On the Efficiency of Internet Markets for Consumer Goods

On the Efficiency of Internet Markets for Consumer Goods Brian T. Ratchford* Xing Pan Venkatesh Shankar August, 2002 * Brian T. Ratchford ([email protected]) is the PepsiCo Chaired Professor of Marketing at the Robert H. Smith School of Business, University of Maryland, College Park, MD 20742, Xing Pan ([email protected]) is a doctoral student at the Robert H. Smith School of Business, University of Maryland, College Park, MD 20742, and Venkatesh Shankar ([email protected]) is a Ralph J. Tyser Fellow and Associate Professor of Marketing and Entrepreneurship at the Robert H. Smith School of Business, University of Maryland, College Park, MD 20742. The authors gratefully acknowledge the helpful comments of the special issue editor and two anonymous referees. On the Efficiency of Internet Markets for Consumer Goods Abstract Despite claims that electronic commerce lowers search costs dramatically, and therefore makes it easy for consumers to spot the best buy, empirical studies have found a substantial degree of price dispersion in electronic markets for consumer goods. This study investigates the consumer welfare implications of observed price levels and price dispersion in electronic markets. We examine the consumer welfare implications of changes in the structure of electronic commerce markets employing comprehensive data sets on e-tailer prices and services collected from BizRate.com in November 2000 and 2001. We find that price dispersion decreased substantially between these two periods, and that measured differences in e-tailer services bear little relation to e-tailer prices. 2 On the Efficiency of Electronic Markets for Consumer Goods Introduction The Internet fits the definition of an electronic marketplace, which is defined by Bakos (1997, p.1676) as an “interorganizational information system that allows the participating buyers and sellers … to exchange information about prices and product offerings.” Bakos (1997) argues that electronic markets allow buyers to get easier access to price and other information, thereby decreasing search costs. In turn the lower search costs will increase competition, and lead to a better allocation of resources (Bakos 1997). Based on this analysis, one might expect online competition to force prices charged by e-tailers to a uniformly low level, and for consumers to benefit greatly from the presence of Internet markets. What actually happened, however, seems, on the surface, to be quite different. While Brynjolfsson and Smith (2000) do find that Internet prices are lower than conventional retailer prices for books and CDs, they still find substantial dispersion in the prices posted by different e-tailers, and that this dispersion is comparable to that found in conventional markets. Using data from a broader sample of items collected in November 2000, Pan, Ratchford and Shankar (2001) document that the range of prices posted by e-tailers for a given item often exceeds the average price of the item. Pan, Ratchford and Shankar (2002) show that only a small proportion of this large degree of price dispersion is explained by differences in services provided by e-tailers. Clay, Krishnan and Wolff (2001) find evidence of large and persistent price dispersion in the Internet market for books over August 1999-- January 2000. Similarly Baye, Morgan and Scholten (2001) find evidence of large and persistent price dispersion in a sample of 3 1000 items over August 2000-March 2001. Evidently Internet markets are not as frictionless as Bakos (1997) and others would have predicted. This raises the question of whether the Internet actually does enhance allocative efficiency, thereby providing substantial benefits to consumers. Related to this question is how large degrees of dispersion in posted prices can exist in a market that is supposedly frictionless, or approximately so. Another question is whether the observed dispersion in offerings is symptomatic of an immature market, and has disappeared more recently as Internet markets have matured. The demise of many e-tailers in the recent economic downturn of 2001 may have enhanced the functioning of e-tail markets. To shed light on these questions, we present a framework based on Ratchford, et al. (1996) for measuring the relationship between consumer’s surplus and the dispersion of retail prices. This framework will provide guidance to our empirical analysis. To provide further guidance to our empirical analysis, we go on to summarize what is known about pricing when search costs and imperfect information are present, and about the impact of services offered by e-tailers on prices when information is imperfect. We apply our theoretical analysis to interpret the comparisons between prices at e-tailers and conventional retailers that are presented in Brynjolfsson and Smith (2000). We also analyze two comprehensive data sets on e-tailer prices and services that were collected mainly from BizRate.com in November 2000 and November 2001. We use these data sets to draw conclusions about changes in price dispersion and consequent changes in consumer welfare resulting from the maturation of e-tail markets that took place between these two time periods. We find that price dispersion decreased substantially between 4 these two periods, and that measured differences in e-tailer services bear little relation to e-tailer prices. Related Literature Since the pioneering work of Alba et al. (1997) and Bakos (1997) there has been a great deal of theoretical and empirical work about the impact of electronic commerce on markets and welfare. A number of theoretical models explore subtleties associated with the emergence of Internet shopping. While Bakos (1997) postulates that the Internet will enhance competition, Lal and Sarvary (1999) and Lynch and Ariely (2000) show that this is not necessarily the case. Lal and Sarvary (1999) present a theoretical model in which Internet can decrease search by lowering the cost of buying a preferred item relative to searching in the store. Welfare, defined as the sum of consumer and producer surplus, increases, but the seller is able to appropriate the cost saving to consumers through a higher price. Lynch and Ariely present a simulated web search environment for wine. They show that designing the web environment to make quality comparison easy decreases price sensitivity for unique items, giving them more monopoly power. Conversely designing the web environment to facilitate comparison makes common items more price sensitive, but lowers demand for the unique items. Some studies point out that the Internet lowers search costs for both price information and non-price information such as product and quality information (Degeratu et al. 2000; Shankar, Rangaswamy and Pusateri 2001). While lower search costs on price information may lead to lower prices, lower search costs on non-price information could lead to lower price sensitivity and consequently, to higher prices. 5 Ellison and Ellison (2001) and Baye and Morgan (2001) study the role of price search engines. Both papers note the following paradox that is succinctly stated by Ellison and Ellison (2001, p.4): “If search engines create Bertrand competition then there will be no price dispersion and consumers will be unwilling to pay for the information the search engine provides.” Ellison and Ellison (2001) propose that the problem can be circumvented if firms adopt strategies to make search more difficult by obfuscating prices or by some other means. Baye and Morgan (2001) propose that the shopping service will solve the problem posed in the paradox by pricing the services in a way that preserves ex ante price dispersion. While the shopping service benefits consumers, it will be costly to firms because it increases competition, and Baye and Morgan (2001) show that it will not always increase overall welfare. There is evidence that information available on the Internet can have substantial effects on prices in conventional markets. Brown and Goolsbee (2002) estimate that the presence of Internet referral services led to a reduction of 8-15 percent in term life insurance policies. Interestingly these authors find that these reductions are accompanied by an increase in price dispersion when the share using on-line referrals is low. The authors point out that this is consistent with the search model of Stahl (1989). Morton, Zettelmeyer and Silva-Risso (2001) found that Autobytel customers in California saved an average of about $450 per car, and Ratchford, Lee and Talukdar (2002) document that the Internet has become a major source of information about automobiles. Their estimates indicate that the Internet leads to both timesavings and better buys. A few studies have compared prices and price dispersion at pure play e-tailers and multichannel retailers 6 (e.g., Ancarani and Shankar, Pan, Shankar, and Ratchford 2002, Tang and Xing 2001). We discuss their results later in the paper. As stated in the introductory section of this paper, Brynjolfsson and Smith (2000), Pan, Ratchford and Shankar (2001), Clay, Krishnan and Wolff (2001), and Baye, Morgan and Scholten (2001) have found evidence of substantial and persistent price dispersion in Internet markets. Along the same lines, Clemons, Hann and Hitt (2002) found evidence of substantial differences in the quality of on-line travel agent recommendations. Clay, Krishnan and Wolff (2001), who document a number of strategies that appear to be followed by on-line sellers of books, find that price dispersion, and propensity to discount, are much greater for New York Times best sellers than for other book types. Smith and Brynjolfsson (2001) show that there is a substantial amount of brand preference for established book retailers, which may help to explain price dispersion in this market. While studies show that there is persistent price dispersion, the latest study that we reviewed ends in March 2001 before the recent economic slowdown may have had its full impact. A question that we will address in this study with data collected In November 2001 is whether price dispersion declined as economic conditions became less favorable to e-tailers. Consumer’s Surplus and Price Dispersion In general allocative efficiency, a standard measure of consumer welfare, may be defined, as the sum of consumer and producer’s surplus, where consumer’s surplus is the difference between willingness to pay and price, and producer’s surplus is the difference between price and cost. Here concentrate first on the consumer’s surplus part of this 7 equation, taking prices as given. Our objective is to develop measures of consumer’s surplus that can be operationalized, and can be related to price dispersion present in a market at any time. View the consumer’s problem as acquiring one unit of an item such as a book, CD or PC, at retail.1 To simplify the analysis, assume that the consumer values the item enough to make a purchase at any price offered. At the highest level the consumer is faced with the decision of whether to buy this item in a conventional store or on the Internet at an e-tailer. Once this decision is made the consumer searches among retailers or e-tailers for a good deal, and after putting some effort into searching for a good deal, makes a choice. We consider the case where the search effort optimizes the tradeoff between a better buy and the cost of search. Assume that consumer i wishes to allocate his/her budget between the focal item m and another composite commodity, which represents all other goods, and that the utility function is separable between the focal product class and all other goods. If the price of the focal item at retailer r from channel j (bricks and mortar or Internet) is pmrj, the amount available to spend on other goods can be expressed as yi – pmrj – Si – cmrj, where y is income, S is expenditure on search for the best retailer or e-tailer at which to buy the focal item, and cimrj is transaction costs of purchasing the focal product. The latter may include travel, waiting at the checkout, entering credit card information on-line, or any other costs of completing the transaction. These costs are reduced by services provided by the e-tailer or retailer. The sum of pmrj and cimrj is the full price of the item (Ehrlich and Fisher 1982). Given these assumptions, the consumer’s indirect utility function, conditional on the purchase of m, can be written as: 1 Alternatively the consumer could purchase a predetermined market basket of items. 8 U imrj = z im + α i ( y i − p mrj − S i − cimrj ), (1.) where zim is the utility of the focal item to consumer i. Without loss of generality, we can rescale Equation 1 so that it is measured in monetary units by dividing through by α. Define Vimrj = U imrj α i and φ im = z im α i . Also, noting that expenditure on search is mainly comprised of time spent on search, t, times a unit time cost, w, we can write S i = wi t i . This leads to the following expression for the monetary value of the consumer’s purchase of the focal item: Vimrj = φ im + ( y i − p mrj − wi t i − cimrj ). (2.) Assume for simplicity that the consumer specializes his/her search on the channel that provides the highest expected utility. Let πirj be the probability that consumer buys at retailer r conditional on buying in channel j. This can be interpreted as the probability that a consumer who has stopped searching after searching a given amount will buy at that store. Then the expected consumer’s surplus (without considering search costs) from the purchase in that channel is (DePalma, Myers and Papageorgiou 1994; Small and Rosen 1981): (3.) [ ] [ ] CS ij = ∑ nr=j 1 π irj φim − p mrj − cimrj = φim − ∑nr=j 1 π irj p mrj + cimrj , where nj is the number of retailers in channel j. Consumer’s surplus net of search costs would be obtained by subtracting search costs from Equation 3. There are two polar cases. One is when the consumer does not search at all (or has no prior knowledge), and ( ) just buys at random, in which case π irj = 1 n j . The other polar case is when the consumer searches until locating the best retailer offering the best buy, b, so that π irj = 1 9 if r = b, 0 otherwise. Abstracting from search costs, the difference between the maximum consumer surplus (b) and consumer surplus from a random choice (a) is: (4.) [ ] [ ] CS ijb − CS ija = p mj + cimj − p mbj + cimbj , that is, the difference between the average full price and the lowest full price. Equation 4 provides a measure of the potential gains to search for a consumer facing a given set of full prices. Clearly Equation 4 will become larger as the amount of price dispersion in channel j increases. The consumer searches in channel j to maximize expected utility, which is equivalent to optimizing the tradeoff between expected consumer surplus and the cost of search. This tradeoff is optimized when: (5.) dCS ij dt ij = − ∑nr=j 1 dπ irj dt ij [p mrj ] + cimrj = wi , where tij is time spent by consumer i at search in channel j. If the time is spent wisely, the search will lead to decreases in dπ irj dt ij for stores with a high full price, and to increases in dπ irj dt ij for stores with a low full price, thereby increasing consumer surplus. Eventually diminishing returns to time will set in (for example if the consumer finds the lowest full price, returns to spending more time become zero), and the consumer will stop searching. The solution to Equation 5 leads to a set of values π*irj and t ij* for each channel. Suppose that the consumer solves Equation 4 for each channel prior to deciding which channel to buy in. Denote bricks and mortar as retailer type B, and the Internet as retailer type I. Then we can define the expected full price at which the consumer is indifferent between buying on-line and going to the brick and mortar retailer as: 10 (6.) [ ] * * * * ∑nr=I 1 π irI [ p mrI + cimrI ] =∑ nr=B1 π irB [ p mrB + cimrB ] + wi t iB − t iI , where higher on-line full prices make the brick and mortar retailer a better deal, and lower on-line full prices make the e-tailer a better deal. Two basic conclusions can be drawn from Equation 6. One is that any saving in search time due to the Internet is more valuable to those with high time costs. Consumers with high search costs will be willing to pay a premium over the standard retailer, and will benefit from the Internet even if they have to pay higher prices for this channel. These are also the consumers who will search less, as shown in Equation 5. They will be more willing to accept a risk of paying a high price. Because of time savings the Internet can be beneficial even though consumers may wind up paying a relatively high price (Lal and Sarvary 1999 reach a similar conclusion in a somewhat different context). The benefits of timesavings increase with search costs. Thus one explanation for large observed price dispersion on the Internet is that its time saving properties make it valuable to a group of consumers with high time costs that is willing to accept high prices rather than incurring additional search costs. In effect the high prices are a payment to the e-tailer for the privilege of using this timesaving medium. At the same time other users of the Internet who do not have such high time costs can afford to search, and can search efficiently on this medium, creating a demand for low prices. Price dispersion results from the differences in incentives to search.2 The other conclusion to be obtained from Equation 6 is that the channel that the consumer can search most efficiently, as indicated by values of π*irj and t ij* obtained from solving Equation 5, gains an advantage. Thus consumers who are relatively efficient at 2 We will pursue this more formally in the next section. Brynjolfsson and Smith (2000) make a similar argument but do not provide a formal demonstration of this point. 11 using the Internet will be willing to pay more to use this channel. However, these consumers are more likely to be able to locate a relatively good buy using this medium. A general conclusion to be drawn from this section is that if choice probabilities and full prices associated with any outlet can be computed for a given state of information, the consumer surplus expression presented in Equation 3 can be used to make statements about potential gains to information. Equation 4 presents an example of such a statement for potential gains from going from no information to complete information. It is important to realize that Equations 3 and 4 are quite general, and do not depend on any assumption of optimizing behavior on the part of consumers. While Equation 4 presents a measure of gains to search, we can also say something about the highest search cost in the market if we are able to impose more structure. Consider a model in which consumers search sequentially among retailers (or products) for the lowest price, as in Carlson and McAfee (1983).3 Prior to search, consumers are assumed to know the general distribution of prices, but not the exact price charged by any seller.4 Consumers are assumed to use a stopping rule in which they search if the expected gain relative to the best price they have observed so far is greater than the cost of another search, but stop otherwise. The stopping rule can be used to place a lower bound on the search cost of the consumer who buys the highest priced item in the market. If a consumer encounters a price of ph, where ph is the highest price, the expected gain from searching further is equal to ∑r ≠ h (1 / nr )( p h − p r ). If the consumer is content to 3 To keep notation simple we assume c = 0 for this argument, and that search is for price. Alternatively the objective could be to find the lowest full price. We will drop the consumer subscript i to simplify notation. 4 Carlson and McAfee (19832) point out that their results can also be derived from a more complex model that relaxes the assumption that consumers know the distribution of prices. 12 buy at the highest price rather than searching further, search cost must exceed this expected gain. So if T is the highest search cost, we know that: T > ∑r ≠h (1 / nr )( p h − p r ) = ((nr + 1) nr ) p h − p , (7.) where “bar” denotes mean, and nr is the number of firms with a price lower than ph. Thus knowledge of the highest price in the market and the average price allow one to say that the highest search cost is at least T. Carlson and McAfee (1983) show that if there is a uniform distribution of search costs between 0 and the upper bound T, and if consumers apply the reservation price rule, a downward sloping demand curve in which quantity is a linear function of the difference between price and average price, will result. If they choose randomly, consumers with the highest range of search costs will spread their purchases evenly among all stores. Consumers with the next highest range will, however, avoid the highest-priced offering, but will spread their business evenly among all others. So the highest-priced retailer (etailer) gets only the business of the highest search cost consumers. The next highestpriced seller gets the business of only the two highest search cost groups, and so on. Summary The following general conclusions can be drawn from the analyses in this section: 1. Differences in consumer surplus gross of search costs between channels, time periods or consumers can be computed as a weighted sum of full prices, where the weights are choice probabilities. 2. The difference between the unweighted average price and the lowest price reveals the difference in expected surplus (gross of search costs) between a completely uninformed and a fully informed consumer. The difference between the highest price and the average price reveals a lower bound on the highest search cost in the market. Given their interpretability, these measures can be used in empirical analyses of price dispersion. 3. Because of the time saving properties of the Internet, it will appeal to consumers with high time costs who are willing to pay a high price. Despite paying a high 13 price these consumers will benefit from the time saving afforded by the Internet, and will be better off with its existence. In the next section we will discuss alternative explanations for price dispersion, and use this discussion to develop hypotheses about drivers of price levels and price dispersion, and how these variables might change as Internet markets mature. Market Behavior Search Costs as an Explanation of Market Behavior There is a large literature that shows that price dispersion can be an equilibrium outcome when some consumers find it too expensive to search for complete information. Examples are Varian (1980), Salop and Stiglitz (1982), Carlson and McAfee (1983), Stahl (1989), Bakos (1997), Burdett and Coles (1997). To keep all prices from converging to the monopoly level, these models typically assume that some consumers are perfectly informed or have zero search costs. On the supply side, equilibrium price dispersion can be driven by differences in firm costs (e.g., Carlson and McAfee 1983), or by firms with identical costs randomizing their prices to capture some mixture of informed and uninformed consumers in a competitive game (e.g. Varian 1980, Stahl 1989). A general result of these models is that prices and price dispersion fall, and welfare increases, as search costs decline.5 To the extent that consumers learn to use the Internet more efficiently, their costs of searching among e-tailers should decline over time, which leads to the following hypothesis: H1: Due to declining search costs and improved consumer information, price levels and price dispersion in e-tail markets should decline over time. 5 Price dispersion can increase when the number of informed consumers is small before it eventually decreases (Stahl 1989). 14 Since there are sites that provide complete lists of prices (e.g. Shopper.com), one might conclude that the Internet already provides low search costs and complete information. However, search costs might still be present for at least some consumers if awareness of these sites is imperfect, or non-price attributes of e-tailers are important. Even if these qualifications do not hold, price dispersion may still exist. Baye and Morgan (2001) show that providers of price lists will preserve ex ante price dispersion (and a demand for their services) through the imposition of fees on firms and consumers. Alternatively Ellison and Ellison (2001) argue that firms using Internet price listing services will attempt to preserve price dispersion by obfuscating their prices. These arguments suggest that Internet price dispersion will persist because sellers and service providers will take steps to maintain it. At least one search model also suggests that price dispersion will persist if consumers enter and exit the market at a steady rate. Burdett and Coles (1997) construct a model in which entrants compete aggressively for the business of consumers, while established firms take advantage of their customers’ switching costs by continually increasing their prices. The result is on going price dispersion with the newest firms at the low end of the distribution, the oldest ones at the upper end. Comparative statics show that prices and price dispersion still decrease as search costs decline in this model. Still the model indicates that entry and aging of firms and consumers, which alter search and pricing incentives, creates another explanation of persistent price dispersion. In sum, service provider and e-tailer incentives, and dynamics of entry and exit, suggest an alternative to H1. H1A: Price levels and price dispersion in e-tail markets will persist over time. 15 Another consideration in predicting the behavior of prices over time is changes in the number of firms selling an item, a subject investigated in detail by Baye, Morgan and Scholten (2001). Over time, the number of e-tailers increased due to the dot.com boom and due to more bricks-and-mortar retailers going online, but subsequently declined due to the bursting of the Internet bubble. These authors show that existing theoretical models make conflicting predictions about the effect of number of firms on price dispersion, with some predicting an increase in price dispersion with number of firms, others predicting a decrease.6 Their empirical results tend to show an inverted-U relationship, which leads to the following hypothesis. H2: Price dispersion will increase at a decreasing rate with number of firms selling an item, then decline after reaching a maximum. Differences in E-Tailer Service as an Alternative Explanation for Price Dispersion The consumer’s maximum utility in Equation 1 would result if he/she minimized the full price of the focal item defined as FPimr = p mr + cimr . The transaction costs are affected by services offered by the e-tailer: cimr = cimr (RS mr ), where RSmr is a vector of services. Full price would be unchanged if ∂p mr ∂RS kr = − ∂cimr ∂RS kr , i.e. if the increase in price due to a marginal increase in any service k equaled the reduction in transaction cost due to a marginal increase in service k. This relationship can be used to trace out indifference or willingness to pay curves for services (Rosen 1974, Ehrlich and Fisher 1982). Similarly the e-tailer would be willing to offer more of the service if he/she could get a large enough price increase to cover its cost. Under perfect competition and 6 The authors contend that the best measure of price dispersion is the gap between the lowest and secondlowest prices, since other ex ante prices could be irrelevant to consumer choices. However, since it is evident from our data that the largest firms in many Internet markets are somewhere in the middle of the price distribution, and do not have the lowest or second-lowest prices, we decided to stick with more conventional measures of price dispersion in this study. 16 perfect information the simultaneous decisions of consumers and sellers will lead to a regression relationship of the form (Rosen 1974): p mr = h(RS mr ). (8.) In this model, often termed the hedonic regression model, the dispersion in prices of an item is completely determined by the dispersion in services. Thus variation in services across e-tailers offers an alternative explanation of price dispersion to costly search. Empirically this hypothesis would be supported if regressions of prices on attributes could explain variation in prices up to random error due to omitted attributes. Thus we can state the following hypothesis: H3: Variation in prices across e-tailers is largely explained by variation in services offered by e-tailers, and measured price dispersion after correcting for differences in services is negligible, over time. The empirical strategy for testing H3 is to estimate Equation 8, compute qualityadjusted prices of each item, and compute the dispersion of these quality-adjusted prices. For example if we estimate the linear relation p mr = a + B m (RS mr ), the quality-adjusted price for any seller is: (9.) a p mr = p mr − Bˆ m (RS mr − R S m ), where the quality-adjusted price is expressed as the price of that item at an average level a of service. Adjusted price dispersion is then the dispersion of p mr , which should approach zero if H3 is correct. Under the maintained hypothesis of perfect information and perfect competition this adjustment process is valid (Rosen 1974). When information and competition are imperfect, Pakes (2001) shows that the function in Equation 8 still exists, but that the coefficients are a complex function of services offered by competing retailers and the distribution of consumer preferences. Therefore the coefficients no 17 longer have a clear interpretation, or even expected sign. While we will employ it as an empirical approximation, the adjustment outlined in Equation 9 can be inaccurate under significant departures from the conditions underlying H3. In the following sections we perform two empirical analyses that apply the models and hypotheses developed thus far in the paper. In our first analysis, we attempt to make statements about the efficiency of the Internet compared to conventional retailers by interpreting the results of Brynjolfsson and Smith (2000) in light of our formulas for consumer surplus. In our second analysis, we employ data on prices and service levels for a large number of categories to study changes in the efficiency of the Internet as a retail medium between November 2000 and November 2001. This time period coincides roughly with the periods just prior to and just after the shakeout of dot.com firms in the economic downturn of 2001. Comparison of Conventional and Internet Retailers Brynjolfsson and Smith (2000) tracked the prices of 20 book titles and 20 CD titles at 8 Internet outlets and 8 conventional outlets on a monthly basis over February 1998-May 1999. Their basic findings were that prices are generally lower on the Internet than in conventional outlets, but that price dispersion is roughly comparable across the two channels. Price dispersion was found to be higher on the Internet for books, lower for CDs. In an effort to make welfare statements about the value of the Internet relative to conventional channels, we will interpret the results of Brynjolfsson and Smith (2000) in light of our models. Table 1 summarizes the results of Brynjolfsson and Smith (2000) that are relevant to this comparison. The first panel compares an un-weighted average of conventional 18 channel prices with a weighted average of Internet prices, where the weights are web traffic shares, a (possibly crude) proxy for market shares. Abstracting from differences in time costs and transaction costs across the two channels, these estimates provide a rough estimate of the difference in expected cost across the channels for a representative consumer of the type outlined in Equation 6. The Internet is clearly superior in this aspect.7 (Table 1 about here) The second panel compares unweighted means. These can be interpreted as expected prices for a consumer who does not search. Again, the Internet appears to provide better buys on average for such a consumer. The differences in minimum prices in the third panel indicate that a fully informed consumer would also get a lower price on the Internet. Brynjolfsson and Smith (2000) estimated the full prices (p + c) for items in both channels, taking account of taxes, shipping and handling and transportation costs likely to be incurred by consumers. Though this comparison tends to narrow the advantage of the Internet, the Internet still shows up as having lower prices for both an uninformed and fully informed consumer. Of course this comparison does not consider intangible services, such as the ability to get a book or CD right away at a conventional retailer. On the other hand, it does not consider the consumer’s opportunity cost of time, which should favor the Internet. The final panel of Table 1 lists the range of full prices on the Internet, which averages $5.98 for books and $4.45 for CDs. This may be interpreted as the difference 7 The authors were understandably unable to obtain share weights for conventional retailers. To the extent that consumers in conventional channels are able to locate the best buys, Table 1 may over state expected retail prices for conventional retailers. 19 between the price that an unlucky uninformed consumer will settle for, and the price paid by a fully informed consumer. Because our earlier discussion indicates that the expected gain from searching one more price for such an unlucky consumer is roughly the difference between the maximum and average price, which is roughly half the range, consumers who accept the highest price rather than search further should have a search cost of $5.98/2 = $2.99 for books, $4.45/2 = $2.225 for CDs. Other studies comparing prices and price dispersion at different types of retailers show interesting differences. Tang and Xing (2001) find that the prices of pure play Internet retailers are significantly (about 14%) lower than those of online multichannel retailers, consistent with Zettelmeyer’s (2000) analytical result. Pan, Ratchford, and Shankar (2002) find that prices are lower for pure play e-tailers than they are for bricksand-clicks e-tailers for CDs, DVDs, desktop and laptop computers; they are similar for PDAs and electronics and higher for pure play e-tailers for books and software. Pan, Shankar, and Ratchford (2002) analytically and empirically show that prices at pure play e-tailers are lower than those at multichannel retailers in eight categories, apparel, gifts and flowers, health and beauty, home and garden, sports and outdoors, computer hardware, consumer electronics, and office supply. Ancarani and Shankar (2002) show that for books and CDs when list prices are considered, traditional retailers have the highest prices, followed by multichannel and pure play e-tailers, in that order. However, when shipping costs are included, multichannel retailers have the highest prices, followed by pure play e-tailers and traditional retailers, in that order. With regard to price dispersion, pure play e-tailers have the highest range of prices, but the lowest variability 20 (standard deviation); multichannel retailers have the highest standard deviation in prices with or without shipping costs. These findings suggest that online pricing is complex. In general these results indicate that, even if there is a large degree of dispersion of Internet prices, the price savings that can be found on the Internet can provide benefits to consumers who can efficiently use this medium. Obviously those who do not have access to the Internet cannot obtain these benefits without incurring the costs of obtaining access and learning how to use the Internet. If the advantage of the Internet persists as this medium matures (and conventional channels decline), efforts to enhance accessibility may be justified. To see what happens as Internet markets mature, we study a wide variety of Internet prices at two time periods in the next section. Comparison of Internet Prices – November 2000 and November 2001 The data for this part of the study are drawn mainly from BizRate.com, one of the well-known price comparison web sites. This site searches and updates daily the product, price and deal information for a large number of e-tailers. To overcome the potential shopbot participation effect, we also tried our best efforts to search and collect prices of those e-tailers who are not listed at BizRate.com, though BizRate’s list is quite complete in general. Moreover, by comparing e-tailers’ prices at BizRate and on their own websites, we verified they are identical for most e-tailers, except for a few who offer lower prices at Bizrate than their websites and then whose prices from their web sites are collected. In November 2000, we collected 6739 price quotes for 581 identical items sold by 105 e-tailers; in November 2001, we collected 6762 price quotes for 826 identical items sold by 89 e-tailers. Since these are posted prices, a critical assumption is that some 21 transactions took place at each observed price. Since an average item had 11.60 sellers in 2000, 8.17 in 2001, there did appear to be some attrition of sellers between the two periods. In addition to prices, we collected ratings of various e-tailer services published by BizRate.com. We purposely focus on identical items to avoid the potential problem of unmeasured product heterogeneity. Such products are found in the following categories: books, CDs, DVDs, computer software and hardware, and consumer electronics. As an example, the Toshiba Satellite 2775XDVD laptop computer with part number of PS277U-6M9J0K and features of PIII 650 MHz processor, 64 MB memory, 12 GB hard disk, 8x DVD, 56 Kbps modem, and 14.1” TFT screen sold by any e-tailer is the same. We compare the prices if such homogeneous items across the e-tailers in our sample selling them at any point in time. Unfortunately, however, we were unable to track prices of identical items over time because the model numbers and identities of items for sale tend to change over the course of a year. Thus our analysis is mainly useful for studying how the general dispersion in prices of identical items changed between the two time periods. It is less useful for tracking changes in price levels.8 In our analysis we work with basic prices as the dependent measure, and do not directly add in shipping and handling costs. This is because there are usually a number of options for shipping and handling, making it problematic to construct shipping and handling costs that are consistent across e-tailers.9 Since we do, however, consider 8 The rapid technology evolution for computer and electronic products, and the quick change in popularity for music, movies, and books, lead the prices of these products change drastically over their short life cycles. Thus comparing the changes of their price levels will have the effect of market maturation confounded with the effect of product maturation and be inappropriate. 9 This may be a manifestation of the obfuscation referred to in Ellison and Ellison (2001). 22 consumer ratings of shipping and handling costs in constructing our estimates of qualityadjusted prices, these costs are incorporated into our analysis. Table 2 provides a summary of the means and standard deviations of the average item prices for both samples. Differences in average price levels between the two samples are due to some degree to differences in the mix of items sampled. For example our 2001 desktop computer sample included a few relatively expensive servers, while our 2000 sample included more low-end items. Similarly, our software and consumer electronics included more and relatively more sophisticated items in 2001. The fall in average prices for DVDs, laptops and PDAs may reflect general price trends for these categories. In general, prices are not directly comparable between the two samples because of differences in items sampled, and general market trends. However, the dispersion of prices among sellers of physically identical items can be compared between the two periods. (Table 2 about here) Hedonic Analysis. The first step in our analysis of the BizRate data is to examine the extent to which service differences account for measured prices, and the hedonic model accounts for price dispersion. Our general approach will be to employ hedonic regressions of price levels on services to develop measures of prices adjusted for the effects of service quality. The first stage in our analysis of services is to define a set of measures of services on available data. BizRate.com presents the consumer evaluations of e-tailers on the first 9 attributes described in Table 3. The items are scored on 10-point scales, where higher scores measure better performance. While the first 9 attributes in Table 3 capture 23 functional dimensions of service, Brynjolfsson and Smith (2000) conjecture that trust is an important dimension of e-tailer service: one would go to a trusted e-tailer to avoid the time needed to resolve problems that might crop up otherwise. To attempt to capture the trust dimension, we employ two variables, which are the final two measures listed in Table 3. One, called “Certify 5,” is a count of the number of certifications that an e-tailer receives from the following certifying agencies: Better Business Bureau, Gomez.com, BizRate.com, Truste.org, VeriSign.org. The other measure of trust, called “years certified,” is the number of years the seller has been certified by BizRate.com. The rationale is that a high number of years indicates that the e-tailer has been active long enough to develop a reputation. (Table 3 about here) Since these 11 measures of e-tailer services are not independent, and some are likely to be measures of the same underlying construct, we subjected them to a factor analysis. Because we wish our measures to be invariant to time period, the factor analysis was done on the pooled 2000 and 2001 data. The results of the factor analysis of the service measures indicate the existence of four underlying factors, which capture 84 percent of the variance in the original data. Table 4 provides the component matrix obtained using Varimax rotation. The factors are labeled: reliability, shopping convenience, certification, and shipping and handling. Since the factors explain a high proportion of the variance in the data, we employ factor scores as our measure of e-tailer services. The service measures employed in our analysis are related to the dimensions of retail services specified by Betancourt and Gautschi (1993). Reliability corresponds to Betancourt and Gautschi’s assurance of product delivery dimension, shopping 24 convenience is related to their assortment, accessibility and ambiance dimensions, product information is related to their availability of information dimension, The certification dimension is related to their assurance of product delivery dimension and also, as pointed out above, to the trust dimension specified by Brynjolfsson and Smith (2000). As the statement about shipping and handling is worded in BizRate, this dimension relates mainly to shipping and handling charges and options. (Table 4 about here) Using scores on the service factors and the measures of trust as independent variables, hedonic regressions of the form outlined in Equation 10 were run on the pooled data for 2000 and 2001. (10.) ( p mrt p mt − 1) = ∑6k =1 bk (RS mrkt − R S mkt ) + ν mrt , where m is item, r is e-tailer, k is attribute, t is time period, and ν is an error term. Because effects get magnified as service levels increase, we divide by the mean of that item’s price in the corresponding time period to stabilize the error variance.10 Measuring all effects as deviations from item means within a given time period eliminates item effects due to generally high or low levels of attributes for sellers of that item. It has the same general effect as including a set of item dummy variables, and creates a zero intercept. We pool across time periods to make our quality adjustment consistent over time. Regressions were run for each major category. Results are presented in Table 5. Though all are statistically significant, none of the regressions in Table 5 has a high Rsquared value. Differences in e-tailer services, at least the ones measured in our data, do 10 The form in Equation 10 gave better results than deviations from average prices, and the ratio of log of price to its mean. However, different functional forms gave similar results. 25 not explain a great degree of the variation in e-tailer prices, contrary to the hedonic hypothesis that services explain price dispersion.11 In addition, one would generally expect positive signs on the various coefficients if this hypothesis is true, since these coefficients would measure marginal willingness to pay under this hypothesis, and one would expect consumers to be willing to pay non-negative amounts for each attribute.12 This phenomenon of wrong signs in regressions of prices on e-tailer service characteristics was also noted by Brynjolfsson and Smith (2000), who rejected using hedonic regressions partly for this reason. While the negative signs provide evidence that Rosen’s (1974) model of hedonic prices under perfect competition does not hold for our data, Pakes (2001) shows that they are possible in more general settings. Thus the estimates in Table 5 are not necessarily biased or otherwise problematic. Consequently we use these in developing our estimates of quality-adjusted prices. (Table 5 about here) Among the variables in our regressions, shipping and handling tends to have the largest effect, which is positive in 7 out of 8 cases. Empirically favorable shipping and handling charges tend to be accompanied by higher prices. Even when significant, however, effects are generally small relative to the observed variation in prices. For example, an increase of one unit (standard deviation) in the factor score for shipping and handling increases the ratio of price to its mean by .058 (increases price by approximately 5.8 percentage points) for books. Among items, prices of books are explained best by the four service factors, prices of CDs second best. However most of 11 The same point is made from our 2000 data in Pan, Ratchford and Shankar (2002). While the within-category nature of our analysis does create some correlations between the four attributes, these are small so that multicollinearity does not appear to be a serious problem in our regressions. 12 26 the explanatory power for CDs comes form the negative effect of certification, which is difficult to interpret. A general conclusion is that the observed dispersion in e-tailer prices is not explained to any great degree by the variation in e-tailer services. The hedonic explanation for price dispersion can be rejected. 13 While the hedonic explanation appears not to hold, we still wish to make empirical comparisons between quality-adjusted prices and unadjusted prices. Using the regression coefficients from Table 5, we calculated a quality-adjusted price for each item according to the above formula. (11.) ( ) a = p mt (1 + ν mrt ) = p mt ( p mrt p mt ) − ∑6k =1 bˆk (RS mrkt − R S mkt ) . p mrt Using the logic outlined in Equation 9, Equation 11 expresses the item’s price adjusted for service quality at time t as price less the effect of deviations from the average level of attributes on price. It is therefore an estimate of what the price of any item would be if it had an average level of attributes. Analysis of Prices. In this section we compare changes in dispersion in prices and quality-adjusted prices between the 2000 and 2001 samples. The comparison in Table 6 employs two general measures of price dispersion that are commonly used in studies of this phenomenon. One is percentage difference, where percentage price difference is defined as 100*(range of item prices/mean item price). For example, across the 104 books in the 2000 sample, the average book has a price range of 48.9% of its mean price. Our other measure is the coefficient of variation. Because both measures expressed relative to price, they have the advantage of controlling for price differences across categories and years. 13 Running regressions on raw attributes rather than factor scores did not lead to substantial improvements in fit. 27 (Table 6 about here) Table 6 presents estimates for both unadjusted prices and quality-adjusted prices defined as in Equation 11. Except for books and CDs in 2001, the quality adjustment procedure does not have much effect on the dispersion measures, which is not surprising given the low R2 values in the corresponding regressions. The dispersion in prices reported in Table 6 appears to be quite large, and our estimates of price percentage difference for books and CDs are somewhat larger than Brynjolfsson and Smith (2000), possibly because our sample contains more e-tailers. However, with the exception of books and PDA, dispersion declined significantly for all measures between 2000 and 2001, consistent with Hypothesis 1. Incorporating the quality adjustment also led to a decline in dispersion for books. The decline in measured dispersion between 2000 and 2001 is especially large for items in the desktop computer and software, with the average dispersion cut approximately in half between those periods. Earlier we showed that the return to search for an uninformed consumer is equal to the difference between the average and lowest price (Equation 4), while an upper bound on search costs is related to the difference between the maximum price and the average price (Equation 7). We calculated these measures for the items in our sample using both unadjusted and quality-adjusted prices. Results for the 2000 and 2001 samples are presented in Table 7. As shown in Table 7, differences between mean price and minimum price show no clear pattern of change between the two time periods. On the other hand Table 7 shows a clear pattern of decrease in all categories for differences between maximum and mean price, which are related to the highest cost of search. These decreases are significant at the .05 level or better in five of the eight cases for both 28 unadjusted and adjusted prices.14 Further inspection of Table 7 and Table 5 shows that the distribution of prices was skewed above the mean in 2000, but is roughly symmetric in 2001. Evidently there were fewer really high prices on the market in 2001, which is consistent with a decrease in the highest search costs. (Tables 7 about here) To further examine whether there were general changes in price dispersion between the samples, we regressed the measures of dispersion in Table 6 on the following variables: • • • • Ln (item average price) – a control for the possibility that dispersion relative to price might decline with the price level since search costs are unlikely to increase proportionally with prices. The natural log gave slightly better results than a linear term (overall results were insensitive to this choice). Linear and quadratic terms in number of firms – to capture the effect of number of firms on dispersion, which Baye, Morgan and Scholten (2001) and Pan, Ratchford and Shankar (2002) found to be nonlinear. Category dummies – to control for effects that are idiosyncratic to category. A dummy = 1 in 2001, 0 otherwise. The results of this analysis are presented in Table 8. Holding other factors constant, the coefficients of the 2001 dummy in Table 8 indicate a significant decline in price dispersion. For unadjusted prices the decline is approximately 18 percent relative to the 2000 mean for both dispersion measures. For adjusted prices the corresponding decline is approximately 24 percent for both measures. The general results in Table 6 hold up across categories when price levels and numbers of firms are controlled for. Consistent with H2, the coefficients of the number of firm variables imply that price dispersion increases with number of e-tailers until it hits its maximum at about 15 for the percent 14 Weighting the highest price by (n+1/n) as derived in our theoretical section gave qualitatively similar results. We preferred to present the unweighted results in Table 6 since the distributions are easier to interpret. 29 price difference measure, and about 10 for the coefficient of variation measure.15 Finally the results in Table 8 indicate that relative dispersion declines with price. (Table 8 about here) Conclusions One of the results in this paper is that price dispersion in the Internet markets studied declined substantially between November 2000 and November 2001. This reflects a maturation of these markets, and, while other explanations are possible (e.g., increased collusion), these results are consistent with improvements in information and consequent gains in consumer welfare. Our finding of decreased price dispersion is contrary to findings of no trend in Clay, Krishnan and Wolff (2001) and Baye, Morgan and Scholten (2001). Both of these studies were done on data from an earlier time period (until March 2001, beyond the Internet markets started witnessing significant shake-outs), which may account for the difference in results. Our data reject differences in e-tailer services as a major driver of observed price dispersion over time. While it is possible that this was due to shortcomings of our measures of services, our measures were consistent with existing theories of retail services, and were obtained from the best source of this information that we are aware of. It seems much more likely that explanations for price dispersion that rest on costly information are correct. With the exception of our results for the book category, it did not make much difference to our results whether unadjusted or prices adjusted for measured differences in service were used. An implication is that analysts are generally safe in 15 Our results for the relation between number of firms and coefficient of variation are similar to Baye, Morgan and Scholten (2001); however they found that dispersion increased with number of firms for a percentage of price difference measure. 30 working with unadjusted prices, as has generally been done in the theoretical and empirical literature on price dispersion. Our model of consumer surplus may provide a useful insight into why there is a large degree of price dispersion in Internet markets even though these markets allow information to be gathered relatively quickly without traveling to retailers. Because it is a time saver, the Internet should appeal to those with very high time costs who do not find it cost effective to search. These consumers will be willing to accept very high prices. At the same time, if the Internet allows relatively efficient search, consumers who do not have such high time costs might be able to locate attractive selling prices expeditiously. The existence of groups with radically different search costs may help to drive Internet price dispersion. To the extent that it indicates that Internet prices are generally lower than for comparable items at brick and mortar retailers, the existing evidence indicates that the Internet improves consumer welfare. We also should point out at this point that price dispersion in Internet markets does not in itself indicate allocative inefficiency. While consumers who pay high prices may lose, producers may capture corresponding gains, and gains and losses may cancel. While active intervention in markets is currently unfashionable, and is not something that we would advocate ourselves, there has been a long standing interest in intervening to eliminate inefficiencies in retail markets (Maynes and Assum 1982). Given this history, this issue could again surface for Internet markets. Our finding that online price dispersion declined over a one-year period suggests that interest in this issue may be 31 premature. Further study of trends in the behavior of prices in e-tail markets would be helpful for monitoring their efficiency. It would also be useful to learn if our conjecture about the identity of those paying high prices on the Internet is correct. Consumers with a very high value for time saving are likely to be wealthy, and not the type that most people would like to help. They are quite different from the poor people who must pay high prices in conventional retail markets because of their lack of mobility. This leads to another potential policy issue of making the benefits of the Internet more accessible to those who currently lack access to this medium or the knowledge of how to use it, that is, the issue of “digital divide” that has been gaining a lot of attention. If the Internet does lead to lower prices, and is able to overcome mobility constraints, steps to promote its use may be warranted. A key missing piece of data limits the applicability of this and other studies of pricing behavior in Internet markets. We generally observe only posted prices, and do not know how many sales take place at each price. Actual sales data similar in scope to storelevel scanner data available in conventional markets are needed for Internet markets. Another data need is more comprehensive data on prices in conventional retail markets. This would allow more comprehensive comparisons of price levels between markets, and allow more general statements about price levels in the Internet vs. conventional markets than can be drawn from the limited number of product categories that have been compared to date. 32 Table 1 Summary of Relevant Results in Brynjolfsson and Smith (2000) Type Conventional Internet Difference Mean Prices – Internet Share Weighted Book $13.90 $11.74 $2.16 CD $16.07 $13.49 $2.58 Mean Prices - Unweighted Book $13.90 $12.68 $1.22 CD $16.07 $13.78 $2.29 Minimum Prices Book na na $1.29 CD na na $1.40 Mean Full Prices - Internet Share Weighted Book $15.04 $13.69 $1.35 CD $17.41 $15.15 $2.26 Minimum Full Prices Book na na $1.09 CD na na $1.23 Range of Full Prices Book na $5.98 na CD na $4.45 na 33 Table 2 Price Level by Category: Comparison of 2001 and 2000 E-tailer Samples Mean & Mean & Obs. Difference Obs. Std. Dev. t-value Category Std. Dev. (2001) in Mean (2000) (2001) (2000) Book 20.65 (22.89) 105 19.26 (26.04) 134 -1.39 -0.44 CD 13.51 (1.66) 43 14.64 (6.53) 120 1.13 1.74# DVD 26.64 (18.73) 96 22.53 (10.79) 103 -4.11 -1.88# Desktop 1209.7 (1077.7) 105 2509.7 (2766.1) 107 1300 4.52** Laptop 2391.6 (653.77) 78 1981.3 (563.55) 96 -410.31 -4.38** PDA 446.86 (317.97) 37 350.70 (205.44) 52 -96.17 -1.62 Software 281.42 (685.26) 51 597.31 (1383.1) 120 315.9 1.99* Consumer Electronics 440.24 (498.74) 66 671.94 (710.78) 94 231.7 2.42* ** Significant at .01. * Significant at .05. # Significant at .10. 34 Table 3 Measures and Explanation of e-Tailers’ Features by BizRate.com Measure Explanation Ease of Ordering Convenience and speed of ordering Product Selection Breadth/Depth of products offered Product Information Information quantity, quality and relevance Web Site Navigation and Looks Layout, links, pictures, images and speed On-Time Delivery Expected vs. actual delivery date Product Representation Level and Quality of Customer Support Tracking Product description/depiction vs. what you received Status updates and complaint/question handling Shipping and Handling Shipping and handling charges and options Certify 5 No. of certifications from 5 agencies Years Certified No. years certified by BizRate.com Tracking order status Table 4 Factor Analysis of e-tailer Services: Rotated Component Matrixa Component Variable 1 2 Ease of Ordering 0.165 0.891 0.064 0.248 Product Selection 0.212 0.833 0.197 0.083 Product Information 0.453 0.629 0.068 -0.099 Web Site Navigation 0.160 0.914 0.144 0.138 On-Time Delivery 0.921 0.166 0.109 0.116 Product Representation 0.729 0.392 0.325 0.060 Customer Support 0.908 0.152 -0.015 0.263 Tracking 0.908 0.232 0.051 0.109 Shipping & Handling 0.314 0.222 0.116 0.853 Certify 5 0.136 0.124 0.898 -0.114 Years_Certified 3 4 0.046 0.177 0.352 0.810 Reliability Shopping Certification Shipping and Factor Name Convenience Handling a Rotation method is Varimax. 35 Table 5 Results of Pooled Regressions of Normalized Prices on Attributes by Categorya Independent Variable Shopping Shipping & Category Parameter Reliability Convenience Certification Handling R2 N Book Estimate 0.054 -0.009 0.008 0.058 0.222 2172 t Value 11.93 -3.68 1.83 21.58 CD Estimate 0.014 0.008 -0.069 0.008 0.176 1275 t Value 3.42 1.73 -15.52 1.36 Desktop Estimate 0.002 -0.010 -0.016 0.027 0.082 1721 t Value 0.60 -2.83 -5.03 10.19 DVD Estimate 0.002 -0.015 -0.023 -0.026 0.084 2309 t Value 0.74 -5.86 -7.78 -7.25 Electronics Estimate 0.011 0.002 0.006 0.017 0.077 1478 t Value 5.07 0.70 2.18 8.39 Laptop Estimate 0.000 0.007 -0.002 0.008 0.021 1871 t Value -0.11 3.51 -0.71 3.94 PDA Estimate 0.007 -0.011 -0.004 0.016 0.040 1039 t Value 2.01 -2.19 -0.85 4.83 Software Estimate -0.012 -0.012 0.012 0.025 0.105 1636 t Value -4.65 -3.88 4.00 12.40 a Regressions have the functional form outlined in Equation.10. 36 Table 6 Change in Price Dispersion Relative to Average Price Across Categories Sample Size Dispersion in Prices Dispersion in Adjusted Prices 2000 2001 2000 2001 Diff. t 2000 2001 Diff. t Percentage of Price Difference Book 105 134 48.90 48.08 -0.82 -0.47 49.16 34.38 -14.77 -8.31** CD 43 120 51.04 39.30 -11.74 -3.62** 49.61 31.31 -18.30 -5.13** Desktop 105 107 34.39 15.01 -19.38 -6.83** 36.19 15.83 -20.36 -7.68** DVD 96 103 43.67 32.28 -11.39 -4.57** 38.34 33.61 -4.73 -2.02* Electronics 66 94 30.99 22.12 -8.87 -4.38** 31.18 22.31 -8.87 -4.36** Laptop 78 96 25.70 17.87 -7.82 -3.44** 25.91 17.78 -8.13 -3.68** PDA 37 52 37.10 30.26 -6.84 -1.47 36.37 30.88 -5.49 -1.23 Software 51 120 35.58 18.95 -16.63 -4.44** 36.09 17.15 -18.94 -4.36** Price Coefficient of Variation Book 105 134 15.29 16.63 1.34 2.47* 15.47 11.93 -3.54 -7.00** CD 43 120 15.46 13.02 -2.45 -2.61** 14.93 10.96 -3.97 -3.98** Desktop 105 107 10.78 5.46 -5.32 -6.36** 10.56 5.71 -4.85 -6.42** DVD 96 103 13.05 10.22 -2.84 -3.25** 11.94 10.42 -1.52 -1.79# Electronics 66 94 9.65 8.22 -1.44 -2.33* 9.33 7.81 -1.51 -2.50* Laptop 78 96 7.55 6.11 -1.44 -1.97* 7.54 6.05 -1.48 -2.08* PDA 37 52 10.49 9.86 -0.62 -0.48 10.22 9.77 -0.46 -0.35 Software 51 120 10.55 6.51 -4.04 -3.90** 10.10 5.79 -4.31 -4.35** ** Significant at .01. * Significant at .05. # Significant at .10. 37 Table 7 Change in Dispersion Above and Below Average Price Across Categories Sample Size Mean Price Difference Mean Adjusted Price Difference 2000 2001 2000 2001 Diff. t 2000 2001 Diff. t Difference Between Average and Minimum Price Book 105 134 4.56 4.18 -0.39 -0.62 4.76 2.84 -1.92 -3.23** CD 43 120 2.43 2.44 0.02 0.07 2.41 2.26 -0.16 -0.70 Desktop 105 107 94.32 171.64 77.32 3.01** 143.70 183.10 39.40 1.40 DVD 96 103 4.35 3.19 -1.16 -2.35* 4.26 3.43 -0.83 -1.77# Electronics 66 94 63.19 79.74 16.55 1.16 57.28 67.33 10.05 0.87 Laptop 78 96 300.35 157.21 -143.13 -4.10** 299.37 156.67 -142.70 -4.26** PDA 37 52 39.68 52.70 13.02 1.35 40.53 50.55 10.01 1.03 Software 51 120 34.87 52.89 18.02 1.02 39.47 52.09 12.61 0.69 Difference Between Maximum and Average Price Book 105 134 5.69 3.80 -1.89 -2.50* 5.76 3.63 -2.13 -2.32* CD 43 120 4.57 3.29 -1.28 -2.91** 4.40 2.32 -2.08 -4.61** Desktop 105 107 213.83 188.71 -25.13 -0.87 209.88 196.24 -13.64 -0.46 DVD 96 103 7.31 3.79 -3.52 -4.88** 5.90 3.85 -2.05 -3.36** Electronics 66 94 80.30 61.41 -18.90 -1.25 82.35 70.10 -12.25 -0.80 Laptop 78 96 336.16 186.67 -149.49 -4.60** 340.53 185.81 -154.72 -4.80** PDA 37 52 95.73 49.41 -46.32 -2.50* 92.20 53.59 -38.61 -2.55* Software 51 120 102.82 54.78 -48.04 -0.81 99.96 44.53 -55.43 -0.99 ** Significant at .01. * Significant at .05. # Significant at .10. 38 Table 8 Determinants of Price Dispersion Relative to Average Price Pct. Price Difference Dependent Price Coefficient of Variation Dependent Unadjusted Adjusted Unadjusted Adjusted Variable Estimate t Value Estimate t Value Estimate t Value Estimate t Value Intercept 44.005 8.65 27.221 5.57 18.742 11.73 13.874 9.13 Ln (Avg. Price) -3.229 -5.41 -2.221 -3.87 -1.123 -5.99 -0.777 -4.36 No. Firms 2.685 3.50 4.052 5.49 0.345 1.43 0.662 2.89 Firms Sq. -0.085 -2.67 -0.140 -4.58 -0.019 -1.87 -0.032 -3.32 CD -3.617 -2.10 -1.445 -0.88 -2.084 -3.86 -0.932 -1.82 Desktop -8.732 -2.83 -4.043 -1.36 -3.062 -3.16 -1.976 -2.14 DVD -11.255 -6.71 -6.138 -3.81 -3.832 -7.27 -1.885 -3.76 Electronics -11.641 -4.53 -6.777 -2.74 -3.422 -4.24 -2.263 -2.95 Laptop -11.978 -3.52 -9.266 -2.84 -3.522 -3.30 -2.692 -2.65 PDA -6.546 -2.34 -1.722 -0.64 -2.081 -2.37 -0.668 -0.80 Software -15.565 -7.01 -11.034 -5.17 -5.144 -7.38 -3.913 -5.91 2001 Dummy -6.998 -6.25 -9.008 -8.38 -2.126 -6.05 -2.816 -8.43 R-square N 0.325 1407 0.279 1407 39 0.311 1407 0.254 1407 References Alba, Joseph, John Lynch, Barton Weitz, Chris Janiszewski, Richard Lutz, Alan Sawyer and Stacy Wood (1997), “Interactive Home Shopping: Consumer, Retailer, and Manufacturer Incentives to Participate in Electronic Marketplaces,” Journal of Marketing, 61 (July), 38-53. Ancarani, Fabio and Venkatesh Shankar, (2002). Price Levels and Price Dispersion: A Comparison of Pure Play vs. Bricks-and-Mortar vs. Bricks-and-Clicks e-Tailers. Working Paper, SDA Bocconi, Milan, Italy. Bakos, Yannis (1997). “Reducing Buyer Search Costs: Implications for Electronic Marketplaces.” Management Science, 43 (December), 1676-1692. Baye, Michael R. and John Morgan (2001), “Information Gatekeepers on the Internet and the Competitiveness of Homogeneous Product Markets.” American Economic Review, 91 (June), 454-474. _________, _________ and Patrick Scholten (2001), “Price Dispersion in the Small and in the Large: Evidence from an Internet Comparison Site.” Working Paper, Indiana University. Betancourt, Roger R. and David Gautschi (1993). “Two Essential Characteristics of Retail Markets and Their Economic Consequences.” Journal of Economic Behavior & Organization, 21 (August), 277-294. Brynjolfsson, Erik and Michael Smith (2000). “Frictionless Commerce? A Comparison of Internet and Conventional Retailers.” Management Science, 46 (April), 563-585. Burdett, Kenneth and Melvyn G. Coles (1997), “Steady State Price Distributions in a Noisy Search Equilibrium.” Journal of Economic Theory, 72, 1-32. _________ and Kenneth Judd (1983). “Equilibrium Price Dispersion.” Econometrica, 51 (July), 955-969. Brown, Jeffrey R. and Austan Goolsbee (2002), “Does the Internet Make Markets More Competitive? Evidence from the Life Insurance Industry.” Journal of Political Economy, 110 (June), 481-507. Carlson, John and Preston McAfee (1983). “Discrete Equilibrium Price Dispersion” Journal of Political Economy, 91 (June), 480-493. Clay, Karen, Ramayya Krishnan and Eric Wolff (2001), “Prices and Price Dispersion on the Web: Evidence from the Online Book Industry.” Working Paper, Carnegie Mellon University. 40 Clemons, Erik, Il-Horn Hann and Lorin Hitt (2002), “Price Dispersion and Differentiation in Online Travel: An Empirical Investigation.” Management Science, 48 (April), 534-549. Degeratu, Alexandru, Arvind Rangaswamy and Jianan Wu (2000), “Consumer Choice Behavior in Online and Traditional Supermarkets: the Effect of Brand Name, Price, and other Search Attribute,” International Journal of Research in Marketing, March, 55-78. DePalma , Andre, Gordon M. Myers and Yorgos Y. Papageorgiou (1994). “Rational Choice Under Imperfect Ability to Choose.” American Economic Review, 83 (June), 419440. Ehrlich, Isaac and Lawrence Fisher (1982). “The Derived Demand for Advertising: A Theoretical and Empirical Investigation.” American Economic Review, 72 (June), 366388. Ellison, Glenn and Sara Fisher Ellison (2001), “Search, Obfuscation and Price Elasticities on the Internet.” Working Paper, MIT. Lal, Rajiv and Miklos Sarvary (1999), “When and How is the Internet Likely to Decrease Price Competition?” Marketing Science, 18 (4), 485-503. Lynch, John and Dan Ariely (2000), “Wine Online: Search Costs Affect Competition on Price, Quality and Distribution.” Marketing Science, 19 (Winter), 83-103. Maynes, E. Scott and Terje Assum (1982). “Informationally Imperfect Consumer Markets: Empirical Findings and Policy Implications.” Journal of Consumer Affairs, 16 (Summer), 62-87. Morton, Fiona Scott, Florian Zettelmeyer and Jorge Silva-Risso (2001), “Internet Car Retailing.” Journal of Industrial Economics, 49(4), 501-519. Pakes, Ariel (2001), “A Reconsideration of Hedonic Price Indices with an Application to PC’s.” Working Paper, Harvard University. Pan, Xing, Brian T. Ratchford, and Venkatesh Shankar (2001). “Why Aren’t the Prices of the Same Item the Same at Me.com and You.com?: Drivers of Price Dispersion Among E-Tailers,” Working Paper, University of Maryland, College Park, MD 20742. _______ , _______, and _______ (2002). “Can Price Dispersion in Online Markets be Explained by Differences in e-Tailer Service Quality?” Journal of the Academy of Marketing Science, 30 (4), 429-441. ________, Venkatesh Shankar, and Brian T. Ratchford (2002) “Price Competition Between Pure Play vs. Bricks-and-Clicks E-Tailers: Analytical Model and Empirical 41 Analysis,” Advances in Microeconomics, Vol. 11, The Economics of the Internet and ECommerce, forthcoming. Ratchford, Brian T., Jagdish Agrawal, Pamela E. Grimm, and Narasimhan Srinivasan (1996). “Toward Understanding the Measurement of Market Efficiency.” Journal of Public Policy and Marketing, 15 (Fall), 167-184. _______ , Myung Soo Lee and Debabrata Talukdar (2002), “The Impact of the Internet on Information Search for Automobiles.” Journal of Marketing Research, forthcoming. Rosen, Sherwin (1974). “Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition.” The Journal of Political Economy, 82 (1), 34-55. Salop, Steven and Joseph Stiglitz (1982). “The Theory of Sales: A Simple Model of Equilibrium Price Dispersion with Identical Agents.” The American Economic Review, 72 (December), 1121-1130. Shankar, Venkatesh, Arvind Rangaswamy and Michael Pusateri (2001). “The Online Medium and Customer Price Sensitivity.” Working Paper, University of Maryland, College Park, MD 20742. Small, Kenneth A. and Harvey S. Rosen (1981). “Applied Welfare Economics With Discrete Choice Models.” Econometrica, 49 (January), 105-130. Smith, Michael and Erik Brynjolfsson (2001), “Consumer Decision Making at an Internet Shopbot.” Sloan Working Paper 4206-01, MIT Sloan School of Management.. Stahl, Dale (1989), “Oligopolistic Pricing and Sequential Consumer Search.” American Economic Review, 79 (September), 700-712. Tang, Fang-Fang and Xing, Xiaolin (2001), “Will The Growth Of Multi-Channel Retailing Diminish the Pricing Efficiency of the Web?,” Journal of Retailing, 77, 319333. Varian, Hal (1980), “A Model of Sales.” American Economic Review, 70 (4), 651-659. Zettelmeyer, Florian (2000), “Expanding to the Internet: Pricing and Communication Strategies When Firms Compete on Multiple Channels,” Journal of Marketing Research, 37 (3), 292-308. 42