wedded bliss or tainted love? stock market reactions to ...moorman/marketing... · alina sorescu...

This article was downloaded by: [152.3.153.225] On: 23 April 2015, At: 16:50Publisher: Institute for Operations Research and the Management Sciences (INFORMS)INFORMS is located in Maryland, USA

Marketing Science

Publication details, including instructions for authors and subscription information:http://pubsonline.informs.org

Wedded Bliss or Tainted Love? Stock Market Reactions tothe Introduction of Cobranded ProductsZixia Cao, Alina Sorescu

To cite this article:Zixia Cao, Alina Sorescu (2013) Wedded Bliss or Tainted Love? Stock Market Reactions to the Introduction of CobrandedProducts. Marketing Science 32(6):939-959. http://dx.doi.org/10.1287/mksc.2013.0806

Full terms and conditions of use: http://pubsonline.informs.org/page/terms-and-conditions

This article may be used only for the purposes of research, teaching, and/or private study. Commercial useor systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisherapproval, unless otherwise noted. For more information, contact [email protected].

The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitnessfor a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, orinclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, orsupport of claims made of that product, publication, or service.

Copyright © 2013, INFORMS

Please scroll down for article—it is on subsequent pages

INFORMS is the largest professional society in the world for professionals in the fields of operations research, managementscience, and analytics.For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org

http://pubsonline.informs.org

http://dx.doi.org/10.1287/mksc.2013.0806

http://pubsonline.informs.org/page/terms-and-conditions

http://www.informs.org

Vol. 32, No. 6, November–December 2013, pp. 939–959ISSN 0732-2399 (print) � ISSN 1526-548X (online) http://dx.doi.org/10.1287/mksc.2013.0806

© 2013 INFORMS

Wedded Bliss or Tainted Love? Stock Market Reactionsto the Introduction of Cobranded Products

Zixia CaoWest Texas A&M University, Canyon, Texas 79016, [email protected]

Alina SorescuTexas A&M University, College Station, Texas 77843, [email protected]

We examine whether cobranding—the practice of using two established brand names on the same product—increases the market value of parent firms. Using data from the consumer packaged goods industry, we

document that the average stock market reaction to the announcement of cobranded new products is approxi-mately +1.0%. We hypothesize that this reaction is significantly higher than it would have been if these sameproducts were single branded, and we find evidence consistent with this hypothesis. We also examine thedeterminants of this stock market reaction. We find that the consistency between the two brand images, the inno-vativeness of the product, and the exclusivity of the cobranding relationship significantly increase the marketreaction to cobranding announcements. Our findings provide important managerial guidelines for enhancingfirm value through cobranding partnerships.

Key words : cobranding; new products; stock market reaction; marketing alliances; propensity score matchingHistory : Received: September 9, 2012; accepted: July 2, 2013; Preyas Desai served as the editor-in-chief

and Scott Neslin served as associate editor for this article. Published online in Articles in AdvanceSeptember 5, 2013.

IntroductionCobranding is the practice of using the establishedbrand names of two different companies for the samephysical product (e.g., Aaker 2004, Blackett and Boad1999). From Dell Computers with Intel Processors,to Kellogg’s Star Wars cereal, to Philips shavers dis-pensing Nivea shaving cream, cobranded productstake many forms across industries, at times connect-ing seemingly unlikely alliance partners. Industriessuch as the credit card industry have significantlyincreased product offerings through the practice ofcobranding (Spethmann and Benezra 1994), and arecent Nielsen study reports that “co-branded creditcards comprise approximately 50% of all credit cardspending.”1 The business press has touted cobrand-ing as a source of competitive advantage, calling it“a holy grail in 0 0 0 differentiating your brand, estab-lishing consumer trust, gaining new channels of dis-tribution or launching a new product successfully”(Thompson 1998, p. 22). Cobranding success storiesare regularly featured on the news. A recent exam-ple is Taco Bell’s popular Doritos Locos Tacos, a 2012introduction leveraging Yum! Brands’ Doritos brand,which now sells about one million units every dayand is credited with having helped create 15,000 jobsfor the Yum! Brands subsidiary (Schriffen et al. 2013).

1 See http://www.mastercard.us/merchants/cobrand-cards.html,accessed April 20, 2013.

However, despite practitioners’ apparent enthusi-asm toward cobranding, its effect on shareholdervalue is not well understood. Most academic researchon cobranding is conducted in a lab setting.2

Although this research shows that cobranded prod-ucts are generally viewed favorably by consumers,such preferences do not necessarily translate intocorporate profits. Profits are also dependent oncontextual variables that are typically not includedin experimental research (e.g., size of the market,competition, marketing support). Cobranded prod-ucts could be costlier to develop and riskier to marketcompared with single-branded products. For exam-ple, the parent firm may incur significant coordinationcosts for the development of infrastructure, researchand development (R&D), and manufacturing knowl-edge required to properly combine the two brandsinto a new product. In addition, negative associationscould transfer from partner brands to the cobrandedproduct, hindering its market success. Alternatively,negative associations could also transfer from thecobranded product to one of the partner brands. In anexperiment intended to assess preferences for brown-ies made from a cobranded mix, Levin et al. (1996)found that if one partner brand is thought to be infe-rior (in their case, the brand of chocolate chips used

2 For an exception, see Venkatesh and Mahajan (1997), who analyt-ically modeled the optimal price of cobranded products.

939

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.

Cao and Sorescu: Stock Market Reactions to the Introduction of Cobranded Products940 Marketing Science 32(6), pp. 939–959, © 2013 INFORMS

in the brownie mix), it brings down not only the per-ception of the cobranded product but also that of theother partner brand.

We are the first, to our knowledge, to providemarket-based evidence on the relation betweencobranding and shareholder value. We do so by exam-ining abnormal stock returns around the introductionof new cobranded products. We seek to understand thedeterminants of these abnormal returns and whethercobranding adds value to these products. Specifically,we investigate the following two questions:

1. What is the sign and magnitude of the stock mar-ket reaction to the introduction of cobranded productsthat can be attributed to the cobranding decision?

2. Among cobranded products, which attributesare most valued by stock market investors?

We rely on two streams of research to build ourtheoretical framework. First, we draw on studies inconsumer research that examine consumer attitudestoward branding, brand extensions, and cobrandedproducts. These studies suggest that attitudes towardpartner brands affect attitudes toward cobrandedproducts, and vice versa (e.g., Park et al. 1996,Simonin and Ruth 1998, Walchli 2007). Second, wedraw on studies that examine the stock market reac-tion to corporate announcements about branding,new products, and alliances. Although these studiesfind that new product announcements usually affectstock prices, contingencies typically affect the mag-nitude and direction of the stock price reaction. Forinstance, the market reacts positively to brand exten-sion announcements, but only in the case of brandsthat enjoy positive consumer attitudes and high famil-iarity (Lane and Jacobson 1995).

The empirical context for our research is the con-sumer packaged goods industry. We assemble asample of 316 announcements of cobranded newproducts made during the 2000–2010 period. Theseannouncements correspond to 61 primary brandfirms that are publicly traded. We use product-level data obtained from Datamonitor’s ProductLaunch Analytics database, archival data on firmannouncements obtained from LexisNexis and Factiva,firm-level data obtained from COMPUSTAT and theCenter for Research in Security Prices (CRSP), and sur-vey data obtained from Amazon’s Mechanical Turk.

We document positive and significant abnormalstock returns around cobranded product announce-ments. We also measure the average treatment effect ofthe treated (ATET), which is the average value thatthe cobranding aspect alone adds to the cobrandedproducts. We find that ATET is significantly posi-tive, which suggests that abnormal returns observedaround cobranded product announcements are higherthan they would have been if these same productswere single branded.

Our analysis also shows which products are morelikely to be cobranded. At the product level, firmsare more likely to cobrand products targeted to chil-dren and products with multiple stock-keeping units(SKUs). At the firm level, having a house of brandsincreases the likelihood of cobranding, as does firmsize and prior experience with cobranding partner-ships. Firms are also more likely to use one of theirbigger brands in cobranding partnerships, rather thana smaller niche brand. At the category level, firms aremore likely to introduce cobranded products in rela-tively smaller categories but ones in which cobrand-ing has been frequently used in the past.

We also find that among products that arecobranded, abnormal returns measured aroundannouncement dates are higher when consumers per-ceive the two cobrands to be consistent with eachother, when the product itself is innovative, and whenthe cobranding relationship is exclusive.

Data from Product Launch Analytics can shedlight on the prevalence of cobranding in the con-sumer packaged goods industry over the past decade.Specifically, this database lists 4,659 cobranded prod-ucts that have been introduced in the United Statesby all firms in that industry (public and private) from2000 through 2010. The annual number of cobrandedintroductions has steadily increased in the early 2000s,has declined from 2007 to 2009, and has resumedan upward trend in 2010 with 461 cobranded prod-ucts introduced that year. Overall, an average 420cobranded products per year were introduced duringour sample period.

Although our results have been derived within theconsumer packaged goods context, they nonethelessprovide a guiding principle for product managersin other industries—not all products benefit fromcobranding, but for some products, cobranding cansignificantly enhance shareholder value. The positiveATET documented in this paper indicates that man-agers can use cobranding as a tool to earn economicrents from products that are suitable for this strategy.

The rest of the paper is structured as follows. Thenext section outlines the conceptual foundation of ourtheoretical framework and proposes hypotheses aboutthe stock market reaction to cobranding announce-ments. This is followed by a methods section wherewe test our hypotheses and provide additional analy-ses of returns to cobranded new products. The paperconcludes with a discussion of managerial implica-tions and suggestions for future research.

Theoretical FrameworkCobranding as a Type of Brand AllianceBrand alliances can take many forms, from productbundling, to dual branding, to cobranding. Product

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.

Cao and Sorescu: Stock Market Reactions to the Introduction of Cobranded ProductsMarketing Science 32(6), pp. 939–959, © 2013 INFORMS 941

bundling is a strategy in which two or more differ-ent products are sold together for one price (Gaethet al. 1990, Yadav 1994). In many instances the com-ponents of the bundle carry the same brand, but thereare cases in which different brands are sold togetherin one package (e.g., the fragrance or skincare multi-brand packs sold by Sephora). Product bundling isalso encountered in promotions, where typically onebranded product is offered for free with the purchaseof another branded product (e.g., Varadarajan 1986).Dual branding is the concept of hybridized retailersusing a single location site, such as Sears and JiffyLube or Arby’s and John Long Silver’s sharing thesame retail space (Levin et al. 1996).

In this paper we focus on a specific type of brandalliance: cobranding. Cobranding involves two brandsthat are typically independent before, during, andafter the commercialization of the cobranded prod-uct but lend both their names to a single physicalproduct for the duration of the alliance. Cobrand-ing alliances differ from typical marketing alliances inthat they also involve fixed costs in R&D and manu-facturing related to the development and productionof the new product. Thus, they require a higher levelof firm commitment since it might be easier and lesscostly to discontinue an advertising or promotionalcampaign involving two partners than to discontinuea cobranded product for which the manufacturer hascommitted production infrastructure and retail space.

Multiple terms have been used in the literatureto label the two brands involved in a cobrandingalliance: modifier and modified brand, primary andsecondary brand, leader and partner brand, baseand supplemental product, for example (e.g., Levinet al. 1996, Uggla and Asberg 2010). In our paperwe adopt the primary and secondary brand termi-nology. The term primary brand denotes the brandof the firm that manufactures the cobranded prod-uct, and the secondary brand refers to the other brandinvolved in the partnership (Helmig et al. 2008). Forinstance, Edy’s/Dreyer’s Grand Ice Cream introducedits Galactic Chocolate and Vanilla Treasure flavorsthrough a cobranding partnership with Disney’s Trea-sure Planet movie in 2002. In this case, Edy’s/Dreyer’sis the primary brand and Disney’s Treasure Planet isthe secondary brand.

We also note that not all cobranding partnershipsare structured in the same manner. Across cobrandedproducts, the two partners bring different levels ofcontribution. We investigate and control for thesedifferences in our empirical analysis. Our empha-sis, however, is on the stock returns to the primarybrand parent. We do not examine the parents of sec-ondary brand products. Our sample contains onlyproduct manufacturers—the parents of primary brandproducts.

Financial Returns to CobrandingPrior research shows that financial returns tonew products accrue mostly to radical innova-tions: products that are significantly new on somedimension of relevance to consumers (Sorescu andSpanjol 2008, Srinivasan et al. 2009). Cobrandedproducts are typically incremental (as opposed toradical) innovations, suggesting that—at least froman innovation perspective—they might not elicit largestock market reactions to their announcements. How-ever, new cobranded products also share two uniquefeatures that are likely to be viewed more favorablyby investors when compared with single-brandedproducts.

First, cobranding could signal quality to consumers(Rao et al. 1999) and can improve consumer attitudestoward individual partner brands. Positive brandassociation spillovers have been documented from theindividual brands to the cobranded product, and viceversa (Simonin and Ruth 1998). Cobranded productscan command a price premium, possibly because theyare viewed as unique and different by consumers.For instance, Venkatesh and Mahajan (1997) foundthat including an Intel 486 instead of a baseline chipinto a Compaq computer would yield a price pre-mium of $140. Cobranded products have also beenfound to elicit more positive perceptions than single-brand extensions, arguably because cobranded prod-ucts might benefit from the secondary brand’s equity(Desai and Keller 2002, Park et al. 1996).

Second, cobranding alliances could offer opportuni-ties for improved operational efficiencies. Cobrandingpartners can gain access to new markets and shareeach other’s resources in terms of manufacturing,managerial knowledge, and advertising. Cobrandingalliances involve both a marketing aspect and a tech-nological aspect, because they involve the creationand commercialization of a new product.

The effect of the marketing aspect is uncertainbecause marketing alliances do not always increaseshareholder value—the stock market reaction ispositive in some studies (e.g., Swaminathan andMoorman 2009) and insignificant in others (e.g., Daset al. 1998, Koh and Venkatraman 1991), potentiallyreflecting differences between investors’ perceptionsof such alliances. Moreover, the volatility of stockreturns seems to increase following announcementsof marketing alliances, perhaps because of additionalrisks such as opportunistic partner behavior (Daset al. 1998). In contrast, the effect of the technologicalaspect is likely positive, consistent with the marketreaction to new product and technological alliances(e.g., Kalaignanam et al. 2007).

These prior studies suggest that cobranding canincrease shareholder value for some but not nec-essarily all products. The extent to which a prod-uct stands to benefit from cobranding depends on

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


product-, firm-, and industry-specific factors. A cost-benefit analysis should be performed before eachcobranding decision. Assuming that managers arerational, on average, they will choose to cobrandproducts for which the benefits of cobranding areexpected to exceed the costs. Thus, we should observethat, on average, the increase in shareholder valueassociated with cobranded products will be largerthan it would have been had these same productsbeen single branded.

Hypothesis 1 (H1). Average abnormal returns to theannouncement of cobranded product introductions arelarger than they would have been if these same productswere single branded.

Determinants of Abnormal Returns to theIntroduction of Cobranded ProductsOnce a decision to launch a new cobranded productis made, managers might want to understand whichproduct attributes can determine the success of thenew product in the marketplace. We identify the con-sistency between the two brands, the exclusivity ofthe cobranding agreement, and the innovativeness ofthe cobranded product as attributes that are likely tobe valued by stock market investors.

Consistency. The literature highlights one char-acteristic that can elicit positive brand associationsfor cobranded products: the consistency between theimages of the two partner brands. Brand consis-tency, defined as the congruence or fit between two(or more) brand images (Keller 1993), is positivelyrelated to attitudes toward brand extensions (Aakerand Keller 1990, Batra et al. 2010) and to attitudestoward brand alliances (Simonin and Ruth 1998).

The fit between two brands can override the asso-ciations that consumers might have with individ-ual brands. Park et al. (1996) found that cobrandedproducts enjoy better recognition when they carrytwo complementary brands rather than two brandsthat are viewed as highly favorable but not com-plementary. Their findings support the predictionsof cognitive consistency theory, which suggests thatindividuals are more likely to view an object favor-ably (and by extension, choose that object amongalternatives) if it does not involve dissonant elements.

In sum, when brands have a consistent image,spillovers of positive attitudes and perceptions ofquality are more likely to transfer between the twobrand partners or between the respective brands andtheir cobranded products. For the primary brandmanufacturer, these positive attitudes should trans-late into higher and less volatile cash flows for thecobranded product. If investors recognize the value ofconsistency, the increase in shareholder value shouldbe higher for cobranding announcements that involveconsistent brand partners.

Hypothesis 2 (H2). Abnormal returns associated withthe announcement of cobranded product introductions arepositively related to the consistency between the two part-ner brands’ images.

Exclusivity. An important dimension in cobrand-ing agreements is the exclusivity of the partnership.In line with industry practice, we focus only on casesof exclusivity from the perspective of the secondarybrand partner. These are cases where the secondarybrand agrees to participate in a cobranding agreementwith a single primary brand firm and does not partic-ipate in similar agreements with the primary brand’scompetitors (e.g., Bucklin and Sengupta 1993). Forinstance, Kellogg’s partnership with Disney specifiesthat only Kellogg’s can use select Disney characterson the packages of its breakfast cereal, but it does notprevent Kellogg’s from entering into future cobrand-ing partnerships with other firms (Verrier 2001).

Exclusivity in cobranding can function as a com-mitment mechanism that limits the secondary brandpartner’s ex post options and protects the pri-mary brand partner from opportunistic behavior(Williamson 1983). In an exclusive partnership, thesecondary brand partner has stronger incentives tohelp the cobranded products turn into enduringassets, which can further enhance the value of thepartnership. In addition, the primary brand partnercan more freely contribute its capabilities since theexclusivity provision makes it less likely that criticaltechnology and skills would transfer to rival firms(Aghion and Bolton 1987). Thus, from a strategicand operational standpoint, an exclusive agreementshould be beneficial to the brand partnership. Like-wise, from the standpoint of consumer perceptions,exclusivity could strengthen brand image for the pri-mary partner, although the lack of exclusivity coulddilute it (e.g., Park et al. 1986). In sum, exclusivityincreases the uniqueness of cobranded products andshould be a source of competitive advantage for theparent firm (Krattenmaker and Salop 1986).

Hypothesis 3 (H3). Abnormal returns associated withthe announcement of cobranded product introductions arehigher for exclusive cobranding partnerships than fornonexclusive partnerships.

Innovativeness. In a meta-analysis, Henard andSzymanski (2001) found that there is no system-atic relationship between product innovativeness andnew product performance, on average. However,Sorescu (2012) observed that, in many studies, innova-tiveness is significantly related to performance (in par-ticular, stock performance), but most of these studiesare based on samples of highly salient or radical inno-vations in the high-tech industries.

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


This calls into question whether product innova-tion in non-high-tech industries would elicit a posi-tive stock market reaction, especially since innovativeproducts are known to increase the risk of their par-ent firms as a result of uncertainty about adoptionby the marketplace (Sorescu and Spanjol 2008). Weargue, however, that this uncertainty is reduced bybranding these products with not one but two estab-lished brands. A transfer of positive associations fromeither partner brand should increase the credibility ofthe product’s new features and its overall perceptionsof quality, which should translate into higher share-holder value.

Innovativeness might also minimize losses fromlaunching an unsuccessful product. When a prod-uct brand extension fails, the parent brand is dilutedmostly if the product is similar to other firm productsthat carry the same brand (Keller and Aaker 1992,Loken and Roedder John 1993). An innovative prod-uct, particularly a cobranded product, is more likelyto be dissimilar from the parent’s portfolio of single-branded products, which should limit negative asso-ciations and brand damages in the case of marketfailures. Overall, we expect that the increase in share-holder value will be higher for cobranded productsthat are innovative.

Hypothesis 4 (H4). Abnormal returns associated withthe announcement of cobranded product introductions arepositively related to product innovativeness.

Taken together, our four hypotheses predict thatthe stock market reaction to cobranded productannouncements is stronger than it would be ifthese same products were single branded. Oncethe cobranding decision is made, we predict ahigher increase in shareholder value for productsthat are innovative, for products with consistentcobrands, and for products with exclusive cobrandingagreements.

MethodEmpirical ContextWe test our hypotheses using data from the consumerpackaged goods industry. We select this industry fortwo reasons. First, consumer packaged goods accountfor a sizeable portion of the U.S. economy. Accordingto the Grocery Manufacturers Association (2013), thefood, beverage, and consumer packaged goods indus-try employs 14 million U.S. workers and contributesmore than $1 trillion to the nation’s economy. Second,cobranding is a prevalent practice in this industry.As mentioned previously, a total of 4,659 cobrandedproducts were introduced in the United States overthe last decade, as reported by Datamonitor’s ProductLaunch Analytics database.

Data and SampleTo test our hypotheses, we use Product LaunchAnalytics to build a sample of cobranded prod-ucts for which we can identify both the primaryand secondary brand parent firms. A comprehen-sive and detailed source of product information thatincludes consumer packaged goods launched aroundthe world since the early 1980s, this database pro-vides, among other things, the date of product intro-duction, the manufacturer, an assessment of the prod-uct’s innovativeness, and a tag identifying productsthat are cobranded or that carry a double trade-mark. Moreover, products are added to this databaseat the time they are launched, eliminating potentialmemory biases related to new product selection andto classification along relevant dimensions (such asinnovativeness).

We obtain from Product Launch Analytics all con-sumer packaged goods introduced in the UnitedStates between 2000 and 2010 that carry the“cobranded” or “double trademark” tag. Of the prod-ucts listed, 4,659 meet this initial screening criterion.We identify manufacturers for all products and retainonly those that are publicly traded. For cobrandedproducts introduced by publicly traded firms, weidentify the primary and secondary brands. The pri-mary brand is usually the manufacturer’s corporatebrand or one of the brands under its umbrella. Thesecondary brand is the other brand that appears onthe product’s package and is identified by ProductLaunch Analytics as being the cobranded partner.

Product Launch Analytics reports the date onwhich each product is added to the database. Thisis also the date when it publicizes the new productthorough their internal Product Alert news releaseservice. In some cases an earlier announcement inother publications could precede the Product Alertnews release date. To identify these cases, we con-duct, for each cobranded product launched by pub-licly traded firms, searches in Factiva and LexisNexisto identify the earliest date when public informa-tion about that product became available. To ensureaccuracy, two researchers independently conductedsearches to identify cobranded products announce-ments and subsequently reconciled their differences.When no formal announcement was found and thefirst publicly available mention of that product was inProduct Alert, we used the Product Alert date as theannouncement date.

In most cases, a single cobranding announcementincludes several cobranded products to be introducedover a period of time. For example, on February 19,2001, Coca-Cola announced a cobranding partnershipwith the Harry Potter and the Sorcerer’s Stone movie

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


that would yield multiple products carrying the Coca-Cola brand. In such cases we treat the announce-ment as a single event because all relevant informa-tion regarding the variables of interest is conveyedon the announcement date. Subsequent product intro-ductions that result from the cobranding partnershipare no longer a surprise to investors.3

After eliminating products introduced by privatelyheld firms and foreign firms, and after eliminatingmultiple products covered by a single announcement,we narrowed our sample to 403 announcements ofcobranding partnerships. We used the RavenPackNews Analytics (a Dow Jones database that consoli-dates corporate news for 28,000 publicly traded com-panies in 86 countries) to determine whether therewere confounding events on the day of these cobrand-ing announcements. A total of 87 announcementshad confounding events and were eliminated fromour sample, because the stock market reaction tothe cobranding announcement cannot be disentangledfrom the reaction to the confounding event. Eliminat-ing announcements with confounding events is in linewith other event studies in marketing (e.g., Boyd et al.2010, Chen et al. 2012).

The final sample contains 316 announcementsof cobranded partnerships during the period from2000 to 2010. These announcements correspond to61 publicly traded primary brand firms. Of these316 announcements, 195 involve secondary brandsbelonging to 42 publicly traded companies, with theremaining secondary brands belonging to private orforeign firms. During our sample period, the 61 pri-mary brand firms introduced a total 21,130 single-branded products and 1,543 cobranded products.

Testing H1 requires that we account for selectionbias because the decision to cobrand is likely endoge-nous. We do so by estimating a switching regressionmodel that accounts for selection based on unobserv-able characteristics and, alternatively, by estimating apropensity score model. The latter requires that wematch each cobranded product with a similar single-branded control product introduced by the same firm.To this end, we collected additional data on all prod-ucts introduced by the 61 primary brand firms in oursample from 1990 to 2000: the total count of all prod-ucts introduced by these 61 firms from 1990 to 2000is 14,539. These single-branded products introducedprior to 2000 are used exclusively as candidates forcontrol products in the propensity score model. Wecollected these additional products back to 1990 to

3 We manually ascribed each cobranded product retrieved fromProduct Launch Analytics to its relevant announcement. Wheneverthe partnership included an innovative product, this was intro-duced first. Thus, the information contained in the innovativenessrating is also available at the time of the original announcement.

reduce data censoring in the process of selecting con-trol products for the propensity score model. Had weselected control products from among new productsintroduced during our sample period (2000–2010),cobranded products introduced at the beginning ofthe sample would have likely matched with single-branded products launched at a subsequent date.

After constructing our product sample, we performarchival searches in Factiva and LexisNexis to obtaindata on the exclusivity of the cobranding agreements;the exact process is described later in the next sec-tion. The next section also describes the survey-basedmethod we used to collect data on cobranding con-sistency. Data on firm-level control variables (such asfirm size) are obtained from COMPUSTAT and Mer-gent, and stock returns are obtained from CRSP.

MeasuresWe now develop empirical measures for our depen-dent and independent variables. These measures arediscussed below and summarized in Table 1.

Dependent Variables: Short-Term Abnormal Re-turns. Short-term event studies have frequently beenused to measure the stock market reaction to corpo-rate announcements such as new product introduc-tions (Chaney et al. 1991), brand extensions (Laneand Jacobson 1995), alliances (Swaminathan andMoorman 2009), and additions of Internet channels(Geyskens et al. 2002). The methodology is well estab-lished and well specified over short-term horizons(Brown and Warner 1985).

We use the market model to estimate the short-term market reaction to the introduction of cobrandedproducts (Brown and Warner 1985). Specifically, weestimate abnormal returns (ARs) for each firm thatintroduces a cobranded product as follows:

ARit =Rit − 4�i + �iRmt51 (1)

where Rit is the rate of return of stock i on day t,Rmt is the rate of return on the stock market index onday t, and � and � are the parameters of the mar-ket model estimated from an ordinary least squaresregression of Rit on Rmt over a period ranging from200 to 30 trading days prior to the announcementdate. The daily abnormal returns are then cumu-lated over a time window (t11 t2), which includes theannouncement day:

CAR4t11 t25=

t2∑

t=t1

ARit0 (2)

For robustness, we also compute abnormal returnsusing the Fama-French-Carhart four-factor model,which augments the market model with three addi-tional risk factors that have been shown to explain the

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Table 1 Variables and Data Sources

Conceptual variable Equation (4) Equation (7) Description Data source

Dependent variablesCobranded product

√Dummy (1 if the product is cobranded and 0 otherwise) Product Launch Analytics

CAR√

Cumulative abnormal return (over a five-day window)computed using the market model and theFama-French-Carhart four-factor model

CRSP

Independent variablesExclusivity

√The manufacturer of the secondary brand has not been

involved in other cobranding agreements over theprevious 10 years (110)

Factiva and Product LaunchAnalytics

Consistency√

Three-item scale of brand fit (Helmig et al. 2007) Amazon Mechanical Turksurveys

Innovativeness√

Extent to which the first introduced product is new to theindustry (110)

Product Launch Analytics

Cobranding equation-independentvariables/Control variables

Type of cobranding agreement√

Dummy variables for(a) endorsement cobranding(b) ingredient cobranding(c) composite cobranding

Factiva and Product LaunchAnalytics

Preannouncement√

Dummy variable (1 if the product is ready for introductionand 0 otherwise)

Factiva

News outreach√

Categorical variable on a scale of 1 to 5, where 1 = lowreach and 5 = high reach

Factiva

Category cobranding√

Number of cobranded products introduced over the pastyear into the category in which the focal product isintroduced


Experience√ √

Number of cobranding products introduced by the pri-mary brand in past five years before the introduction ofthe focal product


Firm size√ √

Total assets (log) COMPUSTATCorporate brand

√ √Dummy (1 for corporate brand and 0 for house of brands) Mergent

Brand strength√ √

Percentage of the primary brand’s products that carry thefocal brand name introduced five years before the intro-duction of each cobranded product


Market size√ √

Number of products introduced in the same category asthe cobranded product during the five years precedingits introduction


Targeted to children√ √

Dummy (1 if the product has the “kids” tag, 0 otherwise) Product Launch AnalyticsNumber of SKUs

√ √Number of SKUs associated with each product introduced Product Launch Analytics

cross section of stock returns (see Carhart 1997, Famaand French 1993):

ARit =Rit−4�+�Rmt+�SMBt+ �HMLt+�UMDt51 (3)

where Rit and Rmt are as previously defined, SMBt isthe return differential between portfolios of small andlarge market capitalization stocks, HMLt is the returndifferential between portfolios of high (value) and low(growth) book-to-market ratio stocks, and UMDt isthe momentum factor computed as the return differ-ential between portfolios of high and low prior-returnstocks.

To choose the appropriate length of the eventwindow, we computed cumulative abnormal returns(CARs) for various event windows, beginning twodays before the announcement and ending two daysafter the announcement. We tested the significance

of the CARs in each event window and selected theevent window with the most significant t-statistic(Geyskens et al. 2002, Swaminathan and Moorman2009). Our event window begins two days before theannouncement and ends two days after the announce-ment [t − 21 t + 2].

Independent Variables. Independent variables in-clude the focal variables required for testing H2–H4(consistency, exclusivity, and innovativeness), as well asadditional control variables and variables that areused to estimate the selection equation.

Consistency. Consistency in brand images can beconstrued as consistency on one or more dimensions:quality perceptions, brand associations, or specificfacets of the two brands’ personalities. Assessingconsistency along multiple dimensions of brand imageis beyond the scope of this paper (see, alternatively,

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Batra et al. 2010). Instead, we follow prior research inbranding and treat consistency as a one-dimensionalconstruct (e.g., Park et al. 1991, Simonin andRuth 1998).

To evaluate the consistency between each pair ofprimary and secondary brands, we use the three-itemscale of brand fit developed by Helmig et al. (2007).Specifically, using a seven-point scale (1 = stronglydisagree, 7 = strongly agree), we measure consumeragreements with the following statements:

• The following primary and secondary brands arecomplementary and fit well together.

• The brand images of the primary and secondarybrands are endorsing each other.

• The combination of brand images of the primaryand secondary brands leads to a consistent new brandimage for the cobranded product.

We included these scales in a survey that waslaunched on the Amazon Mechanical Turk. To ensurethat survey participants would not have an overlylengthy task, we split the sample in eight parts andincluded no more than 40 primary–secondary brandpairs to be evaluated within each survey. Each sur-vey contained a block of three questions (based on theHelmig et al. 2007 scale) for each pair of brands andconcluded with a few demographic questions. Eachsurvey instrument was completed by 40 MechanicalTurk respondents. We requested that respondents beU.S.-based and have a previous task approval rateof greater than equal 95%. The average age of therespondents is 36, with respondents ranging from 18to 83 years old. Of the respondents, 42% were maleand 58% female.

We measured the reliability of this three-item scaleand found that Cronbach’s � across all surveys is 0.75or higher. We thus averaged, for each respondent, theresponses to the three consistency questions, and thenwe averaged the responses across each set of 40 sur-vey participants to obtain one consistency measure foreach pair of primary–secondary brands.4

Exclusivity. We use two different methods to iden-tify cases where the secondary brand has an exclu-sive partnership with the primary brand. The firstmethod is implicit: we search for evidence of previ-ous partnerships in Product Launch Analytics, Fac-tiva, and LexisNexis, beginning with 10 years prior tothe introduction of the cobranded product. To illus-trate, a cobranding partnership with NutraSweet isnot exclusive, since this brand has partnered withmany food manufacturers. Alternatively, the fitnessbrand Curves’ partnership with General Mills, which

4 We note that the consistency measure is not contemporaneous tothe introduction of the product and could have been potentiallyinfluenced by past branding strategies of the parent firm. This is alimitation of our study.

led to Curves’ branded cereals and cereal bars, isexclusive because General Mills is the only consumerpackaged goods manufacturer that has established apartnership with Curves. Exclusivity is coded as adummy variable that takes value a equal to 1 if thecobranding agreement is exclusive and is 0 otherwise.

The second measure of exclusivity is explicit: weread the cobranded product announcements to deter-mine whether the agreement contains an exclusiv-ity provision. We found, however, that cobrandingannouncements rarely include information on exclu-sivity. Thus, our explicit measure of exclusivity mayincorrectly classify some of the exclusive partner-ships as nonexclusive if the information provided inthe published announcement is incomplete. Resultsobtained with this explicit measure are reported in therobustness section.

Innovativeness. To identify innovative products, weuse the “innovative” rating available in ProductLaunch Analytics. This rating is assigned by thedatabase’s staff experts at the time of product intro-duction and identifies products that are new tothe market in terms of formulation, packaging, ormerchandising. An example of a cobranded prod-uct that is innovative on a formulation dimen-sion is Budweiser & Clamato Chelada, a flavoredmalt beverage that combined Budweiser beer andClamato tomato juice. Proctor & Gamble’s IntelliCleanToothbrush System (a rechargeable toothbrush witha liquid toothpaste container that carries both theSonicare and Crest brands) is innovative in terms ofboth formulation (the liquid toothpaste) and techno-logical innovation. If the cobranding agreement pro-vides for a series of products to be introduced throughtime, we use the innovativeness rating of the prod-uct (or products) launched at the time of the initialannouncement. Innovativeness is coded as a dummyvariable that is equal to 1 in the case of innovativeproducts and 0 otherwise.

Type of Cobranding Agreement. Cobranding agree-ments can be classified into three categories, depend-ing on the relative contribution of the two partnersto the cobranding relationship: ingredient, composite,and endorsement. We use dummy variables to controlfor the type of cobranding agreement. To classify ourcobranding announcements into one of these threecategories, we used the following heuristics:

i. Endorsement cobranding occurs when the sec-ondary brand makes no contribution to product for-mulation but only appears on the package of thecobranded product for promotional purposes. In thisarrangement the two brands are “endorsing” eachother as they look to leverage the other brand’s pos-itive associations. For instance, Crest Barbie tooth-paste offers additional visibility to Hasbro’s Barbie; inturn, Barbie should presumably increase the appeal of

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


the toothpaste among children in the target market.In most of these cases, the secondary brand belongsto the entertainment industry, and thus classificationalong this dimension is straightforward.

ii. Ingredient cobranding is an agreement wherebythe secondary brand is an identifiable ingredient thatcontributes to product formulation, for products thatwere previously available in similar forms when theywere single branded. The secondary brand is featuredon the package; however, the primary brand’s charac-teristics remain clearly dominant.

iii. Composite cobranding occurs when both brandshave a significant contribution to the formulation andpositioning of the cobranded product, and when nosimilar version of this product was available in themarketplace prior to the cobranding agreement. Bothprimary and secondary brands are prominently fea-tured on the package and are an integral part of thecobranded product. The secondary brand appears tobe a partner in product design, rather than a supplier.For example, Kraft’s Handi-Snacks Baskin-Robbinsready-to-eat pudding, a type of Handi-Snacks pud-ding with flavors inspired by Baskin-Robbins’ icecream, is a composite cobranded product that promi-nently leverages the characteristics of both brandsand is quite different from the original Kraft prod-uct. By contrast, Coke with Splenda is an example ofingredient cobranding because it is essentially a dietsoda with a new ingredient.

Two experts independently classified all cobrandedproducts into one of these three categories. The initialagreement was 96%, and remaining differences wereresolved through discussion. Examples of cobrandedproducts in each category are presented in Table 2.

Experience 4Prior Cobranding Experience of the PrimaryBrand Partner5. The prior cobranding experience of theprimary brand partner can affect how stock pricesreact to the introduction of cobranded products. Along history of cobranding helps reduce investors’information asymmetry when estimating future cashflows, and it may also suggest that the firm hassuccessfully managed past cobranding partnerships.

Table 2 Examples of Cobranded Products

Type of cobranded product Name of product Primary brand (parent firm) Secondary brand (parent firm) Introduction date

Endorsement branding Crest Barbie toothpaste Crest (Procter & Gamble) Barbie (Mattel) 9/15/2006Bertie Bott’s Every Flavour

BeansCap Candy (Hasbro) Harry Potter (Warner Bros.

Consumer Products)2/11/2000

Ingredient branding Tuna Helper Complete withStarkist Tuna

Betty Crocker (General Mills) Starkist (Heinz) 2/11/2002

Diet Coke with Splenda Coke (Coca-Cola) Splenda (Johnson &Johnson)

2/7/2005

Composite branding Duncan Hines Fun Frosterswith Nestlé Crunch Candy

Duncan Hines (Aurora Foods) Nestlé Crunch (Nestlé) 5/2/2001

Sonicare Crest IntelliCleantoothbrush system

Crest (Procter & Gamble) Sonicare (Royal PhilipsElectronics)

10/1/2004

Thus, investors might be more optimistic about theprospects of cobranded products introduced by firmswith prior cobranding experience. On the other hand,cobranding announcements made by firms with priorcobranding experience may no longer contain a sur-prise element and could already be incorporated intostock prices. We measure prior cobranding experienceusing the number of cobranded products introducedby the primary brand during the five-year periodpreceding the announcement of the cobranded prod-uct. We collect this information from Product LaunchAnalytics and through archival searches in Factivaand LexisNexis.

Firm Size. We use the book value of firm assets tocontrol for the effect of firm size on abnormal returns.This is standard practice in event studies, since largerfirms typically have smaller percentage changes intheir stock prices in response to corporate announce-ments. As in prior studies, we use the log of firmassets as a proxy for firm size (e.g., Boyd et al. 2010).

Corporate Brand vs. House of Brands. We control forthe position of the primary brand in its parent brandportfolio. A partnership where the primary brand isa corporate brand is likely to be more salient than apartnership where the primary brand is from a house-of-brands portfolio. On average, corporate brandshave better established brand associations, and moreresources are available to support the brand and thepartnership. Thus, we expect a stronger market reac-tion when the primary brand is corporate as opposedto belonging to a house of brands. We use a dummyvariable that equals 1 for corporate primary brandsand 0 otherwise.

News Outreach. The reach of the news outlet wherethe cobranding agreement is announced could affectthe magnitude of the stock market reaction. Theannouncement is likely to have a larger audience ifit is publicized through a wide-reaching wire service(such as the Associated Press) as opposed to a tradepublication. We classified news outlets into five cat-egories, listed below in decreasing order of proba-ble outreach in the investment community: newswire,

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


marketing publications (e.g., Brandweek), trade publi-cations (e.g., Dairy Field), newspapers (e.g., the StarTribune, a Minneapolis newspaper that frequentlyreports on General Mills’ product introductions), andProduct Alert (Product Launch Analytics’ own infor-mation releasing service). We explored two specifica-tions of news outreach in our empirical analysis: (1)separate dummies for four out of the five outreachcategories and (2) a categorical variable capturing thereach of the news outlet.

Market Size. The variable market size refers the sizeof the category in which the cobranded product isintroduced. A product introduced in a large marketcould elicit a stronger stock market reaction com-pared with one introduced in a niche market. ProductLaunch Analytics provides a categorization of prod-ucts by market. We measure the market size by com-puting the number of products introduced in thatmarket during the five years preceding the introduc-tion of the cobranded product.

Brand Strength. The variable brand strength refersto the economic significance of the cobranded prod-uct brand to the primary brand parent firm. Stockprices may have a stronger reaction to a cobrandingannouncement if the firm cobrands a product that car-ries its flagship brand rather than a less importantbrand from its portfolio. To control for this possibility,we compute the strength of the primary brand usedon each cobranded product as the percentage of theprimary brand parent firm’s new products that carrythat particular brand name (as opposed to a differentbrand name that belongs to the same firm), calculatedover a five-year period prior to the introduction ofeach cobranded product.

Preannouncement vs. Introduction. Some cobrandingannouncements refer to products that are beinglaunched at the time of the announcement. Others arepreannouncements that mention a future introductiondate. The stock market reaction to preannouncementscould be more subdued as a result of residual uncer-tainties about the product’s ultimate prospects (Soodand Tellis 2009). We control for preannouncementsusing a dummy variable.

Targeted to Children. Among cobranded productsin the consumer packaged goods industry, manyare targeted to children. This is especially preva-lent among endorsement-type cobranding (see alsoGallagher 2007). We use a dummy variable for prod-ucts targeted to children, constructed using the “kids”tag available in Product Launch Analytics.

Number of SKUs. Cobranded products might carrymore SKUs, on average, than single-branded productsbecause the parent firm might seek to Also, a largenumber of SKUs might signal that the manufacturerbelieves that there is sufficient demand for multiplevariants of the product. Thus, products with higher

numbers of SKUs could be perceived more favorablyby investors.

Category Cobranding. A cobranded product ismore likely to be introduced in a category wherecobranding activity is high than in one where it israre. The prevalence of cobranding in each productcategory—a concept we call category cobranding—canbe used as an instrument for the decision to cobrand.We expect category cobranding to be a determinant ofthe cobranding decision—the higher the cobrandingactivity in a given product category, the more likelyit is that products in that category will be cobranded.This is because cobranding is a more salient mar-keting tool among managers in product categorieswith high cobranding prevalence. Yet, at the sametime, category cobranding should have no effect onabnormal returns because it is not directly relatedto the cobranded product. A higher percentage ofcobranded products in a given category does not nec-essarily imply that new products in that categorycan extract higher rents, particularly after accountingfor other product-, firm-, and category-specific vari-ables. We measure category cobranding as the countof cobranded products introduced in a product cat-egory during the year preceding the introduction ofthe focal cobranded product.

ModelsTesting H1Testing H1 requires that we estimate the counterfac-tual CARs that cobranded products would generateif they were single branded. The metric of interestin this hypothesis is the difference between the mea-sured CARs and the counterfactual CARs, the ATET.This difference accounts for selection bias—the factthat the decision to cobrand is more likely strate-gic than random, reflecting systematic differencesbetween cobranded and single-branded products.5

A selection bias could arise if firms use cobrandingto further leverage an advantage that the new prod-uct may already hold, such as high brand awarenessor extensive distribution network. Another case of aselection bias is one where managers use cobranding tomitigate potential shortcomings that products wouldface if they were single branded. For instance, a firmmight enter into an ingredient cobranding agreementwith a secondary brand firm if it lacks the expertiseor credibility that the secondary brand has in pro-ducing that particular ingredient (e.g., using Splendamay be more appealing to certain firms than trying todevelop a proprietary artificial sweetener). We addressthe selection bias problem through a propensity score

5 We thank a reviewer for this insight.

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


model as well as through a switching regression modelthat controls for selection based on unobservables.

Basic Propensity Score Model. The propensityscore model produces an explicit measure of the coun-terfactual CARs required for testing H1. These arethe CARs earned by control products that are, in theideal case, similar to cobranded products in all aspectsexcept branding. To identify these control products,we estimate a probit model of the decision to cobrand.The dependent variable takes a value of 1 if the prod-uct is cobranded and 0 otherwise. In line with previ-ous research, we use a stepwise estimation to ensurethat only relevant covariates are included in the finalmodel (Rosenbaum and Rubin 1984):

Pr4Cij = 1 � Covariatesij1�j5

=ê4�0 +�1Experienceij +�2Corporate brandj

+�3Firm sizeij +�4Brand strengthij

+�5Market sizeij +�6Targeted to childrenij

+�7Number of SKUsij

+�8Category cobrandingij +�j51 (4)

where j denotes the firm, i denotes the announce-ment, Cij equals 1 if product i by firm j is cobrandedand 0 otherwise, �j is a random, firm-specific effect, êis the standard normal cumulative distribution func-tion, and other variables are as previously defined.

From this model we extract the propensity scoreas the predicted probability that a product would becobranded given its characteristics (Bronnenberg et al.2010, Huang et al. 2012, Rosenbaum and Rubin 1983).We then match each cobranded product with a single-branded product using, respectively, each of the fol-lowing six matching methods:

1. the single-branded product having the highestpropensity score from among all single-branded prod-ucts introduced by the same firm within a five-yearwindow,

2. the single-branded products having the closestpropensity scores to that of the cobranded productfrom among all single-branded products introducedby the same firm within a five-year window,

3. a matching method similar to method 1 but withthe matching window extended to the entire sample,

4. a matching method similar to method 2 butwith the matching window is extended to the entiresample,

5. a matching method similar to method 2 exceptthat the matching is done across firms rather thanwithin each firm, and

6. a matching method similar to method 4 exceptthat the matching is done across firms rather thanwithin each firm.

In sum, matching methods 1–4 perform matcheswithin the same firms, whereas matching meth-ods 1, 2, and 5 further restrict the matching periodto be within five years of the cobranded productintroduction.

Once the control products are obtained, we estimatethe counterfactual CARs as the actual CARs earnedby the control products. We then subtract the coun-terfactual CARs from the cobranded CARs and testwhether the mean and median differences are signif-icantly positive. This is our first formal test of H1.

Pairwise-Difference Propensity Score Model. Ifthe matching process in the propensity score modelis not perfect, control products might still differ fromcobranded products along the dimension of interest.To account for this possibility, we perform a sec-ond formal test of H1: we regress the difference inCARs (between cobranded and control products) onpairwise differences in the independent variables ofinterest:

DIFF_CARij

= �0 +�1DIFF_Innovativenessij

+�2DIFF_Brand strengthij +�3DIFF_Experienceij

+�4DIFF_Firm sizeij +�5DIFF_Corporate brandj

+�6DIFF_Market sizeij

+�7DIFF_Targeted to Childrenij

+�8DIFF_Number of SKUsij + �ij1 (5)

where j denotes the firm, i denotes the announce-ment, and DIFF_CAR is the difference between theabnormal returns of cobranded and control products.The independent variables are pairwise differences inthe previously defined variables.

We estimate this model for each of the six controlsamples that correspond to the six matching meth-ods described in the previous subsection. A signifi-cantly positive intercept supports H1 because it indi-cates that cobranded CARs are significantly higherthan counterfactual, single-branded CARs, even afteraccounting for known differences between cobrandedand control products.

Switching Regression Model. The propensityscore model implicitly assumes that there are nounobserved determinants of the decision to cobrand.It is, however, quite plausible that unobserved factors(such as a lack of experience with a key product fea-ture) might influence both CARs and the cobrandingdecision. To account for this possibility, we usea switching regression model (e.g., Verbeek 2008).6

6 We thank the associate editor for suggesting this methodology.

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Unlike the propensity score model, the switchingregression model does not produce a specific estimateof the counterfactual CARs (an intermediate step inthe propensity score model). Instead, the switchingregression model produces a direct estimate of theATET, the metric of interest for H1.

In this model, the equation governing CARsdepends on the branding regime. For each product,the value added of cobranding (ATET) is the sumof three separate components: a constant, a compo-nent that captures the effect of observable determi-nants of cobranding, and an idiosyncratic componentthat captures the effect of unobservable determinants(Verbeek 2008, p. 254). In simple terms, the switchingregression model can be described as follows:

CARij =�0 +è�kXijk + �Cij +è�kCijXijk + �ij1 (6a)

Cobrandij = �0 +è�kXijk

+ �Category cobrandingij +�ij1 (6b)

Cij = 1 if Cobrandij > 01 (6c)

where j denotes the firm, i denotes the announce-ment, and Xij are the determinants of theCARs and of the decision to cobrand (Xijk ∈

8Targeted to childrenij1Number of SKUsij , Brand strengthij ,Market sizeij1Experienceij , Corporate brandj1Firm sizeij95.Cobrandij is a latent variable that captures the truevalue of cobranding, and Cij is the previously definedcobranding dummy of the actual cobranding decision.Category cobranding in Equation (6b) is the instrumentalvariable defined previously.

From this model we derive ATET and test for itsstatistical significance. This is the third formal testof H1. Support for H1 is obtained if ATET is sig-nificantly positive. The appendix presents completedetails of the switching regression model, its deriva-tion, and its estimation.

Comparing the Models. There are strengths andweaknesses in both the propensity score models andthe switching regression model. The two propen-sity score models do not control for selection uponunobservables. The advantage is that they do notrequire an instrument for identification. In contrast,the switching regression model controls for unobserv-ables but ideally requires an instrument for identifica-tion. Ultimately, it is the weight of the evidence acrossthese three models that will determine whether H1 issupported in our study.

Testing H2–H4These three hypotheses are conditional on thecobranding decision having already being made and

thus do not require that we account for selectionor endogeneity. To test H2–H4, we estimate a ran-dom effects model of the determinants of CARs. Themodel is estimated on the subsample of cobrandingannouncements only:

CARij = �0 +�1Ingredient cobrandingij

+�2Composite cobrandingij +�3Consistencyij

+�4Exclusivityij +�5Innovativenessij

+�6Experienceij +�7Firm sizeij

+�8Corporate brandj +�9Brand strengthij

+�10Market sizeij +�11News outreachij

+�12Preannouncementij +�j + �ij1 (7)

where j represents the firm and i represents theannouncement. CARs are the short-term abnormalreturns earned by the primary brand’s parent. Theremaining variables are as previously defined. Wecontrol for firm-specific unobserved heterogeneity(�j ), and we also correct standard errors for het-eroscedasticity. We run a Hausman test to determinethe appropriateness of random effects.

ResultsWe begin by examining the descriptive statisticsof cobranded products and their parent firms. Theresults are presented in Tables 3(a)–3(d). Table 3(a)compares cobranded products to single-brandedproducts introduced by the same 61 primary brandparent firms. During the 11 years covered by our sam-ple, these firms introduced 1,543 cobranded productsand 21,130 single-branded products. The proportionof cobranded to total products is 6.8%. In contrast,these same firms introduced only 1,043 innovativeproducts (as rated by Product Launch Analytics), or4.6% of the total new products. These results sug-gest that, at least in this sample, cobranded productsare more prevalent than innovative products. This isan important finding on its own, given the dispro-portionate amount of academic research dedicated toinnovation in relation to cobranding. Other univari-ate statistics presented in Table 3(a) show little differ-ence between cobranded and single-branded productsin terms of innovativeness, importance of their par-ent brand in the firm’s portfolio, and average num-ber of SKUs per product. The major difference is thepercentage of products targeted to children: 31.95%for cobranded products versus only 6.45% for single-branded products. Cobranded products are also morelikely to be introduced in product categories withsmaller market sizes and, as expected, in categorieswhere cobranding is more prevalent.

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Table 3(b) summarizes key characteristics ofcobranding partnership announcements. Exclusivityis implicitly implied in 23% of cobranding agreements(first row) and is mentioned explicitly in only 3%of announcements (second row). The average consis-tency between the primary and secondary brands isfairly high, 4.80 on a scale of 1 to 7 (where a highernumber indicates higher consistency), in line withour expectations that firms are more likely to pursuecobranding partnerships with brands that provide agood fit with their own. Finally, 6% of cobrandingannouncements include at least one innovative prod-uct, consistent with the overall level of innovativeness

Table 3(a) Descriptive Product-Level Statistics

Cobranded Single branded

Count 1,543 21,130Innovative (%) 4080 4059Targeted to children (%) 31095 6045Average brand strength (%) 21095 26097Average size of the product category in 546009 713024

which they are introducedAverage number of SKUs per product 2055 2072Average number of products introduced 12044 6023

over the past year into the category inwhich the focal product is introduced

Note. Between 2000 and 2010, 22,673 total new products were introduced.

Table 3(b) Descriptive Announcement-Level Statistics

Mean SD Min Max

Exclusivity (implicit) 0.23 0.42 0 1Exclusivity (explicit) 0.03 0.17 0 1Consistency 4.80 0.65 2.76 6.24Innovativeness 0.06 0.24 0 1

Note. There are 316 announcements in total: 210 were classified as endorsement cobranding, 77 as ingredientcobranding, and 29 as composite cobranding.

Table 3(c) Descriptive Parent Firm-Level Statistics, Characteristics, and Associated Variables

Parent firm-level variable

Primary brand Secondary brand(61 publicly traded firms) (42 publicly traded firms)

Variable Mean SD Min Max Mean SD Min Max

Firm size 21,123 29,928 7.03 180,663 21,251 41,945 4.69 392,647($ million)

Experience 22.2785 29.38394 0 147

Table 3(d) Average CARs Around Cobranding Announcements

Model Coefficient SE

Market (%) 0089∗∗∗ 0014Fama–French (%) 0079∗∗∗ 0015

∗p < 0010; ∗∗p < 0005; ∗∗∗p < 0001.

in the sample. For Table 3(b) we also note the break-down of the various types of cobranded products. themost common type is endorsement cobranding: thereare 210 cases of endorsement cobranding in our sam-ple, compared with 77 cases of ingredient cobrandingand 29 cases of composite cobranding. This break-down is not surprising because endorsement cobrand-ing requires minimal commitment from the primarybrand parent, as it typically affects only the packageof a product it already sells in the marketplace.

Table 3(c) presents descriptive statistics of firm-level variables included in the analysis. Cobranding isused by firms acress the size spectrum. Further, firms’

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


cobranding experience ranges from using a cobrand-ing strategy for the first time to having used it onclose to 150 products over a period of five years.

Table 3(d) presents descriptive statistics of the(unadjusted) CARs computed around announcementsof cobranded products. These CARs are positive andsignificant both in the case of the market model(0.89%, p < 0001) and in the case of the Fama–Frenchmodel (0.79%, p < 0001). These results suggest that,on average, cobranded products create value for theshareholder. The tests associated with H1, describedbelow, will determine whether this value enhance-ment comes from the cobranding aspect per se orfrom the underlying new product announcement.

Test of Hypotheses

Results for H1—Basic Propensity Score Model.We begin with the basic propensity score model andextract propensity scores from Equation (4). Thesescores are used to form control samples of single-branded products announcements. The CARs of thesecontrol product announcements provide estimates ofthe counterfactual CARs required for testing H1.

Table 4 presents the estimated coefficients of Equa-tion (4). A stepwise probit model was used in theestimation process, and thus only the significantdeterminants of cobranding (p < 0001) are retained inthe final version.

After computing propensity scores for all prod-ucts, we use the six matching methods described inthe previous section to obtain six control samples of316 single-branded products each. These control sam-ples are drawn from the population of 37,212 newproducts introduced by the 61 firms in our sampleduring 1990–2010. Table 5(a) provides a formal defi-nition of the six matching methods.

Table 5(b) presents descriptive statistics of thecobranded sample and of the six control samples interms of the eight determinants of the cobrandingdecision from Equation (4). An important questionis whether control samples are sufficiently similar to

Table 4 Determinants of the Decision to Cobrand

Independent variable Coefficient SE

Experience 00005∗∗∗ 00000Brand strength 00333∗∗∗ 00059Market size −00001∗∗∗ 00000Corporate brand −00265∗∗∗ 00075Targeted to children 00751∗∗∗ 00033Firm size 00051∗∗∗ 00014Number of SKUs 00011∗∗∗ 00003Category cobranding 00041∗∗∗ 00002Constant −2027∗∗∗ 00134Wald �2 1,843.96

Note. The dependent variable is the cobranding dummy (n = 37,212).∗p < 0010; ∗∗p < 0005; ∗∗∗p < 0001.

the cobranded sample along these eight dimensions.We conduct a Hotelling test to determine whether thevector of means in the cobranded sample is equalto the vector or means in each of the six controlsamples. The results are presented at the bottom ofTable 5(b). In the case of control samples 1, 3, and 4,the Hotelling test does not reject the null hypothesisof equal vector means. These three matching methodsare likely to produce adequate matches that accountfor all observable characteristics of the selection pro-cess. For the remaining matching procedures (2, 5,and 6), the Hotelling test rejects the null, suggestingthat these methods are likely to produce imperfectmatches. However, the problem of imperfect matchesproblem is mitigated by using the pairwise-differenceregression from Equation (5).

We next check and update, through archivalsearches, the announcement dates for all single-branded control products. As we did with cobrandedproducts, we use the Product Alert date as theannouncement date if there are no other announce-ments or if that date precedes other announcements.We also check for confounding events on announce-ment dates. When a confounding event is identified,we select the next single-branded product announce-ment indicated by that particular matching method.

Having identified six control samples, we mea-sure the CARs corresponding to all control productannouncements. These are estimates of the counter-factual CARs required for testing H1. These CARs arethen subtracted from the cobranded product CARs,and the mean pairwise difference provides an esti-mate of ATET that accounts (only) for observableselection factors. The mean and median of these pair-wise differences are presented in Table 6(a). All meansare significantly positive (p < 0001) for all match-ing subsamples, for both the market model and theFama–French model. Only 1 of the 12 medians is notsignificantly positive (p = 0043). These results providestrong support for H1.

Pairwise-Difference Propensity Score Model. InTable 5(b) we show that in three of the six matchingmethods, cobranded products appear to be differentfrom their respective control products. To account forthese possible differences, we regress the pairwisedifferences in CARs on pairwise differences ondeterminants of CARs from Equation (5). For com-pleteness, we estimate this model for all six controlsamples. The intercepts from these regressions pro-vide estimates of ATET that account for observableselection factors, as well as for the possibility ofimperfect matches. These intercepts are presentedin Table 6(b). All values are significantly positive(p < 0001), supporting H1.

Switching Regression Model. We estimate next theswitching regression model of Equations (6a)–(6c) to

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Table 5(a) Description of Matching Procedures

Matching method Description

1 Announcement with the highest propensity score made by the same firm within five years of the cobrandingannouncement

2 Announcement with the closest propensity score made by the same firm within five years of the cobrandingannouncement

3 Announcement with the highest propensity score made by the same firm4 Announcement with the closest propensity score made by the same firm (smallest difference between the two

propensity scores)5 Announcement with the closest propensity score made by any firm in the sample within five years of the

cobranding announcement6 Announcement with the closest propensity score made by any firm in the sample (smallest difference between the

two propensity scores)

Table 5(b) Means of the Determinants of the Decision to Cobrand

Control sample (matching method)

Variable Cobranded sample 1 2 3 4 5 6

Experience 38011 49024 50000 37070 37043 60012 45044Brand strength 00307 00287 00269 00321 00269 00221 00226Market size 537057 343025 481029 330038 489047 509046 464097Corporate brand 00094 00084 00084 00094 00094 00051 00028Targeted to children 00478 00470 00386 00472 00383 00304 00361Log of firm size 9023 9049 9053 9003 9017 9076 9051Number of SKUs 2035 5069 3005 2083 2087 3003 2046Category cobranding 13012 9039 8046 12052 11027 8097 11081Hotelling F -statistica 1037 6024∗∗∗ 00756 1049 14023∗∗∗ 3014∗∗∗

aThe Hotelling F -statistic tests the null hypothesis that the vector of means are equal for the two groups, the cobranded sample and the sample of matchedsingle-branded products.

∗p < 0010; ∗∗p < 0005; ∗∗∗p < 0001.

Table 6(a) Tests of Differences in CARs Between Cobranded and Single-Branded Announcements Matched by Propensity Scores

Matching method

1 2 3 4 5 6

Market FF Market FF Market FF Market FF Market FF Market FF

Meana (%) 0070∗∗ 0053∗∗ 0084∗∗∗ 0074∗∗∗ 1062∗∗∗ 1035∗∗∗ 0080∗∗∗ 0066∗∗∗ 1010∗∗∗ 1002∗∗∗ 0089∗∗∗ 0087∗∗∗

400275 400265 400235 400225 400265 400255 400225 400225 400235 400235 400245 400255Medianb 0040∗ 0031 0058%∗∗∗ 0061%∗∗∗ 1012%∗∗∗ 0092%∗∗∗ 0034%∗∗∗ 0012%∗∗∗ 0086%∗∗∗ 0042%∗∗∗ 0054%∗∗∗ 0030%∗∗∗

400085 400225 400005 400005 400005 400005 400005 400005 400005 400005 400005 400005

Notes. We used the market and Fama–French (FF) models to compare the CARs to the announcement. Pairwise difference in CARs between cobranded andsingle-branded products were matched by propensity scores. Standard errors are shown in parentheses.

aWe used t-tests to determine the significance for means.bWe used the Wilcoxon sign-rank test to determine the significance for medians.∗p < 0010; ∗∗p < 0005; ∗∗∗p < 0001.

Table 6(b) Intercept of a Model Run on the Pairwise Difference in CARs Between Cobranded and Single-Branded Products Matched byPropensity Scores (Equation (5))

Matching method

1 2 3 4 5 6

Market FF Market FF Market FF Market FF Market FF Market FF

Intercept (%) 0083∗∗ 0069∗∗ 0076∗∗∗ 0072∗∗∗ 1093∗∗∗ 1064∗∗∗ 0084∗∗∗ 0065∗∗∗ 1038∗∗∗ 1034∗∗∗ 0092∗∗∗ 0089∗∗∗

400345 400325 400265 400255 400295 400295 400235 400235 400275 400275 400265 400275

Notes. We used the market and Fama–French (FF) models to compare the CARs to the announcement. Standard errors are shown in parentheses.∗p < 0010; ∗∗p < 0005; ∗∗∗p < 0001.

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Table 6(c) Average Effect of the Cobranding Decision on the CARsto Cobranded Product Announcements (ATET)

Model Coefficient SE

Market 1008∗∗∗ 0003Fama–French 0094∗∗∗ 0003

∗p < 0010; ∗∗p < 0005; ∗∗∗p < 0001.

account for the possibility that the selection decisionmight be based, at least in part, on characteristics thatare unobservable. We note that Equation (6b) is identi-cal to selection Equation (4), but it is presented here ina different format to highlight that Category cobrandingis used as an instrument in (6b) and is not used in theCAR equation (6a).7 Table 6(c) provides estimates ofATET obtained from the switching regression model.ATET is 1.08% for market-model CARs and 0.94% forFama–French CARs. Both values are significantly pos-itive (p < 0001), providing support for H1.

Taken together, the results in Tables 6(a)–6(c) pro-vide support for H1 across three different econometricmodels, each of which is based on different sets ofassumptions. The ATET associated with the cobrand-ing decision varies from 0.53% to 1.93%, dependingon the method. Averaging these ATET values withinthe method and then across methods, we concludethat the typical ATET associated with the cobrandingdecision is approximately 1.0%.

Results for H2–H4. We test H2–H4 using the ran-dom effect model of CARs described in Equation (7).The results are presented in Table 7. The top threerows show the coefficients of the focal indepen-dent variables that are directly linked to H2–H4. Forthe exclusivity variable, we use the implicit measure;results obtained with the explicit measure are dis-cussed in the robustness section. The subsequent rowsin Table 7 present the coefficients of the control vari-ables, and the last row presents the Wald �2 statisticfor model significance. Both models are significant atthe 1% level, and they include firm effects. Hausmantests were performed for both models; p-values ofmore than 0.10 indicate the appropriateness of firm-specific random effects.

We find support for all three hypotheses (H2–H4).Consistency is a positive and significant deter-minant of CARs (p < 0001), as are exclusivity(p < 0001) and innovativeness (p < 0005).8 Examin-ing the control variables, we find CARs are lower

7 The validity of Category cobranding as an instrument can beassessed using a test suggested by Sargan (see Verbeek 2008,p. 156). The test is used to reject the validity of a variable as aninstrument. We performed this test and were unable to reject thevalidity of Category cobranding as an instrument (p = 0052 for themarket model and p = 0081 for the Fama–French model).8 We also examine an alternate formulation of hypotheses H2–H4.Because H2–H4 are conditional on the cobranding decision having

for preannouncements (as opposed to actual productlaunch announcements) (p < 0005). Marginally posi-tive effects on CARs are observed for the size of thecobranded product category (p < 0010 for the mar-ket model), and a positive effect of brand strengthon CARS (p < 0005) is obtained in the Fama–Frenchmodel. The other variables do not significantly affectCARs, with the exception of the ingredient branddummy, which has a marginally significant and neg-ative coefficient in the Fama–French model only. Alarge sample is needed to draw robust conclusionsabout differences between the types of cobrandingproducts.

Robustness TestsWe perform two additional tests to assess the robust-ness of our results. First, we retest all hypothesesusing market-adjusted CARs. The results are similarto those reported in Tables 6 and 7. Second, we retestH2–H4 using an explicit measure of exclusivity. Sinceonly 9 of the 316 announcements make explicit refer-ence to the exclusivity provision, explicit exclusivityis a highly skewed and potentially noisy variable; itseffect on CARs is no longer significant, although thecoefficient remains positive.

Additional AnalysisNew Markets or Product Categories. We seek to

understand whether CARs are higher for cobrandedproducts that open a new market or a new productcategory. Using the product category codes providedby Product Launch Analytics, we identify cobrandedproducts introduced in categories where the primary

already been made, we do not need to control for selection or endo-geneity in testing these hypotheses. However, an alternate way ofrestating H2–H4 would be in terms of attributes that maximizethe value-added of only the cobranding aspect (which is the pair-wise difference between the CARs of the cobranded product andthe CARs of a counterfactual single-branded product). It wouldbe interesting to determine whether the same three attributes thatmaximize CARs in H2–H4 are also attributes that maximize thevalue-added of (only) the cobranding aspect of these new prod-ucts. To this end, we estimate Equation (7) with a different depen-dent variable: the ATET obtained from the propensity score model(with CARs computed from the Fama–French and market mod-els). Among the six matching methods described in Table 5(a), weselect Method 4 to run this test. This method selects as controlproducts the single-branded products introduced by the same firmand with the closest propensity score as the cobranded products.The method produces a high-quality match, as indicated by thenonsignificant Hotelling F -statistic in Table 5(b). The results (notshown) are consistent with those obtained in Table 7. Using pair-wise differences in CARs computed from the Fama–French andmarket models, the coefficients on all three variables of interest(consistency, exclusivity, and innovativeness) are positive and signifi-cant at the 5% level. Overall, we conclude that the three attributesthat maximize shareholder wealth conditional on the cobrandingdecision are also attributes that maximize the value-added of thecobranding aspect of these new products.

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Table 7 Determinants of Cumulative Abnormal Returns

CAR

Market Fama–French

Variable Coefficient SE Coefficient SE

Consistency 00009∗∗∗ 00002 00010∗∗∗ 00002Exclusivity 00008∗∗∗ 00003 00009∗∗∗ 00003Innovativeness 00009∗∗ 00005 00011∗∗∗ 00004Ingredient −00005 00004 −00006 00003

cobrandingComposite 00007 00006 00004 00005

cobrandingPreannouncement −00006∗∗ 00004 −00007∗∗ 00004News outreach 00001 00001 001 00001Experience −8002 × 10−6 9.19 × 10−5 5.38 × 10−5 7.45 × 10−5

Brand strength 000004 00002 00004∗∗ 00002Market size 5.59 × 10−6∗ 3.96 × 10−6 3.86 × 10−6 3.40 × 10−6

Corporate brand 000001 0001 −00002 00005Firm size −00001 00001 −00001 00001Constant −00029 00012 −00034 00012Wald �2 74.33 73.10

Note. The dependent variable is CAR (n = 316).∗p < 0010; ∗∗p < 0005; ∗∗∗p < 0001.

brand firm was already present. We ran an alter-nate model where we added a corresponding dummyvariable to Equation (7), but its effect on CARs isinsignificant.

Labeling vs. Innovation Effects. Cobranding part-nerships could affect consumer perceptions thougheither a labeling effect or an innovation effect. Thelabeling effect occurs when consumer perceptions areaffected only by the presence of the secondary brandon the product’s label, even in the absence of achange to that product’s formulation. Endorsement-type cobranded products fall into this category. Theinnovation effect occurs when consumer perceptionsare influenced by the product’s new features. Weexpect the innovation effect to be prevalent for ingre-dient and composite cobranding agreements.

We also seek to understand whether the label-ing effect is sufficient to generate a positive ATETor whether an innovation effect is also necessary.We repeat the analysis in Table 6 for three separatesubsamples: endorsement, ingredient, and compos-ite cobranding. The average ATET values are signif-icantly positive in each subsample, suggesting thateach type of cobranding agreement adds value tothe underlying product. The positive ATET valuesfor endorsement cobranding indicate that the label-ing effect is valuable even in the absence of changesto product formulation. We did not find any signifi-cant differences between the three types of cobrand-ing strategies. Composite cobranding appears to yieldmarginally higher ATET values, but the sample is toosmall to conclusively determine that this is indeed asuperior strategy.

We next turn to H2–H4 to see whether they con-tinue to hold when only the labeling effect of cobrand-ing is present. We reestimate Equation (7) for twoseparate subsamples: (i) endorsement cobranding and(ii) combined ingredient and composite cobranding.The results in each subsample are substantively simi-lar to those obtained with the full sample. This indi-cates that the effects of consistency, exclusivity, andinnovativeness documented in Table 7 are not relatedto the labeling effect of cobranding.

Summary and DiscussionThe reader interested in learning about cobrandingcan rely on a substantial stream of research that doc-uments how consumers react to brand partnerships.This literature has identified a set of circumstanceswhen cobranding can strengthen perceptions of thepartner brands, but it also warns of potential brandequity damage to one or both partners if the brandalliance is not successful. We extend this literatureby focusing on the relation between cobranding andshareholder value.

We find evidence consistent with the hypoth-esis that the stock price increase surroundingcobranded product announcements is attributed tothe cobranding feature of the product, rather than tothe product itself. In the consumer packaged goodsindustry, where new product introductions are mostlyincremental, cobranding appears to be a valuable dif-ferentiating strategy. However, not all products bene-fit from cobranding, and we offer empirical evidenceon a set of characteristics shared by cobranded prod-ucts in this industry over the past decade. We alsofind that consistency between the two cobrands has apositive and significant effect on the market reactionto the introduction of cobranded products. Likewise,investors appear to place more value on partnershipswhere the secondary brand has agreed to an exclusivecobranding agreement with the primary brand.

Our results have several implications for both the-ory and practice. We discuss them below.

1. Cobranding can substantively contribute to firmvalue. Prior research has found that incremental inno-vations in consumer packaged goods contribute tonormal profits but are not a source of economic rents(Sorescu and Spanjol 2008). In such a mature andcompetitive industry, one where incremental, ratherthan radical, innovations are prevalent, it is difficultto elicit strong investor enthusiasm with new productintroductions. Yet our results show that by cobrand-ing an appropriate subset of new products, man-agers can earn significant economic rents. We findthe cobranding aspect alone could increase firm valueby approximately 1.0% for products that benefit fromcobranding. We also identify a set of characteristics

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


shared by cobranded products in this industry. Takentogether, our results provide managerial guidelineson the determinants and possible financial gains ofcobranding.

2. New products need not be technological innovationsto create shareholder wealth3 cobranding can be effectivelyleveraged to significantly increase the value-added of newproducts. The value-added of cobranding (approxi-mately 1.0%) is a remarkable finding given that mostof the products in our sample are not technologicallyinnovative. Even endorsement-type agreements thatleverage the labeling effect of cobranding appear toadd value. Given how difficult it is to achieve innova-tion through product formulation in a mature indus-try such as consumer packaged goods, cobrandingprovides managers with an alternative path to cre-ate shareholder value through brand-driven productdifferentiation.

3. Exclusivity is valued. The value of cobrandedproducts is higher in cases where the secondary brandpartner has not previously engaged in cobrandingagreements with other primary brand firms. Weacknowledge, however, that this could be a simplemanifestation of pioneering advantage. For instance,Cadbury Schweppes’s Diet Rite, the first beverageto incorporate Splenda sweetener, saw a signifi-cant boost in sales after its reformulation with theSplenda brand, but the same was not true for otherentrants that subsequently partnered with Splenda(Esfahani 2005).

4. Consistent brand associations might provide an eas-ier path to successful new product introductions whencompared to exclusivity or innovativeness. The set of suit-able secondary brand partners willing to engage inexclusive cobranding partnerships could be too small,and such partners could require significant profitsharing. Likewise, creating highly innovative prod-ucts could be too elusive a goal. In contrast, selectinga partner with a consistent brand image might wellbe the easiest way to design new products that exciteinvestors and consumers alike.

5. Not all products benefit from cobranding. Althoughthe average cobranding announcement is associatedwith an increase in firm value of approximately1.0%, we expect that only a relatively small propor-tion of products benefit from cobranding. The resultsin Table 4 show that companies operating underhouse-of-brand strategies are more likely to introducecobranded products. However, only 12 of the 61 pub-licly traded firms in our sample fall into this cate-gory. In addition, cobranded products tend to use oneof the parent’s more prominent brands, and cobrand-ing is more prevalent among products targeted tochildren. Firms are more likely to cobrand productswith multiple SKUs and products that carry one oftheir flagship brands. And firms are also more likely

to cobrand in categories where cobranding has beenmore prevalent in the past. Because only a small frac-tion of new products are likely to fulfill these con-ditions, we expect cobranding to be a relatively rarestrategy.

Limitations and Future Research. Our study is,to our knowledge, the first attempt to investigatethe stock market reaction to the introduction of newcobranded products. Similar to many papers thatopen a new research area, ours has some limita-tions that could serve as topics for future research.First, our data set contains limited information aboutthe revenue model behind cobranding agreements. Incases where a licensing fee is paid, the fee is typi-cally not reported in the cobranding announcement.Yet the market reaction to cobranding announcementscould be affected by the magnitude of this licens-ing fee. Second, we used a backward-looking mea-sure of consistency between the two partner brandimages, because, to our knowledge, data on contem-poraneous brand consistency are not available forour sample. Third, additional factors (e.g., competi-tor reactions) could moderate the relation betweenstock returns and cobranded product introductions.Finally, our stock return metrics limit the cobrandingsample to publicly traded firms. The sample couldbe extended to privately held corporations by usingaccounting measures of performance such as sales orreturn on investment when such measures are avail-able at the product level.

Future research could also examine the length andsuccess of cobranding relationships and investigatethe extent to which the initial market reaction cananticipate the longevity and success of cobrandedproducts. Finally, cobranding research from otherindustries, particularly services where cobranding isincreasingly frequent, could explore additional impor-tant dimensions of cobranding that are unique to eachindustry.

AcknowledgmentsThe authors thank participants of MSI Marketing StrategyMeets Wall Street II conference at Boston University andparticipants of the Marketing Science conference at Rice Uni-versity for helpful suggestions.

AppendixWe show how to estimate the ATET when measuringreturns to announcements of cobranded products and whenthe cobranding decision is based, in part, on unobserv-able factors. We closely follow Verbeek’s (2008, pp. 253–256)switching regression model that accounts for selection uponunobservables. In our context ATET represents how mucha cobranded product gains from the cobranding feature.

Let CAR0 be the cumulative abnormal return to theannouncement of a single-branded product. Let R∗ be theincremental CAR that can be attributed to the cobranding

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


feature of a product, i.e., the added value of cobrandingto the CAR of the announcement. Then, the CAR to theannouncement of a cobranded product i is

CAR1i = CAR0i +R∗

i 1 (8)

where i denotes the announcement. For simplicity of expo-sition, we omit j , the firm subscript. By construction, CAR1and CAR0 pertain to the same product under two differ-ent scenarios: one in which the product is cobranded andanother in which it is single branded. In reality, only one ofthese scenarios is observed. The other is the counterfactualscenario that will be estimated by our model.

The magnitude of the CARs to a new product announce-ment is determined by industry-, firm-, and product-levelfactors. For simplicity of exposition, in our models belowwe only include one observable determinant of the CARs,denoted X, and one unobservable determinant of the CARs,denoted U ; however, the derivation of the ATET is easilygeneralizable to multiple dimensions of X and U . Thus,the CARs to single-branded products, and the incrementalCARs attributable to cobranding, R∗, can be written as

CAR0i = �0 +�0Xi + k0Ui +�0i1 (9)

R∗

i = �+�Xi + kUi +�i0 (10)

From (8), (9), and (10), we derive CAR1 to be

CAR1i = �0 +�+ 4�0 +�5Xi + 4k0 + k5Ui +�0i +�i0 (11)

When considering cobranding a product i, managersneed to estimate R∗

i , the added value of cobranding for thatparticular product. If they could perfectly assess R∗

i with-out error (i.e., with �i = 0), they would always make thecorrect branding decision; in that case, ATET could not beestimated. Perfect knowledge of R∗

i is, however, an unre-alistic assumption. More realistically, managers will makecobranding decisions using heuristics derived from theirexperience, which determine what they perceive to be thevalue of cobranding a specific product. The error term, �i,captures these heuristics.

It follows that Ci, the decision to cobrand product i, isCi = 1 if R∗ > 0 and Ci = 0 if R∗ ≤ 0. We note that R∗ isa latent variable and U is unobservable. Therefore, theobservable part of the decision to cobrand can be mod-eled as

Ci = �0 + �1Xi + �′

i0 (12)

The ATET is defined as

ATET ≡ E8CAR1i − CAR0i �Xi1Ui1Ci = 190

More specifically, ATET is the difference between theobserved CARs of cobranded products (CAR1i) and thecounterfactual, single-branded CARs of these same prod-ucts (CAR0i), contingent on these products being cobranded(Ci = 1). If variable U were observable, ATET could be esti-mated from Equations (9) and (11). However, because U isunobservable, we can only estimate reduced forms of (9)and (11) as follows:

CAR0i = �0 +�0Xi + �0i0 (13)

The reduced form of (11), CAR1i = �0 +�+ (�0 +�5Xi + �1i1can then be rewritten as

CAR1i = �1 +�1Xi + �1i1 (14)

where �0i = k0Ui +�0i and �1i = 4k0 + k5Ui +�0i +�i.Following Verbeek (2008, p. 254), the average treatment

effect in this case is given by

ATE ≡ E8CAR1i − CAR0i �Xi1Ci90 (15)

To obtain the ATET, we simply condition Equation (15)on Ci = 1:

ATET ≡ E8CAR1i − CAR0i �Xi1Ci = 190 (16)

We obtain estimates of E8CAR1i � Xi1Ci9 and E8CAR0i �

Xi1Ci9 from (13) and (14):

E4CAR0i �Xi1Ci5= �0 + �0X +E4�0i �Xi1Ci51 (17)

E4CAR1i �Xi1Ci5= �1 + �1X +E4�1i �Xi1Ci50 (18)

Verbeek (2008, p. 256), citing Vella and Verbeek (1999),provided formulas for the conditional expectations of theerror terms �0i and �1i:

E8�0i �Xi1Ci9= �02�i4Ci51 (19)

E8�1i �Xi1Ci9= �12�i4Ci51 (20)

where• Ci are the fitted values from the probit model

in (12), and

• �i4Ci5=Ci −ê4Ci5

ê4Ci541 −ê4Ci55�4Ci50 (21)

The term �i4Ci5 is the generalized residual from the probitmodel. For Ci = 1, it also corresponds to Heckman’s lambda.

• Finally, �02 and �12 are the covariances of �0i and �1i,respectively, with �′

i from Equation (12). If the unobservablequantity U is a determinant of both the CARs and the deci-sion to cobrand, these covariances are different from zero,and we must account for them in the estimation of ATET tocorrect the endogeneity caused by U .

Verbeek explains that for Equations (13) and (14) to beidentified, it is desirable to find an instrumental variable, Z,which affects the decision to cobrand but does to affect thebenefits of cobranding.9 Thus, we estimate Equation (12) byadding an instrument, Z, to the right-hand side as follows:

Ci = �0 + �1Xi + �Zi + �′

i0 (22)

To compute the ATET, we substitute (17)–(21) into (16),conditioned upon Ci = 1, and obtain

ATET ≡ E8CAR1i �Xi1Ci = 19−E8CAR0i �Xi1Ci = 191

ATET = 6�1 + �1Xi +�12�4Ci5− �0

− �0Xi −�02�4Ci5 �Ci = 170

(23)

9 In the absence of an instrumental variable, identification rests onthe nonlinearity of �i4Ci5; the model can only be identified if thisterm is sufficiently nonlinear to prevent multicollinearity.

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


If we define � = �1 − �0 and � = �1 − �0, then we canwrite (23) as

ATET = 6�+ �Xi + 4�12 − �025�4Ci5 �Ci = 170 (24)

To compute ATET from (24), we need estimates of �, �,�12, �02, and �i4Ci5. Estimates of Ci are obtained from (12),and estimates of �i4Ci5 are obtained from (21).

To obtain estimates of �, �, �12, and �02, we return to thedeterminants of the CARs and combine Equations (13) and(14) into a single equation for all products (single-brandedand cobranded) as follows:

CARi = �0 +�0Xi + �0i +Ci

[

4�1 −�05+ 4�1 −�05Xi

+ �1i − �0i

]

0 (25)

Taking expectations in (25), we obtain

E4CARi �X1Ci5 = �0 + �0X + �Ci + �CiXi + �02�4Ci5

+Ci�12�4Ci5−Ci�02�4Ci50

Rearranging the terms, we obtain

E4CARi �X1Ci5 = �0 + �0X + �Ci + �CiXi + �12Ci�4Ci5

+ �0241 −Ci5�4Ci50 (26)

From (26), it follows that the estimates of coefficients �,�, �12, and �02 needed to compute (24) can be obtainedfrom the following regression ran over the sample of single-branded and cobranded products:

CARi = �0 +�1Xi +�2Ci +�3CiXi +�4Ci�i4Ci5

+�541 −Ci5�i4Ci5 + �′′

i 0 (27)

Comparing (26) with (27), we see that the coefficients ofinterest for estimating ATET (�1 �1 �121 �025 are identical tocoefficients �2, �3, �4, and �5 in Equation (27). These coef-ficients can then be used in Equation (24) to complete theestimation process of ATET.

ReferencesAaker DA (2004) Brand Portfolio Strategy: Creating Relevance, Differ-

entiation, Energy, Leverage, and Clarity (Free Press, Glencoe, IL).Aaker DA, Keller KL (1990) Consumer evaluations of brand

extensions. J. Marketing 54(1):27–41.Aghion P, Bolton P (1987) Contracts as a barrier to entry. Amer.

Econom. Rev. 77(3):388–401.Batra R, Lenk P, Wedel M (2010) Brand extension strategy plan-

ning: Empirical estimation of brand-category personality fitand atypicality. J. Marketing Res. 47(2):335–347.

Blackett T, Boad B (1999) Co-branding: The Science of Alliance(St. Martin’s Press, New York).

Boyd DE, Chandy R, Cunha M Jr (2010) When do chief market-ing officers impact firm value? A customer power explanation.J. Marketing Res. 47(6):1162–1176.

Bronnenberg BJ, Dubé J-P, Mela C (2010) Do digital video recordersinfluence sales? J. Marketing Res. 47(6):998–1010.

Brown S, Warner JB (1985) Using daily stock returns: The case ofevent studies. J. Financial Econom. 14(1):3–31.

Bucklin LP, Sengupta S (1993) Organizing successful comarketingalliances. J. Marketing 57(4):32–46.

Carhart MM (1997) On persistence in mutual fund performance.J. Finance 52(1):57–82.

Chaney PK, Devinney TM, Winer RS (1991) The impact of newproduct introductions on the market value of firms. J. Bus.64(4):573–610.

Chen Y, Liu Y, Zhang J (2012) When do third-party productreviews affect firm value and what can firms do? The caseof media critics and professional movie reviews. J. Marketing76(2):116–134.

Das S, Sen PK, Sengupta S (1998) Impact of strategic alliances onfirm valuation. Acad. Management J. 41(1):27–41.

Desai KK, Keller KL (2002) The effects of ingredient brandingstrategies on host brand extendibility. J. Marketing 66(1):73–93.

Esfahani E (2005) Finding the sweet spot. Business 2.0 6(10):49–51.

Fama EF, French KR (1993) Common risk factors in the returns onstocks and bonds. J. Financial Econom. 33(1):3–56.

Gaeth GJ, Levin IP, Chakraborty G, Levin AM (1990) Consumerevaluation of multi-product bundles: An information integra-tion analysis. Marketing Lett. 2(1):47–57.

Gallagher J (2007) Boosting brands. Supermarket News 55(12):39–41.

Geyskens I, Gielens K, Dekimpe MG (2002) The market valuationof Internet channel additions. J. Marketing 66(2):102–119.

Grocery Manufacturers Association (2013) About GMA. AccessedApril 5, 2013, http://www.gmaonline.org/about/.

Helmig B, Huber J, Leeflang PSH (2007) Explaining behaviouralintentions toward co-branded products. J. Marketing Manage-ment 23(3/4):285–304.

Helmig B, Huber J, Leeflang PSH (2008) Co-branding: The state ofthe art. Schmalenbach Bus. Rev. 60(4):359–377.

Henard DH, Szymanski DM (2001) Why some new products aremore successful than others. J. Marketing Res. 38(3):362–375.

Huang Q, Nijs V, Hansen K, Anderson E (2012) Wal-Mart’s impacton supplier profits. J. Marketing Res. 49(2):131–143.

Kalaignanam K, Shankar V, Varadarajan R (2007) Asymmetric newproduct development alliances: Win–win or win-lose partner-ships? Management Sci. 53(3):357–374.

Keller KL (1993) Conceptualizing, measuring and managingcustomer-based brand equity. J. Marketing 57(1):1–22.

Keller KL, Aaker DA (1992) The effects of sequential introductionof brand extensions. J. Marketing Res. 29(1):35–50.

Koh J, Venkatraman N (1991) Joint venture formations and stockmarket reactions: An assessment in the information technologysector. Acad. Management J. 34(4):869–892.

Krattenmaker TG, Salop SC (1986) Anticompetitive exclusion: Rais-ing rivals’ costs by gaining power over price. Yale Law J.96(2):209–214.

Lane V, Jacobson R (1995) Stock market reactions to brand extensionannouncements: The effects of brand attitude and familiarity.J. Marketing 59(1):63–77.

Levin AM, Davis JC, Levin IP (1996) Theoretical and empiricallinkages between consumers’ responses to different brandingstrategies. Adv. Consumer Res. 23(1):296–300.

Loken B, Roedder John D (1993) Diluting brand beliefs: Whendo brand extensions have a negative impact? J. Marketing57(7):71–84.

Park CW, Jaworski BJ, MacInnis DJ (1986) Strategic brand concept-image management. J. Marketing 50(4):135–145.

Park CW, Jun SY, Shocker AD (1996) Composite brand alliances:An investigation of extension and feedback effects. J. MarketingRes. 33(4):453–466.

Park CW, Milberg S, Lawson R (1991) Evaluation of brand exten-sions: The role of product feature similarity and brand conceptconsistency. J. Consumer Res. 18(5):185–193.

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Rao AR, Qu L, Ruekert RW (1999) Signaling unobservable productquality through a brand ally. J. Marketing Res. 36(2):258–268.

Rosenbaum PR, Rubin DB (1983) The central role of the propen-sity score in observational studies for causal effects. Biometrika70(1):41–55.

Rosenbaum PR, Rubin DB (1984) Reducing bias in observationalstudies using subclassification on the propensity score. J. Amer.Statist. Assoc. 79(387):516–524.

Schriffen J, McLean B, Kim S (2013) Surprising story behindthe Doritos Locos Taco. Yahoo! News (April 15), http://news.yahoo.com/blogs/newsmakers/surprising-story-behind-doritos-locos-taco-133112115.html.

Simonin BL, Ruth JA (1998) Is a company known by the companyit keeps? Assessing the spillover effects of brand alliances onconsumer brand attitudes. J. Marketing Res. 35(1):30–42.

Sood A, Tellis GJ (2009) Do innovations really pay off? Total stockmarket returns to innovation. Marketing Sci. 28(3):442–456.

Sorescu A (2012) Innovation and the market value of firms.Ganesan S, ed. Handbook of Marketing and Finance (EdwardElgar, Cheltenham, UK), 129–154.

Sorescu A, Spanjol J (2008) Innovation’s effect on firm value andrisk: Insights from consumer packaged goods. J. Marketing72(2):114–132.

Spethmann B, Benezra K (1994) Co-brand or be damned. Brandweek35(45):20–25.

Srinivasan S, Pauwels K, Silva-Risso J, Hanssens D (2009) Prod-uct innovations, advertising and stock returns. J. Marketing73(1):24–43.

Swaminathan V, Moorman C (2009) Marketing alliances, firm net-works and firm value creation. J. Marketing 73(5):52–69.

Thompson S (1998) Brand buddies—Co-branding meal solutions.Brandweek 39(8):22–30.

Uggla H, Asberg P (2010) A psycho-semiotic research agenda forstrategic brand alliances. J. Brand Management 7(1/2):92–104.

Varadarajan PR (1986) Horizontal cooperative sales promotion:A framework for classification and additional perspectives.J. Marketing 50(2):61–73.

Vella F, Verbeek M (1999) Estimating and interpreting modelswith endogenous treatment effects. J. Bus. Econom. Statist.17(4):473–478.

Venkatesh R, Mahajan V (1997) Products with branded components:An approach for premium pricing and partner selection. Mar-keting Sci. 16(2):146–165.

Verbeek M (2008) A Guide to Modern Econometrics, 3rd ed. (JohnWiley & Sons, Hoboken, NJ).

Verrier R (2001) Disney, Kellogg sign marketing alliance. Los AngelesTimes (September 6) http://articles.latimes.com/2001/sep/06/business/fi-42641.

Walchli SB (2007) The effects of between-partner congruity on con-sumer evaluation of co-branded products. Psych. Marketing24(11):947–973.

Williamson O (1983) Credible commitments: Using hostages to sup-port exchange. Amer. Econom. Rev. 73(4):519–540.

Yadav MS (1994) How buyers evaluate product bundles: A modelof anchoring and adjustment. J. Consumer Res. 21(2):342–353.

Dow

nloa

ded

from

info

rms.

org

by [

152.

3.15

3.22

5] o

n 23

Apr

il 20

15, a

t 16:

50 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.

wedded bliss or tainted love? stock market reactions to ...moorman/marketing... · alina sorescu...

Documents