Information Systems Research Vol. 16, No. 4, December 2005,. 400 417 issn 1047-7047 eissn 1526-5536 05 1604 0400 informs doi 10.1287/isre.1050.0069 2005 INFORMS Managing Piracy: Pricing and Samling Strategies for Digital Exerience Goods in Vertically Segmented Markets Ramnath K. Chellaa Goizueta Business School, Emory University, 1300 Clifton Road, Atlanta, Georgia 30322-2710, ram@bus.emory.edu Shivendu Shivendu Deartment of Economics, University of Southern California, 3620 South Vermont Avenue KAP 300, Los Angeles, California 90089, sshivendu@gmail.com Digital goods lend themselves to versioning but also suffer from iracy losses. This aer develos a ricing model for digital exerience goods in a segmented market and exlores the otimality of samling as a iracy-mitigating strategy. Consumers are aware of the true fit of an exerience good to their tastes only after consumtion, and as iracy offers an additional (albeit illegal) consumtion oortunity, traditional segmentation findings from economics and samling recommendations from marketing, need to be revisited. We develo a two-stage model of iracy for a market where consumers are heterogeneous in their marginal valuation for quality and their moral costs. In our model, some consumers irate the roduct in the first stage allowing them to udate their fit-ercetion that may result in re-evaluation of their buying/irating decision in the second stage. We recommend distinct ricing and samling strategies for underestimated and overestimated roducts and suggest that any otential benefits of iracy can be internalized through roduct samling. Two counterintuitive results stand out. First, iracy losses are more severe for roducts that do not live u to their hye rather than for those that have been undervalued in the market, thus requiring a greater deterrence investment for the former, and second, unlike hysical goods where samling is always beneficial for underestimated roducts, samling for digital goods is otimal only under narrowly defined circumstances due to the rice boundaries created by both iracy and segmentation. Key words: digital roducts; exerience goods; vertical segmentation; ricing; iracy; samling History: Sanjeev Dewan, Senior Editor; Alok Guta, Associate Editor. This aer was received on Aril 16, 2004, and was with the authors 6 months for 3 revisions. 1. Introduction Publishing, software, music, movie, and videogame vendors are transforming their business models due to ongoing technological changes. Three common themes bind these roduct categories. First, they are moving from hysical to largely digital forms. Second, they belong to a class of economic goods known as exerience goods. Third, they are subjected to a articularly disturbing trend of iracy due to increasing ease of dulication and availability of illegal coies. Even as these digital exerience goods industries are struggling with online strategies and ricing issues, trade associations (global industry revenues are in arentheses) such as the Business/ Entertainment Software Alliance (BSA/ESA $60 B), the Recording Industry Association of America (RIAA $9 B), and the Motion Picture Association of America (MPAA $14 B) have also been reorting iracy losses to the tune of $22 billion in non-u.s. markets alone. Industry secific resonses to iracy have been varied; while software and games businesses are exerimenting with vertical segmentation or even samling, those that belong to the RIAA and the MPAA have been fighting iracy largely through legal actions against irates and eer-to-eer (2) services such as Morheus and Kazaa (Goldstein 2002) and through technological tools such as decoy strategies, watermarking, and tagging (see whiteaer at htt://www. bayts.com). Technological deterrence efforts are often limited in their efficacy (Berst 2002) as they are only as good as the first successful hacker, 400
Information Systems Research 16(4),. 400 417, 2005 INFORMS 401 while legal deterrence relies on enforcement and consumers awareness of the law. The existence of legitimate demand for digital exerience goods, albeit with variable rice boundaries, iracy deterrents, and quality/feature references has largely been ignored. While the industry has been engaged in iracy roofing, academic literature has largely focused on ricing issues without adequately addressing the exerience goods character of these digital roducts. The imending arrival of digital television combined with digital recorder technologies such as TiVo and RelayTV is only likely to exacerbate iracy losses (Mount and Caulfield 2002), and hence it is imerative to re-evaluate the overall imlication of ricing, segmenting, samling, and deterrence strategies. Our research is rimarily based on analytical models of vertical segmentation (Mussa and Rosen 1978), exerience goods (Nelson 1970), roduct samling (Ailloni-Charas 1984), and emirical research on iracy (Goal and Sanders 1998). While recent research (Sundararajan 2004) has studied rice discrimination under iracy albeit through a quantity-rice schedule, we adot a classical verticalsegmentation aroach and model a quality-rice schedule (contract) along the lines of Chellaa and Shivendu (2003a). It is known that ricing of hysical exerience goods whose valued attributes cannot be fully ascertained before consumtion needs to take into account the initial information or knowledge that consumers may have about them (Liebeskind and Rumelt 1989), and that differential ricing strategies are needed when a market overestimates versus underestimates the valued attribute (Shairo 1983). A distinguishing characteristic of digitized exerience goods, in contrast to their hysical counterarts, is that vendors cannot let consumers exerience their roducts without fear of it being irated. Further, simle rice-based exlanations offered by early analytical models (Conner and Rumelt 1991) need to be ooled with nonrice, iracy-related asects such as consumers exectation of deterrence, revalent ethical indices (Solomon and O Brien 1991), and consumers ethical roensity to irate (Goal and Sanders 1997), as well as institutional and cultural asects of the market (Ginarte and Park 1997, Marron and Steel 2000). Recent analytical work (Chellaa and Shivendu 2003a) has abstracted the above nonrice deterrence factors through consumers moral cost of irating and exected unitive costs. We extend this model to include a recent emirical observation that the second most imortant reason why consumers irate is their need to try out (Cheng et al. 1997), suggesting that some iracy could have been revented if some other legitimate means of trying out was rovided. As the concet of return is ill-defined for digital roducts, even the alicability of otimal return olicies is somewhat limited (Che 1996). Generally, this legitimate consumer need has largely been ignored in academic research with the excetion of recent emirical work that examines the role of consumer search (Goal et al. 2002, Hui and Png 2001). While for most hysical goods one-time consumtion without lifetime ownershi can be easily rovided through samles, exiration of samles or other forms of limited ownershi have to be exlicitly built into the roduct. In this aer, we examine ricing and samling strategies through a two-stage consumer iracybehavior model and by abstracting consumers ercetion of a digital roduct s fit to their tastes based on their indirect and direct exeriences. Secifically we examine segmentation and samling under iracy when roducts may be underestimated or overestimated by the market. Our research recommends a artroduct samling aroach where the otimal samle size not only deends on the revalent deterrence costs but also on the relative division of this cost across the two stages. This aer is organized as follows: In 2, we develo our two-stage iracy model and analyze vendor s ricing strategies in a vertically segmented market. In 3, we investigate the efficacy of samling and deterrence requirements to mitigate iracy. Section 4 concludes with a discussion of theoretical and managerial imlications while noting the limitations. The roofs of the lemmas and roositions are given in the aendix. 2. Model Transactions involving digital roducts may be modeled as rincial-agent contracts in that there is an agreement (to not use a irated coy), a otential for breach (by irating), some enforcement ossibilities
402 Information Systems Research 16(4),. 400 417, 2005 INFORMS (through deterrence mechanisms), and some information that is rivate to the agent (their valuation for quality and roensity to irate). Our model consists of a rincial (digital roduct vendor) and two tyes of agents (consumers), each of whom have either a high or a low < marginal willingness to ay for quality x. The market is divided in roortion and 1 in these consumer tyes and the vendor knows only the oint distribution that is common knowledge, while consumers know their own tyes. This is osed as a contract-theoretic framework along the lines of Laffont and Martimort (2002), who observe that such discrete setus turn out to be sufficient to highlight the main henomena arising around adverse selection without having to deal with the technicalities of continuum of tyes (. 31). In our model, the vendor can offer two qualities of a digital roduct a high x and a low quality x. Note that as in most economic literature, we use the term quality to refer to any valued attribute of a roduct (Liebeskind and Rumelt 1989, Nelson 1970, Shairo 1983). Information goods readily lend themselves to versioning or quality segmentation (Varian 1997), and one can indeed observe different forms of versioning and segmentation strategies today (see Table 1). Let the vendor s cost of offering a quality x be a + cx, where a is some fixed cost and cx is the cost of roducing quality x, where c > 0, c > 0. This cost function merely states that the cost of roducing a higher quality (or feature) becomes increasingly exensive, e.g., imroving uon a base 2D videogame to create a 3D multilayer interactive game becomes increasingly costly as each new feature is added. As with other digital roducts, the marginal cost of serving an additional customer is assumed to be zero. Note that the quality/features of the digital good x reresents the nonexerience art of the good (known to consumers before consumtion), and secifying a fixed/constant marginal cost of serving additional consumers does not affect the qualitative arguments resented below. 2.1. Consumers Percetion of a Digital Product s Fit to Their Tastes Our digital goods have a lifetime utility, i.e., consumers need to own a coy of the full roduct by urchasing or irating, be it software, music, movies, or games. While the quality of the good is known, consumers only have beliefs about the good s fit (exerience art) to their references or tastes, known as the rivate information contained within an exerience good (Nelson 1970, Shairo 1983). In the case of software, the exerience art becomes aarent only after installing it in one s own work environment and using it for secific tasks; for entertainment roducts the hedonic asects reresent the exerience art. The initial ercetion of a digital good s fit to the consumers references and needs is known to be shaed by indirect exeriences created by advertisements (McGuinness et al. 1992), word-of-mouth communication, and signals such as warranties. However, as Shairo (1983,. 498) oints out, these sources of nonconsumtion information have limitations, e.g., advertising is subject to credibility roblems warranties are limited by adverse selection or moral hazard roblems. Thus, the true value to a consumer of an exerience good is known only after consumtion. Table 1 Common Segmentation Strategies for Digital Goods Software Music Movies Video games Small versus large number of features/comonents included Network versus nonnetwork versions Automatic versus manual future udates Availability of online technical suort, manuals, etc. e.g., Microsoft Office standard versus remium Lossy versus lossless comression (influencing the ability to be layed in HiFi audio systems) Burning and multiformat conversion enabled versus single-format, single-machine requirement Inclusion of live versions versus studio recordings e.g., AAC (Ale) versus MP3 Wide screen versus regular format Seamless branching of movie versus ability to handle multile story lines Menus, interactive features, and other extra content versus lain vanilla otions Broadband versus narrowband for online delivery Hi-definition versus regular format e.g., Siderman double disc release Multilayer versus single layer versions Multilevel versus low-level games Virtual collaborative team lay versus indeendent lay e.g., Many RealArcade games have two versions
Information Systems Research 16(4),. 400 417, 2005 INFORMS 403 In our model, we denote the consumers ercetion of how a roduct fits their tastes or references by k 0 1, and as fit ercetions have no natural scale, we effectively choose a scale by setting the maximal k at 1 (when a roduct fully fits the consumers tastes). We reresent the initial ercetion (common to all consumers) by k = k 0, where k 0 maybe lesser than or greater than the true fit of the roduct. We classify our digital exerience goods to be one of two tyes such that after consumtion, we have k = 1, suggesting that the good fully matched the consumers tastes and was initially underestimated, or k = 0, imlying that the good s fit was overestimated initially. A more general form of this abstraction is to assume that the initial ercetion of fit k 0 has a distribution F and the true fit, which is realized only after consumtion, is given by a distribution G. An overestimated roduct will imly that F first-order stochastically dominates G, while an underestimated roduct will imly the converse. In this aer for the sake of simlicity, we assume that F and G are oint functions as discrete models are often sufficient to highlight the qualitative results. 2.2. Consumers Utility from Purchasing or Pirating a Digital Product The Two-Stage Model Consumers in our model have a linear utility function k 0 x with zero reservation value, i.e., they will buy at rice if their utility is nonnegative. An imortant difference between hysical exerience goods (as modeled by research in economics) and digital exerience goods (as modeled in this aer) is that the latter offers two ossible consumtion channels to the consumers to udate their belief on the exerience art of the good: buying and iracy. When irating, consumers do not incur a monetary cost but deending uon individual secific factors (age, gender, and other ethical indices), they are known to suffer an internal cost of committing an illegal activity (Solomon and O Brien 1991) catured by a construct called the moral cost of irating (Chellaa and Shivendu 2003a). This is given by a arameter U0, where consumers closer to suffer a high moral cost reresenting the most moral individuals with the least ethical roensity to irate (Goal and Sanders 1997). A arameter E catures nonindividual secific iracy costs due to both technological and legal factors and is a measure of the revailing difficulty of irating that can be influenced by vendors. While country-secific coyright laws and track record of enforcement contribute to the consumers exectation of getting caught and the resultant fine (Ginarte and Park 1997, Globerman 1988), vendors can also make investments in technological and legal deterrence to increase E, e.g., through digital rights management (DRM) tools and law suits (Borland 2002, Gentile 2003). Because the vendor is creating two versions of the same roduct, we assume that consumers suffer the same cost of irating for either version. This is consistent with the intuition that once a vendor has a certain DRM, he is unlikely to dumb it down for a lower quality of the same roduct, just as a consumer is unlikely to feel a greater moral cost in irating a high-fidelity version of a song as comared to its comressed, low-quality counterart. Therefore, for any quality, the consumers combined cost of iracy is given by + E UE + E. Note that in this model, we assume that the irated coy suffers no deterioration in quality. Piracy literature has largely modeled the consumers choice between buying and irating as a single-stage decision, where the irated goods act as a cometing seller s roduct line (although illegal) and consumers comare rices with iracy costs (see Chellaa and Shivendu 2003b for a discussion). Such models have not considered an imortant asect of the consumers iracy behavior that some consumers irate roducts such as software to ascertain their usefulness (Cheng et al. 1997). Thus, the consumers decision to irate may not always be final. It may be an intermediate ste of exeriencing where if the irated good is not found useful it will be discarded. If the irated good is found useful, consumers may ossibly kee the irated coy or erhas be incentivized to buy a legitimate coy. This iracy behavior may be otentially beneficial to vendors in that irates are informed of a roduct s usefulness, and some research has exlored this otentially beneficial asect of iracy in the context of ositive externalities (Takeyama 1994), diffusion of legitimate software demand (Givon et al. 1995), and increased revenues (Goal et al. 2002, Hui and Png 2001). Note that rior
404 Information Systems Research 16(4),. 400 417, 2005 INFORMS literature assumes that iracy benefits will somehow roagate through the market to increase aggregate demand to benefit the vendor. However, no light is shed on how vendors may oerationally be able to aroriate or extract iracy benefits. By modeling individual consumer behavior, rather than assuming a demand function, we are able to inoint the imact of iracy on an individual s buying decision so as to develo oerationalizable vendor strategies. If consumers may kee or erhas even discard a roduct after exeriencing it, any costs of irating need to be slit between the stage when they first irate and the ste where they finally decide to kee or discard the irated roduct. An imortant oint to note here is that whether or not irates kee or discard the irated good, their knowledge of the roduct s fit to their taste has been udated due to direct consumtion. This creates an interesting situation where some irates who did not consider urchasing based on their initial ercetion may now, based on their udated ercetion, decide to discard the irated version and buy a legitimate coy. Therefore, we develo a two-stage model of iracybehavior. In the first stage, consumers decide between buying, irating, or not articiating in the market. While the buying and not articiating actions are final, irating is transient because in the second stage some irates may choose to discard the coy while some may buy a legitimate one. To cature the fact that consumers will incur the full cost of committing an illegal activity only if they continue to own the irated coy, we introduce a transient cost arameter 0 1, such that the irating cost in the first stage is given as +E. The first or the transient stage can be interreted as the stage where the irates may not feel the full moral obligation toward coyright or fully suffer from the fear of getting caught until they have decided to kee the irated coy in the second stage. If consumers decide to buy the roduct in Stage 2, then they do not carry over their cost of irating (as they had only temorarily held the illegitimate coy) because they have now aid the full rice for the legitimate roduct (Figure 1). Thus, while all irates suffer the transient cost in the first stage, those who buy a legitimate coy in the final stage do not kee this cost. When consumers move from Stage 1 to Stage 2, their ercetion of the roduct s fit now changes due Figure 1 To Pirate or Not: The Consumers Two-Stage Decision Tree Realized fit k = 1 (Not considered as the condition does not satisfy the IRs of otimal contract) Discard if Kee irated if Buy if θx (1 δ)(β + E) < 0 θx < 0 θx (1 δ)(β + E) > 0 > (1 δ) (β + E) Consider iratingif Do not buy if Buy if k 0 θx δ(β + E) > 0 > δ(β + E) k 0 θx > 0 < δ(β + E) Realized fit k = 0 Discard irated coy θx > 0 < (1 δ) (β + E) (Not considered as the condition does not satisfy the IRs of otimal contract) k 0 θx < 0 k 0 θx δ(β + E) < 0 Stage 1 Stage 2
Information Systems Research 16(4),. 400 417, 2005 INFORMS 405 Table 2 Notation Descrition Notation Descrition Notation Marginal willingness to ay for quality of consumers (heterogeneous, two tyes) Moral cost (heterogeneous, distributed) Deterrence cost Transient cost Initial ercetion of a digital good s fit to tastes (known to consumers before consumtion and based on advertising/hye-related ercetions) Realized fit of the digital exerience good (known to consumers only after consumtion, legally or otherwise, of the full roduct ) Percetion of the digital exerience good s fit after consumtion of samle of size s 0 1 (high tye) (roortion (low tye) (roortion 1 U0 E Digital good quality (any valued attribute and known to consumers) Cost of roduction variable in roduct quality and zero marginal cost of serving additional consumers) x (high quality) x (low quality) cx (high quality) cx (low quality) k 0 (all roducts) Otimal rices in the absence of iracy (high quality) (low quality) k = 1 (underestimated roduct) k = 0 (overestimated roduct) k s = k 0 + 1 k 0 s (underestimated roduct) k s = k 0 1 s (overestimated roduct) Otimal rices in the resence of iracy Otimal rices in the resence of iracy and when samling is emloyed s s (high quality) (low quality) (high quality) (low quality) to directly exeriencing the good, and the direction in which it has changed deends uon whether the good was initially under or overestimated as shown in Figure 1. Thus, k = k 0 is now udated to k = 1 for the underestimated roduct, while it becomes k = 0 (or some negligible value) for the overestimated roduct. In the latter case, the utility from the roduct becomes near zero, and hence consumers will neither buy nor kee the irated coy. Note that it is the iracy cost that is slit across two stages and if the consumer (based on his udated fit) buys a legitimate coy after having irated in the first stage, then he does not carry over the interim iracy costs incurred in the first stage. Further, consumers will only irate the highquality roduct even if they may have been otential urchasers of the low-quality roduct. The simle intuition is that the low-tye consumer s utility from irating the high-quality roduct is greater than irating the low-quality roduct at the first stage because the cost to a consumer in either situation is the same + E, while the value to the consumer from the high-quality roduct is higher x >x. Therefore, in the second stage the tyes will trade off crosssegment irating with own-segment buying, i.e., x 1 +E versus x. Now we consider the vendor s rofit-maximizing behavior. We first consider the case where iracy is absent and subsequently incororate consumers two-stage iracy behavior in the vendor s ricing decision. (Table 2 rovides a summary of notation used throughout the text and the aendix.) 2.3. Vendor s Problem In the absence of iracy, the vendor s ricing decisions that maximize his rofit deend only on the roduction cost of creating two qualities and the roortion that will buy each quality, given by max cx + 1 cx a (1) Along the lines of well-known vertical-segmentation strategies under incomlete information (Liebeskind and Rumelt 1989), the vendor would take into account the consumers individual rationality (IR) and incentive comatibility (IC) constraints to determine rices for the two qualities. If the vendor finds it otimal to maintain two segments (see 2.4 for cases when segmentation is subotimal), he will set rices for the high- and low-quality roducts to be = k 0 x x and = k 0 x, resectively (see the aendix for details). In models with asymmetric information (the vendor does not know the consumers tyes), vendors have to ay an information rent so as to revent the high-tye consumer from being temted to buy the low-tye roduct (given by k 0 x in our model). Note that as k 0 increases, the
406 Information Systems Research 16(4),. 400 417, 2005 INFORMS information rent to be aid increases and the overall revenue k 0 x x + 1 x also increases. Hence, in terms of actionable strategies in the absence of iracy, the vendor can erhas focus on raising k 0 so that consumers believe that the roduct is highly likely to fit their tastes. Indeed, when we observe the advertising strategies of many firms and in different roduct categories, the focus of television, rint, or Web messages is often to make the roduct aear closer to the consumers references. The strategy of roduct hying erhas needs to be revisited for digital exerience goods. If a vendor has over-hyed a hysical exerience good to make it attractive to consumers and because consumers will not be able ascertain the full exerience asect until they have urchased and consumed it, the vendor may be able to extract a higher rice and can in some sense get away with it! On the other hand, for digital exerience goods, the otion of iracy rovides an indirect oortunity for some consumers to exerience the roduct without urchasing it, thus otentially limiting the value of hying to the vendor. Piracy can also otentially affect the segmentation strategy of the vendor. Note that a common strategy to mitigate iracy losses is to lower rices (Chellaa and Shivendu 2003a); however, the strategy of lowering or changing rices is itself constrained by rice boundaries that are needed to maintain segmentation. Hence, before we develo any rice-related strategies, we first need to identify the ortions of consumers who will change their iracy/buying behavior due to the udating of their fit in the second stage. Lemma 1a. The roortion of high-tye consumers that will buy a high-quality good during the first stage is r e = 1 ( ) + E and the roortion of low-tye consumers that will buy a low-quality good at this stage is r e = 1 ( k 0 x + E The consumers will irate in Stage 1 only if their transient cost of irating + E is less than the corresonding rices charged for the highand low-quality roducts. From our earlier discussions, we can see that while the high-tye consumer ) comares the rice and costs of irating the highquality roduct, the low-tye consumer will comare his utility from irating a high-tye roduct k 0 x + E with the utility from urchasing the lowquality roduct k 0 x. In other words, what was hitherto unavailable to the low-tye consumer due to rices is now available due to iracy. In this aer, we are mostly interested in the general case when both the high- and low-quality roduct markets encounter a mix of irates and legitimate buyers. This will be true in Stage 1 only if E < < + E for the hightye and E < k 0 x < + E for the low-tye, and because >k 0 x, we can safely say that there will always be a mix of irates and buyers whenever < + E and E < k 0 x. If these two conditions are not satisfied, then either all consumers irate (a trivial roblem) or no consumers irate (the solutions for which have been derived earlier). Thus, the boundaries on the legal/technological deterrence E and the highest combined iracy costs in the market + E are not a restrictive assumtion; rather they deict a general case where some iracy does occur in both segments. Note that these boundaries also ensure that the roortions of buying or irating segments discussed in Lemmas 1a and 1b are bounded between 0 1. Because the decision of those who buy in the first stage is final, for the second stage we only need to analyze the behavior of first-stage irates. Because exeriencing the digital good now fully informs the irates of the true fit of the good to their references, deending uon whether a roduct was overestimated or underestimated the realized ercetion of fit k is now either zero or one. It is also easy to see that if the roduct was overestimated (realized k = 0), indeendent of their marginal willingness to ay for quality, consumers will discard the irated roduct in the second stage. Hence, we are interested in the roortion of consumers who will engage in the further and final action (buying, keeing the irated coy, or discarding it) only for underestimated goods. Lemma 1b. For an underestimated roduct in the second stage, the roortion of high-tye irates (consumers who had irated in the first stage) that will buy a high-quality roduct is given by r f = 1 ( + E ) 1
Information Systems Research 16(4),. 400 417, 2005 INFORMS 407 and the roortion of low-tye irates who will buy the low-quality roduct is r f = 1 ( + E x ) 1 Shairo (1983) oints out that exerience good vendors should ursue two very different ricing aroaches based on whether they exect their consumers to be essimistic or otimistic (in estimating a valued attribute) and there cannot be an otimal generalized aroach. This observation is significant to our iracy context as well, as in the second stage, an overestimated digital roduct will neither be bought by the irates nor will they kee the irated roduct. Thus, we need to consider searate objective functions for the two roduct tyes. The combined roortions of the high- and low-tye consumers buying the roducts meant for their resective segments can be written as Pr buy x = r e + r f r e r f Pr buy x = r e + r f r e r f Thus, we write the rofit-maximization roblem for an underestimated roduct as [ = max r e + 1 r e r f cx (2) + 1 r e + 1 r e r f cx a ] (3) and for the overestimated roduct r f r f = 0, the rofit function is written as [ = max r e cx + 1 r e cx a ] (4) First, because both consumer tyes only irate the high-quality roduct, there is nothing to be gained by changing the rices of the low-quality x digital good, and therefore the vendor will charge the same rices as in the absence of iracy = = k 0 x for the low-quality roduct. However, he may need to lower the rice of the high-quality roduct to incentivize some would-be irates to buy. Note that while rices under iracy are a function of the deterrence and transient costs, the rice the vendor can charge for the high-quality roduct is still constrained by segmentation rice boundaries. The vendor cannot rice above as it will incentivize the high-tye consumers to buy the low-quality roduct, and similarly he will not rice below k 0 x because at such low rices the vendor might as well not maintain segmentation. The difference in rice has been referred to as the moral rent in single-stage iracy models and is aid to the high-tye consumers as an incentive not to irate (Laffont and Martimort 2002). Proosition 1. The otimal rice for an underestimated digital roduct is = if k 0 x< < if where = E + E 2 31 E 2 2 3 The otimal rice for an overestimated digital roduct is = + E/2. Proosition 1 rovides some imortant insights about the vendor s ricing strategies under iracy when he finds segmentation to be otimal. A first insight is that while the rice of the low-quality roduct is indeendent of the transient cost and the initial estimation of the roduct s fit, the rice of the high-quality roduct is critically deendent on both. A second insight is that for an overestimated roduct, the moral rent aid is decreasing in the transient cost arameter, which imlies that for an overestimated roduct, the vendor is better off when consumers find irating even at the first stage for exeriencing the good to be highly immoral and illegal. On the other hand, the transient cost has an ambiguous effect on the moral rent to be aid if the roduct is underestimated. For underestimated roducts, the moral rent is the highest when the cost of irating is evenly sread across the two stages = 1/2. Hence, for underestimated roducts, the vendor is better off under two extreme situations: (1) when consumers consider irating to exerience the roduct to be accetable 0 but keeing the irated coy to be highly immoral, and (2) when they consider irating to exerience the good itself to be highly immoral 1. Hence, in these extreme situations, the vendor might as well rice the underestimated
408 Information Systems Research 16(4),. 400 417, 2005 INFORMS good based on the realized fit after irating, i.e., ricing with resect to x, that is, k 0 x. The higher rice oint suggests that iracy after all may be good in some consumer segments in that it could generate externality-tye benefits. This is consistent with recent research (Heiman et al. 2001) that observes that some engage in iracy to samle digital roducts and that reducing the iracy costs may be beneficial to vendors (Ailloni-Charas 1984). However, a significant claim of our research is that simly relying on iracy alone to generate these benefits is erhas detrimental as the vendor is now solely deendent on the deterrence abilities to revent the consumer from keeing the irated coy in the second stage. Such investment in iracy-roof deterrence (even if only for the second stage) may erhas not be otimal and is discussed in 3.2. 2.4. Piracy and Viability of Segmentation Strategies While we have redominantly discussed ricing under segmentation, even in the absence of iracy, it is not always otimal for the vendor to create two roduct qualities. These extreme conditions are commonly referred to as the shutdown conditions in economics literature on vertical segmentation (McGuinness et al. 1992). Proosition 2. Under some extreme market conditions (that differ based on the resence or absence of iracy), the vendor may consider doing away with segmentation. However, whether a high- or low-quality good is offered and whether only the high-tye consumer (only low-tye being served is never an otion) or the entire market is served is determined by the relative valuations and roortions of the two consumer segments. Whether a vendor will ersist with segmentation is a decision that is closely related to the variable cost of quality, the relative marginal value for quality, and the roortion of consumers in each segment. Consider one examle of an extreme market, where the marginal valuation of the low-tye is significantly low <. In such a situation and for certain cost functions (given in the aendix), the vendor will consider offering only the high-quality roduct and will set rices such that he serves only the high-tye k 0 x rather than the entire market k 0 x. The decision to adot segmentation will further deend on the cost of offering the low-quality roduct, where if cx>k 0 1 x x, he will offer only the high-quality roduct. Similarly, one can derive conditions for other extreme cases; however, throughout this aer we consider the more general case where offering both qualities is otimal to the vendor. It is salient to note here that the vendor may engage in such an evaluation (segmentation or not) even in the resence of iracy and may consider shutting down the roduction of one quality tye, albeit for additional reasons. During vertical segmentation with information asymmetry, the quality offered to the low-tye consumers is lower (just as the rice charged to the high-tye consumers is lowered) than when the vendor knows the consumer tyes / > x/x. Thus, if the rice derived in Proosition 1 is low (as combined iracy costs are low) such that k 0 x, the vendor will only offer one quality, namely the high quality. The intuition behind this observation is that if the iracy costs are so low such that the derived rice is lower than the utility for the low-tye from consuming the high-quality roduct, the lowtye consumers will begin to buy the high-quality roduct at that rice. Hence, it does not make sense for the vendor to offer the low-quality roduct as well. Again, whether he will serve only the high-tye k 0 x or all of the market k 0 x will further deend on the relative valuations and consumer roortions, just as whether he will consider offering the lowquality roduct rather than the high-quality roduct will additionally deend on the roduction costs. In addition to the conditions described in the absence of iracy, the iracy costs to the consumers ose stricter bounds on when it is viable to maintain segmentation. In 3, we evaluate the efficacy of legitimate samling strategies (rather than deending uon the consumers to irate and learn about the roduct) and otimal investments in deterrence abilities. 3. Samling and Investments in Deterrence Mechanisms to Mitigate Piracy A main theme of our digital roduct samling strategy is that it is feasible to convert some of the consumers who irated in the transient stage into buyers by roviding a samle. We can indeed observe various dimensions along which such roduct samles
Information Systems Research 16(4),. 400 417, 2005 INFORMS 409 Table 3Common Dimensions of Samling for Digital Goods Software Music Movies Video games Exires after a certain time Has only a ortion of songs or albums Online clis and trailers Very similar to the software industry Has only a ortion of features Limited number of lays DVDs, online clis with and emloyed by most game Cannot handle full data sets Retailers like Amazon and CDNow and a montage of clis ublishers (restricted based on the number online stores such as Rhasody Studios such as Paramount Many RealArcade games have two of columns or file size) offer samles in both Windows and Disney commonly versions Is unable to save coies a certain and Real formats for downloads rovide these samles number of uses directly from their after Trialware by Adobe, Macromedia, etc. websites are being tested for different digital exerience goods (Table 3) even if their efficacy and strategic benefits have not been studied in academic research. From a ractical ersective, a samle is a art of the full roduct. Such roduct samles may refer to a full roduct that exires after a certain time, or it could be a subset of the value roviding features of the roduct. Our concetualization is one of lifetime utility where the consumer must own (through urchasing or keeing the irated coy) the full roduct. Samling has been used to good effect by hysical exerience goods vendors, and research in marketing sheds light on the usefulness of samling in increasing goodwill in the long term and roensity to buy in the short term (Heiman et al. 2001). It is a way of allowing consumers to exerience roducts with little risk or obligation by giving a trial size ortion of roducts (see Ailloni-Charas 1984 for a comlete discussion on samling). Because samles serve as a direct source of information to consumers, it is indeed suerior to influencing them to buy than advertising, which only offers indirect exeriences (Liebeskind and Rumelt 1989). Conveniently for hysical roducts, the consumers needs cannot be fully met as samles are not a relenishable source, and if consumers like the roduct they have to eventually urchase it when they run out of samles. However, note that it is not always clear that samling is a suitable marketing strategy for exerience goods that can be irated. Intuitively, we can see that samling may reduce the segment of buyers (or the rice they are willing to ay) if the roduct fit has initially been overestimated and hence is not a suitable strategy, but even for underestimated roducts where samling increases the fit ercetion, samling may not be otimal. For examle, a samle not only rovides the otential buyers with a greater incentive to urchase, but it also rovides the otential irates with a greater incentive to irate. It is also not clear if samling will have the same imact on both the high- and low-tye consumers given that the lowtye consumers are temted to irate only the highquality roduct. Taking these factors into account, it is also not obvious as to what ortion of the roduct should be rovided as a samle. To address these issues, we introduce an abstraction of consumer learning from samling similar to that of Heiman et al. (2001). Let the free samle size of the digital roduct be given by s 0 1, where s = 1 refers to the digital roduct in its entirety. Uon exeriencing the samle, a roduct s initial fit ercetion k 0 is udated to k s. This ost-samling ercetion of a digital roduct s fit may be higher k s >k 0 or lower k s <k 0 deending uon whether the good was initially underestimated or overestimated. For analytical tractability, we assume that learning from directly exeriencing the samle is linear where the realized fit is given by k s = k 0 +1 k 0 s for underestimated roduct and k s = k 0 1 s for the overestimated case. Because samling is costless to consumers in our model, all consumers exerience the samle rior to making the buy/irate decision and udate their ercetion of fit from k 0 to k s k s for underestimated (overestimated) roducts. If the new rice contracts are s for quality x and s for quality x when a samle of size s is rovided, the roortions derived in Lemmas 1a and 1b can be rewritten as in Lemma 2. Lemma 2. For underestimated roducts, the roortion of low-tye consumers who will buy a low-quality good at Stage 1 itself is r es = 1 ( k ) s x + E
410 Information Systems Research 16(4),. 400 417, 2005 INFORMS and in Stage 2aroortion of irates from this segment given by r fs = 1 ( x ) 1 + E will buy the low-quality good. The corresonding roortions for the high-tye consumers are given by r es = 1 ( ) s + E and r fs = 1 ( + E ) s 1 resectively. Before we consider the case where a roduct is underestimated, we first analyze the simler case of overestimation. The vendor has to reduce rices for buyers in the low-tye segment, and those who irate will neither kee the coy nor buy as their realized fit after exeriencing the roduct is zero. However, the high-tye consumers may ursue one of three otions: (1) buy the high-quality roduct if the samle offered is s 1 ( k 1 k 0 0 x x ) (2) buy the low-quality roduct (when the samle size is greater than in (1)), or (3) irate the high-quality roduct (in which case they will not kee the irated roduct after fully realizing the fit). The samle given in otion (1) is the maximum samle size that the vendor can rovide that will kee the high-tye consumer from buying the low-quality roduct taking into account the information and moral rents. Even if the consumers may buy under otions (1) or (2), the rofits are decreasing as comared to not offering a samle at all, and hence the vendor should never rovide samles for overestimated goods irresective of other consumer arameters. The reasoning behind offering samles for underestimated roducts is clouded by both the cross-segment iracy behavior of the low-tye consumers and the otential overall increase in the irating oulation, and hence requires a careful analysis. Proosition 3. For underestimated digital roducts, samling directly affects the roortion of low-tye consumers who buy and it will indirectly affect the revenue from the roortion of high-tye buyers by affecting the amount of surlus that can be extracted. Counterintuitively, even for underestimated digital goods, samling is not an unambiguously suerior strategy as iracy by the low-tye consumers go u. The reason behind this occurrence is that while the benefit from consuming the high-quality good (through irating) has increased from k 0 x to k s x, the cost of irating has remained the same. In ricing the low-quality roduct the vendor will attemt to extract full surlus, i.e., = k s s x, and hence at Stage 1, iracy becomes an attractive otion for more low-tye consumers. Note that the rice that the vendor can charge for the low-tye roduct has increased by xk s k 0, but the roortion of irates has also increased by x/ k s k 0. For the high-tye consumers, samling increases the value from k 0 x to k s x, while the cost of irating remains unchanged. The otimal rice of the high-quality roduct will again only deend uon the deterrence level and transient cost arameter as in Proosition 1; however, both the lower and uer bounds for this rice now increase so as to maintain vertical segmentation. The lower and uer bounds of the otimal rice s of the high-quality roduct will now be k s x and k s /k 0, resectively, while the vendor will set the rice given in Proosition 1 if k s x< <k s /k 0. Interestingly, the effect of samling on rices and rofits in the high-valuation segment deends on whether or not assumes boundary values as given in Proosition 1. Counterintuitively, even when roducts are likely to be underestimated, there may be situations where samling can lead to a decrease in rofits, e.g., when is set at the lower bound. This is because the lower bound is chosen only when rice reduction for iracy begins to affect the segmentation strategy, i.e., the vendor cannot reduce rices below the lower bound as high-tye consumers will then be incentivized to buy the low-quality roduct. Hence, increase in fit from k 0 to k s due to samling moves the rice of the high-quality roduct further away from the rofit-maximizing rice. However, if the uer bound was chosen as, then offering samles will lead to an increase in both rices and rofits as rices move closer to the rofit-maximizing rice. If was chosen as the, then samling does not change the rices or rofits in the high-tye segment as this rice is indeendent of change in fit and only deends uon deterrence and the transient arameter.
Information Systems Research 16(4),. 400 417, 2005 INFORMS 411 While the rices for both the high- and low-quality roducts are decreasing with samling for the overestimated goods, for the underestimated goods, the rice of low quality is strictly increasing and that of high quality is weakly increasing with samling. Note that while the information rent that is aid is strictly higher with samling, the moral rent to be aid remains unchanged when the rices are set at the uer bound and is higher otherwise. Hence, the effectiveness of samling as a strategy and the otimal samle size will deend uon the relative values of the willingness to ay of the two segments, the deterrence level, and the transient cost arameter. 3.1. The Otimal Samle Size The effectiveness of samling deends uon the trade-off between the marginal increase in revenue from increased surlus extraction and the marginal loss in revenue from increased iracy. As both the surlus and iracy loss are deendent uon the samle size offered, we seek to find the otimal level. Proosition 4. If the transient cost arameter >1 x/e, then samling is an otimal strategy and the otimal samle size is given by s = 1 [ 1 2 E 2 + Ex k 1 k 0 2 xx 1 E 0 ] Samling is not a good strategy for underestimated digital roducts if the transient cost is sufficiently small < 1 x/e, and even if samling is aroriate, there is an otimal size of the samle imlying that under threats of iracy, the vendor cannot simly offer the entire roduct as a samle. In the case of hysical exerience goods that are not under threats of iracy as discussed in Shairo (1983), Kim (2002), Besen and Raskind (1991), and others, roviding direct exerience through samling or other means is always good for the vendor when the roducts are underestimated. From a managerial ersective this raises some imortant issues, i.e., when should a vendor invest in advertising, hye, and other indirect exeriences (thus increasing the initial ercetion of the fit), and when should he rovide direct exeriences to the consumer? The constraint on the transient arameter rovides us with some intuition; when the consumers suffer a sufficiently high transient cost of iracy in Stage 1, then samling may be an otimal strategy as iracy does not rovide the urorted externality benefits for underestimated roducts. Investments in indirect exeriences (e.g., advertisements and other roduct hyes) are useful only if accomanied by investments in deterrence mechanisms as well. However, there aears to be no ideal exogenously given level of deterrence other than to suggest that it should be so high that no consumer (irresective of their moral costs) irates. The iracy-roof level may be economically detrimental or even infeasible, and hence we need to exlore a vendor s otimal investments in technological and legal deterrence mechanisms. 3.2. Investments in Deterrence Mechanisms I wonder what tye of coy rotection will come next. Maybe they will ban markers! 1 There are two rimary forms of deterrence technological and legal. Technological rotection solutions are often both exensive and short lived; sometimes breaking these technical barriers may be far simler than exected. For examle, Sony and Universal Studios have adoted a very aggressive anti-iracy stance and recently began offering coyroof CDs to revent consumers from illegally distributing songs. However, it has recently come to light that this rorietary rotection mechanism is easily overcome by simly blackening out the edges of the shiny side of a CD using a felt-ti marker, leading to the comment at the beginning of this subsection (Berst 2002). Similarly, even as entertainment firms are ursuing individual irates to make them an examle of successful legal efforts, they are increasingly facing legal challenges due to the comlex nature of online information sharing. For examle, in tracking and roving that an end-user was resonsible for irating, the RIAA has run into the territory of rivacy and an individual s legal rights as evident from a recent court case involving Verizon Communications Cor and RIAA (Conner and Rumelt 1991). Further, the global nature of iracy couled with differences in laws and legal enforcements in different countries (see 1 A music fan s osting on alt.music.rince after it was discovered that Sony s coy-roof CDs could be coied by blackening out the edges on the shiny side of a CD using a ermanent marker (Reuters 2002).
412 Information Systems Research 16(4),. 400 417, 2005 INFORMS Conner and Rumelt 1991 and Takeyama 1994 for a full discussion on the legal asects of coyright enforcement) have led some to observe that, The battle over music iracy is like the war on drugs: You can t win it, but you can fight it forever, and send millions on the battle (Berst 2002). Our research does not suggest an end to deterrence investments; rather we examine if there are other economically sound alternatives to iracy-roof objectives. Prior research on iracy (Chellaa and Shivendu 2003a) finds that rofits and rices may increase in rotection technologies. However, in a vertically segmented market, there may not be roortional surlus extraction as rice cas are also deendent uon the information asymmetry. Because creating deterrence is costly, it becomes imerative to the vendor to consider the rice deterrence interlay in determining its level in different situations. For examle, while a iracy-roof aroach requires a level of deterrence given by E = /, to ursue the same ricing strategies as in the absence of iracy, the vendor only needs to set the level of deterrence to be E u = 2 1 2 2 1 for the underestimated digital roduct, or E o = 2 / for the overestimated roduct (from Proosition 1). This tells us that ursuing a iracyroof aroach will imly very high investments in deterrence if the transient arameter is very small. Further, in attemting to thwart iracy comletely (as some industries aear to be intent uon), the vendor does not make use of the fact that some consumers suffer moral costs at the individual level. Note that in our analysis, we had not considered any exlicit cost function for deterrence E. Let this cost be given by a convex cost function ge, i.e., high levels of deterrence become increasingly costly. This cost can otentially be urely a sunk cost or can have a marginal comonent as well. To determine the otimal investment, we simly need to incororate this cost in the rofit function and set the derivative /E = 0 and simultaneously solve for rices and otimal deterrence. Intuitively, we can see that if this is a urely sunk cost, then the otimal E will be alied to all roduct versions as a higher E is always beneficial if there are no costs to alying it. However, if this arameter has a marginal alication cost (either to each roduct version or each legitimate buyer), then there may be conditions where differential deterrence may be adoted for the two consumer/roduct segments. These conditions may be a function of the size of the high- and low-tye segments and whether the roduct was overestimated or underestimated by the consumers. For examle, if there are marginal costs to deterrence (such as due to licensing or cost of law suits), then a differential deterrence can be adoted for underestimated and overestimated roducts where the otimal deterrence will be bounded by E u and E o, resectively, at the uer limit. Proosition 5. For underestimated digital goods, where consumers do not consider irating to exerience in Stage 1 to be an issue 0, it may be otimal for the vendor to simly ignore iracy and rice the digital roduct assuming that k = 1. Losses from iracy are lower for underestimated digital goods than for their overestimated counterarts, requiring higher deterrence for the latter goods. Proosition 5 discusses aroaches required under certain secific instances of consumers iracy behavior, e.g., when they have negligible cost of irating (due to negligible ) at Stage 1. These may realistically reresent many instances where consumers do not necessarily intend to commit an illegal activity such as irating but merely intend to check out the roduct s fit. In such situations, the intended benefit from samling can be extracted through iracy itself and the vendor s efforts are better sent on managing iracy in the Stage 2. Otherwise, if the vendor attemts to rice his roduct so as to incentivize some consumers to buy in the first stage, then he may end u greatly underricing the digital good. At the second stage, irates are fully aware of the roduct s fit, and hence their oint of value assessment will be x and not k 0 x or k s x, allowing higher rices to be set. However, note that for this to be a viable rice, the vendor is heavily deendent uon the deterrence abilities at the second stage. When rior research on iracy alludes to the externality benefits of iracy, it is this articular ability of the vendor to raise rices (or cature a greater market segment when valuations are distributed) that is being referred to (Shairo
Information Systems Research 16(4),. 400 417, 2005 INFORMS 413 1983). Our results oint out that the so-called externality benefits of iracy can indeed exist, however, only under very narrowly defined circumstances. Intuitively, one might suggest that vendors should invest more in deterring iracy for those roducts that are highly valued by the consumers; however, the two-stage iracy model reveals that it is more imortant to revent consumers from exeriencing those roducts that may be hyed so that consumers are in favor of urchasing at Stage 1. Note that there are two ways in which the vendor can otentially incentivize the consumer to buy at the first stage: (1) by lowering rices, and (2) by increasing the cost of irating through investments in deterrence. In a vertically segmented market, while the vendor can consider lowering rices of the high-quality roduct, he cannot reduce rices of the low-quality roduct as it will create cross-segment buying by the high-tye consumer, and hence his otions are limited to investments in deterrence. A critical issue with deterrence is that the return on investments (ROI) is not always guaranteed as it may take just one hacker to effectively reduce the returns to zero. Our analysis oints out that going after a iracy-free world may not erhas be the best strategy for digital exerience good industries; rather they should consider a combination of roduct and consumer characteristics in determining the deterrence level. Further, if vendors have invested heavily in increasing the initial exectations of the consumer through advertisements, then the vendor should also simultaneously invest in a higher deterrence level. On the other hand, roducts that are underestimated may require lesser deterrence as some amount of iracyrelated externality benefits can be internalized by the vendor. 4. Discussion and Conclusions We develo a two-stage iracy regime for a vertically segmented market where consumers are heterogeneous in their willingness to ay for quality and iracy costs, and where irates from the first stage decide between buying and keeing the irated coy in the second stage after having udated their ercetion of fit. We then derive otimal rices for underestimated and overestimated digital goods in the resence and absence of iracy, and study the efficacy of samling strategies by deriving an otimal samle size and conditions under which it is beneficial to the vendor. Subsequently, we analyze deterrence levels couled with ricing as an alternative to iracy roofing. 4.1. Imlications to Theory and Practice From a theoretical ersective, we integrate research from literature in economics on ricing, information systems research on iracy, and marketing literature on samling. We add to the research in exerience goods ricing by searating the consumers ercetion of a good s valued attribute from indirect exeriences and direct exeriences, and analyzing the imact of iracy in ricing these goods. Our research also extends the single-stage iracy model for vertically segmented markets to incororate a critical emirical observation that some iracy occurs because consumers want to exerience a good first hand, legally or otherwise. In develoing the two-stage iracy model, we add to the emerging literature on roduct iracy and digital exerience goods ricing. From the ersective of a manager, digital exerience goods exhibit two characteristics: (1) the digital art imlies that these goods have negligible marginal cost of roduction, are easily amenable to vertical segmentation, but are suscetible to iracy, and (2) the exerience asect imlies that some roduct information is bundled with the roduct itself and is revealed only uon consumtion. Together, these two characteristics create unique challenges for digital roduct vendors. Digital exerience goods vendors rimarily face two investment choices: (1) invest in roduct advertisement to increase hye, and (2) invest in technological and legal deterrence for coyright rotection. Examining these choices together with the viability of artial roduct samling, we draw some imortant managerial insights for vertically segmented markets. If the vendor invests in raising the initial ercetion of fit for an underestimated roduct, then it ositively influences the vendor s return in two different ways: (1) it can influence the low-tye consumers to urchase rather than irate, and (2) it can allow a greater surlus to be extracted from the high-tye consumers. However if this initial ercetion is a hye, even then
414 Information Systems Research 16(4),. 400 417, 2005 INFORMS vendors can extract higher surlus but this ability is critically deendent on reventing iracy. Hence, an imortant rule of thumb is that a vendor should not hye a roduct that he cannot rotect. Piracy itself is useful to the vendor in limited cases, such as when consumers find irating as accetable but have issues with keeing the irated coy, as this allows the vendor to extract greater surlus from the market. However relying on iracy to generate this surlus imlies that the vendor assumes that iracy costs are tilted toward the second stage. Instead, our results suggest that the vendor can adot a roactive samling strategy to internalize these externality benefits. However, as samling may inform both the otential buyers as well as irates, it is critical that the vendor understands when samling is otimal and how much of the roduct is to be offered as a samle. When the vendor engages in hye advertising early on to raise consumer exectations, it is obvious that he should not engage in samling. Even if informing the consumers of the value is beneficial, however, samling increases the benefit of irating the high-quality good for the lowtye consumers, hence making iracy more attractive for them. Samling does allow the vendor to extract a greater surlus from the high-tye consumers but the amount by which he can raise his rices is constrained by the limits of vertical segmentation, i.e., if rices become too high for the high-quality roduct, the high-tye consumers may be incentivized to buy the low-quality roduct. Thus, an imortant managerial conclusion is that unlike hysical roducts, digital roduct samling is beneficial to the vendor only in narrowly defined cases when the rices for the high-quality roduct are constrained by segmentation requirements and not by deterrence levels. While the samle size is a function of consumer and deterrence arameters, in determining the form of the samle, the digital roduct industry can learn from the software industry that has a long history of roviding trial versions of software that exire and lite versions that rovide a reduced set of features. Finally, our analysis suggests that iracy roofing may never be a feasible solution as investments in deterrence mechanisms are costly. The vendor should determine the otimal deterrence level based on his ricing, advertising, and samling strategies and the consumers iracy costs. Piracy threatens hyed-u roducts more than roducts that are truly valued by the market, and if consumers suffer little or no iracy costs in checking out a roduct, it may be otimal for the vendor to rice roducts assuming that all consumers are aware of the true value, ignoring iracy to a certain degree. In this case, a vendor should ideally invest in legal and watermarking technologies more than coy-rotection technologies as legal ramifications can lay an imortant role in dissuading consumers from keeing irated coies, while coy-rotection technologies rimarily focus on creating or obtaining irated coies. 4.2. Limitations and Future Research As with all analytical models, our model of iracy also has assumtions built in, mostly for reasons of analytical tractability. We do not endogenize quality by assuming a functional form for the cost function, and while this does not detract from our analysis of ricing and samling, allowing quality itself to be endogenously determined is known to be a valuable subroblem. Our model also assumes that irating either version rovides an accurate assessment of the fit because while technical quality may differ across versions, the content (governing the fit) is the same in both versions. Some exerience goods such as movies also have a unique feature where the first consumtion exerience is more valuable than subsequent ones. While intuitively we can see that for such goods, iracy can be more detrimental, and that even goods that have been underestimated are likely to fall in valuation after the first consumtion exerience; the imact of samling is somewhat unclear and can make for interesting future research. Emirical research on the rice oints of vertically segmented digital goods would rove to be invaluable, secifically in comuting otimal deterrence investments to be made. It may also be interesting to exlore contracts between coyright owners and distributors and the nature of royalty negotiation when iracy is involved. Aendix. Proofs of Lemmas and Proositions (Notation Given in Table 2) The rofit function in the absence of iracy is max cx + 1 cx a. The vendor s roblem is subject to individual rationality (IR) constraints (k 0 x 0 IR
Information Systems Research 16(4),. 400 417, 2005 INFORMS 415 & k 0 x 0 IR and incentive comatibility (IC) constraints k 0 x k 0 x IC & k 0 x k 0 x IC. Adding IC&IC, we get x x. Because > 0, x x. IR& IC immediately imly IR, therefore we ignore IR, and IR is binding. Similarly, IC is binding. Therefore, both IR and IC must be binding at the otimum. Hence, in the absence of iracy, the vendor will offer two rice contracts given by = k 0 x x and = k 0 x. Proof of Lemma 1a. We have U0 Prx < = 1/ x x 0 1. The robability that the highvaluation tye will buy at the first stage is ( ) Pr + E = Pr E 1 ( ) + E In the lower-valuation segment, if irates the high-quality tye, she gets a utility k 0 x, and hence she will irate if k 0 x + E. Therefore, the robability that the lowvaluation tye will buy at the first stage is ( ) k0 x Prk 0 x +E = Pr E 1 ( k ) 0 x +E Proof of Lemma 1b. If a digital roduct is underestimated, then the arms of the decision tree in the final stage will be symmetrical to those in Stage 1. We can now substitute the new costs, i.e., + E by 1 + E, and the new fit realized from irating, i.e., k 0 by 1. The total robability of the high and low tyes buying can be given as Pr buy x= r e + r f r e r f and Pr buy x = r e + r f r e r f where r f = 1 ( ) 1 + E and r f = 1 ( + E x ) 1 Proof of Proosition 1. The rofit-maximization roblem for underestimated roducts is [ u = max r e + 1 r e r f cx + 1 r e + 1 r e r f cx a ] Differentiating with resect to, setting the first-order condition to zero, and noting that all the robabilities of the lower tyes are indeendent of, we can solve u to get the roots as = E ± E 2 31 E 2 2 3 The term under the root is always ositive because 0 1 and the second-order condition is negative for the ositive root. The lowest rice for the high quality is determined by the utility that the low-tye consumer gets by consuming the high-quality roduct. Because IR is also binding, the lower bound is given by k 0 x. Because IC is binding, the vendor cannot fix a rice higher than. For the low-tye consumers, IR is binding and they irate only the highquality roduct so the vendor will not change the rice of the low-quality roduct. For an overestimated roduct, anyone who irates will not urchase, and hence the rofit function is given by [ o = max r e cx + 1 r e cx a ] Because all the robabilities of the lower tyes are indeendent of, solving for o,weget = + E/2. Proof of Proosition 2. In the absence of iracy, substituting for the rices, the otimal rofit under segmentation can be written as x x = k 0 x + x x cx + cx + a. Deending uon whether the marginal value for quality of the low tye is rather low < or not, the vendor will set rices such that only the high tyes buy or all consumers buy. In the former case, when the high (low)-quality roduct is roduced, the rice will be k 0 x k 0 x, and in the latter, the rice will be k 0 x k 0 x. Consider the first case when only the high quality is offered (imlies c 0<k 0 ), resulting in a rofit given by x = k 0 x cx + a; comaring this with the segmented rofit, we get the condition that segmentation is not otimal when cx>k 0 1 x x. Similarly, shutdown under other extreme conditions can be derived (the roof is available uon request from the authors). Note that if the consumer tye is known, then x = ; x =, and the quality offered to the low-tye consumers under asymmetry will be lowered when the consumer tye is not known, i.e., x <, and hence / > x/x. While in the body of this aer we show when only a highquality roduct will be offered during iracy, we can also see that if the enforcement is so oor that the combined iracy cost is lower than k 0 x such that k 0 x (the rice charged for the low-quality roduct), then the vendor will offer only the low-quality roduct at rice. Proof of Lemma 2 and Proosition 3. Similar to the roofs of Lemmas 1a and 1b by relacing k 0 by k s, and given that k s >k 0, the robability of buying for the low-tye consumers is lower with samling. Similarly, the uer and lower bounds of s will be weakly higher than, imlying otential for greater extraction for surlus. Proositions 3 and 4 follow from Lemma 2. Proof of Proosition 4. To find the otimal samle size for underestimated roducts, we need to solve s = max s r e + 1 r e r f cx s s s + 1 r s es + 1 r es r f cx a The robabilities of the high-tye consumers are indeendent of s, and we can substitute k s x for s.ass and k s are linearly related, we differentiate with resect to k s, and setting the first-order condition to zero, we get s k s = k s k s xr es + 1 r es r f = 0
416 Information Systems Research 16(4),. 400 417, 2005 INFORMS By simlifying, we have k s = 1 2 E 2 + Ex 2x 2 2 1 E x From the equation k s = k 0 + 1 k 0 s, weget 1 2 E 2 +Ex k s 2x = 2 2 1 E x 0 1 k 0 The second-order condition is 2 s = ks 2 2 1 [ 2xx 2 x E1 For otimal k s, the second-order condition is negative, imlying >E1 /x, and because s = ks k 0/1 k 0 and k s k 0 1, we have s 0 1. Further, /k s > 0, 2 /ks 2 < 0ifk s <ks, /k s < 0, 2 /ks 2 < 0ifk s >ks, and /k s = 0ifk s = ks. Proof of Proosition 5. When 0, the vendor can assume that all consumers irate at the first stage. For an underestimated roduct, the vendor can charge rices for the high-quality roduct given by x x if x x E and + E/2 ife< x x < + E. We can also see that that the terms r f and r f are absent when a roduct is overestimated as there is no buying in the second stage. So, all rofits have to be realized in the first stage, where the iracy threshold is only + E. Formally, we need to rove that the loss from iracy when the roduct is overestimated as oosed to when it is underestimated is weakly higher for all relevant values of the deterrence level E, i.e., to show o E u E E 0E, where = u E = E = o E = E is the rofit when there is no iracy (where E is the iracy-roof deterrence level). At zero enforcement, we need to show that u E = 0 o E = 0 0. We can exand this exression in terms of rices and robabilities as 1 1 r u eu r fu + u r eu o r eo + u 1 r eu r fu 0. Given that all robabilities and arameters are nonnegative, it will suffice to show that u r eu o r eo + u 1 r eu r fu 0, which is always true as it comares the underestimated revenue with the overestimated revenue. The worst that the vendor can do is to set the rices at the overestimated level where he has to extract all revenue at Stage 1 itself. He will raise or lower u as comared to o only if he believes that doing so will yield some revenue from consumers who buy at Stage 2 such that the total revenue will be maximized. Thus, u r eu + u 1 r eu r fu o r eo. Now, if we can show that u /E and o /E are monotonically increasing, then we have shown that u E o E E 0E. Substituting the otimal rices in u and o and differentiating with resect to E, we have u E = 1 [ 2 1 E 9 2 + 2 ( A1 31 2 1 91 1 k 0 q ) E ] + 9k 0 xx 2 1 + 3 2 1 ] where A = E 2 31 E 2 2 and o E = 1 [ E + 2k0 q1 + ] 2 We can see that both u /E and o /E are ositive and monotonic. References Ailloni-Charas, D. 1984. Promotion: A Guide to Effective Promotional Planning, Strategies, and Executions. John Wiley and Sons, New York. Berst, J. 2002. Why technology can t sto music iracy. ZDNET AnchorDesk. htt: /www.zdnet.com/anchordesk/stories/story/ 0,10738,2677668,00.html. Besen, S. M., L. J. Raskind. 1991. An introduction to the law and economics of intellectual roerty. J. Econom. Persectives 5(1) 3 27. Borland, J. 2002. The record labels new target: Users. CNET News.com. htt://news.zdnet.com/2100-9595_22-941590.html. Che, Y.-K. 1996. Customer return olicies for exerience goods. J. Indust. Econom. 44(1) 17 24. Chellaa, R. K., S. Shivendu. 2003a. Economic imlications of variable technology standards for movie iracy in a global context. J. Management Inform. Systems 20(2) 137 168. Chellaa, R. K., S. Shivendu. 2003b. Economics of technology standards: Imlications for offline movie iracy in a global context. R. Srggne, ed. Proc. Hawaii Internat. Conf. System Sci., Vol. 36. IEEE Comuter Society Press, Washington, D.C., 199 208. Cheng, H., R. Sims, H. Teegen. 1997. To urchase or to irate software: An emirical study. J. Management Inform. Systems 13(4) 49 60. Conner, K. R., R. Rumelt. 1991. Software iracy: An analysis of rotection strategies. Management Sci. 37(2) 125 139. Gentile, G. 2003. Downloading fight targets cororations. The Washington Post. htt://www.washingtonost.com/ac2/wdyn/a6917-2003feb14. Ginarte, J. C., W. G. Park. 1997. Determinants of atent rights: A cross-national study. Res. Policy 26(3) 283 301. Givon, M., V. Mahajan, E. Muller. 1995. Software iracy: Estimation of lost sales and the imact on software diffusion. J. Marketing 59 29 37. Globerman, S. 1988. Addressing international roduct iracy. J. Internat. Bus. Stud. 19(3) 497 504. Goldstein, P. 2002. A music lesson on iracy for Hollywood. Los Angeles Times (March 12). Goal, R., L. Sanders. 1997. Preventive and deterrent controls for software iracy. J. Management Inform. Systems 13(4) 29 47. Goal, R., L. Sanders. 1998. International software iracy: Analysis of key issues and imacts. Inform. Systems Res. 9(4) 380 397. Goal, R., S. Bhattacharjee, L. Sanders. 2002. Economics of online music sharing: Cui bono? Working aer, University of Connecticut, Storrs, CT. Heiman, A., B. Mcwilliams, Z. Shen, D. Zilberman. 2001. Learning and forgetting: Modeling otimal roduct samling over time. Management Sci. 47(4) 532 546.
Information Systems Research 16(4),. 400 417, 2005 INFORMS 417 Hui, K. L., I. P. L. Png. 2001. Piracy and the demand for legitimate recorded music. Working aer, National University of Singaore, Singaore. htt://www.com.nus.edu/ ing. Kim, J. 2002. A story of iracy and rivacy. The Washington Post. htt://www.washingtonost.com/ac2/w-dyn/a38034-2002se4. Laffont, J.-J., D. Martimort. 2002. The Theory of Incentives: The Princial-Agent Model. Princeton University Press, Princeton, NJ. Liebeskind, J., R. P. Rumelt. 1989. Markets for exerience goods with erformance uncertainty. RAND J. Econom. 20(4) 601 621. Marron, D. B., D. G. Steel. 2000. Which countries rotect intellectual roerty? The case of software iracy. Econom. Enquiry 38(2) 159 174. Mcguinness, D., G. Phili, S. Mathew. 1992. The effect of roduct samling on roduct trial, urchases and conversion. Internat. J. Advertising 11 83 92. Mount, I., B. Caulfield. 2002. Erasing RelayTV This digital video recorder does things Hollywood can t forgive. Business 2.0 (May). htt://www.business2.com/articles/mag/ rint/0,1643,39432,00.html. Mussa, M., S. Rosen. 1978. Monooly and roduct quality. J. Econom. Theory 18 301 317. Nelson, P. 1970. Information and consumer behavior. J. Political Econom. 78(2) 311 329. Reuters. 2002. Coy-roof CDs foiled by a marker: Sony s attemt to sto digital coying quickly rebuffed. MSNBC Tech. Section. htt://www.msnbc.com/news/754854.as. Shairo, C. 1983. Otimal ricing of exerience goods. Bell J. Econom. 14(2) 497 507. Solomon, S. L., J. A. O Brien. 1991. The effect of demograhic factors on attitudes toward software iracy. R. DeJoie, G. Fowler, D. Paradice, eds. Ethical Issues in Information Systems. Boyd & Fraser Publishing Comany, Boston, MA, 161 181. Sundararajan, A. 2004. Managing digital iracy: Pricing and rotection. Inform. Systems Res. 15(3) 287 308. Takeyama, L. N. 1994. The welfare imlications of unauthorized reroduction of intellectual roerty in the resence of demand network externalities. J. Indust. Econom. 42(2) 155 166. Varian, H. R. 1997. Versioning information goods. B. Kahin, H. R. Varian, eds. Internet Publishing and Beyond: Economics of Digital Information and Intellectual Proerty. MIT Press, Cambridge, MA, 1 13.