Products vs. Advertising: Media Competition and the. Relative Source of Firm Profits

Transcription

1 Products vs. Advertising: Media Copetition and the Relative Source of Fir Profits David Godes, Elie Ofek and Miklos Sarvary February 2003 The authors would like to thank Dina Mayzlin, and participants at the North East Marketing Consortiu (Dartouth, 2002) and the Marketing in Israel Seinar (Tel-Aviv, 2002) for valuable suggestions. paper has also benefited fro coents at the Marketing Science conference in Edonton, Canada (2002). Preliinary draft, coents welcoe. David Godes and Elie Ofek are Assistant Professors of Business Adinistration at Harvard Business School, Soldiers Field, Boston MA Miklos Sarvary is Associate Professor of Marketing, INSEAD, Bd. de Constance 77305, Fontainebleau, France The

2 Products vs. Advertising: Media Copetition and the Relative Source of Fir Profits Abstract In this paper we ask when will a fir earn ore profits fro selling the attention of its custoers to advertisers than fro selling the underlying product itself. In other words, when will a fir becoe an advertising ediu? We investigate this decision as a function of the intensity and nature of copetition. We show that regardless of inherent product value to custoers, when the fir faces high within-industry copetition it will always earn ore profits fro the product arket; iplying that firs cannot advertise their way out of intense copetition. However, for products of oderate inherent value, we find that the product odel is ore attractive when copetition is at either extree (very high or low) but the advertising odel is ore attractive when copetition is in the iddle range. This results in an inverse-u pattern for relative source of profits as a function of within-industry copetition. We also look at the level of copetition between-edia and identify conditions for fir profits in one industry to increase as a result of heightened copetition with another industry. Moreover, we show that as two edia are ore substitutable (hence copeting ore head-to-head for advertising capaigns) their source of profits will diverge. In addition, the paper considers the ipact of the disutility created by advertising for the product consuer. Interestingly, we find that low levels of consuer disutility ay actually increase the proportion of profits fro advertising as copared to products. Keywords: Media, Circulation Industries, Business Model Selection, Copetitive Strategy, Gae Theory

3 1 Introduction The (Phoenix s) decision to give away copies... cae partly in response to the hordes of Internet sites giving inforation away for free. It hopes free copies will boost its readership, and thus its advertising revenue. - The Boston Herald on the decision to eliinate the $1.50 newsstand price for the Boston Phoenix, an alternative newsweekly (2002). We have to condition the (web users) otherwise... They have to either pay with their attention or pay with their pocket book. - Patrick Hurley, Senior Vice President of business operations, salon.co (2001). By their very nature, soe products 1 attract the attention of their custoers during use. Viewers of Monday Night Football spend about three hours every week watching and listening to ABC television s production. pages of the newspaper. To consue the content of the Boston Globe, readers ust look at the Visitors to Broadway theaters spend two hours sitting in a dark roo, paying attention to little else but the usic, dialog and storyline behind the play being produced on stage. In each of these cases, the product cannot be used without attracting the awareness and cognitive focus of the custoer. Firs that produce such attention-generating products have the option of participating in the advertising arket, which is essentially a arket for the attention of consuers. These firs ust decide to what extent their profits will coe principally fro advertising as opposed to product sales. The observed answer to this question differs greatly across industries. Television and radio derive effectively all of their revenue fro the advertising arket. At the other extree, live theater has typically not participated directly in the advertising arket. In between, newspapers earn on average 80% of revenues fro advertising while this nuber in agazines has historically ranged fro 10% to 70%. In this paper, we investigate the forces behind this decision on source of profitability and offer a theory for when and to what extent a fir will becoe an advertising ediu. Interestingly, recent years have seen significant flux regarding this decision. 1 Throughout this paper, we ll use the ter product to refer to both products and services. The U.S. ovie 1

4 theater has traditionally been an advertising-free zone. However, this began to change in the id-90 s as ore and ore theaters ran specially-produced ads before the feature presentation. As of 2001, nearly 2/3 of the 37,000 screens had soe for of advertising on the. 2 This nuber does not include the rise in product placeents in the fils theselves. The econoics of the agazine business have changed draatically as well. In the past two decades, we have seen a general shift fro subscription-driven odels to ore advertising-driven odels. It is particularly interesting to copare the trends in two edia, alternative newsweeklies and the Web. The best-known exaples of the forer are probably the Village Voice in New York and the Boston Phoenix. When the Phoenix decided to give its papers away for free in January 2001, it was the last holdout aong the ajor newsweeklies. The Voice had eliinated its newsstand price several years earlier. As the quote above suggests, one significant otivating factor for this decision was the increased copetition these papers were facing fro free sources. On the other hand, several dot-cos, particularly those providing content, present a very different exaple. Many such firs thestreet.co and salon.co, for exaple began by giving the product away for free in hopes of earning advertising revenues. In fact, soe copanies AllAdvantage.co and iwon.co, for exaple even paid custoers for their attention. However, we have recently witnessed a shift toward pay-for-content odels. According to a 2002 Forrester Research report, 75% of Internet executives surveyed felt that the proportion of their revenues coing fro subscription fees would increase within two years. About half of the felt that they would earn ore than 10% of their revenues fro subscribers within that tie frae. Yahoo! alone is expected to earn over 40% of its revenues by year-end 2002 fro non-advertising sources, as copared with 23% the previous year. 3 Thus, on the one hand we re seeing soe newspapers shifting their focus toward becoing ore advertising-driven, while on the other hand, we re seeing dot-cos shifting their focus away fro advertising revenues toward becoing ore product-driven. We would expect that the optial proportion of a fir s profits coing fro advertising as opposed to products should depend in part on the relative value that the product provides to 2 Source: Catch the when they re attentive, settled in their seats and eating popcorn at the ovies. by Karen J. Bannan. New York Ties, October 8, 2001, Section C; p Source: Can Yahoo! Make E Pay?, Business Week, Septeber 9, pp See also Profits at Last-Good News fro the Internet in the Econoist Dec. 21, 2002 pp

5 custoers as copared with the value that the custoers attention provides to advertisers. For exaple, we ight assue that a Broadway production would offer ore value to a viewer than a 60-inute television draa. Since custoers ight be willing to pay ore for the Broadway production, we d expect that, all else equal, theater owners would be less inclined to becoe an advertising ediu. Another iportant diension of the proble is the role of copetition. Copetition will liit the fir s ability to earn profits in both the product and advertising arket. However, the nature of copetition in these two arkets is not the sae. While product arkets are relatively independent fro one another, and hence face direct copetition ainly fro other firs within their respective industries, there is substantial cross-industry copetition for advertising. That is, while Tie agazine doesn t directly copete with NBC for readers, it certainly does so for advertisers. 4 Thus, copetition ay have a differentipactonproductarketsand advertising arkets, even though the sae firs copete in both. Note also that the relationship between the advertising arket and product arket exhibits an asyetric dependence. Selling a custoer s attention to an advertiser is contingent on first selling a product to that custoer, but the reverse is not true. 5 This relationship has two principal econoic iplications which we refer to as subsidization and output-doinance. These forces work in opposite directions and drive the core results in this paper. On the one hand, the product business is shown to subsidize the advertising business in the sense that the fir optially produces ore products than it would if it were not able to sell advertising space. Increasing product sales to custoers to benefit fro advertising, of course, depresses product profits. The iplication of this subsidization is that, as copetition increases and the fir scales back product output, the subsidy declines. This has a (partial) positive ipact on product profits and a negative ipact on advertising profits. On the other hand, it is clear that if prices fall by $1 in each arket, it will generally have a bigger negative ipact on the product 4 While there could be soe interdependence in the product arket (for exaple, viewing uch ore television could potentially take away fro tie to read agazines), any dollar spent in one ediu is directly taken fro another in the edia plan/buy given an advertiser s budget (see, for exaple, O Guinn et al. (2000)). 5 Our characterization of firs that sell to consuers in a product arket and then sell their attention to advertisers, is consistent with the definition of a circulation industry by Chaudhri (1998). It is also siilar to the characterization of a cross-arket network effect by Chen and Xie (2002). 3

6 arket than on the advertising arket since the output in the forer is always doinant. This stes fro the fact that the nuber of advertising ipressions can never be high without high product quantities, though product quantities are not liited by advertising ipressions. 6 So, as copetition increases total industry production increases and prices fall, and this is likely to hurt the product arket ore than the advertising arket given that output is high in the forer to begin with. We show that the relative ipact of copetition on advertising profits and product profits is driven priarily by the interplay between these two forces, oderated by the inherent value of the product to custoers. Not surprisingly, when the inherent value of the product is high, the fir earns ore profits fro product sales than fro advertising. In contrast, when this inherent product value is very low, advertising profits are initially higher than product profits. As copetitive intensity increases, however, the fir earns ore profits fro the product arket than fro the advertising arket even for products with arbitrarily low inherent value. The reason for this is that the effect of less subsidization causes ad profits to decline faster than product profits. The decline in product output has a negative effect on ad profits but a arginally positive effect on product profits, as the fir oves closer to the optial output level it would choose if it didn t overproduce to benefit fro the ad arket. When the inherent value of the product is at oderate levels, we get the intriguing result that there is an inverse-u shaped relationship between the copetitiveness of an industry and the difference between equilibriu profits that coe fro advertising and product sales. Hence, for these products, it is only at interediate levels of copetition that we are likely to observe the eergence of true advertising edia, where advertising represents the predoinant source of profits. This pattern eerges because at low levels of copetition, firs ake ost of their profits fro product sales, but the effect of output-doinance (as explained above) tends to ake advertising becoe ore and ore profitable as copetition increases. However, as copetition becoes high enough, the uch lower incentive to subsidize takes over and product profits relatively benefit, producing the inverted-u shape. We point out that these results cobined suggest that 6 Soe Broadway plays are seen by illions of patrons who are not shown a single ad. However, the Super Bowl cannot charge advertisers for illions of GRPs (gross rating points) without delivering illions of viewers. 4

7 when within-industry copetitive intensity is high enough, equilibriu profits are always higher in the product arket than in the advertising arket. This is true for any product. Firs cannot advertise their way out of intense copetition. We also exaine how the level of copetition between edia 7 for advertising dollars affects overall profitability. We find that an increase in the substitutability of the edia (fro the advertiser s perspective) ay actually increase total profits for soe edia. Moreover, as the edia copete ore intensely with each other due to increased substitutability, their equilibriu business odels diverge - soe of the derive ore of their revenue fro advertising while others derive less. We use the odel, as well, to exaine the total nuber of advertising ipressions all custoers in the product arket are exposed to (or the level of advertising clutter ). We find that advertising clutter is increasing in the nuber of firs within a ediu, but only if interedia copetition is high. If inter-edia copetition is low, the aggregate level of advertising ipressions is decreasing in the nuber of firs. To the extent that arketers or regulators seek to onitor the aggregate level of consuer exposure to advertising for purposes of gauging overall clutter, these findings highlight the iportance of explicitly considering the ipact of copetition both within and between edia. We extend the odel to incorporate the realistic possibility that advertising creates disutility for consuers during product consuption. This issue has becoe especially acute for Internet content providers with the advent of intrusive advertising forats. For exaple, ivillage.co has announced plans to decrease significantly its use of the pop-up ad forat on its web site after consuer feedback indicated high levels of annoyance. In this context, our analysis reveals that as the arginal disutility to consuers fro advertising initially increases, profits generated fro advertising actually increase while those fro selling products to custoers decrease. The rest of the paper is organized as follows. In the next section we relate our work to the relevant literature. In Section 3 we develop a parsionious odel of edia copetition where firs can generate revenue in both a product arket and an advertising arket. We present the core 7 We use the ters intra-ediu and within-ediu copetition to refer to copetition a fir faces fro other firs in its own ediu. We use the ters inter-edia and between-edia copetition to refer to copetition a fir faces fro other firs in a separate ediu. throughout the paper. We also use the words ediu and industry interchangeably 5

8 results on advertising profits versus product profits as a function of within-industry copetition. In addition, we consider the ipact of between-industry copetition and the iplications of copetitive intensity on total consuer exposure to advertising. In Section 4, we extend the basic odel to explicitly consider the ipact of the disutility that advertising ay create for consuers. The paper concludes with a suary of the results and a discussion of anagerial iplications. 2 Relevant Literature The extant literature relevant to our analysis can roughly be categorized into three broad areas: (1) research that has epirically analyzed the nature of edia firs business odel decision, (2) other analytical odels of edia, and (3) general odels of ulti-product firs in which there exists soe relationship or externality between the different products. We relate our work to each of these streas in turn. Several epirical papers docuent differences in revenue source for a particular ediu over tie. Data presented in Suner (2001) suggests, for instance, that the predoinant fraction of agazine revenue has shifted fro circulation (charging custoers) to advertising between 1980 and Several reasons are put forward for this shift, including the possibility that the inherent value of agazines to advertisers (relative to other edia such as network television) has increased and that consuers find agazine advertising less annoying and ore satisfying than television advertising. We control for both of these factors in solving for the fir s optial decision. Hall (1976) analyzes the deise of ass circulation agazines in the late 60 s. His findings suggest that agazine publishers followed a strategy of increasing editorial content pages when increasing the volue of advertising pages (the latter tended to increase when circulation increased). Using a systes dynaics ethodology, he shows that the cobined effect of increasing circulation followed by increasing ad page volue and decreasing arginal advertising rates can lead to reduced overall profitability. Siilar relationships are shown epirically in Krishnan and Soley (1987), who present evidence that firs have becoe ore conscious of this potential negative dynaic and correct for it. Liited epirical work considers the relationship between ultiple edia. Liebowitz (1982) exained the ipact of cable networks entry on television broadcasters revenues and advertising rates in Canada. He found that, overall, broadcasters revenues slightly increased as a result of 6

9 cable operators entry while their advertising rates reained largely unchanged. To the extent that cable operators have exclusive territories while broadcasters have overlapping networks, this result is consistent with our finding (in Section 3.2) that the entry of a new but less copetitive ediu ay result in higher profits for the incubent ediu. 8 There have also been a liited nuber of theoretical investigations into questions specifically related to edia, but quite different in focus fro the question we ask here. Chaudhri (1998) addresses the issue of edia onopolies. In particular, he shows that the traditional view that a onopolist will restrict output, to the detrient of overall welfare, is not necessarily true in circulation industries. He shows that the output of a onopolist in the product arket will be higher due to the existence of a perfectly copetitive (i.e., price taking) advertising arket as copared with a case in which an advertising arket is not present. This basic result is consistent with our odel. He also identifies conditions under which the product is priced above or below arginal cost. Since his odel, by design, specifies the ediu as a onopoly, it does not allow for an explicit analysis of the ipact of copetitive behavior. Moreover, he does not address the relative profitability of the advertising business and the product business which fors the core of the present analysis. Chen and Xie (2002) study a duopoly where cross-arket network effects (see footnote 5) are present. Their paper focuses on how the size of an ex-anté loyal custoer base with which an incubent fir is endowed affects product pricing and profitability. Interestingly, they show that, in soe cases, the fir not possessing a loyal base (an entrant) ay earn higher profits than the incubent. Their analysis has iplications for incubent source of profits as a function of the size of the exogenous loyal segent. Masson et al. (1990) look at the relationship between advertising and products with a specific focus on the disutility caused by an increase in the relative proportion of advertising vs. content. They odel the ipact on the consuer s edia consuption decision so that an increase in advertising inutes will cause soe consuers to switch (or be diverted) to other firs. Their odel yields the counter-intuitive result that as copetition within the ediu increases, prices in the ad arket ay actually rise if this diversion effect is large enough. The specification of our odel allows us to analyze the ipact of advertising 8 The particular edia chosen for the study ay exhibit iportant idiosyncrasies related to regulation and the relationship between Aerican and Canadian edia arkets. This ay liit the generalizability of the results in Liebowitz (1982). A study presenting results consistent with Liebowitz (1982) can be found in Wildan et al. (2002). 7

10 disutility (see Section 4). However, we do not assue an ex-anté diversion effect and thus our results are driven by very different phenoena. Specifically, our odel draws heavily on the fact that different edia (radio and television, for exaple) copete for the sae advertising dollars and on the interdependence between the product and advertising arkets. 9 Dukes and Gal-Or (2003) odel the edia fir s decision as to whether or not to enter into an exclusive contract with an advertiser. Such a contract precludes the television station, for exaple, fro accepting ads fro other firs in the sae category. To address this, they odel (again, in a single-ediu context) the interaction between consuer decisions (which stations to listen to and which products to buy), edia decisions (how uch to charge for advertising) and anufacturer decisions (how uch to pay for advertising). Their analysis yields the key insight that exclusive advertising contracts ay benefit anufacturers by softening price copetition at the product level. This occurs because the overall aount of (inforative) advertising declines with such an arrangeent. This is likely to be true when the firs in the ediu are highly differentiated. Finally, Dukes (2002) addresses the question of the optial level of advertising in society. He shows that the ore differentiated the stations (for exaple, the ore different the prograing across radio stations), the ore excessive the aggregate level of advertising. This odel is also useful in explaining why soe product categories (specifically, those with a low level of differentiation) ay by characterized by a higher level of advertising than others. It is iportant to note that, given their focus, both Dukes and Gal-Or (2003) and Dukes (2002) liit their analyses to contexts in which the firs charge only for advertising but not for content. Thus, these odels are not able to address the question of relative profitability of products and advertising in a general way. 10 Finally, there have been a nuber of recent papers odeling copetition aong firs producing several products that are soehow interrelated. While these papers do not explicitly address edia questions, they are nonetheless relevant to our analysis as there are soe siilarities in the odeling approach. Van Alstyne and Parker (2002) analyze a context with an exogenous 9 Masson et al. (1990) focus on a single-ediu context. While they do extend their odel in an appendix to allow the fir to charge consuers a subscription price in addition to generating revenue by selling to advertisers, their goal in this extension is to re-affir that if the audience diversion effect is large enough, increased copetition will raise ad prices. 10 See also Steiner (1952) and Beebe (1977) for early odels of intra-ediu copetition. 8

11 internetwork externality. While the odel does not directly relate to edia, it ight predict the attractiveness of the product relative to the advertising business as a function of the level of disutility associated with advertising. In their odel, it would see that ore disutility decreases the relative attractiveness of the advertising odel. However, as we show in Section 4, a coplete analysis of the copetitive forces at work ight yield the opposite prediction. At a ore general level, Cabral and Villas-Boas (2003) identify conditions for a ultiproduct fir to be worse off when there is a direct positive interdependence (either in the deand or cost structure) between the different products it sells. This occurs because strategic copleentarity in prices can in soe cases cause firs to price ore aggressively as the degree of interdependence increases. Hence, the negative strategic effect arising fro copetition ore than outweighs the positive direct effect arising fro the interdependence. They also show that the ain findings of Strauss (1999) are a special case. Our paper differs in that the priary focus is on the relative source of fir profits fro each arket rather than on total profits. In addition, our set-up differs fro theirs in that, consistent with ost edia contexts, firs set quantities rather than prices (see footnote 14). Furtherore, the dependence between the two arkets in our odel is not entirely isoorphic to the conditions necessary to obtain their results. Lastly, we depart fro the approach taken by each of these papers in that we odel the relationship between products and advertising at a ore structural level. This affords us several advantages in answering our priary question of interest. First, we believe that it accurately captures the specific context (edia) in which we re interested. Moreover, it allows for the ipact of each business on the other to vary as a function of all of the paraeters of the odel. Thus, we endogenize this relationship rather than specify it exogenously. In su, neither the theoretical nor epirical literatures have offered a general answer to the question of where the fir s profits should optially coe fro. We consider this to be the priary contribution of this paper. Moreover, we believe that our odeling approach in particular our explicit consideration of both intra- and inter-edia copetition represents a significant contribution. This is iportant because revenue sources fro consuers and advertisers are strongly linked within the fir but they are characterized by entirely different copetitive environents. There have also been no attepts to our knowledge to explicitly exaine the joint roles of inherent product value and advertising disutility in this context. 9

12 3 Model We odel copetition within and between two industries A, B. By industry, we think of newspapers, agazines or television, for exaple. Let industry consist of N firs. So, N easures the copetitiveness within each respective industry (we denote the nuber of firs in the rival industry as N ). For exaple, if industry is broadcast television in the US, then the firs within the industry would include ABC, CBS and NBC. If the industry is general news agazines, then firs within the industry would include Tie and Newsweek. Each fir in the odel copetes in two arkets: (1) a product arket in which each fir sells its product to interested custoers, and (2) an advertising arket in which it sells the attention of these custoers to advertisers. Product arket copetition is assued to take place only within industry, while advertising arket copetition takes place both within and between industries. 11 In the product arket, firs face the standard linear (inverse) deand: p = v N X i=1 x i, (1) where x i represents the quantity chosen by fir i in industry, andv represents the inherent value of the products offered by firs in industry. So, while ABC and NBC are product-arket substitutes and salon.co and thestreet.co are product-arket substitutes, custoer deand for watching an ABC show has no ipact on that of custoer deand to visit the salon.co website. We specify the price per custoer ipression q an advertiser pays as follows: q = w N X i=1 NX y i γ y j, (2) where y i is the nuber of advertising capaigns sold by fir i in edia type. For exaple, this would be the nuber of ads appearing in a given issue of the Boston Globe or the nuber of j=1 coercials on an NBC show. The inherent value of a custoer ipression to the advertiser in ediu is given by w. This ight capture, for exaple, the attractiveness of the audience 11 In a separate analysis (available fro the authors) we show that our ain results hold even if there is copetition between industries in the product arket as well. Note also that we assue advertisers are drawn fro a set outside of the (N + N ) firs in the odel. 10

13 reached by the ediu or the level of attention generated. Once again, subscript ( ) refers to the other ediu, 12 and γ easures the substitutability between edia types, i.e. to what extent advertiser deand for placing ads in ediu affects deand in ediu. In our forulation, each capaign provides the advertiser with access to the fir s x i custoers, and total payent by an advertiser to fir i for running the capaign is then q x i. This rate reflects the fact that the ore custoers will be exposed to the ad, the ore the advertiser is willing to pay for running the capaign. This characterization is consistent with the observed practice in ost edia. For exaple, online advertising payent is typically rendered on a CPM (cost per thousand ipressions) basis. In television, advertisers pay per GRP (gross rating point) and in radio on a cost per point basis. 13 We point out that the specification in Equation (2) assues that the price for a custoer ipression is not itself an ex-anté function of the nuber of custoers delivered. Besides atching the observed practice, as explained above, this specification is also attractive in that it iplies that advertiser payent (linearly) increases in the nuber of custoer ipressions. This characterization is also consistent with Chen and Xie (2002) and Dukes (2002), though in both of these papers the authors assue that price-per custoer ipression is exogenously given. Given (1) and (2) and assuing, without loss of generality, zero arginal cost of production in both arkets, the profit function for fir i in industry {A, B} is the su of profits generated in the product and advertising arkets: " Π i = p x i + q y i x i = v N X i 0 =1 x i 0 # x i + w N X i 0 =1 NX y i 0 γ y j y i x i. (3) Each fir selects quantities x i and y i in a standard Cournot fashion. 14 Solving for the equilib- 12 So, for exaple, in the profit functionforfirs of edia type A, refers to firs of edia type B. 13 In these schees, one point corresponds to 1 percent of the target audience reached through the ediu. In the newspaper and agazine industries, it is custoary to charge an advertising rate for the aount of space allocated to the ad (full page, colun lines of classified ad etc.). However, it is coonplace for the edia buyer to translate this rate to an effective CPM (see O Guinn et al. (2000), pp ). Moreover, several studies (Thopson (1989), Chen and Xie (2002)) have clearly shown that such rates are strongly and positively correlated with overall circulation (note that advertising rate would correspond in our odel to the ter q x i ). 14 The specification of the odel as a Cournot gae is a natural one for several reasons. On one hand, firs in j=1 these industries tend to think in ters of the underlying quantities. Magazine publishers explicitly consider how 11

14 riu yields: y = C K, x = " 1 v + (N +1) µ # 2 C hv +(y) 2i =, K (N +1) (4) Π = " 1 (N +1) 2 v + µ # 2 2 C =(x K ) 2, where K 1+N + N + N N (1 γ 2 )andc w + N (w γw ). Note that the equilibriu is unique and syetric within industry. Thus, we drop the subscripts i and j that denote individual firs unless this distinction is necessary. We require that prices in both arkets be non-negative. 15 It is easy to verify that prices in the advertising arket are q = C K. Since K>0, a necessary and sufficient condition for non-negative advertising prices is: C 0 w w > γn, {A, B}. (5) N +1 For the condition in (5) to hold for both edia, we siply need the relative intrinsic value of a custoer ipression in each ediu to be not too different (since the RHS of (5) is always strictly less than 1). In the product arket, p = v N (C /K) 2 N +1 so a sufficient condition for any pages of advertising to have in each issue, radio broadcasters consider how any stop sets to run per hour and the anagers of websites decide how any banners and/or pop-ups to serve (Gotha (2002)). The dependence of ad revenue on the underlying nuber of custoers fro the product arket cause edia firs to consider output as the relevant strategic variable; Newspaper publishers tend to think in ters of circulation, Internet sites in ters of page views and visits and radio broadcasters in ters of nuber of listeners. Moreover, the Cournot odel is generally considered to be a reasonable way to odel copetition when the firs have capacity constraints (Kreps and Scheinkan, 1983). This characterization clearly holds with respect to both of the arkets in which our firs copete. For exaple, in the product arket there is significant fixed cost associated with launching a new agazine, newspaper, radio station or television station. Siilarly, in the advertising arket, there are clear constraints on the nuber of ads a fir can run during a progra and the nuber of banner ads a website can serve up in a reasonable tie. Finally, this approach is consistent with Masson et al. (1990) which utilizes a Cournot specification as well as with Dukes (2002) which specifies the edia fir as choosing the aount of advertising it will run. 15 Note that if we included variable costs, the copany ight actually be selling products at an operational loss. 12

15 non-negative product prices is that v be high enough. 16 Given these results, Π a and Π p,the firs equilibriu profits in the advertising and product arkets respectively can be written as: µ " 2 Π a = qy x = y 2 x C v + C K 2 # =, (6) K N +1 " Π p = p x v N CK 2 #" v + C K 2 # =. N +1 N +1 It is interesting to copare the results of this odel to a baseline odel in which the firs are not able to act as advertising edia. 17 In this case, the firs axiize the following profit function: " # XN eπ i = ep ex i = v ex i. (7) The standard Cournot equilibriu solution with only a product arket is characterized by: i 0 =1 ex i 0 ex = ep = v N +1, (8) µ 2 eπ v =. (9) N +1 Before presenting our results, we highlight several observations that will aid in providing intuition for the ain findings. Reark 1 (Subsidization) Copared to the case in which it cannot sell advertising, the fir always (a) produces strictly ore products and (b) sets product prices that are strictly lower. These are clear fro a coparison of Equations (4) and (8). We also see that total profits are strictly higher when the fir has the option of being an advertising ediu (this is consistent with previous work on circulation industries, e.g. Chaudhri (1998)). However, an iportant iplication of this overproduction is that product profits are lower than they would be if the fir could not advertise: µ Ã Π p Π e C 2 = Ãv N K (N +1) v + µ!! 2 C < 0. (10) K 16 Specifically, we assue that v N CK 2 throughout. 17 This ight be for technical reasons (for exaple, it is difficult for services firs to sell the attention of their clients to advertisers), or because the value of a custoer ipression to the advertiser is negligible. Another reason could be related to regulatory constraints. 13

16 One interpretation of (10) is that firs product business optially subsidizes their advertising business. Itisiportanttopointoutthatsinceall firs in the ediu have an incentive to overproduce, there is a strong cobined negative effect of subsidization on product price (and hence product profits). As we discuss below, this subsidy is a drain on product profits in favor of the advertising business. Reark 2 (Output-Doinance) High advertising sales iplies high product sales. However, thereverseisnottrue. This can be seen in the following inequality: x (y ) 2 = p 0. (11) Since prices are constrained to be non-negative, high y necessarily eans that x is high. This is related to Reark 1 in that the extent of the fir s overproduction of products is increasing in (y ) 2. We also point out that p is increasing in both γ and N. The reason for this is clear: higher value of both of these paraeters iplies ore inter-edia copetition, which akes the advertising arket less attractive and less worth subsidizing. y and x ; the latter effect has a positive ipact on p. This results in lower levels of both On the other hand, when there is little effective inter-edia copetition (either γ or N are low), then there is a great deal of subsidization since the advertising arket becoes very attractive. It can also easily be verified that when v is not too low (or γ and N high enough), price in the product arket is decreasing in N. This iplies fro (11) that: p 0 x (y ) 2 < 0. (12) Thus, output in the product arket is ore sensitive to intra-ediu copetition than advertising output is in this case. Furtherore, it is straightforward to show that equilibriu product output in our odel is ore sensitive to N, as copared to the benchark case (7)-(9) where linkage with an advertising arket is not possible, i.e., ex < x The sae is not true for the advertising arket. The equilibriu nuber of capaigns offered by each fir (y ) would be the sae if we assued a benchark scenario, whereby the nuber of custoer ipressions is fixed. 14

17 3.1 Advertising Profits vs. Product Profits: The Ipact of Within-Industry Copetition We now turn to our ain inquiry which is an understanding of what drives relative profitability in the two arkets. We define the difference between advertising profits and product profits in industry as Π a Π p. We re interested here in the ipact of copetitive intensity, as easured by the nuber of firs in the industry (N ), on. 19 The following Proposition offers insight into this issue and shows that the inherent value of the product (v ) plays an iportant oderating role. (All proofs are contained in the Appendix). Proposition 1 (a) There always exists a v and a N such that for all v <v, when N < N, > 0 and when N N, < 0. (b) If γ and N are high enough, then there also exists a v >v and a pair N, N ª such that for v <v < v,whenn N, N, > 0 and < 0 for N / N, N. (c) For all other values of v that don t satisfy (a) or (b) above, < 0 for all N. A graphical representation of Proposition 1 is shown in Figure 1. The first thing to notice is that in all cases, < 0whenN gets high. The advertising arket never yields ore profits than the product arket in highly copetitive industries. This is true even when w is high so that ads are likely to be very effective at attracting the attention of custoers and will thus be of great interest to advertisers. High levels of copetition hurt the advertising business ore than the product business. The second point to notice, as expected, is that when the inherent value of the product is high the advertising arket is never attractive, and when the inherent value of the product is low the advertising arket tends to generate ore profits (if copetition is not too intense). However, when v is oderate, the relative profitability of the advertising odel as copared with the product odel is characterized by an inverted-u shape over a certain range. 19 It is natural in the context of our odel to easure the intensity of copetition within the ediu as a function of N. As Tirole (1988) (pp. 222) states:...consider syetric firs...the only reasonable easures of concentration are then equivalent to the nuber of firs in the industry. See also Seade (1980) for a justification of differentiating with respect to the nuber of firs to study the effects of entry. We also point out that in an alternative specification one could have a fixed cost of operation that deterines the nuber of firs active in equilibriu. Then, one could study changes in the fixed cost (that serves as a easure of barriers to entry) and obtain qualitatively siilar results. 15

18 D (a) : v < v (b): v < v < v N (c) : v > v Figure 1: Relative Profitability of Selling Advertising vs. Products The core intuition behind the results can be understood by focusing on the interplay between the two econoic forces introduced above: output-doinance and subsidization. To see how these forces relate, note that the direct effects, i.e., holding fir choices constant, of copetition on fir profits in the two arkets are: Π a = x y 2, Π p = x 2. (13) Fro Reark 2, we know that the direct effect is ore pronounced on products than it is on advertising. Specifically, Π a Πp = x y 2 + x 2 = x x y 2 > 0. (14) Thus, output-doinance the fact that the fir will not have high advertising output without having high product output results in positive pressure on. That is, ceteris paribus, will be increasing in N. Yet another key factor to consider is the ipact of ore copetition on 16

19 product output choice and what this eans for profits in each arket: 20 Π a x i x, µ Π p x i +(N 1) Πp x j 6= i. (15) x j The first expression in (15) is a cross-arket strategic effect related to how a change in product output ipacts advertising profits, while the second expression is the ipact of the change in all firs product output choice on product profits. 21 Looking at the net effect of these ters on we get: Π a x i µ x Π p x i +(N 1) Πp x = x j y 2 x y 2 (N 1)x x x = 2y 2 +(N 1)x < 0. (16) Thus, subsidization has the opposite ipact fro output-doinance on with increased copetition (recall that x < 0). Specifically, the degree of subsidization declines with ore intense copetition and advertising profits suffer while product profits benefit. To suarize, output-doinance drives up while subsidization pulls it down. We now turn to the three cases of Proposition 1 and explain their intuition in ters of output-doinance and subsidization. 20 It is useful to decopose the ipact of a paraeter into direct effects and strategic effects (Tirole (1988), pp ). The forer capture the ipact of changes in the paraeter holding fir decisions constant. The latter capture the ipact of changes in fir decisions induced by changes in the paraeter. Clearly, the expression in (14) is the difference in direct effects in the two arkets, while the quantities in (15) represent a central subset of the strategic effects of copetition on relevant in edia arkets. These strategic ters are related to the subsidization effect of product output discussed in connection with Reark 1. Note that it is less instructive to look at strategic effects related to changes in y because these do not ebody uch added insight beyond a benchark odel with no endogenous dependence on the product arket (see also footnote 18). For copleteness, we provide in the Appendix a full decoposition of the direct and strategic effects. 21 Note also that in the above Π p i / x i is not zero. This should not be confused as a contradiction of the envelope theore which says that the derivative of the total profit function for fir i with respect to x i should be zero at the interior optiu. This is because here we are looking at the derivative of product profits only: Π p i and not of total profits: Π i = Π p i + Π a i (it is obviously the latter that is optiized in our case). 17

20 3.1.1 Low-Value Products: case (a) When v is low, p is low and products are ainly produced in order to attract custoers whose attention is sold in the advertising arket. More iportantly, x is close to y 2. ipact of output-doinance is negligible, and subsidization is the key driving factor. As such, the Hence, decreases in N and becoes negative at soe point. It is particularly noteworthy that, in this case, even for low v the product odel contributes ore profits when copetition is high enough High-Value Products: case (c) When v is high, the opposite is the case. Here, the subsidization effect becoes negligible. To see this, note that x is strictly decreasing in v. Higher-valued products are less sensitive to copetition. Moreover, since x is increasing in v and y is not a function of v, the ipact of output-doinance (which is proportional to x 2 )grows. Thus, is strictly increasing in N in this region. However, it always reains negative. When the product is of high value, product output is principally for profitable sale in the product arket. Thus, product prices are very high relative to advertising prices Moderate-Value Products: case (b) Here we face an interplay between the ipacts of output-doinance and subsidization, resulting in the inverted-u shape shown in Figure 1. To understand why this occurs, observe that at low N outputs per-fir are high. Hence, output-doinance (14) initially prevails over subsidization (16). As N increases, both effects are decreasing in absolute value, however the forer declines ore rapidly than the latter. In particular, since product output is ore sensitive to copetition than advertising output (see (12)), x x y 2 gets very sall as N gets large. At the sae tie the existence of a ter proportional to N in (2y 2 +(N 1)x ) x, which originates fro the fact that all firs in the ediu are overproducing, itigates the decline of subsidization. This results in subsidization having greater ipact at higher N, yielding the inverted-u shape. As stated in the Proposition, we need for γ and N to be high enough to yield the pattern above. When these paraeters are sall (when inter-edia copetition is not very intense), then p is sall. This is an intuitive iplication since a decline in the intensity of copetition fro 18

21 the other ediu akes the advertising business ore attractive. Since p vanishes, it ust also bethecasethat x approaches y 2 of N is always below that of. cross-over, the inverted-u shape cannot occur. In this case, the curve of x x y 2 as a function (2y 2 +(N 1)x ) x. As the two decreasing curves do not In su, we find that the advertising odel is relatively attractive under very specific conditions. First, we need for copetition within the ediu to not be too intense. Second, as expected, we need the inherent value of the product to neither be too low nor too high. When this value is very low, the advertising odel is relatively attractive for low levels of copetition. It is only when the value is oderate, that we get the intriguing result that the advertising odel is attractive when the within-industry copetitive intensity is neither too low nor too high. These results ight help explain the apparent contradiction in the revenue odel dynaics discussed in the Introduction. There, we noted the difference between the alternative newsweekly category as copared with the dot-co content category. Both seeed to be facing ore copetition. However, in the forer, we observed an increased focus on advertising profits while in the latter the opposite was true. It sees reasonable to assue that the alternative newsweeklies (such as the Boston Pheonix) originally faced a relatively low degree of intra-ediu copetition as copared with dot-cos (such as salon.co). Thus, if we assue that both of the deliver oderate value, an increase in copetitive intensity would explain a relative increase in advertising revenues at the newsweeklies and a decrease in advertising revenues at dot-cos. 3.2 Media Substitutability The traditional odel of intra-industry copetition generally suggests that the lower the differentiation, or the higher the substitutability, between firs the lower are fir profits. The classic exaple of this result is found in standard Hotelling odels. In these odels, fir profits are strictly decreasing with lower transportation costs, which is a easure of substitutability. However, this is not necessarily the case in the inter-industry context that we odel here (i.e., where two separate industries copete). Specifically, as the next Proposition shows, it ay be the case that greater substitutability (higher γ) is beneficial for one industry but not the other. This will be shown to be due to the ipact of substitutability when the two industries are asyetric. 19

22 Proposition 2 Let K 0 1+N + N + N N 1+γ 2. If 2γN (N +1)w >w K 0 (17) then overall fir profits in industry ( ) are increasing (decreasing) in the degree of interedia substitutability. Otherwise, overall fir profits in industry are decreasing in inter-edia substitutability. So, greater substitutability, i.e. higher values of γ, ay actually help soe firs. Note, by syetry, that Proposition 2 iplies that if 2γN (N +1)w >w K 0,thenprofits in industry are increasing in edia substitutability and profits in industry are decreasing. This gives rise to the following Corollary: Corollary 1 For any given set of paraeters, at ost one of the two industries can have profits that are increasing in the degree of inter-edia substitutability. The intuition behind these results lies in the fact that an increase in substitutability has two priary effects on a fir in industry. First, for any given level of output y,thisiplies lower prices. Thus, ceteris paribus, this leads to lower choices of y. However, the sae forces are at work in industry. If the latter respond by choosing a lower y,thenitayinfact be optial for firs in industry to select higher y. That is, the two industries are facing two changes to their output (and therefore pricing) decision: a negative direct ipact on price as a result of higher γ and a potential positive strategic ipact on price if the other industry drops output. As we show in the proof to the Proposition (see Appendix), when condition (17) holds, then an increase in γ results in higher y and q. As suggested by (4), x will also be higher. Thus, firs in industry produce ore products and ore capaigns and sell the latter at a higher price. This results in higher profits. The key factor that drives this result and the reason why we don t observe it in standard intraindustry copetition settings is that when industries and are asyetric (N 6= N ) they respond differently to the level of edia substitutability. 22 More specifically, Equation (17) 22 To see this, assue that N = N N and that w = w w. Then, Equation (17) becoes 2γN (N +1)w>wK 0. But, then, it should hold for both industries and which contradicts Corollary 1. Thus, it is the asyetry that is driving the result. 20