INTERNET-ENABLED MARKET TRANSPARENCY: IMPACT ON PRICE ELASTICITY OF DEMAND IN THE AIR TRAVEL INDUSTRY Nelson Granados Ph.D. Candidate [email protected] Alok Gupta Associate Professor [email protected] Robert J. Kauffman Professor and Director, MIS Research Center [email protected] Information and Decision Sciences Department Carlson School of Management, University of Minnesota Minneapolis, MN 55455 Last revised: May 8, 2006 ______________________________________________________________________________ ABSTRACT The Internet revolution brought about significant changes in the availability of market information to consumers in many industries. It reduced the costs of information search, offering consumers multiple purchasing channels and product options. For example, air travelers are now able to browse the Web for hundreds of travel offers to their destination, compared to typically few offers from a preferred travel agent or airline prior to the Internet era. We analyze a dataset of airline tickets and empirically examine the impact of information technology-enabled market transparency on air travel demand and on consumers’ sensitivity to prices. We examine the effects of both product and price transparency by comparing travel agencies that differ in the product and price information provided by their selling mechanisms. Price transparency is related to information about market prices, while product transparency is related to information about product characteristics and quality. We find that in a more price transparent channel the price elasticity is relatively higher, while for a more product transparent channel the price elasticity is relatively lower. Therefore, sellers can partially offset the negative consequences of price transparency with innovative mechanisms that display product information. In addition, we find that demand in the online channel is elastic, while demand in the offline channel is inelastic. However, the drivers of the higher elasticity of the online channel differ between the opaque and the transparent channels. Hotwire, an opaque Web site, is relatively price elastic due to the lack of product information. On the other hand, OTAs such as Expedia and Travelocity are price elastic due to the relatively high level of price transparency. Keywords: Air travel industry, asymmetric information, consumer demand, e-commerce, electronic markets, empirical research, product transparency, price transparency, search costs.

1 1. INTRODUCTION The Internet revolution brought about significant changes in the availability of market information to consumers in many industries (Tapscott and Ticoll, 2003). It reduced the costs of information search, offering consumers multiple purchasing channels and product options. For example, air travelers are now able to browse the Web for hundreds of travel offers to their destination, compared to typically few offers from a preferred travel agent or airline prior to the Internet era. The Internet has increased the level of market transparency, the ability of buyers, sellers, and intermediaries to observe market information. Two forms of market transparency have been particularly affected, product and price transparency. Product transparency is related to information about the product characteristics and quality. Price transparency is related to information about market prices, including historical transaction prices and price quotes. Advanced information technologies (ITs) such as the Internet provide a platform that reduces information search costs, increasing the potential for market transparency. Information systems (IS) researchers have found analytical support for the hypothesis that an increase in market transparency benefits consumers because they are able to better discern the product that best fits their needs at a better price, at a lower cost (Bakos, 1997). On the other hand, IT-enabled market transparency is a strategic paradox for organizations. Due to better informed consumers, the very benefit of the Internet—making information available to facilitate product marketing and distribution—also makes it difficult to capture profits (Porter, 2000). Therefore, firms are faced with the challenge to balance the trade-off between the benefits of attracting consumers with market information, and the risk of losing information advantages. However, there is little research regarding the impact of market transparency on consumer demand. In this paper, we empirically examine how consumer

2 demand is influenced by different levels of market information and provide insights that will help firms to develop economically-sound transparency strategies in the presence of advanced ITs. This study lays a foundation for future research on the broader welfare effects of market transparency. We address the following research questions: What is the impact of IT-enabled market transparency on consumer demand? What are the implications for firm strategy? To answer these questions, we provide empirical evidence of the impact of market transparency on consumers’ sensitivity to prices. We estimate an air travel demand function using a data set of airline tickets sold by U.S. travel agencies during a one-year period, and use it to compare price elasticities across travel agencies that exhibit different transparency levels. The main contribution of this study is that we provide empirical evidence of the relationship between market transparency and price elasticity in both online and offline channels. The air travel industry is well suited for this type of study, since it is a representative case of how IT has enabled multiple B2C market mechanisms with different levels of information disclosure. In 2001 major U.S. airlines launched an online travel agency (OTA) called Orbitz, claiming that it was the most transparent travel website in the market. Other competitors have introduced niche strategies with innovative, non-transparent mechanisms, such as Priceline.com (Granados et al., 2005a). In a sense, OTAs are performing field experiments with different levels of market transparency, providing an ideal research setting to test how the observed variance in information disclosure affects consumers’ economic behavior. The rest of this paper is organized as follows. In the second section, we summarize the existing research on market transparency in the IS, marketing, and economics disciplines. In the third section, we provide an account of market transparency in the air travel industry. In the

3 fourth section, we provide details on the data set and on the air travel demand model. In the fifth section, we present the hypotheses, methodology and results of our econometric analysis. In the sixth section, we discuss the results and derive strategic implications for firms. Finally, we conclude with the contributions that we have made to build a new theory of market transparency, the limitations of our empirical analysis, and some directions for future research. 2. THEORETICAL BACKGROUND AND PROPOSITIONS We will introduce the concept of market transparency and related theory, and formulate propositions regarding the impact of market transparency on consumer demand. We define market transparency as the availability and accessibility of product and price information. Our focus is on the impact of market transparency on consumers, and so our references to market transparency will be related to the information that is available and accessible to consumers. Sellers and intermediaries make choices about the information they provide to consumers in the selling process. With the Internet, these choices have expanded significantly to that multiple selling mechanisms now exist with different levels of price and product information. Next, we examine theory on these dimensions of market transparency in more detail. The Impact of Price Transparency Lower search costs for price information may lead to lower prices, to the benefit of consumers. Bakos (1997) modeled the informational effect of electronic marketplaces and concluded that lower search costs reduce the ability of sellers to extract monopolistic profits. Respective reductions in price premiums may occur as sellers lose their ability to exploit high search costs, or alternatively, due to an increased ability of consumers to find lower market prices (Stigler, 1961). For example, Brynjolfsson and Smith (2000) examined the book and music markets and found that Internet retailers have lower prices than conventional retailers. The

4 authors suggest that lower search costs on the Internet may help explain this price differential. Although they consider the full price charged to the consumer based on shipping, handling charges and transportation costs, they did not include the distribution cost differential across channels. This leaves distribution costs as another possible explanation for the difference in prices across channels. In the air travel industry, given the existing price dispersion reported by Clemons et al. (2002) and others, the price transparency offered by OTAs may drive prices down, as consumers discover lower fares. Based on existing theory of price transparency, we hypothesize that higher price transparency will increase consumers’ sensitivity to prices. Analogously, through a meta-analysis of previous studies, Kaul and Wittinck (1995) found that price advertising is associated with higher price sensitivity among consumers. This leads to the following proposition regarding the impact of price transparency on price sensitivity. •

Proposition 1 (The Price Transparency – Price Sensitivity Proposition): Price transparency increases consumers’ sensitivity to prices.

The Price Transparency – Price Sensitivity Proposition (P1) suggests that availability and accessibility to more prices may increase sensitivity to price changes as consumers become aware of prices of substitutes and horizontally differentiated alternatives. Product Transparency Reduced search costs for product information may improve market efficiency and prevent market failure (Bakos, 1997). Consider a perspective that is consistent with Akerlof (1970): a diminution in information asymmetry between buyers and sellers regarding product quality will help to shore up the fundamentals of a sound market. For example, Brynjofsson et al. (2003) found an increase in consumer surplus due to increased product variety offered by online

5 bookstores. Lynch and Ariely (2000) performed experiments of Internet-based online wine sales and found that transparency about product quality and store comparison ability increased consumer retention. Based on existing literature related to the impact of product transparency, we hypothesize that product transparency reduces consumers’ sensitivity to prices. Analogously, Kaul and Wittinck (1995) found that product advertising leads to lower price sensitivity on the part of consumers. Higher product transparency is analogous to advertising, which seeks to inform consumers to differentiate sellers and products. Moreover, Lynch and Ariely (2000) found that higher product transparency led to lower price sensitivity. The following proposition summarizes these findings: •

Proposition 2 (The Product Transparency – Price Sensitivity Proposition): Product transparency decreases consumers’ sensitivity to prices.

The Product Transparency – Price Elasticity Proposition (P2) suggests that higher product transparency reduces consumers’ attention to price differences and price changes, as they find alternatives that better satisfy their needs. Product and Price Transparency Combined Market transparency may have a different effect on consumers depending on the industry context. In particular, the weight that consumers place on price and product information may be different in commodity versus differentiated markets. In commodity markets, the effect of making prices more transparent should prevail over the effect of making product quality transparent (Lynch and Ariely, 2000). Conversely, lowering price search costs should matter less in industries that exhibit a high degree of product differentiation (e.g., luxury items). On the other hand, price transparency may have a higher impact as the number of sellers increases

6 (Bakos, 1997). Therefore, other things being equal, in an industry with low concentration and multiple product offerings consumers may be more sensitive to price information. In summary, market transparency can have different effects on consumer behavior depending on the information that is disclosed and the characteristics of the industry. In markets where consumers enjoy high price transparency, there will likely be a higher sensitivity to prices. But product transparency—particularly when it enables differentiation—is likely to have a positive impact on consumers, particularly by reducing price sensitivity. The combined effects of product and price transparency will vary depending on industry characteristics, such as the number of sellers and their ability to differentiate their product offers.

3. EMPIRICAL CONTEXT: THE AIR TRAVEL INDUSTRY In this section, we provide an account of IT-enabled market transparency strategies in the air travel industry to set the stage for the analysis of the impact of market transparency on air travel demand. IT-Enabled Market Transparency in the Air Travel Industry Since the first Internet travel website was launched in 1995, there has been an unprecedented growth in online airline ticket sales. A 2003 industry survey estimated that the percentage of tickets sold over the Internet had reached 16% worldwide and 40% in North America (O’Toole, 2003). The fast early growth of Internet-based airline ticket sales was facilitated by existing global distribution systems (GDS), which were originally developed as interorganizational systems for airline and travel agency distribution. The first computerized reservation systems (CRS) were set up in the 1980s at travel agency locations to provide agencies with the ability to distribute airline seat inventory (Copeland and McKenney, 1988).

7 In 1996, Sabre Holdings, operator of the GDS formerly owned by American Airlines, capitalized on the technological opportunities offered by the Internet. It established Travelocity (www.travelocity.com), one of the first online travel agencies (OTAs). Soon after that, multiple market players emerged to create competition. Existing reservation systems served as readily available search engines for new Internet-based air travel distributors, including non-travel firms (Chircu and Kauffman, 2000). For example, in 1996 Microsoft introduced its own travel services and reservation-making website, Expedia, (www.expedia.com). Priceline.com (www.priceline.com) emerged as the first low-transparency OTA, by developing a selling mechanism that shielded product and price information from the customer until after she committed to a contract-binding bid. Hotwire (www.hotwire.com), a low product transparency OTA launched by major airlines to compete with Priceline.com, began its operations in 2000. In 1999, American, Delta, Northwest, Continental, and United Airlines announced they would create a new OTA called Orbitz. Its website (www.orbitz.com) was launched in June 2001, and since then Orbitz has grown to become a technology leader in its quest to update the legacy systems of airline reservations. The airlines claimed that Orbitz would dramatically decrease the high costs of making reservations. For that purpose, Orbitz was designed and powered by ITA Software (www.itasoftware.com), a pricing and airfare shopping technology developer launched by researchers from the Artificial Intelligence Laboratory at MIT. This software obtains fares directly from the Airline Tariff Publishing Company (www.atpco.net), which collects and distributes fares from airlines worldwide, and it obtains airline travel schedules from the Official Airline Guide (www.oag.com). So, by using ITA software, Orbitz avoids reliance on legacy system infrastructures and their high transaction fees. Figure 1 shows the technological structure of fare distribution in the air travel industry in light of Orbitz.

8 Figure 1. The Technological Structure of Air Travel Distribution

ITA Software

Airline Fares

Airline Schedules

AirlineTariff Publishing Company (ATPCO)

Official Airline Guide (OAG)

Orbitz

Global Distribution Systems (GDSs)

Computer Reservation Systems (CRSs)

Online Travel Agencies (OTAs)

Traditional Travel Agencies

Airline Portals, Reservation Offices

Note: Before OTAs, the dominant electronic systems for air travel distribution were GDSs and CRSs. With Internet-based OTAs, however, a new distribution structure emerged (indicated by the dashed box). In particular, Orbitz introduced technology to distribute airline tickets without reliance on GDS and CRS distribution (indicated by the arrows in bold). Orbitz provides these services to other distributors such as travel agencies as well.

OTAs have aggressively competed for customers by matching the transparency level offered by Orbitz. However, because Orbitz enjoys more powerful technology without reliance on legacy reservation systems, it continues to be at the forefront. For example, while most OTAs can display 20 to 40 product offers for a customer’s search request, Orbitz can normally display more than 100 options. On the other hand, existing differences in market transparency may be driven by strategic choices reflected in the characteristics of the selling mechanism. For example, while Orbitz, Expedia, and Travelocity offer a comparison of products and prices in one table, Hotwire only displays one or two prices upfront, without information about the airline or the itinerary. Air travel channels display different levels of IT advancement. Traditional travel agencies, which typically interact via phone or in person with customers, can provide detailed information

9 regarding product characteristics, such as schedule, stopovers, and fare restrictions. However, the channel is inherently not price-transparent, because there is a limit to the number of priced itineraries that can be relayed without an electronic interface. With the advent of OTAs, not only the product transparency of traditional travel agencies was replicated, but travelers can observe multiple price offers with sorting capability. Most OTAs are dependent and restricted by the GDS technology, which was selected as the fare search and booking engine in many original business models, such as those of Travelocity and Expedia. The path dependency generated by these technological decisions has enabled other innovate firms to enter the market and compete aggressively, such as Orbitz with their transparent mechanism, and Hotwire with their opaque mechanism. Due to these technological imperatives, it is likely that the variety of market transparency levels across agencies will remain. Therefore, travel agencies need to better understand the impact of different levels of product and price transparency on consumers to effectively strategize in this diverse transparency and technological environment. The U.S. travel agency industry is a convenient setting to study the impact of IT-enabled market transparency on consumer demand. Online and offline travel agencies have different levels of market transparency, which represents an ideal setting to research how different degrees of product and price information affect consumers’ economic behavior. In addition, these different transparency levels are influenced by technological choices, such as the OTAs’ decisions to power their fare search engines. Next, we present the data and the air travel demand model that we will use to assess the impact of market transparency on demand.

10 4. DATA AND THE AIR TRAVEL DEMAND MODEL We analyzed a cross-sectional data set of economy class tickets sold by OTAs and traditional travel agencies in the United States from September 2003 to August 2004 in 46 city pairs, such as San Francisco-New York or Boston-Honolulu. The data set contains the outbound portion of all tickets sold via GDSs to these 46 city pairs, so excluded from this sample are airline direct sales, including frequent flyer award tickets, which are usually transacted through airline portals or reservation offices. Tickets sold were aggregated by agency type, time of purchase, and season. The agency types are Expedia, Travelocity, Orbitz, Hotwire, Online Other, and Offline. Agencies that did not sell tickets online were classified as Offline. Time of purchase was defined in terms of weeks before departure up to 20 weeks, so a value of 2 indicates that the ticket was sold between 8 and 14 days before flight departure. Therefore, the data contained bookings up to 5 months before departure, which constitutes the period of main booking activity. The seasons were peak and off-peak; peak season tickets were sold for travel in June, July, August and December 15 to January 14. Each record contained the tickets sold and the average one-way price paid by consumers for each city pair, week before departure, agency type, and season, for a total of 11,040 records with information for 2.15 million tickets. For the purpose of this analysis, we excluded high-season tickets because the quantities sold are more likely to reflect industry capacity or supply restrictions rather than unconstrained demand patterns. In addition, we excluded Orbitz’s sales; due to its booking technology, most of its sales are not done via GDSs. These exclusions reduced the data set to 4,420 records with aggregate information for 1.32 million tickets. There are two advantages of this information structure to examine the impact of market transparency on consumer demand. First, the data set contains prices and tickets sold by agency

11 type. Therefore, the aggregation at the agency type level allows a straight forward estimation of the variation in sales based on agency-specific factors while controlling for other factors that influence demand. Second, we were able to measure the relative levels of market transparency by agency in order to assess the impact on demand. Modeling Preliminaries To test the impact of market transparency on demand, we first consider an air travel demand model of the form QUANTITY = f(PRICE; CONTROL VARIABLES), where QUANTITY represents air travel demand, and CONTROL VARIABLES represents a vector of control variables. Next, we introduce the control variables in the model. (See Table 1.) Table 1. Air Travel Demand Model Variables VARIABLE TYPE Dependent Independent Control

VARIABLE

DEFINITION

QUANTITY PRICE INCOMEOD

Tickets sold Average price paid Sum of gross product per capita of origin and destination cities Time of purchase in weeks before flight departure ADVPURCH City-pair category, coded as (0,0) for no hub (HUB1, HUB2) operations, (1,0) for origin city hub operation, (0,1) for destination city hub operation, and (1,1) for both origin and destination city hub operation. Trip time in terms of block hours, or the scheduled BLKHRS time of taxiing from/to gates plus the travel time. Agency Type Dummies Dummies to capture agency-type specific factor that influence demand (D)

Note: The data source for U.S. income per capita and population was the U.S. Bureau of Economic Analysis (BEA) Time-Series Data for Metropolitan Statistical Areas (www.bea.doc.gov). Trip time was estimated by averaging the block hours of non-stop flights reported in the Official Airline Guide.

Control Variables Income (INCOMEOD). Income is a standard predictor in demand models (Brynjolfsson et al., 2003). In airline route planning, an airport’s catchment area is the area within which

12 travelers commute to fly from the airport. (See Figure 2.) We are interested in estimating the income level of consumers in a city-pair’s catchment area. Figure 2. Sample Airport Catchment Area for Fort Wayne, Indiana

Source: Airlines Reporting Corporation (www.arccorp.com/img/catchmentmap.gif), December 2005.

In the database, the income per capita of an airport’s catchment area is estimated based on official 2003 gross product statistics of its corresponding city or metropolitan area. For example, for U.S. airports the source is the Metropolitan Statistical Area (MSA) personal economic data from the Bureau of Economic Analysis (www.bea.doc.gov). INCOMEOD is measured by adding the gross product per capita of the MSA for the origin and destination airports. Advance Purchase (ADVPURCH). Another relevant variable is the time of purchase prior to departure. A pervasive and well-recognized difference between consumers is the urgency of purchase (Stigler, 1964). In particular, demand may be significantly affected by consumers’ sense of urgency in markets with perishable products. Airlines have used this feature of the air

13 travel product to price discriminate across the booking period prior to flight departures (Clemons et al., 2002). In this study, we incorporate time of purchase measured in weeks before departure. City-Pair Hub Operation Category (HUB1, HUB2). One component of service quality is the number of non-stop flights offered for a given city-pair. Airlines have historically used hub operations to lock in consumers who benefit from the convenience of non-stop operations to multiple destinations. We coded origin-destination city-pair hub operation categories as follows. No hub operation at either origin or destination was coded (0,0). If one or more airlines had a primary hub operation at the origin but not at the destination (e.g., United Airlines in Chicago, or Delta in Atlanta), then we coded the city pair with (HUB1, HUB2) = (1,0). Similarly, if one or more airlines had a hub operation at the destination, then the city-pair was coded (HUB1, HUB2) = (0,1). If hub operations were present at both the origin and destination, the coding was (1,1). Travel Time (BLKHRS). We refer to travel time as the time that passengers spend traveling from one city to another. We measure this variable in terms of block hours, or the time between departure from the gate at the origin airport and arrival at the gate in the destination airport. The advantage of using block hours to measure travel time is that it represents several aspects of the travel experience. First, it incorporates information about the travel distance or stage length, a typical measure to control for route effects. Generally, the shorter the distance, the shorter is the travel time. Shorter trips are more commodity-like than long distance trips, given the lower need for comfort and reliability. However, the average consumer will be more aware of the time that it takes to travel in a city-pair, rather than the air miles, so we use block hours as the estimate of trip length. Second, block hours also include the time for taxiing from and to the gates, take-off, and landing, which increasingly influence travel times and perceptions of quality due to congestion in U.S. airports, and since consumers are increasingly able to learn about average

14 delays and on-time performance. Third, demand for shorter trips is influenced by alternate modes of transportation, such as trains or automobiles, and all-in time to destination is important. We measured BLKHRS based on the average historical block hours for a city-pair’s non-stop flights reported by airlines in the Official Airline Guide (www.oag.com) during the period January to March 2004. The Empirical Model Air travel demand models commonly use linear or log-linear specifications (Lee, 1990; McGill and van Ryzin, 1999; Talluri and van Ryzin, 2004). We specify a log-linear model, consistent with prior studies of Internet-based demand (Brynjolfsson et al., 2003), of the form: QUANTITY = e CONSTANT * PRICEη * INCOMEOD β1 * ADVPURCH β2 * e β3HUB1 * e β4 HUB2 * BLKHRS β5 *

∏e

α jDj

(1)

* eε ,

j

where -η is the price elasticity of demand, Dj’s represent dummy variables for each agency type,

βi and αj are coefficients to be estimated, and ε is a random error term. The log-linear transformation of Equation 1 is: ln(QUANTITY ) = CONSTANT + η ln( PRICE) + β1 ln( INCOMEOD) + β 2 ln( ADVPURCH ) + β 3 HUB1 + β 4 HUB2 + β 5 ln( BLKHRS ) +

∑α D j

j

+ε

(2)

j

Table 4 lists the descriptive statistics of the model variables. (See Table 4.) There was an average 300 tickets sold per city pair, week before departure, and agency type, at an average oneway price of $160 and 4.62 block hours. We next evaluate the soundness of this econometric model of air travel demand. Modeling Diagnostics The classical linear regression model assumes specific characteristics of the error term and the variables in the model (Greene, 2003). We next diagnose whether the model specification

15 Table 4. Descriptive Statistics for Model Variables VARIABLES QUANTITY PRICE INCOMEOD (000s) ADVPURCH HUB1 HUB2 BLKHRS

MEAN 299.74 $159.55 $73.57 10.5 0.82 0.68 4.62

STANDARD DEVIATION 1403.95 $95.45 $6.98 5.77 0.38 0.46 2.65

MIN

MAX

1 $27 $61 1 0 0 1.41

46,021 $1078 $90 20 1 1 11.00

Note: N = 4,420 tickets sold with 9 origin cities and 17 destinations for each origin.

and data are consistent with these assumptions. Correlation and Multicollinearity. We examined the pair-wise correlations of the variables in log-transformed formats. (See Table 3) The only pair-wise correlation of any concern was between ln(PRICE) and ln(BLKHRS) at 0.69. This correlation may exist if airlines charge higher prices for longer travel times due to higher variable costs of operation (Duliba et al., 2001) or due to price-differentiated service for long-haul and international trips. Table 3. Pairwise Correlation Matrix of Log-Transformed Variables ln(VARIABLES)

PRICE INCOMEOD ADVPURCH exp(HUB1) exp(HUB2) BLKHRS

PRICE

1 -0.28 -0.19 0.04 -0.04 0.69

INCOME ADV exp(HUB1) exp(HUB2) BLKHRS OD PURCH

1 -0.00 -0.28 -0.13 -0.25

1 0.00 -0.00 0.00

1 -0.06 0.02

1 -0.06

1

Note: The highest pair-wise correlation was between BLOCKHOURS and PRICE at 0.69. Correlations for the log-transformed variables are similar in magnitude and sign as the raw values.

Kennedy (2003) suggests that pair-wise correlations at the 0.80 level or higher typically ought to be a source of concern. The inclusion of highly correlated variables often results in multicollinearity and unstable parameter estimates. To diagnose these problems beyond simple pairwise correlations, we calculated the variance inflation factors (VIFs) (Kennedy, 2003). The

16 mean VIF was 1.97 for all the variables, and the maximum VIF value was for ln(PRICE) at 3.14. These VIFs are far less than the typical cautionary threshold value of 10 or 20 (Kennedy, 2003) that could signal a serious problem with multicollinearity. Non-linearity. By taking the log-transformation of some variables, we expect to approximate linear relationships that can be examined with a classical OLS regression. Partial regressions of each independent variable on ln(QUANTITY) suggest a linear relationship for the logtransformed variables. (See Figure 3).

-4

-4

-2

-2

e( lnquantity | X ) 0

e( lnquantity | X ) 0

2

2

4

4

Figure 3. Partial Regression Plots of Log-Transformed Variables

-1

-.5

0 .5 e( ln price | X )

1

1.5

-.2

coef = -.51874815, se = .05581296, t = -9.29

-.1

0 .1 e( lnincomeod | X )

.2

.3

-4

-4

-2

-2

e( lnquantity | X ) 0

e( lnquantity | X ) 0 2

2

4

4

6

coef = 2.4076088, se = .18996382, t = 12.67

-3

-2

-1 0 e( lnadvpurch | X )

coef = -1.474402, se = .02188453, t = -67.37

1

2

-2

-1

0 e( lnblockhours | X )

1

2

coef = -.1547856, se = .04390474, t = -3.53

Note: Partial regression plots for each dependent variable were derived by plotting of residuals e(ln(VARIABLE) | X) against e(ln(QUANTITY) | X), where X represents regressors other than the variable evaluated (i.e., VARIABLE).

Heteroskedasticity. The classic OLS regression model assumes that the variances of the error terms are constant. We performed a Breusch-Pagan (1979) Lagrange multiplier test for heteroskedasticity against the fitted values of ln(QUANTITY), in other words, at the level of model. The hypothesis of homoskedasticity or constant variance was rejected (χ2 = 69.25, d.f.=1, p < 0.01). We conclude that there is heteroskedasticity in the econometric estimation, although

17 this test cannot diagnose exactly what its source is. Intuitively though, an apparently obvious potential source of heteroskedasticity is income level in the catchment area, INCOMEOD. In air travel, wealthy regions may exhibit a higher variance in sales than less wealthy regions. Based on the observation that INCOMEOD might account for heteroskedasticity, we ran a second, less general test by Goldfeld and Quandt (1965). We consider a known source of heteroskedasticity (i.e., Var[εi] = σ2i = σ2i zi, with zi = INCOMEOD). We were not able to reject the null hypothesis of homoskedasticity at a significance level of p < 0.05 (p = 0.07). But if we view p < 0.10 as the criterion level of significance to determine the presence of heteroskedasticity, the results we obtained would suggest the known source of INCOMEOD. Thus, it is necessary to estimate the model with weighted least squares or regression with robust standard errors. Endogeneity. In demand models, endogenously generated prices can lead to misspecification because there may be a correlation between the observed prices and the residuals (Villas-Boas and Winer, 1999). In the air travel industry demand and price may be simultaneously determined, as suppliers strategize with prices based on the supply of seats and their estimates of demand. In our model we are interested in measuring relative demand across agency types. However, prices are generally set by airlines, so we consider the risk of simultaneity between prices and relative demand across agency types low in this scenario. To verify this theoretical assessment, we performed a two-stage least squares (2SLS) regression with instrumental variables that are correlated with price. These variables are: •

STAGELENGTH: We define stage length as the distance in travel miles between two citypairs. Based on our theoretical model of optimal price setting, prices depend on the marginal cost, or the cost per available seat mile. Therefore, the higher the mileage, the higher the total costs of operation.

18 •

ONLINE: In the Internet channel, marginal costs may be lower due to a higher level of automation (Riggins, 2004), lower costs of facilitating consumer search (Zettelmeyer, 2000), or less reliance on labor-intensive face-to-face or phone service. This cost differential across channels may influence pricing policy. We use the dummy variable ONLINE with a value of 1 for tickets sold by OTAs (i.e., Expedia, Travelocity, Hotwire, Online Other) and 0 otherwise.

•

MKT_CONC: Monopolistic and oligopolistic market environments may lead to higher prices. We measured the market concentration of each city-pair using the Herfindahl index, which is the sum of squares of the market shares of the different airlines that serve a city-pair. To estimate the Herfindahl index, we included all airlines that serve a city-pair, regardless of the number of stops. The 2SLS regression with the above instrumental variables led to similar coefficients as in

the original regression. We performed the Hausman test to test the null hypothesis that the difference in coefficients is not systematic. We were not able to reject the hypothesis (χ2 = 2.39, d.f.=10, p=0.99). Thus, endogenously-generated prices probably do not lead to a misspecification of the log-linear demand model of air travel demand. 1 Demand Estimation Results The results of the OLS robust standard errors and 2SLS regressions are shown in Table 4. To address the possibility of heteroskedasticity, we report both regression results with standard errors based on the Huber-White sandwich estimators. In the OLS regression, the coefficient 1

Another means to assess endogeneity is to analytically examine the underlying model. Consider a monopolistic airline in a market with a simplified demand function that resembles our industry-level model: DEMAND = a * PRICEη * ε . Assume that the variables influencing the error term cannot be observed by the researcher, but they can be observed by the firm. We can show that solving the profit maximization problem for the monopolist leads to an optimal price that is not dependent on the error term. In contrast, this argument cannot be made for linear demand models, in which the optimal price is a function of the error term (Villas-Boas and Winer, 1999). Thus, our model diminishes misspecification risk due to endogenous prices.

19 estimates and standard errors were stable in comparison with the 2SLS results, except for the coefficient of ln(PRICE), which changed from -0.50 to -0.71 (both significant at the p < .01 level). This stability of the coefficients is in line with the Hausman test described above, which suggests no evidence that the difference between coefficients in the two regressions is systematic. Therefore, both regressions lead to the same conclusions, so we now turn our attention to the results in the 2SLS results. We continue to report the OLS results. Table 4. Air Travel Demand Model: OLS and 2SLS Regression Results 2SLS

OLS

VARIABLES

Coefficient

Robust Std. Error

Coefficient

Robust Std. Error

CONSTANT PRICE

-1.56* -0.83***

1.27 0.15

-3.56*** -0.52***

0.86 0.06

• Control Variables 2.24*** INCOMEOD -1.52*** ADVPURCH 0.30*** HUB1 0.06* HUB2 0.02 BLKHRS

0.20 0.03 0.04 0.04 0.09

2.41*** -1.47*** 0.30*** -0.05 -0.15***

0.18 0.03 0.04 0.04 0.05

• Agency Type Dummies 2.59*** 0.08 2.48*** 0.06 EXPEDIA 2.40*** 0.07 2.31*** 0.05 TRAVELOCITY 1.17*** 0.07 1.08*** 0.05 ONLINEOTHER 4.80*** 0.11 4.62*** 0.06 OFFLINE R2 (Adj. R2) 74.48% (74.42%) 74.65% (74.59%) Note: N = 4,220. The significance levels for the coefficients are * = p < 0.10, ** = p < 0.05, and *** = p < 0.01.

Control Variables. INCOMEOD has a positive relationship with demand (β1 = 2.24, SE = 0.20, p < 0.01), a rather intuitive finding. ADVPURCH has a negative relationship with demand (β2 = -1.52, SE = 0.03, p < 0.01), which is in line with the notion that demand is higher closer to departure. This result corroborates the common practice of airlines to set higher prices closer to departure. HUB1 has a positive relationship with demand (β3 = 0.309, SE = 0.04, p < 0.01), which suggests that, other things being equal, the higher the number of non-stop flights in the

20 origin city, the higher will be the demand. HUB2 had a marginally significant positive relationship with demand (β4 = 0.06, SE = 0.04, p < 0.10). The coefficient for BLKHRS was not significant (β5 = 0.02, SE = 0.09, p = 0.81).

5. HYPOTHESES, METHODOLOGY AND RESULTS We now formulate the hypotheses regarding the impact of market transparency on air travel demand. We also present the methodology that we used to test hypotheses of the impact of product and price transparency on price elasticity. Research Hypotheses Price sensitivity is usually measured as the slope of the demand curve or as the price elasticity of demand (η), defined as η = − ∂x ∂p ⋅ p / x , where x is demand and p is price. We use price elasticity as the measure of price sensitivity. In line with the Price Transparency – Price Sensitivity Proposition (P1) and the Product Transparency – Price Sensitivity Proposition (P2), our hypotheses are: •

H1 (Price Transparency-Price Elasticity Hypothesis): Price transparency has a positive relationship with the price elasticity of air travel demand.

•

H2 (Product Transparency-Price Elasticity Hypothesis): Product transparency has a negative relationship with the price elasticity of air travel demand.

To test these hypotheses, we compared agency types in our data set based on an assessment of their relative levels of price and product transparency. Therefore, we needed to identify agency types that differed significantly in one level of transparency and not in the other. Next, we describe the methodology that we used to assess product and price transparency levels by agency type.

21 Market Transparency Measures by Agency Type There is no established set of measures of product and price transparency. There is only one study that we are aware of with explicit measures of transparency in a B2C electronic market, albeit an experimental one. Lynch and Ariely (2000) measured product and price transparency in wine store sales. They referred to product transparency as quality usability. The authors’ definition of high quality usability consisted of a consumer’s ability to view wine quality information up front, and the ability to sort the wines by type. In addition, subjects were able to drill down to see more differentiating characteristics for a given wine. Price transparency was labeled price usability. The authors gave a contrasting definition of high price usability as the ability of a consumer to view prices up front, with sorting and drill-down capabilities for further price comparison. Consistent with Lynch and Ariely’s experiments, we consider not only the availability of the information, but also the ability of consumers to view information in an organized way, such that it provides value towards a purchase decision. The definitions that we use for these two constructs are: • Price Transparency: The availability and accessibility of information about market prices. • Product Transparency: The availability and accessibility of information about product characteristics for a given set of alternatives. These definitions suggest that transparency is not only determined by the availability of market information, but also how it is made accessible, or how it is organized for consumption by consumers towards a purchase decision. Price Transparency. We measured the availability of price information in terms of the number of priced itineraries provided in the first screen. The higher the number of priced

22 alternatives, the higher is the ability of consumers to find an appropriate alternative at a lower price. This measure is appropriate in the U.S. air travel industry, where the combination of suppliers and travel itineraries can lead to numerous options for consumers in most city-pairs. Therefore, selling mechanisms that organize and present these options based on price are likely to have a significant impact on consumers (Bakos, 1997). In the context of this study, the higher the number of priced itineraries, the higher was the measure of price transparency that we assigned to an OTA. (See Table 5.) Table 5. Priced Itineraries by OTA to Proxy for Price Transparency AGENCY TYPE

PRICED ITINERARIES AVG. ST. DEV.

DESCRIPTION

Online Travel Agencies Expedia 25 Orbitz 183 Travelocity 32 Hotwire 1 Offline Travel Agencies

6 28 3 0.2

Average number of priced itineraries of 20 search requests for representative city-pairs and dates

Miscellaneous

NA

Agencies typically provide one price for specific search request.

1

Product Transparency. We measured the extent to which product characteristics and quality information are provided to the traveler prior to purchase. (See Table 2.) To measure the availability of product information, one of the authors assigned points for the presence in the first screen of relevant attributes of airline tickets to consumers (Clemons et al., 2002; Granados et al., 2005a). Accessibility was measured in terms of the ability to sort and compare products and their characteristics, in line with the Lynch and Ariely (2000) experiments. The methodology that was used to evaluate product transparency by OTA is described in detail in Appendix A. Market Transparency and Price Elasticity of Demand Estimation Results Based on the above air travel industry demand model, we next present the methodology and

23 Table 6. Product Transparency Measures by Agency Type AGENCY TYPE

PROD

AVAILABILITY

ACCESSIBILITY

TRANS

AIRLINE

ITIN

SORT

COMPARE

EXPEDIA

63

50

50

100

50

ORBITZ

92

100

100

66

100

POINT DEDUCTIONS Return itinerary not available in first screen. Comparison table only by airline. Cannot sort by arrival times.

Return itinerary not available in first screen. Comparison table only by airline. Major airline and specific flight 0 0 0 NA NA HOTWIRE times not specified. Average of Expedia and Travelocity; most other OTAs are 63 NA NA NA NA OTHER OTAs powered by GDSs. Accessibility impaired by non75 100 100 50 50 OFFLINE electronic interface. Note: PRODTRANS is the composite measure of product transparency, and equivalent to the average of availability and accessibility measures. NA means “not available.” JUSTIFICATION? TRAVELOCITY

63

50

50

100

50

results of the tests regarding on market transparency and price elasticity of demand. Estimation Model for Effects of Price Transparency. The Price Transparency-Price Elasticity Hypothesis (H1) suggests that price transparency increases price elasticity. To test this hypothesis, we compared estimates of the price elasticities of offline travel agencies and the GDS-based OTAs Travelocity and Expedia (GDSOTA). We used these agency types because, on a relative basis, they differ in the level of price transparency but not in the level of product transparency. (See Tables 5 and 6.) For example, a traveler who calls a traditional travel agency usually gets price quotes for one or two itineraries. On the other hand, with an Internet-based search request in Travelocity or Expedia, consumers have access to the same information that an offline agent can view, but can compare prices for multiple itineraries, compared to just a few quotes by the travel agent via phone or in person. In a sense, by using an OTA, the consumer is displacing the travel agency in the search process, and gaining immediate access to prices. Both offline and the GDS-based OTAs, Travelocity and Expedia, exhibit relatively similar

24 levels of product transparency. For a given product offer, an offline travel agency can quickly provide numerous details about the travel itinerary, the airline carrier, and different rules and restrictions. Travelocity and Expedia provide similar product information online, because they are powered by GDSs, the same reservation system used by offline travel agencies. Therefore, it is reasonable to assume that differences in price elasticity between these two agency types are driven by their different levels of price transparency as we define it. To compare price elasticities between agency types we consider the empirical model QUANTITY = e CONSTANT * PRICE η + λGDSOTA * INCOMEOD β1 * ADVPURCH β 2 * e β 3 HUB1 * e β 4 HUB2 * BLKHRS β 5 *

∏e

α jDj

* eε ,

(3)

j

In this model, GDSOTA is a dummy variable for Expedia and Travelocity. The coefficient, -η, is the base price elasticity for an offline travel agency, and -λ captures the price elasticity of GDSOTA relative to an offline travel agency. The log-linear transformation of this function is ln(QUANTITY ) = CONSTANT + η ln( PRICE) + λ * GDSOTA * ln( PRICE) + β 1 ln( INCOMEOD) + β 2 ln( ADVPURCH ) + β 3 HUB1 + β 4 HUB2 + β 5 ln(BLKHRS ) +

∑α

j

(4)

Dj + ε

j

The Price Transparency-Price Elasticity Hypothesis can then be expressed as H1: λ < 0. Results. Table 7 provides the results of the OLS regressions as specified in Equation 4. The estimated coefficients for the base case price elasticity of demand for offline agencies, and the dummy variable for GDSOTA were negative and significant, with estimates η = -0.82, SE = 0.12, p < 0.01, and λ = -0.47, SE = 0.01, p < 0.01. These estimates suggest that there is a price elasticity differential on the order of 0.47, with demand in GDS-based OTAs being more price elastic than in the offline channel. Therefore, we find support for the Price Transparency-Price Elasticity Hypothesis (H1), that higher price transparency leads to higher price elasticity of demand. We note that in both the OLS and the 2SLS regressions the price elasticity of OFFLINE

25 is inelastic with η < 1 , while the demand of GDS-based OTAs Travelocity and Expedia is priceelastic with η + λ > 1 . Table 7. Offline and GDSOTA-Based Estimation Results: The Price Elasticity Differential

VARIABLES

2SLS REGRESSION Robust Coefficient Std. Error

PRICE GDSOTA * ln(PRICE) CONSTANT

-0.82*** -0.47*** 0.73

• Control Variables INCOMEOD ADVPURCH HUB1 HUB2 BLKHRS R2 (Adj. R2)

2.79*** 0.23 -1.77*** 0.04 0.45*** 0.05 -0.02 0.05 -0.01 0.08 71.9% (71.9%)

OLS REGRESSION

Coefficient

Robust Std. Error

-0.57*** -0.45*** -0.45

0.08 0.01 1.10

0.12 0.01 1.31

2.91*** 0.22 -1.72*** 0.04 0.45*** 0.05 -0.02 0.04 -0.16*** 0.06 72.1% (72.0%)

Note: N = 2,760. * = p < 0.10, ** = p < 0.05, and *** = p < 0.01.

Estimation Model for Effects of Product Transparency. To test the impact of product transparency on price elasticity, we compared the price elasticities between offline travel agencies, again coded as OFFLINE, and the OTA Hotwire, coded as HOTWIRE. These exhibit different levels of product transparency, but similar levels of price transparency. (See Tables 5 and 6). Although an offline travel agency can provide numerous details about a travel itinerary, Hotwire has an opaque mechanism that does not provide any details about the airline name or the itinerary until after purchase. Offline agencies, in general, have a high level of product transparency, while Hotwire has a low level of product transparency. As we noted, the two agency types exhibit similar levels of price transparency. A traveler who calls a travel agency usually receives price quotes for one or two offers up front. The Hotwire mechanism exhibits a similar level of price transparency, since each search request usually yields one or two price offers. Hotwire does not provide product details until the

26 consumer commits to purchase. Thus, product transparency is likely to drive differences in price elasticity between these agency types. The Product Transparency-Price Elasticity Hypothesis (H2) suggests that product transparency has a negative relationship with price elasticity. To test the hypothesis, we compare price elasticities between OFFLINE and HOTWIRE with a similar econometric model as in the Price Transparency-Price Elasticity model, as follows: ln(QUANTITY ) = CONSTANT + η ln(PRICE) + γ * HOTWIRE * ln(PRICE) + β 1 ln(INCOMEOD) + β 2 ln( ADVPURCH ) + β 3 HUB1 . + β 4 HUB2 + β 5 ln(BLKHRS ) +

∑α

j

(5)

Dj + ε

j

In this model, HOTWIRE is a dummy variable for agency type Hotwire. Therefore, -η is the base price elasticity for OFFLINE and -γ captures the price elasticity of HOTWIRE relative to OFFLINE. The Product Transparency-Price Elasticity Hypothesis can then be expressed as H2: γ < 0. Results. Table 8 provides the results of the regression as specified in Equation (5). Table 8. OFFLINE and HOTWIRE Results: The Price Elasticity Differential

VARIABLES PRICE HOTWIRE*ln(PRICE) CONSTANT

2SLS REGRESSION Robust Coefficient Std. Error -0.67*** -1.01*** 3.14

0.21 0.03 1.94

OLS REGRESSION

Coefficient

Robust Std. Error

-0.45*** -0.90*** -2.70**

0.10 0.02 -1.30

• Control Variables 1.94*** 0.30 2.35*** 0.27 INCOMEOD -1.54*** 0.05 -1.36*** 0.04 ADVPURCH 0.33*** 0.07 0.35*** 0.07 HUB1 -0.18*** 0.06 -0.12** 0.05 HUB2 0.41*** 0.13 -0.11 0.07 BLKHRS R2 (Adj. R2) 83.65% (83.58%) 84.61% (84.54%) Note: N = 1,480. The significance levels for the coefficients are * = p < 0.10, ** = p < 0.05, and *** = p < 0.01.

27 The estimated value η, which represents the base price elasticity of OFFLINE, was negative and significant (η = -0.67, SE = 0.21, p < 0.01). The estimated value of the differential γ between OFFLINE and HOTWIRE was negative and significant (γ = -1.01, SE = 0.03, p < 0.01). Therefore, we find support for H2, the Product Transparency-Price Elasticity Hypothesis, that lower product transparency leads to higher price elasticity of demand. In addition, we note that in both the OLS and the 2SLS regressions the price elasticity of OFFLINE is inelastic with η < 1 , while the demand of HOTWIRE is price-elastic with η + γ > 1 . In summary, we found support for the Price Transparency-Price Elasticity Hypothesis (H1), which suggests that price transparency leads to a higher price elasticity of demand. We also found support the Product Transparency-Price Elasticity Hypothesis (H2), that product transparency leads to lower price elasticity. In the next section, we discuss the implications of these findings and the implications for firm strategy.

6. DISCUSSION Consistent with existing theoretical propositions, we observe that the impact of market transparency on consumers depends on the information disclosed. We found support for the hypothesis that price transparency tends to increase price elasticity, consistent with theories from IS, economics, and marketing literature, and with empirical evidence in Lynch and Ariely (2000). Not surprisingly, many firms are reluctant to develop transparent electronic markets. In addition to the competitive pressure to reduce prices, demand in a more price transparent channel is more elastic. At first glance, our findings suggest that a technologically-enabled increase in potential transparency will lead firms to lose in the battle for consumer surplus. However, our findings suggest that the negative effect of price transparency can be at least partially offset with the positive effect of higher product transparency.

28 Product transparency turned out to have a positive relationship with consumer demand. This is consistent with the theoretical proposition that product information will improve market efficiency and will improve consumer surplus. Rational consumers will be willing to pay more to take advantage of the incremental benefits of improved information, as long as the price premium does not exceed the benefits of reduced search costs. Therefore, firms are in a position to benefit from the positive effect of product transparency on consumer demand. For example, in an environment where technology allows firms to compete with product and price information, a viable strategy alternative is to develop technologies that provide an edge in the provision of product transparency, analogous to the move of U.S. airlines to implement advanced technology to overcome limitations imposed by the legacy GDSs. This strategy can lead to an increase in consumer demand, as Orbitz experienced, when it became one of the top three OTAs in market share just two years after its launch.

7. CONCLUSION To our knowledge, this is the first empirical investigation that tests these theoretical arguments in a real-world B2C setting using historical sales and prices. Our findings complement the experimental research of Lynch and Ariely (2000). Despite the different research design and industry contexts, the results in both studies are consistent: Price transparency has a positive relationship with price elasticity, while product transparency has the opposite effect. With this study, we believe we are contributing to build a body of empirical literature that can be used by practitioners to manage the informational challenges of the Internet revolution. In particular, our results echo the notion that practitioners should strategize in a transparent market environment by designing market mechanisms that counter the effect of price transparency with detailed and valuable product information (Bakos, 1997).

29 The results of the impact of market transparency on price elasticity should be treated with some degree of caution. Despite our best efforts to correctly compare price elasticities across agency types, our analysis does not allow full control of exogenous factors that may affect sales by travel agency, such as advertising budgets or the maturity of the firm. In addition, we could not fully control for the effect of product transparency when testing the impact of price transparency, and vice-versa. We are in the process of designing a more sophisticated research design that will bring greater clarity to our testing with respect to this issue. Nevertheless, the results are consistent with the controlled experiments of Lynch and Ariely, and this adds face validity to our tests and results. Our findings also open new opportunities for research in Internet-based selling and electronic commerce. For example, economic models of firm strategy can be developed to provide normative guidelines for practitioners whose firms operate in the presence of IT-enabled market transparency. In closing, we note that differentiation with product information may be a valuable alternative in addition to traditional product differentiation strategies, because it can offset the negative effect of increases in the availability of price information. With this conclusion, we echo the perspective of Tapscott and Ticoll (2003), who argue in their book, The Naked Corporation: How the Age of Transparency will Revolutionize Business, that a transformation of organizational strategy and the marketplace is in the offing. Firms will make strategic choices relative to how “naked” they wish to be in product and price terms, and ultimately identify the business value and profitability limits of market transparency.

REFERENCES [1]

Akerlof, G. A. “The Market for Lemons: Quality Uncertainty and the Market Mechanism,” Quarterly Journal of Economics, 84(3), 1970, 488-500.

30 [2] [3] [4]

[5] [6] [7]

[8] [9]

[10] [11]

[12] [13] [14] [15]

[16] [17] [18] [19] [20] [21] [22]

Bakos, J. Y. “Reducing Buyer Search Costs: Implications for Electronic Marketplaces,” Management Science, 43(12), 1997, 1676-1692. Breusch, T. and A. Pagan. "A Simple Test for Heteroskedasticity and Random Coefficient Variation." Econometrica, 47, 1979, 1287-1294. Brynjolfsson, E., Hu, Y., and Smith, M. “Consumer Surplus in the Digital Economy: Estimating the Value of Increased Product Variety at Online Bookstores,” Management Science, 49(11), 2003, 1580-1596. Brynjolfsson, E. and Smith, M.D. “Frictionless Commerce? A Comparison of Internet and Conventional Retailers,” Management Science, 46(4), 2000, 563-585. Chircu, A. M. and Kauffman, R. J. “Reintermediation Strategies in Business-to-Business Electronic Commerce,” International Journal of Electronic Commerce, 4(4), 2000, 7-42. Clemons, E. K., and Row, M. C. “Information Technology at Rosenbluth Travel: Competitive Advantage in a Rapidly Growing Service Company,” Journal of Management Information Systems, 8(2), 1991, 53-79. Copeland, D.G., and McKenney, J. L. “Airline Reservation Systems: Lessons from History,” MIS Quarterly, 12(3), 1988, 41-60. Duliba, K. A., Kauffman, R. J., Lucas, H. C. “Appropriating Value from Computerized Reservation Systems Ownership in the Airline Industry,” Organization Science, 12(6), 2001, 702-728. Goldfeld, S. and Quandt, R. “Some Tests for Homoscedasticity,” Journal of the American Statistical Association, 60, 1965, 539-547. Granados, N., Gupta, A., and Kauffman, R.J. “Transparency Strategy in Internet-Based Selling,” in K. Tomak (Ed.) Advances in the Economics of Information Systems, Idea Group Publishing, Harrisburg, PA, 2005a, 80-112. Greene, W. Econometric Analysis, 5th Ed., Prentice Hall, Englewood Cliffs, NJ, 2003. Kaul, A. and Wittink, D.R. “Empirical Generalizations about the Impact of Advertising on Price Sensitivity and Price,” Marketing Science, 14(3), Part 2 of 2, 1995, G151-G160. Kennedy, P. A Guide to Econometrics, 5th Edition, MIT Press, Cambridge, MA, 2003. Lee, A. O. “Airline Reservations Forecasting: Probabilistic and Statistical Models of the Booking Process,” Ph.D. thesis, Flight Transportation Laboratory, Massachusetts Institute of Technology, Cambridge, MA, 1990. Lynch, J. G. Jr. and Ariely, D. “Wine Online: Search Costs Affect Competition on Price, Quality, and Distribution,” Marketing Science, 19(1), 2000, 83-103. McGill, J. I., and van Ryzin, G. J. "Revenue Management: Research Overview and Prospects," Transportation Science, 33, 1999, 233-256. O’Toole, K. “IT Trends Survey 2003,” Airline Business Magazine, August 2003, 40-52. Porter, M. E. “Strategy and the Internet,” Harvard Business Review, 79(1), 2001, 62-78. Riggins, F. J. “A Multi-Channel Model of Separating Equilibrium in the Face of the Digital Divide,” Journal of Management Information Systems, 21(2), 2004, 161-170. Stigler, G. “A Theory of Oligopoly,” Journal of Political Economy, 72(1), 1964, 213-225. Stigler, G. “The Economics of Information,” Journal of Political Economy, 69(3), 1961, 213-225.

31 [23] Talluri, K., and van Ryzin, G. J. The Theory and Practice of Revenue Management. Kluwer Academic Press, Boston, MA, 2004. [24] Tapscott, D. and Ticoll, D. “The Naked Corporation: How the Age of Transparency will Revolutionize Business,” Free Press, New York, NY, 2003. [25] Villas-Boas, J. M. and Winer, R.S. “Endogeneity in Brand Choice Models,” Management Science, 45(10), 1999, 1324-1338. [26] Zettelmeyer, F. “Expanding to the Internet: Pricing and Communications Strategies When Firms Compete on Multiple Channels,” Journal of Marketing Research, 37(3), 2000, 292308. APPENDIX A – Coding of Product Transparency in OTAs We measured product transparency as follows. First, the Orbitz mechanism was evaluated based on availability and accessibility of product information. The other OTAs were scored in relation to an index of 100, which corresponded to Orbitz’s level of product transparency. The measures of availability and accessibility were: a. Availability measure: Average of measures of Airline Name Availability and Itinerary Availability (i.e., origin, destination, number of stopovers, departure and arrival times) - Airline Name Availability: Indexed at 100 if airline name was available for all alternatives displayed in the first screen for both outbound and return flights. - Itinerary Availability: Indexed at 100 if itinerary details were available for all alternatives displayed in the first screen for both outbound and return flights. Example 1: Travelocity did not provide the airline name and itinerary of the return flight upfront, so it had an index of 50 for both airline name availability and itinerary availability, for a total availability index of 50. (See Figure A1 for an illustration of Travelocity’s selling mechanism). Example 2: Hotwire’s opaque mechanism did not provide airline name or itinerary upfront, so both its airline name availability and itinerary availability measure was 0, for a total availability index of 0. (See Figure A2 for an illustration of Hotwire’s selling mechanism). b. Accessibility measure: Average of sort and comparison indices. - Sort: Indexed at 100 if sorting capability was possible by (i) departure time, (ii) arrival time, and (iii) trip duration. - Comparison: Indexed at 100 if comparison tables were provided in the first screen by (i) airline and (ii) number of stopovers. Example 3. Travelocity allows sorting of flight alternatives by departure time, arrival time, and travel time, so its sort index was 100. Travelocity only provided a comparison table by airline, so it had a comparison index of 50, for a total accessibility measure of 75. Hence, the product transparency index of Travelocity was 63, or the average of the availability measure (50) and the accessibility measure (75), for Example 4. Hotwire’s opaque mechanism only provides one or two prices per search request, so the sort and comparison indices did not apply. Therefore, the product transparency measure is based only in the availability index of 0.

32 Figure A1. Travelocity’s Selling Mechanism – First Screen Search Results

Comparison Table Sorting

Departure flights only

Source: www.travelocity.com. Accessed April 2004. Note: Only after selecting a departure flight would the return flight option appear. This mechanism is based on the technological structure of some GDSs, which first search for a departing flight and then a return flight in order to price a ticket.

33 Figure A2. Hotwire’s Selling Mechanism – First Screen Search Results