THE DOT COM EFFECT REVISITED: AN EVENT STUDY OF THE VALUE PROPOSITIONS OF EC INITIATIVES ON THE MARKET VALUE OF FIRMS Mani R. Subramani Assistant Professor in Information and Decision Sciences msubramani@csom.umn.edu Eric A. Walden Doctoral Candidate in Information and Decision Sciences ewalden@csom.umn.edu Carlson School of Management University of Minnesota Minneapolis, MN 55455 Last Revised: October 2, 2001 Abstract: The value drivers of electronic commerce are not well understood. Based on pioneering work in the field of information technology value, we propose and empirically test a general theory of electronic commerce value. Specifically, we propose that the important factor to consider is the creation of intangible, electronic commerce technology complementing assets. We use the event study methodology to test the hypothesis that organizations that are most easily able to create and deploy these intangible assets will derive the greatest return for investors. Not only do we use the short run event study common to the information systems literature, but we also apply new techniques for long run event study analysis hitherto unused in information systems research. Our results show support for the theory of the primacy of intangible assets in determining returns to electronic commerce initiatives. We further supplement our analysis by evaluating returns through the lens of electronic commerce business models. The results of this analysis suggest that consideration of intangible assets explains return to electronic commerce initiatives better than business model considerations. Keywords: Intangiable assets, electronic commerce, event study, financial methodologies, business value of information technology Acknowledgements: The authors would like to thank Gordon Davis, Tom Holmes, and George John for their comments, Kelly Slaughter for his coding expertise, and the participants of the University of Minnesota IDSC workshop. All omissions, errors, and blunders remain the sole responsibility of the author. DRAFT VERSION available online at http://www.ericwalden.net "Bottom Line: Nobody Really Knows" {|Neuborne, 2000 #217, page 98|} INTRODUCTION The turn of the century bore witness to an entirely new type of information technology (IT) enabled way of doing business—electronic commerce (EC). The Internet, the purview of a few research scientists and enterprising college students, became accessible to the public at large. Hardware, such as inexpensive personal computers and fast modems, combined with software, such as Mosaic and Netscape, to allow worldwide access to the existing network infrastructure. Joined with location independent transaction systems, such as credit cards, and established parcel delivery systems EC was an inevitable and logical extension to traditional commerce. Now with the number of internet users approaching a billion <ref> and the number of web pages well beyond that number <refs>, EC is clearly an established way of doing business. However, the value propositions of EC are far from understood. Considerable upfront IT investments are required to be a viable player in the current EC environment. The Gartner Group estimates that firms pursuing EC initiaitives spend $1 million in the first five months, and $20 million “for a place in cyberspace that sets them apart from the competition” {|Diederich, 1999 #230|}. Moreover, these costs are projected to increase considerably over time {|Satterthwaite, 1999 #231|}. Enough dotcom firms failed in 2001 to prompt Fortune Magazine to add a feature entitled Deathwatch1, which puts the number of firm failures in through September 2001 at 303 {|Schlosser, 2001 #232|}. The quintessential Internet firm, Amazon, has lost billions and has not yet shown a profit, while CDNow and Pets.com have both folded. However, other foremost Internet firms like Yahoo, eBay and AOL have consistently posted profits. We propose that the primary driver of EC value is not the considerable investments in IT or the business model used, but the ability of the organization to create and deploy intangiable assets to complement the IT. We investigate this using capital market reactions to these different EC initiatives as a measure of business value created. Not only do we examine these reactions through the common short run event study {|Dos Santos, 1992 #176; Im, 2001 #184; Subramani, 2001 #185; Chatterjee, 2001 #209|}, but we also add a long run event study analysis as is becoming accepted in finance research {|MacKinlay, 1997 #175; Cowan, 2001 #187; Fama, 1998 #194|}. In combining both a short run and a long run analysis, we hope to shed additional light on the important issue of how EC creates value. The paper is organized as follows. In the next section, we review the literature on EC value and the case for using capital market reactions to measure it. Then we discuss the theoretical foundations underlying number important EC initiative characteristics. We then describe the event study methodology for both short run and long run event studies. Following that we discuss the data and collection methodology, after which, we present our findings. We close by discussing the findings, recognizing opportunities for future research and drawing conclusions. 1 See also http://www.downside.com/deathwatch.html. 1 DRAFT VERSION available online at http://www.ericwalden.net LITERATURE REVIEW Measuring value creation through IT has historically been a difficult task. While the work of a number of researchers has established that IT does create business value, {|Brynjolfsson, 1998 #154|}, actual quantification of IT benefits remain difficult. Prior approaches to measure returns from IT and complementary investments have used a variety of accounting data, such as return on assets {|Barua, 1995 #172|}, cost savings {|Mukhopadhyay, 1995 #173|} or return on investment {|Hitt, 1996 #174|} to understand the value of these investments. While they have certainly expanded our understanding these accounting-based measures present special difficulties in the measure of EC business value. First, they tend to be insensitive to the strategic nature of IT investments that often create benefits to firms in the form of flexibility and expanded operating choices in future periods {|Benaroch, 1999 #183|}. Moreover, as these benefits often accrue over time periods as long as eight years {|Brynjolfsson, 1998 #154|}, for which EC data is not yet available. The use of forward-looking measures is suggested as one way to overcome these deficiencies {|Brynjolfsson, 1997 #215; Bharadwaj, 1999 #182|}. Consistent with this view, we examine the impact of individual firms' EC initiatives on the expected stream of future benefits by focusing on capital markets’ reactions to EC initiatives. The abnormal returns generated by an EC initiative are the consensus estimate of a large number of informed investors of the expected stream of future benefit streams associated with the initiative. In an efficient capital market, investors are assumed to collectively recognize future benefit streams accruing from initiatives and reflect them in the stock price of the firm. If EC activities of firms are expected to enhance future cash flows, the capital market would respond favorably to unanticipated EC announcements, resulting in an increase stock price. The event study methodology {|Fama, 1969 #162; Brown, 1985 #160|} is designed specifically to take advantage of this aspect of financial markets, making it a very useful tool for management researchers to examine consensus estimates of the future benefits streams attributable to organizational initiatives {|McWilliams, 1997 #159|}. This methodology is well accepted and has been used to study the effect on the economic value of firm actions such as IT investments {|Dos Santos, 1992 #176; Im, 2001 #184|}, appointment of new chief information officers (CIOs) {|Chatterjee, 2001 #209|} corporate acquisitions {|Chatterjee, 1986 #177|}. More importantly, for our purposes the event study has been used to examine EC initiatives in several fields such as information systems (IS) {|Subramani, 2001 #185|}, economics {|Chen, 2001 #208|} and finance {|Cooper, 2001 #216|}. To the existent literature on EC value, we add a number of innovations. First and foremost, we develop theory to explain the value drivers of EC. Specifically we propose and test the idea that EC value is driven by intangiable, It complementing assets. We test this by examining the asset stucture inharent in several stratgic choices including customer type, firm type, production type, governace structure and innovativeness. Second, we examine EC initiatives based on the atomic business model used. Business models are an aspect of EC that has garnered a great deal of theoretical attention {|Aksen, 2000 #206; Timmers, 1998 #207; Weill, 2001 #205|}, but has had very little empirical examination. What empirical analysis has been done has been limited to descriptive case studies {|Weill, 2001 #205|}, but to the best of our knowledge no research has yet to measure the value derived from different EC business models. 2 DRAFT VERSION available online at http://www.ericwalden.net We also apply a methodology new to the IS literature—the long run event study—in addition to the well known short run event study. While the event study methodology, and this paper, assume efficient capital markets, we acknowledge that markets can be wrong. Market efficiency means that all of the information available about a firm is quickly incorporated into the price of the firm’s stock. Hence, the traditional short run event study measures the abnormal change in a stock price over a very short window— typically two or three days. However, in the initial phases of a new commerce method markets may not have access to high-quality information about that method because it has not yet been generated. In such a situation, the short run evaluations might be lacking. However, in the long run as investors gain more information about how the firm pursues the initiative, how customers react to the initiative, how competitors respond to the initiative and how overall perceptions of EC change the may very well revise their initial opinions. Given the new information gained over the long run investors may rebalance their portfolios thereby changing the level of abnormal returns. To the best of our knowledge a long run event study methodology has not yet been applied in the IS literature. By doing so, we not only offer greater insight into our own domain of question, but add (or perhaps subtract) validity to other short run event studies in the IS literature. THEORY DEVELOPMENT While the tangible costs of IT can be quite large—Amazon.com’s 10-K lists an eye popping $187,345,000 in computer equipment and software—they are only the “tip of the iceberg” {|Brynjolfsson, 2000 #218; Brynjolfsson, 1997 #215|}. Research carried out by Brynjolfsson and others suggest that to make computing technology productive requires sweeping organizational change {|Brynjolfsson, 2000 #218; Brynjolfsson, 1997 #215; Bresnahan, 1998 #155|}. The true gains in productivity stem not from the IT itself, but form the considerable investments in intangible, complementary assets such as new business processes, better information sharing and new distribution channels. The ratio of intangiable assets to IT assets has been estimated to be as high as 19:1 {|Brynjolfsson, 1997 #215; Brynjolfsson, 2000 #218|}, which means that Amazon.com may have over $3.5 billion worth of intagiable assets complementing its IT. David {|David, 1990 #219|} explains how this intangible, complementary investment combines with technology to produce value in a historical examination of electrical power. Steam powered shops were organized with a large boiler powering a steam turbine at one end. The turbine turned a large shaft which ran across the shop floor providing, through a system of gears and smaller shafts, motive power to equipment. Thus, the equipment had to be arranged according to power needs, with the machines requiring more power positioned closer to the shaft. Initially shops replaced the steam turbine with an electrically powered turbine, producing some minor cost savings because electrical plants were more efficient at generating power than individual boilers. However, the more important productivity gains came when shops were restructured to take advantage of the characteristics of electricity. Rather than using a central shaft electricity could be brought throughout the shop floor providing equal power at any point. This allowed firms to organize machines by process order rather than power needs. A similar illustration can be drawn using EC technologies. One of the first major groups of commercial users of the internet were universities. Originally the application 3 DRAFT VERSION available online at http://www.ericwalden.net procedure involved the student sending a letter requesting application materials, the admissions department then mailed the application materials, which the student completed and mailed back to the university. With the advent of the internet, many admissions offices would accept email requests for materials, saving one mailing. However, the same technology would pay off to a much greater degree if the whole process were streamlined so that applicants filled out their application online and the computer system automatically distributed it to the appropriate persons2. This would require significant organizational change, redefining (or maybe eliminating) the role of the admissions department. It is likely that the organizational restructuring costs would be much greater than the IT costs in theis scenerio. Our central argument is firms that are more able to make the investments necessary to develop intangible, complementary assets will be more successful. We start with the hypothesis that EC initiatives, in general, generate positive abnormal returns. EC initiatives undertaken by firms reflect the active engagement of firms to build resources and capabilities for the new medium {|Peteraf, 1993 #104|}. These initiaitves are expected to position the firms advantageously to exploit opportunities created by the growth in electronic commerce, thus creating benefits to the firm in future periods. Consistent with the original spirit of the event study methodology {|Fama, 1969 #162|}, announcements of EC initiatives convey information to markets of the willingness of an innovative, forward-looking, profit-oriented management team to make the investments in both IT and the intangible, complementary assets necessary to address growing online markets. These arguments suggest that firms announcing EC initiatives are likely to realize significant strategic and operational advantages in the future. If so, investors should react positively to EC announcements, creating a positive abnormal stock market return - a risk-adjusted return in excess of the average stock market return - around the date of the EC announcement by firms. This leads to the hypothesis that EC initiatives are associated with enhanced benefits streams in the future and consequently enhanced market valuation. We describe our overall hypothesis H1. H1: The abnormal returns attributable to announcements of EC initiatives are positive FIRM TYPE AND MARKET VALUE Firms undertaking EC initiatives fall into two basic categories. The first category is the pure play net firm, like Amazon or eBay. These firms maintain no physical presence in the traditional sense operating in space rather than place {|Weill, 2001 #205|}. The other category we term click-and-mortar, although the popular vernacular seems to now be brick-and-click. These are traditional types of firms with physical presence that also operate in a non-physical EC channel. Those firms which are best able to make the investments in organizational changes needed to complement the EC focused IT should generate significant profits in the future. The goal of a firm pursuing an EC initiative is to create intangible assets to complement IT. There are three basic starting points for the creation of these assets. One is that a 2 The University of Texas at Austin has made a move in this direction, having an online application form. 4 DRAFT VERSION available online at http://www.ericwalden.net firm may start with no intangible, complementary assets. By virtue of the fact that EC is a new phenomena, all net firms start from this point and have to build all of their assets from the ground up. Another staring point is that a firm has intangible assets in place that are actually detrimental to the creation of EC complementing, intangible assets. The resources created by firms to compete in conventional markets, in some cases, may be ill suited or even constraining in changing environments, a phenomenon described as the incumbents' curse {|Chandy, 2000 #233|} or the late mover advantage {|Shankar, 1998 #234|}. That certain components of the resource stock of the firm developed in one environment may turn into serious limitations in another parallels the observation that core rigidities often have their roots in core competencies {|Leonard-Barton, 1992 #235|}. There has been a variety of literature which has suggested that the internet changes everything, and hence completely new forms of organization must be developed to address EC {|Downes, 1998 #220|}. Net firms have significant advantages that make them particularly suited to the current EC environment. These environments are characterized by considerable volatility and are described as a parallel universe requiring radically different organizational strategies and managerial mindsets. Net firms tend to be technology-driven and have significant capabilities related to Internet technologies. Evidence suggests that they are characterized by entrepreneurial cultures and have the ability make rapid changes to their strategies to leverage and align with changes to the fluid technological and market environments {|Yoffie, 1999 #88|}. The final possibility is that a firm starts with some intangible assets that are already complementary to EC technologies. These firms already have assets in place and thus have less cost to complete the development of the proper complementary assets. We believe that click-and-mortar firms fit this description because they already have significnat IT complementing assets in place. Trust and reputation are two such intangible assets, which are certainly complementary with EC, usually possesed in greater degree by click-and-mortar firms. Moreover, click-and-mortar firms are not entirely ignorant of EC technologies. These firms have had MIS department and CIOs for decades. Many firms transformed their processes through reengineering in the 1990’s, and many click-and-mortar firms are exemplars of effective IT use. Not only is it a mistake to assume that such firms start with negative EC complementing assets, but it should be recognized that they actually have many such assets already in place. Furthermore, we should consider the likely possibility that those firms making the transition from brick-and-mortar were exactly the firms that already had a significant advantage in intangible, IT complementary assets. That is to say that the firms who were truly at a disadvantage for building those complementary assets choose not to move into the internet arena. Merrill Lynch is a good example of a firm that was slow to move into the internet arena because of its disadvantage in being able to build the intangible, IT complementary assets needed. Weill and Vitale give a good example of this in their introductory chapter {|Weill, 2001 #205|}. Another tremendous intangiable asset possessed by click-and-mortar firms is reasonable cost of capital. Click-and-mortar firms have the collateral, reputation and cash flow to secure financing at reasonable rates. The ability to secure capital is especially important in the EC context where all storefronts are global. There are tremendous costs of computing capital and software, and even greater costs of building 5 DRAFT VERSION available online at http://www.ericwalden.net the complementary, intangible assets to make the computers and software productive. Net firms, in contrast, are often financed by venture capital which may claim as much as 50% of the firm in exchange for capital. Net initial public offerings (IPOs) were notorious for the amount left on the table. For example, theglobe.com raise $28 million on its IPO price of $4.50 per share, but the price of the stock on the close first day of trading was $31.75, making the value of the IPO shares $197 million {|Tully, 1999 #222|}. Thus, to raise $28 million, theglobe.com had to sell almost $200 million worth of equity. And this trend was widespread during the timeframe of data, no less than $62 billion was left on the table in 1999 and 2000 {|Tully, 2001 #223|}. Clearly if a firm like Ford or Coca-Cola wanted to raise $28 million, they would have to sell only $28 million in equity. Clearly, this is a tremendous advantage of click-and-mortar firms over net firms. One argument in favor of net firms is that, by virtue of having no complementary, intangible assets they can more easily implement the changes necessary to develop such assets. While that is certainly true, we argue that the set of firms that choose to pursue brick-and-mortar operations in 1999 was a group that had almost the right structures in place and a head start in building intangible, complementary assets. Thus, while the net firms had an easy road to travel, it was a long road. The brick-and-mortar firms had a harder road to travel, but it is much shorter, and it was not that much harder. Because, brick-and-mortar firms have a significant head start in building IT complementary, intangible assets, we conjecture that investors will recognize this and reward them with abnormal returns. We explore the complex issue by examining the two subgroups in the data. H2: The abnormal returns attributable to EC announcements of conventional firms are different from the abnormal returns attributable to EC announcements of net firms. CUSTOMER TYPE An important distinction in the popular press and the literature is between different customer types, specifically the differences between business-to-business (B2B) and business-to-consumer (B2C) EC {|Chen, 2001 #208; Subramani, 2001 #185; Kauffman, 2001 #72|}. The tantalizing promise of B2C EC derives from the possibility of the large and rapidly growing population of web users being a market for goods and services. The concept that customers would order goods online to be delivered directly to their doors, bypassing traditional intermediaries such as retail outlets is epitomized by highly publicized firms such as Amazon.com. B2B on the other hand, taps a market estimated in to be worth trillions of dollars {|Junnarkar, 1999 #225|} and derives from a small number of firms making purchases on a massive scale. Following the theory laid out above, we conjecture that the firms which are able to develop IT complementary, intangible assets at the lowest cost will generate the most value. Firms have been using EC technologies to facilitate transactions in the form of electronic data interchange (EDI) for several decades prior to the advent of consumer transactions via the internet {|Swatman, 1992 #92; Emmelhainz, 1993 #90; Riggins, 1999 #71|}. With this long timeframe of both theory and practice of EC for B2B transactions comes a great deal of knowledge about what sorts of intangible assets are most complementary to the EC technologies and how to most efficiently implement the 6 DRAFT VERSION available online at http://www.ericwalden.net changes needed to build those assets. Thus, even for firms new to the B2B marketplace the know-how is available and the asset building should be relatively straightforward. B2C, in contrast, is an entirely new way of using EC technologies. Even firms with the commitment to building the necessary complementary assets have no easy access to the know-how about what those assets might be or how to build them. Thus, B2C firms face a relatively more difficult task of first understanding what must be done, then doing it. Furthermore, the B2C marketplace is characterized by masses of arms-length transactions making it very difficult to determine how customers perceive the firm. This contrasts with B2B marketplaces, where the smaller number of more important customers leads to closer relationships. One very large area where B2C is lacking is in transaction processing. B2C is primarily driven by credit card transactions, which charge a considerable clearing fee3. B2B firms have adopted a variety of lest costly transaction clearing processes. Shipping costs are another area in which the B2C players lack cost efficient complementary assets. Trust is another intangible asset, which is lacking in B2C markets, though organizations such as eTrust (www.etrust.com) are helping to address these issues. Simply stated for small transactions the costs to the customer of resolving improper or defective orders are often greater than the value of the transaction. With large B2B transactions the relative cost of legal or other action is much smaller. In sum, the infrastructure, processes and organizational forms constituting the intangible, IT complementing assets are much less developed for the B2C marketplace than for the B2B marketplace. To inform the debate on the payoffs to firms from B2B and B2C EC, we propose to explore the relative benefits to firms from such initiatives with the following hypothesis: H3: The abnormal returns attributable to B2B EC announcements are different from the abnormal returns attributable to B2C EC announcements. TYPE OF PRODUCT AND MARKET VALUE Several authors make a distinction between products which are digital and those that are tangible {|Negroponte, 1995 #196; Shapiro, 1999 #197; Subramani, 2001 #185|}. Tangible products are those such as books, toys and automobiles which are composed of atoms. Digital products on the other hand are those composed of bits {|Negroponte, 1995 #196|}. One of the defining characteristics of a digital product is that is has virtually zero marginal cost of production{|Shapiro, 1999 #197|}. While it is not immediately obvious what effect marginal cost should have on business value, it does seem reasonable to argue that products composed of bits would naturally lend themselves to EC {|Subramani, 2001 #185|}. However, this viewpoint is based solely on a consideration of the IT assets of an EC firm. Considering the intangible, complementary assets this may not be the case. The mass distribution of digital products has only been possible for a few years, and for the most part has been limited to distribution on inexpensive media such as disk or CD ROM. EC technologies allow for rapid, media free distribution. However, because of the newness of this capability, there is very little supporting infrastructure for selling 3 Typically Visa and Master Card charge merchants 2% of the transaction, plus a batch processing fee. The minimum charge is $0.25, which makes small transactions cost prohibitive. 7 DRAFT VERSION available online at http://www.ericwalden.net digital products. This would include legal rules, deterrent technologies and transaction systems. As the case of Napster demonstrates, there is very little legal president for addressing pure digital distribution. Certainly, we must consider the ruling against Napster as an intangible, IT complementing asset of firms like Sony or BMG. In general, enforcement efforts have focused on individuals selling bootlegged products or firms violating copy write and licensing, but have ignored free sharing of digital products {|Gopal, 2002 #226|}. Thus, the legal environment is not conducive to making profit from selling digital products via the internet4. Further, firms have been reluctant to install preventive controls such as passwords and encryption on digital products because of the trouble it causes legitimate customers {|Gopal, 2002 #226|}. Also, the actual tools and standards for such preventive controls are not well established. Real Juke Box (www.real.com) offers individual users the choice of encrypting the music that they record, so they may only be used by the original user. However, without an agreed upon encryption standard this makes such encrypted music unusable by other legitimate players on the users machine. As we touched upon in the B2B case, the transaction systems available online are effectively limited to credit cards—though some firms like PayPal (www.paypal.com) are attempting to address this problem. If the marginal cost of digital products is close to zero, then in competition the price of digital products should be close to zero {|Hurley, 1999 #227; Shapiro, 1999 #197|}. However, credit cards are very inefficient for charges close to zero. This helps explain why so many digital products are free. A perfect example is games.com, which offers a number of arcade games like Asteroids for interactive play. Millions of people have demonstrated a willingness to pay a quarter to play Asteroids, but games.com can not charge a quarter, because the minimum credit card charge is a quarter and there is a batch processing fee. Charging a quarter would actually cost games.com money and thus, they provide the games for free. In sum, while the EC technologies for distribution of digital products are easily accessible, the intangible, complementary assets are very costly to develop because of the current state of the EC environment. To begin to address these issues, we hypothesize that: H4: The abnormal returns attributable to EC announcements involving tangible goods are different from the abnormal returns attributable to EC announcements involving digital goods. GOVERNANCE STRUCTURE EC initiatives my be pursued unilaterally or in tandem with another firm. We define three different types of potential alliance partners{|Chen, 2001 #208|}—traditional, computer and pure play internet. Following the theory developed above, we conjecture that shareholder wealth effects will accrue to the governance structures which allow for the least costly building of intangible, IT complementing assets. In general, the 4 The reader may argue that one of the most successful companies of our time, Microsoft, has made billions of dollars from the sale of digital products. We would point out that such products are not sold via the internet and are almost always bundled with some sort of hardware when sold to individuals. Only where license enforcement is relatively easy—in sales to businesses—does Microsoft sell products in free form. 8 DRAFT VERSION available online at http://www.ericwalden.net coordination required to develop assets between multiple firms would be a barrier to the creation of those assets. To the degree that those intangible assets were specific or proprietary we would expect to find a number of well document difficulties arising from joint asset development {|Williamson, 1985 #121|}. Furthermore, joint ownership can lead to a number of incentive difficulties that make it difficult to develop the optimal level of intangible assets {|Grossman, 1986 #13; Hart, 1988 #5; Bakos, 1993 #3|}. There is some preliminary empirical evidence to support the idea that having greater numbers of partners reduces the shareholder return to EC initiatives {|Subramani, 2000 #186|}.To begin to address this question, we hypothesize that: H5: The abnormal returns attributable to EC initiatives undertaken unilaterally are different from the abnormal returns attributable to EC initiatives undertaken with alliance partners. INNOVATIVENESS OF INITIATIVE One factor that has repeatedly been demonstrated to generate positive abnormal returns in IT event studies is the level of innovation of the announcement {|Chatterjee, 2001 #209; Dos Santos, 1992 #176; Im, 2001 #184|}. Initiatives can be either executional/extensional or transformational. Executional/extensional initiatives are those in which the firm is streamlining or augmenting an existing process or product. Transformational initiatives are those in which the firm is developing a new business process or new product type. In terms of intangible, IT complementary assets this is a complex issue worthy of its own paper. We will touch briefly on an argument for applying the theory of complementary assets. It would seem that executional/extensional initiatives would be better able to leverage the assets already in place. However, the literature has consistently found that transformational initiatives generate significantly greater returns than executional/extensional initiatives {|Chatterjee, 2001 #209; Dos Santos, 1992 #176; Im, 2001 #184|}. This may be the case because transformational initiatives move the firm into an entirely new space without excluding the old space. Furthermore, if the marginal returns to complementary asset building investments are diminishing, then for the same level of investment the firm can build greater complementary assets in the new space than in the old space where some assets are already in place. Roughly speaking, a firm is face with the choice of working very hard to add a small amount of complementary assets to its existing IT, or working very little to add a great deal of complementary assets to an entirely different set of IT. This argument depends on the nature of the returns to asset building investment, which is an empirical question beyond the scope of this work. However, such a conjecture does help explain the stylized facts of prior event studies in the IT literature. To begin to address this question, we hypothesize that: H6: The abnormal returns attributable to transformational EC initiatives are different from the abnormal returns attributable to executional/executional EC initiatives. BUSINESS MODELS Another popular way of looking at EC initiative is through the lens of the business model {|Aksen, 2000 #206; Timmers, 1998 #207; Weill, 2001 #205|}. While there is no 9 DRAFT VERSION available online at http://www.ericwalden.net universally agreed upon definition, a business model describes how a firm does its business. There are a variety of ways to separate out different business models, but we choose to follow Weill and Vitale’s framework {|Weill, 2001 #205|}. This is a particularly appealing framework because it focuses on atomic business models of EC firms. An initiative is composed of some combination of atomic business models. The eight business models examined are described in Table 1. We present the business model analysis as an exploratory, alternative or complement to our other examination. Thus, we propose that: H7: The abnormal returns attributable to different EC business models are different from one another. OWNERSHIP In addition to business models Weill and Vitale propose several different classes of ownership, which will give rise to profits. Namely they propose that a firm may own the relationship with the customer, the customer data and the transaction. The authors propose that different business models will, in general, have ownership of different sets of these three items and that the ownership of these items will lead to profits. Without specifying the relative magnitudes the authors posit that ownership of more aspects will lead to greater profits. We examine this claim through the hypothesis that: H8: The abnormal returns attributable to each ownership type is positive. METHODOLOGY Event study methodology draws on the efficient market hypothesis that capital markets are efficient mechanisms to process information available on firms {|Fama, 1969 #162|}. The logic underlying the hypothesis is the belief that investors in capital markets process publicly available information on firm activities to assess the impact of firm activities, not just on current performance but also the performance of the firm in future periods. When additional information becomes publicly available on firm activities that might affect a firm’s present and future earnings, the stock price changes relatively rapidly to reflect the current assessment of the value of the firm. The strength of the method lies in the fact that it captures the overall assessment by a large number of investors of the discounted value of current and future firm performance attributable to individual events which is reflected in the stock price and the market value of the firm (see {|McWilliams, 1997 #159|} for a detailed review). Though financial markets may be efficient information processors, even an efficient processor is limited by the availability of information. EC is a new phenomena, and even now it is not clear exactly what form successful EC companies will take. In evaluating the value of EC events the market is limited by the fact that there is little information available for decision making. Thus, short run event studies with windows of a one to three days may not result in accurate measures of capital markets true valuations for EC initiatives. Moreover, all of the information about an event is not likely to be available in one announcement. Over time, more information about the success or failure will become available to the market, and expectations about the event will be adjusted. For example, a firm may announce the release of a new EC technology, like Amazon.com’s one click shopping patent. However, the entire value cannot be known until customers 10 DRAFT VERSION available online at http://www.ericwalden.net adopt (or fail to adopt) the technology, as other firms develop competing technologies, and as courts choose how to enforce intellectual property rights involved in the technology. As information about these factors becomes available, investors will revise their expectations, and reward firms with good events with additional abnormal return, while punishing those with bad events with negative abnormal returns. For the EC context in particular, markets will also learn meta-information about the probably success of an industry, over time. Thus, investors are likely to revise estimates of EC value across the board over time, as it becomes more clear what the likely industrial profits will be. This effect has been documented by several IS researchers, who have found that the market reaction to identical events has changed over time {|Im, 2001 #184|}. Anecdotally, we would expect that the market’s expectation of EC has declined over time, as it is widely believed that the market was in a bubble. Thus, we might expect to find negative abnormal returns for all EC firms. However, these returns should be significantly less negative for firms having good events. Recently, the methodology has enjoyed great success in the IS literature. One of the earliest works examined the impacts of innovative verses incremental investments in IT, and finds that financial markets react positively only to innovative announcements {|Dos Santos, 1992 #176|}. This finding has been expanded using the original data and over 100 new data points {|Im, 2001 #184|}. The new data suggest that not only do markets react, by they react more in the later part of the data set, and they react more for smaller firms. IS researchers have also used event studies to understand the impacts of EC announcements on firm value. Subramani and Walden have found that capital markets react positively to EC announcements in general, but that the reactions are greater for announcements focused on physical products than on those focuses on digital products, and greater for those announcements focused on B2C initiative than on B2B initiatives {|Subramani, 2001 #185|}. The authors further explore the value drivers of the B2B announcements and find that market reaction is only significant for internet firms undertaking B2B initiatives {|Subramani, 2000 #186|}. While the event study methodology has been widely used by IS researchers none have yet taken advantage of the relatively new innovations in financial econometrics to look at the long run impacts of events. <<NEED TO ADD OTHER STUDY FROM MISQ>> The event study methodology provides management researchers a powerful technique to explore the strength of the link between managerial actions and the creation of value for the firm {|McWilliams, 1997 #159|}. This methodology is well accepted and has been used in a variety of management research to study the effect on the economic value of firm actions such as IT investments {|Dos Santos, 1992 #176|}, corporate acquisitions {|Chatterjee, 1986 #177|}, CEO succession {|Davidson, 1993 #178|}, joint venture formations {|Koh, 1991 #181|}, celebrity endorsements {|Agrawal, 1995 #179|} and new product introductions {|Chaney, 1991 #180|}. DATA Following Subramani and Walden, we define the event as a public announcement of a firm’s EC initiative in the media. We collected the data from a full text search of company announcements related to EC in the period between January 1, 1999 to December 31, 1999, using two leading news sources: PR Newswire, and Business Wire. Based on an examination of several candidate announcements, we used the online search 11 DRAFT VERSION available online at http://www.ericwalden.net features of Lexis/Nexis to search for announcements containing the words launch or announce within the same sentence as the words online or commerce5, and .com. The search yielded 2170 potential announcements. CODING Coding involved a full text examination of the announcements. To consider an announcement as an event it had to be a major initiative focused on new EC product offerings or new EC processes implementation. Thus, we excluded several categories of announcements from consideration. We excluded announcements of customer acquisition as not being a major new initiative. Rather we feel, customer acquisition, while important, is a day-to-day business activity, that all firms do on a regular basis. We code such an announcement as an event if the method of customer acquisition was a new EC initiative (e.g. firm xyz signs up first online customer). We also code the announcement as an event for the customer if the product was used for a new EC initiative (e.g. firm xyz acquires firm abc as a new customer for xyz’s online procurement auctions). Another excluded category is minor promotions. EC firms often announce give-aways or other devices to attract customers. These sorts of promotions are not coded. We also do not code earnings announcements as events, though firms frequently issue a press release every quarter detailing the company financials. We do not code website redesign, unless the redesign is part of some larger initiative. Furthermore, we do not code mergers and acquisitions as events. Though these may be important EC initiatives, the value of EC is hopelessly entangled with the value of the merger process. Out of the 2170 potential announcements 881 were coded as announcements of EC initiatives, while the others were earnings announcements, management change announcements, merger announcements, announcements of promotions, or announcements of things other than actual EC initiatives. After removing those firms with not enough data for analysis the total came to 616. Following Subramani and Walden {|Subramani, 2001 #185|} we removed firms with an average share price of less than one dollar (14 firms) and firms with an average trading volume of less than 50,000 shares per day (101 firms), leaving 501 initiatives for examination. FIRM TYPE The first coding task involves separating firms into internet-only businesses and clickand-mortar businesses. We follow we followed the classification system devised for the Dow Jones Internet Index that considers net firms to be those deriving more than 50% of revenue from Internet activities {|Subramani, 2001 #185|}. Of the 501 announcements 145 were coded as involving internet-only firms and 356 were coded as involving clickand-mortar firms. The coding of announcements by type was based on detailed analysis of the full text of the announcement. The announcements were coded as falling into three categories. 5 We observed considerable variation in the wording of announcements related to electronic commerce. Using the word “commerce” captured the most common variants: EC, e commerce, and electronic commerce. 12 DRAFT VERSION available online at http://www.ericwalden.net TYPE OF VALUE ADDING ACTIVITY Other than firm type, which is a characteristic of a firm, the rest of our hypothesis revolve around the nature of the activity undertaken by a firm. Thus, the unit of analysis is the initiative. This allows us to draw general conclusions that can be applied to firms regardless of their other characteristics. The focus on initiatives requires us to look only at the characteristics of the initiative independent of the characteristics of the firm. To this end we define the EC value adding activity. This allows us to focus on the EC part of the announcement of a firm that has a variety of value adding activities. For example, Yahoo sells advertising to industrial clients. These agreements are sold by a sales force that negotiates with clients, in the same way vendors in souks, bazaars, and flea markets have been doing it for thousands of years. At the same time Yahoo provides search and content aggregation value adding services to end consumers in an EC mode. Thus, the initiatives undertaken by Yahoo that will be of interest to us, are the ones focusing on the EC part of the firm. We would not be interested in announcements of the acquisition by Yahoo of a new industrial client (unless that acquisition was done electronically). We will be interested in two questions about the EC value adding activity detailed in an announcement—for whom does the initiative add value and how is the value produced. We turn now to the latter issue. PRODUCTION TYPE One of the main arguments about EC, and the new economy in general, concerns the means of production of products. Many arguments are advanced that suggests that producing digital products has zero marginal cost, which is a large departure from traditional economic analysis {|Bakos, 1991 #100; Choi, 1997 #200; Kauffman, 2001 #72; Shapiro, 1999 #197|}. We use this ideal of digital production being zero marginal cost production, for the coding of announcements as digital or tangible. Specifically, an announcement was coded as being digital if the amount of EC value added could be increased significantly without significant increases in costs. Conversely, if an increase in EC value added entailed a similar increase in costs the announcement was considered as tangible. For example, initiatives where the goods or services were made available online for use or downloaded for use as involving digital products. For instance, announcements of firms offering products or services such as rock concerts on demand, online trading, signup for telecom services and purchase of insurance services were coded as involving digital products. Similarly, announcements by firms of online forums for exchange or trade were coded as involving digital products. Announcements of online availability of products such as sports merchandize or books were classified as involving tangible goods. Of the 501 events, 214 events were coded as involving tangible goods and 287 as involving digital goods. CUSTOMER TYPE To develop the coding scheme for the orientation of the announcement we use a modified version of the coding scheme of {|Subramani, 2001 #185|}. Their coding scheme was based on the revenue model of the firms involved in the announcement. Announcements were coded as B2B if the revenue source of the associated firm was primarily from institutional customers. An announcement was coded as B2C if the 13 DRAFT VERSION available online at http://www.ericwalden.net revenue source was primarily from end consumers. This scheme is based on the firm in question, whereas our focus is on the initiative. Thus, we use the concept of EC value added rather than revenue. We consider a firm to make a B2B announcement if the EC value added is primarily value for another firm. In contrast a firm is considered to be making a B2C announcement when the EC value added is value added for an end consumer6. When there were multiple firms mentioned in announcements, the decision about whether the event was B2B or B2C was made for each firm involved. For example, the announcement "Reel.com Opens Virtual Video Store Front on America Online" was coded as B2B for AOL and as B2C for Reel.com. The logic was that the EC value added by AOL was the one-time fee paid by Reel.com as well as periodic payments based on traffic channeled to Reel.com. The EC value added by Reel.com is from video sales to individuals channeled to the company website. On the other hand the announcement "Digital River Adds Kmart Corporation to Growing Network of Online Software Dealers" was coded as B2C for Kmart and as B2B for Digital River as Digital River would receive periodic consolidated payments based on individual purchases by customers at Kmart's online site. On the other hand, Kmart was delivering EC value added for the end consumers shopping it’s site. Instances where the revenue arrangements were unclear were dropped. Of the 501 announcements, 162 were coded as B2B and 339 coded as B2C. GOVERNANCE TYPE Coding for unilateral vs. alliance involved an examination of the text of the announcement. An announcement was coded as unilateral if it involved only one firm, or if it involved tow firms, but one firm was simply implementing the other firm’s technology. The second condition was added because frequently software firms will announce new customer acquisition in announcements that read, firm ABC has been added to firm XYZ’s list of clients implementing software PQR. While we consider this an EC initiative for ABC, we do not consider this an alliance. Alliances were coded by the type of partner {|Chen, 2001 #208|}. If the alliance was with an Internet firm it was coded as pure-play. If the alliance was with a large and wellrecognized computer industry leader (e.g. IBM, Microsoft, SAP) then the alliance was coded as computer. If the alliance was with a click-and-mortar or brick-and-mortar firm other than a computer industry firm, it was coded as traditional. This classification allows for not only new economy EC firms and old economy traditional firms, but also for middle age information economy firms in the computer industry. INNOVATIVENESS Following prior literature we coded initiatives as either being innovative or noninnovative {|Im, 2001 #184; Dos Santos, 1992 #176|}. innovative announcements are those where new technology is being developed or applied, while non- innovative 6 Note, we do not address the question of how much of the value gets to the other firm or the end consumer, we are only concerned that the EC value added would primarily benefit one or the other types of entities. How the value is split is a topic for another day, and well beyond the scope of this work. 14 DRAFT VERSION available online at http://www.ericwalden.net initiatives are marketing announcements, follow-up announcements about a technology, or announcements of use of well established technologies. ATOMIC BUSINESS MODEL There are a whole range of business models for electronic commerce being put forward by a variety researchers {|Aksen, 2000 #206; Timmers, 1998 #207; Weill, 2001 #205|}. We adopt the categorization developed by Weill and Vitale. This categorization looks at atomic business models as building blocks for EC initiatives. The authors develop eight atomic business models that can be combined in various ways to develop a much larger number of EC initiatives. The advantage of focusing on the atomic models is that they are stable building blocks over time and that there are a small number of them to consider. Weill and Vitale liken these atomic business models to atoms, while the full scale business model of a firm is likened to a chemical compound. Thus, each atomic business model can be combined with other such models to accurately represent the overall business model of a given firm. These eight atomic business models are listed in Table 1. As these are atomic models, and our analysis focuses on initiatives, each initiative may contain multiple atomic business models. Table 1: Atomic Business Models of Electronic Commerce from Weill and Vitale (2001) ATOMIC BUSINESS MODEL DESCRIPTION OF MODEL Content Provider Provides content via intermediaries. Direct to Customer Provides products directly to consumers. Full Service Provider Provides a full range of services (aggregation) directly and via complemetors. Focus on owning customer relation. Intermediary Brings together buyers and sellers by concentrating information. Shared Infrastructure Brings together multiple competitors to cooperate by sharing common IT infrastructure. Value Net Integrator Coordinates activities across the value net by gathering, synthesizing, and distributing information. Virtual Community Creates and facilitates an online community of people with a common interest. Whole of Enterprise Provides a firm-wide single point of contact, consolidating all services provided by a large multiunit organization 15 DRAFT VERSION available online at http://www.ericwalden.net OWNERSHIP OF VALUE ADDING CUSTOMER ASSETS In addition to the atomic business models Weill and Vitale propose that firms may own different value adding customer assets. The three assets that they propose are the customer relationship, the customer data and the customer transaction. While, they classify ownership of each of the assets as being a function of the chosen business model, we seek further empirical validation and code ownership separately from the business model type. For a given initiative, a firm may own none of these assets, some of these assets, or all of the assets. DATA DEMOGRAPHICS Table 2, below, details the number of initiatives coded into each of the subcategories. Those marked with an “*” sum to the total number of announcements. The other categories sum to greater than the total number of initiatives, because an initiative may entail more than one sub-category. For example, an initiative can entail a firm owning both the customer data and the customer relation. For the short run part of the study we required firms to have 251 trading days prior to the event plus one day beyond the event. For the long run part of the study we required firms to have one trading day prior to the event and to still be trading one year after the event. For the long run we also winsorise the returns as described below. The final column looks at the long run returns with only firms who undertook a single initiative in the one year time period. If a firm undertook two initiatives during 1999, then we have no way of separating the abnormal returns to the different events. This is a particular problem in long run event studies. One way to deal with it is to look only at the first event {|Mitchell, 2000 #192|}, but because it is relatively prevalent in our sample we choose to remove all firms with multiple events and present both the full sample and the truncated sample results. Note that while there is a great deal of overlap the long run and short run datasets are not subsets of one another. Thus, we create a final dataset which contains only the unique initiatives for which we can evaluate both the short run and the long run returns. This is called the compatible unique dataset. 16 DRAFT VERSION available online at http://www.ericwalden.net Table 2: Number of announcements coded in each category * * * * Short Run Variable (n=501) Content Provider 23 Direct to Customer 329 Full Service Provider 15 Intermediary 124 Shared Infrastructure 39 Value Net Integrator 1 Virtual Community 21 Whole of Enterprise 2 Relation 443 Data 437 Transaction 308 Net 145 Click-and-Mortar 356 B2B 162 B2C 339 Digital 287 Tangible 214 Unilateral 275 Pure-play Ally 167 Computer Ally 20 Traditional Ally 57 Executional / Extensional 375 Transformational 126 Long Run (n=601) 31 366 18 176 45 2 26 4 528 524 346 243 358 191 410 356 244 315 191 27 84 452 149 Long Run Unique (n=244) 11 168 6 50 17 1 9 2 215 218 170 66 178 88 156 113 131 147 66 7 28 181 11 Compatible Unique (n=192) 8 136 5 35 13 1 7 1 168 171 138 31 161 72 120 84 108 120 52 7 17 144 48 PRICES VOLUMES AND RETURNS Table 3: Demographics for Datasets Short Run Variable Before1 Event After1 ABRET R2 Price Volume 17 N Mean Std Dev Minimum Maximum 501 0.6% 9.7% -74.4% 131.8% 501 2.0% 13.5% -37.8% 169.1% 501 1.4% 14.5% -87.1% 138.1% 451 -10.2% 84.0% -105.4% 591.1% 501 0.124 0.135 0.000 0.810 501 $29.24 $23.82 $1.15 $118.56 501 2618117 5684251 50340 28083361 DRAFT VERSION available online at http://www.ericwalden.net Long Run Variable Before1 Event After1 ABRET R2 Price Volume N Mean Std Dev Minimum Maximum 601 0.6% 8.5% -74.4% 131.8% 601 1.6% 11.8% -37.8% 169.1% 447 1.4% 14.3% -87.1% 138.1% 601 -27.0% 67.0% -105.4% 294.8% 601 0.097 0.133 0.000 0.810 601 $27.56 $21.80 $1.15 $107.66 601 2243990 5252828 50340 28083361 Variable Before1 Event After1 ABRET R2 Price Volume N Mean Std Dev Minimum Maximum 244 1.0% 11.1% -29.0% 131.8% 244 2.6% 15.5% -33.5% 169.1% 192 2.5% 17.2% -39.8% 138.1% 244 -13.3% 80.3% -105.4% 294.8% 244 0.067 0.102 0.000 0.549 244 $21.48 $17.37 $1.15 $107.66 244 663240 1109213 50340 10683254 Variable Before1 Event After1 ABRET R2 Price Volume N Mean Std Dev Minimum Maximum 192 1.3% 12.6% -29.0% 131.8% 192 3.3% 17.5% -33.5% 169.1% 192 2.5% 17.2% -39.8% 138.1% 192 -3.9% 78.8% -100.8% 285.0% 192 0.085 0.107 0.000 0.549 192 $22.16 $18.25 $1.15 $107.66 192 708080 1221720 50340 10683254 Long Run Unique Compatible Unique DATA ANALYSIS SHORT RUN EVENT STUDY The methodology for the short run event study is relatively well documented in the IS literature {|Dos Santos, 1992 #176; Im, 2001 #184; McWilliams, 1997 #159; Subramani, 2001 #185; Chatterjee, 2001 #209|}. A brief description follows. First we estimate a market model for each individual firm using 250 trading days of returns regressed against a market index7. We then calculate abnormal returns (ARs) for each day of a three day window (the day before the announcement, the day of the announcement, and the day after the announcement) as: ARshort Rs ,t ˆ s ˆ s Rm,t . (1) The term in parenthesis is the expected return on the stock where α and β are estimates from the day 251 to day 2 before the event. The R terms are returns with the subscript t 7 In this case we use the S&P 500, which is the standard index. Other studies have found very little difference between the NASDAQ, the Dow Industrials and the S&P 500 {|Subramani, 1999 #70; Subramani, 2001 #185|}. 18 DRAFT VERSION available online at http://www.ericwalden.net denoting time, the subscript s refers to a specific stock and the subscript m refers to the market return. The cumulative abnormal return (CAR) is the summation of abnormal returns. The value of interest is the CAR for the three day event window starting on the day before the event and ending on the day after the event {|Rajgopal, 2000 #229; Chatterjee, 2001 #209|}. Some authors use only a two day window consisting of the day before and the day after {|Dos Santos, 1992 #176; Im, 2001 #184|}. We use this value because when the announcements are made late in the day, or after market close, the market may not be fully able to adjust in the same day. We also look one day before to capture any pre announcement drift {|McWilliams, 1997 #159|}. TEST STATISTICS There are two candidate test statistics to analyze the significance of CAR. The first and simplest is to use the standard formula to estimate the mean and standard deviations of all the CARs in the sample. This yields 1 N CAR CARs , (2) N s 1 and 2 1 N var( CAR) CARs CAR . (3) N 1 s 1 The test statistic is CAR (4) t ~ t( , df N 1) . var( CAR) However, as CAR is the sum of several independent ARs, and the distribution of those ARs can be estimated from the prediction window, the variance of CAR can be estimate from the variance of AR. In this case the variance is: 1 N 1 var (CAR) (5) var( ARs ,t ) , N 1 s 1 t 1 where, 2 Rm, t R m 1 2 , var( ARs , t ) S s 1 P (6) P 2 Rm ,i R m i 1 and 2 1 P ARshort, s ARshort . (7) P 1 s 1 In this case P is the number of days used to predict the market model in (1) or 250. Thus, to calculate the variance, we find the variance of the abnormal return in the prediction interval, then use (6) to calculate the standard error of prediction. This takes into account not only the variance of abnormal returns, but also the variance of the predicted value of the market model. In other words, (6) recognizes that α and β are estimates of the true parameter values and thus the predicted return has variance arising S s2 19 DRAFT VERSION available online at http://www.ericwalden.net not only from the random nature of returns, but also from the inaccuracy of using estimates rather than true parameters for calculation of AR. The variance calculation in (5) is thus preferred to the variance calculation in (3), because the calculation in (3) ignores the additional information available in the prediction interval. More simply stated, we have 253 observations of AR for each event. The variance in (5) uses all 253 observations while (3) essentially drops 250 of them. Furthermore, (5) takes into account the prediction accuracy of the model used to estimate returns while (3) does not. In the limit, (3) and (5) should converge, but in finite samples (5) is preferred because it uses all of the information and uses it properly. Note also that (3) is the form calculated by regression analysis, thus applying regression to analyze the differences will result in an improper test statistic. LONG RUN EVENT STUDY To analyze long run market reactions to events requires application of new techniques from financial economics specifically designed and tested to evaluate long run returns given the many stylized facts about stock price performance {|Cowan, 2001 #187; Barber, 1997 #189; Kothari, 1997 #188|}. These techniques make use of the buy-andhold abnormal return (BHAR) in contrast to the use of cumulative abnormal return (CAR) used in short run event studies (see {|McWilliams, 1997 #159; Subramani, 2001 #185|} for a detailed explanation of CAR). An abnormal return for firm i at time t is defined as: ARit Rit ERit . (8) CAR then is T CAR ARit . (9) t 1 In contrast, BHAR is BHAR 1 Rit 1 ERit . T T t 1 t 1 (10) CAR then is simply the day by day sum of abnormal returns and does not represent the actual return over the holding period. BHAR on the other hand is actual return over the entire holding period minus the expected return over the holding period. To illustrate, consider a firm that has an expected return of zero over a two day period. On the first day after the event the stock price increases by 100%, and on the second day the stock price decreases by 50%. In this case the CAR is 50%, but the BHAR is 0%. Thus BHAR is the better measure because the stock price is identical at the end of the two day period. Barber and Lyon document this discrepancy between CAR and BHAR in the general population of stocks, showing that CAR is generally higher than BHAR at low levels of return {|Barber, 1997 #189|}. This can be particularly problematic for EC research as firms can experience returns as high as 900% in one day{|Subramani, 2000 #190|}, which means that no drop in stock price could result in negative CAR for at least two trading weeks. The discrepancy between CAR and BHAR is smaller at shorter time periods. In fact, the two measures are identical for a one period return, and thus CAR is a good measure 20 DRAFT VERSION available online at http://www.ericwalden.net for typical short run studies. However, the difference becomes more pronounced as the number of periods under consideration increases. This, combined with the fact that the problem is greater for more volatile stocks, means that BHAR is a much more appropriate measure for a long run study of EC value. ISSUES IN LONG RUN EVENT STUDIES One major concern in using BHAR rather than CAR is that BHARs are more skewed than CARs. It is well known that stock returns are skewed {|Cowan, 2001 #187|}. Conversely, by the central limit theorem, the returns of a portfolio approach a normal distribution as more firms are added. Subtracting the skewed firm return from the (approximately) normal portfolio return yields a skewed distribution, which makes classical t or z tests mis-specified. In fact, even non-parametric tests may be misspecified8. The impact of this skewness bias declines with sample size, but increases with the length of the return horizon. A final, important source of potential measurement difficulty is in the construction of the test statistic. In general, the test statistic is the estimated mean of the returns divided by the estimated standard error of that mean. The standard error is typically calculated assuming that observations are independent, so that the variance of each observation can be summed. However, this is not likely to be a good assumption, as returns are cross sectionally correlated in time {|Fama, 1998 #194; Brown, 1985 #160; Mitchell, 2000 #192; Cowan, 2001 #187|}. By assuming independence when returns are actually correlated, the estimated variance is too small, leading to over rejection of the null hypothesis. Research has found that “[i]t is common for t statistics to fall from over 6.0 to less than 1.5 after accounting for cross-correlations {|Mitchell, 2000 #192|} p 291.” We expect that this will be a particularly important factor in our work, because the market sentiment, regarding EC firms, seemed to vary considerably from month to month. METHODOLOGICAL CORRECTIONS The skewness bias can be corrected for in two ways. The first is winsorization. Winsorization corrects for skewness by limiting how far from the mean an extreme observation is allowed to be. Observations beyond some threshold are set equal to the threshold, effectively weighting them less than observations below the threshold (see {|Cowan, 2001 #187|} for a detailed description). This mitigates the impact of the skewness so that test statistics converge more quickly to hypothesized distributions. The second method to control for skewness bias is to choose a reference portfolio to be a single stock {|Barber, 1997 #189|}. This controls for skewness because the reference stock is also drawn from a skewed distribution. Subtracting the event firms’ skewed return from a reference firm’s similarly skewed return produces a symmetrical 8 A typical test is to count the number of positive and negative returns with the idea that if there is no effect then the positive and negative returns will be spilt 50-50 (see for example {|Karpoff, 1994 #191; Subramani, 2000 #190|}). If the distributions are skewed then it is not the case that a zero mean would lead to a 50-50 split on positive and negative returns, because the 50-50 split requires the median and mean to be identical. 21 DRAFT VERSION available online at http://www.ericwalden.net distribution, which converges more quickly to normal, and which reduces bias toward over rejection in either tail. These ideas are illustrated in Figure 1. Panel A shows a hypothetical, skewed distribution of observed returns. Panel B represents the abnormal return, where the expected return is based on a reference portfolio. Because, the reference portfolio is the average of several stock’s returns it is assumed to be normally distributed by the central limit theorem. The abnormal returns in this case will be the difference between a normal and a skewed variable, and will thus be skewed as the panel shows. The final panel, C, illustrates the reference firm abnormal return. In this case, the abnormal return is calculated as the difference of two (identical) skewed distributions. This results in a symmetrical distribution. Note that in all three cases the mean of the distribution is zero. Figure 1: Distributions of Returns and Hypothetic Abnormal Returns Event Firm - Reference Firm Mean = 0 Mean = 0 -10% Frequency Frequency Event Firm - Portfolio Mean = 0 Frequency Event Firm Returns -5% 0% 5% 10% Observed Return Panel A: Event firm hypothetic return distribution -10% -5% 0% 5% 10% -10% -5% 0% 5% 10% Observed Abnormal Return Observed Abnormal Return Panel B: Abnormal returns as observed returns minus reference portfolio with normal distribution Panel C: Abnormal returns as observed returns minus reference firm return. Reference firm comes from same distribution as event firm Barber and Lyon {|Barber, 1997 #189|} find that using a single stock as the reference portfolio produces good results. However, they only examine samples of size 200. Cowan and Sargent {|Cowan, 2001 #187|} compare the winsorization procedure to the single firm procedure and find that the single firm procedure is very sensitive to number of events in the sample. While they do find good results in samples of size 200, they find relatively poor performance in samples of size 1000, while they find that winsorization produces good results regardless of sample size. Because we use samples of different sizes, we choose to apply winsorization to correct for skewness, and follow Cowan and Sargent by removing observations beyond +/- three standard deviations. RESULTS SHORT RUN RETURNS The results of the initiative characteristics (see Figure 2) are consistent with the theory that greater returns will accrue to those firms most easily able to build intangible, IT complementing investments. The absolute return to B2B is larger than the return to B2C, 22 DRAFT VERSION available online at http://www.ericwalden.net and only the B2B return is significant. Production type follows a similar theme with returns to tangible production being significant, and greater than the non-significant returns to digital production. Returns to click-and-mortar firms about twice as great as returns to pure-play net firms, and the net firm return is non-significant while the clickand-mortar return is significant. Unilateral initiatives produced positive and significant returns, while none of the alliance choices produced significant results, and both pureplay and computer alliances actually produced negative returns. Lastly, transformational initiatives produced positive and significant results, while extensional/executional initiatives produced smaller non-significant results. This accords well with prior research, which has almost unanimously found transformational initiatives to be an important factor in IT deployment. Figure 2: Short Run Returns to EC Initiatives (CAR for three day window) 4.0% 3.0% 2.5% 3.0% 1.6% 2.0% 1.0% 3.0% 2.3% 0.8% 0.9% 0.8% 0.7% 0.8% 0.0% -1.0% -0.5% -1.0% EXECUT TRANSFOR TRAD_AL C_AL PP_AL UNI MORTAR NET TANGIBLE DIGITAL B2B B2C -2.0% The returns to business models and customer asset ownership show interesting results as well (see Figure 3). Each of the ownership types lead to positive and significant return as put forward by Weill and Vitale {|Weill, 2001 #205|}. However the only business models which result in significant returns are the direct to customer model and the content provider model. The content provider model results in a negative 5.2% return. Full details of the analysis for both the business model and initiative characteristics are presented in Table 4. 23 DRAFT VERSION available online at http://www.ericwalden.net Figure 3: Short run Return to Business Models (CAR of three day window) 3.7% 3.2% 2.5% 2.7% 2.3% 1.6% 1.7% 0.2% -2.0% -2.9% TRANS DATA REL WOE VC VNI SI INTER FSP D2C -5.2% CP 5.0% 4.0% 3.0% 2.0% 1.0% 0.0% -1.0% -2.0% -3.0% -4.0% -5.0% -6.0% Table 4: Impacts of Initiative Characteristics for Short Run Data Whole Sample (n = 501) -1 0 1 Mean Return 0.6% 2.0% 1.4% Std 0.3% 0.4% 0.5% P-Val Customer Type B2C (n = 339) 0.061* 0.000*** 0.012** -1 0 1 Mean Return 0.3% 1.5% 0.8% -1 0 1 Mean Return 1.1% 2.9% 2.5% Std 0.3% 0.5% 0.6% P-Val 0.359 0.003*** 0.176 Business Models CP (n = 23) -1 0 1 Mean Return -0.9% -2.3% -5.2% Std P-Val 1.5% 2.1% 2.6% 0.543 0.285 0.052* B2B (n = 162) Std 0.6% 0.9% 1.1% P-Val 0.073* 0.001*** 0.023** Production Type D2C (n = 329) -1 0 1 Mean Return 1.3% 3.0% 2.5% Std P-Val -1 Mean Return -0.9% 1.7% 0 2.7% 1 3.2% -1 0 Mean Return -0.6% 0.7% FSP (n = 15) Std -1 0 1 Mean Return 0.1% 1.1% 0.7% 0.609 -1 Mean Return 1.2% 0.4% 0.004*** 2.4% 0.273 0 3.0% 0.6% 0.000*** 2.9% 0.293 1 2.3% 0.7% 0.002*** Std P-Val 0.7% 1.1% 0.429 0.488 0.3% 0.5% 0.6% P-Val DIGITAL (n = 287) 0.000*** 0.000*** 0.000*** TANGIBLE (n = 214) Std P-Val 0.4% 0.6% 0.8% 0.821 0.070* 0.382 Std P-Val Firm Type INTER (n = 124) 24 NET (n = 145) -1 0 Mean Return 0.3% 1.6% Std P-Val 0.6% 0.9% 0.626 0.075* DRAFT VERSION 1 0.2% -1 0 1 Mean Return 0.2% 1.5% 2.3% SI (n = 39) available online at http://www.ericwalden.net 1.3% 0.857 Std P-Val 1.0% 1.4% 1.7% 0.832 0.271 0.184 1 0.8% -1 0 1 Mean Return 0.7% 2.1% 1.6% MORTAR (n = 356) 1.1% 0.453 Std P-Val 0.4% 0.5% 0.6% 0.052* 0.000*** 0.011** Governance Type VNI (n = 1) -1 0 1 Mean Return 0.6% 0.4% -2.9% -1 0 1 Mean Return -0.7% -1.1% -2.0% -1 0 1 Mean Return 1.9% 2.8% 3.7% VC (n = 21) WOE (n = 2) Std 5.4% 7.7% 9.4% P-Val UNI (n = 275) NA NA NA Std P-Val 1.3% 1.8% 2.2% 0.610 0.539 0.386 Std P-Val 1.6% 2.3% 2.8% 0.452 0.443 0.417 -1 0 1 Mean Return 1.0% 3.2% 3.0% -1 0 1 Mean Return -0.5% 0.1% -1.0% -1 0 1 Mean Return 1.1% -1.0% -0.5% -1 0 1 Mean Return 1.1% 0.9% 0.9% PP_AL (n = 167) C_AL (n = 20) Std 0.5% 0.6% 0.8% P-Val 0.036** 0.000*** 0.000*** Std P-Val 0.5% 0.7% 0.8% 0.338 0.873 0.218 Std P-Val 1.3% 1.9% 2.3% 0.411 0.623 0.817 Std P-Val 0.8% 1.1% 1.3% 0.173 0.440 0.499 Ownership REL (n = 443) -1 0 1 Mean Return 0.6% 2.1% 1.6% -1 0 1 Mean Return 0.6% 2.2% 1.7% Std 0.3% 0.5% 0.6% P-Val 0.066* 0.000*** 0.005*** TRAD_AL (n = 57) Innovativeness DATA (n = 437) Std 0.3% 0.5% 0.6% P-Val TRANSFOR (n = 126) 0.088* -1 0.000*** 0 0.004*** 1 Mean Return 0.3% 2.5% 3.0% Std 0.6% 0.8% 1.0% P-Val 0.639 0.003*** 0.004*** TRANS (n = 308) Mean Std P-Val EXECUT Mean Std P-Val Return (n = 375) Return -1 1.3% 0.4% 0.000*** -1 0.7% 0.4% 0.061* 0 3.2% 0.5% 0.000*** 0 1.8% 0.5% 0.001*** 1 2.7% 0.6% 0.000*** 1 0.8% 0.6% 0.198 * represents significance at the 10% level, ** at the 5% level and *** at the 1% level. All tests are two tailed. While all of the initiative characteristics are mutually exclusive neither the business models nor the customer asset ownership structures are exclusive. A single initiative can consist of multiple business models and multiple ownership of customer assets. In general, firm initiatives used one business model, and occasionally two, with the maximum being four. However, initiatives often entailed multiple customer asset 25 DRAFT VERSION available online at http://www.ericwalden.net ownerships, and ownership of all three customer assets was not uncommon. This leads to a problem in analyzing each model in isolation as the returns attributable to an initiative are not properly divided among the business models used in the initiative. For example, an initiative in which the firm owns all three customer assets would inappropriately attribute the entire return to each of the assets. At the extreme, if all initiatives resulted in ownership of all customer assets then each asset would appear to have the overall mean return, which in this case, would be a positive and significant return. To address this issue and properly partition the return among the non exclusive models and asset ownerships, we regress the return from each initiative against all of the business model and customer asset ownership dummy variables. This resulted in the coefficients estimated in Table 5. Table 5: Short Run Joint Analysis of Business Model Impacts (Dependant Variable CAR Day +1) F 1.150 Intercept CP D2C FSP INTER SI VNI VC WOE REL DATA TRANS P-val 0.320 Return -2.9% -2.8% 3.8% 2.1% 1.9% 3.6% 0.0% -0.8% 2.9% -0.1% 0.0% 1.9% R2 0.025 Std 3.4% 4.0% 2.6% 3.8% 2.5% 3.2% 14.7% 3.6% 10.3% 2.3% 2.3% 1.8% Adjusted R2 0.003 P-Val 0.398 0.485 0.142 0.577 0.434 0.266 0.997 0.834 0.781 0.983 0.986 0.303 These results seem to indicate that there are no significant effects of either business models or customer asset ownerships. Not only are none of the coefficients significant, but the F test indicates that we do not have a regression. We must note at this point that, while regression will properly estimate the coefficients, it will mis-estimate the standard errors as discussed above. However, even accepting slightly mis-specified standard error the joint analysis does not show much promise or offer additional insight. LONG RUN RETURNS The time period in question, 1999-2000, represents a quite unpleasant time to invest in EC related stocks. In all cases the one year returns to EC initiatives were negative and significant. Overall the 601 initiatives in the long run data set averaged a –27% BHAR. This makes interpretation difficult, as the goal is to find initiatives that resulted in less negative returns. However, judging based on this criteria, we find that the results still support the theory laid out above. Figure 4 illustrates this. B2B one year BHAR were less negative than B2B BHAR, and were not significant at the 1% level. Click-andmortar firms experienced less of a drop than pure-play net firms. Unilateral action resulted in less negative BHAR than any type of alliance and Transformational initiatives were less negative than executional/extensional initiatives. The only difference between 26 DRAFT VERSION available online at http://www.ericwalden.net the relative values of short and long run returns was that in the long run there was virtually no difference between digital and tangible initiatives. Figure 4: Long Run BHAR EXECUT TRANSFOR TRAD_AL C_AL PP_AL UNI MORTAR NET TANGIBLE DIGITAL B2B B2C 0.0% -5.0% -10.0% -15.0% -14.4% -20.0% -16.4% -17.7% -20.1% -25.0% -30.0% -27.2%-26.7% -30.1% -35.0% -31.7%-30.8% -32.9% -40.0% -40.5% -45.0% -42.7% -50.0% The results for the business model analysis were also similar in the short and long run. As illustrates, content providers did relatively worse than other business models and ownership of the transaction resulted in less absolute negative BHAR. The numeric results are presented in . Figure 5: Long Run BHAR for Business Models 0.0% -10.0% -20.0% -18.5% -19.4% -22.1% -29.4% -30.0% -40.0% -50.0% -60.0% -22.8% -26.5%-24.9% -34.7% -25.1% -47.3% -70.0% -80.0% -90.0% -86.4% 27 TRANS DATA REL WOE VC VNI SI INTER FSP D2C CP -100.0% DRAFT VERSION available online at http://www.ericwalden.net Table 6: Long Run BHAR Whole Sample (n = 601) CP (n = 31) D2C (n = 366) FSP (n = 18) INTER (n = 176) SI (n = 45) VNI (n = 2) VC (n = 26) WOE (n = 4) Mean Return -27.0% -47.3% -22.1% -18.5% -29.4% -19.4% -86.4% -25.1% -34.7% 2.7% 10.5% 3.5% 21.2% 5.0% 11.1% 11.9% 14.8% 31.0% 0.000*** 0.000*** 0.000*** 0.396 0.000*** 0.088* 0.087* 0.103 0.344 Ownership REL (n = 528) DATA (n = 524) TRANS (n = 346) -26.5% -24.9% -22.8% 2.9% 2.9% 3.6% 0.000*** 0.000*** 0.000*** Customer Type B2C (n = 410) B2B (n = 191) -32.9% -14.4% 2.9% 5.9% 0.000*** 0.016** Production Type DIGITAL (n = 356) TANGIBLE (n = 244) -27.2% -26.7% 3.6% 4.2% 0.000*** 0.000*** Firm Type NET (n = 243) MORTAR (n = 358) -42.7% -16.4% 4.2% 3.5% 0.000*** 0.000*** PP_AL (n = 191) C_AL (n = 27) TRAD_AL (n = 84) -20.1% -31.7% -30.8% -40.5% 4.1% 4.4% 10.9% 5.6% 0.000*** 0.000*** 0.009*** 0.000*** TRANSFOR (n = 149) EXECUT (n = 452) -17.7% -30.1% 6.0% 3.0% 0.004*** 0.000*** Business Models Governance Type UNI (n = 315) Innovativeness Std P-Val To consider the impacts of business models, controlling for the non-exclusive nature of the coding, we regress the BHAR against the eight business models and the three ownership structures. These results are reported in Table 7. They seem to indicate that over the long run, the direct to customer business model produced significantly greater BHAR returns than having no business model. Likewise the intermediary model and the shared infrastructure model produce marginally greater BHAR than having no model at all. Ownership of the various customer assets produced no significant BHAR over having no ownership. Further, given that the sum of the models is less than the value of the model with the highest score, it seems unlikely that even owning all of the assets would result in a significant return over owning none of the assets. These results should be interpreted with caution as the F statistic fails to reject the null that all coefficients are simultaneously different from zero. Table 7: Long Run Joint Analysis of Business Model Impacts (Dependant Variable BHAR) F 1.359 P-val 0.188 Return 28 R2 Adjusted R2 0.025 0.007 Std P-Val DRAFT VERSION Intercept CP D2C FSP INTER SI VNI VC WOE REL DATA TRANS available online at http://www.ericwalden.net -57.3% 4.0% 24.7% 9.0% 17.4% 25.3% -36.6% 17.7% 6.9% -6.2% 11.7% 4.0% 14.2% 16.0% 10.5% 16.1% 10.4% 13.5% 48.4% 15.1% 34.5% 9.5% 9.5% 7.4% 0.000*** 0.802 0.019** 0.574 0.094* 0.061* 0.450 0.242 0.842 0.515 0.219 0.584 LONG RUN UNIQUE One potential problem with the examination of the long run BHAR is that firms pursuing multiple initiatives may have overlapping return periods. For example, if a firm starts an initiative in March of 1999 the starts another initiative in June of 1999, there will be a period of overlap from June 1999 to March 2000. The return in this period will be attributed to both initiatives, and thus double counted. This is similar to the problem of non-exclusitivity of the business models. Unfortunately, there is no way to separate the returns and properly attribute them. To account for this, we remove all of our sample firms pursuing more than one initiative in 1999. The results of the initiative characteristics are presented visually in Figure 6. The results are somewhat different from those of the previous analyses, but still show a similar pattern. In the long run unique sub sample the BHAR for firms undertaking B2B initiatives were positive though not significantly so, while B2C was significantly negative. The relative value of digital and tangible initiatives switch so that digital initiatives are not significantly different from zero, while tangible initiatives are significantly negative. Unilateral actions are not significantly different from zero, while both pure-play and traditional alliances are significantly negative. However, computer alliances generated positive, albeit non-significant, BHAR. Transformational initiatives were not significantly different than zero, while executional/extensional initiatives we significantly negative. 29 DRAFT VERSION available online at http://www.ericwalden.net Figure 6: Long Run Unique BHAR 20.0% 10.4% 10.0% 1.5% 0.0% -10.0% -7.0% -4.6% -9.9% -20.0% -15.6% -16.2% -30.0% -26.7% -40.0% -6.8% -21.4% -30.4% EXECUT TRANSFOR TRAD_AL C_AL PP_AL UNI MORTAR NET TANGIBLE B2B DIGITAL -44.2% B2C -50.0% The results of the business model analysis are presented in Figure 7. They show the same basic pattern as the previous two data sets. One exception is that ownership of the data produced less negative BHAR than ownership of the transaction. Full statistics are displayed in . Figure 7: Long Run Unique BHAR 10.0% 1.4% 0.0% -10.0% -20.0% -30.0% -1.1% -6.2% -12.3% -20.5% -14.3%-11.2%-12.9% -20.1% -20.2% -40.0% -50.0% -60.0% -70.0% -80.0% TRANS DATA REL WOE VC VNI SI INTER FSP D2C CP -74.4% Table 8: Long Run Unique BHAR Business Models Whole Sample (n = 244) CP (n = 11) D2C (n = 168) FSP (n = 6) 30 Mean Return -13.3% -20.5% Std P-Val 5.1% 24.7% 0.010** 0.426 -12.3% 1.4% 6.0% 59.9% 0.042** 0.983 DRAFT VERSION available online at http://www.ericwalden.net INTER (n = 50) -20.2% 11.9% 0.096* Ownership SI (n = 17) VNI (n = 1) VC (n = 9) WOE (n = 2) REL (n = 215) -6.2% -74.4% -1.1% -20.1% -14.3% 23.3% NA 28.3% 71.8% 5.4% 0.793 NA 0.971 0.826 0.009*** Customer Type DATA (n = 218) TRANS (n = 170) B2C (n = 156) -11.2% -12.9% -26.7% 5.5% 5.9% 5.1% 0.043** 0.031** 0.000*** Production Type B2B (n = 88) DIGITAL (n = 113) 10.4% -9.9% 10.6% 8.2% 0.326 0.227 Firm Type TANGIBLE (n = 131) NET (n = 66) -16.2% -30.4% 6.5% 11.6% 0.014** 0.011** Governance Type MORTAR (n = 178) UNI (n = 147) -7.0% -4.6% 5.5% 7.1% 0.209 0.519 PP_AL (n = 66) -21.4% 9.2% 0.023** C_AL (n = 7) TRAD_AL (n = 28) TRANSFOR (n = 63) 1.5% -44.2% -6.8% 31.2% 9.8% 10.1% 0.964 0.000*** 0.505 EXECUT (n = 181) -15.6% 6.0% 0.010*** Innovativeness Again we regress BHAR against the business model characteristics to examine the joint impact. This is presented in Table 9. In this case no variables attain significance, and the F statistics fails to reject the joint hypothesis that all coefficients are zero. Table 9: Long Run Unique Joint Analysis of Business Model Impacts (Dependant Variable BHAR) F 0.373 Intercept CP D2C FSP INTER SI VNI VC WOE REL DATA TRANS 31 P-val 0.965 Return -11.9% -7.0% -5.0% 15.3% -16.0% 0.2% -73.5% 3.6% -12.7% -16.4% 27.4% -6.4% R2 Adjusted R2 0.017 -0.029 Std 31.6% 34.4% 23.8% 33.9% 23.0% 27.5% 85.4% 34.0% 62.7% 18.4% 20.0% 15.8% P-Val 0.707 0.840 0.836 0.653 0.487 0.995 0.390 0.917 0.839 0.374 0.172 0.683 DRAFT VERSION available online at http://www.ericwalden.net COMPARABLE DATA SETS Thus, far we have looked at BHAR separately from CAR. We construct a final data set which contains data for firms that have both the 250 trading days prior to the announcement need for CAR analysis and the year of trading days post announcement need for BHAR analysis. We eliminate the non-unique firms from the sample and present the final results in Table 10. The results produce an interesting contrast to the short and long run. B2B is superior in the long run, but B2C is superior in the short run, though not significantly so. Digital initiatives are superior in the long run but tangible initiatives are superior in the short run. Click-and-mortar firms are superior to pure play net firms in both the long and short run. Unilateral initiatives seem to produce more value for shareholders than initiatives with alliance partners. Transformational initiatives produce more value in the long run, but executional/extensional initiatives produce greater return in the short run, although the short run difference is not significant. In terms of business models the direct to customer model seemed to produce better than average returns in both the long and short run. The full service provider model produced great short run returns, but performed particularly badly in the long run. Ownership of the transaction seemed to produce good returns in both the long and short run. Table 10: Comparable BHAR and CAR BHAR Business Models Whole Sample (n = 192) CP (n = 8) D2C (n = 136) FSP (n = 5) INTER (n = 35) SI (n = 13) VNI (n = 1) VC (n = 7) WOE (n = 1) Ownership REL (n = 168) DATA (n = 171) TRANS (n = 138) Customer Type Production Type Firm Type B2C (n = 120) B2B (n = 72) 32 CAR [-1,+1] -3.9% 5.7% 0.497 2.5% 0.5% 30.0% 0.988 -2.5% -0.8% 6.7% 0.900 4.4% -57.3% 14.6% 0.017** 10.3% -17.7% 12.4% 0.164 -0.1% 14.0% 28.3% 0.629 0.0% -74.4% NA NA -2.9% -37.6% 18.4% 0.087* -4.9% 51.7% NA NA 2.6% -4.2% -0.4% -0.7% Std 6.0% 6.1% 6.6% P-Val 0.489 0.949 0.915 -17.5% 5.5% 0.002*** 18.8% 11.7% 0.113 9.3% 7.1% Std P-Val 0.8% 3.7% 0.9% 5.7% 1.9% 2.7% NA 4.3% NA 0.002*** 0.520 0.000*** 0.145 0.970 0.999 NA 0.293 NA 2.9% 2.7% 4.1% 0.8% 0.8% 0.9% 0.001*** 0.001*** 0.000*** 2.8% 2.0% 1.0% 1.2% 0.007*** 0.110 DIGITAL (n = 84) TANGIBLE (n = 108) -1.0% -6.1% 0.911 0.393 0.2% 4.2% 1.3% 1.0% 0.854 0.000*** NET (n = 31) MORTAR (n = 161) -5.5% 18.4% 0.766 -3.5% 5.8% 0.542 -1.0% 3.2% 2.7% 0.8% 0.701 0.000*** DRAFT VERSION available online at http://www.ericwalden.net Governance UNI (n = 120) Type PP_AL (n = 52) C_AL (n = 7) TRAD_AL (n = 17) Innovative- TRANSFOR (n = 48) ness EXECUT (n = 144) 3.1% 7.9% 0.692 -13.7% 9.1% 0.136 1.5% 31.2% 0.964 -28.2% 13.6% 0.054* 4.2% -0.6% -0.1% 0.3% 1.0% 1.4% 4.1% 2.5% 0.000*** 0.677 0.973 0.919 11.0% 12.0% 0.366 -8.8% 6.4% 0.171 2.1% 2.6% 1.6% 0.9% 0.193 0.005*** In order to examine the relation between short and long run returns we regress the short run return against the long run return and a constant. If the returns are related we would expect a positive and significant coefficient. This is presented in Table 11. The Fstatistic suggest that this does not produce a regression, and neither coefficient is significant. Table 11: Relation between BHAR and CAR (Dependant variable is BHAR) F-stat 1.193 P-val 0.276 R2 0.006 R2 Adjusted 0.001 Intercept CAR[-1,+1] Coefficient -0.030 -0.363 Std 0.057 0.332 P-val 0.606 0.276 DISCUSSION INITIATIVE CHARACTERISTICS The analysis is, in general, supportive of the theory that intangible, IT complementing investments are necessary to successfully create value via EC activities. We postulate that EC initiatives which leverage a firm’s existing intangible assets or a firm’s ability to build the necessary assets will result in higher return. We find that, in the short run, all of our predictions hold suggesting that investors also believe this. In the long run we find that click-and-mortar firms produce greater return than pure-play net firms, that B2B initiatives produce greater returns than B2C firms, and that unilateral initiatives, in general, produce greater return. This is consistent with our discussions and expectations concerning intangible, IT complementing assets. However, the returns to digital and tangible production in the long run were quite similar. BUSINESS MODELS Relative to business models, the analysis suggested that the direct to customer model was a reliable model. There was also some evidence that ownership of the transaction was a preferred customer asset. However, when we accounted for the double counting of returns introduced by the non-exclusitivity of the business models and ownership structures, we consistently failed to find any evidence that the overall impact of both business models and customer assets was significantly different from zero. Thus, we are 33 DRAFT VERSION available online at http://www.ericwalden.net lead to presume that neither business model nor customer asset ownership influences the return to EC initiatives. This does not mean that the atomic business models examined are not useful, and in fact, the creators of the models make no claim that any model is better than any other. Rather the purpose of the models is to help firms conceptualize their business and to point out the issues specific to them. In terms of our theory, this would suggest that while different models may require different complementary assets, there is no systematic difference in the overall level of complementary assets required. It seems somewhat comforting to acknowledge that initiatives of many different types can be successful if they are carried out properly. The lack of results in the area of customer asset ownership is no less satisfying for the theory of this paper. It simply suggests that the intangible, IT complementing assets are the important assets to own rather than the customer assets. FUTURE RESEARCH Based on prior literature, we have postulated that EC initiatives create value through the formation and application of intangible, IT complementing assets. A logical extension to this line of theory development is to identify those intangible assets. Once identified theory should be developed to explain the conditions and contributions of those assets, and empirical validation performed to measure validate that theory. We offer three suggestions of possible candidates. First, we suggest that standardization is an asset that complements IT. The advantage of IT is that it can reliable process large amounts of structured data. If the elements of data are standardized that will give them the structure necessary to be efficiently processed by IT. This is not to say that all processes in a firm must be standardized, be cause the environment may be quite non-standard. However, standardizing those processes which can reasonably be standardized should help IT perform more efficiently and free up firm resources to deal with non-standard issues. The second IT complementing asset is an organizational learning focus. By this we mean that an organization puts forth considerable effort not only to perform its functions, but also to position itself for the future through continual learning. This complements IT, because IT lacks inertia. Once in place IT faithfully performs the same tasks. Without an organizational impetus to drive IT changes there will be no changes and the environment will shift, rendering the fixed IT obsolete. A third IT complementing asset is the technical expertise of the user. We see this clearly in EC, where there were few viable transactions until a large portion of the population became familiar with the use of the hardware and software necessary to access the internet. CONCLUSIONS We have proposed that the ability to deploy and create intangible, IT complementing assets is necessary to the creation of business value in the EC context. We examine this hypothesis using the event study methodology in two ways. First we apply the traditional short run methodology, then we apply a newer long run methodology. In both cases we find reasonable evidence that firms which can more easily deploy the relevant intangible assets produce greater return for their owners. We compare this to the return generated by firms based on business model and customer asset ownership, and find no compelling 34 DRAFT VERSION available online at http://www.ericwalden.net evidence that business model or customer asset ownership determines return on EC investments. We note that this does not mean business models do not have academic value, but rather that different business models may be successful if the proper intangible assets are in place. This work contributes to the literature in a variety of ways. First it puts forward the theory that intangible IT complementing assets are an important determinant of EC success. A second contribution lies in introducing the long run event study methodology into the Is literature. Short run event studies have recently become popular, but as our analysis shows it is not clear that the long run returns are well predicted by short run returns. This occurs simply because more information comes available in the long about the intangible assets being put into place to support the IT. A third contribution is the first empirical examination of the value created by EC business models. We find that the form of the business model is not a great determinant of value, but rather the implementation of the business model is important. We echo Michael Porter’s recent commentaries that suggest the strategy supporting EC is more important than either the for of the business model or the technology involved {|Porter, 2001 #228|}. REFERENCES 35