Essays on Empirical Industrial Organization DISSERTATION

Transcription

1 Essays on Empirical Industrial Organization DISSERTATION Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University By Junqiushi Ren, M.A. Graduate Program in Economics The Ohio State University 2017 Dissertation Committee: Jason R. Blevins, Advisor Maryam Saeedi Bruce A. Weinberg

2 Copyright by Junqiushi Ren 2017

3 Abstract This dissertation consists of three essays. The first essay studies an online platform, namely, Groupon, and the second and third essays explore the US airline industry. All of the essays are in the field of empirical IO. The first essay, The Role of Reputation in Daily Deal Markets: The Case of Groupon, addresses the question of whether and how business reputation alters the sales and promotion results of daily deals. Daily deals are online discount vouchers offered by local businesses and advertised on deal sites. They have become very popular in recent years with the popularity of Internet and mobile apps. To study the role of reputation in daily deal markets, I scrape daily deals in 17 large cities from Groupon (the biggest daily deal site), which is then matched with a demographic dataset extracted from Year American Community Survey. I first use the restaurant category as a baseline, and show that reputation, measured by the percentage of positive reviews on Groupon, is positively associated with the sales of vouchers. I then include reputation from external platforms, namely, the average star ratings from Yelp and Google, and find that although these ratings are positively associated with voucher sales, such relationship is much weaker compared to that with Groupon ratings. Subsequently, I use the number of Yelp reviews that mentions the keyword Groupon as a proxy of customer flows that were brought in ii

4 by Groupon directly, and show that reputation is positively associated with the promotion results of daily deals. Finally, I extend the analysis to categories other than restaurants. Results show that business reputation is positively related to the sales of vouchers in all categories, though the extent varies. In particular, the relationship is relatively strong for restaurants and health, and relatively weak for beauty and entertainment. The second essay, A Re-examination of Southwest s Entry, studies how Southwest Airlines, the biggest low-cost carrier in the US, affects legacy carriers pricing. While much of the literature on low-cost carriers focus on nonstop markets, this paper extends this body of scholarship to connecting markets, and find that the entry of Southwest imposes a substantial downward pressure on the prices in both nonstop and connecting markets. Moreover, I find that the price drops are much larger in connecting markets than in nonstop markets. My findings also provide evidence that connecting and nonstop markets are relevant to each other, by showing that a connecting entry affects nonstop prices to a considerable extent, and a nonstop entry significantly affects connecting prices as well. The third essay, Profits and Entry Decisions: The Effect of Southwest Airlines, looks at how Southwest affects other airlines profits and entry decisions. As shown in my second essay, Southwest can lead to significant drops in airfares. Meanwhile, the fare drops are accompanied with dramatic increases in passenger traffic. As a result, it is not clear of whether and how Southwest influences other airlines profits and entry decisions. In this essay, I estimate a static game of simultaneous entry closely following Ciliberto and Tamer (2009), and demonstrate that Southwest has a very remarkable, negative impact on the iii

5 profit functions of other carriers. Subsequently, I break down airlines entries into nonstop entries and connecting entries, and find that an airline s entry decision is determined by its own and each competitor s presence in both nonstop and connecting markets. Hence, I also confirm that connecting and nonstop markets are related to each other. Finally, by conducting counterfactual experiments, I show that Southwest has a significant influence on the entry decisions of other airlines and the equilibrium number of non-southwest carriers in a given market. iv

6 Dedication This dissertation work is dedicated to my parents, Youxue Ren and Hong Liu, who have always loved me unconditionally and whose great examples have taught me to buckle down for the things that I try to accomplish. This work is also dedicated to my fiancé, Yang Shi, who has been a steady source of support and inspiration during the doctoral level studies and life. v

7 Acknowledgments I am extremely thankful to my advisor, Jason Blevins, for not just offering me the freedom to investigate the topics of research and urge me to build up my own capacity as a researcher, but also giving me numerous supportive recommendations. I benefited a lot from the thoughtful talks we had at the crucial stage of my job market paper. Without his steady support and valuable advice, I could not have finished this work. I am also profoundly indebted to Maryam Saeedi for her valuable insights, suggestions, and patience. She was my advisor in the third and fourth year, and it was such a great chance to discuss my work with her. Through her guidance, I turned out to be better in thinking and writing. My heartfelt thanks also go to Bruce Weinberg for his support, advice, and time. He not only helped me to make my essays more concrete, but also gave me a lot of good suggestions to make my job talk better. I would also like to thank Javier Donna for teaching me and building up my enthusiasm for empirical IO. I also own thanks to Kurt Lavetti, who served as the fourth member of my candidacy committee. What s more, I am grateful to Daeho Kim, Lucia Dunn, and Stephen Cosslett, who gave me a lot of good advice during my graduate studies. vi

8 I would extend my gratitude to the Department for providing a decent environment of researching. I want to express gratitude toward Hajime Miyazaki for having my best interests as the top priority whenever he gave me his remarks. John-David Slaughter helped me a lot with technical issues. I am also thankful to Rick Tobin, Miroslava Marshall, and Ana Ramirez for their extraordinary administrative work. vii

9 Vita B.A. Economics, B.S. Mathematics, Renmin University of China M.A. Economics, The Ohio State University 2013-present... Graduate Teaching Associate, Department of Economics, The Ohio State University Fields of Study Major Field: Economics viii

10 Table of Contents Abstract... ii Dedication...v Acknowledgments... vi Vita... viii List of Tables... xii List of Figures... xiv Chapter 1: The Role of Reputation in Daily Deal Markets: The Case of Groupon Introduction Background About Groupon About Yelp About Google Reviews Data and Empirical Methodology Data Collection Estimation Specification Baseline Results Consumer Traffic...15 ix

11 1.5.1 A Descriptive Analysis A Regression Analysis Other Categories Conclusions and Discussions...20 Chapter 2: A Re-examination of Southwest s Entry Introduction Data Data Description Market Definition Data Aggregation Market Selection Empirical Model Empirical Results: Fare Changes Compare with Previous Literature Empirical Results: Number of Passengers Conclusions...42 Chapter 3: Profits and Entry Decisions: The Effect of Southwest Airlines Introduction Data Data Construction Market Definition Data Aggregation...50 x

12 3.2.4 Market Selection Models and Game Estimation Empirical Models Variable Definitions The Game Error Terms Simulation Identification How Does Southwest Influence Profits? Empirical Results: Aggregate Across Service Types Empirical Results: Distinguish Service Types Compare Different Research Settings Counterfactual How Does Southwest Influence the Entry Decisions of Each Airline? How Does Southwest Influence the Number of Non-Southwest Firms? Conclusions...73 References...75 Appendix A: Additional Figures and Tables for Chapter Appendix B: Additional Tables for Chapter Appendix C: Cities with Multiple Airports for Chapter Appendix D: Additional Tables for Chapter Appendix E: Hubs and Focus Airports for Chapter xi

13 List of Tables Table 1.1: Basic Results...12 Table 1.2: Results by Regular Price Ranges...14 Table 1.3: Consumer Traffic...18 Table 1.4: Beauty, Entertainment, and Health...20 Table 3.1: Profit Estimations: Aggregate Across Service Types...60 Table 3.2: Competitive Effects: Distinguish Service Types...64 Table 3.3: Competitive Effects: Compare Different Research Settings...67 Table 3.4: Changes in the Mean Upper Bounds when Southwest is Counterfactually Removed...70 Table 3.5: Mean Number of Non-Southwest Firms...72 Table 3.6: Comparison: Mean Number of Firms...72 Table A.1: Summary Statistics...83 Table B.1: Operating Costs for Carriers of Interest...84 Table B.2: Descriptive Statistics...84 Table B.3: Legacy Carriers Respond to Entry: Mean Fare...85 xii

14 Table B.4: Nonstop Price Changes: Previous Literature...88 Table B.5: Legacy Carriers Respond to Entry: Mean Number of Passengers...89 Table B.6: Robustness Check with Alternate Definition of Market: Legacy Carriers Respond to Entry: Mean Fare...92 Table B.7: Robustness Check with Alternate Data Aggregation: Legacy Carriers Respond to Entry: Mean Fare...95 Table D.1: Summary Statistics xiii

15 List of Figures Figure 1.1: The Average Number of Reviews per Merchant per Month Before and After Their Groupon Offers...16 Figure 2.1: Price Changes Surrounding Southwest s Entry...34 Figure 2.2: Nonstop Price Changes: Compare with Previous Literature...38 Figure 2.3: Passenger Traffic Changes Surrounding Southwest s Entry...41 Figure A.1: A Sample Deal on Groupon s Webpage...81 Figure A.2: Popular Categories Consumers Purchased on Groupon in Figure A.3: Popular Categories Daily Deal Sites Offered in xiv

16 Chapter 1: The Role of Reputation in Daily Deal Markets: The Case of Groupon 1.1 Introduction Daily deals are online discount vouchers that are typically offered by local businesses and advertised on deal sites. Popular categories include restaurants, beauty salons, movie theaters, and health workshops. Each business can design its own coupon and then post on daily deal sites, such as Groupon, LivingSocial, etc.. Coupons are usually available for sale for 24 to 36 hours. Consumers could purchase these vouchers via computers or mobile apps, and then normally have 120 to 360 days to redeem at the businesses. From vendors perspective, the primary goal of offering daily deals is to conduct online promotions at a low cost. From consumers perspective, the main benefit of purchasing daily deals is to enjoy goods/services at a deeply discounted rate. While daily deals share many common functions with traditional promotional tools such as newspaper coupons, mailers, or yellow pages regarding offering discounts and attracting new consumers, they differ from traditional promotional tools in at least three important aspects. First, daily deal sites often encourage customers to leave feedback and display their reviews and ratings prominently on deals webpages. The information provided in peer reviews and ratings help potential buyers to update their beliefs about 1

17 product quality and alter their propensity to buy. Second, daily deals have been associated with deep discounts that are rarely found elsewhere. Daily deal sites usually recommend businesses to sell coupons for half price, and sometimes offer up to 60 to 70 percent discounts off regular prices. Last but not least, it is very easy for consumers to compare similar deals or businesses on daily deal sites, because typically there are many similar businesses offering coupons at the same time, and the search costs on the Internet are very low. In other words, promoting on these sites can be very competitive. With the popularization of computers and mobile apps, daily deal markets have been growing exponentially. According to IBISWorld 1, the current U.S. daily deal industry generates over $4 billion revenue a year, and Groupon, the leader in this industry, owns over 60% of industry revenue. However, although daily deals appear to be a new and promising format of online promotions, there is very limited research to examine their underlying economic mechanism. Edelman, Jaffe, and Kominers (2011) tried to explain the mechanism using price discrimination. Dholakia (2011) did a survey to study businesses willingness to offer daily deals again. Byers et al. (2012) found that voucher sales are significantly correlated with soft incentives, such as the time of listing, the position of a deal on the web page, and so on. Luo et al. (2014) found that consumers purchase likelihood is affected by deal popularity. Li and Wu (2014) found that deal sales are simultaneously driven by a herding effect and a word-of-mouth effect. 1 Last Accessed: October 20,

18 Among these works, few of them take consumers ratings into consideration. Nevertheless, there are a couple of reasons why including ratings are important. First, ratings and reviews are typically prominently displayed on deals webpages. So they are among the first things that consumers will see whenever they browse a deal. As such, one can expect that they will modify consumers purchase decisions. Second, ratings from previous customers have been proved to affect sales in many other markets, especially the markets for experience goods. 2 Scholars have identified that reputation often serves as a useful tool to correct the problem of asymmetric information in those markets. 3 Therefore, it is reasonable to propose that ratings also make a difference in daily deal markets. My objective in this paper is to study the role of reputation in the daily deal industry, with Groupon as an example. I address whether and how business reputation, measured by the percentage of positive reviews or the average star rating, affects the sales and promotion results in this industry. To the best of my knowledge, my paper is the first empirical work to solve this problem. Hence, it not only adds to the literature on daily deals, but also extends prior work on reputation and e-commerce by exploring a new market, namely, the daily deal market. I first use the restaurant category, the most popular category on Groupon, as a baseline. Based on a cross-sectional dataset consisting of 3380 restaurants in Groupon marketplace, I first find that reputation is positively associated with sales of vouchers, which is reflective of the demand for daily deals. Such relationship is defined as the 2 See, for example, Bickart and Schindler (2001), Chevalier and Mayzlin (2006), Zhu and Zhang (2010), Moe and Trusov (2011), Anderson and Magruder (2012), Mayzlin et al. (2014), and Luca (2016). 3 See, for example, Bolton et al. (2004), Brown and Morgan (2006), and Saeedi (2014). 3

19 reputation effect. Ceteris paribus, on average, if the percentage of positive reviews increases by 1, sales will increase by about 1.5%. When I classify the restaurants by their normal price ranges, I find that the reputation effect is larger for expensive restaurants and smaller for cheap restaurants. This indicates that consumers pay more attention to ratings when they buy expensive restaurants vouchers, but do not care that much when they buy cheap restaurants vouchers. The reason is that expensive restaurants vouchers usually have a high face value and a high price; when buyers purchase these vouchers, they face the risk of losing a large sum of money. Such risk results in consumers paying more attention to previous ratings. Noticing that Groupon is not the only source where potential consumers can obtain reviews and ratings, I then include the reputation from external platforms, namely, the average star ratings from Yelp and Google. The results show that these external ratings are also positively associated with sales of vouchers. However, such relationships are much weaker compared with Groupon ratings. Subsequently, I discuss how reputation modifies the promotion result of daily deals. While restaurant revenue data or sales data is the ideal source to measure promotion results, such data is not publicly available. Instead, I consider a novel, alternative approach to capture the promotion result qualitatively. The idea is to use the number of reviews on Yelp as a proxy of customer flows. That being said, a large number of reviews on Yelp imply that the business has a lot of customers, while a small number of reviews imply that the business is not very popular. 4

20 I first start with a descriptive analysis. I find that, on average, the monthly number of Yelp reviews immediately rises following Groupon offers, and then gradually drops to the initial level. Based on the proportion of Yelp reviews that contain the word Groupon, Groupon deals appear to be the main driver behind this rise. Next, I use the number of Yelp reviews that mentions the keyword Groupon as a proxy of customer flows that are brought in by Groupon directly. Regression results show that reputation is positively related to the number of reviews noting Groupon. This means that, ceteris paribus, when a good reputation restaurant and a poor reputation one post very similar deals on Groupon, these deals will generate different results, with significantly more customers being attracted to the good reputation restaurant. Finally, I extend the analysis to categories other than restaurants. The results show that reputation is positively related to the sales of vouchers in all categories, but the extent varies. In particular, the relationship is relatively strong for restaurants and health services, and relatively weak for beauty services and entertainment. The rest of the article is composed as follows: Section 1.2 briefly introduces the background knowledge of Groupon, as well as the rating systems of Yelp and Google; Section 1.3 describes the data and empirical methodology; I discuss the relationship between reputation and coupon sales in Section 1.4, and discuss the relationship between reputation and consumer traffic in Section 1.5; I then extend the analysis to other categories in Section 1.6; Conclusions are offered in Section

21 1.2 Background About Groupon To study the role of reputation in daily deal markets, I choose Groupon, which is the biggest daily deal website in the US, as my research setting. Groupon was launched in Chicago in November As of 2014, Groupon is available in 47 countries worldwide, and have over 200 million subscribers. 4 Groupon encourages businesses to offer deep discounts. A typical Groupon deal can be a $80 Italian meal for $40. When a customer buys this deal, the Italian restaurant will take approximately $20 and Groupon will take approximately $20. Consumers can browse and purchase deals on Groupon s website and mobile app. Figure A.1 in Appendix A is a typical Groupon deal that consumers could see. For each deal, the following information is available: a brief description of the deal, the vendor s information, the face value of the voucher, the percentage of discount, the discount amount, and the discounted price. Besides, the percentage of positive reviews and the number of reviews are prominently displayed, allowing shoppers to see vendors reputation easily. Groupon s products cover a wide range from clothes and electronic devices to restaurant meals and Yoga classes. In this paper, I mainly focus on deals offered in the restaurant category. I select the restaurant category because it is not only the most popular category customers tend to purchase on Groupon, but also the most popular category daily deal sites are offering deals, as Figure A.2 and A.3 in Appendix A show. In the last part of 4 9DCF-F125D369ED28/Groupon_Q2_2014_Public_Fact_Summary.pdf. Last Accessed: October 20,

22 Section 1.6, I extend the discussion to other categories, namely, beauty, entertainment, and health About Yelp Yelp is a popular review platform which allows users to browse and write reviews about all kinds of local businesses. Founded in 2004, Yelp has 135 million monthly visitors and 71 million reviews as of Whenever a user leaves a review, she must first assign a star rating from 1 to 5 stars in whole-star increments. Yelp will automatically aggregate all reviews for a specific business, and then displays the average star rating in half-star increments. In this process, Yelp always rounds off to the nearest half-star. For instance, an average rating of 3.11-star will be rounded to 3-star, and an average rating of 3.41-star will be rounded to 3.5-star. People can check reviews and ratings from Yelp s website and mobile app; businesses can not pay to modify reviews. Although only the average star rating is displayed prominently, one can easily find other details about ratings, such as the distribution of ratings, the date of each review, the text of each review, etc.. Yelp also displays other information about the business, including the location, zip code, operating hours, and regular price ranges. This information is very helpful in my subsequent analyses. 5 Last Accessed: October 20,

23 1.2.3 About Google Reviews Besides Yelp, Google is another popular platform where users can read and write reviews about different kinds of businesses. Google reviews are easily found if one searches on Google. The map of Google often comes with reviews as well. Similar to Yelp, Google also allows users to leave a rating from 1 to 5 stars in whole-star increments, and then computes and displays the average star rating. However, while Yelp always rounds the average rating to the nearest half-star, Google does not do that. I am interested in Google reviews because they are very easy to obtain for consumers. Whenever a potential consumer searches a business on Google, she can easily find the ratings, and thus may influence her purchase decision of daily deals. 1.3 Data and Empirical Methodology Data Collection All data in my study are collected from public sources. Specifically, data about deal characteristics, the percentage of positive reviews and the sales of vouchers are extracted directly from Groupon website by using R. Data about community facts comes from the Year American Community Survey, which is publicly available on the U.S. Census Bureau website. Besides, I also gather ratings and reviews from Yelp and Google. I sample 17 cities in the US, including Atlanta, Boston, Chicago, Dallas, Detroit, Houston, Las Vegas, Los Angeles, Miami, New Orleans, New York, Orlando, Philadelphia, San Diego, San Francisco, Seattle, and Washington DC. The cities are selected in the same 8

24 way as Byers et al. (2012) suggested. The basic selection criteria includes city population and geographic distribution. My dataset consists of deals listed on Groupon for the sample cities as of July Accordingly, the dataset includes deals. For each deal, I collect the quantity sold, the original face value, the percentage of discount, the discounted price, a brief description of the deal, vendor s business name, cuisine type, and zip code. I also collect how many consumers have left reviews, and the percentage of positive reviews. I notice that rating is unreliable without enough number of reviews; therefore, I drop the observations that have few reviews, and only keep those that have at least 10 reviews. This results in a sample of deals, among which 5688 are from the restaurant category, 2932 are from the entertainment category, 2476 are from the beauty category, 2092 are from the health category, and 4505 are from other categories such as pets, home service, shopping, etc. I also extract demographic facts from Year American Community Survey. For each zip code, the survey has the demographic information in that zip code area, including population density, sex ratio, median household income, and educational attainment. For the restaurant deals, I further gather business information from Yelp and Google, which includes the business name, zip code, regular price range, the number of reviews, and the average star rating (up to 5). Again, I drop the observations that have less than 10 reviews. For each Yelp review, I then collect the date it was written, the reviewer ID, and the text of the review. The Yelp and Google ratings are merged the Groupon dataset using the business name along with the zip code, which results in 3380 matched restaurants. 9

25 The final dataset for the restaurant category is a cross-sectional data which contains demographic information, Yelp ratings, Google ratings and all details about the Groupon deal. The final dataset for beauty, health, and entertainment is a cross-sectional data which contains demographic information and all details about the Groupon deal. Table A.1 in Appendix A presents summary statistics Estimation Specification I am interested in estimating the reputation effect in terms of moderating voucher sales and voucher promotion results. The key regression specification is given as: = β T 4 demo + β T 5 type + β T 6 city (1.1) where is the outcome variable, such as the logged sales of vouchers or a proxy of customer flows; is consumers ratings (Groupon displays the percentage of positive reviews; Yelp and Google display the average star rating with a maximum of 5 stars); the coefficient captures the relationship between reputation and ; is the logged price of a voucher; is the logged discount amount, which captures the gap between the face value and the discounted price, and is the amount that a consumer could save by purchasing the voucher; demo is a vector that captures demographic information in the zip code area of the business, including population density, sex ratio, median household income, and educational attainment; type is a dummy-coded vector representing the type of the deal in a given category: for restaurant deals, the types refer to the cuisine types, which include American/European/Latin American/ Asian/others; for 10

26 beauty deals, the types include hair salon/nail salon/spa/massage/others; for entertainment deals, the types include movie/theatre show/sightseeing/diy/others; for health deals, the types include yoga/weight loss/dancing/boxing/others. city is a dummy-coded vector representing the city. is a dummy that equals one if the deal is sold out. is a dummy that equals one if the deal is available on Groupon for more than one week. Lastly, is an error term that represents any additional variations in the outcome variable. 1.4 Baseline Results I begin by applying the key regression (1.1) to the restaurant sample which contains 5688 restaurant deals in the Groupon marketplace. Column (1) of Table 1.1 presents the results. I find that the price is negatively related to voucher sales, with a price elasticity of This implies an elastic demand. The discount amount is positively associated with sales. Holding all else constant, a one percent increase in discount amount is associated with a 0.673% increase in sales. In addition, the percentage of positive reviews is positively associated sales. All else equal, if the percentage of positive reviews increases by 1, sales will increase by about 1.4%. 11

27 Table 1.1: Basic Results (1) (2) Dependent variable: ( ) ( ) Sales of vouchers Sales of vouchers ( ) *** *** (0.082) (0.108) ( ) 0.673*** 0.722*** (0.074) (0.092) Pct. Positive Reviews on Groupon 0.014*** (0.002) Rescaled Groupon Rating 0.789*** (0.120) Avg. Rating on Yelp 0.108* (0.047) Avg. Rating on Google 0.202** (0.062) R-squared # of observations Notes: Coefficients are estimated from regression model (1.1). The numbers in the parentheses are standard errors. *** Significant at the 1% level. ** Significant at the 5% level. * Significant at the 10% level. Noticing that Groupon is not the only source where potential consumers can obtain reviews and ratings, I then include the reputation from third-party platforms, namely, the ratings from Yelp and Google. While Groupon displays the percentage of positive reviews, both Yelp and Google display the average star ratings with a maximum of 5 stars. In order to compare with Yelp and Google, I rescale the Groupon rating by dividing the percentage of positive reviews by 20. Thus, a restaurant with 100% positive reviews receives a score of 5, and a restaurant with 80% positive reviews receives a score of 4, etc. I then re-run the 12

28 key regression (1.1) on the matched sample, where the variable is now a vector that contains the rescaled Groupon ratings, the average star ratings on Yelp, and the average star ratings on Google. Column (2) of Table 1.1 presents the results. I find that all the three ratings are positively associated with voucher sales, but to different extents. In particular, an additional star of Groupon rating is associated with a 78.9% increase in sales, an additional star of Yelp rating is associated with a 10.8% increase in sales, and an additional star of Google rating is associated with a 20.2% increase in sales. These results indicate that consumers put more weights on Groupon ratings. This is likely because Groupon ratings are already on the deal page, and thus they are the easiest to obtain; Yelp ratings and Google ratings, on the other hand, need extra clicks. Another reason is that Groupon ratings are solely from previous Groupon buyers, and thus they are more valuable to potential Groupon buyers than ratings from third-party platforms where most reviews are from non-groupon users. Next, I apply the key regression (1.1) to a few subsamples. I divide restaurants into groups based on their regular price ranges. Cheap, moderate and expensive corresponds to restaurants whose regular price per person (without any coupons) runs from $1 to $10, from $10 to $30, and over $30, respectively. 6 Table 1.2 reports the estimates. In all cases, the results indicate that prices are negatively related to sales, while discount rates and the percentages of positive reviews are positively associated with sales. When comparing the coefficient of log( ), I find that the demand for cheap restaurants vouchers is the most price elastic, and the demand for expensive restaurants vouchers is the least price elastic. 6 Recall that the regular price information is obtained from Yelp. 13

29 Comparing the coefficient of reputation, I find that the reputation effect is the smallest for cheap restaurants vouchers, and the largest for expensive restaurants vouchers. In particular, if the percentage of positive reviews increases by 1, voucher sales will increase by 0.2% for cheap restaurants, 1.4% for moderate-price restaurants, and 2.6% for expensive restaurants. The reason is that expensive restaurants typically offer high-value deals. When buyers purchase such deals, they face the risk of losing a large sum of money, which results in consumers paying closer attention to business reputation. Table 1.2: Results by Regular Price Ranges (1) (2) (3) Dependent variable: ( ) ( ) ( ) Sales of vouchers Sales of vouchers Sales of vouchers Cheap Moderate Expensive ( ) *** *** *** (0.192) (0.097) (0.253) ( ) 1.042*** 0.577*** 1.073*** (0.170) (0.083) (0.260) Pct. Positive Reviews from Groupon (0.009) 0.014*** (0.002) 0.026*** (0.006) R-squared # of observations Notes: Coefficients are estimated from regression model (1.1) using different subsamples. The numbers in the parentheses are standard errors. Cheap, moderate, and expensive corresponds to restaurants whose regular price per person (without any coupons) runs from $1 to $10, from $10 to $30, and over $30, respectively. *** Significant at the 1% level. ** Significant at the 5% level. * Significant at the 10% level. 14

30 In short, my baseline results demonstrate the reputation effect on sales. Consumers form demand based on ratings. A good rating implies that the vendor has good food and service, which lures purchase. Although there are several third-party platforms that provide external ratings, consumers are more rely on the peer reviews directly from Groupon. Last but not least, the reputation effect varies with restaurants regular price ranges, with expensive restaurants having larger effects than cheap restaurants. 1.5 Consumer Traffic The ultimate goal of daily deals is to bring customers to the businesses that offer the deals. According to Groupon s survey in 2014, 93% of merchants agree that they benefited from Groupon deals and had many new customers brought in by vouchers. However, do these merchants receive the same level of benefits? More specifically, does reputation play a role in determining how many customers to bring in? When a poorreputation merchant and a good-reputation merchant post similar deals on Groupon, will the good reputation one be able to attract more customers? The discussion in Section 1.4 concludes that a good reputation helps to increase voucher sales, which implies that it may also help to entice customers to the business. In this section, I am going to discuss this issue in a more straightforward way. 15

31 1.5.1 A Descriptive Analysis To see if reputation has an effect on bringing customers in, the ideal way is to use restaurant revenue data, or restaurant sales data. However, such data is not publicly available. Instead, I consider an alternative approach to examine the promotion result qualitatively. The idea is to use the number of reviews on Yelp as a proxy of customer flows. The underlying assumption is that receiving more reviews means more people have visited the restaurant. Figure 1.1 depicts the average review volume across all restaurants in my matched sample. Figure 1.1: The Average Number of Reviews per Merchant per Month Before and After Their Groupon Offers 16

32 In Figure 1.1, the x-axis represents the time from the Groupon offer, and the y-axis represents the average monthly number of reviews. The blue curve represents the total number of Yelp reviews for a given business. The orange curve represents the number of reviews that contain the keyword Groupon. From this figure, I find that the trend of the average monthly number of reviews appears to be relatively flat prior to the Groupon offer. Once the Groupon offer is introduced, the number of reviews significantly rises. Based on the proportion of reviews that contain the keyword Groupon, Groupon deals appear to be the main driver behind this rise. After four months since the Groupon offer, the average number of reviews drops to the original level. Starts from the fifth month, the trend appears to be flat again. Thus, I conclude that the Groupon effect is significant within the first four months. This is not surprising, since most deals expire after 120 days of purchase. In short, the above descriptive analysis indicates that Groupon offers could result in a significant rise in the monthly number of Yelp reviews. This rise is considerable within the first four months since the Groupon offer, but is negligible since the 5th month A Regression Analysis Although Groupon offers appear to be the proximate cause for the rise in the monthly number of Yelp reviews, there are other factors that may also lead to the rise. Thus, it is important to study Groupon s promotion result in a more direct way. While the number of reviews on Yelp may serve as a proxy of customer flows, the number of reviews on Yelp 17

33 that mentions the keyword Groupon may be a proxy of customer flows that were brought in by Groupon directly. Therefore, if I can show that reputation has an influence on the number of Yelp reviews noting Groupon, I should be able to throw some light on the reputation effect in terms of moderating promotion results. I apply key regression (1.1) to matched restaurant sample, where the dependent variable is the number of reviews on Yelp that mentions the word Groupon within the first four months since the Groupon offer. Table 1.3 reports the results. Table 1.3: Consumer Traffic Dependent Variable: the number of reviews on Yelp that mentions the word Groupon (1) (2) (3) (4) OLS Poisson OLS Poisson Pct. Positive Reviews from Groupon Rescaled Groupon Rating Avg. Rating from Google 0.009* (0.005) 0.011* (0.006) 0.302* (0.164) (0.117) 0.266* (0.139) (0.128) R-squared/Pseudo R # of observations Notes: Coefficients are estimated from regression model (1.1). I use the number of reviews on Yelp that mentions the word Groupon as a proxy of customer flows that are brought in by Groupon directly. The numbers in the parentheses are standard errors. *** Significant at the 1% level. ** Significant at the 5% level. * Significant at the 10% level. 18

34 In columns (1) and (2), I measure reputation solely on the percentage of positive reviews from Groupon. Both OLS and Poisson regressions show that reputation is positively associated with the number of reviews noting Groupon. In columns (3) and (4), reputation is measured as a vector which includes the rescaled Groupon rating and the average star rating from Google. The results show that although both ratings are positively associated with consumer traffic, ratings from Groupon have a more significant impact compared to ratings from Google. 1.6 Other Categories In this section, I extend my baseline approach to categories other than restaurants. In particular, I examine beauty deals (e.g., hair salons, spas, massages), entertainment deals (e.g., movie tickets, theater shows, sightseeing), and health deals (e.g., yoga classes, boxing, weight loss classes). Table 1.4 reports the results. After comparing the coefficient of log ( ), the estimates indicate that Beauty and Restaurant have elastic demand, while Entertainment and Health have inelastic demand. When comparing the coefficient of reputation, I find that in all four categories, reputation is positively associated with the sales of vouchers, though the extent varies. Specifically, the relationship is relatively strong for restaurants and health, and relatively weak for beauty and entertainment. The reason is that, in practice, the deals from restaurants and health workshops are more homogeneous than the deals from beauty and entertainment. For example, while restaurants often sell deals in the format of $11.50 for $20 worth of food and customers can choose anything from the menu, nail salons normally sell detailed 19

35 deals, such as $28 for one shellac manicure, $35 for one gel manicure, etc.. In the latter case, shoppers may make purchase decisions more based on the characteristics of each deal, and thus ratings are less valuable. Table 1.4: Beauty, Entertainment, and Health Dependent variable: ( ) (Sales of vouchers) ( ) ( ) Pct. Positive Reviews from Groupon (1) (2) (3) (4) Beauty Entertainment Health Restaurant *** (0.184) 0.777*** (0.118) 0.009** (0.003) ** (0.228) (0.215) 0.007** (0.003) ** (0.168) (0.126) 0.016*** (0.003) *** (0.082) 0.673*** (0.074) 0.014*** (0.002) R-squared # of Observations Notes: Coefficients are estimated from applying model (1.1) to different categories. The numbers in the parentheses are standard errors. *** Significant at the 1% level. ** Significant at the 5% level. * Significant at the 10% level. 1.7 Conclusions and Discussions In this work, I study the role of reputation in daily deal markets. I address whether and how reputation, measured by either the percentage of positive reviews or the average ratings, influences the sales and promotion results in these markets. For the largest daily deal site Groupon, focusing on the most popular daily deal category restaurants, I show that the value of reputation is positive: reputation is positively related to the sales of 20

36 vouchers, which is reflective of the demand for daily deals; moreover, reputation is positively related to the promotion results of daily deals. When extending the analysis to other categories, I find that reputation is still positively associated with voucher sales, but the extent varies across categories. In particular, for restaurants and health, reputation is strongly associated with sales, while for beauty and entertainment, the relationship is weaker. To enable readers to interpret the results accurately, it is necessary to recognize the limitations of this study. First, although examining the number of reviews can throw some light on promotion results, it is not accurate. Some customers may not leave reviews even if they visit the businesses. Similarly, for those who leave reviews, it is not necessarily that their review text will contain the word Groupon even if they are brought in by Groupon deals. Therefore, it is still interesting to use other data, such as restaurant revenue data, when available, to infer the promotion results of daily deals. Second, it is worth trying to use panel data to re-do the process presented here, so that the unobserved time-invariant heterogeneity can be controlled. For instance, instead of taking the cumulative sales of vouchers, one can monitor the daily sales of vouchers in each day when the deal is available, and examine how reputation alters daily sales. While my work here has limitations and is preliminary, I believe it still sheds light on several crucial questions in modeling and data mining daily deal sites and similar e- commerce platforms, and shows the importance of reputation in daily deal markets. 21

37 Chapter 2: A Re-examination of Southwest s Entry 2.1 Introduction Southwest Airlines is the biggest low-cost carrier (LLC) by far, worldwide. Over the past few decades, its fast expansion has become the foremost thrust behind the development of LCCs, the drops in airfares, and the structural changes in the US airline industry. Unlike legacy carriers, whose structures are based on the connections among hubs and spokes (the so-called hub-and-spoke structure), Southwest utilizes a point-to-point structure and uses only the Boeing 737 aircraft 7. This allows Southwest to operate at low costs and, at the same time, implement considerable flexibility in the routes it flies. In fact, the Form 10-K financial reports have shown that Southwest has substantially lower average cost compared to legacy carriers. For example, in 2014, the operating cost per ASM (available seat mile) for Southwest was $0.125, while for major legacy airlines it was, at least, $ Table B.1 in Appendix B provides a breakdown of operating costs on an ASM basis for major carriers and Southwest. Southwest s low-cost strategy has proved to be extremely successful. It is not surprising that Southwest has attracted a growing amount of empirical attention within the 7 Southwest Airlines Details and Fleet History Planespotters.net Just Aviation. Available at: Last Accessed: October 8,

38 industrial organization literature. A lot of research has been done to prove its dramatic downward pressure on airfares, including Dresner et al. (1996), Morrison (2001), and Goolsbee and Syverson (2008). However, most existing studies on Southwest only examine nonstop flights. Economists have studied how the fares for the incumbents nonstop flights change when Southwest starts, or threatens to start, operating nonstop flights in the same market. But the airline industry is more than nonstop flights. Connecting flights comprise a majority of all air travel. For example, according to the 2010 Airline Origin and Destination Survey (DB1B) data, the shares of domestic passengers that traveled with nonstop service, onestop service, and two-stop service are 39%, 57%, and 4%, respectively. The share of passengers traveled with connecting service is even higher on less popular routes. For example, on routes with less than 25,000 passengers per year, the share of nonstop travelers and one-stop travelers are, respectively, 27% and 70%. Moreover, connecting trips differ from nonstop trips in a variety of aspects, so it is inappropriate to use nonstop trips to represent the entire air travel market. For instance, from consumers perspective, connecting trips are less convenient and usually cheaper than nonstop trips. From airlines perspective, on the other hand, connecting trips are less costly and more flexible. It should be noted that although Southwest is famous for the art of point-to-point (P2P) routing, it provides connecting services as well, only not to the same degree as traditional legacy carriers, like Delta and American. For instance, in 2010, Southwest provided connecting service on 1223 domestic routes, and provided nonstop service on

39 routes, while for Delta the numbers of routes were 1625 and 610 for connecting and nonstop service respectively. 8 Given that connecting trips are very common, yet are different from nonstop trips, it is worthwhile to treat nonstop and connecting services as different products on the same routes, and combine them into a single research. By doing this, my paper generates very comprehensive findings. The main objective of this paper is to study how Southwest s entry influences legacy carriers pricing in both nonstop and connecting markets. In doing, I use a panel dataset that covers air travel from 1993 to 2014, and regress prices on a set of time dummies surrounding Southwest s entry into a market. My primary results indicate that when Southwest enters a route with connecting service, legacy carriers on that route, on average, will cut nonstop prices by 4.1%, and cut connecting prices by 8.1%; on the other hand, when Southwest enters a route with nonstop service, legacy carriers will reduce nonstop and connecting prices by 5% and 8.8%, respectively. These results reveal three important conclusions. First, legacy carriers cut both connecting and nonstop prices upon Southwest s entry, no matter it is a connecting entry or a nonstop entry. Second, Southwest s nonstop entry leads to more significant fare drops compared to its connecting entry. The reason is that nonstop entry creates more threats than connecting entry. Third, legacy carriers cut connecting fares more than they do with 8 The numbers of routes are identified from the 2010 DB1B data. 24

40 nonstop fares. This is rational behavior. Due to the inconvenience of connecting flights, legacy carriers perceive the need to cut their prices more to remain attractive to customers. In addition, I find that in all cases, there is significant preemptive price cutting. For Southwest s nonstop entry, legacy carriers start to reduce prices approximately six quarters prior to the actual entry. For Southwest s connecting entry, preemptive price cutting does not begin until the quarter before entry. The reason why legacy carriers start to reduce prices earlier in response to nonstop entry may be because nonstop entry is usually preannounced in most cases, Southwest announces its new nonstop routes for at least one year, and sometimes the announcement could be as early as five years ahead. Therefore, legacy carriers can begin to plan and conduct strategic preemptive price cutting earlier. In contrast, connecting entry is harder to predict since Southwest seldom advertises its connecting services. So legacy carriers could not plan ahead, and thus they fail to reduce prices until closer to the actual entry when more information makes the entry to be predictable. As for the reason of doing preemptive price cutting, a lot of scholars have proposed various explanations, including entry deterrence, occupying market shares, and developing customer loyalty. 9 Next, I examine the changes in passenger traffic as a response to the fare changes. Not surprisingly, the fare drops are accompanied with rises in passenger traffic. For example, legacy carriers have approximately 3.8% more passengers traveling on nonstop flights and 4.6% more passengers traveling on connecting flights upon Southwest s 9 See, for example, Milgrom and Roberts (1982), Klemperer (1987), Goolsbee and Syverson (2008), and Ma (2016). 25

41 connecting entry. When it comes to Southwest s nonstop entry, the numbers of passengers increase by 4.1% and 6.3% respectively for legacy carriers nonstop and connecting flights. To the best of my knowledge, this paper is the first empirical work that analyzes Southwest s entry in the context of both connecting and nonstop services. There are three important ways in which this paper adds to the current literature. First, while most of the existing research ignores the connecting part in the airline industry, this paper takes both nonstop and connecting trips into the discussion, and thus it can correct the possible bias from ignoring connecting trips, and analyze the responses of legacy carriers to LCCs much more comprehensively. Second, although some scholars have separately analyzed connecting trips from nonstop trips 10, they usually use fare variations across routes. In contrast, here I focus on fare changes within a route. Third, by showing that a connecting entry significantly affects nonstop prices and a nonstop entry significantly affects connecting prices, my findings imply that the connecting and nonstop markets are relevant to each other. In other words, airline competition exists both within and across service types. Therefore, my work also extends previous work on airline competition. 11 The rest of the paper are composed as follows: the data and the steps taken to construct the sample are described in Section 2.2. Section 2.3 outlines the econometric model. Section 2.4 discusses the effect of Southwest s entry on airfares. Section 2.5 compares the results in Section 2.4 with existing literature. Section 2.6 examines the effect of Southwest s entry on passenger traffic. Section 2.7 concludes the article. 10 See, for example, Lijesen et al. (2002), Berry and Jia (2010), and Brueckner et al. (2013). 11 See, for example, Reiss and Spiller (1989), Dunn (2008), and Gayle and Wu (2011). 26

42 2.2 Data Data Description The data employed in this study s analyses came from the Airline Origin and Destination Survey (DB1B), which consists of a 10% random sample of all tickets sold and operated by US carriers. All tickets in this survey are aggregated by their actual travel quarter, which means that for each ticket, I know in which quarter it flew, but I do not know the exact date or the time of departure. Besides the quarterly information, the database provides rich ticket-level facts, including the ticketing, operating, and reporting carriers, the price, the number of passengers, the fare class, the airport sequence (the origin and destination airports and all transfer points), the actual flight distance, and the nonstop flight distance between the two endpoint airports. However, the data does not have passengerlevel facts such as if a passenger purchased the ticket in advance, if a passenger is a frequent flyer, etc.. In addition, for connecting trips, the associated layover time is unknown. This is why I use the difference between the actual flight distance and the nonstop distance to measure the quality of a connecting trip in later sections. The sample used in this study includes data from the first quarter of 1993, through to the last quarter of That being said, this study is based on a panel dataset that covers 22 years. I impose the following data restrictions. I drop: (i) trips that are part of international travel; (ii) trips that involve more than one intermediate stop to simplify the analyses; (iii) trips with fares of less than $20 in light of the high likelihood that they might come from 27

43 coding errors or are unusually discounted trips; and (iv) trips that involve two or more carriers since these are mostly codeshare trips whose pricing rule is more complicated Market Definition Following Borenstein (1989), Goolsbee and Syverson (2008), and Ciliberto and Tamer (2009), I characterize a market by the two endpoint airports, independent of any intermediate transfer stops and the heading of the flight. For example, a flight from John F. Kennedy International Airport (JFK) to San Diego International Airport (SAN) and a flight from SAN to JFK are considered as being in the same market. If a city has multiple airports, I assume that flights from (to) different airports in the same city belong to different markets. For instance, a flight from JFK to SAN and a flight from LaGuardia Airport (LGA) to SAN are regarded as being in different markets. In Table B.6 in Appendix B, I conduct a robustness check by defining a market by the two endpoint cities 12. Although the coefficients of interest become smaller under this alternate definition, the trends of the price paths do not change much Data Aggregation The data is aggregated into market-carrier-quarter-service type cells, where the carrier refers to the ticketing carrier. An example cell can be JFK-SAN-Delta-2010Q1- onestop, where JFK-SAN represents the market between JFK and SAN, Delta 12 More specifically, I consider all airports that are within 100 miles from the city center as being in the same city. There are 19 US cities with multiple airports. A list of these cities and their airports are given in Appendix C. 28

44 represents the carrier of this trip, 2010Q1 is the quarter in which this flight took off, and onestop is the service type. Note that within each cell, I don t distinguish the intermediate transfer points. The reason is that the transfer point is normally associated with the carrier, and thus by including carrier identification, I am able to control the transfer point as well in most cases. For example, in the above JFK-SAN market, American usually operates connecting flights through Phoenix, while Delta normally transfers via Los Angeles. Even if a single airline transfers through various airports, the results should not be very different, as the airline will charge very similar prices for those trips. In Table B.7 in Appendix B, I perform a robustness check which distinguishes between transfer airports, i.e., I aggregate data into airport sequence-carrier-quarter-service type cells. The results do not change a lot. Hence, it should be safe to neglect the transfer points in data aggregation. I then compute the average ticket price (weighted by the number of passengers), the average flight distance (weighted by the number of passengers), and the total number of passengers within each cell Market Selection To study how legacy carriers respond to Southwest s entry, I restrict the sample to markets where Southwest started flying between 1993 and The criteria to identify a nonstop (connecting) entry is as follows: Southwest must not have operated nonstop (connecting) flights in the market for at least 12 quarters preceding the quarter of entry, and must stay in the market for no less than four successive quarters following the quarter 29

45 of entry. Southwest must also have transported at least 25 passengers on its nonstop (connecting) flights in the quarter of entry and the four quarters following. In all, I observe Southwest s entry into 651 markets with connecting flights, and 424 markets with nonstop flights. For each of the selected markets, I further delete the cells where the total number of passengers is less than 25 (for a predicted quarterly traffic of 250). The reason for doing so is to exclude very unpopular trips. In this study, I focus on the seven major legacy carriers in the sample period, including American, Delta, United, US Airways, Northwest, TWA, and Continental. Ultimately, the final sample has 686 markets and 70,512 observations (cells). 2.3 Empirical Model I use a set of fixed effects models to identify how legacy carriers respond to Southwest s entry in terms of prices and passenger traffic. Two dependent variables are used for prices, namely, the logged mean fare of nonstop trips ( _ ), and the logged mean fare of connecting trips ( _ ). Similarly, two dependent variables are used for passenger traffic, including the logged number of passengers on nonstop trips ( _ ) and on connecting trips ( _ ). Table B.2 in Appendix B provides descriptive statistics for the dependent variables. I control for a couple of fixed effects, namely, the market-carrier-pair and the timecarrier-pair fixed effects. The aim to include these fixed effects is to control for any unobserved differences across market-carrier-pairs and time-carrier-pairs, respectively. 30

46 The empirical specification is as follows: = + +, +, + (1 + ) + (2.1) where is the dependent variable for carrier in market in quarter (in later analyses, is one of the following variables: _, _,, or _ ); is the market-carrier-pair fixed effect; is the time-carrier-pair fixed effect; and is an error term that captures any additional variations in the dependent variable., are time dummy variables surrounding Southwest s nonstop entry into a market. There are 18 time dummies that account for from the eighth quarter before entry through to infinite quarters after entry. In particular,, is equal to one if the observation occurred eight quarters before Southwest began to operate nonstop flights in market,, is equal to one if the observation occurred seven quarters before Southwest began to operate nonstop flights in market, etc.. in which Southwest started nonstop services in market., indicates the quarter, indicates the first quarter after Southwest started nonstop services in market,...,, indicates the eighth quarter after Southwest started nonstop services in market. Finally,, indicates the ninth to infinite quarters after Southwest started nonstop services in market. If Southwest never entered the market with nonstop services,, = 0,. Similarly,, are time dummy variables surrounding Southwest s onestop entry into a market. There are another 18 time dummies that account for from the eighth 31

47 quarter before entry through to infinite quarters after entry., indicates the quarter in which Southwest started onestop services in market., indicates the first quarter after Southwest started onestop services in market,...,, indicates the eighth quarter after Southwest started onestop services in market, and, indicates the ninth to infinite quarters after Southwest started onestop services in market. If Southwest never entered the market with onestop services,, = 0,. measures the quality of the trip. Following Dunn (2008), I calculate the quality of a connecting trip as the total distance flown on the connecting flight minus the nonstop distance between the two endpoint airports. As this variable decreases, the quality of connecting service raises. If a trip is nonstop, this variable will equal to zero. The coefficients of interest are s and s. Since I omit time dummy variables that indicate the ninth to infinite quarters preceding the quarter in which Southwest entered, my estimated s and s should capture the value of the dependent variable in the dummy periods relative to its size in the excluded period (the ninth to infinite quarters preceding the quarter of actual entry). 2.4 Empirical Results: Fare Changes Table B.3 in Appendix B summarizes the results from using model (2.1) to estimate price changes. In the table, column (I) shows the results when _ is the dependent variable, and column (II) shows the results when _ is the dependent variable. 32

48 In order to track the price changes of legacy carriers, I graph price paths based on the estimates of time dummy coefficients ( s and s), as shown in Figure 2.1. The estimates are transformed into percent changes in the average price relative to its value in the excluded period. In this figure, Southwest enters in quarter 0. Negative quarter values represent quarters before entry and positive quarter values represent quarters after entry. The red solid line shows the percent changes in the average price. The blue dotted lines show the 95% confidence interval. If the price path is downward sloping, this means that legacy carriers reduce prices in the market. Picture (a) in Figure 2.1 illustrates legacy carriers nonstop price changes surrounding Southwest s nonstop entry. According to this picture, legacy carriers started to reduce prices since six quarters prior to the actual entry. By the quarter prior to Southwest s entry (quarter 1), legacy carriers set the nonstop prices to be 7.1% lower, 13 on average, compared to prices in the excluded period (the ninth to infinite quarters before entry). There are many reasons to explain the preemptive price-cutting behavior. For example, legacy carriers may resort to using it to deter entry, by signaling that they too are low-cost. Another explanation is that legacy carriers aim at developing customer loyalty in the hope that those customers would stay with them even after Southwest entered the market in later periods. The average price further decreased to 12.1% lower in the quarter of entry (quarter 0). As such, an immediate price drop because of Southwest s entry should be 5% (=12.1%-7.1%). The average price then maintained at a relatively low level. By the eighth quarter after entry (quarter +8), the average price was down by 12.7%. These results 13 The percent change in average price is computed as exp(-0.074)-1 = = -7.1%. 33

49 corroborate previous findings in the literature, specifically, that legacy carriers decrease their nonstop prices in response to Southwest s nonstop entry into a market. Figure 2.1: Price Changes Surrounding Southwest s Entry 34

50 Picture (b) in Figure 2.1 shows legacy carriers connecting price paths surrounding Southwest s nonstop entry. Similar to Picture (a) in this figure, legacy carriers began to decrease prices since approximately six quarters before Southwest s entry. By the quarter prior to entry (quarter 1), the average price decreased by 7.7%. The price further reduced to 16.5% lower in the quarter of entry (quarter 0). After that, the price kept to be low, and reached 21.2% lower by the eighth quarter after entry (quarter +8). Comparing Picture (a) and Picture (b), it is clear that legacy carriers nonstop and connecting prices experienced similar paths: both started to decrease since six quarters before entry, both decreased fast around the quarter of entry, and then kept low after entry. Nonetheless, the fare drops are much larger for connecting services than for nonstop services. Legacy carriers cut connecting prices more because those connecting flights are less competitive when competing with new entrants due to their inconvenience. Thus, it is necessary to reduce prices more to compensate consumers. Picture (c) in Figure 2.1 portrays legacy carriers nonstop price changes around Southwest s connecting entry. According to this picture, the average price didn t systematically drop until the quarter prior to entry (quarter 1). By the quarter of actual entry (quarter 0), the average price was down by 5.5%. The price fluctuated a little after entry, and arrived at 6% lower by the eighth quarter after entry (quarter +8). Comparing Picture (a) and Picture (c), legacy carriers nonstop fares exhibited different paths in response to Southwest s nonstop and connecting entry. When facing nonstop entry, the prices not only started to drop much earlier, but also declined more. The reason why the prices dropped earlier in the case of nonstop entry may be because nonstop 35

51 entry is usually pre-announced in most cases, Southwest announces its new nonstop routes for at least one year, and sometimes the announcement could be as early as five years ahead. Therefore, legacy carriers could begin to plan and conduct strategic preemptive price cutting earlier. In contrast, connecting entry is harder to predict since Southwest seldom advertises its connecting services. So legacy carriers could not plan ahead, and thus they fail to reduce prices until closer to the actual entry when more information makes the entry to be predictable. As for why the prices dropped more in response to nonstop entry, the reason is that nonstop entry creates more threats than connecting entry, as nonstop services are more convenient and have better quality. Picture (d) in Figure 2.1 shows legacy carriers connecting price paths around Southwest s connecting entry. Similar to Picture (c) in this figure, legacy carriers didn t cut prices until the quarter prior to entry (quarter 1). The average price decreased to 10.1% lower by the quarter of entry (quarter 0), and remained to be low after entry. By the eighth quarter after entry (quarter +8), the average price was down by 12.5%. Comparing Picture (c) and Picture (d), it is clear that these two cases have similar patterns: both prices started to drop since the quarter before entry, both significantly decreased in the quarter of entry, and then kept low. However, the magnitude of the drop is much larger in Picture (d), which means that legacy carriers connecting prices fell more than their nonstop prices. Comparing Picture (b) and Picture (d), the preemptive price cutting patterns are different: for nonstop entry, preemptive price cutting started since six quarters prior to actual entry; for connecting entry, there was no preemptive price cutting until one quarter before entry. 36

52 As a brief summary, these results suggest that Southwest s nonstop entry affects legacy carriers pricing in both nonstop and connecting markets, and so does its connecting entry. The fare drops caused by Southwest s nonstop entry are much larger than those caused by its connecting entry. When facing the same type of entry, legacy carriers cut connecting fares more than they do with nonstop fares. Besides, there is obvious preemptive price reduction in all cases. Legacy carriers start to reduce prices approximately six quarters prior to Southwest s nonstop entry, and one quarter prior to its connecting entry. 2.5 Compare with Previous Literature As discussed in Section 2.1, most of the existing articles on Southwest s entry tend to ignore the connecting part and only focus on the nonstop part. My first innovation is that I use data on connecting fares in a second regression. As a result, my work extends existing literature by broadening the analyses to connecting markets. My second innovation is that I add a term, to the models. By doing this, I can correct potential bias that previous literature may have from ignoring connecting entries. In particular, while previous literature studies a situation very similar to that in Picture (a) in Figure 2.1, there are still some differences. To compare these differences, I first estimate a model very similar to what s widely used in previous articles: = + +, + (2.2) where all variables are as defined in Section

53 The results from estimating model (2.2) are presented in Table B.4 in Appendix B. I then graph the price path based on the estimates of s, and compare it with the s in Picture (a) in Figure 2.1. The green solid line in Figure 2.2 represents the estimated price changes from model (2.2). The red solid line is the same as that in Picture (a) in Figure 2.1. In both cases, Southwest enters with nonstop services in quarter 0. Figure 2.2: Nonstop Price Changes: Compare with Previous Literature 38

54 From this graph, there are two important differences between the green line and the red line. First, the estimated price drops are generally larger in the green line. Moreover, the s of these two lines are statistically significantly different. Second, the preemptive price cutting is more significant in the green line. As a comparison, in the red line, these additional variations are explained by adding a series of time dummies that indicate Southwest s connecting entry into a market. In other words, I correct the bias in the estimates by including connecting entry. In addition to the differences shown above, my work extends existing literature by broadening the analyses to connecting markets (Pictures (b), (c), and (d) in Figure 2.1). As a result, I provide compounding findings to understand Southwest s impact. While it seems natural that connecting fares respond to connecting entry and nonstop fares respond to nonstop entry, my findings that such reactions also exist across service types, namely, connecting fares respond to nonstop entry and nonstop fares respond to connecting entry, provide evidence that the nonstop and connecting markets are related to each other. Therefore, it is not good to consider them as separate markets as suggested by Goolsbee and Syverson (2008). Instead, my findings prove that it is necessary to take the crossmarket competition into account. 2.6 Empirical Results: Number of Passengers The results in Section 2.4 suggest that legacy carriers cut both connecting and nonstop prices upon Southwest s entry. A following question is whether such price drops could entice more travelers to fly with those carriers. Table B.5 in Appendix B summarizes 39

55 the results from using model (2.1) to estimate the changes in passenger traffic. In the table, column (I) shows the results when _ is the dependent variable, and column (II) shows the results when _ is the dependent variable. The estimates are based on the equilibrium numbers in the market, and thus they cannot accurately represent the changes in demand. However, they shed light on the size of the market. I transform the estimates of time dummy coefficients to percent changes in passenger traffic, which are represented by the green solid line in Figure 2.3. The blue dotted lines show the 95% confidence interval. The red dotted line pictures the changes in average price in the corresponding market. In all cases, Southwest enters in quarter 0. Picture (e) in Figure 2.3 shows legacy carriers nonstop passenger traffic in response to Southwest s nonstop entry. As the average price drops, the number of passengers increases. By the quarter of entry (quarter 0), the passenger traffic rises by 11.4%. The number slowly increases in later periods, and reaches 16.3% higher by the eighth quarter after entry (quarter +8). Recall that the average price drops to 12.7% lower in this quarter. Hence, the revenue should increase. Picture (f) in Figure 2.3 illustrates legacy carriers connecting passenger traffic in response to Southwest s nonstop entry. Similar to Picture (e), the decrease in prices is accompanied with an increase in passenger traffic. The number of passengers rises to 12.1% higher by the quarter of entry (quarter 0), and further rises to 18.5% by the eighth quarter after entry (quarter +8). In addition, the changes in quantity are smaller than the changes in price, which implies a decrease in revenue. 40

56 Figure 2.3: Passenger Traffic Changes Surrounding Southwest s Entry Picture (g) in Figure 2.3 represents legacy carriers nonstop passenger traffic surrounding Southwest s connecting entry. By the quarter of entry (quarter 0 ), the 41

57 passenger traffic rises by 6.5%. By the eighth quarter after entry (quarter +8), the number increases to 12.4% higher. As the changes in quantity are more significant than the changes in price, the revenue should increase. Picture (h) in Figure 2.3 illustrates legacy carriers connecting passenger traffic surrounding Southwest s connecting entry. The number of passengers increases to 7.7% higher by the quarter of entry (quarter 0), and keeps rising afterward. By the eighth quarter after entry (quarter +8), the passenger traffic increases by 15.5% compared to the excluded period. Note that the average price decreases by 15.3% in this quarter, the revenue slightly increases. In short, the results demonstrate that the price reductions are accompanied with an increase in passenger traffic. While legacy carriers get less revenue from connecting services when Southwest enters the market with nonstop flights, they should have more revenue in all other cases. 2.7 Conclusions This paper has examined the responses of legacy carriers to Southwest s entry in the context of both connecting and nonstop services. My findings indicate that Southwest, the leading LCC in the US, imposes significant downward pressure on the airfares of both services. Upon its entry, legacy carriers reduce prices in both nonstop and connecting markets, with connecting prices dropping more, probably to compensate for the inconvenience of changing planes. In addition, I find that the declines in airfares are 42

58 accompanied with an increase in the number of passengers. As a result, legacy carriers should gain more revenue in most markets. While most of the previous studies ignore the connecting part in the airline industry, this paper takes both nonstop and connecting trips into the discussion, and thus it can correct the possible bias from ignoring the connecting part, and analyze the responses of legacy carriers to LCCs much more comprehensively. Additionally, by showing that Southwest s connecting entry significantly affects nonstop prices and that its nonstop entry significantly affects connecting prices, my findings provide evidence that the connecting and nonstop markets are relevant to each other. In other words, airline competition exists both within and across service types. There are two important limitations to these findings. First, in all analyses, I assume that Southwest s entry is exogenous. It would be interesting if one could find an instrumental variable to correct for the potential endogeneity. Second, I exclude codeshare flights in this article. It is worth trying to see if codeshare flights also experience the same changes as non-codeshare flights. Despite these limitations, this article still contributes to the literature as the first empirical work that analyzes Southwest s entry by incorporating both connecting and nonstop markets. 43

59 Chapter 3: Profits and Entry Decisions: The Effect of Southwest Airlines 3.1 Introduction In the United States airline industry, there are legacy carriers and low-cost carriers. Today, Southwest Airlines is the largest low-cost carrier. It has long been believed that the introduction of Southwest can cause a drop in airfares. As early as 1993, Bennett and Craun concluded that Southwest s operations on the Oakland-Burbank route resulted in a 55% decrease in prices. Later researches found similar results by studying different routes and different settings, including Richards (1996), Dresner et al. (1996), Morrison (2001), and Goolsbee and Syverson (2008). At the same time, some scholars also find that the price drops are accompanied with significant increases in passenger traffic. 14 Consequently, it is not apparent of whether and how airlines profits have changed on those routes. In this paper, I estimate a game of simultaneous entry, and demonstrate that Southwest negatively affects other airlines profit functions. In particular, I assume that airlines decisions to enter a route are based on if they can get non-negative profits from entry. I then estimate the parameters in determining profit 14 See, for example, Goolsbee and Syverson (2008). 44

60 functions such that the entry outcomes, predicted by the equilibrium computed from the model, match the actual entry outcomes. The existence of multiple equilibria constitutes a big challenge in estimating this kind of model. As an example, suppose there is a market with three potential entrants: one large and two small. The possible market structure could be one where the large firm has a monopoly, but it could also be a duopoly of the two smaller firms. It is not easy to get around such multiplicity. To deal with this problem, I use the partial identification method proposed by Ciliberto and Tamer (2009). Based on the 2010 Q1 DB1B data, my results show that in a given market, Southwest s presence enters into other airlines profit functions with a negative sign. As the profit functions obtained with this method represent indices, I cannot tell in dollar value of how much the real profits are influenced by Southwest. Nonetheless, from the coefficients that represent Southwest s presence, I can infer that such influences are airlinespecific: I find that Southwest has a larger impact on the profits of medium airlines 15 and small LCCs 16 than its impact on larger airlines 17. Among the large airlines, Delta s profit is influenced the most. In addition to identify Southwest s impact on each airline s profit, I also attempt to distinguish and examine different service types, namely, connecting and nonstop services. In the past, most literature on the US airline industry, especially those on LCCs, focused on nonstop trips. The reason why they didn t discuss connecting flights is because nonstop 15 In this article, medium airlines include Continental and Alaska Airlines. 16 In this article, small LCCs include AirTran Airways, Allegiant, Frontier, Spirit, JetBlue, Virgin, and Sun Country Airlines. 17 In this article, large airlines include American, Delta, US Airways, and United. 45

61 service and connecting service can be considered separate markets, or at least substantially differentiated products (Goolsbee and Syverson 2008). Nonetheless, there is no solid evidence that nonstop and connecting markets are truly separate from each other. In fact, this question has caused lengthy discussion in antitrust practices, and no agreement has been reached. For example, in 2008, Department of Justice (DOJ) investigated the alliance of five airlines 18 and concluded that... nonstop service is a separate product from connect service,... Put another way, for a substantial number of nonstop consumers, connecting competition does not discipline price in any meaningful sense However, in 2010, Department of Transportation (DOT) reevaluated the data and challenged DOJ s empirical model. It reported that... all of the nonstop overlap markets are disciplined to an extent by connecting competition Therefore, it is obvious that there is no agreement among policy makers of whether nonstop and connecting markets are relevant or not. In academia, there is no consistent understanding of this issue, either. While Gayle and Wu (2011) argued that connecting and nonstop products only weakly compete with each other, Reiss and Spiller (1989) and Berry and Jia (2010) proposed that these two service types are considerably relevant. In a word, neither academia nor antitrust authorities are certain about whether and to what extent nonstop and connecting markets can affect each other. By distinguishing 18 The alliance includes American Airways, British Airways, Iberia, Finnair, and Royal Jordanian Last Accessed: March 20, Last Accessed: March 20,

62 nonstop and connecting services in each airline s profit function, my paper sheds light on this very important question. I find that each airline s profit function is determined by its own and all other airlines presence in both nonstop and connecting markets. For instance, Southwest s nonstop presence on a given route affects all other airlines nonstop and connecting profits from the same route. Southwest s connecting presence has similar influence, too. Moreover, the nonstop presence generally has larger impact on other airlines profits than the connecting presence. When having Southwest s presence, other airlines connecting profits are more affected by it than their nonstop profits. Using the estimated parameters, I perform several counterfactual experiments to assess the extent to which Southwest s entry influences the entries of other airlines. The counterfactuals are conducted by setting Southwest s competitive effects equivalent to zero (like removing Southwest from each market), computing the new equilibria, and then comparing the equilibria with Southwest being removed to the initial equilibria where Southwest operates. I find that removing Southwest increases the probability of observing each individual carrier. For example, the likelihood of small low-cost carriers entering a market boosts by up to 41.4% when Southwest is removed, while the likelihood of observing Delta increases by up to 39.8%. Not surprisingly, the equilibrium number of non-southwest carriers in the market also increases with Southwest being removed. According to the model prediction, the mean number rises from to These results further confirm that Southwest has a substantial impact on the entry decisions of other carriers and the market structures. 47

63 The findings of this paper contribute to the current literature in three key ways. First, while the majority of the literature on Southwest focuses on its impact on other airlines pricing, this paper extends the discussion to profits and market structures. Second, it adds to the literature on the empirical analysis of entry games. Currently, most articles estimating entry games in the context of airline industry tend to either only count for nonstop services or aggregate different service types. 21 However, connecting services differ from nonstop services in a variety of aspects, which makes it inappropriate to use nonstop services to represent the entire air travel market or to directly aggregate them. For instance, from consumers perspective, connecting trips are less convenient and usually cheaper than nonstop trips. From airlines perspective, on the other hand, connecting trips are less costly and more flexible. In this paper, I distinguish connecting and nonstop entries in the game estimation, by adding separate variables that indicate airlines connecting and nonstop presence into their profit functions. As a result, I get some very interesting conclusions. Third, my paper also throws light on the long-existing debate of whether nonstop and connecting markets are relevant to each other. I reveal that at least from the standpoint of profit determination and entry decisions, these two service types are considerably related to each other and affect each other. The remaining portion of this manuscript is outlined as follows. Section 3.2 introduces the data and the steps taken to construct the sample. Section 3.3 explains the empirical models and the game. Section 3.4 evaluates the effect of Southwest s presence on other airlines profit functions. Section 3.5 presents counterfactual exercises to examine 21 See, for example, Bresnahan and Reiss (1990), Berry (1992), and Ciliberto and Tamer (2009). 48

64 how Southwest influences the entry decisions of each competitor and the equilibrium number of firms in the market. Conclusions and comments are made in Section Data Data Construction The primary data employed in this study s analyses came from the Airline Origin and Destination Survey (DB1B), which consists of a 10% random sample of all tickets sold and operated by US carriers. All tickets in this survey are aggregated by their actual travel quarter, which means that for each ticket, I know in which quarter it flew, but I do not know the exact date or the time of departure. Besides the quarterly information, the database provides rich ticket-level facts, including the ticketing, operating, and reporting carriers, the price, the number of passengers, the fare class, the airport sequence (the origin and destination airports and all transfer points), the actual flight distance, and the nonstop flight distance between the two endpoint airports. The sample used in this study includes data in the first quarter of That being said, this study is based on a cross-sectional panel dataset. The reason to use cross-sectional data instead of panel data is that the game is a static game, and I only examine the singlestage equilibrium. Hence, one period should be enough. I impose the following data restrictions. I drop: (i) trips that are part of international travel; (ii) trips that involve more than one intermediate stop to simplify the analyses; (iii) trips with fares less than $20 in light of the high likelihood that they might come from 49

65 coding errors or are unusually discounted trips; and (iv) trips that involve two or more carriers since these are mostly codeshare trips and are more complicated Market Definition Following Borenstein (1989), Goolsbee and Syverson (2008), and Ciliberto and Tamer (2009), I characterize a market by the two endpoint airports, independent of any intermediate transfer stops and the heading of the flight. For example, a flight from John F. Kennedy International Airport (JFK) to San Diego Internation Airport (SAN) and a flight from SAN to JFK are considered in the same market. If a city has multiple airports, I assume that flights from (to) different airports in the city belong to different markets. For instance, a flight from JFK to SAN and a flight from LaGuardia Airport (LGA) to SAN are regarded in different markets Data Aggregation The data is first aggregated into market-carrier-service type cells, where the carrier refers to the ticketing carrier. An example cell can be JFK-SAN-Delta-onestop, where JFK-SAN represents the market between JFK and SAN, Delta represents the carrier of this trip, and onestop is the service type. Note that within each cell, I don t distinguish the intermediate transfer points. The reason is that I only concern about if a given airline is present in a specific market. So as long as it transports passengers between the two endpoint airports with onestop flights, I will count it as being present in the market onestoply, regardless of the actual transfer point. 50

66 Moreover, in reality the transfer point is normally associated with the carrier, and thus by including carrier identification, I am able to control the transfer point as well in most cases. I then compute the total number of passengers within each cell, and delete the cells where the total number of passengers is less than 50 (for a predicted quarterly traffic of 500). The reason for doing so is to exclude very unpopular trips. Subsequently, I aggregate the data further by market segments. In doing, I create several dummy variables to represent the carriers operating in each market. I make two versions of this aggregation. In the first version (consistent with Ciliberto and Tamer s method), for a particular carrier and a particular market, I create a dummy variable that equals1 if the carrier is present in the market, regardless of the service type, and equals 0 if it is not present in the market. It is worth mentioning that I do not need price information in the analyses. The key interest here is whether a given airline is present in a specific market or not. In the second version, I break down the presence of airlines by service types. Hence, For a particular carrier and a particular market, I create two dummy variables to indicate its market presence: equals 1 if carrier operates connecting flights in the market and equals to 0 otherwise; similarly, equals 1 if carrier operates nonstop flights in the market and equals to 0 otherwise. To simplify the analyses, I create dummy variables individually for five major airlines, including Southwest, American, Delta, US Airways, and United. I then build two more dummy variables to indicate the remaining airlines. The dummy variable equals 1 if either Alaska or Continental Airlines is observed in a specific market. They are denoted as medium airlines (MA). The dummy variable equals 1 if no less than one small 51

67 low-cost carriers (LCCs) are observed in a specific market. Small LCCs refer to LCCs other than Southwest, including AirTran Airways, Allegiant, Frontier, Spirit, JetBlue, Virgin, and Sun Country Airlines. After the construction of dummy variables, the data can be aggregated by market segments. Within each market, there is a vector whose elements are all zeros and/or ones. This vector represents the structure of the market, namely, what airlines serve the market and what do not Market Selection Following Ciliberto and Tamer (2009), besides the markets selected as above, I also incorporate a few markets where, in the sample period, no carrier operated. To pick these markets, I first construct a ranking of airports by the metropolitan statistical areas (MSAs) population. I then identify unserved markets between the top 50 MSAs and add them to the sample. Ultimately, the final sample has 2850 markets. 3.3 Models and Game Estimation Empirical Models The models used in this study are very similar to that suggested by Berry (1992) and Ciliberto and Tamer (2009). I assume that the profit function for airline in market is: = (3.1) 52

68 where is a vector of market characteristics which are common among all carriers in market ; is a vector of carrier s characteristics which enter only into carrier s profit in market ; is a vector that represents other carriers presence in market ; finally, is the part of profits that are unobserved by economists. The coefficients of interest are. They summarize the effect of entry by airline on the profit of airline. This effect is defined as the competitive effect. For example, the effect of Southwest s presence on American s profit is captured by, and the effect of Delta s presence on American s profit is captured by. Subsequently, I break down airlines entries by service types. The profit function for carrier s type service in market is given by: = + +,,, +,, +, + (3.2) where there are only two types of services: a nonstop service and a connecting service. If represents the nonstop service, then represents the other service, namely, the, connecting service, and vice versa., captures the effect of airline s presence in the, market with type service on the profit of its type service in the same market;, measures the effect of airline s presence in the market with type service on the profit, of airline s type service in the same market; and, measures the effect of airline s presence in the market with type service on the profit of airline s type service in the same market. 53

69 3.3.2 Variable Definitions The variables used in this study are defined below. Table D.1 in Appendix D gives summary statistics of these variables. The market characteristics included in are: o : The geometric mean of population at the market endpoint MSAs. 22 o : The average per capita income at the market endpoint MSAs. o h : The average rate of per capita income growth at the market endpoint MSAs. o : The nonstop distance between the market endpoint airports. 23 o : The minimum of the distances from each endpoint airport to the closest airport. 24 o : A dummy variable that is equal to 1 if either one of the market endpoints is a slot-controlled airport, and is equal to 0 otherwise Data for population and income are acquired from the U.S. Census Bureau. 23 Data for nonstop distances are extracted from DB1B. 24 Data on the distances between airports are obtained from the National Transportation Atlas Database, available from the Bureau of Transportation Statistics. They are also used to construct the variable cost. 25 There are four slot-controlled airports in the US, including New York LaGuardia, New York Kennedy, Ronald Reagan Washington, and Chicago O Hare. 54

70 The dummy variables included in are defined in two versions, as described in Section 3.2.3: o In the first version (consistent with Ciliberto and Tamer s method), for a particular carrier and a particular market, I create a dummy variable that equals1 if the carrier is present in the market, regardless of the service type, and equals 0 if it is not present in the market. o In the second version, I break down the presence of airlines by service types. Hence, For a particular carrier and a particular market, I create two dummy variables to indicate its market presence: equals 1 if carrier operates connecting flights in the market and equals to 0 otherwise; similarly, equals 1 if carrier operates nonstop flights in the market and equals to 0 otherwise. The firm characteristics included in are: o : A carrier s airport presence in a market is the average of its presence at the endpoint airports. For each airline, its presence at a particular airport is computed by dividing the number of markets served by this airline from the airport by the total number of markets that are served by at least one airline from the airport. For example, since Delta dominates ATL, its airport presence in all markets from ATL is large. o : A carrier s cost in a market is its minimum connecting distance minus the nonstop distance divided by the nonstop distance between the two endpoint airports. This variable measures the fixed opportunity cost of entering a market, as the closest 55

71 connecting trip should be the next best alternative to a nonstop trip, and this number does not change as the number of passengers change The Game I estimate a static game of simultaneous entry. In doing, the estimation is based on data in one period. I assume that for each market, every airline needs to decide whether or not to enter it in this period, and this decision is independent of history. I also assume that the decision in a market is not affected by the decisions in other markets. An airline will enter a particular market as long as it can realize non-negative profit from entry. Its post-entry profit in the market is specified as a function of market-level variables, of its airport presence and fixed cost, and of the competitive effects brought by its competitors. The parameters in determining the profit function are estimated by ensuring that the entry outcomes, predicted by the model equilibrium, match the observed entry outcomes, shown in the data. Note that the estimation is based on the assumptions that each airline takes its overall network structure as a given when deciding whether to enter a market, and airlines make decisions independently Error Terms The error term in model (3.1),, consists of four parts: (i) a carrier-specific error term, ; (ii) a market-specific error term, ; (iii) two airport-specific error terms, and. is computed by summing up these four errors. 56

72 The error term in model (3.2),, is obtained by adding an additional error term to the described above, where is a service-type-specific error term which is common across all observations whose type is. I assume that,,,, and are all independent and follow the standard normal distribution. This is a game of complete information. So and are observed by all players in the market Simulation Parameters in the profit function are estimated using the Monte Carlo simulation method proposed by Ciliberto and Tamer (2009). The main goal is to find the set of parameter values such that the probabilities of different entry outcomes, obtained from the data, are bounded within the lower and upper bounds predicted by the model equilibrium. In this instance, an equilibrium outcome = (,,, ) is a sequence of zeroes and ones that represent carriers entry decisions (e.g., Delta and Southwest are the only two airlines that enter the market). For each parameter, I then report the smallest connected cube that contains the 95% confidence region Not every parameter in the cube is in the confidence region. 57

73 3.3.6 Identification Ciliberto and Tamer (2009) shows that exclusion restrictions help to identify the parameters, where exclusion restrictions can be some variables that enter into one firm s profit function, but not another firm s. In this paper, there are two exclusion restrictions: the variable and the variable. Both variables are carrier-specific variables that shift an individual carrier s profit function without changing its competitors profit functions. They help to identify the parameters. 3.4 How Does Southwest Influence Profits? Empirical Results: Aggregate Across Service Types Table 3.1 provides the results for estimating model (3.1). Here, I aggregate data across different service types. There are two types of services, namely, nonstop services and connecting services. Aggregating across service types means that a carrier is defined as operating in a market if it transported no less than 50 passengers in the sample period, regardless of the service types. Column (I) in Table 3.1 shows the results when I restrict = and =,. In other words, I assume that the coefficients of the market-level and carrier-level variables are common across airlines. I also assume that an airline s presence affects each competitor s profit to the same degree. For example, the effect of Southwest s presence on Delta s profit is, and on American s profit is also. This is called as Southwest s competitive effect, and it is summarized in the row titled SW. The 58

74 smallest connected cube that contains the 95% confidence region of is estimated to be [ , ]. The negative sign means that other carriers profits decrease when Southwest enters the market, and the numbers indicate how much the profit index declines. Similarly, the row titled AA shows American s competitive effect. The parameter is estimated to be [ , ], which means that American s entry negatively affects other carriers profits, and this effect should be, on average, smaller than Southwest s effect. The competitive effects of Delta, US Airways, United, and medium airlines (MA) are very similar to that of American. On the other hand, small low-cost carriers (LCC) have a similar competitive effect as Southwest, which is significantly larger than other airlines. In column (I), the coefficient of has a negative effect on the profit function ([ , ]), as does ([-4.269, ]). The rest of the variables have positive effects on profits, namely, ([0.112, 0.242]), ([5.435, 5.892]), h ([1.009, 1.884]), ([5.187, 5.899]), ([1.082, 2.050]), and ([0.568, 1.663]). 59

75 Table 3.1: Profit Estimations: Aggregate Across Service Types (I) (II) (III) Competitive Effect Non-SW [ , ] [ , ] AA [ , ] DL [ , ] US [ , ] UA [ , ] MA [ , ] LCC [ , ] SW [ , ] [ , ] SW on AA [ , ] SW on DL [ , ] SW on US [ , ] SW on UA [ , ] SW on MA [ , ] SW on LCC [ , ] Airport Presence [0.112, 0.242] [0.099, 0.132] [0.080, 0.447] Cost [-0.473, ] [-0.244, 0.003] [-0.810, ] Per Capita Income [5.435, 5.892] [2.494, 2.906] [1.088, 2.156] Income Growth Rate [1.009, 1.884] [0.727, 0.832] [0.513, 1.178] Slot Control [-4.269, ] [-2.697, ] [-3.246, ] Market Size [5.187, 5.899] Non-SW [4.201, 4.368] [1.196, 3.235] SW [8.807, 8.869] [6.988, 9.986] Market Distance [1.082, 2.050] Non-SW [0.775, 0.879] [-0.265, 1.153] SW [-7.210, ] [ , ] Closest Airport [0.568, 1.663] Non-SW [0.294, 0.454] [-1.035, 2.435] SW [-3.912, ] [-2.427, ] Constant [-8.600, ] Non-SW [-6.429, ] [ , ] SW [-7.571, ] [-6.888, ] Note: I report the smallest connected cube that contains the 95% confidence region. In this table, SW represents Southwest and Non-SW represents non-southwest carriers. 60

76 Column (II) in Table 3.1 shows the results when I relax the assumption =, but still assume that =,. Moreover, I add an assumption that the competitive effects of non-southwest carriers are the same, i.e., American, Delta, US Airways, United, MA, and LCC have the same competitive effects, which is summarized by. Southwest s competitive effect is still captured by. In this setting, the effect of Southwest s presence on other airlines profits, which is its competitive effect, is estimated to be [ , ]. This is larger than the competitive effect of non-southwest airlines, whose estimate is [ , ]. In column (II), the coefficient estimates for,, and are quite different for non-southwest airlines, than those of Southwest. In particular, the coefficient of is estimated to be [4.201, 4.368] for non- Southwest airlines, while for Southwest the coefficient is [8.807, 8.869]. Regarding and, these two variables have positive effects on the profits of non-southwest airlines ([0.775, 0.879] and [0.294, 0.454]), but have negative effects on the profit of Southwest ([-7.210, ] and [-3.912, ]). Column (III) uses a similar setting as column (II), except it allows for Southwest s competitive effect on various airlines to be different. That is to say, I drop the assumption =. Therefore, measures the competitive effect of Southwest on American s profit, while measures the competitive effect of Southwest on Delta s profit, and. Nevertheless, I still keep the assumption that the competitive effects of non-southwest carriers are the same. For example, the competitive effect of Delta is still captured by. I find that Southwest s presence has a negative impact on all 61

77 non-southwest carriers profits, but to various extents. In particular, the impact of Southwest s presence on LCC s profits is the largest ([ , ]), followed by the impact on MA s profits ([ , ]) and on Delta s profits ([ , ]). The impact of Southwest s presence on American s profits is the smallest [ , ]). Comparing columns (II) and (III), I find some differences in the coefficient estimates of exogenous variables. For example, in column (II) the coefficient of for Southwest is [-7.210, ], but in column (III) the coefficient is [ , ]. This suggests that the assumption of a homogeneous competitive effect of Southwest, on various opponents, leads to some bias in estimation. In short, these results show that Southwest s presence has a very strong negative impact on the profit functions of other carriers. This impact varies by the identity of the carrier. While the profits of small low-cost carriers and medium airlines are influenced to a large degree, the profits of big carriers are generally less impacted, with the exception of Delta Empirical Results: Distinguish Service Types I now break down airlines entries by service types. Each airline can enter a market with either a connecting service or a nonstop service or both. The post-entry profit of an airline s nonstop (connecting) service is determined by its own connecting (nonstop) presence, its competitors connecting presence and nonstop presence, as well as the 62

78 exogenous variables. An airline will provide a nonstop (connecting) service as long as it can realize non-negative profit from the nonstop (connecting) service. Table 3.2 reports the estimates of competitive effects from estimating model (3.2). In this instance, I still assume that the competitive effects of non-southwest carriers are the same, i.e., American, Delta, US Airways, United, MA, and LCC have the same competitive, effects. Therefore,, connecting profit;,, captures the effect of Southwest s nonstop entry on its own captures the effect of Southwest s nonstop entry on a non-southwest carrier s connecting profit, for example, on Delta s connecting profit;,,, captures the effect of a non-southwest carrier s nonstop entry,, on its own connecting profit; and, captures the effect of a non- Southwest carrier s nonstop entry on another non-southwest carrier s connecting profit (e.g., the effect of Delta s nonstop entry on United s connecting profit). Column (I) in Table 3.2 shows the effects of Southwest s connecting presence on its own nonstop profit, on a non-southwest carrier s connecting profit, and on a non- Southwest carrier s nonstop profit. I find that Southwest s connecting presence negatively affects non-southwest carriers connecting and nonstop profits ([ , ] and [ , ]). Moreover, the impact on connecting profits is larger than the impact on nonstop profits. This is because Southwest s connecting service competes head-to-head with non-southwest carriers connecting services, but only partially with their nonstop services. Another reason is that connecting services are relatively low quality because of 63

79 their inconvenience, so non-southwest carriers need to reduce their prices a lot in response to Southwest s entry, which results in larger decreases in their profits. Table 3.2: Competitive Effects: Distinguish Service Types (I) (II) (III) (IV) SW Non-SW Nonstop Connect on on SW Connect on Non-SW Nonstop on SW Connect SW Nonstop [15.669, ] [ , ] [ , ] [9.978, ] [ , ] [ , ] Non-SW Connect Non-SW Nonstop [ , ] [ , ] [ , ] [ , ] On other firms: [ , ] On the same firm: [7.438, 8.322] On other firms: [ , ] On the same firm: [16.895, ] On other firms: [ , ] On other firms: [ , ] Note: I report the smallest connected cube that contains the 95% confidence region. In this table, SW represents Southwest and Non-SW represents non-southwest carriers. Column (II) in Table 3.2 summarizes the competitive effects of Southwest s nonstop service. I find that Southwest s nonstop presence also negatively influences non- Southwest carriers connecting and nonstop profits ([ , ] and [ , ]). Moreover, the effect on connecting profits is larger than that on nonstop profits. 64

80 The reason is that connecting services are less competitive than nonstop services, so non- Southwest carriers perceive the need to cut their prices more in response to Southwest s entry, and thus their profits decline to a larger degree. Comparing columns (I) and (II), it is clear that Southwest s nonstop presence has more significant competitive effects than its connecting presence. As a connecting service can be considered as a low-quality product, and a nonstop service can be considered as a high-quality product, this result implies that the entry of a high-quality product creates more pressure than the entry of a low-quality product. Columns (III) and (IV) summarizes the competitive effects of non-southwest carriers connecting services and nonstop services, respectively. I find that both connecting and nonstop presence have a negative impact on other carriers profits. As for the influence level, a connecting presence affects Southwest s connecting profits by [ , ], which is larger than its impact on Southwest s nonstop profits ([ , ]); similarly, it affects another non-southwest carrier s connecting profits by [ , ], which is larger than its impact on another non-southwest carrier s nonstop profits ([ , ]). On the other hand, a nonstop presence affects Southwest s connecting profits by [ , ], which is also larger than its influence on Southwest s nonstop profits [ , ]; its impact on another non-southwest carrier s connecting profits is [ , ], and on another non-southwest carrier s nonstop profits is [ , ]. 65

81 Comparing columns (III) and (IV), I find that the competitive effect of a non- Southwest carrier s connecting service is generally bigger that of its nonstop service. This finding is consistent with the earlier conclusion when columns (I) and (II) are compared. Comparing columns (I) and (III), I show that the competitive effect of Southwest s connecting service is bigger than the competitive effect of non-southwest carriers connecting services. Similarly, the competitive effect of Southwest s nonstop service is bigger than the competitive effect of non-southwest carriers nonstop services, if columns (II) and (IV) are compared. Overall, these results suggest that when I break down airlines entries by service types, both connecting entries and nonstop entries have negative competitive effects on opponents profit functions. Furthermore, a connecting entry generally has a smaller competitive effect than a nonstop entry. When facing the same type of entry, the profit of connecting services is more affected than that of nonstop services. These findings shed light on the question of whether nonstop and connecting markets are relevant to each other. I show that these two service types are considerably related to each other and affect each other, at least from the standpoint of profit determinations Compare Different Research Settings As said in Section 3.1, most existing literature on airline entries tends to either only count for nonstop services or aggregate different service types. Hence, it is important to compare how the estimation results change when I distinguish between service types. 66

82 Table 3.3 presents the estimates of competitive effects under different research settings. In all columns, I assume that the competitive effects of non-southwest carriers are the same, i.e., American, Delta, US Airways, United, MA, and LCC have the same competitive effects. Table 3.3: Competitive Effects: Compare Different Research Settings (I) (II) (III) Aggregate Service Types Only Nonstop Services Distinguish Service Types (Nonstop) SW on Non-SW [ , ] [ , ] [ , ] Non-SW on SW [ , ] [ , ] [ , ] Non-SW on Non-SW [ , ] [ , ] [ , ] Note: I report the smallest connected cube that contains the 95% confidence region. In this table, SW represents Southwest and Non-SW represents non-southwest carriers. Column (I) in Table 3.3 shows the results when connecting services are completely ignored. Column (II) presents the results when different service types are aggregated. This column is consistent with the setting in column (II) in Table 3.1, except that here I allow the competitive effects of a non-southwest carrier on Southwest and on another non- Southwest carrier to be different. In column (III), I report the estimates for nonstop services when nonstop and connecting services are distinguished. The numbers in column (III) are extracted directly from Table

83 Comparing columns (II) and (III), the estimated competitive effects are generally smaller in column (II). This is easy to interpret. As discussed in Section 3.4.2, connecting services usually have smaller competitive effects than nonstop services. So when these two types of services are combined, the aggregated competitive effects will be smaller than the nonstop competitive effects solely. Comparing columns (I) and (III), although both columns examine the competitive effects of nonstop services, the estimates in column (I) are larger than those in column (III). This implies that ignoring connecting services introduces some bias into the estimates. 3.5 Counterfactual In this section, I perform counterfactual experiments to assess the extent to which Southwest s presence influences the entry decisions of other airlines as well as the equilibrium number of firms in the market. In doing, I set the competitive effects of Southwest equal to zero (as if it is removed from the market), recompute the equilibria, and then compare the equilibria with Southwest being removed to the equilibria with Southwest still exists How Does Southwest Influence the Entry Decisions of Each Airline? To investigate if Southwest s presence affects the entry decisions of a given airline, I can look at the changes in the probability of observing that airline in a market before and after Southwest is counterfactually removed. In particular, if the change is positive, i.e., the probability of observing the airline increases with Southwest being removed, then the entry 68

84 of this airline is discouraged by Southwest. The larger in the change of probability, the more Southwest affects the airline s entry decision. Table 3.4 shows the results. Similar to Ciliberto and Tamer (2009), I report the biggest positive change as well as the median change in the mean upper bounds of the probabilities of observing a particular carrier in any possible market structures. The calculation process is like this: For a particular carrier, there are 32 possible market structures in which it can be observed. For example, for Delta, a possible market structure can be one where it is a monopoly, or where it is a duopoly with US Airways, etc.. For each of the 32 market structures, I compute the means of the upper bounds of its probability across all markets. I do this computation both before and after Southwest is counterfactually removed. Then, I take the changes of mean upper bounds for all 32 market structures where the carrier can be observed, and record the biggest positive change as well as the median change. 69

85 Table 3.4: Changes in the Mean Upper Bounds when Southwest is Counterfactually Removed Airline Biggest Positive Change Median Change AA DL US UA MA LCC Note: I report the biggest positive change and the median change in the mean upper bounds of the probabilities of observing a given carrier in any possible market structures. The estimates in column (III) in Table 3.1 are used. I use the estimates in column (III) in Table 3.1 to compute the equilibria where Southwest is counterfactually removed. The actual parameter values that are used are the middle points of the confidence intervals. For example, the coefficient estimate is [0.080, 0.447]; then I use ( = ( )/2) to compute the equilibria. I find that for every carrier, the changes in the mean upper bounds are on the positive side, but the magnitude varies by the identity of the carrier. For example, for LCCs, the probability of seeing it in a market increases by up to 41.4%, with a median increase of 23.2%. This is the largest increase among all carriers. On the other hand, the probability of observing American in a market increases the least, with a median increase of 11.1%. These results indicate that Southwest has a considerable impact on the entry decisions of each airline. Once Southwest is removed, each airline is more likely to enter the market. 70

86 3.5.2 How Does Southwest Influence the Number of Non-Southwest Firms? In this part, I use Berry (1992) s method to evaluate the changes in the mean number of firms in a market. This method contends that, under certain assumptions, it is possible to get a unique prediction of the number of firms that enter a market, although it needs much effort to predict the identities of the firms that enter. The assumptions include homogeneity in firms competitive effects and heterogeneity only in firms fixed costs. As my assumptions regarding non-southwest carriers in column (II) in Table 3.1 satisfies these requirements, I can use those estimates to predict the equilibrium number of non-southwest firms after Southwest is counterfactually removed from each market. Table 3.5 reports the result. The actual mean number of non-southwest firms is when Southwest is still in place. The predicted mean number of non-southwest firms is when Southwest is counterfactually eliminated from each market. This confirms that Southwest has a substantial impact on the number of non-southwest firms. Once it is removed, the equilibrium number increases significantly. One may question why the increase in the number of firms is less than 1. There are a couple of reasons to explain this. First, while in some markets the number of firms increases by more than 1, there are some markets where the number of firms increases by less than 1. So the average increase could be smaller than 1. Second, the predicted number is calculated under Berry (1992) s assumptions. However, the actual number is not. What s more, those assumptions may not hold in realistic settings, where the actual number is generated. So the change in the number of firms should be suggestive. 71

87 Mean Number of Non-Southwest Firms Table 3.5: Mean Number of Non-Southwest Firms Prediction from the model (with Southwest being counterfactually removed) Actual data (with Southwest still in place) Note: The mean number of firms were predicted by applying Berry (1992) s method. The estimates in column (II) in Table 3.1 are used. As a comparison, I do the same calculation for other carriers, and summarize the results in Table 3.6. Table 3.6: Comparison: Mean Number of Firms Carrier X Predicted Mean Number of Non-X Firms (with Carrier X being counterfactually removed) Actual Mean Number of Non-X Firms (with Carrier X still in place) Change in the Mean Number of Non-X Firms AA DL US UA MA LCC SW Note: The predicted mean number of firms were calculated by applying Berry (1992) s method. 72

88 From Table 3.6, it is apparent that for each carrier, removing it will lead to an increase in the number of opponents. However, the degree of increase is different. While removing Southwest leads to the greatest increase, followed by removing LCCs, eliminating medium airlines generates the smallest change. This result, again, proves that Southwest s impact on the air travel market is remarkable. 3.6 Conclusions This paper studies the effect of Southwest Airlines on the profits and entry decisions of other airlines in the US air travel market. I estimate a static game of simultaneous entry closely following Ciliberto and Tamer (2009), and find that Southwest has a notable, negative impact on the profits of non-southwest carriers. In particular, the profits of medium airlines and small low-cost carriers, such as JetBlue and Spirit, are more impacted than those of larger airlines, like American or United. After breaking down airlines entries by service types, I find that Southwest s connecting services have smaller competitive effects than its nonstop services. Finally, through counterfactual experiments, I confirm that Southwest has a remarkable impact on the entry decision of each carrier and the equilibrium number of non-southwest firms in the market. This paper extends the existing literature in several key ways. It is among the few works that examine Southwest s impact on other carriers profits and entry decisions. Moreover, it adds to the literature on the empirical analysis of entry games, by distinguishing connecting and nonstop entries in the game estimation. Lastly, my paper provides evidence to answer the long-existing debate of whether nonstop and connecting 73

89 markets are relevant to each other. From the perspective of profit determination and entry decisions, I find that these two service types are considerably related to each other and affect each other. It is important to discuss the limitations to these findings. First, I impose strong assumptions on the distribution of the unobservables. It would be ideal if these assumptions could be relaxed and the identification problem in the model could still be studied. Second, I only examine the single-stage equilibrium. It is worth trying to include multiple periods and study a repeated game, to see if the results still hold. 74

90 References Anderson, Eric T., and Inseong Song. Coordinating price reductions and coupon events. Journal of Marketing Research 41.4 (2004): Anderson, Michael, and Jeremy Magruder. "Learning from the crowd: Regression discontinuity estimates of the effects of an online review database." The Economic Journal (2012): Bajari, Patrick, Han Hong, and Stephen P. Ryan. Identification and estimation of a discrete game of complete information. Econometrica 78.5 (2010): Bennett, Randall D., and James M. Craun. The airline deregulation evolution continues: The Southwest effect. Office, Berry, Steven T. Estimation of a model of entry in the airline industry. Econometrica: Journal of the Econometric Society (1992): Berry, Steven, and Panle Jia. Tracing the woes: An empirical analysis of the airline industry. American Economic Journal: Microeconomics 2.3 (2010): Bickart, Barbara, and Robert M. Schindler. Internet forums as influential sources of consumer information. Journal of Interactive Marketing 15.3 (2001): Bolton, Gary E., Elena Katok, and Axel Ockenfels. How effective are electronic reputation mechanisms? An experimental investigation. Management Science (2004): Bolton, Gary E., and Axel Ockenfels. The limits of trust in economic transactionsinvestigations of perfect reputation systems Borenstein, Severin. Hubs and high fares: dominance and market power in the US airline industry. The RAND Journal of Economics (1989):

91 Bresnahan, Timothy F., and Peter C. Reiss. Entry in monopoly market. The Review of Economic Studies 57.4 (1990): Brown, Jennifer, and John Morgan. Reputation in online auctions: The market for trust. California Management Review 49.1 (2006): Brueckner, Jan K., Darin Lee, and Ethan S. Singer. Airline competition and domestic US airfares: A comprehensive reappraisal. Economics of Transportation 2.1 (2013): Byers, John W., Michael Mitzenmacher, and Georgios Zervas. The groupon effect on yelp ratings: a root cause analysis. Proceedings of the 13th ACM conference on electronic commerce. ACM, Byers, John W., Michael Mitzenmacher, and Georgios Zervas. Daily deals: Prediction, social diffusion, and reputational ramifications. Proceedings of the fifth ACM international conference on Web search and data mining. ACM, Chen, Jian, Xilong Chen, and Xiping Song. Comparison of the group-buying auction and the fixed pricing mechanism. Decision Support Systems 43.2 (2007): Chen, Yubo, and Jinhong Xie. Third-party product review and firm marketing strategy. Marketing Science 24.2 (2005): Chen, Yubo, and Jinhong Xie. Online consumer review: Word-of-mouth as a new element of marketing communication mix. Management Science 54.3 (2008): Chevalier, Judith A., and Dina Mayzlin. The effect of word of mouth on sales: Online book reviews. Journal of Marketing Research 43.3 (2006): Ciliberto, Federico, and Elie Tamer. Market structure and multiple equilibria in airline markets. Econometrica 77.6 (2009): Daraban, Bogdan, and Gary M. Fournier. Incumbent responses to low-cost airline entry and exit: A spatial autoregressive panel data analysis. Research in Transportation Economics 24.1 (2008): Dholakia, Utpal M. How businesses fare with daily deals: A multi-site analysis of groupon, livingsocial, opentable, travelzoo, and buywithme promotions

92 Dresner, Martin, Jiun-Sheng Chris Lin, and Robert Windle. The impact of lowcost carriers on airport and route competition. Journal of Transport Economics and Policy (1996): Dunn, Abe. Do low-quality products affect high-quality entry? Multiproduct firms and nonstop entry in airline markets. International Journal of Industrial Organization 26.5 (2008): Edelman, Benjamin G., and Michael Luca. Digital discrimination: The case of airbnb.com Edelman, Benjamin, Sonia Jaffe, and Scott Duke Kominers. To groupon or not to groupon: The profitability of deep discounts. Marketing Letters 27.1 (2016): Ekmekci, Mehmet. Sustainable reputations with rating systems. Journal of Economic Theory (2011): Feinberg, Fred M., Aradhna Krishna, and Z. John Zhang. Do we care what others get? A behaviorist approach to targeted promotions. Journal of Marketing Research 39.3 (2002): Gayle, Philip G., and Chi-Yin Wu. Are air travel markets segmented along the lines of nonstop versus intermediate-stop(s) products?. Unpublished paper, Kansas State University (2011). Godes, David, and Dina Mayzlin. Using online conversations to study word-ofmouth communication. Marketing Science 23.4 (2004): Goolsbee, Austan, and Chad Syverson. How do incumbents respond to the threat of entry? Evidence from the major airlines. The Quarterly Journal of Economics (2008): Haile, Philip A., and Elie Tamer. Inference with an incomplete model of English auctions. Journal of Political Economy (2003): Hui, Xiang, et al. Reputation and regulations: evidence from ebay. Management Science (2016): Imbens, Guido W., and Charles F. Manski. Confidence intervals for partially identified parameters. Econometrica 72.6 (2004):

93 Ishii, Jun, Sunyoung Jun, and Kurt Van Dender. Air travel choices in multi-airport markets. Journal of Urban Economics 65.2 (2009): Jin, Ginger Zhe, and Phillip Leslie. Reputational incentives for restaurant hygiene. American Economic Journal: Microeconomics 1.1 (2009): Klemperer, Paul. Entry deterrence in markets with consumer switching costs. The Economic Journal 97 (1987): Li, Xitong, and Lynn Wu. Herding and social media word-of-mouth: Evidence from Groupon Li, Xitong. How does online reputation affect social media endorsements and product sales? Evidence from regression discontinuity design. The 24th workshop on information systems economics (WISE 2013) Li, Xinxin, and Lorin M. Hitt. Self-selection and information role of online product reviews. Information Systems Research 19.4 (2008): Lijesen, Mark G., Piet Rietveld, and Peter Nijkamp. How do carriers price connecting flights? Evidence from intercontinental flights from Europe. Transportation Research Part E: Logistics and Transportation Review 38.3 (2002): Luca, Michael. Reviews, reputation, and revenue: The case of Yelp.com Luca, Michael, and Georgios Zervas. Fake it till you make it: Reputation, competition, and Yelp review fraud. Management Science (2016): Luo, Xueming, et al. Mobile targeting. Management Science 60.7 (2013): Luo, Xueming, et al. Group-buying deal popularity. Journal of Marketing 78.2 (2014): Ma, Chao. "Does Capital Structure Have Opposite Effects on Incumbents Responses to Entry Threat and Actual Entry?." (2016). Manski, Charles F. Partial identification of counterfactual choice probabilities. International Economic Review 48.4 (2007):

94 Manski, Charles F., and Elie Tamer. Inference on regressions with interval data on a regressor or outcome. Econometrica 70.2 (2002): Mayzlin, Dina, Yaniv Dover, and Judith Chevalier. Promotional reviews: An empirical investigation of online review manipulation. The American Economic Review (2014): Mazzeo, Michael J. Product choice and oligopoly market structure. RAND Journal of Economics (2002): Mela, Carl F., Sunil Gupta, and Donald R. Lehmann. The long-term impact of promotion and advertising on consumer brand choice. Journal of Marketing Research (1997): Milgrom, Paul, and John Roberts. Predation, reputation, and entry deterrence. Journal of Economic Theory 27.2 (1982): Moe, Wendy W., and Michael Trusov. The value of social dynamics in online product ratings forums. Journal of Marketing Research 48.3 (2011): Morrison, Steven A. Actual, adjacent, and potential competition estimating the full effect of Southwest Airlines. Journal of Transport Economics and Policy (JTEP) 35.2 (2001): Narasimhan, Chakravarthi. A price discrimination theory of coupons. Marketing Science 3.2 (1984): Raju, Jagmohan S., Sanjay K. Dhar, and Donald G. Morrison. The effect of package coupons on brand choice. Marketing Science 13.2 (1994): Reiss, Peter C., and Pablo T. Spiller. Competition and entry in small airline markets. The Journal of Law and Economics 32.2, Part 2 (1989): S179-S202. Richards, Krista. The effect of Southwest Airlines on US airline markets. Research in Transportation Economics 4 (1996): Saeedi, Maryam. Reputation and adverse selection, theory and evidence from ebay

95 Seim, Katja. An empirical model of firm entry with endogenous product type choices. The RAND Journal of Economics 37.3 (2006): Shaffer, Greg, and Z. John Zhang. Competitive coupon targeting. Marketing Science 14.4 (1995): Sweeting, Andrew. The strategic timing incentives of commercial radio stations: An empirical analysis using multiple equilibria. The RAND Journal of Economics 40.4 (2009): Venkatesan, Rajkumar, and Paul W. Farris. Measuring and managing returns from retailer-customized coupon campaigns. Journal of Marketing 76.1 (2012): Wu, Jiahua, Mengze Shi, and Ming Hu. Threshold effects in online group buying. Management Science 61.9 (2014): Zhu, Feng, and Xiaoquan Zhang. Impact of online consumer reviews on sales: The moderating role of product and consumer characteristics. Journal of Marketing 74.2 (2010):

96 Appendix A: Additional Figures and Tables for Chapter 1 Figure A.1: A Sample Deal on Groupon s Webpage Source: Authors use of Groupon (August 2015). 81

97 Figure A.2: Popular Categories Consumers Purchased on Groupon in 2011 Source: Figure A.3: Popular Categories Daily Deal Sites Offered in 2011 Source: 82