorbehaviort hatw ould havehappenedan ywa y.Analogously,thenewtechnology couldjustfulfillsubstitute demand, ifin vestorssubstitutetradesacrossdiff erentplatforms. Therefore,simple com- parisons ofin vestorbehaviorpre-andpost-adop tionoranalyses oftr adeson onesingle platformcouldv astl yoverestimatetheeff ectsofthenewtechnology. Furthermore, the policy implicationscouldno tbe anymorediff erent.Is thetechnologyhelping inv estors toachie vetheirgoalsb yfacilitatingt heirtr ades?Oristechnology influencingadopters in profoundwa ysthatcouldstr ayinv est orsawayfromtheir originalgoals?3 In thispaper, weuseunique dataonGerman householdst oo vercome these empirical challengesand to weighinon thequestion iftechnolo gydriv eschangesor justfulfills untapped orsubs tituteinves tordemand.Ourdatacomesfromtwolar ge German retail bankst hathave introducedtradingapplicationsfor mobilede vices.Foro ver 15,000 bank clientst hathave usedthesemobileappsin the years2010-2017,w ecan observ eall holdings andtr ansactions,and,moreim portant,t hespecific platformused foreachtr ade (e.g., personalcomputervs.smartphone).Theseuniquefeatures oft hedataprovefruitful forour analyses. Thatis,we canconduct allour maintestscom paringtr adesdone by the same investorinthesamemont hacros sdifferentplatf orms. Moreover, wecandirectl y testf orsubstitutioneff ects. Wepresent four setofresults.F irst, we study iftheuseofsmartphones induces differencesin the riskinessoftrades. Comparing trades bythe sameinves torint he same year-month,wefindt hattheprobability ofpurchasing riskyassetsincreasesin smartphone tradescompared tonon-smartphoneones. Analogously,smartphone trades involveassetswithhigher volatility andmorepositive ske wness.Thisevidence isbes t

summarized byouranal ysesof lotterytype st ocks.4Smartphones increaset heprobability3In a2020 articletitled "Robinhood HasL uredYoungTr aders,SometimesW ithDevastatingResults, theNe w

YorkTimes featuresaseries ofstories ofin ves tors thathavelos tsubs tantialamountofmoneytradingoff theirmobile phones.

4FollowingKumar (2009),wedefine aslo ttery-types tock st hoseassetswithbelowmedianprices and

2 of buyinglo ttery-typestocksb y67%oftheunconditionalmeanf orsmartphoneusers. Second, weexaminet heeff ectsofsmartphoneson thetendencyt ochase past returns. Wefind that smartphonesincreasethe probabilityof buyingassets inthet opdecile of thepas tperformance distribution.Smartphonesincreasethe probabilityofbuyingassets in thetop 10percentofpas tperf ormance by 12.0percentagepoints (or70.6%ofthe unconditional mean). Third,w einves tigateifinvestorsselectivel yuse smartphonetoexecutetheirrisky, lottery-type,andtrend-chasingtr ades.In this scenario,inves tors couldsimply substitute theirtr adesfromonede vicet oano ther,wit houtanyreal consequencesfortheirover - all portfolioefficiency. Usingadifference-in-diff erencesdesign that comparesiOSand Android users,w efindthat, follo wingthelaunch ofsmartphoneapps,inves torsare-if anything-morelikel ytopurchaseriskyand lottery-type assetsand to chaseho tinvest- ments alsoon non-smartphoneplatforms.Whileinconsis tentwit hsubstitutioneff ects,this evidencepo tentiallysuggestst hatinvestors arelearningt obecomeoverallmorebiased aftert heirinitialuseof smartphonest otr ade. Last,w eevaluate themechanismsthatma ydrivet hesesmartphone effects.We begin byexaminingwhethertheabilitytotradeanytimeandeverywhereviasmartphonesdrives our results.T oevaluate theimportanceoft hischannel,we repeatouranalyses, including year-by-time-of-the-dayfixedeffects.Int hisspecification,oures timatesbecomesmaller but remaineconomicall yandstatis tically significant.Thisfindingsuggeststhat timeof tradeis important inoursetting,but itdoes not fully explainour findings.Consis tent witht hisinterpretation,heterog eneityanalysessho wthatsmartphone effectsares tronger during after-hours(i.e.follo wingex changeclosure).Institutionaldiff erencesbetween tradingon officialex changes andinafter-hoursmar kets dono tdrivethisheterog eneity.

Givent hatindividualsaremore likel yt orel yonthemoreintuitiv esystem1 laterinthe abovemediansk ewness andvolatility.

3 day(Kahneman,2011),strongereffectsduringafter-hoursareconsistentwithsmartphones facilitatingtr adesbasedmoreon system 1t hinking. Alternatively,invest orsmayusesmartphonestotradediff erentinves tmentsand this selection ofriskier assetclasses may drive ourresults.We re-estimateourmain anal- yses, includingy ear-by-asset-classfixedeffects.We findagainsmaller butstills trong smartphone effects,sugges tingthatthechoice ofassetclassesdoesn "tfully explainour findings. Anotherpossibilityis that digitalnudg esmightcontributet oour results.Smartphone tradingapps inf actprominentl yfeatures tockst hathave experienceddramaticpositive (and negative)performance intherecentpast. Ift hesestock sare riskierandwith higher skewness,digitalnudg escould mechanicallyinfluencein ves torbehavior .T otestforthis hypothesis,were-r unourmainspecifications separatelyf ordiff erentasset classes:indi- vidual stocks,mutualfunds,ando ther inv estments (options,certificates,andwarr ants). Wefind that ourresultsares trongacross allasset classesand notjust for individuals tock s thatcan bemore prominently featured inthesmartphone tradingapp.A dditionally , wetes tifaph ysicalattribute ofsmartphones-t heirsmallerscreen-contributest oour findings. Toexploret hismechanism, weseparatel yin vestigate theeffectsoftr adingvia deviceswit hdifferentscreen sizes(iPhonesvs. iPads). Given that wedono tfindstrong er results fortrades viaiPhones,we concludet hatt hisphysicalattribute isno tlikely todrive our findings.Las t,ourresultsdo not appeart obe short-livedanddriv enb ytheinitial enthusiasmor the learningcurv eofthe newtechnology.Oures timatesdo notchange significantlybetw eenthefirs tquarterupt othetent hquarter after theinitialuse ofthe smartphone app. Our findingscontribute to literatureont heeffectsof technologyon invest orbehavior. Barber andOdean (2002)document that inv estorswho switchedfromphone-basedto onlinetradingstarttradingmorefrequently,butlessprofitablythanbefore.Choi,Laibson, 4 and Metrick(2002) documentsimilar resultsin 401(k)plans. Oure videncecom plements theses tudiesbydocumenting thatsmartphonesincrease the purchasesoflotter y-type stockandtrend-chasing. Moreim portantly ,w edocumentdifferentbehaviorswithin the same investorandsame month,butacross platfor ms.This identificationstrategyenables us tomorecon vincingly addressselectioneffectswhenexamining how ane wtechnology impactsin vestorbehavior. Givent helarge diffusionofrobo-advisersinthe past decade,D" Acunt o,Pr abhalaand Rossi(2019) andLoos etal. (2020)ha ve inv estig atedtheeffectsof thisinnovationon investorbehavior.Bo thstudieshighlightt hatrobo-advicehasthepo tentialtoreduce investmentbiasesandim prov eportf olioperformance.Ourevidencepro videamore nuanced pictureof the effectsofne wtechnologiesonin ves torbehavior .Smartphones appear tofos terinvestmentbiases suchasinvestinginlo ttery -typeand hot stocks.Our paper contributesalso to therecentliter atureonthe effect ofmobileappson financial behaviors.Le viandBenartzi(2020) andD" Acunt o,R ossi,and Weber(2020)studythe effectsof mobileapplications onspending behaviors. We contributet otheses tudiesby investigatinginvestment decisions.Oursettingprovidesanicelabor ator yt ounders tand theconsequences ofpro vidingcons tantfeedbackand easeofexecution oftradest oretail investors. More recently,aseriesof studies hav ein vestigated theeffectsoftr adingsmartphone apps onagg regatemarkets.Using datafromtheUSretail broker age compan yRobin Hood,Welch(2020)findsthataportfoliomimickingtheaggregateholdingsofRobinHood investorsdidnotunder perfor mstandard academicbenchmarks.5Using thesamedata, Barber etal. (2020)document that episodesof intensebuyingactivityb yR obinhoodusers

are followedbynegativ ereturns.Using datafromaleadinginv estmentadviserin China,5RobinHood operates entirelyonlinevia awebsiteand mobileapps. Thev ast majorityofits trades are

made usingt hesmartphoneapps. 5 Cen (2019)sho wsthat,after themobileapp introduction,inves torsflow sint omutual funds becomemore volatile andmoresensitive to short-term fundreturnsand market sentiment. Ourresults nicely dovetailwit hthefindingsint hesestudiesand makethree distinctivecontributions.F irst, wefocusont heconsequencesofsmartphones onretail investors,andnotagg regate markets.A ggregateeffectsmight masksubstantialinves tor heterogeneity,makingit difficultt ounders tandpo tentialredistributive effectsoft his technology.Second, ourin ves tortradingdataallowtoshar penthecausal interpretation of smartphoneeff ectsandto inv estigatet hemechanismsdrivingthem.Third,while Robinhoodin vestorsareMillennialswithlittleor notr adingexperience, theGerman investorsthatadopt smartphonetradingare, onaver age, 45yearsold with niney earsof experience investingwiththe banks.Therefore, wecancap turetheeff ectsofsmartphone tradingon moreexperienced usersand amore representative sample oftr aders.

2 HypothesesDevelopment

Newtechnologiescan change the wayhouseholds makeeconomicdecisions, from labor supply,toborro wing,toin vestor behavior.6Broadlyspeaking, we investig ateif smartphones influencefinancial risk-taking andinves tmentbiases. Theeffectsofsmart- phones inbo thsettingsarenot obvious ex-ante.By facilitatingsearchingand monitoring efforts,smartphones canreduce the participationcos tsinthe st ockmarket andpromo te financial risk-taking.Ifin ves torsare,instead,sensitivetoshort-ter mlosses, themore frequent feedbackviasmartphones couldreduce risk-taking, aspredicted int heframe- workofm yopic lossaversionby Benartziand Thaler(1995).Consistentwith myopicloss

aversion,Haighand List (2005)document thatprofessional option traders takelessrisk 6Forexample,Foset al(2019),Jackson(2019)andKoustas(2018) documenttheeffectofride-sharingapps

on labormar ketdecisions;DiMaggioand Yao (2019),Buchak ,Matv os,Pisk orskiandSeru (2018)and Fuster et al.(2018) documentt heeff ectofFintech lendingonborrowing decisions;and D"Acunto, Prahabala, and Rossi(2019) documentt heeff ectofrobo-advisingon investment decisions. 6 when randomlyassignedto the treatmentofreceivingmore frequentfeedback. Psychologistshypo thesizethatwehav etwomodes ofthinking:sys tem1,whichis fast,instinctiv eandemotional,andsys tem2, whichisslow er, moredeliberative, and more logical(S tanovichandWest, 2000;Kahneman, 2003).Byprovidingthe abilityt o almostinstantaneouslytradeinmorerelaxedenvironments,smartphonesmaypotentially allowmore impulsiv e,system1trades. System1reasoning isassociated withpreference forlo tteries(seeKahneman,2011). Kumar (2009)documents that preferencefor gambling are correlatedwit hlottery -typestockswithpositiv elyskewedpay offs.Moreover ,Bali et al.(2019) findt hatin vestor preferencesforlotterystock sareamplified byattention and socialinter action.Collectively ,thisevidencesugges tsthatsmartphones mighthave a strongeff ectsonpreferences for lottery-type investments withpositiveskewness. Newtechnologiesha ve thepotentialto reduceinves tmentbiases.Whilehuman ad- visors mightmak ethesame investment mistak esoftheirclients(Linnainmaa,Melzer, and Previtero,2020),robo-advisers area potential cost-eff ective solutionthatcould in- crease portfolioefficiency(e.g., D"A cunto, Prahabala,andRossi,2019;Looset al.,2020). Smartphones couldg rantubiquitousaccesst oinformation andhigh speedofexecution of trades.Garg anoandRossi(2018)document the moreattentionto inv estment lead tohigher profits.On the ot herhand,aspre viouslydiscussedsmartphonescouldpro- motemore intuitive/ system1t hinkingand,possibly ,higher relianceonvalue-des troying investmentheuristics.F orexample,consumers aremoreinclinedt omakeim pulsive pur- chases suchas ordering moreunhealthy food whenusingmobiledevices.Benartzi and Lehrer (2015)re viewoftheeff ectsof smartphonesonconsumerchoices. Givent hattheireff ectsareambiguous,w etestif smartphonesinfluence financialrisk - taking, preferencesfor lotterys tocks,andthe well-documentedinvestmentbiast ochase pastretur ns. 7

3 Data&Em pirics

This sectiondescribes the datausedint heanal yses,discusses oursam pleanddetails our empiricalstr ategy.

3.1 Data&Summar yS tatistics

Weuse comprehensiv einvestor transaction-leveldatafromtwolarg eGermanretail banks.F oralarg er andomsampleofclientsat thebanks, weobserv et hesecurities traded, thetypeoftrade(buysvs.sell),dayandtimeofthe tradeexecution,priceandunitsofeach transaction,and, moreim portantf orouranalysis, theplatfor mused foreachtr ade.The customersinour datahold their primary accountswit hthesetw obanksand uset hemfor mostof their transactions.Thisdata coversabout sixtyfiv emillion transactionsovert he yearsfrom1999to2017byovertwohundredandtwentyfivethousandinvestors.Thedata from firstbankco vers informationonov erfortyfivemillion transactions byonehundred and tent housandinves torsfrom1999to2016.Thedatafrom the secondbank coversclose totw entymilliontransactions by onehundredandsixteen thousandinv est orsfrom 2003 to2017. At theinv estorle vel,weobservemonthly snapshotsofportfolioholdings and demographiccharacteris ticssuchasgender, age, wealt h,andincome.7 Mostof ouranal ysesuses transactiondatawhere we imposethree sample filters.F irst, welimit oursam plebetw een2010and2016 foronebank andfrom 2013t o2017 forthe otherbank.W echoose theseyears to reflecttheear liestsmartphoneapps introduction foreach bank.8Second, wedroptr adesassociated withsavings plansand wealth managementser vicesbecausethese areaut omatedordon" tin volve anactiv echoicefrom investors.Finally, wedroptradeswithoutinf ormationont heasset traded(e.g.,asset

class). Applyingthese filtersresultsina sample ofo ver twenty twomilliontr ansactions7Wealthandincome areonl yrecor dedat theaccountopening.

8Our resultsare robust ifwesepar ately estimatethem foreachbank

8 byroughl yonehundredand eightyt housandin ves tors. Overeighteent housandofthese investorsusesmartphonetrading appsat least once. Wecom plementtheproprietar ydatafromt hetwobank swit hpublicl yavailable data on prices,retur ns,andot herchar acteristicsforallsecurities tradedwithinGerman y. Table

1 reports summarystatis ticsforvariablesused inouranalyses withinoursam ple.

Smartphone isa dummy variablethat takesav alueofonef ortradesex ecutedusing smartphones. Ona verage,2%oftradesinour sample areplaced usingsmartphones (standarddeviation of0.15).How ev er, conditionaloneverusingthem,invest orsex ecute over15%oftheirtradesviasmartphones.Wefirstmeasurerisktakingasthe probabilityof purchasing riskyassets (i.e.,direct andindirect equityholdings) .The av erag eprobability ofpurchasingriskyassetsinoursampleis93%.Wealsomeasurerisktakingasthevolatility of theassetspurchased, measuredas the annualizeds tandard deviationov era trailing

12-monthrolling window .Themeanvolatilityin oursam pleis 17.27%witha standar d

deviationof 13.14%. Our measuresf orgamblingpref erencesincludeinv estmentske wness,calculated ona

12-monthrollingwindow,andtheprobabilityofpurchasinglotterytypeassets.Following

Kumar(2009), we definelottery -typeassets asthosewithbelowmedian price,and abov e median volatilityandsk ewness. Themeanprobabilityofpurchasinga lotter y-type asset withinour sample is7%.To examinetrend chasing,w eusethe probabilityof purchasing assetsinthetopdecileofthepastreturndistribution(weuseatrailing12-monthwindow). Finally,weuse thebank-reported riskcategories oftheassets purchasedandthe probabil- ity ofpurchasing warr antsorcertificates.Thebanks "risk -categoriesappl yt oalltheassets tradedb ytheclients andrange between oneand five,withhigherv aluesrepresenting greaterrisk s.Theav erag eriskcategoryfortheassetspurchased inour sampleis3.99. The meanprobability ofpurchasing aw arrant is9% (3%foracertificate). In Figure1, weexploret hee volutionoft heusageofsmartphoneso ver oursample 9 period. PanelAplo tst hepercentageof tradesthat occurviasmartphoneson the Y-axis againstcalendary earon theX-axis. Oneof thetwo banksinour sample launcheda smartphone tradingappin 2010.By 2017,t heend ofour sample, over2.5% ofall trades wereoccurringoversmartphones.Theaverage usagedropsin2013becauseweaddtoour data asecond bankwhich launchedits own trading appin 2013.Amonginv estorswit h thebank that firstintroducedt hetradingapp, 4%of alltradesw ereviasmartphonesin

2017. Althoughthe overall fractionoftotaltrades mightappearsmall,t headop tionrates

are steadilyincreasingwith ov er10%ofall investorsusing smartphonetr adingb y2017. PanelB plots thepercentage oftradest hatoccurviasmartphones only forthose invest ors who haveusedthe smartphoneapp atleastonce. Amongt hesein ves tors,the fraction of smartphone tradesismuch higher, reachingo ver20%of allt heirtradesby2017. Thus,if smartphone tradesdiffer fromother trades, theymighthav easignificantim pactonthe overallportfolioefficiency . Sinceinvestorsendogenouslychoosetousesmartphones,adoptersmightbeinherently differentfrom non-adopters. InTable2, wecompare tradingbehavior (PanelA)and investorcharacteristics(P anelB)acrosssmartphoneusersand non-users.Fornon-users, wecom putesummarys tatisticsov eralltheyears inoursample.For smartphoneusers, instead,w euseonly infor mationuntiltheir firstsmartphonetrade.Theref ore,tr ading statisticsfor adoptersdono treflectthe effects ofsmartphones.Compare to non-users, adopterstrademorefrequently(10vs.fivetradespermonth)andplace largertrades(4,477 vs.3,813 eurosin av erag etrades).Smartphoneusersarealsomorelikel ytobuy riskier assets(95%vs.92%)andpurchase morevolatileassets(22%volatilityvs.16.52%).Finally, adoptersdispla yahigherprobability ofbuying lotter y-type assetsand inv estmentsinthe topdecile oft hepas treturndis tribution. PanelB reportsin ves tor-levelcharacteristicsforsmartphoneusersandnon-users. While thereareno differences inter msofincomeand wealt h,adopterstend to bey ounger 10 males withshortertenure att hebank. 9Specifically,smartphoneusers hav eone year shorter tenureat the bank,areabout 8yearsy ounger ,and 13%more likelyt obemales comparedt onon-users.

3.2 EmpiricalChallengesand Methodology

Investigatingtheeffects ofnewtechnologies ontradingactivity posessignificant em- pirical challenges,becauseof selectionand substitution effects. Individualswho use smartphones totrade couldbedifferent fromin ves torsthat useo therplatforms.Inour sample,sm artphoneusersaremoreactive,morelikelytobuyhigher volatilityandlottery- type assets,and to chasepastt opperf ormers.Thesediff erenceshighlightthe importance of conductingwit hin-investoranalyses.Whileawithin-inv estoranal ysiscould address thistype ofselection, inv est orcharacteristicscouldalsochangeov ertime.Forinstance, individualscanbecomemoresophisticatedorstarttradingmoreactivelyovertime.These changesmight drive theirchoiceof thetrading platfor m.Theref ore,theselection effects could operateatt hein vestor -timelevel. Thankst otherichness ofourdata,w eare ablet ogo ones tepfurtherin addressing thisconcer n.Weexploit withinindividual-by -timev ariation,byincluding inourestima- tions individual-by-month(orby-y ear)fix edeffects.Bycomparingtrades acrossdifferent platformsmadeb yt hesameinv estorwit hint hesamemonth(oryear),wecan accountfor time-varyinginv estorcharacteristicsandselectionatt heinvestor -timele vel. Specifically , wees timatethef ollowingmodel: H

8-9-C=(<0AC?>=48-9-C¸"8-C¹"8º ¸&8-9-C(1)

whereHmeasuresbehaviors(suchasrisk-taking,preferenceforlotterystocksandtrend9Income andw ealthbinsaremeasuredat the timein ves tors begintheirrelationship with thebank.

11 chasing)byinvestor8usingplatform9duringyear-monthC.(<0AC?">=48-9-Cisanindicator variableequal to oneforin ves tor8forsmartphone trades inmonthC.8-Care investor- by-month(year)fix edeffectst hataccountfor time-varyingunobser veddifferences att he investorlevel.T oevaluatethe importanceofacross- andwithin-inves torheterogeneity in our setting,w ealsoestimate the modelwithoutan yfixedeff ectsand withthe inclusion of investorfixedeffects( 8) forallour mainresults. Robus ts tandard errorsaredouble- clusteredat the invest orandyear-monthlevel. Fores timatingthesereg ressions,wecollapse oursamplet othein ves tor bymonth bytr adingplatform level.Fort hispurpose,wecategorize thetrading platformsin two groups:smartphones vs. allother devices. Aftercollapsingthedata, theunitof analysis in ourreg ressionsisthe meanv alueofalltr adesb ythesame inv est orinthe samemonth, using thesametr adingplatf orm.Thises timationstrategyallow sus tocontrolf orboth across- andwit hin-inves torheterogeneitythatmaybiases timates(byha vinginves tor- by-timef.e.),while allowing trades withinthe sameinves torandt hesamemonth to be correlated (bydoubleclus teringt hestandard errors). Thereisapotentialtrade-offwhenusinginvestor-by-timefixedeffects.Wegainbenefits intermofidentificationattheexpensesofexternalvalidityofourresults.Withthesefixed effectsw ecanachiev ebetter identificationbyaccountingfor time-var yingunobser ved differencesat the invest orlevel.Nonetheless,ourresultscome onlyfromthosein vestors thattr adeusingbot hplatf ormswithinthe sameyearormonth. Theseinves torsmight or mightno tbearepresentativ esam pleof allthesmartphone traders.To betr ansparent about thistrade-off, werun allourmajor analysesusingdiff erentspecifications. Firs t, wereport resultswit houtan yfixedeff ect.Then,we includeinves torandtime fixed effects.Las t,weintroduce resultswithin ves tor -by-year andinvestors-by-monthfixed effects.As we introducemoreandmore restrictiv especifications, we mov etowards better identification butpossibl yawa yfrommoreexternalvalidity. 12 Anotherconcernwhenestimatingtheeffectsofnewtechnologiesisthatinvestorscould use thenew platformt oexecutespecifictypesof trades(e.g.,buying riskierinves tments), substitutinga wayfromotherplatf orms. Inthepresenceofsubs titutioneffects,wemight mistakenlyattributevariationintradingstrategiestotheuseofsmartphones,whenindeed investorsarejustreallocating their tradesacrossplatf orms. Totest forthis possibility, weconduct adiff erence-in-differences analysis,exploitingthes taggeredintroductionof mobile appsacross different operatingsystems (iOSvs.Android). Bycomparingnon- smartphonetradesf orsmartphoneusersbef oreand after thelaunchof different trading apps, wecanes tablishho wprevalent spilloverand substitutioneffectsare.Section 5 discusses thisanalysis anditsresultsin detail.

4 MainR esults

Weexamine the associationbetweent heuse ofsmartphonesandt hreetradingbe- haviors:risk -taking,preferencesf orlottery -typeassets,andtrend chasing.Asdiscussed in Section 2 , theeffects ofsmartphonesont hesebeha viorsare not obviousex-ante.By facilitatingmore timely informationacquisition, smartphonescanreduceparticipation costsand, theref ore,increasefinancialrisk-taking.Analogousl y, by reducingmonitoring costs,smartphones canpromo temore efficienttradesand potentiall yreduce investment biases. Smartphones,ho wever,providealsoubiquitousaccessand highspeedofex e- cution oftr ades.Thisconstant feedback mightdiscourage risk-taking,ifin ves torsare verysensitive totheir losses(asint hemyopic loss-av ersionfr amew orkby Benartziand Thaler,1995). Additionall y,accessanywhereandanytime mightf ostermoresys tem1 thinking(Kahneman,2011). System 1haslongbeen associatedwit hmore intuitive and impulsiveactions,pref erencesf orgamblingand, consequently,lo ttery-typeassets and could exacerbatebiases, suchas trendchasing. 13

4.1 RiskT aking

Wefirs tanalyzet heeffectsof smartphonesonfinancialrisk -taking.Intable3, we report resultsf orthisanal ysis,estimatingdiff erentversionsof Equation1. Ouroutcome is anindicat orvariablet hatcapturest heprobabilityofpurchasing riskyassets.We define asrisky assetsdirect andindirect st ockholdings, that isstocksand equitymutual funds. Bonds,bond fundsor gold-relatedfunds aretreated asnon-equity inv estments. 10 In Column(1) we donotinclude fixed effects.Int hisspecification wefindt hatthe probability ofpurchasing riskyassets isfiv epercentag epoints (pp)higherfor tradesdone using smartphonesrelativ etoo thertrades. Thiseffectcorrespondsto anincreaseof5.2% oftheunconditionalsamplemeanforsmartphoneusers(0.95).Whilewefindasignificant effectof smartphones,unobser vable (tous)heterogeneitybetw eensmartphoneusersand non-users candriv ethisresult. InColumn(2),w econtrol for time-inv ariantin vestor heterogeneityb yincludinginv est orfixedeffects.Wealsoaccountf ornation-widetime trends byincludingy earfix edeffects.Consis tentwitht hesefactors playingarole, our estimatesare smaller-2.11% ofthesam plemean-but stillstatis ticallysignificantat 1% level. Our estimatesinColumn (2)could alsobe biasedbecause ofomitted time-var ying investorcharacteristics.F orexample,inves torriskpref erencescouldvar yovertimeand thisv ariationcouldbecorrelated with the decisiont oadoptsmartphone trading.We control forthis possibilityinColumn(3) by includingin ves tor -by-year fixed effectsin our estimation.Thisspecification compares trades donebyt hesameinv est orwithin thesame year ,usingsmartphonesversuso ther platfor ms.Usingthis specification,we find thatinv estorsarethreeppmorelikel yt opurchasearisky assetwhen trading using

smartphones. Finally,inColumn(4)w euse ourmos ts tringentspecificationb yincluding 10In thisanalysis, weomittr adesinot herassets suchascertificatesand warr antsthatcannotbe easily

classified. 14 investor-by-monthfixedeffectsandcomparingtr adesdoneby thesamein ves torwithin thesame year -month.Followingthediscussion insubsection 3.2 , werecallt hatwhile thismore string entspecificationallowsf orbetter identification,theseresultsarebased solelyon those invest orsthatexecutemultipletradesacrossdiff erentplatforms during thesame month. Usingthisspecification, we findthatt heprobability ofpurchasinga risky assetincreases by fourpp-4.3%of thesample mean-whenusing the smartphone versuso therplatforms. Since theunconditionalmean ofpurchasing riskyassets for smartphoneusers ishigh (0.95), theeffects previouslyes timatedmightnotfully capturethe increasedrisk taking induced bysmartphoneuse. Therefore, we usethev olatilityofthe assetspurchased as asecond complementar ymeasureofrisk-taking.W emeasure this volatilityast he annualizedstandarddeviationofreturnsovert hepast12months.Wereportt hevolatility results inT able 4 . Usinga specificationwit houtan yfixedeff ects(Column1),w efindthat thev olatilityofassetspurchased usingsmartphones is12.07pp highercom paredt ot he volatilityof ot herassets.Thismagnitudeiseconomicall ylar ge asit correspondst o54.8% of thesample mean.Howe ver ,bothacross-andwithin-invest orheterogeneitymight drivet hisestimate.When wecontrolf orbo thinv est orandyear fixedeffectsinColumn (2), weestimate asmallereffect for smartphones,equal to4.43pp.In ourmosts tringent specification inColumn (4),w efind thatvolatility ofassets purchasedusingsmartphones is 9.28pphigher than thevolatility ofother assetspurchasedby the sameinvestor within thesame year -month.Thismagnitudeiseconomicallylar ge asit correspondst o42.2%of theunconditional mean.

4.2 Preferencesfor lottery-typestock s

Wes tartthein vestigation ofpreferencesforlottery-typeassets by studyingthe skew- ness oft heassetspurchased.R etailin ves torsgener allypreferpositivelyskew edassets 15 (e.g., Kumar,2009).We presentt heseresultsintable 5. InColumn (1),w efind thatsmart- phone useincreases the skewnessof investmentsby 19.23ppor 33.4%ofthestandard deviationof the skewnessf orphoneusers(57.58).Asin previoustables,t hisfirs tcolumn does notincludean yfix edeffects.When weaddfix edeffects,w ees timatesmaller,but still economicallyand statis ticallysignificantresults,consistentlywit hpre viousresults. For example,in Column(4) we findt hataftercontrolling for invest or-by-mont hfix edeffects smartphone useincreases ske wnessofassetpurchasedby 14.40,or 25%of the standard deviationof the skewnessf orphoneusers. In table6, wemeasuremore directly preferences forlotter y-typeassets.F ollowing Kumar(2009), we defineaslotter y-type those assetsthathaveint heirassetclassesbelo w median prices,abo vemedianvolatility, andabo vemedianskewness. InColumn (1),we find that-withoutincludingany fixed effects-smartphonetrades increasetheproba- bility ofpurchasing lotter y-typeassetsby10pp,or 83%of theunconditionalmean for smartphoneusers.W estillfindstatisticallyandeconomicallysignificantresults,evenafter theinclusion oft hesame fixedeffects previousl yused.Under themost restrictivespeci- fication withinv estor-by-monthfixedeffects,wefindthatsmartphonetr adesincreasethe probability ofpurchasing lotter y-typeassetsby8pp,or 67%of theunconditionalmean.

4.3 TrendChasing

Smartphones allowinv estorstoaccessinformationon theirinv estments onamore timelybasis. We investig ateifsmartphoneinfluencethetendencyofinv est orst ochase pastretur nsandbuy"ho t"in ves tments,orassetst hathaveperformedunusuall yw ellin therecent past. Inourov erall sample 68%ofpurchasesinvolve assetsthatear nedabo ve median returnsint herecent past.Even before adoptingthe smartphoneapp,usersha ve

17% oft heirtradesconcentr atedinthe top10th percentileof pastperfor mers.

Intable7,wefindthatsmartphonetradesincreasethistendencyofbuyingassetsintop 16

10thpercentile ofpas tperf ormance.Wit houtfixedeffects,in Column(1),wefindt hatthe

probabilityofbuyingpastwinnersgoesupby16.4pp.Aftercontrollingforindividual-by- monthfix edeffects,w estillfind aneconomicallyand statistically significantresult. Smart phone tradesincreaset helik elihoodofpurchasingpas twinnersby 12.0pp or70.6% of theunconditional mean. Overall,ourresults sugges tt hatsmartphonesaffectinv estortr ades.Ev encomparing tradeswit hinthesame invest or-mont h,westillfindthatinvestorsbuy morevolatile and highersk ewnessassetsusingsmartphones.These tendenciesresult ina significant increase int heprobabilityofpurchasing lotter y-type assets.Moreo ver,investors become significantlymore likel ytochasepast returns.

4.4 Doin vestorssubstitutetheirtradesusing smartphones?

While ourwit hininves tor-timeanalysesmakeprogressinaddressingpo tentialselec- tion problems,in vestorsstillendogenouslydecide whichtradingplatf ormtousef oreach of theirtrades. Theycanpredominantl yexecuteon smartphonest heirhigh-volatility, high skewness,lottery -typeoftrades.Inthiscase, smartphonetr adesarejust substituting tradest hatwouldha veoccurredan ywayindifferent platfor ms.Inthepresenceofsubsti- tution effects,we shouldexpectnon-smartphonetr adest odispla ylo wervolatility ,lower skewness,andt obe lesslikely to inv olvelottery -typeassetsorpas twinners.Ourdata withinformationonbothsmartphoneandnon-smartphonetradesallowustodirectlytest forsubs titutioneffects. Toidentify these spill-overeff ects,weuseadiff erence-in-differencesapproacht hat exploits thestagg eredadoptionofthe smartphoneappby different clientsofthe two banks.This empirical approachallows ust ocomparedifferentusers before andaftert hey startusing the tradingapp.In practice,int hisem piricaldesign wecompare earl yvs.late smartphone users.Em pirically,weestimatet hefollowing equation: 17 H

8-C=(<0AC?">=4*B48-C¸8¸C¸&8-9-C(2)

whereHmeasures risk-taking,volatility ,skewness,preferences forlottery-type assets and pastwinnersf ortr adesinnon-smartphone platformsbyin vestor8during year-month C.(<0AC?">=4*B48-Cis anindicat orvariableequal toonef orin vestor 8in themonths followingthersttradeusingthesmartphoneapp.8representsinvestorxedeectsthat control fornontime-v arying unobserveddierencesat theinvest orlevel. Crepresents year-monthxede ects. Wepresent these estimatesintable 8, panelA .Thecoecientof interest, , ispositiv e forall outcomesand statis tically signicantforfouroutoft hev eoutcomevariables (witht hesoleexcep tionbeing theprobabilityofbuyingrisky assets).Af terusingthe smartphone app,in vestorsstartbuyingalsoonnon-smartphone platformsassets with higher volatilityandmore positive ske wness,andbecomemore likelytopurchase lotter y- type assetsand past winners.Althoughsmaller ineconomic magnitudethanour main eects,w endpositive spillov ereectson non-smartphonetrades.Thisevidence goes againstsubstitution eectsandt hehypot hesist hatinves torslargelyselectsmartphones toex ecutetheirhigh volatility,high ske wnesstrades.Theseresultsare consistent with investorslearningfromsmartphone tradingto ot herplatf orms. Apotentialconcernwiththisdesignisthatinvestorsendogenouslychoosetoadoptthe smartphone tradingapp.In ot herw ords,thisanalysis suersfromthepo tentialselection eectsbetw eenearly andlateusers.Too vercome this limitation,w erunan additional dierence-in-dierencesanalysis thatexploitst hestaggered launchof tradingappsf or dierentsmartphone operating systems(iOSv s.Android).??This empiricalapproach

allowsust ocom paredierentusers beforeandaf terthetr adingapp fortheir smartphone??This datais only availablef oroneofthetw obanksin oursam ple.Hence, welimitt hisanalysist othis

one bank. 18 operatingsys temislaunched.In practice, we estimate thefollo wingequation: H

8-C=0(<0AC?">=4!0D=2"8-C¸0

8¸0C¸&0

8-9-C(3)

whereHmeasures ouroutcome ofinteres tf ortradesin non-smartphone platformsby investor8duringyear-monthC.(<0AC?">=4!0D=2"8-Cisanindicatorvariableequaltoone forin vestor8in themonths followingt helaunchofthetr adingappfor their smartphone operatingsys tem.0

8represents investorfixedeffectsand 0Crepresents year-monthfixed

effects.W epresentthese estimates inpanelBof table 8 . Consistentwith theresultsin panel A,we alsofindpositive spillov ereff ectsforall theoutcomevariables,wit hthreeout of fivevariables beingstatisticall ysignificant. The identificationassum ptionforthis analysisist hatofparallel trends,i.e.int he absence oft heapplaunch,t hetr adingbeha viorofinv estorso wningdifferenttypes of smartphones-iOS vs.Androidde vices-would haveevol vedinaparallel way.Al- thought hisassumption cannotbefullytes ted,weexamine itsv alidityinthe pre-period byes timatingthedynamics ofsmartphoneeffects ov ertime. Figure 4plotst hecoefficients of specificationsin whicht hesmartphone typeisinteracted with ev ent-timein quarters. Weplo testimatesf ortheprobability ofpurchasingriskyasset (panelA),volatility (panel B), skewness(panelC),and trend-chasing(panel D).A crossall outcomes,w efind no statisticallysignificantdifferences for invest orsowningdiff erentsmartphonesinthetwo- yearperiod before theapplaunch. Afterthe launch,w edo notdetectneg ativeeffects, a findingt hatisinconsistent with substitutioneffect. Ifanything, weobserv edelayed positivespillo vereffectsonnon-smartphone trades.Int hisspecification, the effectson non-smartphone tradesarefurt herdela yedby thefactthat notallt heinves torsstart using theappimmediatelyafteritslaunch.Moreover,ifinves torslearnfromsmartphonetrades, spillovereffects couldtaketime to manifest.Ov erall, thisevidence isconsistentwithin- 19 vestorslearningfrom smartphonetradesand adopting similarbeha viorsalsowhenno t tradingusing smartphones. Overall,ourresults areinconsis tentwit hsubs titutioneffectpla yingarole.If anything, our evidencesugges tthatthereare positivespillov erandthat inv estorslear nfromtheir smartphone trading.

5 Mechanism

In thissectionw ein vestigate whatdrivesthedifferentialtradingbeha viorassociated withsmartphones. Firs t,wetestif usingsmartphonesto tradeatspecific timesof the dayor to tradespecificassets canexplainourresults. Then,w es tudyif digitalnudg esor thede vicescreensizeg enerate ourresults. Last,weinv estig ateif smartphoneeffectsare short-livedor moreper manent.

5.1 Doin vestorsusesmartphonestotrade during differenthours?

Smartphonespotentiallyallowanimmediateaccesstotradingoveranextendedperiod oftime.T oevaluateifthisextendedaccesstotradingdrivesourresults,wefirstinvestigate tradingdynamics ov erdifferenthoursoft heday. Infigure 2 panel A, weplot thedensity of tradesperhour oft heda yf orourentiresam ple,includingbot hsmartphoneandnon- smartphoneusers.There aretwopeaksintradingactivity.Theycoincidewiththeopening (9:00 to10:00am)and the closingof thefinancialmar kets inGer many(4:00t o5:00pm). In panel Bw eplott hesamedensitysepar atelyfor smartphoneand non-smartphoneusers. The twodensityplo tslar gelyo verlap,withsmartphoneusersmar ginallymorelikelyto tradearound closinghours. Inpanel C,w elimit ouranal ysist osmartphoneusersand plotsepar atelytheirsmartphonev s.non-smartphonetrades. Ag ain,thereis noapparent differencein the twodensityplo ts.Traders usewit hsimilarfrequencysmartphones and 20 othertrading platformsduring theday.

In table

9 , weinv estigatemoreformallythe effects oftradinghoursonourresults, by including inour analyses bothin vestor-by -monthandtradinghour-by-yearfixedeffects. This specificationallo wsustocom parealso tradesmadeduring thesamehour oft he day(e.g., 9:00am) inthesame year .Allourpre viousresults arerobusttot hisadditional specification. Investorsonsmartphonearemorelik ely to buyrisky ,lo ttery-type,and top-performingassets,andin ves tin morevolatileandhigherskewness assets.Com pared toour previous resultsintables 3 t o7, theeconomicmagnitudes areattenuated. They rangefrom35% oft hepre viouses timatefort heprobabilityofpurchasing riskyassets (1.4pp vs.4pp)t o52.6% forthe volatility oftheassets purchased(7.6%vs. 14.4)%). All theresultsremain economically significant.F orexample,t heprobability ofbuying lottery-typeassetsviasmartphone increaseb y3.2 pp,or 26.7%of theunconditionalmean forsmartphone users(12%). Althoughin vestorsdonotusesmartphonesmore frequently thanot herplatf orms at specific hoursof the day,t heeffectsofsmartphoneson tradingappearmitig atedwhenwe comparetr adesexecutedduring similarhours(by includingtr adinghour -by -yearfixed effects).This evidencesuggeststhattheeffectsofsmartphonesmightvaryacrossdifferent hours oft heday. Wedirectlytest thishypo thesisby rerunningourmain specifications separatelyfor tradesduringmar ket-hours(9amt o5pm) vs.tradesduringaf ter-hours (5pm to10pm).W edefine theafter -hourwindo wbasedont hefactthatlocal German marketmakers allowinv estorst otradebetween5pmand10pm,ev enifnationals tock exchangesareclosed. We reportt heresultsoft hisanal ysisintable10. Theeff ectsof smartphones vs.ot hertradingplatforms aresignificantlys trongerduringafter-hours (panel B)as compared tomark et-hours(panelA).A veragingacrossalloutcomes, our estimatesare 80%higher duringaf ter-hours, ranging froma27%increasefor skewnessof assets purchasedt oa175%increase for the probabilityof buyinglottery stocks. 21
Strongereffects duringafter-hours areconsis tentwithsmartphonesfacilitating trades based moreon system 1thinking(Kahneman,2011). Duringaf ter-hours,inv est orsare more likelytobe outofthe wor kplaceand inmore informallocations suchathomeor at restaurants.Moreover ,laterinthedayin ves tors arealsomoreprone tothe effectsof decision fatigue(Baumeister etal.,1988).F ort hesereasons, inv estorscould bemorelikel y torel yonthe moreimmediate andautomaticsys tem1 thinking andt oavoid system2 thinkingt hatrequiresmoreconscious effort, energy andattention. Smartphonesappear tof acilitateorfos tert hishigherrelianceonsystem 1. Apotentialconcernwiththisinterpretationofourevidenceisthatinstitutionalfeatures couldbesystematicallydifferentwhentradingduringmarkethoursvs.after-hours,when marketsareclosed. Thesediff erentins titutionalf eatures-andnota higherrelianceon system1-could drive ourresults.T ohelp addressthisconcer n,wer una falsification testb yestimatingsmartphone effectsint hemor ning,between8am and9am. Duringthis hour marketsarestill closedin Germany. Nonet heless,ear lierinthemorninginv estors are lesslik elytobein morerelaxeden vironmentsand shouldno tsuff erdecisionfatigue. If institutionalfeatures driveourresults wewould expectt ofind similarresultsduring after-hoursandt hismor ninghour.Alter natively, ifhigherrelianceon system1drives our results,w ewouldexpect strongersmartphone effects duringafter-hours.Consistent witht hislatterinterpretation, we documentinpanelC thatsmartphoneeff ectsare ver y similar int hemorninghour andinmark ethours, butw eaker thantheeffects oftr ades during after-hours. Althoughinvestorsdonottradeviasmartphonesmorefrequentlyatspecifichours,the effectsof this newtechnologyare strongerin trades duringaf ter-hours,wheninves tors relymore onsys tem1 thinking.Collectivel y, thisevidence suggeststhathours-of-the-day effectscan contributet oexplain, butnotfull yaccount for ourevidence.That is,even withint hesametrading hours,in vestors aremore likelytobuyriskier,lotter y-type, and 22
hotassets.

5.2 Doin vestorsusesmartphonestotrade different assetclasses?

Investorscouldusesmartphonest otr adespecific assetclasses. Thisselectioneffect could driveourresults. We test forthis possibilitybyincluding inourmainspecifications asset-class-by-yearfixedeff ects.Fort hisanalysis, weclassifyassets intosixcategories: individual stocks,bonds,mutualfunds,w arrants, certificates,and options. Whilethe economicmagnitudesareattenuated,smartphoneeffectsareeconomicallyandstatistically significant alsoin trades withinthe sameassetclass,in thesamey ear. For example, the volatilityof the assetspurchasedincreasesb y2.5% or11.4% oft heunconditionalmean forsmartphone users(22%). Analogously ,t heprobabilityofbuyinglotter y-typeassets increases by2.4pp, or20.0% oft heunconditional mean.Alt houghim portant,asset-class effectscanno tfullyaccount forourresults. Even withinthe sameasset class,in vestors when usingsmartphones aremore likel yt opurchaseassetsthatare riskier,with lotter y- type characteristics,andthat hav erecentlyperf ormedveryw ell.12

5.3 Dodigit alnudgesdr iveourresults?

Choice architectureand nudges cansignificantlyaff ecteconomic decisions,fromper- sonalinvestmentstosavingforretirement,fromcreditcardstomortgages(forareviewsee ThalerandSunstein,2008).Smartphoneappsarev eryeffectiveinnudgingconsumersand changing theirconsump tionandspendingbehaviors(Le viand Benartzi,2020; D"Acunto, Rossi,and Weber ,2020).Analogously,in ves tingappscaninfluencebehaviors byusing

push notificationsorb ygiving moresalienceto specificinf ormation. For example,the 12When werun specificationswithbo th hour-of-the-day andasset-classfixedeffects, wefindsmallerbut

stilleconomicall yandstatis tically significantsmartphoneeffects.Wereportt heseresults int heAppendix

Table A3 . 23
Robinhoodtr adingappprominently features thewinningand losingstock softhe previ- ous day.13Welch(2020) andBarber etal. (2020)document that Robinhood inv est orsare more likelytobuy topwinnersand top losers.Thusprominently displaying "top mover" stocksinthe appcould contributetog enerate these trading patterns.Similarl y,inour setting informationdisplayed inthesmartphoneappcould mechanically generatetr ades thatf avorriskierandlottery -typeassets, andpas twinners. Todirectl ytestf ortheeff ectsofdigitalnudg esonourresults, wewould needt o observehow informationis displayedonthe mobileapps vs.onother platfor ms.This informationisonl ypartiall yavailable tous.14Therefore,w eovercome thisdatalimitation byr unningafalsification test. Giventhat smartphoneappstendt oprominentlyf eature onlyindividual st ocks,weinvestig ateifoursmartphone effectsarepresentalsoinother asset classessuch asmutual fundsand options. Ifdigital nudges driveourresults, we wouldexpect smartphoneeff ectst obestrong eroronly presentin individualstocksand weakerorno tpresent atallint heo ther assetclasses. Our findingsin table11document thatthe inclusionofasset-class-by -year fixed effects does notfully explainourresults.While this evidence isconsis tentwit hnotone specific asset classdriving ourresults, we cannot ruleoutt hatsmartphoneeff ectsarestrong erin individual stocksvs.ot herassetclasses.In table12, wemoredirectl ytes tfort hesehetero- geneouseff ects.Inpractice, we runourmain specificationswithin vestor -by -month fixed effectssepar atelyfordifferent assetclasses:individuals tocks,mutual funds,and ot her classes (certificates,op tionsandwarr ants).In panelA,wedocument that smartphone effectsare economically andstatisticall ysignificant forallouroutcomesintrades related

toindividual st ocks.Moreimportantly,in panelB, wedocumentsimilarlystrong-if 13Undert herecentnew s,R obinhooddisplaythe" TopMov ers" listwhichpresents the four stocks with

highestabsolute return sincethemar ket closeofthe previousday.By clickingon the "

Show More

" option, thein vestorscouldseeanexpandedlist oft he20 st ocks withthe larg estpricemo vements.

14While weareable to observ ethecurrent appforoneoft hetwobank s,w edon "tknow the infor mation

displaywhen the appwasfirs tintroduced andifany meaningfulchangehas happened. 24
anythingstrong er-effectsfortradesinmutualfunds. Thisevidencesugg eststhat digi- tal nudges,suchas saliently featuring winnerstock s,arenot likel ytodriveourresults. Evidence fromtr adesinot herasset classessuchasoptionsand warr antsconfir mt his interpretation.Alt houghoursample islimited tofe wt housandobser vationsandonl y one oft heresults(volatility ofassets purchased)isstatis tically significantat the 1%le vel, ourpointestimatesareallpositiv eandsimilarin magnitudetotheestimatesinotherasset classes. Collectively,thesefindings suggestt hatdigital nudgesdonotdriv ethesmartphone effectsw edocument.Onecould argue that ev enifthese nudgeswere tomechanically driveour results,t hey arefeaturesofthe smartphoneapp and,ultimately, just thechannel throughwhich smartphonesinfluence trading behavior .Whiledocumentingt hischan- nel wouldstill beinteresting,sho wingt hatsmartphoneshav eeffectsabo veandbe yond automaticnudg eshasmoreprof oundim plications.F irst,given thateachsmartphone app hasspecific features andpotentially emplo ysdifferentnudges,our results-notbeing drivenb yanyspecific nudge-aremorelik ely togener alizet osmartphonetradingapps in general.Second,the policyim plicationsarestar kly different.Ifdigital nudges drive tradingbeha vior,regulatingthemcould limitt heeffectsof smartphones.Alter nativel y,if thesenudg esarenot the onlymajor driveroftr adingbehaviors,an ypolicyinterv ention regulating thechoicearchitecture int heseapps mightno tbeaseff ective ashoped.

5.4 Doesdevice screensize driv eour results?

Smartphoneshaveasmallerscreen,whereinformationcanbemoredifficulttonavigate and wheremore prominentf eaturescan capturemuchof the invest orattention. This physicalattribute ofsmartphones canexacerbate existing trading biasesor createnew ones (forare view seeBenartziandLehrer,2015). Therefore, we test ifsmartphonesmaller screen sizecontributes to ourresults. 25
Fort heyears2010 to2014f orone bank,we canobser veiftr adesoccur through smartphone (iPhone),iP ad,ordesktop, thus providingvariation inthede vicescreensize. In thisanalysis, weestimate theeffect ofsmartphonesandiP adsseparately bycomparing themt oother platforms.15Wereport ourresults intable 13. Inpanel Aw einclude individual andy earfixedeff ects,whileinpanel Bweinclude only year fixed effects. We do nothav eenoughpowert oinclude individual-by-monthfix edeffectsas inour previous analyses,because suches timatesw ouldbebasedonl yonthose inv est orswhotrade int he same monthusingat least three platforms, thatisa smartphone,aniPad,andadesktop (or otherplatform). TheestimatesinpanelA areless restrictiveas the yuse only variation from thoseinv estorswhomakeatleastone trade acrossthet hreediff erentplatforms anytimeduring oursam pleperiod. Usingthisspecification, we findt hatboth iPhones and iPadsincreaset helik elihoodofbuyingriskier andlottery -typeassets, andtrend chasing. Themagnitudes arev ery similarforriskyassetsand, possibly ,stronger iniP ad tradesf orlottery -typeassetsandpastwinners. The estimatesinpanel Aare identifiedb ycom paringtr adesof thesamein vestors across deviceswith differentscreensizes. Nonetheless,in ves tors thatusethreediff erent platformscouldbe anon-representativ esam pleof theot hertradersat the twobanks. In otherwor ds,gainsinterms ofidentificationcouldcome att heexpenseofexter nalv alidity of theseresults.T oaddress thistrade-off, inP anelBwe includeonlyy earfix edeffects and weexploitbo th within-andacross-individualvariation. Consistentwith outresults in panelA ,wefind alsointhis specificationt hatt heeff ectsofiPhonesand iPad arevery similar acrossall ouroutcome variables. Collectively,thise videncesuggests thatthe smallerscreensizeofsmartphonesdoes notdriv eourmainresults. Ourfindings areconsis tentwit he videncein Liaoet al.(2020)

thatdiff erencesinthe devices "physicalattributes persedonotdriv einves tor behavior in15In ourmain analyses, thesmartphoneplatf ormincludedbo th smartphonesandtabletssuchasiP ads.

26
a peer-to-peerlendingplatfor m.

5.5 Aresmar tphoneeffectstransitor y?

Last,w einvestigatethedynamicsofsmartphoneeffects.Doinvestorsgetexcitedabout thisne wtechnologyandtem poraril ychang etheirbehavior?Oraresmartphone effects persistento vertime?Ifinv est orshea vilyrelyonthisnewtechnology justinthef ew monthsaf tertheadop tion,ourestimates mightovers tatet herelevance ofsmartphone effects.By relying oninves tor -by-timefixedeffects,ourresultsreflectinfactonlytr ading behaviorin those monthswhenin vestors activel yusesmartphonestotrade. Weprovideagraphicalrepresentationoftheresultsofthisanalysisinfigure 4 . Weplot theinter actionofthe indicator forsmartphonetr adesinequation1withindicat orsfort he quarters afterthe adoptionofsmartphone trading.We includein allour specifications investor-by-monthfixedeffects.InpanelA, wereportresults forthe probabilityof buying risky assets.The effects ofsmartphonesares tablefrom the first quarterofusage upt o quarter nineor after wards.Theeffectsonvolatility(panel B)and skewnessof trades (panel C),and probabilityof purchasingpas twinners (panelD) arealsostable ov ertime. Overall,this evidencesugges tsthatin vestors"initial excitementorwillingness toex- periment riskierand moreg ambling-typetr adesviasmartphonesare notdrivingour results. Theeff ectofsmartphonesdoes not appeart obe short-livedandtr ansitor y.

6 Conclusion

Smartphonesrepresentoneofthemostwidelyusedtechnologies,withover250million devicesin the USalone.Larg eonline brokers reportthatover 20%of allretailinv est or annual tradeshav ebeenexecutedusingmobile devices andestimatet hispercentag et o 27
double int henextfe wy ears.16 Using ano veldatasetfromtw olar ge German retailbanks,w einvestigate ifandhow smartphones influencein vestors.Comparingtradesdoneb ythesame invest orin the same monthacrossdiff erentplatf orms,we documentthattraderson smartphonebuy more riskyassets, chasehigher volatility andhigher skewnessin ves tments,and lottery- type assets.Moreo ver,investors aremorelikelytobuypast winners. Weconduct sev eraladditionalanalysestobetter understand themechanismbehind theseresults. Although invest orsarenotmorelikelyto uset hisnewtechnology atspecific hours oft heday, smartphoneeffectsarestrong erduring after-hours.The selectionof specific timesof the dayorspecific assetclasseswhenusing smartphonescontribute-but do notfully explain-ourresults.After usingsmartphones, inv estorss tartbuying higher volatility,highersk ewness, morelottery-typeassets alsoin theirnon-smartphonetrades. This evidencehelpst or uleoutsubstitution effectsacrossdiff erentplatf orms. Collectively,ourevidence sugges tsthatin vestorsmake moreintuitive(sys tem1- type) decisionswhile usingsmartphones. Thistendency leadst oincreased risk-taking, gambling-likeactivity, andmoretrendchasing. Previous studies hav elink edthesetr ad- ing behaviorsto lowerportf olioefficiencyandperformance. Therefore,the conv enience

of smartphonetr adingmightcomeat acos tf orman yretail invest ors.16Sources:https://www.statista.com/topics/2711/us-smartphone-market/;

https://www.cnbc.com/2018/11/29/td-ameritrade-sees-more-people-trading-on-their-phones.html 28

References

[1] Bali,T .,D.A. Hirshleifer ,L.Peng andY.Tang.2019. Attention, SocialInteraction,and

InvestorAttractiont oLotterySt ocksWorkingPaper .

[2] Barber,B.M., andT .Odean. 2002.Online Invest ors:Do the slow diefirst?Review of

FinancialS tudies15:455-87.

[3] Barber,B.M., X.Huang, T. Odean,and C.Schw arz.2020. Attention InducedT rading and Returns:EvidencefromR obinhoodUsers WorkingPaper . [4] Baumeister,R.,Bratsla vsky ,E.,Muraven,M.,andD.M.T ice.1988.Egodepletion:Is theactiveselfalimitedresource?JournalofPersonalityandSocialPsychology74(5):1252- 1265.
[5] Benartzi,S. andJ. Lehrer. 2015.The SmarterScreen:Surprising Wa yst oInfluence and ImproveOnlineBehavior. PenguinBooks, New York. [6] Benartzi,S. andR. H.Thaler .1995. Myopic LossAversion andt heEquityPremium Puzzle.The QuarterlyJournalofEconomics, 110(1),73-92. [7] Buchak,G.,G. Matvos, T. Piskorski,andA .Seru.2018. Fintech,Regulat ory Arbitrage, and theRiseof Shadow Banks. Journal ofF inancialEconomics, forthcoming. [8] Cen,X., 2019.Going Mobile,In ves tor Behavior,andFinancialFragility(December 15,

2019).WorkingPaper , AvailableatSSRN:https://ssr n.com/abstr act=3312411

[9] Choi,J. J.,D. Laibson,and A. Metrick.2002. How doestheInter netaff ecttr ading? Evidencefrominvestorbehaviorin401(k)plans,JournalofFinancialEconomics64:397- 421.
[10] D"Acunto,F.,N. PrabhalaandA .Rossi.2019. ThePromisesandPitf allsof Robo- AdvisingThe Reviewof Financial Studies, Forthcoming. [11] D"Acunto,F.,Rossi, A.G.,and M.Weber. 2020.CrowdsourcingP eerInf orma- tion toChange SpendingBehavior. Chicago BoothResearch PaperNo.19-09,F ama-

Miller WorkingPaper

, AvailableatSSRN:https://ssr n.com/abstr act=3339335or http://dx.doi.org/10.2139/ssrn.3339335 [12] DiMaggio, M.,and V. Yao. 2019.FinT echBorrowers: Lax-Screening orCream-

Skimming?WorkingPaper .

[13] Fos,V., N.Hamdi,A .Kalda,andJ. Nickerson. 2019.Gig-Labor:T rading Safety Nets forS teeringWheels.WorkingPaper . [14] Fuster,A.,Plosser, M.,Schnabl,P. andJ.Vick ery .2018. Theroleoftechnology in mortgagelending.The Reviewof Financial Studies, forthcoming. 29
[15] Gargano,A.and A.Rossi. 2018.Does ItPay toPa yA ttention?,Review ofF inancial

Studies31,12:4595-4649.

[16] Haigh,M. S.and J.A .Lis t.2005. DoProfessionalT radersExhibitMy opicLoss Av er- sion? AnExperimental Analysis, Journal ofF inance60,1:523-534. [17] Jackson,E.,2019. Av ailabilityof thegigeconomyand longr unlaborsupply effects fort heunemploy ed,WorkingPaper . [18] Kahneman,D., 2003.A perspective onjudg ementandchoice.American Psychologist

58(9):697-720.

[19] Kahneman,D., 2011.Thinking, Fas tand Slow.NewY ork:Farrar, StrausandGiroux. [20] Koustas,D.,2018,Consum ption insurance andmultiplejobs:Evidencefrom rideshare drivers.WorkingPaper . [21] Kumar,A.20 09.WhoGamblesintheStockMarket?,JournalofFinance64,4:1889-1933. [22] Levi,Y. andS.Benartzi,2020. Mindt heApp: MobileA ccesst oF inan- cial InformationandConsumerBeha vior. WorkingPaper , AvailableatSSRN: https://ssrn.com/abstract=3557689orhttp://dx.doi.org/10.2139/ssr n.3557689 [23] Liao,L., Wang, Z.,Xiang,J.,Y an,H., andJ. Yang. 2020.Userinterface andfirs thand experience inretail inv esting.The Reviewof Financial Studies, forthcoming. [24] Loos,B., Previtero, A.,Scheurle, S.,andA. Hackethal,2020. Robo-advisers and investorbehavior.WorkingPaper . [25] Linnainmaa,J., B.Melzer ,and A.Previtero,. 2020.The MisguidedBeliefsofF inancial

Advisors,Journal ofF inanceforthcoming.

[26] Stanovich,K.E.and R.F .West, 2000.Individualdifferenceinreasoning: Implications fort herationalitydebate?. BehavioralandBr ainSciences 23(5):645-726. [27] Thaler,R.H., andC. R.Suns tein.2008. Nudg e:Im provingdecisionsabout health, wealth,andhappiness. YaleU niversityPress. [28] Welch.2020.R etailRa w:Wisdom oftheRobinhoodCro wdandthe Covid Crisis.

WorkingPaper

. 30

Figure 1:

Smartphone Usage

This figureplo tsthefr actionoftrades thatoccuro ver smartphonesthroughtime. Panel A plotsthis usagefor theentiresam plewhilePanel Bplotst hisconditional forinv est ors

who uset hesmartphone.123Percent of SmartPhone Trades20102011201220132014201520162017PanelA 5101520Percent of SmartPhone Trades20102011201220132014201520162017PanelB 31

Figure 2:

TradingHourDensity

This figureplo tsdensityfor hourof theday that trade occurs.PanelA plotsthis forthe samplewhile Panel Bcomparest hisdensity forphoneusers versusnon-users.P anelC

plotst hisdensityfor phoneusers andcomparessmartphone andnon-phone trades. .02.04.06.08.1Density0123456789101112131415161718192021222324

HoursPanelA .02.04.06.08.1Density0123456789101112131415161718192021222324

HoursNon-UsersPhone UsersPanelB 32

.02.04.06.08.1Density0123456789101112131415161718192021222324

HoursNon-Phone TradesPhone TradesPanelC 33

Figure 3:

SpilloverEffects: Dynamics

This figureplo tsthedynamics ofthespillo ver effects onother tradesestimatedusing difference-in-differencesregressions. Thefirstdiff erencecomes frombeforeand after launch dateof smartphoneapp whilet hesecond difference comesfromthe typeof smartphone anin vestorowns(e.g.iPhonevs android).Eachcoefficientrepresents the effectof the useofsmartphoneon risktaking by the sameindividual ono ther platforms fordiff erentquartersrelative to thelaunchdate ofthetr adingapp.Theoutcome variables include Probabilityof purchasingrisky assets(panel A),v olatility(panel B),sk ewness (panel C)of assetspurchased, andprobability ofpurchasing top 10percentile perfor mers

(panel D).The confidenceinter vals areplottedat5%le vels. -.010.01.02Probability of Purchasing a Risky Asset-8-7-6-5-4-3-2-1012345678+

Quarters relative to eventPanelA -1012Volatility-8-7-6-5-4-3-2-1012345678+

Quarters relative to event

PanelB

-20246Skewness-8-7-6-5-4-3-2-1012345678+ Quarters relative to event

PanelC -.020.02.04Purchasing Top 10 pctl Performers-8-7-6-5-4-3-2-1012345678+

Quarters relative to eventPanelD 35

Figure 4:

Dynamics ofSmartphone Effects

This figureplo tsthedynamics ofoureffects relative to the firstuseof smartphone.Each coefficient representst heeffectof theuseof smartphoneon risktakingfor different quarters relativeto thefirst use.Theoutcomev ariablesincludeProbabilityof purchasing risky assets(panel A),v olatility(panel B),skewness (panelC) ofassets purchased,and probabilityofpurchasingtop10percentileperformers(panelD).Theconfidenceintervals

are plottedat5% lev els..02.03.04.05.06Probability of Purchasing Risky Assets12345678>=9

Quarters following first smartphone tradePanelA 81012Volatility12345678>=9

Quarters following first smartphone trade

PanelB

101520Skewness12345678>=9 Quarters following first smartphone trade

PanelC .08.1.12.14.16Purchasing Top 10 pctl Performers12345678>=9

Quarters following first smartphone tradePanelD 37

Table1:

SummaryS tats

T

Investing Documents PDF, PPT , Doc

[PDF] 100 minus age rule investing

1. Business

2. Finance

3. Investing

[PDF] 100 minus investing

[PDF] 401k vs investing

[PDF] about investing in bitcoin

[PDF] about investing in real estate

[PDF] about investing in stocks

[PDF] about investing money

[PDF] about investing.com app

[PDF] adblock plus investing.com

[PDF] additional investment besides 401k

12345 Next 200000 acticles