The effect of permanent product discounts and order coupons on purchase incidence, purchase quantity, and spending

(1)

The effect of permanent product discounts and order coupons on purchase incidence,

purchase quantity, and spending

Liu, Huan; Lobschat, Lara; Verhoef, Pieter; Hong, Z.

Published in: Journal of Retailing DOI:

10.1016/j.jretai.2020.11.007

IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.

Document Version

Version created as part of publication process; publisher's layout; not normally made publicly available

Publication date: 2020

Link to publication in University of Groningen/UMCG research database

Citation for published version (APA):

Liu, H., Lobschat, L., Verhoef, P., & Hong, Z. (2020). The effect of permanent product discounts and order coupons on purchase incidence, purchase quantity, and spending. Journal of Retailing.

https://doi.org/10.1016/j.jretai.2020.11.007

Copyright

Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).

Take-down policy

If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.

Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.

(2)

JournalofRetailingxxx(xxx,xxxx)xxx–xxx

The

effect

of

permanent

product

discounts

and

order

coupons

on

purchase

incidence,

purchase

quantity,

and

spending

Huan

Liu

a,∗

,

Lara

Lobschat

b

,

Peter

C. Verhoef

c

,

Hong

Zhao

d

a_Nankai_University,_Nankai_Business_School,_Department_of_Marketing,_Weijin_Road_No.94,₃₀₀₀₇₁_Tianjin,_China

b_Maastricht_University,_School_of_Business_and_Economics,_Department_of_Marketing_&_Supply_Chain_Management,_{Tongersestraat}_53,

6211LMMaastricht,TheNetherlands

c_University_of_Groningen,_Faculty_of_Economics_and_Business,_Department_of_Marketing,_Nettelbosje_2,₉₇₄₇_AE_Groningen,_The_Netherlands

d_University_of_Chinese_Academy_of_Sciences,_School_of_Economics_and_Management,_Zhongguancun_East_Road_No.80,₁₀₀₁₉₀_Beijing,_China

Abstract

Thispaperexaminestheinfluenceofapermanentdiscountstrategyoncustomerpurchasebehavior,i.e.,purchaseincidenceineachweek, purchasequantity(inunits),andtotalorderspending(inCNY).Permanentdiscountsaredefinedasdiscountscontinuouslyprovidedbyretailers. Weidentifytwotypesofpermanentdiscounts,namely,product-specificpricediscounts(PD)andordercoupons(OD,whichcanberedeemed foratotalorder).WecollecttransactionaldatafromaChineseonlineretailerandempiricallyexaminetheeffectsofthetwotypesofpermanent discountsand customers’expectationsofPDand OD.Wefindnonlinearrelationshipsbetweenpermanentdiscountsandcustomerpurchase behavior.PDsnegativelyinfluencespendingwhentheyarelowerthan19%butshowapositiveeffectbeyondthisthreshold,hencedepictinga U-shapedrelationship.Theyalsoaffectpurchasequantitypositivelybutatadecreasingrate.CustomerexpectationsofPDinfluencepurchase incidence,spending,andpurchasequantityfollowingaU-shapedpatterwithapositiveinfluenceappearingwhenPDexpectationsarehighthan 31%,27%,and18%respectively.Ontheotherhand,ODsinfluencespendingandpurchasequantitypositivelyatanincreasingrate.Customer expectationsofODinfluencepurchaseincidence,spending,andpurchasequantityfollowingaU-shapedrelationshipwherethepositiveinfluence onpurchaseincidenceshowsbeyondODexpectationsof426CNY,andthepositiveeffectappearingonspendingandpurchasequantitywhenthese expectationsarehigherthan34CNY.Wealsofindthatcustomerexpectationsofdiscountsinteractwithcurrentdiscountlevelsintheirinfluence onspending.Combiningtheseresultsandconsideringthatordercouponsnegativelyaffecttheprofitmarginofthetotalbasket,wesuggestthat retailersshouldofferordercouponswithrelativelylowvaluebutproduct-specificpricediscountswithhighdiscountdepth.

Keywords: Digitalchannels;Permanentdiscounts;Product-specificpricediscounts;Ordercoupons;Customerspending

Globaldigitalcommerceamountedto$2.79trillionin2019 andis expectedtoexceed $4trillionin2024 (Statista2020). China in particular “is already more digitalized than many observersappreciate.Chinaisoneoftheworld’slargestinvestors andadoptersof digitaltechnologiesandishometoone-third of the world’s unicorns” (Woetzel et al. 2017, p. 3). Digital commerce in Chinaaccounts for more than40% of all digi-taltransactions,promptingmassivecompetitionamongonline retailers(TheEconomist2017).Althoughmorethan2600online

∗_{Corresponding}_author.

E-mailaddresses:huan.liu@nankai.edu.cn(H.Liu),

l.lobschat@maastrichtuniversity.nl(L.Lobschat),p.c.verhoef@rug.nl (P.C.Verhoef),zhaohong@ucas.ac.cn(H.Zhao).

retailers competein this market (ChinaZ.com 2018), the top threedigitalretailersinChina(Alibaba,JD.com,and PinDuo-Duo)account for nearly 80% of total retailsales (eMarketer 2018).Accordingly,manyonlineretailersfail,andanestimated 90%operateatadeficit(iResearch2013).

Forsmall and medium-sizeddigital retailers, the struggle tosurvive drives themtolookfor waystoattract customers, oftenrelyingonpermanentdiscountsforalltheirproductsfor everycustomer. Arguably, customers enjoy deals, and it has beenshownthatdiscountsgenerallyincreasewebsitetrafficand purchase intention (e.g., Gong, Smith, and Telang 2015). To determinewhetherthisstrategyiseffective(andefficient),we explicitlyconsiderhowtwo differenttypesof permanent dis-counts,providedcontinuouslybyretailersindigitalchannels,

https://doi.org/10.1016/j.jretai.2020.11.007

(3)

influencecustomerpurchasingbehavior:product-specificprice discountsandordercoupons.Aproduct-specificpricediscount isofferedonlywhenpurchasingaparticularproduct,whereas anordercouponisnotrestrictedtoaspecificproductbutcan beredeemed forawholeorder. Thesetwotypesofdiscounts are widelyused in digital retailing, especially in China. For example, onlinestores onthe Taobaoplatform routinely pro-videadiscountedpriceforaproduct(i.e.,product-specificprice discount),aswellasdiscounteventssuchas“Spend300 Chi-neseyuan(CNY)andgeta30CNYcoupon”(i.e.,the30CNY couponcanberedeemedagainstthecurrentorder;insome dis-countevents,couponscanbeusedwithoutanylimitations).In the lattercase, the 30CNYcoupon isnot assigned toa spe-cificproductbutcanberedeemedagainstthewholeorder.This papercomparesproduct-specificpricediscounts(forexample, whenaproductwiththeregularpriceof100CNYisofferedata discountedpriceof90CNY)with“amount-off”ordercoupons (forexample,acouponthatgives10CNYoff,withorwithout spendingconditions)thatareprovidedcontinuously.

Mostpreviousstudiesdiscusstemporarydiscountstrategies in traditional channels, such as brick-and-mortar stores(e.g.,

Gedenk and Neslin1999; Horváth andFok 2013;Srinivasan et al. 2004). The present paper aims toadd to this research byaddressingthe followingthreeaspects.First, inour study, we compare differentdiscount strategies inadigital context.

Searchingandcomparingprices(and/ordiscounts)iseasierand requireslesseffortfromconsumersindigitalchannelsthanin physicalstores (e.g., clickingtodifferent websites vs. travel-ingtodifferentofflinestores;Raghubir2004).Manyretailers evenprovideautomaticpricecomparisonswithintheir digital channels.Thestimulatingeffectofdiscountsoncustomer spend-ingmaythusbeweakerindigitalchannelsthanintraditional channels,becauseconsumerscaneasilyfind(orexpecttofind) betterdealsatexceptionallylowsearchcosts(Reibstein2002). Althoughseveralstudieshavelookedatdigitaldiscountsin dif-ferent contexts, including the cross-channel effects of online discounts (Breugelmans andCampo 2016;Gong, Smith, and Telang2015)andtheeffectsofmobilediscountsonpurchasing (Fong,Fang,andLuo2015),yetlittleisknownaboutemerging digitaldiscountstrategiessuchasprovidingonlinepermanent discounts.

Second,ourcurrentknowledgeofdiscountsmainlyconcerns temporarydiscountsofferedduringalimitedtimeperiod(e.g.,

NeslinandvanHeerde2009).However,whenexposedto fre-quent(orevenpermanent)pricediscounts,consumersdevelop

discountexpectationsandmaypurchaseonlyifaproductison discount(KalwaniandYim1992).Hence,ifaretailerprovides permanentpricediscountsforalargepartofitsoffering, con-sumerslearnfromtheirexperienceofobservingdiscountsand purchasingwiththem(Grewaletal.2010)andmaybelievethat discountswillalwaysbeavailable.Theywilldevelopdiscount expectationsbasedontheirobservationsandexperiences,and thismayinfluencetheirresponsestocurrentdiscounts.Thus,it isquestionablewhetherresearchfindingsfromstudieson tem-porary discounts can be generalized topermanent discounts. Therefore,in thepresent paper, we analyzeandcompare the effectsoftwodifferentpermanentdiscounttypes.

Third,moststudiesdiscusshowproduct-specificprice dis-countsinfluencecustomerspendingbehavior(e.g.,Biswasetal. 2013;Fong,Fang, andLuo 2015),andsomepapersfocuson theeffectsofproductline-specificandproductcategory-specific pricediscountsonspending(e.g.,Jiaetal.2018;Huietal.2013). Nevertheless,researchonothertypesofdiscountsisscant.For instance,offeringordercoupons,adiscount strategyrecently appliedby B2Cretailers and/or platforms suchas Taobao, is widelyneglectedincurrentresearch.Hence,thepresentpaper willanalyzeordercouponsthatarenotrestrictedtoaspecific product(oraproductlineorcategory).Inourcontext,similarto

Jiaetal.’s(2018)statementforproductline-specificprice dis-counts,consumersdecidenotonlywhethertoredeemthecoupon butalso whichproduct(s) to buy withthe coupon.This may leadtodifferenttotalbasketspendingamountsacrosscustomers who have different choices of product combinations relative toproduct-specific price discounts. The latter mainlyinduce customerstopurchasethediscountedproducts.Therefore,this studybridgesanotherresearchgapbyexaminingtheeffectof ordercouponsoncustomerpurchasebehaviorandbyidentifying potentialdifferencesintheeffectsofproduct-specificdiscounts. Insum,weseektoaddressfoursalientresearchquestions: (1) Whataretheeffectsofpermanentproduct-specificprice

dis-countsindigitalchannelsoncustomers’purchaseincidence ineachweekaswellastheirtotalspending(inCNY)and purchasequantity(inunits)foreachplacedorder?

(2) Whataretheeffectsofpermanentamount-offordercoupons indigitalchannelsoncustomers’purchaseincidenceineach weekaswellastheirtotalspending(inCNY)andpurchase quantity(inunits)foreachplacedorder?

(3) Whataretheeffectsofcustomers’discountexpectationson customers’purchaseincidenceineachweekaswellastheir totalspending(inCNY)andpurchasequantity(inunits)for eachplacedorder?

(4) Do customers’ discount expectations also influence cus-tomers’responsestocurrentdiscounts?

Byanswering the abovequestions, we are abletoprovide insightsintowhetherapermanentdiscountstrategyleads con-sumerstospendmoreorlessandwhetherproduct-specificprice discounts influence customer purchasingbehavior differently comparedtoordercoupons.Weproposeaconceptual frame-worktodetailpossiblemechanismsbywhichthetwotypesof discountsaffectcustomerpurchasebehavior.Totestour propo-sitions,we rely on data from aChinese e-commerce retailer thatsellsthroughonlineandmobilechannels.Thedatainclude informationabout allorders by 3866 uniquecustomersfrom January1–December 31,2015.Weobserve twotypesof dis-counts:product-specificpricediscounts andamount-offorder coupons.Totesttheeffectsofbothdiscounttypesonconsumers’ purchaseincidenceaswellastheirpurchasequantityand spend-ingforeachplacedorder,weformulateasimultaneousequation systemthatcorrects forthepotential endogeneitybiascaused bythefirm’sdiscountstrategyimplementation.

Wefindthatpermanentproduct-specificpricediscountsand amount-offordercoupons influencespending levelsand

(4)

pur-chasequantityindifferent,nonlinearpatterns.Product-specific pricediscountsnegativelyinfluencecustomerspendingatlower discountlevelsandpositivelyaffectspendingathigherdiscount levels (i.e.,higher than19%), whiletheypositively influence purchase quantity atadecreasing rate. Moreover,customers’ expectations ofproduct-specific pricediscounts,formed from previousdiscountexperiences,exertaU-shapedeffecton pur-chaseincidence,spending,andpurchasequantitywiththeeffect turningpositivewhentheseexpectationsarehigherthan31%, 27%,and18%respectively.Therefore,wesuggestthat retail-ersshouldofferrelativelyhighpermanentproduct-specificprice discountstoattractmorecustomerstospendandpurchasemore. Ordercouponspositivelyinfluencecustomerspendingand pur-chasequantityatanincreasingrate.Customers’expectationsof ordercouponsdemonstrateaU-shapedpatternwithapositive effectappearingonpurchaseincidence,spending,andpurchase quantitypositivelyinfluencewhenexpectationsarehigherthan 426 CNY, 34 CNY, and 34 CNY respectively. Nevertheless, takeninto accountthatordercouponsaffecttheprofit margin ofconsumers’totalorder,wesuggestthatretailersshouldkeep thevaluesofpermanentordercouponsrelativelylow(wewill explainthisinmoredetailintheimplicationsection).Besides, weshowthatcustomers’expectationsofthetwotypesof dis-countsmoderatetheeffectofacurrent(permanent)discounton customerspending.Wesuggestthatretailersshoulddesigntheir permanentdiscountstrategycarefullytoavoidnegativeeffects; to stay competitive,they should also takecustomer discount expectationsintoaccount.

Literaturereview

AsthesummaryinTable1indicates,mostpreviousstudieson discounteffectsdiscuss(1)offlinediscountsand(2)temporary discounts,butignore(3)consumers’previousdiscount experi-encesand(4)otherdiscounttypesbesideproduct-specificprice discounts.Findingsfrompriorresearchrevealpositivediscount effects on purchaseintentions (Grewal etal.1998), purchase quantity (Mela, Jedidi, and Bowman 1998), andrelationship duration(Thomas,Blattberg,andFox2004).YetKalwaniand Yim(1992)cautioned thatconsumersform discount expecta-tionsafterbeingexposedtofrequentdiscountsandthusmaystart to purchase discountedproducts only. In their meta-analysis,

DelVecchio, Henard, and Freling (2006) indicated that price discounts greater than 20% negatively influence sales in the long run, although other studieshave claimed that a moder-atediscountdepth(about30%)ismosteffective(e.g.,Andrews etal.2014; DelRioOlivaresetal.2018).Hence,prior stud-ies also point tothe importance of accountingfor consumer discountexpectationsandfornonlineareffectsofdiscountson consumerpurchasebehavior.Inthefollowing,wewilldiscuss priorresearchondiscounteffectsinmoredetail.

First, themajority ofexisting studiesfocus onoffline dis-counts,althoughonlinediscountsarewidelyappliedinpractice. When online shopping was new, Reibstein (2002) identified price as the most important element for attracting customers toshoppingwebsites,butthesemotivationshavelikelychanged overtime,whichsuggeststhatanonlinediscountstrategymight

inducenoveleffectstoday.ArecentpaperbyValentini,Neslin, andMontaguti (2020)hasshownthat mostomnichannel cus-tomers tend to focus on one channel (i.e., online or offline) toobtain andusepromotions.Inaddition,mobilepromotions haveattracted researchers’attention,because informationcan bedeliveredeasilyviamobilephonesatverylowcostandwith appropriatelocationtargeting(e.g.,Fong,Fang,andLuo2015;

Huietal.2013).Danaheretal.(2015)consideredhowmultiple factorsinfluencetheredemptionofmobilecoupons;theyfound thatbothtraditionalcouponfeatures(e.g.,facevalue,expiration length)andmobilecouponfeatures(e.g.,locationandtimeof delivery)havesignificantinfluence.However,withtheincreased applicationofonlinechannelsinretailing,onlinediscount strate-gies are emerging that we have little knowledge about. For example,astrategyofprovidingdiscountsforcustomers contin-uouslycanbedevelopedinonlinechannelsbecauseofthelower costofofferingdiscounts(e.g.,noneedforpaperbillboards)and thegreaterflexibilityinadjustingdiscountsonlinecomparedto brick-and-mortarstores.Aswenotedabove,giventhe signif-icantfeaturesof continuousdiscounts(e.g., theirpermanence andtheroleplayedbydiscountexpectations),existingstudiesof offlineandonlinediscountsdonotallowustodrawconclusions astowhetherorhowapermanentdiscountstrategyinfluences customerspendingbehavior.Therefore,wefocushereontwo specifictypesofpermanentdiscountsandtheirdistincteffects oncustomerpurchasebehavior.

Second,priorstudiesgenerallyfocusondiscountsthatexist for aspecificperiod of time(e.g., Fanget al.2015; seealso

Table 1). A permanent discount strategy differssignificantly from a temporary one, because the latter evokes a sense of urgency, forcing consumers to purchase during the discount periodtoobtainthediscountbenefit(e.g.,BlattbergandNeslin 1990).Whentheyencounterpermanentdiscounts,consumers learntoexpectdiscounts atthenextpurchaseoccasion(or in thefuturemoregenerally)andfeelnocompulsiontoaccelerate their purchases or to stockpile products. Despitethe superfi-cialresemblance,everydaylowpricing(EDLP),whereretailers chargestable,lowpricesforarangeofproductsona continu-ousbasis(Hoch,Dreze,andPurk1994),isquitedifferentfrom this.EDLPisapositioningstrategy(LalandRao1997) asso-ciatedwithclaimssuchas“guaranteedlowprices”(Ortmeyer, Quelch,andSalmon1991).Thus,itpromisesconsumerslower averagepricesandreducestheirneedtotrackdealsorswitchto competitors.Incontrast,permanentdiscountsarea“pure” dis-countstrategy,withoutanypositioningemphasisorguarantee ofofferingthelowestpricesinthemarket.

Third,manystudieshavedemonstratedthatconsumersform aninternalreferencepricefromtheirpastpurchaseexperiences (e.g.,Mazumdar,Raj,andSinha2005).Accordingly,they per-ceiveagainoralosswhenacurrentpriceisbeloworabovetheir referenceprice(vanOest2013).LattinandBucklin(1989),as wellas Kalwani and Yim(1992),have shownthat price and promotionexpectationsinfluenceconsumerpurchases,andthat ignoringtheseexpectationsleadstobiasedpredictionsof con-sumer decisions. Nevertheless, some recent discount studies havenottaken discountexpectations intoaccount (e.g., Fang etal.2015;Park,ParkandSchweidel2018).Moreover,asthe

(5)

Table1

SummaryofRecentDiscountLiterature.

Paper ResearchContext DiscountType MajorDiscountVariable Channel

Temporary Discounts

Permanent Discounts

OneorTwoTypes Current Discount

Discount Expectations

Offline Digital

KalwaniandYim(1992) √ 1 √ √ √

Brand-specificprice discount

GedenkandNeslin(1999) √ 1 √ √

Srinivasanetal.(2004) √ 1 √ √

HorváthandFok(2013) √ 1 √ √

Brand category-specific discount Jiaetal.(2018) √ 1 √ √ √ Productline-specific coupons

Fong,Fang,andLuo(2015) √ 1 √ √

Product-specificprice discount Fangetal.(2015) √ 1 √ Product-specificprice discount √

Gong,Smith,andTelang(2015) √ 1 √ √

Category-specific pricediscount

BreugelmansandCampo(2016) √ 1 √ √ √ √

Category-specific pricediscount Pastpromotion frequency Danaheretal.(2015) √ 1 √ √ Store-specificprice discount Huietal.(2013) √ 1 √ √ Category-specific pricediscount

ZhangandBreugelmans(2012) √ 2 √ √

Categoryprice discounts;reward pointpromotionswith loyaltyprogram

Park,Park,andSchweidel(2018) √ 2 √ √

Pricediscount coupons;non-price freesamplecoupons

Thisarticle √ 2 √ √ √

Product-specificprice discount;order coupons(notspecific toproducts)

Pastdecaying weightedaverage discountlevel

inherentdiscontinuityofshort-termdiscountsmightnotleadto expectationsaboutsubsequentdiscounts,studiesthatfocuson temporarydiscountswillhavelimitedgeneralizability.

Fourth, previous studies have mainly considered product-specific price discounts (e.g., Biswas et al. 2013), product line-specific price discounts (e.g., Jia etal. 2018), and prod-uct category-specific price discounts (e.g., Hui et al. 2013). Intuitively,sinceallthesediscounts arerestrictedtoaspecific product, a specific product line, or a specific product

cate-gory,their influenceoncustomerspending willresult mainly fromspendingontheseparticularproducts(forexample, peo-plestockpilingdiscountedproducts;Ailawadietal.2007).On theotherhand,ordercouponsarenotrelatedtoaspecific prod-uct (or productline or category), and consumers may try to useanordercouponbyconsideringcombinationsofmultiple differentproducts.Inthiscase,thenumberofproductsin con-sumers’ considerationsets when redeeming ordercouponsis largerthanthe numberof products that areconsidered when

(6)

Fig.1.Conceptualframeworkoftherelationshipbetweenproduct-specificprice discountsandamount-offordercouponswithcustomerspendingbehavior.

Notes:Thedashedlinewithtwoarrowsrepresentsinteractioneffects.

customersredeemproduct-relatedpricediscounts.Hence,order couponsmaygenerateeffectsoncustomerspendingthatare dif-ferentthanthosegeneratedbyproduct-relatedpricediscounts. As Levy et al. (2004) indicated,different types of discounts mayhavedistincteffectsoncustomerpurchasingbehavior.In thisconnection,DelVecchio,Krishnan,andSmith(2007) com-pareddifferentformatsofdiscounts(i.e.,percentageoffversus centsoff),andtheyfoundthat thediscountformatinfluences consumer perceptions of the discount price. Park, Park, and Schweidel (2018) explored the difference betweentwo types of discounts (i.e.,pricediscounts andfree samplesinmobile channels).They foundthat bothtypesof discountspositively influencepurchaselikelihoodandspendingduringthediscount period,withfreesamples alsohaving positiveeffects on pur-chaseincidenceafterthepromotion.Intheircomparisonofprice discountwiththerewardpointpromotionsofaloyaltyprogram,

ZhangandBreugelmans(2012)foundthatconsumersweremore responsivetothelatter.Inthepresentpaper,wewillfocuson thedifferencebetweenordercouponsandproduct-specificprice discountsthatareprovidedtocustomerspermanently.

Conceptualframework

Weproposethe followingconceptualframeworkto under-standthe effects of the two typesof permanent discounts on keycustomeroutcomeswhilealsoconsideringconsumers’ dis-countexpectations(seeFig.1).Weassumethatproduct-specific pricediscountsdifferfromordercouponsintermsoftheir influ-ence onspending andpurchasequantity.Wealso discuss the effects of customer expectations of the two types of perma-nentdiscounts.Customerexpectationsofdiscountsareformed on the basis of previous experiences or observations; once customershaveexpectations, thesewill influencetheir future purchase decisions. Thus, we consider the influence of cus-tomerexpectationsofthetwotypesofpermanentdiscountson purchase incidenceandactualpurchasebehavior (i.e., spend-ingandpurchasequantity).Moreover,as customers’discount expectationsmayinfluencetheirresponsestocurrentdiscounts, weexplorewhetherdiscountexpectationsinteractwithcurrently provideddiscounts.Overall,withinourconceptualframework, weanalyzetheeffectsoftwotypesofdiscountsonconsumers’

purchasequantityandspendingandtheirinteractionswith dis-countexpectations,whilediscountexpectationsinfluenceboth purchaseincidenceandactualpurchasebehavior(i.e.,spending andpurchasequantity).1

Effectsofpermanentproduct-specificpricediscountson purchasebehavior

Pricediscountsofferaneffectivewayforconsumerstoobtain economicsavings,soshopperstendtoincreasetheirspendingin responsetodiscounts(e.g.,Kendrick1998).AlthoughRaghubir (1998)suggestedthatconsumersmightperceivehigherproduct pricesinresponsetoahigherdiscount,otherstudieshaveargued thatconsumersderivefurtherbenefitsinadditiontosavingsfrom discounts,suchasopportunitiestobuyhigher-qualityproducts, abettershoppingexperience,andmeansforvalueexpression, entertainment, andexploration (e.g., Chandon, Wansink, and Laurent2000).Discountsalsomayincreasecustomers’mental budgetsandencouragethemtopurchasemore(Jiaetal.2018).In addition,studieshaveshownthatdiscountshavepurchase rein-forcementeffects(KahnandRaju1991)andthattheyincrease statedependenceovertime(Keane1997).Inthecontextofthe intensecompetitionintoday’smarket,inarecentglobal indus-trystudy57%of firmsreportedbeinginvolvedinapricewar (Simon-KucherandPartners2020).Inthisenvironment, con-sumersmayexperienceavarietyofdiscountsfromwhichthey canderivemorebenefits.

Nevertheless,whenaretailerofferspricediscountsfor spe-cific products continuously (i.e., permanent product-specific price discounts), consumers may recognize that these prod-ucts are always discounted and thus may not feel the need toacceleratetheirfuturepurchasesinresponse(Blattbergand Neslin1990).Moreover,consumersmightmakenegative infer-enceswhendiscountlevelsincrease.Forexample,DellaBitta, Monroe,andMcGinnis(1981)arguedthatadrasticprice reduc-tion might be perceived as exaggerated or fake. Jensen and Drozdenko(2004)foundthatconsumers’perceptionsofproduct qualitydidnotchangeatdiscountlevelsbelow30%. Neverthe-less,ifadiscountexceeds40%,customers’valueperceptions andpurchaseintentionsareundermined.Intheirmeta-analysis,

DelVecchio, Henard,andFreling (2006)noted that discounts greaterthan20%negativelyinfluencecustomerpreferencesfor apromotedbrand.Andrewsetal.(2014)recommendeda30% discountasmost effectivefor increasing customerpurchases, relativetonodiscountor a50%discount. Similarly, DelRio Olivaresetal.(2018)reportedthatdiscountsbetween5%and 35%havepositiveeffectsoncustomerretention,whereas lev-elsbelow5%andabove35%exertanegativeinfluence.Intheir empiricalstudy,Jiaetal.(2018)confirmedaninvertedU-shaped

1 _Note_that_we_observe_particular_{product-specific}_price_discounts_and_order

couponsonlywhentheyareredeemedbyacustomerinagivenweek;thisiswhy wecannotexploretheireffectsonpurchaseincidence.Giventhatweobserve allthediscountsthatacustomerhasexperiencedpriortoagivenweek,weare abletoanalyzetheeffectsofconsumers’discountexpectationsontheircurrent purchaselikelihoodinagivenweek.

(7)

effectofproductline-specificdiscountdepthoncustomer spend-ing.

Taking these considerations together, we thus expect per-manentdiscountstostrengthenconsumers’perceivedbenefits butalsotheirsensitivitytopricesanddiscounts(Mela,Gupta, andLehmann1997).Wethereforeargue,againstprevious stud-ies,thatthenegativeeffectofhigherpricediscountsmentioned abovedoesnotnecessarilyleadtonegativepurchasebehavior. Instead,weexpectthatitwillweakenthepositiveeffectofhigher discountsonpurchasing.

H1. Purchasequantityforaplacedorderwillincreasewithan increaseofthevalueofaproduct-specificpricediscount,though atadecreasingrate.

Theeffectofproduct-specificpricediscountsoncustomers’ spending is relatively unclear. How and to what extent cus-tomerspendingwillbeaffectedwilldependonthecomparison betweenthechangeofpurchasequantityandthechangeofprice duetodiscounts.Itishardtoderivewhichpart(i.e.,thechangeof quantityorprice)overwhelmsanotherfromtheories.Therefore, wedonotstateaspecifichypothesisfortheeffectof product-specificpricediscountsonspendingbutwillofcourseexplore thiseffectinoursubsequentanalyses.

Effectsofpermanentordercouponsonpurchasebehavior

Intuitively, negative inferences caused by high discounts (e.g.,lowerreputationorpoorquality)canoccurinthecaseof ordercoupons.However,consumersmayderivemorebenefits fromordercouponsthanfromproduct-specificpricediscounts. Themajordifferencebetweenproduct-specificpricediscounts andordercouponsisthattheformerrelateentirelytoa partic-ularproduct,whereasthelatterdonot.Therefore,ifconsumers mentally allocate the value of acouponto each productina basket,theymayperceivethatthepricesofalltheproductsin thebasketarereduced;incontrast,aproduct-specificprice dis-countreducesthepriceforoneproductonly.Moreover,some ordercouponsareconditional,suchas“Spend300CNYandget a30CNYcoupon.”Insuch cases,consumersmaysearchfor andpurchasemultipleproductstoachievetheamountrequired toredeemthecoupon(LeeandAriely2006).Ordercouponsof thissortmaythereforestimulatecustomerpurchasequantityand spendingmorestronglythanproduct-specificpricediscounts.

Weassumethatthebenefitsassociatedwithproduct-specific pricediscounts(e.g.,savings,greaterbudgets,andopportunities tobuyhigher-qualityproducts)alsoapplytoordercoupons,with couponsalsohaving apositiveeffectoncustomers’purchase quantity for agivenorder. Moreover,thispositiveeffectwill increasewithhighercouponlevels.Thus,oursecondhypothesis isasfollows:

H2. Purchasequantityforaplacedorderwillincreasewithan increaseofthevalueofanordercouponatanincreasingrate.

Similarly, it is difficult to theorize the influence of order couponsoncustomerspending,whichshouldrelyonthe com-parison between the change in purchase quantity due tothe couponsandthechangeinspendingduetothereductioneffect

of thecoupons(i.e., the higherthe valueof the coupons,the lessmoneyconsumersneedtopay).Hence,wewillexplorethis effectinthecontextofourempiricalanalyses.

Effectsofcustomers’discountexpectations

Whenconsumersobservethataretailerprovidesprice dis-counts and/or order coupons continuously, these permanent discountsmaysignalapoor retailer image(e.g.,low product orservicequality, lowreputation,orpoor management),with theresult thatknowledgeable consumersstarttoquestionthe retailer’sstrategicmotivations(e.g., Biswasetal.2013).This can undermine consumer purchase likelihood (e.g., Dodson, Tybout,andSternthal1978). In addition,consumersevaluate currentoffersandmakepurchasedecisionsonthebasisof com-parisonsbetweentheobservedofferandtheirinternalreference pricepoints (Kalyanaram and Winer1995).They learnfrom experiencehowtoformpriceanddiscountexpectations.Such expectationsareparticularlyrelevantforpermanentdiscounts. Whendiscounts areprovidedcontinuously,consumersexpect thattheywillbeprovidedinthefuture, too.Onceconsumers starttoexpectthemasaruleratherthananexception,discounts maynolongerbeabletoincentivizecustomerspending(Lattin andBucklin1989).

Moreover, Breugelmans and Campo (2016) affirmed that the frequency of price discounts in digital channels reduces the effectiveness of future discounts. Even temporary price discounts reduce consumers’ price expectations, which may reducepurchaseintentionsfor products soldatregular prices (DelVecchio, Krishnan, and Smith 2007). In a permanent discount context, consumers’ reference price or their price expectations will likely equal the discounted price that they observedor experiencedpreviously. Wetherefore expectthat higher discount expectations will negatively influence cus-tomers’purchaseincidenceandactualpurchasebehavior.Thus, weproposehypothesesasfollows:

H3. Customers’ expectations of product-specific price dis-countsnegativelyinfluence(a)purchaseincidence,(b)purchase quantity,and(c)totalspending.

H4. Customers’ expectations of order coupons negatively influence(a)purchaseincidence,(b)purchasequantity,and(c) totalspending.

Higherdiscountexpectations mayalsocause customersto perceivesmallergainsfromlaterdiscounts(KalwaniandYim 1992),iftheyfeeldisappointedbysmallerdiscounts.For exam-ple, if a customer previously received a 30% price discount onaverage,acurrent 20%discountwillnotseemlikeagood deal.Therefore,weexpectthatconsumerdiscountexpectations moderatethe effects ofcurrent discounts.Thus,weprose the followinghypotheses:

H5. Customers’ expectations of product-specific price dis-counts weaken the current effects of product-specific price discountsonpurchasequantity.

(8)

Table2

OperationalizationofVariables. Subscripts

i Theithcustomer

t Thetthweek

Variables ComputedPeriod DescriptionandCalculation

Dependentvariables

PIit Analysisperiod Purchaseincidence=1ifacustomerpurchasesinagivenweek,0otherwise

Sit Analysisperiod Averageorderingspending=ln(Averagespendingineachorderinagivenweek+1)

Qit Analysisperiod Averageorderquantity=ln(Averagenumberofitemsineachorderinagivenweek+1)

Explanatoryvariables

PIit-1 Analysisperiod LaggedtermofPIittocapturestatedependence

Ave.PDt Analysisperiod Averageproduct-specificdiscountinthetthweekprovidedbytheretailer=ln(Averageproduct-specific

discountbasedonproductdiscountsobtainedbyallcustomerswhohavepurchasesintthweek+1) Ave.ODt Analysisperiod Averagecouponvalueinthetthweekprovidedbytheretailer=ln(Averagecouponvaluebasedoncoupons

obtainedbyallcustomerswhohavepurchasesintthweek+1)

PDit Analysisperiod =ln(Averageproduct-specificpricediscountratioineachorderexperiencedbycustomeriinweekt);

product-specificpricediscountratio=discountedamountperitemofaspecificproduct/thisproduct’s regularprice

PDexpit Analysisperiod Product-specificdiscountexpectation=ln(Decayingweightedaverageofproduct-specificdiscountratios

obtainedinpreviousperiods+1).

ThevalueofthefirstperiodistheaverageofPDfortheinitializationperiod. ODit Analysisperiod =ln(Averagetotalcouponvalueineachorderperweek+1)

ODexpit Analysisperiod Ordercouponexpectation=ln(Decayingweightedaveragecouponvalueobtainedinpreviousperiods+1)

Controls

Tenurei Initializationperiod =ln(Numberofdaysbetweenacustomer’sfirstordertoJune30,2015+1)

Recencyi Initializationperiod =ln(Numberofdaysbetweenacustomer’slastordertoJune30,2015+1)

Prespendingi Initializationperiod =ln(Totalspending+1)

Preordersi Initializationperiod =ln(Numberoftotalorders+1)

H6. Customers’ expectations of order coupons weaken the currenteffectsofordercouponsonpurchasequantity.

Giventhatwedonotproposeclearexpectationsforthe influ-ence of the two typesof current discountlevels oncustomer spending,wecannotbuildhypothesesfortheinteractioneffects betweendiscountsanddiscountexpectationsonspendingeither, whichwewillexploreinlaterempiricalanalyses.

Data

Descriptionofdata

Toanswerourresearchquestions,weworkedwitha small-to-medium-sizedChineseretailerthatofferspermanentdiscounts foralmostallofitsproducts.2Whenthefocalretailerwasfirst establishedin2011,itstraditionalbrick-and-mortarshops

pre-2 _To_gather_preliminary_insights_into_the_prevalence_of_permanent_discounts,

wecollecteddataaboutthepricingstrategiesof22smallandmedium-sized Chi-nesedigitalretailersfromtheirwebsitesinFebruary2019,includingourfocal firm.Ofthese22retailers,13providepermanentdiscountsfortheircomplete assortmentorpartthereof,andnineoffertemporarypromotions(e.g.,dailydeal offers).Ofthethirteenretailersofferingpermanentdiscounts,fiveofferthem forallcategoriesbutdonotclaimtobediscountstores.Wealsoobservethat theirclosecompetitors(i.e.,retailersofferingsimilarproductswithverysimilar prices)usesimilarstrategies,signalingthatapermanentdiscountstrategyis

dominatelysoldbabyproducts,butovertimeitaddedawide rangeofotherproductcategoriestoitsassortment(e.g., cosmet-ics,snacks).Afterintroducinganonlinestorein2012,itshifted mostof itssales focus(i.e.,95%) fromofflinetoonline, and in2014itlaunchedmobilesaleschannelstoexpandits multi-channelmix.Theproductassortment,prices,anddiscountsare thesameacrossallchannels,andtheretailerprovidestwotypes ofdiscounts:product-specificpricediscounts(PD)intheform ofdiscountedpricesforproducts,andordercoupons(OD)in theformofamount-offcoupons(e.g., 10CNYoff).3,4Inour dataset,weidentify5686differentproducts,only253(4.45%) ofwhichare offeredwithoutapricediscount. Oneach prod-uctpage, theretailer presentsthe regular anddiscountprices

appliedbysmallandmedium-sizeddigitalretailerstoremaincompetitiveand toattractcustomersawayfromtheirpeercompetitors.

3 _In_Western_countries,_coupons_are_often_presented_in_form_of_“%_off,”_“$

off,”and“Buyone,getonefree”(Raghubir2004;Drechsleretal.2017);for example,whenyousigninonaretailer’swebsiteforthefirsttime,youcanget anX%coupon.InChina,ordercouponsarenormallyformattedasamount-off discounts.

4 _The_focal_retailer_offers_both_{unconditional}_coupons_(e.g.,_a₁₀_CNY-off

coupon,whichcanberedeemedforanyorders)andconditionalcoupons(e.g., a10CNY-offcoupon,whichcanbeonlyredeemedfororderswithacertain amount).Unfortunately, ourdatadoesnot allowusto differentiatethetwo typesofordercoupons.Butfromaseniormanagerweknowthatunconditional couponsaccountsformorethan90%ofallordercouponswhileconditional couponsarenomorethan10%.

(9)

Table3

DescriptiveAnalysisofMajorVariables.

Min Median Mean Max SD SE

PIit 0.00 0.00 0.20 1.00 0.40 0.00 Sit 2.28 2.64 2.86 5.39 0.47 0.00 Qit 0.69 0.69 0.86 2.71 0.27 0.00 Ave.PDt 0.12 0.21 0.21 0.26 0.03 0.00 Ave.ODt 3.67 3.89 3.95 4.32 0.20 0.00 PDit 0.00 0.20 0.21 0.47 0.09 0.00 PDexpit 0.00 0.24 0.27 0.83 0.12 0.00 ODit 1.39 3.81 3.73 5.70 0.75 0.01 ODexpit 0.00 3.68 3.54 6.95 0.78 0.00 Tenurei 0.00 4.78 4.78 5.20 0.54 0.01 Recencyi 0.00 4.76 4.64 5.19 0.68 0.01 Prespendingi 2.39 3.28 3.50 8.86 1.12 0.02 Preordersi 0.69 0.69 0.91 2.56 0.34 0.01

for morethan95% ofitsproducts.Only104 (2.69%)of cus-tomersinourdatasethaveapurchasehistorywithoutanyprice discounts.Theretaileralsosendscouponstocustomerswhich canberedeemedagainstanyorder.Someofthesecouponsare thesameacrosscustomers,butothersarebasedoncustomers’ currentandpreviousspendingbehavior.

Weobtaineddatafortheonlineandmobilechannelsofour focalretailer forthe periodJanuary1 toDecember31,2015. These data include customer order information: order time, online versusmobilechannel,products ineach order,regular prices,price-specificdiscounts,couponsredeemedineachorder, numberofitems,andactualspendinginChineseyuan(CNY). To retain more granulardata information, we aggregated the data toweekly panels instead of monthly panels.To capture customers’priorpurchasebehavior(tenure,recency,frequency, andspending),wesplitthedataintoaninitializationperiodfrom January1toJune31,2015andananalysisperiodfromJuly1to December31,2015.Weidentified3866uniquecustomerswho orderedatleastonceintheinitializationperiodandatleastonce intheanalysisperiod.Intheanalysisperiod,customersentered atotalof32,470orders.

Operationalizationofvariables

In the following, we detail the operationalizations of our focalvariablesfromourconceptualframework(see Table2). WeprovidedescriptiveanalysesinTable3.

Dependentvariables

Wetake bothpurchase incidence (labeledPIit) andactual

purchasebehaviorbycustomersexposedtothefocalretailer’s discountsasdependentvariables(e.g.,BreugelmansandCampo 2016;Jiaetal.2018).Intermsofactualpurchasebehavior,we measurespendinglevelandquantityofeachorderonaverage foreachweek(labeledSitandQit,respectively).

Explanatoryvariables

Theretaileroffersrelativelyhighdiscounts,especiallywith its ordercoupons.The weekly averageproduct-specific price discountratiofor eachorder,i.e.,theaverageweeklyratioof

discountstoregularpricesofallproductspurchasedbyall cus-tomers,is24%;theweeklyaverageordercouponvalueis40.68 CNY,or64%ofanaverageorderintermsofspending.5Theratio ofthe totaldiscountamount(product-specific pricediscounts plusordercoupons)toeachorder’stotalspendatregularprices is85.60%percustomeronaverage.Thedescriptiveinformation forthelogarithmvaluesofthetwovariablesPDit(pricediscount

ratio) and ODit (absolute value of coupons) are in Table 3.6

Wealsoincludeconsumers’expectationsof PDit andODit as

explanatoryvariables(i.e.,PDexpitandODexpit,respectively),

whichwe specify asthe decayingweightedaveragelevels of PDitandODitthataconsumerhadredeemedpreviously.7

Controlvariables

Weusecustomers’purchase-relatedbehaviorinthe initializa-tionperiodtocontrolforcustomerheterogeneity.Specifically, weincludetenure,recency,totalspend,andnumberofordersin ourmodel.

Methodology

Endogeneityofdiscountvariables

Retailersgenerallyusediscountsstrategically,sodiscounts maybeendogenous(Bijmolt,vanHeerde,andPieters2005).We useacopulaapproachtocorrectforpotentialendogeneitydueto thecorrelationbetweenthediscountvariablesanderrorterms.

ParkandGupta(2012)proposedGaussiancopulastoaccount forendogeneityissues,withtheadditionofintegratingthe cop-ulaterms ofallendogenous variablesinthemajormodels as additionalregressors.Thisapproachiswidelyusedin market-ingresearch(e.g.,Datta,Ailawadi,andvanHeerde2017).To addressthelackofavailabilityofgoodinstrumentsfor endoge-nousvariables,acopulaapproachprovidesaneffectivesolution withoutrequiringexclusionrestrictions.

In ourstudy, we encounter six potential endogenous vari-ables(i.e.,themaindiscountvariablesandtheirinteractions), whichmakes itextremely difficultto find appropriate instru-ments.Thus,weapplyacopulaapproachtoobtain

p*= Φ-1(H (p))

5 _These_percentages _are_calculated_from _the_original _data,_without_taking

logarithms.

6 _The_logarithm_was_taken_to_reduce_data_skewness_and_to_reduce_the_variable

rangeofdata.

7 _The_decaying_weighted_average_calculation_was_employed_to_account_for

memorydecayorconsumersforgettingwhattheyexperiencedbefore.Forthe sakeofsimplicity,weassumethatconsumermemorydecayfollowsalinear pattern.Thus,weassumeandassignalinearincreasingsequencewithastarting pointof1,andtheintervalis1(i.e.,1,2,3,4,...,28)forthefirstweektothe lastweekinourtimewindow,whichcanbeproxiedasthevalueoffreshness ofone’smemoryinweekt.Thehigherthevalue,thefresherthememoryis. Wethenusethepercentageofsequencevalueastheweighttoeachweek’s discountsexperiencedbyindividualsto calculateaveragediscountsastheir discountexpectations.Specifically,wecalculatetheweightsforacustomeri’s

discountexpectationinacurrentperiodTwiththefollowingformula:weightit=

sequencevalueit/(sequencevaluei1+sequencevaluei2+...+sequencevalueit

(10)

whereH(p)istheempiricalcumulativedensityfunction(CDF) of anendogenous regressorp, andΦ−1is theinversenormal CDF.AGaussiancopulaapproachrequiresthatthepotentially endogenousvariablesarenotnormallydistributed,soweapply theAnderson–Darlingnormalitytesttoconfirmthatthe endoge-nousvariablesdonotexhibitnormaldistributions.Wealsoapply the Shapiro–Wilktesttoarandomly selectedsampleof 5000 recordstoensureconsistentnon-normaldistributions.

Inadditiontothepotentialendogeneityofspecificdiscounts, specificcustomersmayreceivemorediscounts becauseof the firm’stargetingpolicy.WeapplyaMundlakapproachto cor-rect for potential endogeneity in discounts due to individual differences (Mundlak 1978; Risselada, Verhoef, and Bijmolt 2014).Thisapproachconstructsaverageproduct-specificprice discountratiosandaveragetotalcouponvaluesforeachorder perweekandcustomerintheanalysisperiod,astwoadditional explanatory variables. These variables account for how cus-tomersdifferintheirpossibilityofreceivingdiscounts(labeled Mund.PDiandMund.ODi,respectively).

Modelspecification

As customers’spending and purchase quantity are condi-tional ontheir decisions onwhether tobuy inagiven week, wefirstspecifycustomerpurchaselikelihood.Weuseabinary purchaseincidencevariablethatindicateswhetheracustomer purchasesinweekt.Wecannotobservediscountswithoutany purchaseinagivenweekt;weobserveonlydiscountsthatthe customerredeemedinpreviousweeks.Thus,toexplaina cus-tomer’spurchaselikelihoodinweekt,weincludethedecaying weightedaveragelevelofdiscountsobtainedbeforeweektas aproxyforacustomer’sdiscountexpectations,andwecontrol forpotentialnonlinearrelationships.Wealsoincludethe aver-agediscountlevelforagivenweek,basedondiscountsobtained byallcustomerswhomadepurchasesinthetthweek(Ave.PDt

andAve.ODt),asproxiesofthecurrentdiscountlevelsinweek

tprovidedbytheretailer.Usingcustomers’purchaseincidence inthepreviousweek,wecapturestatedependencebetweenthe two consecutivetimeperiods.TheMundlakterms correctfor potentialindividualendogeneitybias.

We propose that PIit, whichindicates whether customer i

makesapurchaseinweekt,isdrivenbythelatentutility(PIit*)

ofcustomeriforpurchasinginagivenweekt,suchthat

PIit=

1ifPI*_it> 0

0unobservedifPI*_it≤0. (1) Thelatentutilityisspecifiedasfollows:8

PI*_it=PIit-1+PDexpit+PDexp2it+ODexpit+ODexp2it

+ Ave.PDt+Ave.ODt+Mund.PDi+Mund.ODi

8 _To_control_for_general_time_trends_and_holiday_effects,_we_estimate_each

week’s contributioninsteadoflineartime trendeffects.To savespaceand keeptheequationrelativelysimple,weomitcoefficients(␤)forallexplanatory variablesinallequations.

+Tenurei+ Recencyi+Prespendingi+Preordersi

+A set of weekindicators+ξ1i+ε1it (2)

Wethenmodelcustomers’actualpurchasebehavior,i.e.,their averagespendingforeachorderinagivenweekandthe aver-agequantityforagivenorder.Spending(Sit,reflectingaverage

orderspending)andquantity(Qit,reflectingtheaverage

num-berofitems inan orderas purchasequantity)areconditional on observing apurchase in week t by customeri. As previ-ously discussed, customers’ discounts redeemed ina current weekanddiscountexpectationsformedfrompreviousdiscounts influence purchasing behavior. The quadratic terms of these discount-relatedvariablesenableustotestforpotential nonlin-earrelationships.Toaccountforthe hypothesizedmoderating effectofdiscountexpectations, weinclude interactioneffects betweendiscountsanddiscountexpectations.Finally,weagain addcopula(Papies,Ebbes, andvanHeerde2017)and Mund-laktermstocorrectforthepotentialendogeneityofdiscounts aswellasothercontrols,asinEq.(2).Theequationsfororder spendingandquantityareasfollows:

Sit=

S*_itifPI*_it> 0

0unobservedifPI*_it≤0 (3)

S*it=PDit+PD2it+PDit*PDexpit+ODit+OD2it

+ ODit*ODexpit+PDexpit+PDexp2it+ODexpit

+ ODexp2_it+Mund.PDi+Mund.ODi+A set of copula

terms+Tenurei+Recencyi

+Prespendingi+Preordersi+A set of week

indicators+ξ2i+ε2it (4) Qit= Q*_itifPI*_it> 0 0unobservedifPI*_it≤0 (5) Q*it=PDit+PD 2 it+PDit*PDexpit+ODit+OD2it

+ODit*ODexpit+PDexpit+PDexp2it+ODexpit

+ODexp2it+Mund.PDi+Mund.ODi+A set of copula

terms+Tenurei+Recencyi

+Prespendingi+Preordersi+A set of week

(11)

Fig.2.Timetrendsofproductpricediscounts(PD)andordercoupons(OD).

Eqs. (1) and(2) together constitute amixed-effectsprobit incidence model;Eqs.(3)–(6) aremixed-effectspanel regres-sionmodels.ξirepresentscustomerrandomeffectsthatcapture

unobservableindividualdifferences.εitistheresidualfor

cus-tomeriinweekt.Theerrorstructuresareasfollows,wherethe means–μreandμeare3×1vectorsandsetto0:[␰1i␰2i␰3i]

(∼MVN(␮re,re),and[␧1i␧2i␧3i](∼MVN(␮e,e).

Tocheckwhethercustomerswhopurchaseinagivenweek inherentlyspendmore(␴12re2>0)orless(␴12re2<0)thanthose

whodonotpurchaseinthatweek,weallowrandomeffectsand errorstocorrelateacrossequations,Thecorrelationoferrorsof thethreeequationsineindicatesthepotentialselectionbias

thatcouldresultifthesameunobservedfactorscausedifferent dependentvariablestochangeinspecificdirectionsatthesame time.Thevarianceofthepurchaseincidenceequationissetto 1foridentificationoftheequationsystem.Thus,thecovariance matricesreandecanbepresentedasfollows:

Σre=

⎡ ⎢ ⎣

σ_1re2 σ_12re2 σ2_13re σ_12re2 σ_2re2 σ2_23re σ_13re2 σ_23re2 σ2_3re

⎤ ⎥ ⎦ Σe= ⎡ ⎢ ⎣ 1 ρ_12e2 ρ2_13e ρ2_12e ρ_2e2 ρ2_23e ρ2_13e ρ_23e2 ρ2_3e ⎤ ⎥ ⎦

Resultsanddiscussion

Since one of our aims were to compare the influence of product-specific price discounts (PDit) with the influence of

ordercoupons (ODit) oncustomer behavior,we standardized

allcontinuousexplanatoryvariables.Table4presentsthe corre-lationmatrixofPD-relatedvariablesandOD-relatedvariables. All the relatively highcorrelations(higher than 0.50) appear betweenPD-relatedvariablesandOD-relatedvariablesrather thanwithin PD-or OD-related variables, e.g.,the correlation between PDit and ODit is −0.78.Weknow fromthe retailer

that part of its strategy is totreat product-specific price dis-countsandordercouponsassubstitutivemarketinginstruments,

whichissupportedbyourdata(seeFig.2).Theretailertendsto Table

4 Correlation Matrix (Product-specific Price Discount-related V ariables and Coupon-related V ariables). PD it PD it 2 PD it *PDe xp it A v e.PD t Mund.PD it PDe xp it PDe xp it 2 OD it OD it 2 OD it *ODe xp it A v e.OD t Mund.OD it ODe xp it ODe xp it 2 PD it 1.000 PD it 2 0.280 1.000 PD it *PDe xp it − 0.176 0.031 1.000 A v e.PD t 0.080 0.067 − 0.025 1.000 Mund.PD it 0.150 0.041 0.031 0.000 1.000 PDe xp it 0.028 − 0.044 − 0.016 − 0.145 0.340 1.000 PDe xp it 2 − 0.007 − 0.018 0.018 − 0.107 − 0.116 0.586 1.000 OD it − 0.779*** − 0.294 0.161 − 0.059 − 0.116 0.015 0.022 1.000 OD it 2 0.335 0.813*** − 0.057 0.066 0.048 − 0.042 − 0.025 − 0.236 1.000 OD it *ODe xp it − 0.047 0.067 0.543** − 0.010 0.039 − 0.038 − 0.047 0.090 0.080 1.000 A v e.OD t − 0.053 0.005 0.025 − 0.305 0.000 0.159 0.087 0.131 0.020 0.004 1.000 Mund.OD it − 0.105 − 0.012 − 0.028 0.000 − 0.698*** 0.340 − 0.017 0.151 0.009 − 0.044 0.000 1.000 ODe xp it − 0.031 0.015 0.004 0.092 − 0.322 − 0.779*** − 0.507 0.031 0.022 0.058 − 0.053 0.496 1.000 ODe xp it 2 − 0.002 − 0.029 0.010 − 0.093 − 0.034 0.532 0.796*** 0.027 − 0.031 − 0.026 0.104 0.002 − 0.428 1.000 * p < 0.05, ** p < 0.01, *** p < 0.001.

(12)

Table5

SimultaneousEstimationofPurchaseIncidence,Spending,andPurchaseQuantity.

EstimatingProduct-SpecificPriceDiscountEffects EstimatingOrderCouponEffects

PIit Sit Qit PIit Sit Qit Constant −0.510***(0.099) 3.405***(0.094) 1.055***(0.044) Constant −0.310**(0.114) 2.610***(0.061) 0.663***(0.033) PIit-1 0.374***(0.012) PIit-1 0.736***(0.018) Ave.PDt 0.694***(0.036) −0.553***(0.076) −0.243***(0.018) Ave.PDt 0.813***(0.036) −0.072(0.061) 0.016(0.014) Ave.ODt 1.156***(0.126) −1.158***(0.129) 0.570**(0.058) Ave.ODt 1.475***(0.146) −0.253**(0.084) −0.135**(0.041) PDexpit −0.023(0.012) 0.051***(0.011) 0.032**(0.006) ODexpit −0.071***(0.011) 0.008(0.005) 0.001(0.003) PDexpit2 0.028***(0.006) 0.110***(0.005) 0.018**(0.003) ODexpit2 0.011*(0.006) 0.009***(0.002) 0.004**(0.001) PDit 0.051**(0.015) 0.035***(0.008) ODit 0.741***(0.074) 0.300***(0.047) PDit2 0.062***(0.007) −0.004**(0.003) ODit2 0.081***(0.006) 0.041***(0.004) PDit*PDexpit 0.033***(0.006) 0.007(0.003) ODit*ODexpit 0.010**(0.003) 0.002(0.002) Mund.PDi 0.034**(0.012) −0.083***(0.013) −0.037***(0.006) Mund.ODi 0.047***(0.011) 0.027***(0.006) 0.017***(0.003) Copula(PDit) −0.118***(0.014) −0.022**(0.007) Copula(ODit) −0.508***(0.073) −0.179***(0.046) Copula(PDit2) −0.064***(0.008) −0.034***(0.004) Copula(ODit2) 0.041***(0.005) 0.018***(0.003) Tenurei 0.071(0.043) −0.018(0.034) −0.003(0.015) Tenurei 0.095*(0.038) 0.014(0.012) 0.015*(0.006) Recencyi −0.112**(0.037) 0.039(0.027) 0.018(0.012) Recencyi −0.138***(0.029) −0.020*(0.008) −0.007(0.005)

Prespendingi −0.028(0.016) 0.047**(0.014) 0.018**(0.006) Prespendingi −0.007(0.018) 0.009(0.005) −0.002(0.003)

Preordersi 0.771***(0.049) −0.478***(0.037) −0.218***(0.018) Preordersi 0.846***(0.050) 0.071**(0.020) 0.046**(0.014)

Log-Likelihood −43559.80 Log-Likelihood −36778.31

AIC 87365.60 AIC 73804.63

BIC 88540.97 BIC 74989.55

*p<0.05,**p<0.01,***p<0.001.

Note:Therobuststandarderrorsaregiveninparentheses;thecontinuousindependentvariablesarestandardized.Weconductedseveralrobustnesschecks,using

modelsthat(1)controlfortimetrendswithoutdifferentiatingeachweek’scontribution,(2)donotcontrolforaveragelevelsofpricediscountandcouponvalue,(3) donotallowindividualeffectstobecorrelatedacrossequations,and(4)donotallowerrortermstobecorrelatedacrossequations.Allthesemodelsprovidedsimilar findingstotheresultsreportedinthistable.Wealsoexaminedwhetherconsumers’channelpreferences(PCvs.mobile)playamoderatingrole.Ofthe3866customers inourdataset,288usedonlymobilechannelstopurchaseinthefirstsixmonthsofourobservationperiod,3329customersusedonlyPCs,and249customersused bothchannels.Thus,theaveragemobileratioislow(.083).Weaddedmobileratiotoourmainmodelsandre-estimatedtheimpactsofthetwotypesofdiscounts ontheoutcomevariables,correctingtheself-selectionbias,butdidnotfindanysignificantmaineffectsofchannelpreferenceoritsinteractions.Recognizingthat non-significanteffectsofchannelpreferencemightbecausedbyaskeweddistributionofthemobileratio,wereplacedthecontinuousmobileratiowithadummy variablethatcaptureswhetheracustomerprefersmobileoronlinechannels(mobileratio≥.5,dummy=1,otherwise=0)andthenre-estimatedthemodels.Again, theinteractioneffectsofchannelpreferencewerenotsignificant.Theeffectsofthetwotypesofdiscountsonpurchasingbehaviorremainedconsistentwithourmain estimation.Thedetailedresultsareavailableonrequest.

providelowerordercouponswhenitisalreadyofferinghigher product-specificpricediscounts,andviceversa,toavoid con-sumers experiencingtwo typesof discountsat thesametime (whichmayleadtosevererevenuelossesfortheretailer).

Toavoidmulticollinearityissues,wemodeledtheeffectsof PD-related variablesand OD-related variableson purchasing behaviorseparatelyindifferentregressions.Whenmodelingthe effects of PD-relatedvariables, wecontrolledfor the average valueof ordercouponsredeemedbyallcustomersinagiven week(Ave.ODt)asaproxyof thegeneralordercouponvalue

levelprovidedbytheretailerinthatweek.Similarly,when mod-elingtheeffectsofOD-relatedvariablesonpurchasingbehavior, we controlled for the average product-specific pricediscount levelinaweek(Ave.PDt).Wedetailtheresultsfromthe

simul-taneousestimationofEqs.(1)–(6)inTable5.

Nonlineareffectsofproduct-specificpricediscounts

WefirstpresentthefindingsforPDitfromthespendingand

quantityequations.PDit anditsquadratictermsaresignificant

inbothequations.Inthespendingequation,theeffectof PDit

followsaU-shapedcurve(␤PD=.051,p<.01;␤PD2=0.062,

p<.001),whereastheeffectofPDit onpurchasequantity

fol-lowsaninvertedU-shapedcurve(␤PD=0.035,p<.001;␤PD2

=−.004,p<.01).SincePDit isstandardizedintheestimation,

PDitvariesbetween−2.40and2.90.Inthisrange,theeffectof

PDitonspendingindeedfollowsaU-shapedpattern(Fig.3a),

butits effectonpurchase quantityis positiveatadecreasing rate(Fig.3c). Theseresultsindicate that,withanincrease in product-specificpricediscountlevels,customers’spendingfirst decreasesandthenincreases,whiletheirpurchasequantityfirst increaseswitharelativelysteepslopebutthenlevelsoff. Hypoth-esisH1isthereforesupported.

The original product-specific price discount level corre-spondingtothevertexofthe U-shapedeffectinthespending equation is 18.89% (=exp(−0.411*0.09+0.21)−1) after de-standardization and anti-logarithm, meaning that discounts higherthan18.89%positivelyinfluencespending,whereas dis-countslowerthan18.89%haveanegativeeffectonspending. Asalreadyindicatedinourconceptualframework,thechange of spending caused by discounts should rely on the tradeoff betweenthechangeofpurchasequantityandmonetary reduc-tionduetodiscounts.TheexplanationfortheU-shapedpattern maybethattheincreasedspendingresultingfromthelarger pur-chasequantityattractedbydiscountslowerthan18.89%does notoffsetthereductioninspendduetothediscounts(Raghubir 2004);discounts higherthan18.89% encourage customersto purchasemoreitemsandgeneratehigherspending,which com-pensatesforthe spendingreductionthat resultsfrom offering thesediscounts.

(13)

Fig.3.Effectsofpricediscounts(PD)andordercoupons(OD)onspendingandpurchasequantity. Note:BothPDandODarestandardizedvalues.

Nonlineareffectsofordercoupons

ODitanditsquadratictermsaresignificantinthespending

andquantityequationsaswell(spending:␤OD=0.741,p<.001;

␤OD2=0.081,p<.001;quantity:␤OD=0.300,p<.001;␤OD2

= 0.041,p<.001).The standardizedODit ranges from−3.11

to2.62.Withinthisrange,theeffectsofODitonspendingand

purchasequantityareincreasinglypositive(seeFig.3bandd), supportinghypothesisH2.

Theamount-offordercoupons,whichcanberedeemedfor the wholeorder and are not restricted to aspecific product, influencecustomerpurchasebehaviordifferentlythan product-specificpricediscounts. Aswe expected,ordercoupon value influences customers’ purchase quantity positively, and the higherthecouponvalue,thestrongerthepositiveeffecton pur-chasequantity.Moreover,ourresultsshowthatthecouponvalue positivelyinfluencescustomerspendinginasimilarwayasfor purchasequantity.Twoexplanationsarepossible:(1)The stim-ulatedspending from purchasingmoreitems due tocoupons exceedsthe couponvaluethat isdeductedfromthe customer payment;(2)Highercouponvaluesencourageconsumerstobuy moreexpensiveproductsthatwouldnormallybeunaffordable. Totestthesecondexplanation,weconductedadditional analy-sisandfoundthatordercouponvaluehasasignificantpositive effect(␤=0.305,p<.001)ontheaveragepriceofproducts pur-chasedbycustomersinagivenweek.9Bothofthesecasesmay leadconsumerstospendmoreduetoahighercouponvalue.

9 _We_conducted_an_additional_regression_to_model_the_effect_of_order_coupons

ontheaveragepricesofproductspurchasedbyeachcustomerinagivenweek. Thefullestimationresultsareavailableonrequest.

Effectsofdiscountexpectations

Directeffects

Customer expectation of product-specific price discounts (PDexpit)exertssimilareffects onpurchaseincidence,

spend-ing, and quantity. Specifically, its influence on each of the threedependentvariablesisaU-shapedrelationship(incidence: ␤PDexp = 0.027, p<.05; ␤PDexp2 = 0.028, p<.001, Fig. 4a;

spending:␤PDexp = 0.051, p<.001; ␤PDexp2 = 0.110, p<.05,

Fig. 4c; quantity: ␤PDexp = 0.032, p<.01; ␤PDexp2 = 0.018,

p<.01, Fig. 4e). Likewise, customers’ expectations of order coupons(ODexpit)influencethe threeoutcomevariablesina

U-shapedway(seeFig.4b,d,andf),asthequadratictermsare significantandpositiveinthethreeequations(incidence:␤ODexp

=−.071,p<.001;␤ODexp2=0.011,p<.05;spending:␤ODexp2

=0.009,p<.001;quantity:␤ODexp2=0.004,p<.01).

Hypothesis H3 expects that customers’ product price discount expectations will negatively influence the three purchase outcomes. Our empirical estimations do indeed reveal such negative effects, but only in a certain range of discountexpectations.Whencustomers’discountexpectations arelowerthanaparticular threshold,theyinfluencepurchase behavior negatively. When they exceed this threshold, the effectsbecomepositive.Specifically,thethresholdsof product-specificpricediscountsinthethreeequations(i.e.,thevertexes in Fig. 4a, c, and e) are 31.00% (=exp(0*0.12+0.27)−1), 27.40% (=exp(−0.232*0.12+0.27)−1), and 17.74% (=exp(−0.889*0.12+0.27)−1), respectively. Likewise, weexpectnegativeeffectsofordercouponexpectationon pur-chaseincidence(H4a),purchasequantity(H4b),andspending (H4c)inH4.Butwefindordercouponexpectationsinfluence

(14)

Fig.4.Effectsofconsumers’expectationsofpricediscount(PDexp)andofordercoupons(ODexp)onpurchaseincidence,spending,andpurchasequantity.

Note:BothPDexpandODexparestandardizedvalues.

thepurchasebehaviorinaU-shapedpattern.Thethresholdof order couponexpectation inthe purchase incidence equation is426.12CNY(=exp(3.227*0.78+3.54)−1,Fig.4b),whileit is33.47CNY(=exp(0*0.78+3.54)−1,Fig.4dandf)inboth equations for spending and purchase quantity. The negative effects of discount expectations below the cited threshold may result from a decrease in customers’ reference prices (e.g., Kalwani and Yim 1992), because the discounts (both product-specificpricediscountsandordercoupons)redeemed inanordercanbeaveragedforeachproductwithinthebasket, thereby reducing the average price of the products. After consumers experience purchasing such products at “reduced prices,”theyarelesslikelytopurchaseoverallorarelikelyto purchase less when products are not offered at the “reduced prices.” Hence, a higher discount expectation leadsto lower purchasepossibilityandactualpurchases.

Anotherpotentialexplanationforourfindingsontheeffectof PDitandODitonquantityisthatcustomerswithhigherPDexpit

orODexpithadredeemedhighdiscountsbefore,whichledthem

tobuylargequantitiespreviously.Customershavespecific, rel-ativelystabledemandforretailproductsinagivenperiod(e.g., ayear),sothosewhohavepurchased(i.e.,stockpiled)alarge quantityinapreviousperiodarelikelytohavelimitedpurchase demandinthecurrentperiod(BlattbergandNeslin1990).Thus, theirpurchaseincidenceandactualpurchaseamountwilldecline inweekt.However,thepositiveeffectofPDexpit orODexpit

onincidence,spending,andquantityabovethecitedthreshold may result from customers’spending anddemand character-istics;thatis,thefocalretailersendsdiscountstocustomersin accordancewiththeirpreviousandcurrentspending.Customers whohaveexperiencedveryhighaveragediscountvalueson pre-viouspurchaseoccasionsmayalsoexhibithigherspendingon average.Despitealreadypurchasingmorethanothercustomers,

theystillshowhigherpurchasedemand.Thesecustomersalso believethattheywillreceivehighdiscountvaluesinthecurrent period,whichmayimprovetheirspending.

Moderatingeffect

Turningtothemoderatingeffectsofcustomers’expectations of discounts,we find that theexpectations for PDit andODit

positivelyinteractwiththecurrentpricediscountandthe cur-rentordercouponvaluerespectivelyinthespendingequation (␤PDexp*PD=0.033,p<.001;␤ODexp*OD=0.010,p<.01),which

supportstheinteractioneffectsoncustomerspendingbut pro-videsnoevidenceforH5andH6.Ourestimationsuggeststhat thepositiveeffectofproduct-specificpricediscountson spend-ingismoderatedbycustomers’expectationsofpricediscounts. Whenproduct-specificpricediscountsarethesameacross cus-tomers,acustomerwithalowerexpectationofthepricediscount (e.g.,PDexpit=−2.5)spendsmorethanacustomerwithahigher

pricediscount(e.g.,PDexpit=2.5,Fig.5a).Thisfindingis

con-sistentwithourpredictionthat consumerswillperceive more benefitwhenthedifferencebetweenacurrentdiscountandtheir discountexpectationisgreater(LattinandBucklin1989).

However,wealsoseefromFig.5athat,whenthediscount value is relatively high, consumers with a higher price dis-count expectation spend more than customers with a lower expectation.Acustomerwhohasexperiencedhigherprice dis-counts(i.e.,higherexpectation)andwhoredeemsahigherprice discount for the current order may be a customer who has highdeal-proneness.Deal-pronecustomerstendtoutilize dif-ferentdiscountopportunitiestoobtainbest dealsandsavings (Valentini,Neslin, and Montaguti 2020).This customer may alsohaveahigherpurchase demand,sinceacustomerwitha higherdemandmayspendmoreingeneral,andtheretailermay intuitivelytargetmorediscountstosuchacustomer.Similarly,

(15)

Fig.5.Effectsofpricediscounts(PD)andordercoupons(OD)onspendingwithdifferentcustomerexpectationsofpricediscountandcouponvalue.

Note:PD,PDexp,OD,andODexparestandardizedvalues.

wefindthatcustomerexpectationsofordercouponvalueplaya moderatingroleontheeffectofordercouponsonspending(see

Fig.5b).

Conclusionsandimplications

Smallandmedium-sizeddigitalretailersprovidediscounts continuouslytoattractcustomerstopurchaseandinthehopeof remaining competitive by buildinglong-lasting relationships. We find that bothproduct-specific price discounts andorder coupons offered in a digital environment significantly influ-encecustomers’actualspendingandpurchasequantity,butin quitedifferentways.Inparticular,wefindthathigher product-specificpricediscounts donotalwaysstimulateconsumersto spendmore.Theyinfluencecustomerspendingpositivelyonly whentheyarehigherthan18.89%.Pricediscountsbelowthis threshold influence spending negatively. On the other hand, product-specificpricediscountspositivelyaffectthenumberof itemspurchasedbycustomersinagivenorder,withaslopethat levelsoffasthepricediscountvalueincreases.Regardingorder coupons,whicharenotlimitedtospecificproducts,theirvalues alwaysshowpositiveeffectsoncustomerspendingandpurchase quantityintheirbaskets.Moreover,themagnitudesofthe pos-itiveeffectsofordercouponsincreasewithincreasinglevelsof thevalueofcoupons.

Thesefindingscontributetoresearchonconsumerresponses todiscountsbyexploringaspecificdiscountstrategyadopted inmanyB2Cdigitalchannels,i.e.,discountscontinuously pro-videdbyretailersindigitalchannels.Thisstudyalsocontributes tothediscountliteraturebydelineatingtwotypesofprice dis-counts. Although previous studies have noted that different discounts mayexertdivergenteffects oncustomers’purchase behavior(e.g.,Levyetal.2004),wedonotknowofanystudy thathasanalyzedandcomparedproduct-specificpricediscounts andordercoupons.Ourfindingsindicatethatproduct-specific discountsandordercouponsaffectcustomers’purchase behav-iordifferently.

Moreover,our resultsshow that customers’pricediscount expectations,asapproximatedbythedecayingweightedaverage of discount levels received previously, reduce consumer

pur-chaseincidence and actualpurchase behaviorat lowerlevels andincreasepurchaseincidence athigher levels.Weidentify differentinteractionsofexpectationsofdiscountswithcurrent discountsatdifferentcurrentdiscountlevels.Breugelmansand Campo(2016)reportedthatpreviousdiscountfrequencyimpairs discounteffectiveness.Weaddnuancetothisfindingby show-ingthat,fordiscounts(bothproduct-specificpricediscountsand ordercoupons)atlowerlevels,relativelyhighdiscount expec-tationsreduceacurrentdiscount’spositiveeffectonspending; however,whenthediscountvaluesarerelativelyhigh,relatively highdiscountexpectationsenhanceacurrentdiscount’spositive influenceoncustomerspending.

Implicationsforonlineretailers

Our findings provide useful insights for digital retailers, especiallyfordigitalretailersactiveinhyper-competitive envi-ronments in which they strongly and continuously focus on discountstoattract customers. Asbothproduct-specificprice discounts and order coupons influence customer purchase behaviorinanonlinearway,retailersshouldtakecaretodesign theirdiscountstrategiesaccordingly.Inthisstudy,wefocused ontheeffectsofdiscountsonpurchasing,andourresultshave thefollowingimplications.

First,inordertohaveaneffectonpurchasebehavior,retailers shouldproviderelativelyhighproduct-specificpricediscounts. Ifthepricediscountistoolow(i.e.,lowerthan19%),weobserve anegativeeffecton spending.Only whenpricediscounts are higherthan19%dohigherpricediscountsattractcustomersto spendmoreintheirbaskets.Ourresultsalsoshowthatthe pos-itiveeffectonquantitydecreaseswithhigherdiscounts.Taken together,theseresultssuggestthatlowerproductdiscount val-uesare definitely not preferred. Besides, we show that when customers’expectations ofproductpricediscounts arehigher than31%,27%,and18%,theypositivelyaffectcustomers’ pur-chaseincidenceinaweek,spending,andpurchasequantityin their baskets. Thisasks retailers to createsufficient discount expectationsinthemarketinordertokeepattractingcustomers. Takealltogether,inthehyper-competitivemarketunderstudy,

(16)

ourresultssupporttheuseof continuouspricediscountswith sufficientlyhighvalues.

Second,ourstudysuggeststhatfirmscanuserelatively low-valueordercoupons.Therearemultiplereasons.Wefindthat ordercouponvaluespositivelyinfluencespendingandquantity inabasket whatevercoupon valuesare.From thissense, the retailershouldprovidehighercouponvalue.Moreover,if con-sideringtheU-shapedeffectsofcustomerexpectationsoforder coupons,onlywhenordercouponexpectationsarehigherthan 426CNYtheypositivelyinfluencecustomerpurchaseincidence attheretailerinagivenweek;whenordercouponexpectations arehigherthan34CNY,theyaffectspendingandpurchase quan-tityinabasketpositively.Inthisvein,theretailermayneedto keepordercouponvalueshigherthan426CNYcontinuouslyto assurethatbothordercouponvaluesandcouponexpectations positivelyinfluencecustomerpurchaseincidence,spending,and purchasequantity.However,giventhatordercouponsaffectthe marginonthetotalbasket,careshouldbetakentoprovideorder coupons withvery high value.Therefore, we do not suggest that the retailer keepcoupon valueshigher thanthemaximal threshold—426CNY.Instead,wesuggestkeepingordercoupon valuesatrelativelylowerlevels.Therearetwochoiceshere,i.e., couponvalueshigherthan34CNYbutlowerthan426CNYand couponvalueslowerthan34CNY.Inthefirstchoice,coupon valueexpectations between34 CNYand426 CNYmaylead toverylowpurchaseincidence(seeFig.4bwherethevalueon xaxisisbetweenzeroandthevertex).Inthesecondchoice2, couponvaluelowerthan34CNYmayleadcustomerstohave lowerspendingandpurchasequantity(seeFig.3banddwhere thevalueonxaxisislowerthanzero).Therefore,whether select-ingthefirstorthesecondchoicedependsontheretailer’saim, i.e.,whetheritplanstostimulatemorespendingandpurchase quantityortoinducemorecustomerstopurchase.

Third, our results suggest that the effect of a discount increaseswhenthediscountexceedsthemodeledexpectation, which implies that discounts should not be below customer expectations.Thissupportstheuseofdiscounttacticsthatare continuousbutalsoconsistent.

Finally,notethatwedidnotcalculateprofitconsequences. This means that we cannot provide implications for optimal discountlevels,butonlyinsightsbasedonpurchaseoutcomes withoutconsideringmarginconsequences.Giventhatdiscounts erodemarginsandthatourresultssuggestaneedforhigh-value productdiscountsandcontinuousdeepdiscountingtomeet cus-tomer expectations, retailers should consider the (long-term) profitimplicationscarefully.

Limitationsandfutureresearch

This initial study, using actual transactional data, of the effects oftwo typesof permanentdiscounts providedin digi-talchannelspartlyconfirmsourpredictionsandadvancesnew insights.However,it alsosuffersseverallimitations.Wehave accesstodatafromonlyonedigitalretailer,andthedataperiodis justoneyear.Richerdata,includinginformationfrommultiple retailersover alonger period,couldusefully testthe general-izabilityofourfindings.Further,wedonotknowthecostsof

productsinourdataset,informationthatwouldclarifythe influ-enceofthisdiscountstrategyonretailer profits.Although we knowhow muchof the value of acouponis redeemedinan order(inourcase,averageredeemedcouponvalueratiois64% attheindividuallevel),welackinformationaboutcouponexpiry dates.Inaddition,wehaveexaminedtheinfluenceof product-specificpricediscountsonpurchasequantityandspendingatthe basketlevel.Ifappropriatedataisavailable,futurestudiescould examinetheeffectofpricediscountsattheproductleveltosee howapricediscountimpactsthepurchaseoutcomesofa fea-turedproduct.Itwouldalsobeusefultoexplorewhetherhigher pricediscountstriggercross-buyingbehavior,asthiswouldhelp toexplainwhypricediscountsinducehigherspending.Finally, thenatureofourdatapreventsusfromestablishingtheeffectof currentdiscountlevelsonpurchaseincidence.

Despitethese limitations, this study offers useful insights regardingthe uniqueinfluencesof product-specific discounts and coupons on customer purchasing behavior. Accordingly, we notesomepromisingdirections for futureresearch.First, experimentaldatacouldbeusedtodeterminethecausal relation-shipbetweendigitaldiscountsandcustomerspending.Second, researcherscouldtestfortheeffectsofproductcategoryonthe twotypesofdiscountstrategies;consumersmayexhibitvarying (discount)sensitivityacrosscategories,andretailerswillobtain differentprofitlevelsacrosscategories,dependingontheir mar-gins.Third,consumerresponsestodiscountsmayvaryacross onlineandmobilechannels;furtherresearchshoulddetermine whethersuchadifferenceexistsandhowretailerscanbest dif-ferentiatethesechannels intheiroverall marketingstrategies. Fourth,attributionmodels,insteadofaggregatemeasures,could beusedtoaddresstherolesofdifferentdigitalchannelsandtheir contributionstoretailersalesandrevenues.Fifth,advanced tech-nologies(e.g.,artificialintelligence)mayhelpdigitalretailersto applythepermanentdiscountsstrategyinamoreefficientway, forexample,bytargetingdifferentcustomersinrealtimeonthe basisoftheirparticulardemographicandbehavioral character-istics.Weanticipatemorestudiesofthecombinationofdigital channelsandothertechnologiesintermsoftheirinfluenceon customerpurchase behavior.Such researchwould extendthe findingsofthepresentstudyinvaluableways.

Acknowledgements

ThisworkisfinanciallysupportedbytheChineseScholarship Council(CSC),theFundamentalResearchFundsfortheCentral UniversitiesNo.63202030,andthe National NaturalScience FoundationofChinaNo.71972175.

References

Ailawadi,Kusum L.,KarenGedenk, ChristianLutzkyand ScottA.Neslin (2007),“DecompositionoftheSalesImpactofPromotion-induced Stock-piling,”JournalofMarketingResearch,44(3),450–67.

Andrews,Michelle,Xueming Luo,ZhengFangandJaakko Aspara(2014), “CauseMarketingEffectivenessandtheModeratingRoleofPrice Dis-counts,”JournalofMarketing,78(6),120–42.