University of Southern California Dissertations and Theses

An empirical investigation of everyday low price (EDLP) strategy in electronic markets

(USC Thesis Other)

An empirical investigation of everyday low price (EDLP) strategy in electronic markets

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content AN EMPIRICAL INVESTIGATION OF EVERYDAY LOW PRICE (EDLP) STRATEGY IN ELECTRONIC MARKETS by Raymond Gee Han Sin A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (BUSINESS ADMINISTRATION) August 2005 Copyright 2005 Raymond Gee Han Sin R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. UMI Number: 3196892 Copyright 2005 by Sin, Raymond Gee Han All rights reserved. INFORMATION TO USERS The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion. ® UMI UMI Microform 3196892 Copyright 2006 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. DEDICATION To God, my parents, and my beloved wife R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. ACKNOWLEDGEMENTS I would like to express my deepest respect and gratitude to my advisor and chairperson of my doctoral dissertation, Dr. Ramnath K. Chellappa, for his tremendous help and guidance ever since the day I joined the Ph.D. program. His persistence, creativity, and passion for research have inspired me and shaped me to become the researcher I am today. My most sincere appreciation go to my committee members, including Dr. Omar A. El Sawy, Dr. 11-Horn Harm, and Dr. Sivaramakrishnan Siddarth. Their continuous support and insightful comments have made possible the successful completion of this dissertation. I also wish to thank my parents for their unconditional love and support, and am especially grateful to my wife Vivian for her patience, sacrifices, and for her unfailing love and belief in me during the challenging years of my graduate study. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. iv TABLE OF CONTENTS Dedication ii Acknowledgements iii List of Tables vii Abstract viii CHAPTER 1: INTRODUCTION 1 CHAPTER 2: LITERATURE REVIEW 4 2.1 Overview 4 2.2 Literature on Retail Price Format 5 2.2.1 Price Format and Consumer Segmentation 6 2.2.2 Viability of EDLP in Electronic Markets 11 2.2.3 Dimensions of EDLP 14 2.2.4 Category-level Adoption of EDLP 16 2.3 Literature on Price Dispersion 18 2.3.1 Price Dispersion Due to Product Heterogeneity 19 2.3.2 Price Dispersion Due to Consumer Heterogeneity 19 2.3.3 Price Dispersion Due to Vendor Heterogeneity 27 2.3.4 Empirical Findings on Online vs. Offline Price Dispersion 29 2.3.5 Random Pricing 33 2.3.6 Measuring Price Dispersion 35 2.4 Literature on Airline Pricing 40 2.4.1 Market Power 40 2.4.2 Competition 44 2.4.3 Market Structure 46 2.4.4 Cost 47 CHAPTER 3: DATA AND METHOD 51 3.1 Data 51 3.1.1 Overview 51 3.1.2 Collection 52 3.1.3 Bias Control 54 3.1.4 A Note on EDLP Definition 55 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. V 3.2 Method 56 3.2.1 Introduction to Hierarchical Modeling 56 3.2.2 Idea Behind HM 58 3.2.3 Estimation Methods 60 CHAPTER 4: A STUDY OF PRICING STRATEGY OF ONLINE EDLP VENDORS 63 4.1 Conceptual Development 63 4.2 Models 71 4.2.1 Test of Hypothesis 1: Price Levels 73 4.2.2 Test of Hypothesis 2a: Temporal Price Variability 74 4.2.3 Test of Hypothesis 2b: Price Variability within Markets 76 4.2.4 Test of Hypothesis 3a: Intertemporal Price Range 77 4.2.5 Test of Hypothesis 3b: Price Range within Markets 78 4.2.6 Test of Hypothesis 4: Individual Seller’s Mean Price versus Market Minimum 79 4.3 Results 80 4.3.1 Test of Model Specification and Robustness 80 4.3.2 Results of Hypothesis 1: Price Levels 86 4.3.3 Results of Hypothesis 2a: Temporal Price Variability 87 4.3.4 Results of Hypothesis 2b: Results of Hypothesis 2b: Price Variability within Markets 89 4.3.5 Results of Hypothesis 2c: Results of Hypothesis 2c: Price Variability Across Markets 90 4.3.6 Results of Hypothesis 3a: Intertemporal Price Range 92 4.3.7 Results of Hypothesis 3b: Price Range within Markets 93 4.3.8 Results of Hypothesis 4: Firm’s Mean Price vs. Market Minimum 94 4.3.9 Results of Hypothesis 5: Category-level Adoption of EDLP 97 4.4 Discussion 98 CHAPTER 5: PRICE FORMAT AND PRICE DISPERSION 102 5.1 Conceptual Development 102 5.2 Models 106 5.2.1 Test of Hypothesis 6: Price Format and Price Dispersion 107 5.2.2 Test of Hypothesis 7: HILO Reactions 110 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. vi 5.3 Results 111 5.3.1 Results of Hypothesis 6: Price Format as a Source of Price Dispersion 111 5.3.2 Results of Hypothesis 7: HILO Reactions 115 5.3.3 Results of Hypothesis 8: Price Dispersion and Consumer’s Reservation 116 5.3.4 Results of Hypothesis 9: Online vs. Offline Price Dispersion 117 5.4 Discussion 118 CHAPTER 6: SUMMARY AND CONCLUSIONS 120 6.1 Summary of Results 120 6.1.1 Pricing Strategies of Online EDLP Vendors 120 6.1.2 Price Format and Price Dispersion 122 6.2 Managerial and Policy Implications 123 6.2.1 Applications of EDLP in the Airline Industry 123 6.2.2 Cost Structure Does Not Determine Pricing Strategy 124 6.2.3 Strategic Variability in EDLP Pricing Online 125 6.2.4 Policy Implications 126 6.3 Contributions 127 6.4 Limitations and Future Research 130 REFERENCES 133 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. vii LIST OF TABLES Table 1: Explanation of Variables 5 3 Table 2: Hausman Test 81 Table 3: Results of Model 1 - Price Levels (Business Tickets) 83 Table 4: Results of Model 1 - Price Levels (Leisure Tickets) 85 Table 5: Results of Model 2 a - Results of Hypothesis 2a: Temporal Price Variability 87 Table 6: Results of Model 2b - Results of Hypothesis 2b: Price Variability within Markets 89 Table 7: Variability in Median Prices Across Markets 91 Table 8: Results of Model 3a - Intertemporal Price Range 92 Table 9: Results of Model 3b - Price Range within Markets 93 Table 10: Results of Model 4 - Mean Price vs. Market Minimum 94 Table 11: Frequency of an Airline Charging Market Minimum Prices 96 Table 12: Results of Model 6a - Price Dispersion and Price Format 111 Table 13: Results of Model 6b - Price Dispersion Across Markets 112 Table 14: Prices Dispersion and Price Format - Online and Offline Channels 113 Table 15: Prices Dispersion Across Markets - Online and Offline Channels 114 Table 16: Results of Model 7 - Reactions by HILO Sellers 115 Table 17: Price Dispersion Online vs. Offline 117 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. viii ABSTRACT Retail price formats are important strategic tools for traditional brick-and-mortar retailers. The most popular price formats are “Everyday Low Price” (EDLP) and “Promotional Pricing” (HILO). Sellers who adopt EDLP tend to charge relatively stable, below average prices; while those who adopt HILO are promotion-oriented and charge higher prices on average but engage in frequent promotions. While managing the “low price” perception for EDLP sellers is much easier in the physical context given the relatively high costs required for consumers to uncover the difference in prices across stores, the advent of electronic commerce and the ability of consumers to easily price-search multiple stores pose new challenges to sellers adopting this particular price format in electronic markets. This study examines the role of price format adoption in competition in electronic markets and offers insights on alternative explanations for persistent price dispersion online. Using a hierarchical modeling approach and 1,137,500 unique price observations obtained from online travel agents as well as individual airlines’ websites in 2004, this research aims to address the following questions: 1) Do self-declared EDLP sellers indeed charge low, stable prices online? 2) Is EDLP being adopted in electronic markets in the same fashion as in physical markets? 3) How does this practice affect the overall pricing structure in the market? R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. The results of this study show that the extent to which online sellers adopt everyday low price strategy varies with different types of market and product categories, offering the first evidence of category-level adoption of price format. Further, online EDLP sellers focus more on the “within-markef ’ characteristics of this pricing strategy rather than the “across-time” characteristics, implying a diminishing role of intertemporal price consistency to the practice of everyday low price in the online environment. Analysis on the relationship between price format and online price dispersion yields strong evidence that vendors’ adoption of different price formats contributes to price dispersion in both online and offline contexts. Finally, the degrees of dispersion in prices are found to be significantly lower online compared to that in physical markets. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 1 CHAPTER 1 INTRODUCTION Retail price formats have long been recognized as important strategic tools for traditional brick-and-mortar retailers. The most popular price formats are “Everyday Low Price” (EDLP) and “Promotional Pricing” (HILO). Sellers who adopt promotional pricing aim to attract consumers by using frequent “deals” and discounts on certain featured products that change on a weekly or monthly basis. Vendors who adopt EDLP, on the other hand, focus on creating a “low price” image to persuade consumers that regardless of what items they buy or when they buy it, they can always expect below average prices. EDLP has proven to be a successful strategy in the physical context. A classic example is Wal-Mart, the retail giant famous for its “everyday low price” motto1 . Wal-Mart strategically locates their stores away from malls and other specialty stores that engage in frequent price discounts. By making price comparison costly to consumers, they are able to avoid head-to-head price comparison and maintain a low price reputation even though their prices may not be the lowest in the area. Since EDLP is a conscious pricing strategy that aims to create a “low price” perception, it is the goal of this strategy to impart such a perception in consumer’s 1 Recently Wal-Mart further emphasized its EDLP strategy by changing their advertising slogan to “Always low prices. Always". R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 2 mind - even if the actual pricing behaviors of sellers adopting this strategy are not consistent with their claim. A basic premise for the success of EDLP, therefore, appears to be that consumers will encounter difficulties in searching for the lowest price offered by each and every vendor in the market. While managing this perception in the physical context is much easier given the relatively high costs required to uncover different prices across stores, the advent of electronic commerce and the ability of consumers to price-search multiple stores with a few simple mouse-clicks pose new challenges to sellers adopting this particular price format in electronic markets. Is it feasible to adopt the “Everyday Low Price” in the online environment? Are there any such examples? If so, are the pricing behaviors of these sellers truly consistent with the image they deliver? Further, given that sellers who adopt different price formats intentionally vary prices to different extents, can price formats be a factor that contributes to online price dispersion? This study examines price format as an important strategic tool for competition in electronic markets and offers insights on alternative explanations for persistent price dispersion online. Further, it provides managerial guidelines to online firms in adopting different pricing strategies, taking into account the specific nature of electronic markets. Using a hierarchical modeling approach and 1,137,500 unique price observations obtained from online travel agents as well as individual airlines’ websites in 2004, this research aims to address the following questions: 1) Do self-declared EDLP sellers indeed charge low, stable prices online? 2) Is EDLP R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 3 being adopted in electronic markets in the same fashion as in physical markets? 3) How does this practice affect the overall pricing structure in the market? Specifically, does it have any effects on online price dispersion? This research offers new theoretical perspectives and empirical evaluations on the strategic pricing of online retailers with five major contributions. First, this study offers the first evidence of category-level adoption of price format and generalizes research in EDLP to electronic markets. Second, it uncovers new dimensions in which the “everyday low price” strategy is adopted and contrasts the characteristics of EDLP pricing between online and offline markets. Third, this study identifies vendors’ conscious adoption of different price format as an alternative explanation for dispersion in prices, hence furthers the understanding of the nature of online price dispersion. Fourth, it offers a rationalization that reconciles the discrepancies in findings by existing empirical research regarding the random pricing theory. Fifth, based on most recently available data, this study is one of the first that demonstrates price dispersion online to be lower than that in offline markets. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 4 CHAPTER 2 LITERATURE REVIEW 2.1 OVERVIEW The two studies presented in this dissertation are informed by three streams of literature: marketing research on price format, economic theory on pricing as well as information systems research on price dispersion, and literature on airline pricing. The first study investigates pricing behaviors of online sellers who adopt the “everyday low price” strategy. This study is mainly informed by marketing research on retailer’s pricing strategy. Further, research on pricing in electronic markets provides a bridge through which the critical differences between physical and electronic markets can be incorporated into the consideration of the applicability of different price formats and specific ways in which they may be adopted in the online environment. The second section reviews the marketing literature on retail price format, discusses the rationale of the adoption of different price formats, how they influence consumers’ perceptions and purchase behaviors, how these different pricing strategies are characterized and measured in research, their applicability in electronic markets, and the gap in existing literature that will be addressed in this dissertation. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. The second study investigates the relationship of price format and price dispersion. This study is informed by both information systems research in online price dispersion as well as economic theory on pricing. In particular, seller’s conscious adoption of specific price format has not been identified as a factor that contributes to price dispersion by existing research. Section three reviews the literature in pricing theory as well as research on price dispersion, and presents an overview of the sources of price dispersion and the roles they play in affecting the dispersion of prices in electronic markets. The empirical analyses of both studies employ data obtained from the airline industry. Therefore, it is crucial to understand how prices of airline tickets are determined and control for the known factors that affect airline pricing, so that the differences in prices can be attributed to sellers’ adoption of different price formats. The final section of this chapter provides a review of literature in airline pricing and surveys the role of each factor in the pricing of airline tickets. 2.2 LITERATURE ON RETAIL PRICE FORMAT Price formats in retail industry have been studied extensively in marketing. Existing literature identifies two basic price formats, namely “Everyday Low Price” (EDLP) and “Promotional Pricing” (HILO). Although researchers argue that these formats are more appropriately regarded as a continuum rather than a dichotomy (Bell and Lattin 1998, Hoch, et al. 1994, Shankar and Bolton 2004), it is commonly R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 6 agreed that sellers who declare to adopt EDLP tend to charge relatively stable, below average prices with little or no temporary price discounts. These sellers aim to attract consumers who expect “good deals” by creating credibility and a low price image (Bell and Lattin 1998). On the other hand, HILO sellers are promotion-oriented and charge higher prices on average but engage in frequent promotions that allow prices to fall temporarily below the EDLP price level (Lai and Rao 1997). These sellers aim to price discriminate consumers of different preferences and price knowledge (Blattberg, et al. 1981). Therefore, researchers suggest that EDLP and HILO are more than mere price formats but rather important positioning strategies that attract different types of consumers, such as large basket shoppers versus small basket shoppers (Bell and Lattin 1998), and time-constrained consumers versus cherry-picking consumers (Lai and Rao 1997, Ortmeyer, et al. 1991). 2.2.1 Price Format and Consumer Segmentation Hoch, Dreze, and Purk (1994) investigate the profitability of EDLP versus HILO strategies by examining relative demand elasticity of the stores adopting these two price formats. They find that consumer demand does not respond much to changes in everyday price, and that reduction in EDLP prices may not drive volume large enough to compensate the lower profit margins, hence the HILO price format outperforms EDLP in profitability in their pricing experiments. Their finding is consistent with that in the event study of discrete change in a store’s price format R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 7 conducted by Mulhem and Leone (1990), who also find that stores that switched from EDLP to HILO experienced higher sales. In contrast to the finding by Hoch et al. (1994), Lai and Rao (1997) observe the opposite results in their game theoretic analysis that investigates the coexistence and profitability of EDLP and HILO strategies. The focus of their model is to examine factors that contribute to EDLP’s success. In their model, the EDLP store’s offering of constant, everyday low prices is an endogenously determined equilibrium outcome rather than an assumption. Lai and Rao assume that there are two types of consumers in the market: time constrained and cherry-picking. Both types of consumers purchase two products that are offered by two stores simultaneously. Time-constrained consumers, however, are assumed to visit only one store for their purchases. Their analysis on pricing and price format adoption concludes that the EDLP seller enjoy higher profit in equilibrium compared to the HILO seller. The intuition is that the presence of an EDLP competitor makes it difficult for the HILO seller to compete effectively on prices. While the pricing strategy may result in lower margins on the basket, this effect is offset by higher market shares, resulting in higher overall profits for the EDLP seller. Moreover, they suggest that although time constrained consumers find everyday low prices at EDLP attractive while cherry pickers find the promotions at HILO attractive, both formats attempt to attract both kinds of customers. Lai and Rao illustrate the latter point by introducing a nonprice attribute - service - into the theoretical model, where they show an R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 8 interesting result that service offered by the HILO store is higher than that in the EDLP store in equilibrium, thus attracting a larger portion of time constrained customers who are more willing to pay for services than cherry-pickers. This contrasts with the conventional wisdom that EDLP store’s success being based on consumers who are time constrained, but rather the HILO store is targeting this portion of customers with their higher level of service that provides time constrained customers the convenience, such as better parking space and speedier check-out counters, that they are more likely to value. Lai and Rao further demonstrate this finding by a survey that shows customers who patronize EDLP stores are indeed more likely to shop at various other places compared to HILO customers. This observation offers indirect evidence to their argument that HILO store’s clientele consists of more time constrained consumers as their model suggests. Bell and Lattin (1998) examine the relationship between consumers’ shopping behavior and their preferences for store price format, and find that EDLP stores are more likely to attract large basket shoppers while HILO stores are likely to attract small basket shoppers. Large basket shoppers are characterized by lower ability to respond to prices in individual product categories while higher sensitivity to the expected cost of the overall basket. On the other hand, small basket shoppers are typically older in age (less time-constrained), have lower incomes (more price conscious), and with smaller family sizes (lower consumption). In sum, EDLP R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 9 stores are likely to attract customers who are less price elastic and with higher average level of consumption. In contrast to Ainslie and Rossi’s (1998) finding that more frequent shoppers (those who shop for “filling-in” certain supplies that ran out at home) being more price sensitive to brand choice, Bell and Lattin’s empirical results suggest that small basket shoppers appear to be less responsive to marketing activities. In reconciling the discrepancy in the two findings, they suggest that the proportion of such need-driven shopping trips may not necessarily be different between large and small basket shoppers. In other words, while fill-in trips will be identified as small-basket trips, not all small-basket trips are fill-in trips. Due to the evidence of opportunistic purchase behavior among small basket shoppers, Bell and Lattin urge future research to look into trip-specific differences that drive different types of shoppers to stores that use different price formats. Ho et al. (1998) show that the coexistence of EDLP and HILO can be explained by different types of consumers going to stores adopting different price formats: HILO price format induces consumers to shop more frequently but spend less on each trip, while consumers shopping at an EDLP store visit less frequently but spend larger dollar amounts for each visit. This finding is consistent with that of Bell and Lattin (1998) in that “large basket” shoppers, who spend more on the shopping trip, prefer EDLP stores compared to shoppers of smaller basket sizes. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. In sum, in a price-based competition between EDLP and HILO sellers, research on the relationship between consumer type and their preference for price format suggests that EDLP tend to attract consumers who are time-constrained (Lai and Rao 1997), hence do their shopping less frequently (Ho, et al. 1998) and purchase more on each visit (Bell and Lattin 1998). Two predictions on retail price format adoption in electronic markets can be made based on Lai and Rao’s theoretical framework and Bell and Lattin’s empirical finding: First, since physical separation between EDLP and HILO stores is absent in electronic markets, consumers can compare prices directly across stores relatively costlessly. Lai and Rao’s model predicts that both the equilibrium average prices and profits of EDLP sellers would decrease. Second, Bell and Lattin find that the typical small basket shoppers are less time-constrained and more price conscious; while the effects of Internet on price sensitivity are uncertain depending on the availability of product information and whether the same products are available in different stores (Lynch and Ariely 2000), the time (and hence opportunity costs) required for a consumer to compare prices from multiple sellers in electronic markets are unambiguously reduced compared to those in physical markets. This implies that potentially a large portion of online consumers may exhibit behaviors similar to those of small basket shoppers. Bell and Lattin suggest that as more and more consumers exhibit small basket shopper behavior, an EDLP seller has to lower the average price to the HILO seller’s deal price level. Given Hoch et al.’s finding on the failure of EDLP format to generate enough volume to compensate the loss in profit margin from price reduction, a R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. central question this research aims to address is: is it still feasible to adopt the EDLP strategy online? 2.2.2 Viability of EDLP in Electronic Markets Research suggests that the coexistence of EDLP and HILO relies heavily on the choice of relative physical locations by sellers adopting the two strategies. In pure price-based competition, the theoretical model of Lai and Rao (1997) concludes that the coexistence of EDLP and HILO sellers in equilibrium is dependent on two conditions: first, the number cherry-picking consumers (relative to the number of time-constrained consumers) needs to be small. This condition warrants the profitability of EDLP sellers since time-constrained consumers will not shop around and buy from the lowest price; if EDLP sellers set a basket price that is attractive enough, they will gain this portion of consumers. Second, the transport (search) costs of the time-constrained consumers cannot be too low. Even though costs of searching for the time-constrained consumers can still be high relative to that of cherry-pickers, if the search costs are sufficiently low, EDLP sellers will have to further reduce their prices to incentivize these consumers to not search but to purchase from them. In electronic markets, consumer’s ability to compare prices is improved and the associated costs are tremendously reduced compared to having to physically visit different stores and find out price information in traditional markets. Other things being constant, in the online environment more consumers are capable of performing search as cherry-picking consumers, and for those who still suffer R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 12 higher costs of searching, their search costs are largely reduced as well. Therefore, it is unclear whether it is still feasible to practice EDLP online and whether the coexistence of vendors practicing the two different pricing strategies is still possible. In the analysis of the relationship between consumer’s shopping motivations and retail price environment, Pemer (1998) suggests that a consumer’s preference for specific price format is governed by four major factors. These factors are hedonic shopping motivation, ego-expressive shopping motivation, rationality, and risk averseness. Hirschman and Holbrook (1982) suggest that consumer derives pleasure from the shopping experience itself in addition to the simple utilitarian motive to obtain the desired items. Different price formats, therefore, may appeal to different consumer segments that place different values on the excitement of obtaining deals from HILO sellers. For example, for those who values security, the shopping experience at an EDLP store may be reassuring, while others may regard such an experience as less exciting (Pemer 1998). Further, consumers may be motivated by ego-expressive motive in performing extensive price search to bolster one’s self-concept as a smart shopper (Feick and L. 1987, Schindler 1989, Slama and Williams 1990). In electronic markets where prices of both EDLP and HILO stores become more visible and directly comparable, the role of hedonic shopping motives in determining consumer’s price format preference may become less important while ego-expressive R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. shopping behaviors are further encouraged, suggesting that online sellers may find HILO price format to be more favorable. Economic theory suggests that a rational consumer seeks to minimize the sum of prices paid for goods and the opportunity cost of shopping (McCloskey 1985, Samuelson and Nordhaus 1992). When faced with relatively low opportunity cost for shopping, a consumer would therefore be encouraged to engage in extensive comparison shopping (Pemer 1998). Prospect theory (Kahneman and Tversky 1979) suggests that risk aversion may influence a person’s preference of certain but low-value outcome versus uncertain but potentially high-value outcome. Therefore, risk-averse consumers may prefer relatively stable prices of the EDLP stores rather than drawing a “lottery” from the HILO sellers in their higher average price and occasional offering of great bargains on certain products. However, in electronic markets where the costs of shopping different stores are relatively low, the risks incurred by both types of shoppers become less significant. Therefore, even risk-averse consumers may prefer engaging in extensive price search across different stores as the benefits outweighs the costs of finding out the lowest price. These theories suggest that HILO price format would be preferred by consumers in electronic markets where comparison shopping is relatively costless. Therefore, the sustainability of EDLP in the online environment becomes questionable. However, given the success of certain online sellers who clearly advertise themselves as EDLP R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 14 sellers, it is imperative to investigate the extent to which their pricing behaviors are consistent with the “everyday low price” image and how they are pursuing this particular strategy. 2.2.3 Dimensions of EDLP To understand whether the pricing behavior of a seller is consistent with his advertised price image, it is necessary understand what characterize the differences in EDLP and HILO price formats. Empirical research on retail price format has identified three dimensions on which different price formats can be compared. They are price level, intertem poral price variability, and price range. Price Level and Variability Hoch et al. (1994) report that while prices in EDLP stores are on average below those in HILO stores, they engage in as much promotion (though with less deep discounts) as their HILO counterparts. This suggests that EDLP strategy can be adopted in two separate dimensions that do not necessarily go hand in hand - low average price level, and low price variability. Further, their results show that HILO format outperforms EDLP in generating volume sales and in profitability. As opposed to Hoch et al. (1994), Ho et al. (1998) does not find one particular price format that dominates over the other. Ho, Tang and Bell (1998)’s analysis of the differences in specific dimensions of pricing strategy between EDLP and HILO is R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 15 perhaps the closest in spirit to this research. Ho et al. investigate the same dimensions (price level and price variability) and conclude that there are no observable differences between two stores adopting the same (EDLP) format, while differences in the average price and price variation between EDLP and HILO stores are significant. Further, they observe a positive association between variability and price level. This implies that EDLP sellers who charge lower price on average should also maintain low price variability, while prices set by HILO sellers, on the other hand, should exhibit both higher average and higher variations. Price Range Shankar and Bolton (2004) find that the positioning of stores explains a large portion of variation in the range of prices offered, and report that prices offered by the EDLP stores fall within a significantly narrower range compared to those by the HILO stores. Further, they suggest that prices at EDLP stores may be closer to the low prices than those at HILO stores, though this speculation has not been tested directly due to limitation associated with their definition of relative brand price. Relative brand price is defined by Shankar and Bolton as the ratio of price of a particular brand to the average price of all brands within the category. Therefore, low relative brand price of an EDLP store can be attributed to one (or both) of two factors: lower price level of a particular brand, and smaller range of prices within that category. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 16 2.2.4 Category-Level Adoption of EDLP Industry reports suggest that pure EDLP/HILO seldom exists, and point towards a mixed adoption of various elements in the two price formats. Progressive Grocer (1994) reports that there are few if any companies running pure EDLP programs anymore, and most mix in specials to make for a better merchandising appeal to their shoppers...” The industry continues to observe that “ ...As for pricing strategies, in a shift that may more accurately reflect what's really happening in the industry, hybrid pricing (some combination of EDLP and highlow) was the leading choice, bumping out the longstanding favorite EDLP. In the past, it was fairly common for retailers to refer to their pricing as EDLP, even if, in fact, it was not a pure EDLP program...” (Radice 1998) and specifically, EDLP sellers are found to practice HILO in certain product categories (Garry 1996). Hoch et al. (1994) also report that there is anecdotal evidence that some retailers adopted “category-level EDLP” to build store traffic and stave off competition from alternative retail formats. However, they argue that adopting EDLP on a category-by-category basis would not work well because EDLP has to be adopted on a universal basis to benefit from overall store price image. In the electronic world where the cost of obtaining price information is considerably less than in the physical context, the adoption of EDLP on a limited level appears to be even more unviable because of the possible erosion of the low price image the firm is trying to build. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. In their analysis of expected basket attractiveness, Bell and Lattin (1998) conclude that some categories are more salient than others in determining the price image of the store. They point out that future research is needed on the relationship between store price image and category-level pricing. Ho et al. (1998) also call for research to address multi-category pricing. In particular, they suggest the possibility that sellers may adopt different price formats for different product categories to maximize revenue. From the research perspective, speculation on seller’s category-specific pricing had been left unaddressed until a recent paper by Shankar and Bolton (2004). Shankar and Bolton analyze how market, chain, store, category, customer, brand, and competitor factors affect retailer’s pricing strategy. Of the seven factors that they have identified, competitor factors are found to be explaining the majority of variation in retailer pricing strategy, while the effects of chain positioning (EDLP versus. HILO) and category factors (storability versus perishable, and necessity versus luxury) are marginal in most cases. Consistent with the findings by Ho, Tang and Bell (1998), Shankar and Bolton find that prices of EDLP sellers are less variable than those of HILO sellers. Further, prices of storable products and necessity are found to be more consistent. Their results suggest that retailers may choose to be less price-consistent for other categories where change in prices can potentially stimulate increases in primary R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 18 demand. In particular, Shankar and Bolton suggest that EDLP stores may tailor their overall strategy to “recognize differences in consumer demand within and across categories” (p. 39). In other words, along with Bell and Lattin (1998) and Ho et al. (1998), they also suggest that the practice of EDLP may not be consistent across all products that the vendor sells. However, Shankar and Bolton did not study whether EDLP strategy may be selectively employed in specific product categories or types of markets. The question of whether different price formats can be applied to different product categories remains unanswered. This dissertation examines whether sellers employ such a category-specific adoption of different price formats in the online environment. 2.3 LITERATURE ON PRICE DISPERSION From marketing research discussed in the previous section, it is clear that a vendor’s choice of a particular price format affects the way in which the vendor sets prices. Since different prices observed in a market, among other factors to be discussed in this section, can be attributed to different pricing practices of specific vendors, price format can be a potential explanation to the existence of price dispersion even in the online environment. In order to understand the mechanisms through which different price formats adopted by vendors can influence the dispersion of prices in a market, it is critical to identify the sources of price dispersion and understand the role of each of these sources in vendor’s pricing strategy. This section summarizes the findings by literature on pricing theory and price dispersion, and discusses each R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 19 of the three major sources of price dispersion recognized by existing research: heterogeneities in product, consumer, and vendor. 2.3.1 Price Dispersion Due to Product Heterogeneity Product heterogeneity is the most obvious source of price dispersion. If the products are different and deliver different values, then it should not be surprising that they are being priced differently. Products can be heterogeneous along various dimensions, such as differences in how closely they align with consumers’ tastes and preferences (Hotelling 1929, Salop 1979), or differences in quality (Mussa and Rosen 1978). They can also be differentiated with the associated shopping experience, convenience, and services such as delivery and return policy (Smith, et al. 2000). Further, heterogeneity can be introduced into ex ante homogeneous products by incorporating elements that creates switching costs for consumers (Klemperer 1987), which lead to unwillingness of consumers to purchase an otherwise identical product from another seller. Thus the concept of switching cost is closely related to consumer’s brand loyalty. 2.3.2 Price Dispersion Due to Consumer Heterogeneity Brand Loyalty and Awareness Brand is a signal for quality or service levels (Lai and Sarvary 1999). Using game-theoretic analysis, Narasim han (1988) shows that in a market where consumers are heterogeneous in brand loyalty, price dispersion results from seller R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 20 arbitrage and randomizing strategies in equilibrium. The predictions of this model have been supported by empirical findings on price dispersion in electronic markets. Brynjolfsson and Smith (2000) observe that branded retailers and retailers with whom a consumer visited previously hold significant price advantages, and conclude that price dispersion on the Internet may result from heterogeneity in consumer awareness or in retailer branding and trust. Chen and Hitt (2001) demonstrate that the necessary conditions for Bertrand competition to emerge and hence the resulting convergence of prices in a market are full brand awareness and the absence of brand sensitivity, which are unlikely in the Internet context. They show that there is a non-monotonic concave relationship between price dispersion and the proportion of consumers who are informed (brand-aware), and conclude that high price dispersion does not necessarily imply lack of competition. Therefore, consumer knowledge plays a significant role in price dispersion online. Price Knowledge Varian (1980) identifies two forms of price dispersion, namely spatial price dispersion and temporal price dispersion, that are closely related to the relative portions of consumers who are informed and uninformed about the distribution of prices in the market. Spatial price dispersion occurs in a situation where multiple sellers contemporaneously offer a homogeneous product. An example is Salop and Stiglitz’s (1977) study of the coexistence of these two types of consumers in a market of homogeneous products. Salop and Stiglitz show that market equilibrium R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 21 takes the form with some fraction o f sellers sell at competitive price (at average cost) while the rest sell at a higher price (monopolistic price). Although high-priced sellers sell only to uninformed consumers, sufficiently large number of such consumers keeps the business of these sellers. A direct interpretation of Salop and Stiglitz’s result is that consumer heterogeneity can be a source of price dispersion. In particular, they examine a market with consumers having different cost of information acquisition and find that the equilibrium in such market to be monopolistic competition with price dispersion. Similar result is obtained in their later paper, even with the relaxation of assumption about consumer heterogeneity, that finds when all individuals are identical but that information is costly to gather, the only market equilibrium is characterized by price dispersion (Salop and Stiglitz 1982). The necessary assumption underlying Salop and Stiglitz’s analysis is that consumers cannot learn from experience which, according to Varian, is implausible as consumers will gain knowledge over time and the portion of uninformed consumers will decrease to a level such that eventually the existence of high-priced sellers in the market is no longer supported. Varian proposes an alternative form of price dispersion that occurs on the temporal dimension. In the market with temporal price dispersion, sellers intentionally vary prices over time so that consumers cannot learn by experience about sellers that consistently have low prices. Such randomized pricing strategy is an attempt to discriminate between informed and R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 22 uninformed consumers; and as long as the cost of acquiring price information is positive, price dispersion persists. In Varian’s model, consumers “choose” to become informed or to remain uninformed. However, the portions of the two types of consumers are assumed to be exogenously determined and consumer choices are not modeled in the analysis. Baye and Morgan (2001) extend Varian’s model on two dimensions: first, they incorporate a third party that maintains an information gateway to facilitate the exchange between buyers and sellers, known as the “information gatekeeper”, to capture some characteristics of the electronic markets such as the intermediaries. Second, they explicitly model consumer’s choice of being informed (subscribe to the gatekeeper) or uninformed (not subscribing). The information gatekeeper charges fees to firms and consumers for advertising prices and accessing price information from its site, respectively. An interesting finding in their model is that equilibrium dispersion exists in the homogeneous product market even if all subscribing consumers purchase from the lowest price seller. The rationale behind this finding is that firms face the tradeoff between charging higher prices and earning higher profits from the uninformed consumers and losing sales from those who have access to prices from other sellers, and thus will randomize prices as long as there is a positive portion of uninformed consumers who will be purchasing at the monopoly price. A critical assumption in their analysis is that firms cannot price discriminate the two types of consumers by charging different prices on their own website and on the gatekeeper’s R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 23 site. In a later paper, Baye and Morgan (2002) revisit this model and relax the assumption of firm’s inability to price discriminate. They find that even when firms can post different prices on their own websites and on the gatekeeper’s site, prices remain dispersed at the gatekeeper’s site but are strictly lower than the (monopoly) price at the firms’ own websites as long as firms pay a positive fee to list a price at the price comparison site. The fees that consumer pay in gaining access to prices listed by different firms in Baye and Morgan’s (2001) model can be thought of as the cost associated with searching for these different prices and offerings. Search Cost Bakos (1997) models the effects of electronic markets on lowering consumer search costs. His main idea is that an electronic marketplace such as the Internet facilitates the exchange of price and product information among participating buyers and sellers by serving three distinct market functions: matching buyers and sellers, facilitating transaction, and providing an institutional infrastructure (Bakos 1998). Bakos posits that electronic markets match buyers and sellers by reducing the costs involved in their search for the right products and the right buyers. Specifically on the buyer’s side, such search costs are composed of the opportunity cost of time spent on searching, and the associated expenditures referred to as “access costs” (Bakos 1997). As a result of lowering such costs, market becomes more efficient in R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 24 terms of decreased ability o f sellers to extract monopolistic profits and increased ability of the market to optimally allocate resources. Bakos separates consumer’s costs of searching into two components: those that associate with the search for price information, and those that associate with the search for product attribute information. Bakos’s analysis leads to two major findings: first, when cost of product information is positive but cost of price information is close to zero, near perfect competition prevails with sufficiently many firms (i.e. equilibrium price converges to marginal cost). On the other hand, when cost of price information is positive and cost of product information is zero, equilibrium price is decreasing in the cost of price information. However, Harrington (2001) criticize Bakos’s results by demonstrating that as long as the cost of search over price remains positive, regardless of its size, would result in prices being bounded above competitive price. His argument is that firms can still exploit profit from information asymmetry even if the consumer follows optimal search strategy, and there does not exist a symmetric pure-strategy equilibrium involving search. Further, Harrington points out that Bakos’s second finding is based on an unreasonable implicit assumption that the entire price search cost can be recovered if the searched product was not purchased. He asserts that under the normal assumption that such search costs cannot be recovered, the equilibrium price R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 25 is the monopoly price and is independent of the cost of searching over price as long as it is positive. Harrington’s critique is consistent with the predictions by extensive search literature in economics. Existing search models conclude that only when consumers can search costlessly (i.e. perfect price and product information) would a market boils down to Bertrand competition with Walrasian price as the unique Nash Equilibrium (Bertrand 1883). However, when price information is costly to obtain, Stigler (1961) shows that firms will price above marginal cost. Diamond (1987) further illustrates that in a market where consumers search sequentially and have strictly positive and unrecoverable search costs, the unique price equilibrium is where all firms charge the monopoly price. Product Knowledge While search cost theories predict that if and when the Internet reduces consumer search costs, prices will be less dispersed (though not necessarily converge to a competitive level). In contrast to this view, Lai and Sarvary (1999) present an interesting model and shows that, under certain conditions, lower search costs can actually decrease the level of competition and hence lead to higher dispersion in prices on the Internet. In their game-theoretic analysis, Lai and Sarvary divide product attributes into those that can be easily communicated through the electronic medium (digital attributes) and those that cannot (non-digital attributes). They find R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 26 that when consumers are heterogeneous in their knowledge about non-digital attributes of the products, even if the Internet reduces discriminatory power of information regarding product quality, the levels of price competition and consumer search online may decrease. The intuition behind their results is that while the Internet allows consumers to evaluate digital attributes easily, non-digital attributes can still be only evaluated by physical presence. Therefore, for products with important non-digital attributes, consumers are reluctant to take the risk of searching for products with better fit and sellers are able to exploit consumer loyalty and risk aversion. In other words, the Internet can potentially create higher effective search costs for consumers, lower their price sensitivity and induce them to remain with the product they are familiar with. Price Sensitivity Different degrees of price sensitivity and demand elasticity of consumers can also be a source of price dispersion. In their study of price competition between Amazon and Barnes and Noble, Goolsbee and Chevalier (2002) find significant price sensitivity at both merchants, but that the demand at Bames and Noble being much more price-elastic than that at Amazon. Such difference in price elasticity accounts for the different prices charged by the two vendors over even highly homogeneous products like books. In the context of differentiated but comparable products, the effects of the Internet on consumer price sensitivity are more complicated. While price information is made more accessible to consumers online, potentially leading R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 27 to lower consumer price sensitivity, the Internet also offers consumers with much richer non-price information that may lead to lower consumer price sensitivity and hence resulting in more significant dispersion in prices (Degeratu, et al. 2000, Shankar, et al. 2001). One example is Lynch and Ariely’s (2000) finding that when wines carried by different websites vary, lowering cost o f search for quality information leads to lower price sensitivity by online wine shoppers, resulting in increased price dispersion. On the other hand, for the same wines that are available in different websites, consumer price sensitivity increases due to easier cross-store comparison; prices for these products are thus less dispersed. Besides differences in product and consumer characteristics, heterogeneity in vendors has also been proposed as a significant source of price dispersion. 2.3.3 Price Dispersion Due to Vendor Heterogeneity Cost Reinganum (1979) observes that when firms differ in their marginal costs and have perfect information on buyers’ reservation prices and demand functions, they behave as monopolistic competitors and offer a distribution of prices that creates price dispersion in the market. Spulber (1995) find that when vendors’ costs are different but are known to all competitors in the market, vendors with lower costs will undercut the prices set by others with higher costs, eventually forcing high-cost vendors out of the market and leading to convergence in prices. However, in the R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 28 more realistic case where asymmetry of information exists, price dispersion would result. Some argue that by making rivals’ prices observable, the Internet can potentially increase cost transparency (Sinha 2000) and thus reduce price dispersion. Branding and Trust Since transactions on the Internet typically do not involve simultaneous exchange of money and goods, branded retailers and those with whom the consumers trust may extract a premium from consumers who are reluctant to purchase a product from an unknown retailer (Smith, et al. 2000). Therefore, heterogeneity in retailer trust may lead to price dispersion online. Smith and Brynjolfsson (2001) argue that consumers indeed search more among the dominant firms of whom they are aware, allowing such firms to maintain higher prices. This speculation is supported by Bryjnolfsson and Smith’s (2000) finding that branded retailers tend to be able to charge higher prices. Pan et al. (2002) also offer indirect support to this argument. In their analysis of the relationship between e-tailer’s service quality and price dispersion, they find that e-tailer service attributes do not explain much of price dispersion observed online. They suggest that differences in e-tailers’ ability to command price premiums may be attributed to the fact that some e-tailers possess brand names and more trusted websites. Although electronic markets unambiguously reduce the costs of acquiring product and price information by consumers, hence should contribute to lowering price R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 29 dispersion for products that are highly homogeneous, empirical research has found mixed evidence when comparing dispersion of prices in the online and offline contexts. 2.3.4 Empirical Findings on Online vs. Offline Price Dispersion Bailey (1998) examines the relative efficiency of Internet market versus traditional market by comparing prices of books, CDs, and software titles in 1996 and 1997 sold through the Internet and traditional channels. He concludes that price dispersion among online vendors was at least as great when compared to that among traditional retailers. Further, prices on the Internet were found to be higher than those in the offline channel. Brynjolfsson and Smith (2000) conducted a similar study between 1998 and 1999. Their findings are consistent with those by Bailey in that price dispersion in the online channel was not lower than that in the offline channel. Specifically, the average level of price dispersion in books and CDs was 33% and 25% respectively. However, after weighting the posted prices by the vendors’ web traffic, they find that price dispersion was smaller online compared to offline. Lee and Gosain (2002) are among the first to show that online price dispersion differ among the different categories within the same type of products. In comparing the R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 30 dispersion in prices among “oldies” CD albums and “current-hit” CD albums in the online and offline contexts, they find that price dispersion was significantly higher among the oldies albums online (31%) than offline (11%); while dispersions in the two channels were not significantly different for current-hit albums (19%). In their study of price and non-price competition in the online book industry and comparison of price dispersion in online and offline bookstores, Clay et al. (2002) report that the average prices of books are not significantly different across the two channels. However, there is evidence that price dispersion online is still substantial. In particular, they find that percentage price difference range from 27% for randomly selected hardcover books to 73% for paperback bestsellers. They also find that unit prices on Amazon.com were on average 5% higher than those on BamesandNoble.com and 11% higher than those on Borders.com. Clay et al. did not yield conclusive results in their analyses on the relationship between store differentiation and the observed price differences. Besides highly homogeneous products such as books and CDs, research in online price dispersion has also looked at differentiated products in various industries, such as online insurance quote, airline tickets from online travel agents, and Internet car referral services. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 31 Clemons et al. (2002) investigate the difference in prices of comparable airline tickets quoted by different online travel agents in 1997. Not only do they find prices to be significantly different across online travel agents (up to 28%), but that such difference persists (up to 18%) even after controlling for observable heterogeneity in the tickets, such as arrival and departure times, number of connections, and Saturday night stay restriction. In a study of the impact of Internet car referral services on dealer pricing in California between 1999 and 2000, Morton et al. (2001) observe that Internet car purchases through the online referral service amount for only 2.9% of car sales and prices paid by customers who buy through the online channel are on average 2% lower. Further, they find a negative association between the number of cars a dealership sold through the Internet referral service and the range of prices consumers paid. Morton et al. attribute the decrease in price range to increased buyer information and bargaining clout due to Internet referral services. Brown and Goolsbee (2002) study the impact of increased ability of consumers to more efficiently compare prices on the Internet on life insurance market during 1992 to 1997. They find an inverted-U shape relationship between price dispersion in the market and the number of Internet search sites, and conclude that the growth of the Internet reduced term life prices by up to 15%. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 32 In sum, while existing theories have identified various sources o f price dispersion, they fail to arrive at a conclusive prediction on the role of the Internet on the relationship between these factors and online price dispersion. It is apparent from the empirical studies presented above that product heterogeneity does not fully explain price dispersion. Besides heterogeneity in products, a majority of literature on price dispersion focuses on the differences in consumers in explaining the existence of dispersion. However, except for a few studies that infer consumer price sensitivity from their purchases across different websites on comparable products, empirical research is often limited in the ability to address such heterogeneity because differences in consumers are hard to measure and is often constrained by the availability of data at the individual consumer level. Take search costs for example; a consumer’s costs of searching includes both actual time spent on performing the search and individual-specific perceptual costs such as cognitive efforts that cannot be easily measured. Further, consumer heterogeneity need not be taken as exogenous. While analytical models often makes simplifying assumptions on the distribution of consumers in their heterogeneity in certain dimension, and that relative portions of each “type” of consumers are not affected by vendor’s strategy, in reality vendor can adopt different strategies to affect the perceptions of consumers and hence their behavior. For example, marketing research on price format suggest that vendor’s adoption of R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 33 different price formats can serve as mechanisms to keep the consumers from becoming fully informed and hence influence their behavior. In other words, it is very well possible that consumers react to differences in vendor’s pricing strategies, rather than the other way around. The limiting role o f heterogeneity in the understanding of price dispersion has been acknowledged at least as far back as the 1960s by Stigler: “a portion of the observed dispersion is presumably attributable to such differences. But it would be metaphysical, and fruitless, to assert that all dispersion is due to heterogeneity.” (Stigler, 1961 p.214). 2.3.5 Random Pricing Random pricing theory suggests that vendors randomly change prices so consumers cannot leam from experience on which seller has the lowest price (Chellappa and Kumar 2005, Chen and Hitt 2003, Varian 1980). Two empirical studies on the nature of price-rankings of online retailers can be applied to shed lights on this theory. Baylis and Perloff’s (2002) find that a retailer kept its rank or changed by only one position 57% and 75% of the time in digital camera and scanner markets in 1999, respectively. A robust check was performed in 2001 and the authors report largely similar results. Pan et al. (2003a) conclude from these results that random pricing theory is not em pirically supported, because the relatively stable price-rankings of retailers imply that vendors do no play collectively raise or lower prices randomly over time or take turns to undercut each other. On the contrary, a study by Baye et al. (2004) on prices for 36 best-selling consumer electronics posted R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 34 on an Internet price comparison site between 1999 and 2001 find substantial variation in the identity of the low-price firms as well as the level of the lowest price. Baye et al.’s results offer strong evidence to random pricing, and they conclude that such a “hit and run sales” is widely used and effective for avoiding all-out price competition in online markets. The studies by Baylis and Perloff (2002) and Baye et al. (2004) yield largely conflicting evidence. While the former suggest that the relative prices offered by online retailers are stable, the latter find online sellers randomize prices to preclude rivals from being able to systematically undercut a fixed price. How can these two contradictory findings be reconciled? There are potentially two explanations for the discrepancies between the findings of these two studies. First, there are differences in the nature of competition in the two markets under investigation. Such differences could be due to differences of maturity of the Internet market, or differences in the type of products the two markets carry. However, both of the studies were conducted in the same period, therefore the maturity of Internet markets is not a likely candidate to explain the difference. Further, both studies look at consumer electronic products. Although the set of products considered by Baye et al. are much greater in variety and in number, the products under examination in both studies are highly homogeneous in nature and prices for the identical model and make were compared across different sellers. In conclusion, difference in nature of R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 35 competition should not be the reason for the disagreement in findings of the two studies. A second potential explanation is differences in retailers’ pricing strategies. Marketing literature suggests that there are two general price formats that retailers adopt. One is characterized by low average price level and low variability known as “everyday low price” (EDLP), another one is characterized by higher average price level but with frequent price cuts, known as “promotional pricing” (HILO). It appears that the former offers a good description of the pricing behaviors of vendors considered in Baylis and PerlofFs study, whereas the latter of those considered in Baye et al.’s. However, without explicitly examining the nature o f the differences in the price formats adopted by vendors in the two cases, this remains only a speculation. This research offers the first evidence that vendor’s adoption of different price format can contribute to such differences observed by earlier studies. Specifically, one of the empirical analyses presented in chapter 4 investigates whether there is any correlation between price format and the stability in the identity of the seller offering the lowest price. 2.3.6 Measuring Price Dispersion The dispersion of prices can be reflected in a variety of measures, such as the difference between the highest and the lowest price, the variance or standard deviation of the distribution. This section summarizes the different measures R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 36 employed by existing literature in studying price dispersion in the airline industry, conventional retail markets, and in online intermediaries. In an empirical analysis on price dispersion in the airline industry, Borenstein and Rose (1994) report dispersion with the Gini coefficient but observe that results of their study to be very similar when they measure dispersion with the coefficient of variation or the relative interquintile range of fares instead. They find that price dispersion varies substantially across different airlines as well as across different routes. In particular, price dispersion in tourist-oriented routes is found to be substantially less than that in business-oriented routes. Further, they report that the magnitude of price dispersion across airlines within a market can be mainly attributed to market and competitive factors. They find that price dispersion increases with airport dominance and the number of competitors, while decreases with frequency of flights on a route. Borenstein and Rose speculate that the positive relationship between airport dominance and price dispersion is due to frequent-flyer programs (FFPs) that are most effective in airports in which the airline dominates. Borenstein (1989) suggests that FFPs are most attractive when an airline offers extensive service from a traveler’s home airport for the following reasons. First, since the bonus value increases with mileage accumulation, a customer who participates in FFPs has strong incentive to concentrate his travels with only one or a few airlines. Second, the expected value of the bonus to a customer from participating in FFPs (free travel certificates) increases with the R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 37 number of destinations the airline serves and also with the airline’s services from the customer’s likely point of departure. Therefore, FFPs tend to induce loyalty and enhance value most, especially for the high-fare business travelers, when the airline dominates more airports and routes. Schwieterman (1985) study the effects of Airline Deregulation Act (ADA) of 1978 on different types of markets (monopoly, duopoly, and oligopoly) in the air travel industry. He analyzes the change in price dispersion, measured by the ratio of highest to lowest prices to capture the range of prices available from different airlines within a market, in these markets after deregulation. Schwieterman concludes that although airlines were still able to continue to increase profits by segmenting markets as means of price discrimination, deregulation has eroded airlines’ ability in doing so. Baye, Morgan and Scholten (2003) study the dispersion of prices in consumer electronic products available through an online intermediary over a period of 18 months. They show that prices do not converge to the “law of one price” - although average range in prices decreases by 40% to 30% by the end of the observation period, 28% of variation in prices remains unexplained after controlling for firm heterogeneities such as costs, branding, reputation, trust, product availability and shipping costs. Baye et al. (2003) report a wide range of measures of price dispersion. In particular, they consider both “absolute m easures o f price R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 38 dispersion” (i.e. range in prices, differences between average and lowest price, difference between lowest two prices) and “relative measures of price dispersion” (i.e. range in prices as a percentage of lowest price, coefficient of variation, gap between lowest two prices as a percentage of the lowest price, and difference between average and the lowest price as a percentage of lowest price). In a more recent paper, Baye and Morgan (2004) report bounded rationality as a source of price dispersion based on their investigation of lab-generated versus observed price dispersion in the real world. Again using prices obtained for consumer electronic products through an online intermediary, Baye and Morgan observe significant price dispersion in homogeneous product markets. Further, they discover that similar patterns of price dispersion occur in the laboratory settings, where other potential sources of price dispersion such as transaction costs, imperfect information, and variance in seller costs are absent by experimental design. Baye and Morgan construct two statistics, coefficient of variation and price range, to summarize the level of price dispersion. Among the various measures of price dispersion, the most representative ones are the range in prices (which captures the absolute measure of price dispersion) and the coefficient of variation (which captures the relative measure of price dispersion). While the “raw” measure of variability such as variance and standard deviation have R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. also been used by prior studies on price dispersion (Dahlby and West 1986, Pratt, et al. 1979), the scaled measure, coefficient of variation, is more robust because it is expressed as a ratio to the average price. While two markets may exhibit the same variance in prices, the dispersion is considered more significant in the one where the average price is lower. The rationale is there is more room for differences in prices when the average price in the market is high, thus the same variance would mean more severe dispersion in a market where prices are comparatively lower. In addition to the literature summarized above, range and coefficient of variation have been used in a variety of other studies. Carlson and Pescatrice (1980) investigate price dispersions in conventional retail markets of expensive products versus low-price products, and find that the former exhibit smaller scaled price dispersion but wider absolute price dispersion compared to the latter. Sorensen (2000) examines dispersion in the retail prices of prescription drugs and finds that the prices of repeatedly purchased prescriptions exhibit significant reductions both in price dispersion and in price cost margins, offering indirect support of the positive relationship between lower in consumer search costs and decrease in price dispersion. Baye, Morgan and Scholten (2004) report persistent price dispersion in best-selling consum er electronics products sold in an online interm ediary over tim e. Specifically, they observe unpredictable variations in the identity of the firm that sells the offer the identical product at the lowest price. A major implication of their analysis is that a “hit and run” pricing strategy is widely used and is an effective R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 40 weapon in online markets, where consumers can gain access to a list of prices offered by competing firms for a similar products from an “information clearinghouse” that characterize Internet price comparison sites. 2.4 LITERATURE ON AIRLINE PRICING Another stream of literature that is closely related to this study is research on airline pricing. While prices of an airline tickets are affected by many factors, in order to examine a more complete model of EDLP/HILO competition, it is necessarily incorporate factors such as the different costs of firms, control for variations in reservation values (e.g. timing of the purchase of tickets with respect to the departure date), and consider different consumer segments (e.g. business-oriented travelers versus leisure travelers) (Lai and Rao 1997). This section presents an overview of four major factors that affect pricing of airline tickets - market power, competition, market structure, and cost - as well as the corresponding variables to be incorporated into the empirical analyses of this study. 2.4.1 Market Power Hub Hub dominance significantly influences an airline’s ability to mark up prices and has been used as a measure for market power (Borenstein 1989). Berry et al. (1997) suggest that hubbing airlines may offer superior service through their control over airport resources, such as more convenient gates and better departure times, that R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 41 business travelers are more likely to value. They find that flights originating at hubs are more appealing to customers who exhibit characteristics of business travelers - less price sensitive, high willingness to pay for frequency and high disutility from connecting flights. Borenstein (1989) points out that airlines that dominate the traffic at an airport enjoy competitive advantage in two regards. First, these airlines are more likely to induce loyalty through frequent flyer programs by offering more flights to and from a city. Second, dominant airlines at an airport are typically endowed with more slots (right for takeoff and landing during particular time periods, especially during peak hours) and have more control of the gates (building and jetways). They may therefore be able to limit the abilities of potential competitors to obtain gates and other facilities necessary for entry or expansion of service at the airport, or even influence the decision of the airport operator regarding the expansion of the airport to accommodate new entrants. In sum, prior research suggests that hub increases entry barriers and drives up prices for hub-originating passengers. Market share Market share is an indicator of an airline’s power and ability to influence prices in a given market (origin-destination airport pair). Prior literature has used various measures to operationalize this construct. For example, Borenstein (1989) computes “ORGSHARE” (origination share) and “RUTSHARE” (route share) to R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 42 capture the effect of an airline’s share in a market on ticket prices. Origination share is the weighted average of observed carrier’s share of daily passenger originations at the two endpoints of the observed routes. Route share is the observed carrier’s share of all passengers on a route. Firms with larger origination and route shares would be expected to have greater market power and hence set higher prices. The use of market share measures as independent variables in explaining price, however, is problematic. Because market share is essentially a measure of demand, which in turn is a function of price; market share is therefore an outcome variable determined by price and is highly endogenous. Borenstein (1989) uses the instrumental variable approach and two-stage least square to mitigate the problem. Similar use of market share measures as independent variables has not been replicated in other research, including his later paper. Borenstein and Rose (1994) use flight share, measured as the share of flights operated by a carrier relative to the total in the market, as a proxy for market share. Flight share is less likely to suffer from endogeneity problem, because flight schedules are typically set and reported to the corresponding airports one month prior to departure. A majority of airlines schedule their flights and make available of such information to consumers three to six months ahead of time. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 43 Flight share is also found to be highly correlated with origination share (0.71) and route share (0.75) in my data. Higher flight share reflects higher market share, hence market power and expected prices of tickets. Airport Presence Berry (1990) suggests that large airport presence enables an airline to gain large degree of bureaucratic control over airport operations and block entry or expansion of rivals. Airport presence is measured separately by the number of origin and destination airports served by an airline. In addition to being a form of product differentiation, airport presence in essence translates into higher market power “under the umbrella”, not just in particular airport(s) or route(s). In Morrison and Winston’s (1989) demand analysis, it is also found that consumers prefer airlines that have large operations out of the origin city (i.e. high presence at destination airports). Although their work has been criticized as having numerous serious flaws (such as treating price as exogenous), the results resemble that obtained by Berry (1990). In the empirical analysis, I find that the effect of airport presence is largely captured by the various market share measures and does not contribute to improving the model fit. Further, airport presence in the origin and destination are very highly correlated (0.91), and create multicollinearity issues (with VIF ranging from 12 to 19.73). Therefore, airport presence is dropped from the final model. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 44 Ticket Restrictions Landing and takeoffs privilege at an airport is a scarce resource, and is particularly valuable to airlines during high traffics. Since airports tend to be less busy on weekends and on off-peak hours, airlines use price discrimination to alleviate congestion at peak usage times. One major tool that airlines use to discriminate price inelastic consumers (business travelers) from relatively more price elastic consumers (tourists) is through restrictions imposed on tickets, such as weekend stay-over and advance purchase requirements (Clemons, et al. 2002, Dana 1998, Gale and Holmes 1993, Stavins 2001), and differential pricing between peak hours and off-peak hours (Hayes and Ross 1998) . In sum, tickets that include flights departing during peak hours, without weekend stay-over restriction, and closer to the departure dates are expected to be more expensive. 2.4.2 Competition Herfindahl Index Herfindahl index has been widely used in airline pricing literature to capture the extent of competition in the market. It has been operationalized in three different ways: number of passengers, number of flights, and number of carriers serving the route. Borenstein (1989) constructs origination Herfindahl (ORGHERF) as the weighted average of the Herfindahl indices for passenger originations at the two 2 To my best knowledge, none o f the earlier studies have accounted for flights’ departure times o f the ticket due to unavailability o f such data. Prior research has used variance in load factor (lower) and plane size (larger) as measures for peak-load pricing. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 45 endpoints of the observed routes. He also constructs route Herfindahl (RUTHERF) for all domestic origin-destination passengers on the observed route. Borenstein suggests that when competition is characterized by many small firms (Herfindahl indices are small), competition is less intense while collusion possibility is higher when the market is dominated by a few large firms (Herfindahl indices are large). The expected overall effect of Herfindahl on price is ambiguous, and the findings from his empirical analysis are also inconclusive. Stavins (2001) uses the number of flights in each route to calculate each carrier’s market share and the corresponding Herfindahl index as proxy for competitiveness of the market. Similarly, Borenstein and Rose (1994) use Herfindahl of flights to measure concentration. Hayes and Ross (1998), on the other hand, adopt a combined approach by constructing average endpoint Herfindahl from number of scheduled flights while route Herfindahl from the number of carriers serving a route. Stavins (1996) find the Herfindahl index to be negatively associated with price. Borenstein and Rose (1994) and Hayes and Ross (1998) find that Herfindahl constructed from number of flights to be negatively associated with the degree of dispersion in prices in the market, while Hayes and Ross report additional finding on the Herfindahl constructed from number of carriers to be positively associated with dispersion. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 46 Market Concentration Market concentration is a measure of the degree of competition on a route, and has been operationalized in two different ways. Borenstein and Rose (1994) use the concept of “density”, measured by the total number of flights on the observed route, to capture market concentration. Hayes and Ross (1998) and Stavins (2001), on the other hand, use number of carriers offering services in each route as a proxy for the route Herfindahl and as an alternative measure of market competition. I find that number of carriers serving a route to be highly (and negatively) correlated with the three Herfindahl measures - flight Herfindahl (-0.69), origination Herfindahl (-0.73), and route Herfindahl (-0.62) - while providing the best fit to the model. As discussed in the market share section above, the use of origination and route shares suffer endogeneity problems. Therefore, I use the number of carriers offering service on a route as market concentration to capture the competitive aspect of the market. High market concentration implies a more competitive market, therefore lower prices. 2.4.3 Market Structure Slot-constraint Slot is a measure of the extent to which one or both the endpoint airports are congested. Currently there are four slot-constrained airports in the domestic air transportation market in the U.S.: Chicago O’Hare (ORD), Kennedy (JFK) and La Guardia (LGA) in New York City, and Washington National (DCA). Since slot R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 47 corresponds to scarcity of resources at the airport, the opportunity cost o f operating in slot-constrained airports is higher and hence higher prices are expected if one or both endpoints of a route are slot-constrained. Prior studies find mixed results on the relationship between slot and price; while Berry et al. (1997) and Fournier and Zuehlke (2004) find a positive relationship, Stavins (2001) finds that the effect of slot on price is negative after controlling for Saturday-night stay-over and advance-purchase requirements. Distance (Shorthaul) The variable costs of operating a flight are directly associated with flight distance between the two endpoint airports; hence higher prices are expected for flights with longer distance (Borenstein 1989). On the other hand, total distance of a flight varies by intermediate point (Hayes and Ross 1998). As length of a flight increase, there can be more variability in the carrier’s choice of intermediate airports and potentially generate greater economies of scale. Empirical findings from prior research suggest that cost being the dominant factor in explaining the relationship between distance and price (Berry, et al. 1997, Borenstein 1989, Stavins 2001). 2.4.4 Cost Number of stops The number of stops made by passengers traveling on a given route has two effects on price. From the airline’s cost perspective, increase in number of stops implies R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 48 more takeoffs and landings and shorter length of haul. This contributes to increase in cost due to higher fixed cost of operation. On the other hand, stops decrease the quality of the flight and consumer’s willingness to pay for the product. Therefore, Borenstein (1989) posits that the overall effect of number of stops on price is ambiguous. Prior empirical findings suggest that the quality effects dominate the cost effects (Borenstein 1989, Clemons, et al. 2002). Frequency Borenstein (1989) suggests that there is a negative relationship between frequency and per-flight cost because higher frequency implies greater aircraft utilization. High aircraft utilization has been repeatedly suggested as one of the top three factors in cost reduction by low cost carriers, such as the Jetblue Airways and Southwest Airlines3. On the other hand, higher frequency implies more choices on departure times and lower delay, hence increasing value of the products, particularly for business travelers who place higher value on their time. The number of scheduled flights by an airline on a given route can be used as a proxy for market share (Hayes and Ross 1998). Frequency is found to be highly correlated with flight share in my data (0.81), and offers superior fit compared to the 3 See the annual reports o f these two airlines. Available online: http://www.cuiTan-connors.com/ib2003/mda.html and http://www.southwest.com/investor relations/swaar03.pdf. For additional reference on relationship between aircraft utilization and cost, see Transportation Research Circular, April 2001, published by the Federal Aviation Administration and Transportation Research Board. Available online: http://gulliver.trb.org/publications/circulars/ec027/ec027.pdf R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 49 alternative measures of market power (origination share and route share). Therefore, I use frequency to capture the composite effects of market power as well as cost saving from greater aircraft utilization. Equipment Size Larger equipment has a lower per-seat-mile cost on flights of more than 500 miles (Borenstein 1989). It is more cost effective to operate larger aircrafts on middle- to long-haul routes where the fixed cost of takeoffs and landings is offset by savings in variable cost when the flight is airborne (Berry, et al. 1997). Since the mean distance for the origin-destination airport pair in my sample is 1021 miles, airlines are likely to enjoy economies of scale by flying larger aircrafts; therefore average size of the equipment is expected to be negatively associated with price. Cost Per Available Seat Mile (CASM) CASM is a direct measure of unit cost of operation. Carriers with higher CASM are expected to charge higher prices to cover the expense of providing services in a given route. While Borenstein (1989) uses the weighted average of CASM of all competing carriers on the route (other than the observed carrier) to measure competitiveness of other airlines in the market, I find it more appropriate to incorporate this cost variable directly for the observed airline for two reasons: First, there are already alternative measures for the relative competitiveness of the market R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 50 (e.g. market concentration, market share measures); second, the effects of airline’s own cost on its pricing should be more profound that those of competitors’ costs. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 51 CHAPTER 3 DATA AND METHOD 3.1 DATA 3.1.1 Overview I use pricing data of airline tickets available from the Internet to conduct my analyses. The primary reasons for choosing the airline industry and using airline tickets as the product in my investigations are: 1) According to the U.S. Travel Market Forecast by Jupiter Research in 2004, online airline bookings in U.S. were expected to total over $30 billion; accounting for 26% of overall air booking revenues. Over 50% of consumers purchased more than half of their travel needs online, with 29% made all their travel arrangements on the Web - this is one of the fastest growing, multibillion-dollar online business; therefore understanding the role of pricing strategy is very important for this market4. 2) Online travel agents (OTAs) such as Expedia and Travelocity possess all the necessary characteristics o f an intermediated electronic market (IEM). Not only can consumers find price and product information of the various offerings from multiple vendors, but they can also directly purchase the products from these sites. 3) The attributes of an airline ticket are clearly defined: price, airline, departure and arrival airports and times, number of connections, class of service, and fare restrictions, etc. 4 In an updated forecast by Jupiter Research in 2005, it is reported that nearly one-third of US travel will be booked online, with online agencies accounting for nearly half of that total. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 52 3.1.2 Collection The data was collected over a period of four months in 2004. Intelligent agents were developed using open source scripting languages (Perl and Curl). These intelligent agents sent out reservation requests daily to websites where airline tickets can be purchased. These websites include online travel agents as well as individual websites of both EDLP and HILO sellers. Following Clemons et al. (2002), I specify routes (departure and arrival airport-pairs), service class (economy, non-refundable), and travel dates and times in my ticket requests to eliminate variations across websites from these sources. Further, based on extensive literature in airline pricing, I incorporated a number of additional control variables that have been shown to affect the pricing of airline tickets5. The explanations of variables are summarized in Table 1. To eliminate the possibility of periodic fare changes that may influence the results of my analyses, I adopted simultaneous data collection across the different websites by running multiple agents in parallel. As a result, I collected a total of 1,137,500 individual tickets that are characterized by one to four weeks of advance purchase from each website. The final sample includes observations on the fourteen largest domestic airlines and three regional airlines, composes a representative sample of the 500 busiest routes (directional) that account for 86.51% of domestic passenger 5 Sources for data on the routes and carriers in the sample (e.g. slot, CASM) are: Airline Origin and Destination Survey Data Bank (DB1B), Service Segment Data, and Quarterly Airline Financial Data from the Bureau of Transportation Statistics (BTS); and Domestic Airline Fare Consumer Quarterly Report from the Department of Transportation (DOT). R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 53 Variable Description EDLP Dummy = 1 if the ticket is written by an “Everyday Low Price” Seller; 0 otherwise fwcon Number of connections from origin to destination on the forward journey retcon Number of connections from destination to origin on the return journey fwPEAK Dummy = 1 if the flight departs between 6am to 10am or 3pm to 7pm on the forward journey; 0 otherwise retPEAK Dummy = 1 if the flight departs between 6am to 10am or 3pm to 7pm on the return journey; 0 otherwise redeye Dummy = 1 if the flight departs after 10pm or arrives between midnight and 6am on either forward or return journey; 0 otherwise multicarr Dummy = 1 if it is a ticket combined of multiple legs from different carriers; 0 otherwise DD7 Dummy = 1 if the ticket is requested 1 week before departure date; 0 otherwise DD14 Dummy = 1 if the ticket is requested 2 weeks before departure date; 0 otherwise DD21 Dummy = 1 if the ticket is requested 3 week before departure date; 0 otherwise freq Average daily frequency of (round-trip) flights offered by a particular airline on the given route hub Dummy = 1 if either the origin or destination (or both) airport is a hub airport for the carrier writing the ticket; 0 otherwise shorthaul Dummy = 1 if the non-stop distance between the origin and destination airports is less than or equal to 500 miles; 0 otherwise slot Dummy = 1 if either the origin or destination (or both) airport is a slot-constrained airport; 0 otherwise mktcon Dummy = 1 the total number of airlines offering services (non-stop or indirect) on the observed route. CASM Cost per available seat mile for the carrier serving on the observed route avgEQUIP Average equipment (aircraft) size for the carrier serving on the observed route EDLPxDD7 Interaction Dummy = 1 if the ticket is written by an EDLP seller and is requested 1 week before departure date; 0 otherwise EDLPxDD14 Interaction Dummy = 1 if the ticket is written by an EDLP seller and is requested 2 weeks before departure date; 0 otherwise EDLP x DD21 Interaction Dummy = 1 if the ticket is written by an EDLP seller and is requested 3 weeks before departure date; 0 otherwise EDLP x shorthaul Interaction Dummy = 1 if the ticket is written by an EDLP seller and the observed route is characterized by shorthaul; 0 otherwise EDLP x hub Interaction Dummy = 1 if the ticket is written by an EDLP seller and either the origin or destination (or both) airport is a hub airport for the carrier; 0 otherwise Table 1: Explanation of Variables R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 54 enplanements in the U.S. according to the most recently published quarterly report from the Department of Transportation (DOT) available at the time of collection. 3.1.3 Bias Control I adopted numerous measures to control for error and potential bias in my data collection. First, since tickets available online can be separated into those that are obtained from individual airlines’ websites and those that are obtained from IEM such as Travelocity and Orbitz; to separate the price effects of individual airline’s pricing strategy from those due to competition in IEM, I recorded the origin of the ticket (i.e., from which website the ticket is obtained) in addition to the carrier-specific information. For data accuracy, I monitored prices of tickets obtained from third-party websites, such as Yahoo! Travel, that connected to the same OTAs (Travelocity for the case of Yahoo!). Second, tickets of up to 10 routes were chosen in random each day and cross-checked with results generated from web-browser requests for every website from which the agents inquired to ensure consistency of data collected by the agents. I also monitored changes on the individual websites and modified the agents accordingly. Third, in case of interrupted connection or high Internet traffic that caused incomplete collection, the gaps were filled in on the same night or next morning by submitting the identical requests for the missing routes. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 55 Fourth, logical checks were performed on each dataset to ensure that no erroneous information was generated. For example, I checked for the existence of unidentifiable airline information, negative number of stops, zero or negative price, etc. In case when the data did contain error, I investigated the cause of the errors to ensure that it was not due to defected functionality of the agents or change in website design; the data points with inaccurate information were subsequently deleted. Fifth, additional information such as flight number and the departure and arrival times on both forward and return journeys, was included in the data collection. The primary use for such information was to control for the exact, identical tickets across time for my temporal price variability analysis. Further, it was observed that certain outliers (e.g. tickets that cost over $1000 or with unusual number of connections) were due to IEM generating round trip tickets that were composed of multiple one-way tickets offered by different airlines, or tickets with multiple legs from various carriers. I added a variable “multicarr” to identify such tickets and to control for their effects on prices6. 3.1.4 A Note on EDLP Definition Bell and Lattin (1998) point out that although “pure versions of HILO and EDLP seldom exist in practice and EDLP/HILO is best thought of as a continuum” (p.68), 6 Exclusion of these tickets did not change the results. While marginal changes were observed in the coefficient estimates of a few airline-specific variables, the statistical significance and signs of these variables remain the same. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 56 stores can still be differentiated according to their relative tendency to operate one format or the other. For example, they observe from their sample that not only are there clear differences in pricing patterns between the two stores (they observe that both average basket price and price variability over time are lower for EDLP stores), but the differentiation on pricing format is also clearly emphasized in store advertising. Since one of the objectives of my study is to examine the actual pricing behaviors of firms that declare themselves to be “EDLP” sellers, I therefore use the second observation as a guiding principle in defining the type of price formats adopted by the firms in my sample. For a firm to be identified as adopting the EDLP strategy, it must 1) Emphasize such a practice in their advertising messages; and 2) Consciously refer to their pricing practice as “everyday low price” in their financial reports and communications to investors. 3.2 METHOD 3.2.1 Introduction to Hierarchical Modeling I em ploy a m ethod called hierarchical m odeling (HM) for my analyses. Hierarchical models are also known as “hierarchical linear models” (Raudenbush and Bryk 2002, Singer 1998), “multilevel models” (Goldstein 1987, Webster, et al. 1998), “variance component models” (Aitkin and Longford 1986, Longford 1993), “random coefficient regression models” (De Leeuw and Kreft 1986, Longford 1993), “mixed-effects models” and “random effects models” (Hsaio 2003). Hierarchical modeling is a specialized regression technique designed to analyze data that exhibits R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. a hierarchical structure (Goldstein 1995, Raudenbush and Bryk 2002). For example, in the airline industry, it is very common to observe multiple tickets being offered by each airline for a given origin-destination pair. The individual ticket prices are considered “nested” within each airline unit. The major problem in traditional regression approaches such as multiple regression and logistic regression that use ordinary least square (OLS) technique on hierarchical data is that, in addition to the assumptions of linearity and no measurement error in predictors, the random errors are assumed to be independent, normally distributed, and are homoskedastic (constant variance). In other words, individual observations are assumed to be independent and that any one observation is not in any way systematically related to others. This assumption of independence is likely to be violated in nested data where some of the observations are sampled from the same individual unit (firm), because a portion of the random error is a unit-level error that is constant across observations within the units. In the airline industry, prices of tickets offered by a carrier are likely to be correlated because they are written by the same airline with the specific cost structure and pricing strategy. The dependence among observations is also refereed to as Intra-Class Correlation (ICC). The OLS assumption of independent errors is violated in the presence of ICC (Kreft and De Leeuw 1998), and the standard errors of the coefficients are underestimated. This raises the risk of R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 58 inferring that a relationship is statistically significant when it may have occurred by chance alone (type-I error) (Pedhazur 1997). Further, the assumption of constant variance is likely to be violated in airline pricing data, because unit-level random error is likely to vary across airlines. In the OLS approach, statistical tests that involve variables at the higher-level unit are based on the total number of lower-level units and not the number of higher-level units, which can influence the estimates of standard errors and associated statistical inferences (Raudenbush and Bryk 2002, Tate and Wongbundit 1983). 3.2.2 Idea Behind HM Hierarchical models extend traditional linear regression models by relaxing one of the critical assumptions, namely independence of residuals (Snijders and Bosker 1999). Hierarchical models take into account the partial independence of individual observations within the same group as well as the fact that these observations may be more similar to one another compared to those belong to another group. Hierarchical models allow researchers to examine the effects of variables at more than one level, removing the traditional constraint that only one unit of analysis can be investigated. The fundamental idea behind hierarchical modeling is that there are separate analyses for each of the unit at the lowest level of a hierarchical structure, while both R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 59 individual-level and group-level unit variances are examined in the outcome measure by estimating variance between groups and examining the effect of variables at each level simultaneously. The total variance in the outcome is divided into the parameter variance and error variance components. Unlike OLS, hierarchical models estimate residuals from different levels separately and account for the covariance structure among group-level regression estimates, giving more accurate group effect estimates than traditional methods that systematically underestimate them (Raudenbush and Bryk 1989). This also allows one to model explicitly both within- and between- group variances as well as their effects on the outcome while m aintaining the appropriate level of analysis (G riffin and Hofman 1997). For the purpose of this research, one critical advantage of using hierarchical models is that they allow for incorporation of airline and market characteristics into the model of individual ticket prices while at the same time producing accurate estimates of the group-level effects and the construction of the corresponding valid tests and confidence intervals (Mendro, et al. 1995), which are typically ignored by the OLS (Bryk and Thum 1989). Traditional fixed effects models use the dummy variable approach to “absorb” all observable and unobservable heterogeneity across different group units, hence the inclusion of any group level-specific characteristics will cause multicollinearity problems. In other words, when using fixed effects models to address the multilevel nature of hierarchical data, level-two variables cannot be included into the model specification because they will be confounded with the R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 60 group fixed effects and the parameters of the model will be unidentifiable. In the airline pricing context, this implies that airline and market specific attributes cannot be explicitly accounted for in the model, limiting a researcher’s ability in drawing inferences on the mediating effects of these characteristics on the relationship between other explanatory variables (such as pricing strategy) and ticket prices. While typically such effects can be incorporated into the model using interactions between the explanatory variables and group level dummies, when the number of groups (such as origin-destination pairs) is large, the interaction approach becomes impractical and will lead to overidentification of the model due to the large number of parameters. 3.2.3 Estimation Methods There are in general two maximum likelihood methods that are commonly used in estimating hierarchical models; the full maximum likelihood (ML), and the restricted maximum likelihood (REML). In ML, both fixed effects and variance components are included in the likelihood function. Variance-covariance parameters and second-level fixed coefficients are estimated by maximizing the joint likelihood. In other words, ML maximizes the likelihood with respect to all of the parameters in the model simultaneously and thus estimating both fixed and random effects at the same time. An advantage of ML is that the deviance of the data is being minimized, but at the cost of potential downward bias in the variance components estimates; though such biases are generally small (Hox 2002). R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 61 In REML, only the variance components are included in the likelihood function. REML treats estimates for regression coefficients as carrying some amount of uncertainty and are being estimated in the second step of the estimation. Unlike ML, variance-covariance components are first estimated with maximum likelihood that integrates over all possible values of the fixed effects, which are then recovered using generalized least square (GLS) given the variance-covariance estimates obtained from the first step (Raudenbush, et al. 2001). Finally, GLS is applied to yield estimates and standard errors for the fixed effects (Goldstein 1995, Raudenbush and Bryk 2002). Two advantages of REML are that GLS estimates require only a small number of iterations (Goldstein 1995), and by removing the fixed effects from the model in estimating variance-components, REML can lead to theoretically less bias than ML when the number of groups is small. Since REML estimates correct for the loss of degrees of freedom due to estimating the fixed effects, they do not suffer from the downward bias as do ML estimates. However, Raudenbush and Bryke (2002) point out that such difference is usually trivial in practice. Further, when the number of level-2 (group) units is large, the estimates from ML and REML are nearly identical. I choose to adopt REML in this research because of two important differences between ML and REML: First, REML minimizes the deviance of the least squares residuals as opposed to minimizing deviance of the data. Second, although ML is consistent and asymptotically efficient, it does not adjust for the number of fixed R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. effects that are being estimated; hence the variance components will tend to be underestimated with small sample sizes (Jones and Steenbergen 1997). Although the data sets I use in this research do not fall into the “small sample size” category, in some of the models where the dependent variables are aggregated measures (such as range and coefficient of variation), the number of observations reduce to only several thousands. As a precautionary measure as well as to be consistent with the majority of research using hierarchical models, REML is employed in all of my analyses. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 63 CHAPTER 4 A STUDY OF PRICING STRATEGY OF ONLINE EDLP VENDORS 4.1 CONCEPTUAL DEVELOPMENT EDLP has proven to be a successful strategy in building a low price image in the physical context. The geographical distance between EDLP and HILO stores creates a natural barrier that helps EDLP sellers to avoid head-to-head price comparison and maintain a low price reputation even when their prices may not be the lowest in town. A basic premise for the success of EDLP, therefore, appears to be that consumers will encounter difficulties in searching for the lowest price offered by every vendor in the market. While managing this perception in the physical context is much easier, in electronic markets consumers can easily price-search multiple stores and thus threaten the ability of sellers in adopting this particular price format. The first part of my study aims to address three main questions: 1. Are the pricing behaviors of self-declared EDLP sellers truly consistent with the image they are trying to deliver? 2. Are there any differences in the way EDLP is adopted in the online context compared to offline? 3. Do online sellers employ category-specific (hybrid) price format? If so, is the extent to which this is practiced being different from that in offline markets? R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 64 Extant research on retail price formats focuses on three dimensions of everyday low price: price level, temporal price variability, and price range. Studies find mixed evidence in the offline context when examining EDLP sellers’ actual pricing behaviors in these regards. Ho, Tang and Bell (1998) find that EDLP stores charge lower than average prices compared to HILO stores, with EDLP prices also exhibiting lower price variation. Hoch, Dreze and Purk (1994), on the other hand, report that although prices in EDLP stores are on average 9% lower than those at the HILO stores, EDLP sellers engage in price promotions on just as much merchandise as their HILO counterparts. Shankar and Bolton (2004) observe only marginal difference on price variability between EDLP and HILO sellers, but find strong evidence that EDLP price level to be lower and fall within a narrower range compared to that of HILO stores. Following prior research, these three dimensions will be separately examined in this study. The search cost literature suggests that due to low cost of search online, consumers are able to engage in more aggressive price comparison than in the offline context (Bakos 1997). Further, the existence of intermediated electronic marketplaces (IEM) (Chircu and Kauffman 2001), such as Orbitz and Travelocity, allows consumers to receive multiple price quotes and product information from different vendors with a single visit. Online EDLP sellers are therefore subjected to relatively costless and direct price comparison with other vendors, and are at a much disadvantaged situation compared to operating in the physical environment. These R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 65 characteristics of electronic markets would likely discourage the adoption of EDLP strategy and as a result, there could be little or no observable differences between EDLP and HILO pricing online. Empirical research in information systems, however, find substantial evidence that price dispersion persists online even in homogeneous product markets (Brynjolfsson and Smith 2000). Others also suggest that despite the low search costs, not all online consumers would search extensively or buy from the seller with the lowest price (Barua, et al. 1997). Results from existing empirical research point to the existence of barrier to perfect price competition, suggesting the possibility of EDLP strategy in maintaining low price image even in the online context. Therefore I expect to find differences between online EDLP and HILO prices in a similar fashion as observed in the offline context. Hypothesis 1 (HI): Price levels set by EDLP sellers online are lower compared to those by HILO sellers. Most of the existing research in retail price format focuses on the measure of price level and variability across time. In other words, the comparison is based on average level and change of prices of the same product over a period. In contrast to supermarket purchases on which the majority of EDLP research is based, however, a substantial portion of online retailing is not characterized by routine and repetitive R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 66 purchases of the same products, though they may belong to the same categories. For example, while consumers may routinely purchase books and CDs online, they are not likely to repeatedly buy the same books or the same CD titles over the span of time. A significant implication of this difference is that as opposed to supermarket shoppers who are more concerned about the frequency and size of temporal price changes of the specific products they routinely purchase, a majority of online shoppers are more concerned about whether they can get a good deal for the product of interest at a given point in time. On the other hand, in airline markets repetitive purchase behavior may still be observed, particularly for business travelers. Therefore, in addition to low temporal price variability, online EDLP sellers may build their low price image by offering low and less volatile prices within any given market so that they signal to consumers that they can expect “high probability of getting a good deal”, regardless of which specific product they are considering. Further, some online EDLP sellers, most notably Southwest Airlines and Walmart.com, emphasize their offering of low prices not only for certain products in particular markets but low prices for any product in any market. For example, Southwest Airlines follow a strategy similar to that of “99 cents” stores by routinely advertising a fixed price for flying from anywhere to a particular city or between any two cities within a region. Therefore, I hypothesize that online EDLP sellers may appeal to consumers by maintaining low variability in prices i) across time, ii) within a market for any given time, and iii) across markets. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 67 Hypothesis 2a (H2a): For any given product offered by an EDLP seller, the temporal price variability is lower compared to those of HILO sellers. Hypothesis 2b (H2b): Price variability of an EDLP seller within a market is smaller than those of HILO sellers at any point in time. Hypothesis 2c (H2c): Price variability of an EDLP seller is universally lower across different markets compared to those of HILO sellers at any point in time. While the average price level need not be associated with the range of prices offered by a seller, by setting a wider range of prices EDLP sellers increase the probability of , a consumer finding higher prices, and therefore damaging their low price image. In particular, in the online environment where direct price comparison is made easier for the consumers, sellers who engage in the everyday low price strategy should not only offer lower average prices but also restrict the maximum price charged to minimize the chance that a consumer finding better offers from others. Further, from the HILO sellers’ perspective, easier price search implies more severe price competition. While in the physical context they can employ price discounts and deals to discriminate uninformed consumers from those who are informed; in an electronic market, the portion of informed consumers (or those who have increased ability to “become informed” due to lower search costs) becomes larger. This implies deeper price cuts are needed by the HILO sellers’ in their deal prices to attract the cherry-picking, deal-shoppers as consumer’s relative costs of obtaining price information from both EDLP and HILO decreases (Lai and Rao 1997). To R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 68 compensate their loss in profit margin from deals, HILO sellers need to set higher regular prices to extract more premiums from consumers who are less price elastic; which translates into wider price range. Therefore, I propose that the range of prices offered by EDLP sellers to be narrower compared to those by HILO sellers online. Hypothesis 3a (H3a): The intertemporal price range for any given product offered by an EDLP seller is narrower compared to those of HILO sellers. Hypothesis 3b (H3b): The range of prices offered by an EDLP seller in a market is narrower compared to those by HILO sellers at any point in time. Notice that EDLP is not a strategy that delivers the promise of “lowest price available”, but rather one that simply helps the sellers to create a “low price” image. Therefore, it is uncertain whether such an image can be sustained when consumers can easily find out when some (if not all) of the prices charged by the EDLP sellers are not the lowest available online. While always charging the lowest possible price in the market, if not cost-prohibitive, obviously cannot be a profit-maximizing strategy, it would be devastating to the low price reputation of these sellers if online consumers can easily find lower prices elsewhere. It is therefore particularly challenging to online EDLP sellers in handling the delicate balance between revenue generation and potential damage to their low price image. One key insight from Bell and Lattin’s (1998) finding is that it would be optimal for an EDLP store to R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 69 increase its price closer to the average price in the HILO store as a shopper increases her tendency to become a large basket shopper, while EDLP store must lower its price closer to the deal price in the HILO store as the shopper exhibit more characteristics of a small basket shopper. This finding has significant implication to the context of electronic markets. While the net effect of the online medium on consumer price sensitivity towards differentiated products remains ambiguous and depends on whether product information is made more accessible for easy comparison of non-price attributes of the products across stores (Alba, et al. 1997, Degeratu, et al. 2000, Lynch and Ariely 2000, Shankar, et al. 2001), for markets where the products are homogeneous or where product attributes are clearly defined and are easily comparable, we shall expect online consumers to display behaviors more consistent with those of small basket shoppers in the offline context. Therefore, I expect that the difference between the average price available and the minimum price observed in the market to be closer for EDLP sellers as opposed to their HILO counterparts. Hypothesis 4 (H4): The average prices offered by EDLP sellers are closer to the lowest price available in the market compared to those by HILO sellers. In their discussion on future research, Ho, Tang, and Bell (1998) suggest that firms may adopt EDLP for certain products while HILO for others to maximize profits. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 70 Until recently this speculation had remained unverified. Shankar and Bolton (2004) offer the first evidence that vendors engage in category-specific pricing. In the online context, the existence of IEM makes price information more transparent and thus threatens the ability of EDLP sellers to vary their pricing strategy across product categories. This has two effects on the pricing behavior of online EDLP sellers: on one hand it is less feasible for them to adopt EDLP for certain products while HILO for others because it would be easy for consumers to discover the discrepancy in their pricing practice, hence undermining the low price image of these sellers; on the other hand, given that the protection of a low price image is now more costly, consistent EDLP pricing may become less attractive. However, because of cost transparency (Sinha 2000, Zhu 2004), the opportunity cost for sellers in employing everyday low price on all product categories in electronic markets are much higher because EDLP sellers are no longer competing locally with two or three HILO sellers. Further, the lack of physical separation from HILO sellers make their pricing strategy more vulnerable and EDLP sellers are at greater risk of losing sales from consumers who switch to the competitors when they discover better deals elsewhere, which is more likely in online than in physical markets. Intensified competitions along with weakening of barriers that protect the low price image would cause online EDLP vendors to adopt the everyday low price strategy selectively on a limited set of products. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 71 Hypothesis 5 (H5): The extent to which an online EDLP seller practice everyday low price differs from one product category to another. 4.2 MODELS I model the relationship between price, product quality, and firm’s pricing strategy by considering the most relevant and observable dimensions o f the tickets and information on the respective routes and carriers. The functional forms and variables included are based on extensive research on airline pricing (e.g. Berry, et al. 1997, Borenstein 1989, Borenstein and Rose 1994, Brueckner, et al. 1992, Morrison and Winston 1990, Stavins 2001) and online price dispersion in the airline industry (Clemons, et al. 2002). A particular nature of the data is that it is “clustered” in different classes. For example, prices of tickets in the same route will be more similar to each other compared to those of tickets from different routes. This clustering causes errors within each class to be correlated with each other, violating the i.i.d. assumption in OLS. Therefore, hierarchical modeling approach is chosen over the traditional fixed-effects models. However, the data does not exhibit a pure hierarchical structure - i.e. two tickets can belong to the same airline but different routes. Such a lack of unique identity of members within each class is known as cross-classification. I employ log-linear form in the cross-classified hierarchical models for my analyses for two main reasons. First, this formulation is consistent with existing research on airline pricing, and is flexible in allowing for proportional and declining marginal effects of the explanatory variables. Second, it captures the R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 72 percentage change as opposed to absolute change in price, which is consistent with actual pricing behavior in the market. Notice that cross-classification structure of the data is observed only in model 1, where the dependent variable is individual ticket prices. For all other models, unless stated otherwise, the observations are nested only within a particular group (route) due to aggregated measure (such as range and coefficient of variation). Therefore, only one random effect will be modeled. This section presents the models employed in testing the hypotheses discussed earlier in this chapter. Although some of the hypotheses (hypotheses 2c and 5) do not require separate models to be analyzed, for notational convenience and ease of reference, the models are named corresponding to the hypotheses being tested. It should be noted that this first study focuses on comparing the pricing strategies of EDLP and HILO sellers along various dimensions, the data used in the empirical analyses is a subset of the whole data and consists of only markets in which both EDLP and HILO sellers participate. Further, two markets in which EDLP sellers have over 99% of the market share are excluded to control for any bias in estimations caused by potential unusual pricing behaviors in those markets. The resulting data consists of 124 unique origin-destination pairs and the total numbers of observations on individual ticket prices are 138,514 and 133,892 for business and leisure travels, respectively. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 73 4.2.1 Test of Hypothesis 1: Price Levels Model 1: In (priceikm ) = a + Px EDLPk + P2 jwconik m + P3 retconlk m + pJw P E A K l k +p5 retPEAKikm+ p6 redeyeikm+P7 rnulticarrik m +P,DD1 + P9DD14 + PwDD2l +Pu ln { f reclk m ) + PnhuKn + Pnshorthaulm (1 ) +Pu slotm + /?1 5 In ( mktconm ) + p X ( In (CASMk) + PX 1 In (avgEQUIPk) +Pn (EDLPk x DD1) + p x 9 (EDLPk x DD14) + p w (EDLPk x DD2\) +P2 ] (EDLPk x shorthaulm ) + /i2 2 ( EDLPk x hubk m ) + ea " ikm where 0 m Pi =h+u, 1 m uok~ N (0,(p) uom V W 1 mj ■N ' ikm vOy ^0 0 ‘ 'O l Vr io r n j (2) (3) (4) (5) (6) Equation (1) is the basic model to be estimated. Dependent variable priceik m denotes the price of ticket i offered by carrier k in market (route) m at tim et7. Time is represented by the number of weeks between the date at which the ticket is requested and the departure date (i.e. number of weeks of advance purchase). For expositional simplicity, subscript that denotes product category (i.e. business and leisure) is suppressed from all variables. Interactions between EDLP For ease of exposition, subscript t is suppressed from the equations. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 74 and advance purchase, EDLP and market characteristic (short-haul), and EDLP and market power characteristic (hub) are included to capture any variation in sellers’ adoption of this pricing strategy8. The hierarchical structure of the model is presented in equations (2) and (3). y0 represents the overall intercept; u( jk and u0 m are the random carrier and route effects, respectively. uo k is assumed to distributed normally with mean zero and variance tp (equation (4)). The slope of EDLP is considered to be composed of both fixed (y,) and random («lm) effects. In other words, not only are prices related to the carrier’s pricing strategy, but this relationship is hypothesized to also vary across routes. The variance and covariance components for u0 m and u] m are presented in (5). Finally, (6) specifies the white-noise error particular to the individual observation. 4.2.2 Test of Hypotheses 2a: Temporal Price Variability Model 2a: Pv^thn = a + & M o n ik m + p 2 retconlk m + pjnulticarrik m +PJreqk m + Pshubk m + P6 shorthaulm + P-slotm (7) +Pimktconm + P9 EDLP x shorthaulk m + PV jEDLP x hubk m + s ik m 8 Preliminary analysis suggests substantial amount of variations of adoption of EDLP strategy across different routes; a set of interaction variables between (EDLP and market characteristics) were selected based on industry knowledge as potential candidates to improve the fit of data. Stepwise regressions were then performed and the interactions yielding the most improvement were selected. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. Sik m 75 where a = y0+uok+uO m (8) «o»~W (0 ,<p) (9) ^(0,cr2) (10) and uo k denotes the airline fixed-effects and u0 m represents the random route effects9. Dependent variable pvarjk m is the coefficient of variation of the four prices of a particular ticket i offered by carrier k in market (route) m . It is computed as: T f T p riceik m ,t P V arilcm = ~ ---------------F prveuan.t pvcir^m of a given ticket is the coefficient of variation, measured by the standard deviation of its prices across weeks (t = 1 , T ) divided by the mean of these prices. It should be noted that there are several major differences between this model and the previous one (model la). First, since airline effects are treated as fixed, all airline-specific but time- and market-invariant variables (CASM and avgEQUIP) are excluded from the model. Second, several temporal attributes of the ticket have also been removed from the model. This is because the data used in this model is composed of the identical tickets that are observable throughout the four-week period. The resulting data consists of only tickets that have departure times at peak Unless otherwise indicated, the interpretation of U Q m remains the same in the rest of the models R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 76 hours on both journeys. Therefore, the variables “fwPEAK”, “retPEAK” and “redeye” are not included. Finally, because the unit of analysis is variability of prices across four time periods, dummies that represent time (weeks of advance purchase) are therefore omitted from the model as well. (11) 4.2.3 Test of Hypothesis 2b: Price Variability within Markets Model 2b: P varfo » = « + PX DD1 + P2DD\4 + p JJD 2\ + P Jreq k m +P5 hubl c m + P6 shorthaulm + p 1 slotm + P% mktconm +PgEDLPxDD7 + PwED LPxD D \4 + PnED LPxD D 2\ +PUEDLP x shorthaulk m + P^EDLP x hubk m + s jk m where a ~ Y o U 0k U 0m ( 12) uo „ ~ n (0 > < p) (*3) ~ N(0,cr2) (14) and uo k denotes the airline fixed-effects. Dependent variable pvark m is the coefficient of variation of prices of a set of tickets I offered by carrier k in market (route) m at a given time. It is computed as: i i X I *Pric e ih « - y L P ric e a m P Vark m = - _ / R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 77 Model 2b is similar to model 2a in that airline-specific but time- and market-invariant variables are not included in the model because of the airline fixed-effects. Notice that the unit of analysis is variability across all tickets that are written by each airline in a market at a given point in time, therefore two modifications are made: First, because individual ticket-specific attributes are lost in pooling the tickets for the aggregate measure, fwcon, retcon, fwPEAK, retPEAK, redeye, and multicarr are omitted. Second, time dummies (DD7, DD14, DD21) and their interactions with EDLP are added back to this model. 4.2.4 Test of Hypothesis 3a: Intertemporal Price Range Models 3 a and 3b are structurally identical to models 2a and 2b, respectively. Therefore, I will focus the discussion only on the dependent variable. Model 3 a: rangeik m z=a + + fJ2 retconik m + fi3 multicarrik m + fijreq k m + fi5 hubk m + (3(shorthaulm + /31 slotm +/3& mktconm + (39EDLP x shorthaulk m + PwEDLPx.hubk m + sjk m (15) where a = r 0 +UQk + u 0m (16) N{0,<p) (17) ~ jV(0,ct2) ( 18) R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 78 and uQ k denotes the airline fixed-effects. Dependent variable rangeik m is the standardized difference between the maximum and minimum prices of a particular ticket across time. It is measured as: max (priceik m ) - min (priceik m ) rangelk m = ------- 7 — : — {----- 77— : — r- max ( pricem) - mm (pricem) The comparison of unadjusted ranges of prices across airlines and routes does not provide a meaningful interpretation. In order for different ranges of prices across different markets to be comparable, rangeik m is normalize to be bounded between 0 and 1 and is thus represented as the ratio of difference in maximum and minimum prices of a ticket written by a particular carrier to that of the market across the four weeks. 4.2.5 Test of Hypothesis 3b: Price Range within Markets Model 3b: range k m = a + (i,DDl + fi2D D \4 + & D D 21 + P Jreq k m +P5 hubk n , + P6shorthaulm + P-,slotm + fi8 mktconm +]39 EDLPxDD7 + J 3 wED LPxD D \4 + j3nED LPxD D 2l +(3n EDLP x shorlhaulk m + /3 n EDLP x hubk m + s jk m where a = ro+uok+uom (2°) U Om~N {0,<P) (21) s ikm ~ -^(OjCr2) (22) R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 79 and u0 k denotes the airline fixed-effects. Dependent variable rangek m is the standardized difference between the maximum and minimum prices available from an airline in a market at a particular time. Similar to the dependent variable in model 3 a, it is also normalized to fall between 0 and 1: rnax(pricek m ) - min (price k m ) ™nge^ — ; --- ^ -------—7 — ; ---f max ( pricem) - mm ( pricem) 4.2.6 Test of Hypothesis 4: Individual Seller’s Mean Price versus Market Minimum Models 4 structurally identical to model 3b. Therefore, the discussion will focus on the dependent variable. Model 4: meandiffh n = a + f3X DD1 + p 2 DD\4 + /33DD2\ + P3 freqk m +/3shubk m + P6 shorthaulm + fa slotm + famktconm +faEDLPxDD7 + fa0ED LPxD D \4 + p uEDLPx DD2\ +PUEDLP x shorthaulk m + PUEDLP x hubk m + sik m where (X — Yq + U Q k ^ U < S m (24) ' ; V (<).'?) (25) ^ - N i d . a 1) (26) R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 80 and u0 k denotes the airline fixed-effects. Dependent variable meandiffk m is the standardized difference between the average price of tickets from an airline in a market and the minimum price available from the corresponding market at a particular time. It is measured as: mean ( price^ ) - min ( pricem ) mindiffk km min ( pricem ) The rationale for this adjustment is that, a unit dollar difference between the market minimum price and the carrier’s average price carries different weight depending on the relative level of the minimum price in the market. A $20 difference is more pronounced when comparing with a market minimum price of $100 than if it were $250. 4.3 RESULTS 4.3.1 Test of Model Specification and Robustness Model Specification Multicollinearity and Hausman tests were performed to check for misspecification errors in the models. The results indicate that multicollinearity is not an issue in the models. The highest value of VIF is 8.86, with the next highest VIF being 4.29. Both values fall below the critical level of 10. Further, Flausman tests were performed separately for route random effects and airline random effects. The Chi-square values for the two random effects are 0.001 and 0.002, well below the critical value of 30.58 for the 1 percent significance level at 15 degrees of freedom. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. These statistics indicate that the coefficient estimates from the random effects model are not significantly different from those obtained from the fixed effects model. A comparison on the coefficients of the set of variables used in the Hausman test is presented in table 2, with the first column indicating results from the fixed effects model, and the second column indicating results from the hierarchical model with both route and airline effects being treated as random. OLS HM fwcon 0.1398*** 0.1400*** retcon 0.1741*** 0.1743*** fwPEAK -0.0111*** -0.0111*** retPEAK 0.0135*** 0.0135*** redeye 0.0050 0.0052 multicarr 0.3042*** 0.3042*** DD7 0.2359*** 0.2360*** DD14 0.0454*** 0.0454*** DD21 0.0068** 0.0068** lnfreq -0.0537*** -0.0535*** hub 0.0501*** 0.0503*** EDLP*DD7 -0.0718*** -0.0718*** EDLP*DD14 0.0198*** 0.0197*** EDLP*DD21 0.0385*** 0.0384*** EDLP*shorthaul -0.4922*** -0.4927*** EDLP*hub 0.0889*** 0.0882*** Table 2: Hausman Test R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 82 < p - 0.0355*** ro o - 0.0473*** a 1 0.1177*** 0.1177*** N 138514 138514 -2LL 97739.9 97720.2 BIC 97751.7 97755.7 Significance at 1% (***), 5%(**), and 10% (*) levels . Table 2 continued: Hausman Test Robustness Tables 3 and 4 summarize the results of model 1, including the coefficient estimates, covariance parameter estimates for the random effects, and goodness of fit indexes for business and leisure tickets, respectively. Bayesian Information Criteria (BIC) is reported in addition to the conventional reporting of log likelihood in light of the large sample size. Two baseline models are also presented along with the full cross-classified hierarchical model (HM). Both BIC and %21 d f tests show that the full HM model offers superior fit to the data compared to the fixed effects model and the null model for both types of tickets. Additional robust checks were performed by comparing the sum of residual-squared of the fixed-effects model (OLS) and HM based on the differences between actual (y) and predicted (y) values of the dependent variable. HM outperforms the OLS by a difference of 1 0 Same notations are used in all remaining tables. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 83 7.14% and 4.65% lower in the sum of residual-squared for business and leisure tickets, respectively. Furthermore, the intra-class correlations (p ) for carrier = 0.34 (0.23) from the null ( ~ \ f — \ < p = 0.22 (0.23) and route T o o V V + T o o + 0-2 , k < P + t 0Q+ c t 2 y model suggest that there are fair amounts of clustering of prices within both carrier and route for business (leisure) tickets. This suggests that results obtained from OLS analysis of this data would likely be misleading. The amounts of reduction in variance components c p and r 0 0 suggest that 70.95% (67.28%) of explainable variation in carrier means and 23.59% (29.65%) of explainable variation in route means are explained by the variables incorporated in the full model for business (leisure) tickets. Further, the random error is reduced by 32.88% and 29.16% compared to the second baseline model for the two types of tickets respectively. All statistics indicate that the chosen variables provide excellent explanation of the pricing of airline tickets in the sample. BUSINESS OLS HM NULL HM FULL Intercept — 5.7760*** 6.5681 EDLP — 0.0284 fwcon 0.1398*** 0.1083*** retcon 0.1741*** 0.1466*** Table 3: Results of Model 1 - Price Levels (Business Tickets) R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 84 fwPEAK -0.01113*** -0.0113*** retPEAK 0.01347*** 0.0124*** redeye 0.0050 0.0074** multicarr 0.3042*** 0.2916*** DD7 0.2359*** 0.2335*** DD14 0.04541*** 0.0444*** DD21 0.006821** 0.0047* lnfreq -0.05366*** -0.1278*** hub 0.05005*** 0.0790*** shorthaul — -0.0286 slot — 0.0161 InMktCon — 0.1533*** InCASM — 0.1172 InavgEQUIP — -0.2834 EDLP*DD7 -0.07175*** -0.0691*** EDLP*DD14 0.01976*** 0.0163*** EDLP*DD21 0.03845*** 0.0404*** EDLP* shorthaul -0.4922*** -0.5104*** EDLP*hub 0.08891*** 0.2759*** < P — 0.1277*** 0.0371*** r oo — 0.0831*** 0.0635*** * 0 1 — — -0.0435*** x\\ — — 0.1039*** a 1 0.1177*** 0.1630*** 0.1094*** N 138514 138514 138514 -2LL 97739.9 142740 88134.7 BIC 97751.7 142775.5 88193.9 Sum of Residual-Sq 16284.5 15121.4 Table 3 continued: Results of Model 1 - Price Levels (Business Tickets) R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. LEISURE OLS HM NULL HM FULL Intercept -- 5.8525*** 5.8972*** EDLP -- -0.0199 fwcon 0.1487*** 0.1236*** retcon 0.1677*** 0.1485*** fwPEAK -0.0144*** -0.0170*** retPEAK -0.0359*** -0.0396*** redeye -0.0178*** -0.0138*** multicarr 0.2214*** 0.2189*** DD7 0.2602*** 0.2608*** DD14 0.0914*** 0.0911*** DD21 -0.0035 -0.0030 lnfreq -0.0732*** -0.1324*** hub 0.0467*** 0.0721*** shorthaul - -0.1607*** slot - -0.0118 InMktCon - 0.1019* InCASM - 0.0604 InavgEQUIP - -0.0651 EDLP*DD7 0.0249*** 0.0278*** EDLP*DD14 -0.0606*** -0.0604*** EDLP*DD21 0.0898*** 0.0880*** EDLP* shorthaul -0.4587*** -0.4282*** EDLP*hub 0.0188** 0.2094*** < P - 0.07426*** 0.0243*** ^00 - 0.07434*** 0.0523*** 70 1 - - -0.0327*** T1 1 - - 0.0896*** a 2 0.1250*** 0.1684*** 0.1193*** Table 4: Results of Model 1 - Price Levels (Leisure Tickets) R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 8 6 N 133892 133892 133892 -2LL 102550.8 142290.6 96768.5 BIC 102562.6 142326.0 96827.5 Sum of Residual-Sq 16716.4 15938.6 Table 4 continued: Results of Model 1 - Price Levels (Leisure Tickets) 4.3.2 Results of Hypothesis 1: Price Levels The results from model 1 (tables 3 and 4) offers partial support for the hypothesis that prices of online EDLP sellers are lower compared to those of HILO sellers. The first interesting observation is that the main effects of EDLP are insignificant in both types of tickets, implying that there does not exist an “overall” low price strategy by EDLP sellers. The most striking difference between EDLP and HILO prices are observed in tickets for routes less than 500 miles and those in which EDLP sellers have hub airport(s). EDLP prices in short-haul markets are on average 39.97% and 34.83% cheaper than HILO prices for business and leisure tickets, respectively1 1 . On the other hand, prices offered by EDLP sellers are higher compared to those by HILO sellers in hub 1 1 The relative effect o f a dichotomous variable coefficient (c) on the dependent variable in semilogarithmic equations is 100*(exp(c)-l) (Halvorsen, Robert and Raymond Palmquist, "The Interpretation o f Dummy Variables in Semilogarithmic Equations," American Economic Review, 70, 3 (1980), 474-475. , Hann, 11-Horn, Jeff Roberts, Sandra Slaughter and Roy Fielding, "An Empirical Analysis o f Economic Returns to Open Source Participation," Working Paper, (2004),. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 87 markets. For business tickets, EDLP prices in hub markets are 31.77% higher compared to HILO prices while 23.29% higher for leisure tickets. In the temporal dimension, the pricing of EDLP sellers is more complicated. EDLP prices are higher for two- and three-week advance purchase business tickets than HILO prices by 1.64% and 4.12%, respectively. For leisure tickets, their prices are higher for one- and three-week advance purchase tickets by 2.82% and 9.2%. Lower prices are found in EDLP sellers’ one-week advance business tickets (6.68% cheaper) and two-week advance leisure tickets (5.86% cheaper). The relative magnitudes of coefficient estimates for the interactions among EDLP and weeks of advance purchase compared to those among EDLP and market specific attributes reveal that, in terms of the “low price” component, online EDLP sellers employ “everyday low price” strategy more extensively in the market-specific rather than the temporal-specific dimension. 4.3.3 Results of Hypothesis 2a: Temporal Price Variability Business Leisure Intercept 0.1435*** 0.2315*** fwcon 0.0075** 0.0055 retcon 0.0117*** 0.0032 multicarr -0.0345*** -0.0682*** Table 5: Results of Model 2a - Temporal Price Variability R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 88 ffeq 0.0004*** -0.0001 hub 0.0103** -0.0118 shorthaul 0.0288** 0.0550** slot -0.0045 -0.0196 mktcon -0.0024 -0.0025 EDLP* shorthaul -0.1039** 0.0447 EDLP* hub -0.0687 0.0996*** EDLP1 0.0001 2 -0.0925* EDLP2 0.0195 -0.0652* < P 0.0032*** 0.0077*** a 1 0.0096*** 0.0138*** N 7368 2690 -2LL -12783.1 -3518 BIC -12765.3 -3502.2 Table 5 continued: Results of Model 2a-T em poral Price Variability The results of model 2a are summarized in table 5. Coefficients for the EDLP main effects suggest that price variability between EDLP sellers and HILO sellers across time are significantly different only in leisure tickets. For business tickets, the only observable difference is in short-haul markets, where the variability of prices offered by EDLP sellers is lower than that of HILO sellers. While price variability is generally lower for leisure tickets offered by the EDLP sellers, in markets where the origin and/or destination is their hub airport, prices are found to be more variable 1 2 No tickets have been identified for this airline. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 89 than those of the HILO sellers. These results offer partial support for hypothesis 2a. 4.3.4 Results of Hypothesis 2b: Price Variability within Markets Business Leisure Intercept 0.1329*** 0.1613*** DD7 0.0555*** 0.0656*** DD14 0.0231** 0.0465*** DD21 0.0011 -0.0099 freq 0.0006*** 0.0010*** hub 0.0508*** 0.0311*** shorthaul 0.0207 -0.0124 slot 0.0535*** 0.0269* mktcon 0.0039 0.0028 EDLP*DD7 -0.0032 -0.0503*** EDLP*DD14 -0.0040 -0.0339* EDLP*DD21 0.0211 0.0328* EDLP* shorthaul -0.0677*** -0.0248 EDLP*hub -0.0600*** -0.0463** EDLP1 -0.1288*** -0.1208*** EDLP2 -0.1511*** -0.1168*** < P 0.0046*** 0.0023*** a 1 0.0213*** 0.0187*** N 2954 2944 -2LL -2597 -3026.3 BIC -2581.0 -3010.3 Table 6: Results of Model 2b - Price Variability within Markets R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 90 Results from model 2b (table 6) indicate strong support for the hypothesis that prices of EDLP sellers are less variable in the market compared to others who do not adopt this strategy. The main effects of EDLP are negative and significant for both business and leisure tickets. This implies that prices offered by EDLP sellers are more consistent compared to those offered by HILO sellers, regardless of whether the Saturday-night stay restriction is imposed on the tickets and how many days prior to departure when the tickets were requested. Further, the low price variability is even more observable in markets where there is a hub airport for the EDLP sellers, for both business and leisure tickets, and in one- and two-week advance purchase leisure tickets. For business ticket, the low price variability of EDLP sellers is being emphasized in short-haul markets but not with respect to the number of weeks of advance purchase. 4.3.5 Results of Hypothesis 2c: Price Variability Across Markets In order to establish low price variability of EDLP sellers across markets, two conditions need to be satisfied. The first condition should establish that EDLP sellers have narrower price range within a market at any given point in time. This condition, however, is only necessary but not sufficient; because despite the narrower price range in a given market, the prices can still vary greatly from one market to another. In other words, while the price ranges of EDLP sellers may be consistently narrower compared to those of HILO sellers in all markets, the prices in R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 91 one market may be very high while those in another market may be very low. Since hypothesis 2c implies that the absolute difference in prices of any two tickets written by EDLP sellers drawn from any given two markets is low, an additional condition that needs to be satisfied is low standard deviation of median prices of EDLP sellers across different markets. Standard Deviation of Median Prices Across Markets Business 1-wk adv 2-wk adv 3-wk adv 4-wk adv EDLP 107.49 108.73 100.48 90.66 HILO 274.04 215.50 222.32 216.48 Leisure 1-wk adv 2-wk adv 3-wk adv 4-wk adv EDLP 133.79 104.67 124.80 95.18 HILO 228.13 171.01 179.27 177.84 Table 7: Variability in Median Prices Across Markets Analysis of the standard deviation of median prices across markets (table 7) offers support to the second condition. The median price variations across markets for EDLP sellers are consistently lower than those for HILO sellers, regardless of the number of weeks of advance purchase for both business and leisure tickets. To complete the test for this hypothesis, however, the relative range of prices offered by EDLP and HILO sellers need to be examined. We shall come back to this in the discussion of results of hypothesis 3b, which offers insights on the relative range of EDLP and HILO prices. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 4.3.6 Results of Hypothesis 3a: Intertemporal Price Range 92 Business Leisure Intercept 0.5009*** 0.5261*** fwcon 0.0904*** 0.1111*** retcon 0.0556*** 0.0702*** multicarr 0.0036 0.0208 freq -0.0007*** -0.0008*** hub 0.1001*** 0.0561* shorthaul -0.0019 0.0042 slot -0.1581*** -0.0699 mktcon -0.0216*** -0.0053 EDLP* shorthaul -0.1360 0.0742 EDLP* hub -0.2776* -0.0794 EDLP1 o.oooo1 3 -0.0748 EDLP2 0.0149 -0.1162* < P 0.0281*** 0.0324*** cr2 0.0449*** 0.0545*** N 7368 2690 -2LL -1407.7 155.3 BIC -1389.9 171.1 Table 8: Results of Model 3a - Intertemporal Price Range Results of model 3a (table 8) suggest that except for leisure tickets offered by one of the EDLP sellers, there is no significant difference in the range of prices between EDLP and HILO sellers across time. Further, unlike the results from earlier analyses, no observable differences are found between the pricing practice of EDLP 1 3 Due to lack o f observation. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 93 and HILO in either short-haul or hub markets in terms of price range across time. Hypothesis 3a is not supported. 4.3.7 Results of Hypothesis 3b: Price Range within Markets Business Leisure Intercept 0.5345*** 0.5656*** DD7 0.0480*** 0.0505*** DD14 0.0381*** 0.0520*** DD21 -0.0080 -0.0033 freq 0.0009*** 0.0018*** hub 0.0646*** 0.0367** shorthaul -0.0656** -0.0717*** slot -0.0030 -0.0618** mktcon -0.0225*** -0.0216*** EDLP*DD7 -0.0262 -0.0716** EDLP*DD14 -0.0056 -0.0622* EDLP*DD21 0.0614* 0.0744** EDLP* shorthaul -0.0547* -0.1052*** EDLP*hub -0.1168*** -0.0704** EDLP1 -0.2457*** -0.2448*** EDLP2 -0.2887*** -0.2347*** < P 0.0136*** 0.0086*** < T 2 0.06435*** 0.0602*** N 3041 3030 -2LL 647.1 406 BIC 663.1 422.0 Table 9: Results of Model 3b - Price Range within Markets R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 94 Hypothesis 3b is supported. The price range of EDLP sellers in any given market is narrower compared to HILO sellers (table 9). The range of prices offered by EDLP sellers is lower in business tickets by 24.57%1 4 to 28.87% compared to HILO prices, and by 23.47% to 24.48% in leisure tickets. Price range is even smaller for EDLP sellers in short-haul and hub markets for both business and leisure tickets, and in most number of weeks of advance purchase (except for three-week advance purchase) for leisure tickets. Results from this model complete the test for hypothesis 2c. Recall that the condition remains to be shown was that at any given point in time, the range of prices offered by EDLP sellers need to be lower than that by HILO sellers, which is supported in the results of model 3b. These findings offer strong evidence that EDLP sellers adopt a pricing strategy similar to that of the “99cents” stores in restricting the price variability across different markets. 4.3.8 Results of Hypothesis 4: Firm’s Mean Price vs. Market Minimum Business Leisure Intercept 0.7204*** 0.7635*** DD7 0.3409*** 0.2182*** DD14 -0.0439 0.1351*** DD21 0.0005 -0.0142 Table 10: Results of Model 4 - Mean Price vs. Market Minimum 1 4 As a percentage o f the overall market range. This definition holds for all percentages presented in the discussion on model 3b. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 95 freq -0.0055*** -0.0078*** hub -0.2191** -0.1952*** shorthaul 1.5642*** 1.2326*** slot 0.8859*** 0.7493*** mktcon 0.0879*** 0.0565** EDLP*DD7 -0.2724 -0.0566 EDLP*DD14 0.0635 -0.1041 EDLP*DD21 0.0587 0.1868 EDLP* shorthaul -1.9373*** -1.6727*** EDLP*hub 0.9753*** 0.7524*** EDLP1 -0.8575*** -0.4589*** EDLP2 -1.3575*** -0.9564*** (P 0.4381*** 0.3057*** u 1 1.4662*** 0.8207*** N 3041 3030 -2LL 10103.7 8347.8 BIC 10119.7 8363.8 Table 10 continued: Results of Model 4 -M ea n Price vs. Market Minimum Results indicate that the gap between individual carrier’s average prices and the market minimum prices are smaller for EDLP sellers than their HILO competitors (table 10). Therefore, hypothesis 4 is supported. This implies that EDLP sellers tend to charge closer to the lowest price available in the market compared to HILO sellers. Notice that the gap between EDLP sellers’ average price and the market minimum is even smaller in short-haul markets, but widens in the hub markets. This suggests that EDLP sellers more aggressively build their image on the low price R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. dimension in markets where they enjoy cost advantages, while at the same time they may charge higher average prices than HILO sellers in markets where they have high market power. Further, no observable difference is found for the various weeks of advance purchase between EDLP and HILO sellers. The following descriptive statistics offer additional insights on this “minimum price practice” by EDLP sellers: Business Leisure EDLP1 196 193 EDLP2 56 48 HILOl 27 35 HIL02 2 1 24 HIL03 47 77 HIL04 9 2 HIL05 13 16 HIL06 1 1 9 HIL07 4 2 HIL08 98 79 HIL09 29 17 HILO 10 1 0 1 0 HILO 11 6 16 Total 527 528 EDLP1 37.19% 36.55% EDLP2 10.63% 9.09% EDLP total 47.82% 45.64% Table 11: Frequency o f an Airline Charging Market Minimum Prices This simple frequency count (table 11) indicates that the minimum prices offered by the EDLP sellers are indeed lowest available in the market for nearly half of the time, R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 97 for both business and leisure tickets. However, notice the asymmetry in percentage, which indicates that one of the EDLP sellers emphasize more on this strategy by offering prices that are more than three times as likely as those by the other in undercutting all competitors in the market. Additional logit analysis was performed and the results are consistent with the observation from these simple statistics: the probability that the minimum prices available from EDLP sellers are indeed the lowest available in the markets nearly half o f the time; 48.15% for business and 50.49% for leisure1 5 . 4.3.9 Results of Hypothesis 5: Category-level Adoption of EDLP Strong evidence is found on the hypothesis that EDLP sellers adopt price format differently in different markets and product categories. The most consistent pricing behavior is observed in short-haul markets, where online EDLP pricing exhibit the characteristics that are consistent with the “everyday low price” image in all three dimensions (price level, price variability, price range) for both business and leisure tickets. However, the pricing practice of online EDLP sellers differ largely with respect to the number of weeks of advance purchase for business and leisure tickets. For example, while EDLP prices generally exhibit low price variability and narrow price range, the effects are more pronounced for one- and two-week advance purchase leisure tickets. On the other hand, in terms of the “low price” dimension, general price levels between EDLP and HILO sellers are not significantly different. 1 5 In terms o f highest price, only 1.2% o f the time the highest price set by EDLP sellers is indeed the most expensive in the market. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 98 Further, EDLP sellers focus on charging low prices exclusively in specific product categories - one-week advance purchase for business tickets, and two-week advance purchase for leisure tickets - and specific types of markets (short-haul) for both business and leisure tickets. 4.4 DISCUSSION Analysis on the difference between sellers’ lowest prices and the market minimum points to an interesting phenomenon: online EDLP sellers focus less on maintaining low average price in general (model 1), but more on offering lowest prices that undercut those of their competitors. While in the analysis of average price levels I find that online EDLP sellers adopt this low price strategy selectively in certain types of market and product categories, analysis on EDLP prices relative to the minimum price available in the market reveals that EDLP sellers compete aggressively in lower bounds of their prices on all types of products, and even more so in certain markets where they have cost advantage. These results offer evidence for the prediction based on Lai and Rao’s (1997) model on EDLP and HILO competition that, as the distance between the two types of firms decreases - due to either decrease in physical distance or lowering of search costs for the consumers in the context of electronic markets - EDLP sellers need to lower their overall level to keep their prices attractive to time-constrained consumers who still have higher search costs relative to the cherry-picking consumers. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 99 Results from models 1 and 2 suggest that online EDLP sellers adopt the everyday low price strategy to different extents depending on the type of market as well as product categories. While they compete aggressively on price level in a limited set of markets and products, online EDLP sellers’ adoption of the low price variability dimension of the strategy exhibits an “umbrella effect” that covers most markets and types of products. Further, the range of prices offered by EDLP sellers online are consistently narrower compared to that by HILO sellers. These findings suggest that EDLP sellers may find maintaining stable prices to be more effective than offering low prices in building an “everyday low price” online, as it is costly to compete on price level in electronic markets due to higher price transparency. These results also offer a potential explanation for why research in online price dispersion finds contradicting evidence on random pricing theory (Baye, et al. 2004, Baylis and Perloff 2002). I find that online sellers who adopt the EDLP strategy set a relatively narrower range of prices closer to the lower bound of market prices, while HILO sellers on the other hand engage in more promotions and discriminatory pricing at the same time. The result is that in markets where both types of sellers coexist, the price-rank of sellers appears stable (Baylis and Perloff) while in other markets where only HILO sellers compete, substantial evidence of “hit and run” strategy and randomness in price-rank are observed (Baye, et al. 2004). R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 100 Online EDLP sellers’ pricing are most consistent with their image in markets where they have cost advantage. EDLP prices in these markets exhibit the classical behavior of an “ideal” everyday low price practice in all three dimensions: low price level, low price variability, and narrow price range. On the other hand, the most “consistently inconsistent” behavior is observed in markets where EDLP sellers have higher market power. EDLP prices in these markets exhibit more promotional characteristics than their HILO competitors in terms of relative price level. The former result suggests that online EDLP sellers focus on building the “everyday low price” image in markets where they can afford to aggressively undercut competitors while maintaining stable prices at the same time, the latter suggests that these sellers may try to recover the loss in profit margins in markets where they can exercise discriminatory power and extract more premiums from consumers in those markets. These results suggest that the adoption of a “hybrid” strategy (EDLP in some markets, HILO in others) may be valuable particularly in online competition, offering the first evidence of category-specific price format adoption speculated in existing literature (Bell and Lattin 1998; Ho et al. 1998). Further, in reconciling the discrepancy in their image on price consistency, online EDLP sellers adopt a strategy analogous to that of a “99cents” store and emphasize the stability of their low prices across different product markets. The most notable difference between online EDLP pricing and that reported by literature in the offline context is that EDLP prices are not in general lower than R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. HILO prices. While online EDLP prices are consistent with the expectations of low within-market variability and range, these results depart from findings of EDLP pricing in the offline markets in that both temporal price variability and range for online EDLP sellers are not significantly different compared to those of their HILO competitors. In other words, the “within markets” characteristics of EDLP are more observable than the “across time” characteristics, suggesting a diminishing role of intertemporal price consistency in the practice of everyday low price in online markets. Part of this result can be attributed to the perishable nature of the product considered in this study; as consumer’s reservation value increases when the departure date approaches, so does the price discriminatory power of the firms (Baye, et al. 2004). An alternative explanation is that reduction in search costs in electronic markets has differential effects on consumer’s price sensitivity along two dimensions. By making price information more accessible, consumers may focus on the comparison of “spot-prices” when they perform the search. However, due to the large number of prices available at the time of search, recall may be poor. As a result, electronic markets may increase consumer’s price sensitivity at any given point in time when they perform price comparison but have a negative effect on consumer’s intertemporal price sensitivity, hence online sellers may find it profitable to engage in randomized pricing across time. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 102 CHAPTER 5 PRICE FORMAT AND PRICE DISPERSION 5.1 CONCEPTUAL DEVELOPMENT While existing theoretical and empirical studies have identified various factors that contribute to the existence of online price dispersion, mixed evidence are found on the relationships among these factors and the degree o f dispersion. Further, a significant portion of price dispersion remains unexplained even after controlling for observable heterogeneities in consumers, vendors, and products. The second study in this dissertation identifies vendors’ adoption of different price formats as a potential source of price dispersion, and investigates its effects on online price dispersion. This study aims to address three questions: 1. Do conscious pricing strategies contribute to price dispersion in the online environment? 2. Does price dispersion exist in an online market for perishable products? 3. Are prices less dispersed online than offline? Marketing literature (Bell and Lattin 1998, Ho, et al. 1998, Hoch, et al. 1994) suggests that retailers adopt different price formats to target different consumer segments. Empirical studies have found significant differences in the range and variability in prices offered by sellers adopting different price formats (Ho, et al. 1998, Shankar and Bolton 2004). Moreover, price formats are also positioning R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 103 strategies (Alba, et al. 1997, Lai and Rao 1997, Ortmeyer, et al. 1991). Differences in sellers’ positioning create differentiations among sellers, which potentially enable them to extract price premiums from consumers and therefore increase price dispersion in the market. Therefore, we should expect to see two phenomena: First, the extent to which prices are dispersed differ between vendors who adopt EDLP and those who adopt HILO. Second, markets in which both types of price formats are present exhibit different degrees of price dispersion compared to markets in which only one type of price format exists. Therefore: Hypothesis 6 (H6): Vendor’s pricing strategy in the form of conscious price format is an important source of price dispersion in both online and offline markets. In markets where sellers adopting both types of price format coexist, consumer’s reservation value could be lower compared to that in markets where only HILO sellers compete. Because EDLP sellers charge a relatively stable level of prices and make their prices more transparent to consumers, consumers may use them as “references” and the basis for evaluating the “fairness” of prices charged by HILO sellers (Alba, et al. 1997, Ortmeyer, et al. 1991). Everything else constant, this would imply intensified price-based competition and HILO sellers may react by matching, undercutting, or reducing the discrepancy between their prices and the everyday low prices, leading to decrease in price dispersion in the market. Further, Shankar and Bolton (2004) find that variability in prices of a seller decreases as R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 104 competitors’ prices are lower and more consistent. This suggests that HILO firms may react to the presence of EDLP sellers, who offer lower prices with higher consistency, by lowering their own price variability and hence leading to lower overall price dispersion. Further, their results indicate that a seller’s relative brand price increases in competitors’ relative price level, implying the range of prices offered by the HILO sellers would also decrease in the presence of EDLP sellers. In sum, we expect to observe lower price dispersion along both dimensions (variability and range) in markets where both EDLP and HILO sellers compete. Hypothesis 7 (H7): The extent to which a firm exercises promotional pricing is influenced by the presence of EDLP competitors. Dynamic price discrimination is characterized by randomized pricing strategies, which is found to be most profitable for vendors selling products that exhibit two characteristics: First, the product is perishable, or expires at a point in time. Second, capacity is fixed well in advance and can only be changed at a relatively high marginal cost (McAfee and te Velde 2004). Airline tickets are an example of products that possess both characteristics, and hence randomized pricing by airlines is generally observed. Further, based on Varian’s (1980) model of sales, Baye and Morgan (2004) demonstrate that price dispersion increases as consumer’s reservation increases. Since consumer’s reservation value for airline tickets increases as the departure date approaches, it is expected that prices of tickets that are closer the R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 105 departure date would exhibit higher dispersion, regardless o f the specific price formats adopted by the carriers. Research on retail price format also yields similar insights. Lai and Rao’s (1997) numerical illustration shows that the deal offered by the HILO store increases in the willingness to pay by the consumers. Further, not only would the HILO store increase its deal, but the EDLP store also increases its overall discount and at a faster rate compared to the price cut offered by their HILO counterpart1 6 . This result has two significant implications. First, consistent with the prediction drawn from Varian’s model, this implies that as the departure date approaches, prices of tickets would become more dispersed than before. Second, not only are price dispersions observed from the EDLP firms expected to be lower compared to those observed from the HILO firms, but this gap also increases as the departure date approaches. For example, the difference in dispersions between EDLP prices and HILO prices are more salient for one-week advance purchase tickets than that for four-week advance purchase tickets. Hypothesis 8 (H8): Price dispersion increases with consumer reservation for both price formats. 1 6 As consumer’s willingness to pay increases between the range o f 0 and 1. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 106 Empirical research in online price dispersion concludes that prices remain dispersed over time in various types of electronic markets (Baye, et al. 2002, Baye, et al. 2004, Pan, et al. 2003b, Rachford, et al. 2003). Further, studies that compare the relative magnitudes of price dispersions across online and offline channels find that the dispersion is either more pronounced online compared to offline (Bailey 1998, Brynjolfsson and Smith 2000) or is not significantly different across the two channels (Clay, et al. 2002, Lee and Gosain 2002). Even though price dispersion - by virtue of vendors’ constant attempt to price discriminate and their conscious adoption of different pricing strategies - may not completely vanish in electronic markets, lowering of search costs and increased price transparency in the online environment should result in lower price dispersion compared to that in physical markets, ceteris paribus. Controlling for the identical types of markets and products, prices of the same sellers that operate in both channels are expected to be less dispersed online than offline. Hypothesis 9 (H9): While there is indeed price dispersion online, this dispersion is significantly smaller compared to offline price dispersion. ^ M O D E L S This section presents the models employed in testing the hypotheses discussed earlier in this chapter. Although some of the hypotheses (hypotheses 8 and 9) do R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 107 not require separate models to be analyzed, for notational convenience and ease of reference, the models are named corresponding to the hypotheses being tested. The analyses performed in this study employ the full set of data, though there are small variations depending on the models’ objectives. For models 6a and 6b, prices from all markets and tickets from both EDLP and HILO sellers are included. In model 7, since the focus is on the reactions by HILO sellers in the presence of EDLP competitors in the market, tickets offered by EDLP sellers are excluded. In models that estimate offline price dispersion, sales data from both electronic and physical channels is used. The source of the ticket and price data is the Airline Origin and Destination Survey Databank IB (DB1B) obtained from the Bureau of Transportation Statistics for the fourth quarter of 2004. DB1B is a 10% sample of airline tickets from reporting carriers. Data includes origin, destination and other itinerary details of passengers transported. 5.2.1 Test of Hypothesis 6: Price Format and Price Dispersion To show that sellers’ conscious adoption of different price formats contributes as a source of price dispersion, two conditions need to be satisfied: First, the extent of price dispersion differs between EDLP and HILO sellers. Second, at the market level, the degrees of dispersion in prices differ between markets in which both types of sellers coexist and those in which only HILO sellers compete. Model 6a offers a test of the former condition, while model 6b examines the latter. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 108 Model 6a: Dispk m = a + P,DD7 + p 2DDI4 + p 3 DD21 + P4 freqh n + P5hubh n +P6shorthaulm + P7 slotm + p 8mktconm + P9 ( EDLPk x DD7) (27) +Pl0 ( EDLPk x DD14) + p u (EDLPk x DD21) + sik m where c‘ = r«+«M + u< l, ( 28) - N(0,<p) (29) (30) Consistent with existing literature (Borenstein and Rose 1994), route effects are taken as random while airline effects are fixed (represented by uQ k ). The dependent variable is separated into two measures of price dispersion, range1 7 and coefficient of variation, at the carrier-route level (km). Separate models are tested with respect to each of the two measures. M odel 6b: Dispm = a + PIDD7 + P2DD14 + P3DD21 + P4 shorthaulm + p 5 sloti where (31) +P6mktconm + P7EDLPmktm + s m 1 7 Range at the carrier-route level is adjusted as a ratio to the range o f prices in the market. In other words, the maximum value will be 1 for an airline that offers prices that covers the maximum and the minimum prices available in the market. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 109 Similar to model 6a, the dependent variable is separated into two measures of price dispersion, range and coefficient of variation. However, the measures in this model are at the route level (m) since the focus of this model is in examining how dispersions differ between markets in which both EDLP and HILO sellers coexist and those where only HILO sellers are present. For this purpose, an additional independent variable ( EDLPmkt )is introduced to identify the two types of markets. Again, separate models are tested with respect to each of the two measures of dispersion. Notice that since the level of analysis is at the route level (i.e. one observation per route), route and airline fixed-effects are excluded from this model. To demonstrate that pricing strategy is a source of price dispersion in the offline context, models 6a and 6b are estimated again using purchase data that consists of prices from the offline channels. However, slight adjustment needs to be made to 1 o the models . First, since there is no information on when the tickets were purchased (in terms of days prior to departure), all time dummies (D D 1,D D \4,D D 2\) and their interactions with EDLP are excluded from the modified models. Second, analyses on the DB1B data are aggregated for both business and leisure travel since there is no identifying information available from the data. 1 8 For ease o f exposition, mathematical representations o f the modified models are omitted. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 110 5.2.2 Test of Hypothesis 7: HILO Reactions Model 7: Dispk m = a + fi,DD7 + p 2DD14 + p 3 DD21 + P4freqk m + p 5 hubk m +P6shorthaulm + P7 slotm + P8mktconm + p gEDLPmktm + eik m (33) where a = n + u ok+u0 m (34) w om~N (0 ,< p) (35) slkm~ N { 0 ^ ) (36) and u0 k denotes the airline fixed-effects. The dependent variable is defined in the same way as that in model 6a. This model is structurally very similar to model 6a, except that a dummy variable EDLPmkt that identifies the types of competitors in a market is included in place of the EDLP interactions. The reason is that the objective of this model is to study whether there are any differences between pricing reactions of the HILO firms in markets where EDLP sellers are present and those where only HILO competes. As discussed in the beginning of this section, only tickets offered by HILO sellers are included in the data for estimation. Therefore, no EDLP specific effects are being analyzed. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. Ill 5.3 RESULTS 5.3.1 Results of Hypothesis 6: Price Format as a Source of Price Dispersion BUSINESS LEISURE Range CY Range CV Intercept 0.8480*** 0.0278 0.9406*** 0.1306*** DD7 0.0442*** 0.0779*** -0.0081 0.07376***' DD14 0.0629*** 0.05054*** -0.0112 0.06319*** DD21 0.0138 0.02812*** -0.0137 0.0007 freq 0.0017*** 0.000222*** 0.0022*** 0.000364*** hub 0.0287* 0.02326*** 0.0679*** 0.01392** shorthaul -0.0295 -0.0060 0.0040 -0.0039 slot -0.1462*** 0.02566*** -0.1079*** 0.03407*** mktcon -0.0219*** 0.0019 -0.02524*** -0.0020 DD7*EDLP 0.1277** -0.03692** 0.0655 -0.04301** DD14*EDLP 0.0775 -0.0316* 0.0455 -0.04973*** DD21*EDLP 0.1661*** -0.0019 0.1041** 0.0265 EDLP1 -0.3490*** -0.0587*** -0.3457*** -0 112*** EDLP2 -0.1415** -0.0158 -0.2113*** -0.0743*** V 0.0557*** 0.0068*** 0.0395*** 0.0074*** a 1 0.0997*** 0.0114*** 0.0823*** 0.01372*** N 6491 6491 6526 6526 -2LL 4355.1 -9265.2 3168.9 -8226.3 BIC 4367.4 -9252.9 3181.2 -8214 Table 12: Results of Model 6a - Price Dispersion and Price Format The results of model 6a are summarized in table 12. Dispersions in prices of the two EDLP sellers are significantly lower than those of the HILO sellers, in both R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 112 business and leisure tickets. The finding is robust for both dispersion measures, range and coefficient of variation. However, notice the slight difference between the two EDLP sellers, EDLP1 and EDLP2. For EDLP2, the range of prices for three-week advance purchase business tickets is indeed wider than that of HILO sellers. BUSINESS LEISURE Range CV Range CV Intercept 13.46 0.0473*** 55.29* 0.1068*** DD7 130.91*** 0.0634*** 155.62*** 0.0705*** DD14 45.54*** 0.0357*** 110.10*** 0.0670*** DD21 28.28* 0.0227** 2.55 0.0068 shorthaul -30.76*** -0.0074 -51 37*** -0.0077 slot 72.23*** 0.0559*** 90.09*** 0.0656*** mktcon 11.50*** 0.0057*** 12.38*** 0.0038* EDLPmkt -82.29*** -0.0294*** .99 19*** -0.0475*** N 1844 1844 1844 1844 -2LL 25170.2 -1792.5 25564 -1826.5 BIC 25177.7 -1785 25571.5 -1819 Table 13: Results of Model 6b - Price Dispersion Across Markets The results of model 6a are summarized in table 13. Price dispersion in markets where both EDLP and HILO sellers compete are significantly lower than that in markets where only HILO sellers are present. This finding is robust to different dispersion measures and for both business and leisure tickets. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. From these results, we can conclude that price format is a significant source of online price dispersion. To show the second part of the hypothesis on offline price dispersion, I perform similar analyses on a different set of data that consists of purchase price data from both online and offline channels. Range CV Intercept 0.4587*** 0.3860*** freq 0.0012*** 0.0000 hub 0.4118*** 0.1034*** shorthaul 0.0716*** -0.0146 slot 0.0726*** 0.0314** mktcon -0.0109** -0.0006 EDLP1 0.0971** -0.1712*** EDLP2 0.0756 -0.1039*** < P 0.0042** 0.0022*** < 7 2 0.0823*** 0.0229*** N 1387 1387 -2LL 626.5 -970.6 BIC 638.9 -958.2 Table 14: Price Dispersion and Price Format - Online and Offline Channels Tables 14 and 15 present results from DB1B data. When both online and offline channels are considered, we observe that differences in price dispersions between EDLP and HILO sellers are still observable. However, the results are slightly different compared to the previous analysis with online pricing data. While the coefficients of variation remain lower for EDLP sellers, the range of prices offered R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. by EDLP sellers is either not significantly different exceeds that by HILO sellers. One potential explanation could be that for EDLP1, majority of business tickets sales are made one or three weeks prior to departure, which contribute to increase in price dispersion as measured by price range according to the results reported in table 11. Unfortunately, without information on Saturday-night stay restriction and the actual purchase and departure dates of the tickets, this cannot be verified. Range CV Intercept 776.42*** 0.4726*** shorthaul -457.63*** -0.0409*** slot 53.59 0.0151 mktcon 67.80*** 0.0077** EDLPmkt -549.43*** -0.1926*** N 497 497 -2LL 7544.2 -633.6 BIC 7550.4 -627.4 Table 15: Price Dispersion Across Markets - Online and Offline Channels In terms of overall price dispersion in the market, the analysis of data from both online and offline channels suggest that dispersion in markets where both EDLP and HILO sellers coexist is lower, by both measures of dispersion. This result is consistent with that found in online pricing data presented in table 13. In sum, results from these four analyses show that price format is a significant source of price dispersion in both online and offline contexts. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 115 5.3.2 Results of Hypothesis 7: HILO Reactions BUSINESS LEISURE Range CV Range CV Intercept 0.7004*** 0.0300 0.7795*** 0.1304*** DD7 0.0441*** 0.0779*** -0.0079 0.0738*** DD14 0.0630*** 0.0506*** -0.0110 0.0632*** DD21 0.0136 0.0281*** -0.0137 0.0008 freq 0.0036*** 0.0003** 0.00335*** 0.0004*** hub 0.0070 0.0239*** 0.0528*** 0.0149** shorthaul -0.0513 -0.0066 -0.0282 -0.0087 slot -0.1422*** 0.0250** -0.0801*** 0.0363*** mktcon -0.0091 0.0012 -0.0117* -0.0027 EDLPmkt -0.1531*** 0.0145 -0.0261 0.0258 < P 0.0482*** 0.0078*** 0.0306*** 0.0085*** cr2 0.1000*** 0.0119*** 0.0824*** 0.0141*** N 6091 6091 6129 6129 -2LL 4128.1 -8514.5 2905.7 -7594.2 BIC 4140.1 -8502.5 2917.7 -7582.2 Table 16: Results of Model 7 - Reactions by HILO Sellers Results of model 7 are summarized in table 16. Only limited evidence is found on the speculation that pricing of HILO sellers react to the presence of EDLP sellers. Except for price range in business tickets, no significant difference is observed among the dispersions in HILO prices in markets where EDLP sellers participate and those they do not. Therefore, hypothesis 7 is not supported. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 116 5.3.3 Results of Hypothesis 8: Price Dispersion and Consumer’s Reservation Results from model 6a and 6b (tables 12 and 13) allows us to draw inferences on the hypothesized relationship between online price dispersion and consumer’s reservation. According to results in table 12, price dispersion is found to be higher as the departure date approaches at the carrier-route level. However, this result is observed in business tickets only. While the dispersion measured by coefficient of variation increases with consumer reservation for both business and leisure tickets, there is no evidence that the range of prices varies across the four weeks of advance purchase. Further, the relative dispersions of EDLP prices and HILO prices do not vary systematically with respect to increasing reservation; there is no support for the speculation that EDLP prices become less and less dispersed compared to HILO prices as departure date approaches. On the other hand, there is strong evidence that price dispersion increases with consumer’s reservation at the route-level (table 13). Both measures of price dispersion, range and coefficient of variation, increase as the departure date approaches. This observation is robust with respect to both business and leisure tickets. The results validate the theoretical predictions of Varian’s (1980) model of sales and the interpretation by Baye (2003) that as consumer reservation increases, so does dispersion of prices in the market. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 5.3.4 Results of Hypothesis 9: Online vs. Offline Price Dispersion To test this hypothesis, analyses were performed separately on each of the price dispersion measures. First, t-test is performed on the range o f prices for each carrier-route observation between the two data sets. Second, F-test is performed on the ratio of variances. Variance is used in place of coefficient of variation as the second measure of price dispersion because of its desirable distributional properties. Variance for large sample has a Chi-square distribution, and the ratio of variances has an F-distribution, allowing for a straight forward statistical test of the equality of variances drawn from the two samples. The results are summarized in table 17. Range ( H 0 :Equal Ranges) Variance Ratio (H 0 : Equal Variances) 5% significance 1% significance Route Level Rejected (p<0.001) Rejected (Variances are different 79.49% of the time) Rejected (Variances are different 81.20% of the time) Carrier-Route Level Rejected (p<0.001) Rejected (Variances are different 82.20% of the time) Rejected (Variances are different 84.20% of the time) Table 17: Price Dispersion Online vs. Offline Range of purchase prices in both online and offline channels are significantly wider compared to posted prices on the Internet; the variance of the former is also higher than that of the latter majority of the time (averaging over 80%). The results are robust at both route and carrier-route level measures of dispersion, offering strong support for hypothesis 9. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 118 5.4 DISCUSSION The first four analyses presented in this chapter yield two important observations on the relationship between price format and online price dispersion. First, dispersions in prices of sellers who adopt the EDLP strategy are lower compared to those who adopt the HILO strategies. Second, price dispersions in markets where EDLP sellers are present are lower compared to those where only HILO sellers compete. Both of these results are found to be robust for both online and offline contexts and thus offer strong support to the hypothesis that price format is a source of price dispersion in both online and offline markets. While the majority of research in online price dispersion focuses on attributing the differences among online vendors’ pricing behaviors to their reactions to consumer heterogeneity and competitive forces in the market, the analysis on dispersion in HILO sellers’ prices presented in this study shows that dispersions in HILO prices do not vary across markets where EDLP competitors are present and those where they are not. This finding suggests that conscious adoption of price formats is a choice that sellers proactively make, rather than a reactive response. The results offer support for the theoretical argument of random pricing theory in explaining price dispersion, which largely lack empirical support in the existing literature. Results from the analysis on the relationship between consumer’s reservation value and price dispersion indicate that price dispersion is higher among products for R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. which consumers have higher willingness to pay. This finding generalizes the theoretical prediction by Baye (2004) to the context of perishable products. Further, the analysis demonstrates that potentially different conclusions can be arrived depending on the unit of analysis. While high price dispersion within each firm-market unit is likely to lead to high overall dispersion in the market, the reverse is not necessarily true. The implications on price dispersion research are that, if the objective of the study is to verify whether dispersion exists or the extent to which it exists, then market-level dispersion offers a stronger test as opposed to firm-market level. On the other hand, if the objective is to examine how variances in firm’s characteristics affect price dispersion, then firm-market level would be the appropriate level of analysis. In the comparison of the degrees of price dispersion in online versus offline contexts, both the range and variance of prices are found to be significantly lower in electronic markets. This study is one of the first to show that while price dispersion still exists online, the dispersion is indeed smaller compared to that in the offline context. The results offer support to the theoretical prediction that due to lowering costs of search and increased price transparency, prices in electronic markets are less dispersed compared to those in physical markets. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 120 CHAPTER 6 SUMMARY AND CONCLUSIONS 6.1 SUMMARY OF RESULTS 6.1.1 Pricing Strategies of Online EDLP Vendors This study shows that the extent to which online sellers adopt the “everyday low price” strategy varies with different types of market and product categories. In particular, while there is no evidence that the average prices offered by online EDLP sellers are generally lower than others, results indicate that these sellers compete aggressively in the lower bounds of their prices on all types of products, and even more so in certain markets. This finding is consistent with the theoretical prediction extended from Lai and Rao’s (1997) analysis on EDLP and HILO competition in that as the distance between the two types of firms decreases, due to either decrease in physical distance or lowering of search costs for the consumers, EDLP sellers will lower their expected price level to keep prices attractive to time-constrained consumers who still suffer higher search costs relative to the cherry-picking consumers. Online EDLP sellers’ pricing are most consistent with the “everyday low price” image in markets where they have cost advantage. On the other hand, prices of these sellers in markets where they enjoy high market power exhibit more R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. promotional characteristics than those of their HILO competitors in terms of relative price level, and are not significantly different compared to HILO prices in terms of temporal price variability. In reconciling the discrepancy in the “everyday low price” image, online EDLP sellers emphasize the stability of their low prices across different product markets, analogous to that of a “99cents” store. The most notable difference between EDLP pricing online and that observed in physical markets is that EDLP prices are not in general lower than HILO prices. Online EDLP sellers are found to set lower prices only in a restricted set of markets and specific product categories. Further, online EDLP sellers focus more on the “within-market” characteristics of this pricing strategy rather than the “across-time” characteristics, implying a diminishing role of intertemporal price consistency to the practice of everyday low price in the online environment. This study offers the first evidence of category-level adoption of price format speculated by prior literature (Bell and Lattin 1998, Ho, et al. 1998), while being the first to investigate pricing behaviors of vendors adopting different price formats along various dimensions in electronic markets. Results from the analysis on relative price range and minimum price comparison offer a potential explanation for why research in online price dispersion finds contradictory evidence on random pricing theory (Baye, et al. 2004, Baylis and Perloff 2002). Online EDLP sellers price their products within a narrower range and closer to the lower bound of market prices, while HILO sellers vary their prices R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 122 significantly. As a result, in markets where both types of sellers coexist, the price-rank of sellers appears stable (Baylis and Perloff) while in other markets where only HILO sellers are present, substantial evidence o f “hit and run” strategy and randomness in price-rank are observed (Baye, et al. 2004). 6.1.2 Price Format and Price Dispersion Analysis on the relationship between price format and online price dispersion yields strong evidence that vendors’ adoption of different price formats contributes to dispersion in both online and offline prices. Further, results suggest that dispersions in HILO prices do not vary across markets where EDLP competitors are present and those where they are not. In contrast to the majority of research in online price dispersion that attributes dispersion to vendor’s reaction to consumer heterogeneity, this finding suggests that conscious adoption of particular price formats is a choice that vendors proactively make, rather than a reactive response. The results offer support for the theoretical argument of random pricing theory in explaining price dispersion, which largely lack empirical support from the existing literature. Consistent with the theoretical predictions derived from the model of sales (Baye, et al. 2004, Varian 1980), the study on the relationship between consumer’s reservation value and price dispersion reveals that the degree of price dispersion is more severe among products for which consumers have higher willingness to pay. Finally, the extent to which prices are dispersed is found to be significantly lower online R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 123 compared to that in the offline context. In contrast with the empirical studies that find online price dispersion is either higher or not significantly different from offline (Bailey 1998, Brynjolfsson and Smith 2000, Clay, et al. 2002, Lee and Gosain 2002), this study is one of the first to show that though price dispersion still exists online, it is indeed smaller compared to that in physical markets. 6.2 MANAGERIAL AND POLICY IMPLICATIONS 6.2.1 Applications of EDLP in the Airline Industry In the supermarket context, the size of a basket is influenced by two factors: the number of categories and the quantity of each category. Since Bell and Lattin (1998) control for a fixed set of categories in their analysis, it is obvious that large basket size in their sample is due to the large quantity purchased within each category by large basket shoppers. However, it appears counterintuitive that these shoppers exhibit lower price sensitivity; because even if the dollar saving of comparison shopping is small for each item in the basket, the aggregate savings can be substantial when large quantities are bought. One potential explanation is that EDLP stores offer volume discounts that are effective only when the customer purchase a certain number of the same products, which is reflected in their lower average basket price - even if the individual prices of each item may not necessarily be lower compared to those set by their HILO counterparts. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 124 In the context of airline industry where purchases are not characterized by quantities, firms that adopt EDLP may offer discounts based on frequency of travels and mileage of trips. While this is true for all Frequent Flyer Programs (FFPs), by applying the logic derived from Bell and Lattin’s observation, we should expect to see that EDLP carriers offer higher rewards to frequent flyers compared to others. Observations from the airline industry indeed offer support to this view. Southwest Airlines, a well known everyday low price seller in the industry, offers Rapid Rewards that are considerably easier and quicker to redeem than many of their competitors’ reward programs. In fact, Rapid Rewards is “the only frequent flyer program that has ever received the prestigious Freddie Award for ‘Best Award Redemption’ as voted by frequent flyers nationwide” 19. This implies that “everyday low price” alone does not by virtue attract consumers that exhibit certain characteristics; but rather, for this strategy to be effective, it needs to be implemented with peripheral services and discounts that reinforce the low price image. 6.2.2 Cost Structure Does Not Determine Pricing Strategy Hoch et al. (1994) note that “ ...Price is not a defensible point of differentiation for a firm unless it already has the appropriate operating cost structure in place.” (p.26). Specifically, using the airline industry as an example, they assert that major airlines such as American “abandoned the idea o f imitating the low-cost, low-service strategy that has been so successful for Southwest Airlines.” However, it is widely observed 1 9 http://www.southwest.com/rapid rewards/free award ticket.html R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 125 that major carriers do imitate low-cost airlines both in terms of cost management and pricing. Further, Lai and Rao (1997) show that the existence of both EDLP and HILO emerges as the market equilibrium despite the absence of cost asymmetries. This theoretical finding is supported by the empirical results in this study that show cost fails to be a significant factor in explaining the difference in prices set by EDLP versus HILO firms. An important implication of this finding is that rather than the cost-based view on the rationale for setting everyday low price (Hoch, et al. 1994, Ortmeyer, et al. 1991), EDLP is a conscious pricing strategy that does not depend solely on cost savings. In other words, although a low cost structure may be necessarily for a seller to maintain low and stable prices, cost itself is not a driver for the adoption of the everyday low price strategy. 6.2.3 Strategic Variability in EDLP Pricing Online This study shows that EDLP prices are not universally lower than prices offered by HILO sellers. However, EDLP sellers are found to be aggressive in undercutting prices offered by their HILO competitors. These results suggest that while it may be costly for EDLP sellers to compete on the overall price level in electronic markets due to high price transparency, online EDLP sellers may achieve lower expected price level by setting lowest prices that match or undercut the minimum prices offered by their HILO counterparts. Further, the analyses reveal significant discrepancies between the theoretically ideal “everyday low price” practice and the actual pricing behaviors of online EDLP sellers along certain dimensions, most R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 126 notably in the temporal price variability and the application of low price, low variability in only a restricted set o f markets and products. These results suggest that the adoption of a “hybrid” strategy (EDLP in some markets, HILO in others) may be valuable particularly in online competition; because a constant and stable price level can be easily challenged by competitors who can immediately observe and react to these prices in electronic markets. 6.2.4 Policy Implications “ Airline deregulation works because low cost-low fare air carriers make it work. ” 90 Herb Kelleher, co-founder of Southwest Airlines The Airline Deregulation Act (ADA) of 1978 has benefited airline passengers by removing government control from commercial aviation and exposing the airline industry to market forces. The emergence of low-cost carriers into markets traditionally dominated by major carriers has lowered the ability of airlines to exercise monopolistic power and led to lower air fares. Certain low-cost carriers such as the Southwest Airlines have successfully gained market shares and their adoption of the “everyday low price” strategy has proven to be profitable. While it is observed that these EDLP carriers set low prices in markets where they have cost advantage and demonstrate the ability in undercutting prices of competitors because of their low cost structure, results from this study indicate that these low-cost carriers 2 0 Testimony before the National Civil Aviation Review Commission. Available online: http://www.faa.gov/NCARC/testimonv/swa-te.htm R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 127 may no longer be setting competitive prices once they have obtained sufficient market power. The results from this study suggest that it may not be socially optimal for a government regulator such as the Federal Aviation Administration (FAA) to completely withdraw their influences in the airline markets. While in deregulated markets airlines may be driven to price competitively to stave off competitors, and consumers would benefit from increased competition in the short run, once obtained dominant shares in the market, even the self-declared low-price carriers may begin to engage in monopolistic pricing. The regulatory body should therefore continue to promote competition in airline markets, such as by offering subsidies and tax benefits to help traditional carriers in their transition to adopting more effective cost structure and management, protecting them from being “priced-out” by the low-cost entrants who can potentially in the long run become monopolists in airline markets. 6.3 CONTRIBUTIONS This study contributes to existing literature in the following ways. First, this study offers the first evidence of category-level adoption of price format and generalizes research in EDLP to electronic markets. Second, it uncovers new dimensions in which the “everyday low price” strategy is adopted by online sellers, contrasts the characteristics o f EDLP pricing between online and offline markets, and offers explanation to these differences based on theories from multiple disciplines. Third, by identifying conscious price format adoption by sellers as a source that contributes R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 128 to dispersion of prices, this research furthers the understanding of the nature of online price dispersion. Fourth, results from the analysis on relative range and minimum price comparison offers an explanation that reconciles the discrepancies in findings by existing empirical research in regards to the random pricing theory. Fifth, based on most recently available data, this study is one of the first that demonstrates price dispersion online to be lower than that in offline markets. Further, this study proposes a new approach - hierarchical modeling - in analyzing online price dispersion. Hierarchical modeling can be a particularly useful tool in handling data that exhibit a hierarchical, or nested, structure. Such an example can be found in online books and CDs retailing, where each online seller offer multiple, or even tens of thousands of, book and CD titles. While the unit of analysis in dispersion studies are typically at the product level so that potential effects of heterogeneity in products on price dispersion can be controlled for, when analyzing the market as a whole and drawing inferences from estimations across products and vendors, variations due to differences in the second-level (firm) unit need to be accounted for. Existing research commonly adopt the least square dummy variable (LSDV) approach to control for such differences between group variances among retailers. In many cases, the dependence among observations from the same firm is being ignored. Although ordinary least square method still produces consistent estimates R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 129 for the coefficients, the standard errors will be underestimated, leading to increased likelihood of type-I error. Traditionally, generalized least square can be used to allow for consistent estimation of heteroscedastic variance. However, LSDV imposes stringent constraints on the researcher’s ability to make meaningful inferences from heterogeneity in vendors’ characteristics, since all observable and unobservable differences among vendors are being absorbed in the fixed-effects dummies. Hierarchical models not only allow the control for within-group dependence among individual observations from the same firm as well as between group heteroscedastic variances across different firms, but they also allow for the inclusion of additional explanatory variables that capture certain firm-level characteristics of interests to the researcher. This enables researchers to draw inferences on, for example, the effects of differences in vendor’s costs and pricing strategy on price dispersion separately. Moreover, the effects of these different characteristics are allowed to vary across different product markets or different categories of the same type of products, without the need for interaction dummy for each market-vendor characteristic pair that result in enormous number of parameters to be estimated and overidentification of the model. This research has demonstrated one application of hierarchical modeling in analyzing online price dispersion, and has offered insights on the usefulness of this technique in price dispersion research, particularly in efficiently addressing the effects of heterogeneity in different vendor characteristics on dispersion and how such effects may vary across different types of products and markets. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 130 6.4 LIMITATIONS AND FUTURE RESEARCH As with any empirical studies that employ data from a single industry, the results obtained in this research are limited in their generalizability to other industries in electronic markets. It should also be noted that this is a descriptive study of the practice of EDLP online, rather than a normative study of how pricing decisions should be made by sellers adopting this strategy. The focus of this study is to investigate whether, and to what extent, the “everyday low price” strategy is being applied in the online context. By demonstrating that certain characteristics of this strategy are still evident from the pricing behaviors of self-declared EDLP sellers in markets where vendors predominantly employ randomized pricing, this research opens up avenues for future research on the applicability of EDLP strategy in other types of electronic markets. The lack of evidence on prices offered by EDLP sellers exhibiting lower temporal variability compared to those offered by their HILO competitors can be partly attributed to the perishable nature of the product considered in this study; as consumer reservation value increases with approaching departure date, so does the price discriminatory power of the firms (Baye, et al. 2004). However, an alternative explanation is that reduction in search costs in electronic markets has differential effects on consumer’s price sensitivity along two dimensions: by making price information more accessible, consumers may focus on the comparison of “spot-prices” when they perform the search. Due to the large number of prices R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. available at the time of search, recall may be poor. As a result, electronic markets may increase consumer’s price sensitivity at the time when price comparison is performed, but have a negative effect on consumer’s intertemporal price sensitivity, hence online sellers may find randomize pricing across time profitable. Future research is therefore urged to consider the temporal dimension in analyzing the effects of electronic markets on consumer price sensitivity. While reduction in search costs may unambiguously increase consumers’ sensitivity towards prices at one given point in time, consumers may be sensitive to who the low-price seller is rather than how low the prices are compared to the past. Findings on lack of temporal price stability but consistent price-rank of sellers adopting the EDLP strategy in this study offers initial support to this argument. In comparing online and offline price dispersion, the data employed in this study consists of offered prices online and purchased prices in both online and offline channels. A matching comparison of offered prices from online versus those from purely offline channel was not possible due to limitation in available data. However, this actually strengthens the power of the test rather than weakens it. The intuition is that, since prices at which tickets were purchase are necessarily a subset of the prices being offered, dispersion in prices of tickets purchased from the online channel will be lower compared to that in prices posted on the Internet. By symmetry, purchased prices in offline channels are also less dispersed than offered prices in the same channels. Therefore, the results that show higher dispersion in R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 132 purchased prices in the combined channels compared to only posted prices online imply that dispersion o f posted prices in offline channels are indeed much higher that those in electronic markets. The analysis on price dispersion demonstrates that potentially different conclusions can be arrived at depending on the unit of analysis. While high price dispersion within each firm-market unit is likely to lead to high overall dispersion in the market, the reverse is not necessarily true. An important implication on future research in price dispersion is that, if the objective of the study is to verify whether dispersion exists or the extent to which it exists, then analysis at the market level dispersion offers a stronger validation as opposed to that at the firm-market level. On the other hand, if the objective is to examine how differences in firm’s characteristics affect price dispersion, then firm-market level would be the appropriate level of analysis. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 133 REFERENCES Ainslie, Andrew and Peter E. Rossi, "Similarities in Choice Behavior across Multiple Categories," Marketing Science, 17, 2 (1998), 91-106. Aitkin, M. A. and N. T. Longford, "Statistical Modeling Issues in School Effectiveness Studies," Journal of the Royal Statistical Society, Series A, 149 (1986), 1-43. Alba, Joseph, John Lynch, Barton Weitz, Chris Janiszewski, Richard Lutz, Alan Sawyer and Stacy Wood, "Interactive Home Shopping: Consumer, Retailer, and Manufacturer Incentives to Participate in Electronic Marketplaces," Journal of Marketing, 61, (1997), 38-53. Bailey, J. P., "Intermediation and Electronic Markets: Aggregation and Pricing in Internet Commerce," Cambridge, MA, (1998). Bakos, J. Yannis, "Reducing Buyer Search Costs: Implications for Electronic Marketplaces," Management Science, 43, 12 (1997), 1676-1692. Bakos, J. Yannis, "The Emerging Role of Electronic Marketplaces on the Internet," Communications of the ACM, 41, 8 (1998), 35-42. Barua, A., S. Ravindran and A. B. Whinston, "Efficient Selection of Suppliers over the Internet," Journal of Management Information Systems, 13, 4 (1997), 117-137. Baye, Michael R. and John Morgan, "Information Gatekeepers on the Internet and the Competitiveness of Homogeneous Product Markets," American Economic Review, 91,3 (2001), 454-474. Baye, Michael R. and John Morgan, "Information Gatekeepers and Price Discrimination on the Internet," Economics Letters, 76, 1 (2002), 47-51. Baye, Michael R. and John Morgan, "Price Dispersion in the Lab and on the Internet: Theory and Evidence," RAND Journal of Economics, 35, 3 (2004), 449-466. Baye, Michael R., John Morgan and Patrick Scholten, "The Value of Information in an Online Consumer Electronics Market," (2002). Baye, Michael R., John Morgan and Patrick Scholten, "Persistent Price Dispersion in Online Markets," In e. Dennis W. Jansen (Ed.), forthcoming in "The New Economy", University of Chicago Press, Chicago, 2003. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 134 Baye, Michael R., John Morgan and Patrick Scholten, "Temporal Price Dispersion: Evidence from an Online Consumer Electronics Market," Journal o f Interactive Marketing, 18, 4 (2004), 101-115. Baylis, Kathy and Jeffrey M. Perloff, "Price Dispersion on the Internet: Good Firms and Bad Firms," Review of Industrial Organization, 21, 3 (2002), 305-324. Bell, David R. and James M. Lattin, "Shopping Behavior and Consumer Preference for Store Price Format: Why "Large Basket" Shoppers Prefer EDLP," Marketing Science, 17, 1 (1998), 66-88. Berry, S., "Airport Presence as Product Differentiation," The American Economic Review, 80, 2 (1990), 394-399. Berry, Steven, Michael Camall and Pablo T. Spiller, "Airline Hubs: Costs, Markups and the Implications of Customer Heterogeneity," NBER Working Paper #5561, (1997). Bertrand, J., "Theorie Mathematique de la Richesse Sociale," Journal des Savants, 67, (1883), 499-508. Blattberg, R., G. Eppen and J. Liberman, "A Theoretical and Empirical Evaluation of Price Deals for Consumer Nondurables," Journal of Marketing, 45, (1981), 116-129. Borenstein, S., "Hubs and High Fares: Dominance and Market Power in the U.S. Airline Industry," The RAND Journal of Economics, 20, 3 (1989), 344-365. Borenstein, S. and N. L. Rose, "Competition and Price Dispersion in the U.S. Airline Industry," The Journal of Political Economy, 102, 4 (1994), 653-683. Brown, Jeffery R. and Austan Goolsbee, "Does the Internet Make Markets More Competitive? Evidence from the Life Insurance Industry," Journal of Political Economy, 110, 3 (2002), 481-507. Bryk, Anthony S. and Y. M. Thum, "The Effects of High School Organization on Dropping Out: An Exploratory Investigation," American Educational Research Journal, 26, (1989), 353-383. Brynjolfsson, E. and M. Smith, "Frictionless Commerce? A Comparison of Internet and Conventional Retailers," Management Science, 46, 4 (2000), 563-585. Carlson, J. and D. Pescatrice, "Persistent Price Distributions," Journal of Economics and Business, 33, (1980), 21-27. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 135 Chellappa, Ramnath K. and Ravi Kumar, "Examining the Role of "Free" Product — Augmenting Online Services in Pricing and Customer Retention Strategies," Journal of Management Information Systems, 22, 1 (2005), 355-377. Chen, P. and L. M. Hitt, "Brand Awareness and Price Dispersion in Electronic Markets," 22nd Annual International Conference on Information Systems, (2001). Chen, Pei-yu and Lorin M. Hitt, "Limited Awareness, Brand Sensitivity and Price Dispersion," Working Paper, (2003). Chircu, Alina M. and Robert J. Kauffman, "Impacts of Electronic Commerce on Optimal Market Structure," In L. Soete (Ed.), forthcoming in The Economics of Electronic Intermediation (tentative title), Springer-Yerlag, New York, NY, 2001. Clay, Karen, Ramayya Krishnan, Eric Wolff and Danny Fernandes, "Retail Strategies on the Web: Price and Non-price Competition in the Online Book Industry," Journal of Industrial Economics, 50, 3 (2002), 351-367. Clemons, E. K., I. Hann and L. M. Hitt, "Price Dispersion and Differentiation in Online Travel: An Empirical Investigation," Management Science, 48, 4 (2002), 534-549. Dahlby, B. and D. West, "Price Dispersion in an Automobile Insurance Market," Journal of Political Economy, 94, 2 (1986), 418-438. Dana, James D., "Advance-Purchase Discounts and Price Discrimination in Competitive Markets," The Journal of Political Economy, 106, 2 (1998), 395-422. De Leeuw, J. and I. G. G. Kreft, "Random Coefficient Models for Multilevel Analysis," Journal of Educational Statistics, 11, (1986), 57-85. Degeratu, Alexandru, Arvind Rangaswamy and Jianan Wu, "Consumer Choice Behavior in Online and Regular Stores: The Effects of Brand Name, Price, and Other Search Attributes," International Journal of Research in Marketing, 17, (2000), 55-78. Diamond, P., "Consumer Differences and Prices in a Search Model," The Quarterly Journal of Economics, 102, 2 (1987), 429-436. Feick, Lawrence F. and Linda L., "The Market Maven: a Diffuser of Marketplace Information," Journal of Marketing, 51,1 (1987), 83-98. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 136 Fournier, Gary M. and Thomas Zuehlke, "Price Effects of Reciprocal Multi-Market Contacts Among Airline Carriers," Working Paper, November, (2004). Gale, Ian L. and Thomas J. Holmes, "Advance-Purchase Discounts and Monopoly Allocation of Capacity," The American Economic Review, 83, 1 (1993), 135-146. Garry, Michael, "A Cultural Happening," Progressive Grocer, 75, 4 (1996), 38-42. Goldstein, H., "Multilevel Covariance Component Models," Biomietrika, 74, 430-431 (1987). Goldstein, H., Multilevel Statistical Models, 2nd ed., Halstead Press, New York, 1995. Goolsbee, A. and J. Chevalier, "Measuring Prices and Price Competition Online: Amazon and Barnes and Nobel," Technical Report, University of Chicago, 2002. Griffin, M. A. and D. A. Hofman, "Hierarchical Linear Models in Organizational Research: Cross-Level Interactions," Reseach Methods Forum, 2, (1997), 1-3. Grocer, Progressive, "61st Annual Report of the Grocery Industry," 73, 4 (1994), Halvorsen, Robert and Raymond Palmquist, "The Interpretation of Dummy Variables in Semilogarithmic Equations," American Economic Review, 70, 3 (1980), 474-475. Hann, Il-Hom, Jeff Roberts, Sandra Slaughter and Roy Fielding, "An Empirical Analysis of Economic Returns to Open Source Participation," Working Paper, (2004). Harrington, J. E., "Comment on "Reducing Buyer Search Costs: Implications for Electronic Marketplaces"," Management Science, 47, 12 (2001), 1727-1732. Hayes, Kathy J. and Leola B. Ross, "Is Airline Price Dispersion the Result of Careful Planning or Competitive Forces?," Review of Industrial Organization, 13, 5 (1998), 523-541. Hirschman, Elizabeth C. and Morris B. Holbrook, "Hedonic Consumption: Emerging Concepts, Methods and Propositions," Journal of Marketing, 46, 3 (1982), 92-101. Ho, T-H, C. S. Tang and D. R. Bell, "Rational Shopping Behavior and the Option Value of Variable Pricing," Management Science, 44, December (1998), 145-160. Hoch, Stephen J., Xavier Dreze and Mary E. Purk, "EDLP, Hi-Lo, and Margin Arithmetic," Journal of Marketing, 58, October (1994), 16-27. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 137 Hotelling, H., "Stability in Competition," Economic Journal, 39, (1929), 41-57. Hox, J. J., Multilevel Analysis: Techniques and Applications, Lawrence Erlbaum Associates, Mahwah, NJ, 2002. Hsaio, Cheng, Analysis of Panel Data (2nd ed.), Cambridge University Press, Cambridge, 2003. Jones, B. S. and M. R. Steenbergen, "Modeling Multilevel Data Structures," Mimeo, Annual Meeting of the Political Methodology Society, Columbus, OH, (1997), Kahneman, D. and A. Tversky, "Prospect Theory: An Analysis of Decision Under Risk," Econometrica, 47, 2 (1979), 263-291. Klemperer, P., "Entry Deterrence in Markets with Consumer Switching Costs," The Economic Journal, 97, (1987), 99-117. Kreft, I. G. G. and J. De Leeuw, Introducing Multilevel Modeling, Sage Publications, Thousand Oaks, CA, 1998. Lai, R. and R. C. Rao, "Supermarket Competition: The Case of Every Day Low Pricing," Marketing Science, 16, 1 (1997), 60-80. Lai, R. and M. Sarvary, "When and How is the Internet Likely to Decrease Price Competition?," Marketing Science, 18, 4 (1999), 485-503. Lee, Zoonky and Sanjay Gosain, "A Longitudinal Price Comparison for Music CDs in Electronics and Brick-and-Mortar Markets: Pricing Strategies in Emergent Electronic Commerce," Journal of Business Strategies, 19,1 (2002), 55-71. Longford, N. T., Random Coefficient Models, Clarendom, Oxford, 1993. Lynch, John G. and Dan Ariely, "Wine Online: Search Costs Affect Competition on Price, Quality, and Distribution," Marketing Science, 19,1 (2000), 83-103. McAfee, R. Preston and Vera te Velde, "Dynamic Price Discrimination: Theory and Airline Behavior," Working Paper, California Institute of Technology, (2004). McCloskey, Donald N., The Applied Theory of Price, Macmillan, New York, 1985. Mendro, R. L., W. J. Webster, K. L. Bembry and T. H. Orsak, "An Application of Hierarchical Linear Modeling in Determining School Effectiveness," Mimeo, Annual R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 138 Meeting of the American Educational Research Association, San Francisco, CA, (1995). Morton, Fiona Scott, Florian Zettelmeyer and Jorge Silva-Risso, "Internet Car Retailing," Journal of Industrial Economics, 49, 4 (2001), 501-519. Mulhem, Francis J. and Robert R Leone, "Retail Promotional Advertising: Do the Number of Deal Items and Size of Deal Discounts Affect Store Performance?," Journal of Business Research, 21, (1990), 179-194. Mussa, M. and S. Rosen, "Monopoly and Product Quality," Journal of Economics Theory, 18, (1978), 301-317. Narasimhan, C., "Competitive Promotional Strategies," Journal of Business, 61,4 (1988), 427-449. Ortmeyer, G., J. Quelch and W. Salmon, "Restoring Credibility to Retail Pricing," Sloan Management Review, Fall, (1991), 55-66. Pan, Xing, Brian T. Rachford and Venkatesh Shankar, "Can Price Dispersion in Online Markets be Explained by Differences in e-tailer Service Quality?," Journal of the Academy of Marketing Science, 30, 4 (2002), 443-456. Pan, Xing, Brian T. Rachford and Venkatesh Shankar, "Price Dispersion on the Internet: A Review and Directions for Future Research," Working Paper, (2003a). Pan, Xing, Venkatesh Shankar and Brian T. Rachford, "The Evolution of Price Dispersion in Internet Retail Markets," Advances in Applied Microeconomics, 12, forthcoming (2003b). Pedhazur, E. J., Multiple Regression in Behavioral Research: Explanation and Prediction (3rd ed.), Harcourt Brace College, Forth Worth, TX, 1997. Pemer, Lars Erik, "Optimal Stimulation Level As A Moderator of "High-low" And EDLP Pricing Effectiveness," Dissertation, (1998). Pratt, John W., David A. Wise and Richard Zeckhauser, "Price Differences in Almost Competitive Markets," The Quarterly Journal of Economics, 93, 2 (1979), 189-211. Rachford, Brian T., Xing Pan and Venkatesh Shankar, "On the Efficiency of Internet Markets for Consumer Goods," Journal of Public Policy and Marketing, 22, 1 (2003), 4-16. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 139 Radice, Carol, "Turning Up the Heat," Progressive Grocer, 65th Annual Report of the Grocery Industry, Apr (1998), 29. Raudenbush, Stephen W. and Anthony S. Bryk, "Quantitative Models for Estimating Teacher and School Effectiveness," In R. D. Bock (Ed.), Multilevel Analysis of Educational Data, Academic Press, San Diego, CA, 1989, 205-232. Raudenbush, Stephen W. and Anthony S. Bryk, Hierarchical Linearl Models: Applications and Data Analysis Methods, Sage, Newbury Park, CA, 2002. Raudenbush, Stephen W., Anthony S. Bryk, Y. F. Cheong and R. Congdon, HLM5: Hierarchical Linear and Nonlinear Modeling, Scientific Software International, Lincolnwood, IL, 2001. Reinganum, Jennifer F., "A Simple Model of Equilibrium Price Dispersion," The Journal of Political Economy, 87, 4 (1979), 851-858. Salop, S., "Monopolistic Competition with Outside Goods," Bell Journal of Economics, 10, (1979), 141-156. Salop, S. and J. E. Stiglitz, "Bargains and Ripoffs: A Model of Monopolistically Competitive Price Dispersion," The Review of Economic Studies, 44, 3 (1977), 493-510. Salop, S. and J. E. Stiglitz, "The Theory of Sales: A Simple Model of Equilibrium Price Dispersion with Identical Agents," The American Economic Review, 72, 5 (1982), 1121-1130. Samuelson, Paul A. and William D. Nordhaus, Economics, McGraw-Hill, New York, 1992. Schindler, Robert M., "The Excitement of Getting a Bargain: Some Hypotheses Concerning the Origins and Effects of Smart-Shopper Feelings," In T. K. Srull (Ed.), Advances in Consumer Research, 16, Association for Consumer Research, Provo, UT, 1989, 447-543. Schwieterman, J., "Fare is Fair in Deregulation: The Decline of Price Dispersion," Regulation, Summer, (1985). Shankar, Venkatesh and Ruth N. Bolton, "An Empirical Analysis of Determinants of Retailer Pricing Strategy," Marketing Science, 23, 1 (2004), 28-49. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 140 Shankar, Venkatesh, Arvind Rangaswamy and Michael Pusatari, "The Online Medium and Customer Price Sensitivity," Working Paper, University of Maryland, College Park, MD 20742, (2001). Singer, Judith D., "Using SAS PROC MIXED to Fit Multilevel Models, Hierarchical Models, and Individual Growth Models," Journal of Educational and Behavioral Statistics, 24, 4 (1998), 323-355. Sinha, Indrajit, "Cost Transparency: The Net's Real Threat to Prices and Brands," Harvard Business Review, March-April, (2000), 43-54. Slama, Mark E. and Terrell G. Williams, "Generalization of the Market Maven's Information Provision Tendency Across Product Categories," Advances in Consumer Research, 17, (1990), 48-52. Smith, Michael D., Joseph Bailey and Erik Brynjolfsson, "Understanding Digital Markets: Review and Assessment," In Kahin (Ed.), Understanding the Digital Economy, MIT Press, Cambridge, M.A., 2000, 99-136. Smith, Michael D. and E. Brynjolfsson, "Customer Decision Making at an Internet Shopbot: Brand Still Matters," The Journal of Industrial Economics, 49, 4 (2001), 541-558. Snijders, T. A. B. and R. J. Bosker, Multilevel Analysis: An Introduction to Basic and Advanced Multilevel Modeling, Sage Publications, Thousand Oaks, CA, 1999. Sorensen, Alan T., "Equilibrium Price Dispersion in Retail Markets for Prescription Drugs," Journal of Political Economy, 108, 4 (2000), 833-850. Spulber, Daniel F., "Bertrand Competition When Rivals' Costs Are Unknown," Journal of Industrial Economics, 43, 1 (1995), 1-11. Stavins, Joanna, "Price Discrimination in the Airline Market: The Effect of Market Concentration," Review of Economics and Statistics, 83, 1 (2001), 200-202. Stigler, George J., "The Economics of Information," The Journal of Political Economy, 69, 3 (1961), 215-225. Tate, R. L. and Y. Wongbundit, "Randomo versus Nonrandom Coefficient Models for Multilevel Analysis," Journal of Educational Statistics, 8, (1983), 103-120. Varian, Hal R., "A Model of Sales," The American Economic Review, 70, 4 (1980), 651-659. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. 141 Webster, W. J., R. L. Moendro, T. H. Orsak and D. Weerasinghe, "An Application of Hierarchical Linear Modeling to the Estimation of School and Teacher Effects," Mimeo, Annual Meeting of the American Educational Research Association, San Diego, CA, (1998). Zhu, Kevin, "Information Transparency of Business-to-Business Electronic Markets: A Game-Theoretic Analysis," Management Science, 50, 5 (2004), 670-685. R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays on technological evolution and financial returns to innovation

PDF

Augmenting knowledge reuse using collaborative filtering systems

PDF

Information systems capabilities and indirect relational value: Sustaining networked organizations

PDF

Defining the advice -seeking request in electronic solicitations in boundary spanning environments

PDF

Distributed annotation framework supporting collaborative filtering of information

PDF

Exploring, understanding and recording analysis steps in information-management applications

PDF

IT -enabled dynamic capabilities in new product development: Building a competitive advantage in turbulent environments

PDF

Beyond binary theories to agile learning: An empirical investigation of learning methods in warranty management.

PDF

Crisis management and narcissism: A Kohutian perspective

PDF

Information diffusion and the boundary of market efficiency

PDF

An investigation of repeat visit behavior on the Internet: Models and applications

PDF

Does it matter what people say about you: the impact of the content of buzz on firm performance

PDF

Essays on product trade-ins: Implications for consumer demand and retailer behavior

PDF

An empirical examination of customer perceptions of service quality utilizing the extended service quality model under the condition of multiple subunit service providers

PDF

Quality investment and advertising: an empirical analysis of the auto industry

PDF

An investigation of certain pricing practices in restraint of trade

PDF

A comparative examination of the impact of business -government relations on labor market reform in Egypt and Mexico, 1975--1995

PDF

Interaction of low -energy electrons with beams of sodium clusters, nanoparticles, and fullerenes

PDF

Interactive versus self effort: untangling the cognitive elaboration -- learning relationship on internet discussion forums

PDF

A typology of corporate environemental strategy and its driving factors in jultinational corporations

Asset Metadata

Creator Sin, Raymond Gee Han (author)

Core Title An empirical investigation of everyday low price (EDLP) strategy in electronic markets

School Graduate School

Degree Doctor of Philosophy

Degree Program Business Administration

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag business administration, marketing,Economics, Commerce-Business,Economics, Theory,OAI-PMH Harvest

Language English

Contributor Digitized by ProQuest (provenance)

Advisor Chellappa, Ramnath K. (committee chair), El Sawy, Omar A. (committee member), Hann, Il-Horn (committee member), Siddarth, Sivaramakrishnan (committee member)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-457380

Unique identifier UC11336730

Identifier 3196892.pdf (filename),usctheses-c16-457380 (legacy record id)

Legacy Identifier 3196892.pdf

Dmrecord 457380

Document Type Dissertation

Rights Sin, Raymond Gee Han

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA