Essays on Strategic Behavior in the U.S. Airline Industry

Home , Carri, Continental Lite, Legacy carrier, Mesaba Airlines, Regional airline, Song (airline)

Dissertation

Presented in Partial Fulﬁllment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University

Kerria Measkhan Tan, B.A., M.A.

Graduate Program in Economics

The Ohio State University

2012

Dissertation Committee:

Matthew Lewis, Advisor James Peck Huanxing Yang c Copyright by

Kerria Measkhan Tan

2012 Abstract

In my first dissertation essay, “Incumbent Response to Entry by Low-Cost Carri- ers in the U.S. Airline Industry,” I analyze the price response of incumbents to entry by low-cost carriers in the U.S. airline industry. Previous theoretical papers suggest that airlines might respond to entry by lowering prices to compete harder for existing customers or they might increase prices to exploit their brand-loyal customers. This paper tests which effect is more prominent in the airline industry. I find that when one of four low-cost carriers enters a particular route, legacy carrier incumbents respond differently than low-cost carrier incumbents to new low-cost carrier entry. Legacy carriers decrease their mean airfare, 10th percentile airfare, and 90th percentile airfare before and after entry by a low-cost carrier. However, low-cost carriers do not significantly alter their pricing strategy. The differing incumbent responses can be attributed to the finding that low-cost carrier entrants tend to match the price set by rival low-cost carriers in the quarter of entry and tend to enter with a lower price than that of legacy carrier incumbents. The results also suggest that entry does not affect price dispersion by incumbent carriers.

Legacy carriers have increasingly outsourced the operation of certain routes to regional airlines over the past decade. My second dissertation essay, “The Influence of Low-Cost Carriers on the Use of Regional Airlines,” investigates how low-cost carriers influence where legacy carriers decide to use regional airlines. I find evidence

ii that legacy carriers are more inclined to switch to regional airlines on routes where a low-cost carrier exists. Moreover, legacy carriers tend to not only decrease average airfares once they start outsourcing but also price match competing low-cost carriers.

However, I do not find evidence that low-cost carriers are effectively deterred from entering routes where a regional airline is present. The results refute the notion that regional airlines can serve as an effective barrier to entry, while suggesting that legacy carriers exploit the more cost-efficient regional airlines in order to lower price and therefore better compete with low-cost carriers.

My third and final dissertation essay, “The Effect of De-Hubbing on Airfares,” studies the price effect of de-hubbing, which occurs when an airline ceases hub operations at an airport. Legacy carriers dramatically decrease both the frequency of

flights and the number of seats offered once it de-hubs an airport, whereas their competitors generally respond by maintaining their capacity level at the airport. As a result, prices could potentially decrease as the market becomes less concentrated or increase because the de-hubbing airline’s capacity reduction diminishes the availability of substitutes. I perform an event study using four cases of de-hubbing at domestic airports between 2001 and 2006 to test which price effect dominates. Not only do average airfares increase after a legacy carrier de-hubs an airport but also the de-hubbing airline and its competitors at that airport tend to increase their prices by a similar percentage. The results suggest that de-hubbing ultimately leads to softer competition between airlines at the de-hubbed airport.

iii This is dedicated to my parents, who taught me to work hard and dream big.

iv Acknowledgments

It seems like only yesterday that Steven Chen, Jason Yau, and I would meet up in the study rooms at Geisel Library on the University of California, San Diego campus in order to study for the math classes that we took together. I remember a particular instance in which I was working through a linear algebra problem using dry erase markers that we “borrowed” from classrooms on campus while taking advantage of the study rooms’ one-way windows. They both stopped me and told me that I have a particular knack for teaching. It was at this moment that I realized that I truly enjoyed these study sessions and that teaching was something that I wanted to pursue for a living. Fast forward about ten years and now I am on the verge of obtaining my Ph.D. in Economics and starting my new job as an Assistant Professor at Loyola

University Maryland. Surely, there have been others along the way that have helped me achieve my academic goals. This serves as an inadequate, yet sincere thank you to the people who have particularly inﬂuenced my life.

One of the biggest reasons why I came to Ohio State was the opportunity to work with my advisor, Matt Lewis. I remember meeting him during my recruitment trip and discussing my senior honors thesis on the Southwest Eﬀect. I cannot imagine having a better advisor than Matt. He gives me the freedom and encouragement to pursue the topics that I am interested in, but is not afraid to let me know if he thinks

I am not being as productive as I should be. I feel like I can brainstorm openly in

v front of him without the fear that he will belittle me if I am incorrect or start to stray towards the wrong path. He is always available when I need to talk and has given me such great advice on research, presentations, and other things. I have the utmost respect for him and will always be greatly indebted to him for the help he has given me over these past six years.

Jim Peck and Huanxing Yang serve as the other two members of my dissertation committee. I learned greatly from them when I took their second year microeconomic theory courses and wanted their help to ensure that the empirical results in my research are rooted in theoretical foundations. Jim, in particular, helped boost my conﬁdence when I was about to go on ﬂy-outs while on the job market. I will miss discussing the English Premier League with Huanxing, especially when it comes to the rivalry between our two beloved teams, Liverpool and Manchester United.

Other faculty members at Ohio State have also been very inﬂuential. Belton

Fleisher became a close conﬁdant and someone that I could always turn to for advice. I always knew that Hajime Miyazaki had my best interests in mind whenever he gave me his opinions regardless of whether they were solicited or not. Bruce Weinberg always believed in my teaching potential and nominated me for several teaching awards. He also served as a teaching reference when I was on the job market. I got to know Don

Haurin well when we formed arguably the best battery in the history of the Ohio

State summer softball league. Similarly, Bill Dupor and I built up a rapport during

Saturday morning pick-up basketball games and was gracious enough to serve as the fourth member of my candidacy committee. Finally, Lucia Dunn vastly improved my interviewing skills, especially when it came to the ﬁve minute talk on my job market paper. I also have a much ﬁrmer hand shake grip thanks to her.

vi I also would not have been able to get through my graduate studies without a great core of friends. I developed a close friendship with Michael Sinkey after all the time we spent watching and playing sports, as well as working at various coffee shops in and around the Columbus area. Jeff Baird and Neil Dalvi ensured that I struck the right balance between work and play. Saif Mekhari was never shy to give me his advice on how to best excel in the program based on his experiences. I would not have been as confident and prepared for job market interviews had I not spent endless hours practicing with Matt Jones and Brandon Restrepo. Matt Dicker and

Dan Gallardo were always a phone call or Google Chat message away whenever I needed to talk to the friends that knew me best. Last but surely not least, Cassandra

Lissey has become my best friend and someone I will love and trust for the rest of my life.

As wonderful as the aforementioned people have been, they cannot compare to the impact that my family has had in my life. Sophia is the best little sister that I could ever ask for. My grandparents on both sides of my family have always spoiled me with love and attention. However, my parents have had the biggest inﬂuence in my life. I would not have been as fascinated with the airline industry if it were not for my dad, who worked for Korean Air for well over twenty years. He gave me the conﬁdence to chase after my dreams and to learn from my mistakes along the way.

He has taught me to always “turn poison into medicine.” My mom is the most selﬂess person I know and is such a great inspiration to me. She has taught me to work hard and to never give up on my goals. I surely would not have been able to accomplish any of this without her love and support. In the end, I just hope that I have made my family proud.

vii Vita

September 30, 1984 ...... Born - Northridge, California.

2006 ...... B.A. Economics, University of California, San Diego. 2007 ...... M.A. Economics, The Ohio State University. 2006-2012 ...... Graduate Teaching Associate, The Ohio State University.

Fields of Study

Major Field: Economics

viii Table of Contents

Page

Abstract ...... ii

Dedication ...... iv

Acknowledgments ...... v

Vita ...... viii

List of Tables ...... xi

List of Figures ...... xiii

1. Incumbent Response to Entry by Low-Cost Carriers in the U.S. Airline Industry ...... 1

1.1 Industry Background and Potential Eﬀect of Entry ...... 5 1.2 Data ...... 9 1.3 Empirical Analysis ...... 11 1.3.1 Estimation Strategy ...... 14 1.3.2 Incumbent Price Response to Entry: Mean Airfare . . . . . 17 1.3.3 Incumbent Price Response to Entry: 10th Percentile Airfare, 90th Percentile Airfare, and Gini Coeﬃcient ...... 22 1.4 Conclusion ...... 30

2. The Inﬂuence of Low-Cost Carriers on the Use of Regional Airlines . . . 32

2.1 Industry Structure ...... 34 2.2 Data ...... 37 2.3 Motivations for Regional Airline Entry ...... 39 2.4 The Eﬀect of Regional Airline Entry on Pricing ...... 52 2.5 Conclusion ...... 63

ix 3. The Eﬀect of De-Hubbing on Airfares ...... 66

3.1 Data ...... 69 3.2 Empirical Analysis ...... 74 3.3 Conclusion ...... 79

Appendices 81

A. Tables for “Incumbent Response to Entry by Low-Cost Carriers in the U.S. Airline Industry” ...... 81

B. Data Construction for “The Inﬂuence of Low-Cost Carriers on the Use of Regional Airlines” ...... 85

C. Robustness Checks for “The Inﬂuence of Low-Cost Carriers on the Use of Regional Airlines” ...... 94

C.1 Pooled Logit Regression Results ...... 94 C.2 Price Matching Windows ...... 95

Bibliography ...... 104

x List of Tables

Table Page

1.1 Frequency of Price Matching by Entrant ...... 12

2.1 Entry by Regional Airlines ...... 42

2.2 Entry by Low-Cost Carriers (Main Results) ...... 45

2.3 Entry by Southwest Airlines vs. Other Low-Cost Carriers ...... 46

2.4 Summary Statistics (Truncated Sample) ...... 49

2.5 Entry by Low-Cost Carriers (Selected Sample) ...... 51

2.6 Price Response to Outsourcing ...... 56

2.7 Frequency of Price Matching Before Outsourcing ...... 58

2.8 Frequency of Price Matching After Outsourcing ...... 59

2.9 Price Matching by Legacy Carriers ...... 62

3.1 Summary Statistics ...... 70

3.2 Capacity Before and After De-Hubbing ...... 73

3.3 Diﬀerence-in-Diﬀerences Results ...... 76

3.4 Difference-in-Difference-in-Differences Results ...... 78

A.1 Summary Statistics ...... 81

xi A.2 Incumbent Price Response to Actual Entry ...... 82

A.3 Legacy Carrier Incumbent Price Response to Actual Entry ...... 83

A.4 Low-Cost Carrier Incumbent Price Response to Actual Entry . . . . . 84

B.1 Regional Airline Partnerships ...... 91

B.2 Summary Statistics (Entry Sample) ...... 92

B.3 Summary Statistics (Legacy Carrier Subsample) ...... 93

B.4 Summary Statistics (Low-Cost Carrier Subsample) ...... 93

C.1 Entry by Regional Airlines (Pooled Logit Model) ...... 96

C.2 Entry by Regional Airlines (Pooled Logit Model with Time Dummies) 97

C.3 Entry by Low-Cost Carriers Airlines (Pooled Logit Model) ...... 98

C.4 Entry by Low-Cost Carriers (Pooled Logit Model with Time Dummies) 99

C.5 Frequency of Price Matching Before Outsourcing: 15% Window . . . 100

C.6 Frequency of Price Matching After Outsourcing: 15% Window . . . . 101

C.7 Frequency of Price Matching Before Outsourcing: 25% Window . . . 102

C.8 Frequency of Price Matching After Outsourcing: 25% Window . . . . 103

xii List of Figures

Figure Page

1.1 Incumbent Response to Entry: Mean Airfare ...... 18

1.2 Legacy Carrier Incumbent Response to Entry: Mean Airfare . . . . . 19

1.3 Low-Cost Carrier Incumbent Response to Entry: Mean Airfare . . . . 20

1.4 Legacy Carrier Incumbent Response to Entry: 10th Percentile Airfare 23

1.5 Low-Cost Carrier Incumbent Response to Entry: 10th Percentile Airfare 24

1.6 Legacy Carrier Incumbent Response to Entry: 90th Percentile Airfare 25

1.7 Low-Cost Carrier Incumbent Response to Entry: 90th Percentile Airfare 26

1.8 Legacy Carrier Incumbent Response to Entry: Gini Coeﬃcient . . . . 27

1.9 Low-Cost Carrier Incumbent Response to Entry: Gini Coeﬃcient . . 27

2.1 Number of Passengers Flown by Operating Carrier ...... 36

3.1 Number of Flights by De-Hubbing Airline ...... 71

xiii Chapter 1: Incumbent Response to Entry by Low-Cost Carriers in the U.S. Airline Industry

When a firm enters a market consisting of a brand-loyal segment and a price- sensitive segment, there are two effects on the incumbents’ pricing strategy: the competitive effect and the displacement effect. Once the entrant enters, the incumbent would continue to decrease prices in order to keep customers because the incumbent

firm’s individual demand curve decreases and becomes more elastic due to an increase in the number of substitutes. Klemperer [24] and Perloff and Salop [28] refer to this as the competitive effect. On the other hand, Rosenthal [31] and Hollander [20] provide the theoretical foundation for the displacement effect, in which entry can actually cause incumbents to increase their prices due to the existence of the two market segments. If entrants are known to cater toward price-sensitive consumers, then incumbents may be best served by increasing prices. In effect, these incumbents will focus their attention on their brand-loyal consumers, who will continue purchasing from them even if an entrant offers lower prices. This strategy will maximize profits whenever the increase in price dominates the effect of the quantity decrease. Since both effects can occur simultaneously, the net effect on prices depends on which effect is more prominent.

1 The growth of several low-cost carriers over the past decade allows for the ability to study whether the competitive eﬀect or the displacement eﬀect is more dominant in the airline industry. This paper focuses on two types of airlines: legacy carriers and low-cost carriers. Legacy carriers are airlines that operate a hub-and-spoke network1 and were founded prior to the industry’s deregulation in 1978, while low-cost carriers typically implement a point-to-point network2 and emerged after deregulation. The purpose of this paper is to study the price response of both legacy carrier and low-cost carrier incumbents when a low-cost carrier enters a new route.

The key result of the paper is that legacy carrier incumbents react differently than low-cost carrier incumbents to entry by low-cost carriers. First, legacy carrier incumbents significantly decrease average one-way airfares the quarter before and the quarter after actual entry by a low-cost carrier. Moreover, low-cost carrier incumbents do not seem to significantly respond to entry by a rival low-cost carrier. Second, I study how the incumbents’ distribution of prices changes due to entry by a low-cost carrier. The 10th percentile prices decrease by about the same amount as the 90th percentile prices so that no significant change occurs to the overall price distribution of the airfares. As such, there is no statistically significant change to price dispersion. Prices decrease all along the distribution of prices almost equally so that price dispersion does not change. Finally, low-cost carrier entrants are likely to enter with an average price that is around the average price of low-cost carrier incumbents and less than that of legacy carrier incumbents. Hence, one reason why low-cost carrier incumbents do not significantly respond to entry by a rival low-cost carrier is because

1A hub-and-spoke network concentrates passengers from several satellite airports (spokes) at a major airport (hub) en route to their ﬁnal destination airport. 2A point-to-point network provides more direct service with fewer connections than a hub-and- spoke network.

2 the entrant tends to match the price of the low-cost carrier incumbent. Meanwhile, there is downward pressure on legacy carrier incumbents’ prices since the entrant sets a price that is likely to be lower than their price. Although both the story based on the competitive effect and the displacement effect seem to be plausible in the airline industry, the results support the claim that the competitive effect dominates the displacement effect.

The empirical analysis regarding incumbent response to entry resembles that in

Goolsbee and Syverson [16], which examines the eﬀect of potential competition by

Southwest Airlines on rivals’ pricing strategies. They find that carriers decrease their prices when they face potential competition with Southwest Airlines, suggesting that incumbents decrease their prices when entry is merely threatened. They estimate a two-way fixed effects model, incorporating time dummies to estimate the effects of potential competition on prices. In effect, they conduct an event study by examining the incumbents’ prices before, during, and after Southwest Airlines enters both airports of a route. I expand upon their work by modifying their estimation strategy so that I can examine the effect of actual competition3 when entry actually occurs by not only Southwest Airlines but also other low-cost carriers.

Gerardi and Shapiro [13] investigate how an airline’s ability to price discriminate on a given route is aﬀected by competition. They ﬁnd that price dispersion decreases

3It is important to note the differences between the different types of competition in the airline industry. Suppose that Southwest Airlines operates at the San Diego International Airport (SAN) and the San Francisco International Airport (SFO). Suppose further that Southwest Airlines services the SAN-SFO route. Actual competition exists when two airlines service the same route at the same time. United Airlines is said to actually compete with Southwest Airlines if United also services the SAN-SFO route at the same time as Southwest Airlines. Now suppose that Southwest Airlines also operates at the Los Angeles International Airport (LAX), but does not service the SAN-LAX route. Potential competition exists when a firm operates at two airports but does not service the route linking both airports that is served by another airline. United Airlines potentially competes with Southwest Airlines if United services the SAN-LAX route at the same time that Southwest Airlines operates at both airports but does not service the SAN-LAX route.

3 with competition, in stark contrast to Borenstein and Rose [4]. Both my paper and these previous papers studies how a firm responds to competition. However, the previous literature is interested in estimating the effect of competition on price dispersion in the airline industry as a whole, whereas this paper examines how price dispersion changes upon entry by a low-cost carrier. Naturally, endogeneity problems arise with these types of studies. I try to minimize the endogeneity problem by looking at entry as opposed to a smooth measure of competition, such as the route- level Herfindahl-Hirschman Index. Moreover, the previous literature assumes that the effect of competition is the same for all airlines, while I allow the effect of entry on price dispersion to vary across different airlines. I am interested in how the incumbents respond to entry by each low-cost carrier.

One of the key results of this paper is that an increase in competition does not lead to a significant change in the incumbent’s price dispersion, which differs from the findings from both Gerardi and Shapiro [13] and Borenstein and Rose [4]. This can be attributed to the differing identification strategy in this paper from the two previous studies, which regress measures for price dispersion on several control variables, including various proxies for competition. Their key findings stem from the sign and strength of the estimated coefficient for the competition variables. The major difference in the analysis of this study to the previous literature is that this paper uses entry as opposed to the route-level Herfindahl-Hirschman Index to identify com- petitor’s response to competition. I analyze the pricing behavior right around entry by performing an event study that captures the immediate effect of competition on price dispersion. By investigating how the Gini coefficient and the tails of the price

4 distribution change around the entry period, this paper is able to shed new light on

the eﬀect of competition on the price distribution of rival ﬁrms.

1.1 Industry Background and Potential Eﬀect of Entry

The competitive structure of the U.S. airline industry has gone through several

changes since deregulation in 1978. Airlines have since experienced more ﬂexibility

in their route network and pricing strategies. It is easier to enter routes that were

once heavily regulated by the Civil Aeronautics Board. As a result, there has been

an inﬂux of entry in the past two decades by low-cost carriers. These airlines include

AirTran Airways, JetBlue Airways, Southwest Airlines, and Spirit Airlines. Low-cost

carriers are able to charge low prices due to their eﬃcient cost structure, beneﬁtting

from the implementation of a point-to-point network, usage of non-unionized labor,

and operation of the same type of aircraft.4 This is in stark contrast to legacy carriers, which implement a hub-and-spoke network, use mostly unionized labor, and operate with a variety of diﬀerent aircrafts. Legacy carriers, which include American

Airlines, Continental Airlines, Delta Air Lines, Northwest Airlines, United Airlines, and US Airways, get their name because they were founded and operated prior to deregulation.

Low-cost carriers have gained market share in the airline industry, particularly in the past decade. In 1997, low-cost carriers flew over 37 million passengers total and accounted for 21.4% of the market share of all passengers flying domestically. In 2007, the number of passengers flying with low-cost carriers increased to over 75 million passengers, resulting in a 36.2% market share of all domestic travel. This growth can

4For example, Southwest Airlines exclusively uses Boeing 737 jets.

5 be partly attributed to the expansion of the low-cost carriers’ route network. Among the top 1000 most traveled routes, there were 494 instances of entry from 1993:Q1 to 2007:Q4 by low-cost carriers, with AirTran Airways entering 224 routes, JetBlue

Airways entering 68 routes, Southwest Airlines entering 150 routes, and Spirit Airlines entering 52 routes. Each route consists of a particular one-way airport-pair. For example, two routes were considered to be entered when Southwest Airlines started

ﬂying from Orlando International Airport to Philadelphia International Airport and vice versa in 2004:Q2. This paper examines four currently operating low-cost carriers

(AirTran Airways, JetBlue Airways, Southwest Airlines, and Spirit Airlines), who have grown substantially over the past two decades and who remain signiﬁcant players in the airline industry today.

Previous research has studied the eﬀect of brand loyalty on the demand for ﬂying.

Borenstein [2] and Gilbert [14] describe how airlines employ marketing schemes in the form of frequent flier programs in order to create and strengthen consumers’ brand loyalty for that particular airline. Consumers enroll in an airline’s frequent flier program and accumulate credit each time they fly with that particular airline. Members can redeem their credit for free flights, upgrades, or other rewards from that airline.

Brand-loyal consumer effectively experience a switching cost upon enrollment in a particular carrier’s frequent flier program. Kim, Shi, and Srinivasan [23] explore how these marketing programs can create two market segments: brand-loyal consumers and price-sensitive consumers.5 Brand-loyal consumers tend to be members of a particular airline’s frequent flier program and become disposed to purchasing more

ﬂights on that airline. Price-sensitive consumers simply look to ﬂy with the airline

5Kim, Shi, and Srinivasan [23] refer to the brand-loyal consumers and price-sensitive consumers as the heavy-user segment and the light-user segment, respectively.

6 charging the lowest price for a given route. Borenstein [2] explains how consumers are inclined to participate in a particular airline’s frequent ﬂier program when they live in that airline’s hub city. For example, Delta Air Lines uses Hartsﬁeld-Jackson

Atlanta International Airport as a hub. Consumers in Atlanta are more likely to not only fly with Delta but also enroll in Delta’s frequent flier program in order to benefit from the wide selection of markets serviced out of Atlanta. This ultimately serves to hook passengers to that particular airline, who can exploit their brand-loyal segment by increasing prices without the fear of losing a significant amount of their market base. In other words, members of an airline’s frequent flier program will continue to purchase from that carrier even if they were charged a higher price because these consumers want to obtain an award after purchasing a certain amount of trips from that airline. Therefore, brand loyalty serves as a switching cost for consumers.

There is empirical evidence for the displacement effect in industries which parallel the airline industry. Using data on the pharmaceutical industry, Grabowski and Ver- non [17] found that entry by generic drugs induced firms selling branded prescription drugs to target consumers with inelastic demand, leaving generic drugs to focus on consumers with more elastic demand. This led to an increase in the price of branded drugs, exemplifying the case when the displacement effect is more prominent than the competitive effect. The airline industry can be considered analogous to the prescription drug market in the sense that brand loyalty is prevalent in both industries with incumbent carriers similar to branded prescription drugs and low-cost carrier entrants akin to generic drugs.

I ask whether incumbent airlines segmented the market in a similar fashion once a low-cost carrier entered a route. The displacement eﬀect dominates if incumbent

7 airlines focus solely on brand-loyal consumers, resulting in an increase of the incumbent’s average price. Incumbents can focus on the brand-loyal segment of the market and allow entrants to service the price sensitive market segment, which would increase price dispersion. However, the competitive eﬀect dominates if entry by low- cost carriers leads to stronger competition for price sensitive consumers, resulting in a decrease in the incumbent’s average price. Furthermore, the decrease in price at the low end of the price distribution could induce incumbents to also decrease prices at the high end of the distribution in order to prevent brand-loyal consumers from becoming more price sensitive. If there was a substantial diﬀerence between full fares and discount fares, then brand-loyal consumers would substitute between competing carriers. This paper sets out to investigate whether competition for price sensitive consumers induces price competition for brand-loyal consumers as well.

The competitive eﬀect also seems to be a credible story behind how incumbents respond to entry by low-cost carriers. Morrison [27] and Vowles [33] both document evidence that incumbents decrease price when Southwest Airlines enters a new market

– the so-called Southwest Effect. This supports the claim that the competitive effect could dominate the displacement effect. However, given the nature of the airline industry, it is plausible that the displacement effect dominates as in the pharmaceutical industry. Therefore, it could be argued that incumbents would increase their price in response to entry by a low-cost carrier. This paper serves to empirically test whether the competitive effect story or the displacement effect story characterizes the entry effect of low-cost carriers in the U.S. airline industry.

8 1.2 Data

The data used for this paper was collected from the Airline Origin and Destina- tion Survey (DB1B), which is published quarterly by the Bureau of Transportation

Statistics. It is a ten percent sample of airline tickets from carriers ﬂying domestic routes. From this database, I collect information on the origin, destination, non-stop distance between endpoints, ticketing carrier, market fare,6 and number of passengers paying a particular market fare. The market fare is the one-way price paid by a passenger for a speciﬁc origin-destination route on a particular carrier.

I eliminate all observations where the market fare is less than $10 or the distance was equal to zero. Observations with an unidentiﬁed ticketing carrier were dropped.

Only observations related to nonstop ﬂights were kept. Observations pertaining to carriers who have less than 1% of the traﬃc on a given route were eliminated. Finally, the sample was restricted to the 1000 routes with the highest number of passengers from 1993:Q1 to 2007:Q4. The dataset contains information on 2.67 trillion passengers over the 15 year time period, which corresponds to roughly 45 million passengers per quarter.7

In order to be identiﬁed as an instance of actual entry, the entrant must have not operated on the route for twelve quarters prior to the quarter of entry and remain on the route for two quarters after entry. The entrant must also service at least 100 passengers in the quarter of entry. Two robustness checks on the identiﬁcation of entry were performed. There are some cases in which two or more low-cost carriers

6Market fare is calculated by the Bureau of Transportation Statistics as the itinerary yield mul- tiplied by the number of miles ﬂown. 7This paper focuses on the eﬀect of six legacy carriers and four low-cost carriers. The total number of passengers serviced by these ten carriers represents 81.9% of the sample.

9 entered a particular route within the sample period. One concern would be that the incumbents would respond to the first entrant, but not necessarily to the second entrant. The first robustness check isolates the first-entrant response by identifying entry only if there was no low-cost carrier servicing the route prior to entry. Another concern may arise if incumbents attempt a predatory pricing scheme in order to deter entry.8 Since my identification rule is that the entrant must remain on the route for only two subsequent quarters after entry, the price response would capture the initial price decrease and subsequent price increase consistent with a predatory pricing scheme. The second robustness check rules out predatory pricing effects by requiring that the entrant must continue to operate on the route for at least eight quarters after entry. The results for each robustness check remain qualitatively consistent with the main results of this paper.

There are three carrier classifications in the DB1B: reporting carrier, operating carrier, and ticketing carrier. Reporting carrier refers to the carrier who submits the information to the Bureau of Transportation Statistics. Operating carrier refers to the carrier who conducted the actual service of air transportation. Ticketing carrier refers to the carrier who issued the passenger the ticket for the flight. In most cases, the three are the same. However, there are cases in which the three are different. For instance, a regional airline could operate the flight under a codeshare agreement with the ticketing carrier. The scope of this paper focuses on how the entry by a low-cost carrier affects the incumbents’ prices. The brand name competition is based at the ticketing-level rather than at the operating-level. Moreover, the consumer’s decision on a reservation is based on the ticketing carrier. In other words, consumers often

8See Elzinga and Mills [8] for details on the Spirit Airlines v. Northwest Airlines predatory pricing case.

10 ignore who the operating carrier is or the fact that the flight is a codeshare flight with another carrier. At the time that the reservation is made, passengers base their purchase on who they purchase the ticket from. For these reasons, I use the ticketing carrier classification here.

1.3 Empirical Analysis

In order to take a preliminary look at incumbent response to entry by low-cost carriers, I analyze the average prices set by incumbents and the entrant in the quarter of actual entry. I report the frequency and percentage that an entrant enters with an average price higher than, equal to, or lower than that set by the incumbents.

In order to do this, I create a price window of $20 around the average price set by each incumbent.9 Price matching occurs if the entrant’s average price is within the incumbent’s $20 price window in the quarter of entry. In order for the entrant to have been determined to set a price higher (lower) than the incumbent’s price, the entrant’s average price must be at least $20 greater than (less than) the incumbent’s price. In order to check the robustness of the results, price windows of $10, $15,

$25, and $30 were calculated. The results are qualitatively similar. The quantitative diﬀerences between price windows stem from the fact that the percentage of price matching increases as the size of the window increases. Table 1.1 summarizes the results using a $20 price window.

Low-cost carrier entrants tend to set an average price that is lower than the legacy carrier incumbents’ average price in the quarter of entry. For example, Southwest

Airlines enters at an average price that is lower than American Airlines’s price on

9The average price in the sample is $170.35 so the $20 price window accounts for roughly a 10% cushion in prices.

11 Table 1.1: Frequency of Price Matching by Entrant

Entrant’s price > incumbent’s price Incumbent American Continental Delta Northwest United US Airways AirTran JetBlue Southwest Spirit AirTran 10 (8.1%) 8 (8.1%) 4 (2.0%) 8 (8.2%) 7 (6.4%) 10 (6.4%) n/a – 10 (25.0%) 0 (0.0%) JetBlue 1 (2.3%) 3 (9.4%) 5 (8.1%) 4 (22.2%) 4 (11.1%) 3 (6.8%) 2 (11.8%) n/a 2 (100.0%) 0 (0.0%) Southwest 6 (7.9%) 4 (5.2%) 5(3.9%) 4 (5.7%) 2 (2.2%) 6 (5.0%) 2 (5.7%) 0 (0.0%) n/a 4 (50.0%) Entrant Spirit 0 (0.0%) 0 (0.0%) 1 (2.2%) 0 (0.0%) 0 (0.0%) 1 (2.5%) 0 (0.0%) – 1 (6.3%) n/a Entrant’s price = incumbent’s price

12 Incumbent American Continental Delta Northwest United US Airways AirTran JetBlue Southwest Spirit AirTran 36 (29.3%) 34 (34.3%) 60 (29.3%) 45 (46.4%) 28 (25.5%) 53 (33.8%) n/a – 25 (62.5%) 6 (75.0%) JetBlue 20 (46.5%) 14 (43.8%) 24 (38.7%) 4 (22.2%) 8 (22.2%) 12 (27.3%) 9 (52.9%) n/a 0 (0.0%) 4 (100.0%) Southwest 26 (34.2%) 25 (32.5%) 48 (37.5%) 35 (50.0%) 23 (25.8%) 49 (40.8%) 19 (54.3%) 2 (50.0%) n/a 4 (50.0%) Entrant Spirit 5 (16.7%) 6 (24.0%) 11 (23.9%) 1 (4.0%) 1 (3.3%) 8 (20%) 2 (33.3%) – 5 (31.3%) n/a Entrant’s price < incumbent’s price Incumbent American Continental Delta Northwest United US Airways AirTran JetBlue Southwest Spirit AirTran 77 (62.6%) 57 (57.6%) 141 (68.8%) 44 (45.4%) 75 (68.2%) 94 (59.9%) n/a – 5 (12.5%) 2 (25.0%) JetBlue 22 (51.2%) 15 (46.9%) 33 (53.2%) 10 (55.6%) 24 (66.7%) 29 (65.9%) 6 (35.3%) n/a 0 (0.0%) 0 (0.0%) Southwest 44 (57.9%) 48 (62.3%) 75 (58.6%) 31 (44.3%) 64 (71.9%) 65 (54.2%) 14 (40.0%) 2 (50.0%) n/a 0 (0.0%) Entrant Spirit 25 (83.3%) 19 (76.0%) 34 (73.9%) 24 (96.0%) 29 (96.7%) 31 (77.5%) 4 (66.7%) – 10 (62.5%) n/a 44 of 76 (57.9%) instances of entry. In other words, Southwest Airlines is likely to undercut American Airlines’s average price in the quarter that they enter that route, conditional on the fact that American Airlines is an incumbent carrier. It is very rare for a low-cost carrier to set a price that is higher than that of a legacy carrier incumbent. In fact, Southwest Airlines sets a price that is at least $20 higher than the average price set by United Airlines on only 7 of 110 (6.4%) of the routes that

Southwest Airlines entered and United Airlines is an incumbent. The results suggest that legacy carrier incumbents may face downward pressure on their prices since they are being undercut by low-cost carrier entrants.

Low-cost carrier entrants tend to price match the average price set by low-cost carrier incumbents. On 19 of 35 (54.3%) of the routes in which Southwest Airlines enters and AirTran Airways is an incumbent, Southwest Airlines ends up setting an average price that is within $20 of AirTran’s price. The results for the other low- cost entrants suggests that they are more likely to price match rival low-cost carriers than incumbent legacy carriers, whose prices tend to be more expensive than low-cost carriers. Therefore, low-cost carrier incumbents might not need to change their prices since there is weak price competition from the entering low-cost carrier. The diﬀering responses by legacy carrier incumbents and low-cost carrier incumbents foreshadow the results presented in this section of the paper.

I study three different responses to entry in order to give a more complete analysis on the entry effect of low-cost carriers. First, I examine how incumbents change their mean airfare before and after actual entry by a low-cost carrier. Second, I investigate how the incumbents’ price distribution of airfares is affected by that entry.

In particular, I look at how the tails of the distribution (10th percentile airfare and

13 90th percentile airfare) change before and after entry. I also examine the entry eﬀect

on the incumbent’s Gini coeﬃcient, which serves as a proxy for price dispersion. The

Gini coeﬃcient is commonly used10 as the measure for fare inequality to reﬂect the

fact that different passengers end up paying different prices for the same flight serviced

by a particular carrier. The Gini coeﬃcient is constructed to be between zero and

one, where inequality increases as the Gini coeﬃcient increases. In other words, a

Gini coeﬃcient of zero represents perfect equality, whereas a Gini coeﬃcient of one

signiﬁes perfect inequality. In the context of the airline industry, a Gini coeﬃcient

of zero means that everyone pays the same price for a speciﬁc route serviced by a

particular carrier, whereas an increase in the carrier’s Gini coeﬃcient shows that there

is more price dispersion on a particular route. Finally, I investigate whether low-cost

carrier entrants set their price below, at, or above the incumbents’ prices when they

enter a new route.

1.3.1 Estimation Strategy

Following the estimation strategy in Goolsbee and Syverson [16], I use a two-

way fixed effects model to identify the entry effects on incumbents’ prices. Four

dependent variables were used, including the logged mean airfare (lnprice), the logged

10th percentile price (lnp10), the logged 90th percentile price (lnp90), and the log-

odds ratio of the Gini coeﬃcient (loddGini).11 Following Gerardi and Shapiro [13], the log-odds ratio of the Gini coeﬃcient is used to account for the fact that the

Gini coeﬃcient is bounded between zero and one. I control for the carrier’s market

10Borenstein and Rose [4], Hayes and Ross [18], and Gerardi and Shapiro [13] all use the Gini coeﬃcient in their estimation strategy.

11 h G i The log-odds ratio of the Gini coeﬃcient (G) is deﬁned as loddGini = ln (1−G) .

14 share on the route, the arithmetic mean of the market share for a carrier at the two endpoints, the Herﬁndahl Index of the route, the arithmetic mean of the Herﬁndahl

Index at the two endpoints, and the geometric mean of metropolitan statistical area

(MSA) population of the two endpoints. The market share variables are both based on the number of passengers. MSA population data were obtained from Local Area

BEARFACTS published by the Bureau of Economic Analysis. I also include carrier- route fixed effects and carrier-year-quarter fixed effects. I cluster the standard errors by route-carrier to account for correlation between a route-carrier combination over time. Table A.1 in Appendix A provides summary statistics.

The basic speciﬁcation is as follows:

12 X yijt = γij + µt + βτ entryj,t0+τ + Xijtα + ijt, (1.1) τ=−12 where yijt is either lnpriceijt, lnp10ijt, lnp90ijt, or loddGiniijt for carrier i on route j in time t, γij is the carrier-route fixed effects, µt is the year-quarter fixed effects,

entryj,t0+τ are the time dummies that specify the lag/forward of the low-cost carrier actually entering a route, and Xijt are the control variables explained above.

The two-way ﬁxed eﬀects model contains 25 time dummies that account for 12 quarters before actual entry to 12 quarters after actual entry, including the actual quarter of entry.12 The estimates of the time lags/forwards of entry show the relative sizes of logged one-way average airfare in the dummy period versus its average value in the excluded period (the thirteenth to sixteenth quarters before entry). Table A.2 in Appendix A summarizes the results of the time dummies for each low-cost carrier entrant in the case where all incumbent carriers (legacy carriers, low-cost carriers,

12It is important to maintain “clean” windows so particular care was exhibited to ensure that no other carrier entered that route within the 25 quarter window. This reduced the number of entered markets in the sample, but would ensure consistent and accurate regression estimates.

15 and other carriers13) are accounted for. Column 3 depicts the results of all incumbent carriers to entry by Southwest Airlines. Since the dummies are mutually exclusive, an incumbent sets a price that is 12.24% lower,14 on average, in the time period immediately after actual entry (t0 + 1) relative to the excluded period (the thirteenth to sixteenth quarters before entry). In other words, the estimates are not additive.

In order to track the price changes by incumbents in response to entry by a particular carrier, I create price paths based on the coefficients of the time dummies in the two-way fixed effects model. The price data is based only on incumbents’ prices so we can interpret the results as the incumbents’ pricing response to entry by a particular carrier. I transform the estimates in order to interpret the coefficients as relative percent change in price.15 The term “relative” can be interpreted as being relative to prices in the excluded time period. Entry occurs at time period 0 with negative time values signifying the quarter before actual entry and positive time values signifying the quarter after actual entry. The solid line is the transformation of the point estimates from the model with the dotted lines representing the 95% confidence interval. If prices are constant throughout (no change in prices by incumbents), then this can be considered as the incumbents not responding to entry by any sort of price changes. If prices are less than zero and statistically significant before actual entry, then this provides evidence for preemptive price cutting.

13Not all carriers are characterized as either a legacy carrier or a low-cost carrier. For example, ATA Airlines is a charter airline yet was an incumbent when Southwest Airlines entered the Los Angeles International Airport to Philadelphia International Airport route in 2004:Q2. 14The percent change relative to the excluded period is found by exp(-0.1306) - 1 = -0.1224. 15The point on the ﬁgure associated with the relative price change by all incumbents a quarter after Southwest Airlines enters would be -0.1224, instead of the actual regression estimate of -0.1306.

16 1.3.2 Incumbent Price Response to Entry: Mean Airfare

Incumbent airlines can potentially respond to entry by low-cost carriers in either one of two ways. The incumbent could decrease their prices before entry occurs in order to enforce the brand loyalty of their consumers, while enhancing their attractiveness to price-sensitive consumers. Prices could continue to drop even after entry occurs as the incumbent responds to the reduction in their respective demand due to an influx of substitutes. Conversely, entry could induce incumbents to actually increase prices so that they could exploit the switching costs inherent in the brand-loyal market segment. This might occur if the effect of an increase in prices can more than offset the effect of a decrease in quantity so that profits ultimately increase. I check to see which of these stories holds true in the airline industry by examining how the incumbents’ mean airfare changes before and after actual entry by a low-cost carrier.

Figure 1.1 illustrates the price paths for all incumbents (legacy carrier, low-cost carriers, and other carriers) in response to entry by either AirTran Airways (Figure

1(a)), JetBlue Airways (Figure 1(b)), Southwest Airlines (Figure 1(c)), and Spirit

Airlines (Figure 1(d)). These price paths essentially graph out the time dummies from the regression results summarized in Table A.2 in Appendix A. Again, these estimates can be interpreted as the percentage price change relative to the excluded period (the thirteenth to sixteenth period before entry). Morrison (2001) and Vowles

(2001) both examine price changes the quarter before and the quarter after actual entry by Southwest Airlines. They ﬁnd that incumbents signiﬁcantly decrease their prices before and after entry by Southwest Airlines. Thus, I focus my analysis on the price response in the quarter before to the quarter after actual entry occurs. However,

17 I broaden their analysis to examine the type of price eﬀect induced by entry by other low-cost carriers.

(a) Entrant: AirTran Airways (b) Entrant: JetBlue Airways

Figure 1.1: Incumbent Response to Entry: Mean Airfare

Each price path in Figure 1.1 shows the percentage price change relative to the excluded period (the thirteenth to sixteenth period before actual entry) for the twelve quarters before entry to the twelve quarters after entry. According to Figure 1(c), incumbents do not signiﬁcantly change their average prices until Southwest Airlines actually enters the route. Moreover, incumbents’ mean airfare steeply decreases in the quarter of entry and the ﬁrst quarter after entry. In fact, the solid line shows that incumbents’ prices decrease 12.24% on average in the quarter following entry by

Southwest Airlines. Based on the 95% conﬁdence intervals (the dotted lines), Figure

1(c) shows that this decrease is statistically signiﬁcant. This key result corroborates the previous ﬁndings in the literature. Namely, incumbents decrease their prices in

18 response to entry by Southwest Airlines. However, I want to determine whether this eﬀect is induced by other low-cost carrier entrants.

Further examination of the other price paths in Figure 1.1 shows that incumbents tend to decrease their mean airfares the quarter before entry, the quarter of entry, and the quarter after entry. Southwest Airlines had the largest average entry eﬀect, with the aforementioned result of inducing incumbents to decrease prices by 12.24%, on average, the quarter after actual entry. Other low-cost carriers had similar, yet weaker eﬀects. AirTran Airways induced a decrease of 10.81%, while incumbents also reacted to entry by JetBlue Airways and Spirit Airlines, but only by a modest amount of 5.57% and 5.36%, respectively. Nevertheless, each low-cost carrier induced incumbents to decrease their prices before and after actual entry.

(a) Entrant: AirTran Airways (b) Entrant: JetBlue Airways

Figure 1.2: Legacy Carrier Incumbent Response to Entry: Mean Airfare

19 (a) Entrant: AirTran Airways (b) Entrant: JetBlue Airways

Figure 1.3: Low-Cost Carrier Incumbent Response to Entry: Mean Airfare

The results from Table 1.1 suggest that it is worthwhile to examine the variations in the entry response of legacy carrier and low-cost carrier incumbents. Figure 1.2 shows the relative price response of legacy carrier incumbents to entry by low-cost carrier, whereas Figure 1.3 shows the price response of low-cost carrier incumbents.

Figures 1.2 and 1.3 correspond with the regression results in Tables A.3 and A.4 in

Appendix A, respectively.

Based on the price paths in Figure 1.2, legacy carriers respond to entry by low- cost carriers by dramatically decreasing their average airfares. In fact, there is a more pronounced price drop than the eﬀect shown in Figure 1.1, which considers all incumbents servicing the entered route when the entrant actually enters. Southwest

Airlines induces incumbents to decrease their average prices by 13.09%. However,

AirTran Airways induces an even stronger eﬀect than that of Southwest as incumbents

20 cut their mean airfare by an average of 13.31% the quarter after AirTran Airways actually enters a route. Entry by JetBlue Airways and Spirit Airlines invokes legacy carrier incumbents to decrease their prices by 7.07% and 7.98%, respectively. All of these eﬀects are larger than their respective eﬀect implied by Figure 1.1.

The existing literature focuses on the strong entry eﬀect of Southwest Airlines.

Over the past decade, other low-cost airlines have entered the industry and are currently major carriers in the industry. JetBlue Airways and AirTran Airways demon- strate how other low-cost carriers can mirror the entry effects exhibited with South- west Airlines. The upshot is that the Southwest Effect can no longer be considered as a special case relevant to one particular airline, particularly as it pertains to legacy carrier incumbents. Rather, the entry effect pertains to low-cost carriers in general.

Figure 1.3 shows that low-cost carrier incumbents do not signiﬁcantly alter their mean airfare when either a low-cost carrier enters the route. These price paths are in stark contrast with Figure 1.2, where it was shown that legacy carrier incumbents signiﬁcantly decrease their mean price. Therefore, legacy carrier incumbents (Figure

1.2) react diﬀerently than low-cost carrier incumbents (Figure 1.3) in their response to entry by a low-cost carrier.

The diﬀering response by legacy carriers and low-cost carriers can be rationalized by the frequency of price matching by low-cost carrier entrants. Recall that Table 1.1 shows that low-cost carrier entrants are likely to undercut legacy carrier incumbents, yet match the price of low-cost carrier incumbents. The competitive eﬀect story predicts that incumbents would decrease their price after entry occurs in response to an increase in price competition from the entrant, whereas the displacement story would induce a price increase by the incumbent. The results support the claim that

21 the competitive eﬀect story applies to legacy carrier incumbents; however, low-cost carrier incumbents are not susceptible to either eﬀect.

1.3.3 Incumbent Price Response to Entry: 10th Percentile Airfare, 90th Percentile Airfare, and Gini Coeﬃcient

Different passengers who fly on the same flight may pay markedly different fares.

As such, it is possible that entry by a low-cost carrier could aﬀect the price distribution of airfares set by incumbent carriers. Borenstein and Rose [4] show that price dispersion increases as routes become more competitive. The intuition is that entry can induce incumbents to decrease their discount price (i.e. the 10th percentile airfare) to attract price-sensitive consumers, while keeping their full-fare price (i.e. the

90th percentile airfare) high, resulting in an increase in price dispersion. Gerardi and

Shapiro (2009) conclude that price dispersion actually decreases when there is more competition in the route. The intuition here is that an increase in competition erodes the incumbent carriers’ market power, which mitigates the ability for these airlines to eﬀectively price discriminate. Therefore, price dispersion is smaller in markets that are more competitive. In this section, I discuss the eﬀect of entry by low-cost carriers on the incumbents’ price distribution of airfares.

The price paths in this section are constructed based on regression results using either the logged 10th percentile airfare, logged 90th percentile airfare, or the log-odds ratio of the Gini coeﬃcient as the dependent variable. As in Gerardi and Shapiro

[13], the 10th percentile airfare is intended to control for the eﬀect on discount tickets, whereas the 90th percentile airfare proxies for full-fare prices. These two dependent variables eﬀectively account for changes at the tails of the price distribution. The

Gini coeﬃcient measures the price dispersion of a carrier’s prices on a speciﬁc route

22 in a particular time period, and is between zero and one. Since the Gini coefficient emphasizes the middle of the price distribution, a full analysis of the entry effect on incumbents’ price distribution involves analyzing the effects on the tails as well.16

(a) Entrant: AirTran Airways (b) Entrant: JetBlue Airways

Figure 1.4: Legacy Carrier Incumbent Response to Entry: 10th Percentile Airfare

Figure 1.4 shows that legacy carrier incumbents slash their 10th percentile prices immediately before and immediately after entry. In the quarter after Southwest Air- lines actually enters a route, legacy carrier incumbents decreased their 10th percentile prices by 11.56%, on average, relative to the excluded period (the thirteenth to sixteenth quarter before entry). Other low-cost entrants induced similar eﬀects, with legacy carrier incumbents dropping prices by an average of 8.09%, 7.49%, and 7.69%

16See Gerardi and Shapiro [13] for a more in-depth discussion on the pros and cons of the Gini coeﬃcient.

23 when AirTran Airways, JetBlue Airways, and Spirit Airlines entered the route, respectively. These results suggest that legacy carrier incumbents signiﬁcantly decrease their discount prices in response to entry by a low-cost carrier.

(a) Entrant: AirTran Airways (b) Entrant: JetBlue Airways

Figure 1.5: Low-Cost Carrier Incumbent Response to Entry: 10th Percentile Airfare

The analysis from Section 1.3.2 shows that the response by legacy carrier incumbents diﬀers from that of low-cost carrier incumbents, as far as changes to mean airfare is concerned. Figure 1.5 shows that the low-cost carriers do not signiﬁcantly alter their 10th percentile prices in response to entry by a rival low-cost carrier. Just as with mean airfares, the results for 10th percentile airfares show a stark contrast in the response by low-cost carriers from that of legacy carriers to entry by a low-cost carrier.

Figure 1.6 indicates legacy carrier incumbents decrease their full fare prices, on average. Southwest Airlines induces legacy carrier incumbents to decrease their 90th

24 (a) Entrant: AirTran Airways (b) Entrant: JetBlue Airways

Figure 1.6: Legacy Carrier Incumbent Response to Entry: 90th Percentile Airfare

percentile prices by 14.86%, while legacy carriers decreased their 90th percentile price by 14.68% and 14.05% in the quarter after actual entry by AirTran Airways and Spirit

Airlines, respectively. Interestingly, these eﬀects are of similar magnitudes than that on the 10th percentile prices. Full fare prices charged by legacy carriers decreased by 3.35%, on average, in response to entry by JetBlue Airways. Although this is not as strong as their eﬀect on 10th percentile prices, entry by JetBlue Airways still put downward pressure on the incumbents’ full fare prices.

The results of the eﬀect of entry by a low-cost carrier on low-cost carrier incumbents’ full fare prices are illustrated in Figure 1.7. In contrast to the results for legacy carrier incumbents, low-cost carriers do not strongly respond to entry. The analysis on entry by Southwest Airlines continues to show the pronounced eﬀect that they have on incumbents’ prices. As with mean airfare and discount prices, low-cost

25 (a) Entrant: AirTran Airways (b) Entrant: JetBlue Airways

Figure 1.7: Low-Cost Carrier Incumbent Response to Entry: 90th Percentile Airfare

carrier incumbents do not alter their full fares in the same manner as legacy carrier incumbents in response to entry by low-cost carriers.

In order to examine the overall eﬀect of entry on the incumbent’s price distribution,

I calculated the log-odds ratio of the carrier’s Gini coeﬃcient, which measures the carrier’s price dispersion at the route level. It may be the case that there is price polarization, which would cause the Gini coeﬃcient to increase. I use the log-odds ratio as the dependent variable in Equation 1.1 and plot the transformation of the time dummies. Figures 1.8 and 1.9 can be interpreted as the evolution of the incumbent’s price dispersion in the entered route over time.

Figures 1.8 and 1.9 show that the Gini coeﬃcient for the prices set by legacy carriers and low-cost carriers, respectively, do not signiﬁcantly respond to entry by

26 (a) Entrant: AirTran Airways (b) Entrant: JetBlue Airways

Figure 1.8: Legacy Carrier Incumbent Response to Entry: Gini Coeﬃcient

(a) Entrant: AirTran Airways (b) Entrant: JetBlue Airways

Figure 1.9: Low-Cost Carrier Incumbent Response to Entry: Gini Coeﬃcient

27 a low-cost carrier.17 Again, the percentage change in the Gini coefficient is relative to the excluded period (the thirteenth to sixteenth period before entry). Recall that legacy carrier incumbents decrease both their 10th percentile and 90th percentile prices, on average, in response to entry by a low-cost carrier. Although the 90th percentile prices decrease more than the 10th percentile prices, the total effect on the Gini is negligible. In other words, the Gini coefficient for legacy carriers does not significantly change immediately before and after entry because both tails of the price distribution decrease. The mean average airfare decreases as well, indicating downward pressure on the entire price distribution. On the other hand, low-cost carriers do not significantly respond to entry by a rival low-cost carrier as there is no effect on either the mean airfare, the 10th percentile airfare, or the 90th percentile airfare. Consequently, there is no significant effect on price dispersion by low-cost carrier incumbents.

Both Borenstein and Rose [4] and Gerardi and Shapiro [13] determine the effect of competition on price dispersion by estimating regression models consisting of a transformation of the Gini coefficient18 as the dependent variable, while the inde- pendent variables include a proxy for competition. These papers are interested in the estimated sign and significance of the competition variables on price dispersion.

Their results suggest a significant, yet contrasting effect. Gerardi and Shapiro [13] attribute their differing results to the fact that they use panel data, while Borenstein and Rose [4] use cross-sectional data. They argue that the results in Borenstein and

17As a robustness check, I run the two-way fixed effects regression model using the interquartile range of prices as the dependent variable. This serves as a robustness check for the Gini coefficient since the interquartile range would provide further information about the shape of the price distribution. The results support the analysis on the Gini coefficient. 18Borenstein and Rose [4] use logged Gini coefficient as their dependent variable, while Gerardi and Shapiro [13] use the log-odds ratio of the Gini coefficient.

28 Rose [4] suﬀer from omitted-variable bias, which they ﬁx by including route-carrier

fixed effects. I find that the Gini coefficient does not significantly change due to entry by a low-cost carrier, implying that increased competition from entry by low-cost carriers has no effect on price dispersion. Thus, the results of this paper differ from the key findings in Borenstein and Rose [4] and Gerardi and Shapiro [13]. Borenstein and Rose [4] find that price dispersion increases when there is more competition. This would occur if entry induces incumbents respond to entry by decreasing their 10th percentile prices, while keeping their 90th percentile prices high, suggesting that the path for the Gini coefficient should be significantly positive around the time of entry.

However, Gerardi and Shapiro [13] find that an increase in competition would lead to a decrease in price dispersion. My results would have corroborated their finding if the path for the Gini coefficient was negative around the time of entry, suggesting that an increase in competition due to the entry by a low-cost carrier would induce a higher degree of price equality.

The identification strategy used in this paper is different than the strategy used by Borenstein and Rose [4] and Gerardi and Shapiro [13], which could explain for the dissimilar result. My approach is similar to an event study, in which I identify individual events of entry and estimate the immediate effect of entry on incumbents’ prices. I examine how incumbents react differently to different low-cost carrier entrants around the time of actual entry. I also analyze how the incumbent response differs depending on whether the incumbent is a low-cost carrier and legacy carrier.

However, Borenstein and Rose [4] and Gerardi and Shapiro [13] are interested in a more general industry-wide eﬀect of competition on incumbent prices. According to

Borenstein and Rose [4], competition is aﬀected by a change in the Herﬁndahl Index

29 or the total number of flights on the route, whereas Gerardi and Shapiro [13] identify a change in competition by a change in the Herfindahl Index or the total number of carriers servicing the route. The results in this paper show that price dispersion does not significantly change immediately following an increase in competition, specifically when a low-cost carrier enters a new route.

1.4 Conclusion

This paper studies the incumbent response to entry by low-cost carriers. Legacy carrier incumbents tend to decrease their average airfare, discount fares, and full fare price before and after entry by a low-cost carrier. However, low-cost carriers do not significantly alter their prices in response to entry by a rival low-cost carrier. In both cases, the Gini coefficient does not significantly change, implying that entry does not affect the incumbent’s price dispersion. This paper extends upon the work by

Goolsbee and Syverson [16] by going further to identify how incumbents respond to entry by not only Southwest Airlines, but also other prominent low-cost carriers. The key punch line to this paper is that although the strongest entry response occurs when

Southwest Airlines enters a new route, legacy carrier incumbents tend to respond in a similar, yet weaker fashion to other low-cost carriers.

The results suggest that competition does not induce an immediate impact on price dispersion. Entry by a low-cost carrier induces legacy carrier incumbents to decrease their 10th percentile, 90th percentile, and mean airfares. Since legacy carrier incumbents decrease prices all along the price distribution, then there was no net change to the overall dispersion of prices. Low-cost carrier incumbents do not alter their price dispersion as they do not signiﬁcantly respond to entry by a rival low-cost

30 carrier. These ﬁndings extend the results in Borenstein and Rose [4] and Gerardi and Shapiro [13], who focus on the eﬀect of competition on price dispersion in the industry as a whole.

Legacy carrier incumbents react diﬀerently to entry by low-cost carriers than low- cost carrier incumbents. Low-cost carrier entrants tend to undercut legacy carrier incumbents, while matching the prices of low-cost carrier incumbents. Legacy carriers decrease their prices in response to the low prices set by a low-cost carrier entrant.

This downward pressure on prices was not experienced by low-cost carrier incumbents due to the weak price competition that ensued between rival low-cost carriers. This paper sheds light on a previously unknown phenomena: the strategic interaction between low-cost carrier entrants and rival low-cost carrier incumbents.

31 Chapter 2: The Inﬂuence of Low-Cost Carriers on the Use of Regional Airlines

Legacy carriers have recently become more reliant on regional airlines as an important feeder of passengers within their route network. In fact, the number of passengers

ﬂown by regional airlines has increased from 6.56 million in 1998 to 36.5 million in

2009, which is largely due to a growth in outsourcing by legacy carriers during this time period. Under these arrangements, the planes are owned by the regional airlines, but are painted to resemble the legacy carrier’s fleet. Pilots and flight attendants are employed by the regional airline, yet the legacy carrier is responsible for ticketing and operations at the airport. Legacy carriers contract with regional airlines primarily because of their cost advantage.19 This paper examines the factors that contribute to the “make-or-buy” decision regarding the operation of a route. In particular, I investigate how low-cost carriers influence whether legacy carriers operate a route themselves or outsource to a regional airline.

I focus on two potential explanations that could contribute to the growing use of regional airlines. Legacy carriers might outsource to regional airlines as a response to current competition with low-cost carriers. As a result of lower operating costs, legacy carriers may ﬁnd it easier to decrease average airfare once they switch the operation of

19Hirsch [19] found that senior pilots and ﬂight attendants at United Airlines make 80 percent more and 32 percent more than their counterparts at regional airlines, respectively.

32 a route from their own ﬂeet to a regional airline. Thus, a major factor in the growing use of regional airlines could be motivated by increased price competition with existing low-cost carriers. Alternatively, a legacy carrier might switch to a regional airline in an attempt to erect a barrier to entry against prospective low-cost carriers.20 By using a regional airline, legacy carriers could signal that a previously attractive route would no longer be proﬁtable to a prospective low-cost carrier entrant. The purpose of this paper is to study the role of low-cost carriers on the decision of where to use regional airlines.

My ﬁndings are largely consistent with the idea that regional airlines are used to help legacy carriers better compete with low-cost carriers. Legacy carriers are more likely to start using regional airlines on routes where low-cost carriers are present.

When the switch to regional airlines occurs, the legacy carrier typically lowers average price to match that of competing low-cost carriers. In contrast, I ﬁnd no evidence that the likelihood of entry by low-cost carriers is reduced where legacy carriers have outsourced to regional airlines. Therefore, the evidence is consistent with the notion that outsourcing is a competitive pricing response to current competition with low- cost carriers rather than an attempt to preclude future entry by prospective low-cost carriers.

Other studies have examined competitive responses to low-cost carriers. Gools- bee and Syverson [16] ﬁnd that incumbents decrease their price prior to entry by

Southwest Airlines, a low-cost carrier. Gerardi and Shapiro [13] suggest that legacy carriers experience a decrease in their ability to price discriminate when facing more

20Forbes and Lederman [10] mention that outsourcing the operation of a route to a regional airline could serve as an eﬀective barrier to entry to low-cost carriers. Similarly, Borenstein [3] conjectures that partnerships between legacy carriers and regional airlines can increase the cost of entry at airports where the two airlines connect.

33 competition, especially from low-cost carriers. However, the existing literature has not formally studied the use of regional airlines as a potential strategic response to competition from low-cost carriers. By examining how low-cost carriers inﬂuence where legacy carriers decide to use regional airlines, this paper contributes to the literature on strategic interaction and competition in the U.S. airline industry.

2.1 Industry Structure

Before the Airline Deregulation Act of 1978, regional airlines operated as com- muter airlines, servicing thin and short-haul routes. At the time, the Civil Aeronau- tics Board heavily regulated price and entry in the airline industry. Airlines were allowed to set high prices on long-haul routes, which cross-subsidized proﬁt losses made on low-margin short-haul routes. Regional airlines, however, were exempt from regulation as long as their ﬂeet contained planes below a certain size.21 As such, they operated independently from the major airlines at the time.

After the airline industry became deregulated in 1978, legacy carriers22 altered their route structure by developing hub-and-spoke networks, in which passenger traﬃc is concentrated through certain airports in the United States. Under this system, legacy carriers have to decide whether to operate a route themselves or outsource to a regional airline.23 Although passengers purchase their ticket from the legacy carrier, the contracted regional airline supplies the aircrew and ﬂeet used in the operation of

21The size limit eﬀectively limited regional airlines to planes with 20 to 30 seats. 22Legacy carriers get their name from the fact that they have existed prior to deregulation. 23Another recent trend is called codesharing, in which a legacy carrier operates a route on behalf of a rival legacy carrier. See Goetz and Shapiro [15] for a detailed analysis on codesharing.

34 the ﬂight. Legacy carriers typically use regional airlines on short-haul routes linking a legacy carrier’s spoke airport to its hub airport, and vice versa.

Legacy carriers have become more attracted to outsourcing to regional airlines in part because regional airlines have changed the type of aircraft they fly from turbo- props to regional jets, which has increased range, speed, and passenger capacity. As a result, the use of regional airlines by legacy carriers has drastically expanded over time as the total number of routes that legacy carriers outsourced to regional airlines increased from 1,917 in 1998 to 17,111 in 2009. Figure 2.1 shows that 213 million and 6.56 million passengers flew on flights operated by legacy carriers and regional carriers, respectively, in 1998. This corresponds to a market share of 67.3% for legacy carriers and 2.1% for regional airlines. While the number of passengers flown by legacy carriers decreased to 146 million passengers (44.7% market share) in 2009,

36.4 million passengers (11.1% market share) ﬂew with a regional airline operating on behalf of a legacy carrier, an increase of nearly 455% over the twelve year period.

Although legacy carriers continue to increase the amount of routes that are outsourced to regional airlines, this relationship has been partly restricted due to “scope clauses” in the labor agreements between legacy carriers and labor unions. These clauses limit the number of planes that can be operated by a regional airline on behalf of the legacy carrier. Therefore, legacy carriers are faced with a trade-oﬀ when they decide whether to switch to a regional airline as this would take away the opportunity of using that regional airline on a diﬀerent route.

The rise of low-cost carriers is another outcome of a deregulated airline industry.

Since deregulation made it easier for new airlines to enter the industry, start-up airlines emerged that found ways to lower the cost of available seat mile relative to

35 Figure 2.1: Number of Passengers Flown by Operating Carrier

The number of passengers is calculated based on whether the operating carrier was a regional airline, low-cost carrier, or legacy carrier. As such, the number of passengers is not determined by the ticketing carrier. For example, regional airlines would be credited for the number of passengers it flew on behalf of a legacy carrier. Legacy carriers only get credit for passengers who flew on flights that they operated themselves.

that of legacy carriers. These low-cost carriers decreased the cost of operation by

using a point-to-point network, non-unionized labor, and a ﬂeet consisting of the

same aircraft.24 Legacy carriers established low-cost “airline within an airline”25 in order to counter the inﬂux of new airlines, but they have all since been discontinued because they quickly became a ﬁnancial burden to the legacy carrier.

As with regional airlines, low-cost carriers have experienced a remarkable growth in the number of passengers ﬂown between 1998 and 2009. Figure 2.1 illustrates that the number of passengers ﬂown by low-cost carriers has increased dramatically from

24Southwest Airlines, for example, operates only Boeing 737 planes, which decreases the cost of maintenance and inventory. 25Examples include Continental Lite (Continental Airlines), Song (Delta Air Lines), and Ted (United Airlines).

36 58.5 million (18.5% market share) in 1998 to 130 million (39.8% market share) in 2009.

During this time period, there have been 940 instances of entry26 by low-cost carriers into routes with a maximum distance of 1,500 miles, with AirTran Airways entering

424 routes, JetBlue Airways entering 136 routes, Southwest Airlines entering 322 routes, and Spirit Airlines entering 58 routes. The expansion of the low-cost carrier’s route network largely explains their recent growth and their emergence as a major inﬂuence in the U.S. airline industry. This paper studies how price competition with low-cost carriers could induce legacy carriers to outsource the operation of a route to regional airlines.

2.2 Data

In order to investigate how legacy carriers use regional airlines to better compete with low-cost carriers, I use data from three main sources. The main dataset used in this paper is the Airline Origin and Destination Survey (DB1B), which is published quarterly by the Bureau of Transportation Statistics. It is a ten percent survey of domestic air travel and contains data on the origin, destination, non-stop distance between endpoints, ticketing and operating carrier,27 market fare,28 and number of passengers paying a particular market fare. I augment this data with monthly data on the number of delayed ﬂights from the Airline On-Time Performance Data set,

26Each instance is defined by entry into a one-way airport pair. For example, when Southwest started operating the route between Detroit Metro Airport and Philadelphia International Airport in 2004, two routes are entered: the route from Philadelphia to Detroit, and the route from Detroit to Philadelphia. 27The key distinction between the ticketing carrier and the operating carrier is that the ticketing carrier is the airline that the passenger purchased the ticket from, whereas the operating carrier is the airline that is in charge of the aircrew and fleet used on the flight. 28Market fare is calculated by the Bureau of Transportation Statistics as the itinerary yield mul- tiplied by the number of miles flown. Other charges, such as baggage fees, priority seating fees, or the cost of food and beverage purchased on the flight, are not accounted for in the market fare.

37 also from the Bureau of Transportation Statistics. Finally, I use yearly data from the

Local Area Personal Income tables on population, per capita personal income, and

personal income by major source and earnings at the metropolitan statistical area-

level, which are created and distributed by the Bureau of Economic Analysis. I use

the personal income by major source and earnings dataset to obtain information on

both accommodation and nonfarm earnings for metropolitan statistical areas (MSA).

Data from 1998 to 2009 are collected from each of the three data sources.

The following steps are undertaken to clean the data. First, I eliminate all ob-

servations where the distance was equal to zero or the market fare is less than $10.

Observations with an unidentiﬁed ticketing carrier are also dropped. Only observa-

tions related to nonstop ﬂights are kept. I then limit the sample to routes within

the continental United States with a maximum distance of 1,500 miles since regional

airlines would not be used on longer routes and restrict the sample to the 2,500 routes

with the highest number of passengers from 1998 to 2009. This eﬀectively removes

observations on routes involving airports in Alaska, Hawaii, and Puerto Rico. I drop

routes that are never serviced by a legacy carrier in order to focus on routes where

there is the potential for strategic behavior between legacy carriers29 and low-cost carriers.30 In some cases, data on the number of delayed ﬂights or accommodation earnings are not reported. Routes with incomplete information on either of these two variables are eliminated. The data used in this paper are discussed in greater detail in Appendix B.

29The legacy carriers studied in this paper are American Airlines, Continental Airlines, Delta Air Lines, Northwest Airlines, United Airlines, and US Airways. 30The low-cost carriers studied in this paper are AirTran Airways, JetBlue Airways, Southwest Airlines, and Spirit Airlines.

38 2.3 Motivations for Regional Airline Entry

This section analyzes two possible mechanisms through which competitive pres- sures from low-cost carriers could inﬂuence the decision to outsource to regional airlines: 1) outsourcing as a response to current competition with low-cost carriers and

2) outsourcing to erect a barrier to entry against low-cost carriers. I examine the ﬁrst explanation by testing whether legacy carriers are more likely to switch to a regional airline on routes where low-cost carriers are present. This is done using a two-way

fixed effects logit regression of entry by regional airlines on the number of low-cost carriers operating on a route, as well as other control variables. Second, I estimate the effect of regional airlines on entry by low-cost carriers in order to test whether the presence of a regional airline reduces the likelihood that low-cost carriers enter that route. In other words, I investigate whether legacy carriers outsource to regional airlines where they intend to preclude future entry by regional airlines.

Observations from the DB1B and the Airline On-Time Performance Data are aggregated to the year level so that the ﬁnal dataset contains route-year observations.

Entry is identified when an airline starts servicing a route and remains on that route for at least two consecutive quarters. In some cases, airlines are seen in the DB1B to operate on a particular route only to drop out for a quarter and reappear in the subsequent quarter. This is not an example of actual entry but represents an issue with the DB1B being a ten percent sample of airline tickets. Nevertheless, this problem was resolved by qualifying entry when the carrier did not service the route in question for at least four quarters before the identified quarter of entry. The carrier must have also flown at least 100 passengers on the entered route in the quarter of

39 entry. The ﬁnal dataset contains 12,790 observations on 1,161 routes from 1998 to

2009.

There are three types of control variables: market variables, demographic vari-

ables, and competition variables. Market variables include the natural log of the

number of passengers on a route (lndensity) and the percentage of ﬂights on a route

that were delayed at least 15 minutes (pdelay). Demographic variables include the maximum of the ratio of accommodation earnings to nonfarm earnings for each endpoint on a route (tourism), as well as the geometric mean of the population (pop) and

per capita income (income) of the MSA where the origin and destination airports are

located. Finally, I include competition variables to control for the maximum market

share of a servicing airline on that route (maxshare) and the route-level Herﬁndahl-

Hirschman Index (HHIroute). I also control for the number of competing airlines by including the number of legacy carriers (nLEG), low-cost carriers (nLCC), regional airlines (nREG), and other airlines (nOT HER) operating on that route. Summary statistics for this dataset and a detailed description on variable construction can be found in Appendix B.

I use a fixed effects approach to exploit the panel structure of my data in order to test whether the presence of low-cost carriers affect the likelihood that a legacy carrier outsources to a regional airline while controlling for time-invariant, route- specific factors. I am interested in routes where legacy carriers have a choice to operate with a regional airline or not so the data only includes routes where legacy carriers compete at some point in the sample time period. Legacy carriers typically employ a regional airline on routes that they previously operated themselves. Entry in this sense is defined when a legacy carrier starts using a regional airline where it

40 had previously served as both the ticketing and operating carrier. Once a regional

airline has entered the route, the legacy carrier remains as the ticketing carrier, but

the regional airline becomes the operating carrier. Thus, I construct REGentry as a

dependent variable, which assumes the value of 1 when a legacy carrier switches to a

partnered regional airline, and 0 otherwise.

The specification for the two-way fixed effects logit regression model is as follows:

REGentryi,t+1 = γi + νt + αXi,t + βnLCCi,t + i,t, (2.1)

where REGentryi,t+1 is the indicator variable that identifies entry by a regional airline, γi is the route fixed effect, νt is the year fixed effect, nLCCi,t is the number of low-cost carriers operating on route i in year t, and Xi,t are the other control variables explained above. Note that the control variables are in terms of period t, whereas the dependent variable relates to period t + 1. In other words, I am looking at the effect that the control variables in a particular year will have on entry by regional airlines in the subsequent year. I am particularly interested in the sign and significance of the nLCC variable, which controls for the number of low-cost carriers operating on the route. If regional airlines are more likely to be used where low-cost carriers are present, then the estimated coefficient for nLCC should be positive and statistically significant.

Table 2.1 reports the results of the two-way fixed effects logit model. The estimated coefficient for nLCC31 (0.363) is both positive and significant at the 1% level. The estimates show that legacy carriers tend to start using regional airlines on routes where low-cost carriers are present. The results also show that the presence

31Alternatively, I used an indicator variable that is equal to 1 if any low-cost carrier operates on the route and 0 otherwise. The results using this regional presence variable are qualitatively similar.

41 of other regional airlines on the route discourages legacy carriers to start using a regional airline themselves as the estimated coefficient for nREG (-0.354) is negative and statistically significant. Finally, outsourcing would be more likely to occur when the route experiences a high frequency of delay (pdelay) and when the routes connect markets with smaller populations (pop). This is likely to be the case since regional airlines are used to integrate small cities into the legacy carrier’s route network. As a robustness check, I ran logit regressions that includes route-invariant variables, such as distance, hub airports, multi-airport markets, and slot-controlled airports, in lieu of route fixed effects. The results for these specifications, which can be found in

Appendix C, are qualitatively similar.

Table 2.1: Entry by Regional Airlines

Dependent variable REGentry Logit Standard Variable coeﬃcient error Market density (lndensity) -0.003 (0.081) Airport congestion (pdelay) 1.836** (0.395) Population (pop) -0.751* (0.299) Per capita income (income) -0.464 (0.317) Tourist market (tourism) -0.131 (0.125) Route concentration (HHIroute) -0.686 (0.788) Maximum market share (maxshare) 0.033 (0.842) Number of legacy carriers (nLEG) -0.005 (0.062) Number of low-cost carriers (nLCC) 0.363** (0.117) Number of regional airlines (nREG) -0.354** (0.030) Number of other airlines (nOT HER) 0.064 (0.045) N 418

Note: This table presents the results for the two-way fixed effects logit regression model on entry by regional airlines. Entry is defined when a legacy carrier switches operation of a route to a regional airline. Observations are at the route-year level. Route and year fixed effects suppressed. * indicates significance at 5% level. ** indicates significance at 1% level.

42 The results suggest that legacy carriers are more likely to use regional airlines where they currently compete with low-cost carriers. However, it could also be the case that legacy carriers outsource to regional airlines on routes where they intend to deter future entry by low-cost carriers. The rest of this section investigates whether regional airlines could serve as an eﬀective barrier to entry to low-cost carriers.

Previous papers have estimated barriers to entry using a logit model. Cotterill and Haller [7] find that the number of large supermarket chains in a particular market serves as an effective barrier to entry. Cetorelli and Strahan [6] conclude that banks with market power erect a significant financial barrier to entry. These papers generally use entry in the relevant market as a dependent variable and test whether particular market conditions affect entry rates. If a logit coefficient for a particular variable is negative and statistically significant, then that variable is determined to be an effective barrier to entry.

In order to test whether regional airlines serve as a barrier to entry to low-cost carriers, I utilize LCCentry, an indicator variable equal to 1 when a low-cost carrier enters the route in the following year, and 0 otherwise, as the dependent variable in a two-way fixed effects logit regression model with the following specification:

LCCentryi,t+1 = γi + νt + αXi,t + βnREGi,t + i,t, (2.2)

where LCCentryi,t+1 is the indicator variable that identiﬁes entry by a low-cost car-

32 rier, γi is the route fixed effect, νt is the year fixed effect, nREGi,t is the number of regional airlines on route i in year t, and Xi,t are the other control variables explained

32The route fixed effect captures several variables that could potentially affect entry by low-cost carriers, including distance of the route, the use of at least of the endpoints as the ticketing carrier’s hub airport, the existence of substitute airports in a given MSA, or the usage of an airport slot policy. Running logit models that explicitly control for these variables yield qualitatively similar results, which are presented in Appendix C.

43 above. Note that the control variables are in terms of period t, whereas the dependent variable relates to period t + 1. In other words, I am looking at the effect that the control variables in a particular year will have on entry by low-cost carriers in the subsequent year. I am particularly interested in the estimated sign and significance of the nREG variable, which accounts for the number of regional airlines operating on the route. If regional airlines serve as an effective barrier to entry to low-cost carriers, the estimated coefficient for nREG should be negative and statistically significant.

Table 2.2 reports the regression results. The control variable of interest is nREG,33 which is the number of regional airlines operating on the route. The logit coefficient for nREG (-0.016) is negative, yet statistically insignificant, implying that regional airlines have no effect on entry by low-cost carriers. Thus, the results suggest that legacy carriers are unable to preclude entry by a low-cost carrier by outsourcing to regional airlines.

The regression results point to three signiﬁcant factors to low-cost carrier entry.

Both the natural log of the number of passengers on the route (lndensity) and the percentage of delayed flights (pdelay) have a negative and significant effect, implying that low-cost carriers generally tend to avoid congested routes that are likely to cause disruptions to their route network. Moreover, a higher number of low-cost carriers operating on a route (nLCC) significantly inhibits rival low-cost carriers from entering a route. That is not to say that low-cost carrier presence completely blocks entry. In fact, there are some cases where low-cost carriers enter a route despite the presence of a rival low-cost carrier. Nevertheless, the results suggest that this type of competition discourages a possible low-cost carrier entrant from actually entering the route.

33Alternatively, I used an indicator variable that is equal to 1 if any regional airline operates on the route and 0 otherwise. The results using this regional presence variable are qualitatively similar.

44 Table 2.2: Entry by Low-Cost Carriers (Main Results)

Dependent variable LCCentry Logit Standard Variable coeﬃcient error Market density (lndensity) -0.622* (0.271) Airport congestion (pdelay) -3.691** (1.155) Population (pop) -0.193 (0.724) Per capita income (income) -0.802 (0.971) Tourist market (tourism) -0.060 (0.210) Route concentration (HHIroute) -0.112 (2.011) Maximum market share (maxshare) -1.200 (2.184) Number of legacy carriers (nLEG) -0.081 (0.175) Number of low-cost carriers (nLCC) -4.280** (0.298) Number of regional airlines (nREG) -0.016 (0.091) Number of other airlines (nOT HER) 0.239 (0.123) N 3,597

Note: This table presents the results for the two-way fixed effects logit regression model on entry by four low-cost carriers (AirTran Airways, JetBlue Airways, Southwest Airlines, and Spirit Airlines). Observations are at the route-year level. Route and year fixed effects suppressed. * indicates significance at 5% level. ** indicates significance at 1% level.

Industrial organization economists have been interested in Southwest Airlines as a case study on the eﬀect of low-cost carriers in the airline industry. Southwest Airlines has a reputation of not being susceptible to aggressive entry deterrence strategies by incumbents.34 Consequently, it could be the case that the results from Table

2.2 are largely inﬂuenced by Southwest Airlines. In order to check that other low- cost carriers are similarly unaﬀected by regional airlines, I run separate regressions with either entry by Southwest Airlines (W Nentry)35 or by other low-cost carriers

(otherLCCentry) as the dependent variable.

34Bamberger and Carlton [1] ﬁnd that Southwest Airlines has a high survival rate, meaning that Southwest Airlines successfully remains on an entered route for at least a year after entry. 35WN is the IATA code for Southwest Airlines.

45 Table 2.3: Entry by Southwest Airlines vs. Other Low-Cost Carriers

Dependent variable W Nentry otherLCCentry Logit Standard Logit Standard Variable coeﬃcient error coeﬃcient error Market density (lndensity) 0.430 (0.737) -0.809* (0.340) Airport congestion (pdelay) -3.559 (3.044) -3.916** (1.267) Population (pop) -4.569 (2.598) -1.489 (0.845) Per capita income (income) 1.633 (2.739) -2.285* (1.089) Tourist market (tourism) 2.072* (0.885) -0.361 (0.276) Route concentration (HHIroute) -7.875 (5.505) 3.253 (2.283) Maximum market share (maxshare) 3.150 (6.213) -2.505 (2.423) Number of legacy carriers (nLEG) 0.299 (0.459) -0.242 (0.201) Number of low-cost carriers (nLCC) -6.422** (1.144) -3.561** (0.306) Number of regional airlines (nREG) -0.246 (0.268) 0.038 (0.102) Number of other airlines (nOT HER) 0.218 (0.377) 0.266 (0.148) N 869 2,992

Note: This table presents the results for the two-way fixed effects logit regression models on entry by both Southwest Airlines and other low-cost carriers (AirTran Airways, JetBlue Airways, and Spirit Airlines). Observations are at the route-year level. Route and year fixed effects suppressed. * indicates significance at 5% level. ** indicates significance at 1% level.

The results in Table 2.3 suggest that certain factors have heterogenous effects on entry by Southwest Airlines and other low-cost carriers. In fact, some factors that affect entry by Southwest Airlines do not affect entry by other low-cost carriers, and vice-versa. Southwest Airlines is more apt to enter routes in touristy markets (tourism), while other low-cost carriers are not inclined to enter dense routes

(lndensity) that are prone to delays (pdelay). The only determinant that significantly affects both Southwest Airlines and other low-cost carriers is the number of low-cost carriers operating on the route (nLCC), which suggests that the presence of low-cost carriers significantly deters entry by low-cost carriers, in general.

46 The key insight from Table 2.3 is that regional airlines do not have an effect on either Southwest Airlines or other low-cost carriers. The logit coefficients for the number of regional airlines operating on a route (nREG) are statistically insignificant in both regressions. It is not surprising that Southwest Airlines is unaffected by the presence of regional airlines given its reputation of being undeterred by incumbent legacy carriers’ aggressive pricing strategies. It is interesting, however, to see that other low-cost carriers are unaffected as well. These coefficients provide further evidence that regional airlines serve as ineffective barriers to entry to low-cost carriers.

Regional airlines and low-cost carriers compete against each other on certain routes. For example, both United Airlines and Southwest Airlines ﬂy nonstop between Seattle-Tacoma International Airport and Portland International Airport, yet

United Airlines uses a regional airline, SkyWest Airlines, to operate that route. How- ever, these two types of airlines do not service all of the same markets. There are routes that regional airlines service that low-cost carriers would consider too thin to be proﬁtable. These routes typically link a small spoke airport with one of the legacy carrier’s hub airports, such as when Delta Air Lines uses Atlantic Southeast Airlines to operate the route between Hartsﬁeld-Jackson Atlanta International Airport and

Little Rock National Airport. No low-cost carriers service this route. On the other hand, low-cost carriers operate on some routes where the distance between the two endpoints is too far for the aircraft used by regional airlines. United Airlines uses its own ﬂeet and aircrew to operate the route between Baltimore/Washington In- ternational Airport and Denver International Airport, where it also competes with

Southwest Airlines. Thus, certain routes are more likely than others to be serviced by both regional airlines and low-cost carriers.

47 Legacy carriers could focus their use of regional airlines on routes where they anticipate entry by low-cost carriers instead of on routes where low-cost carriers are less likely to operate. If this strategy decreases the attractiveness of that route to a low-cost carrier, then both regional airlines and low-cost carriers would seemingly be attracted to some of the same routes, resulting in an upward biased coeﬃcient for nREG. In order to resolve this potential endogeneity issue, I isolate a selected sample of routes that are most likely to be served by both low-cost carriers and regional airlines. By focusing on cases where regional airlines would have its highest potential eﬀect as a barrier to entry to low-cost carriers, I am able to investigate whether regional airlines make attractive routes unattractive. In order to do this, I start by truncating the full sample to keep routes which are serviced by at least one low-cost carrier and one regional airline in every year of the sample period (1998 -

2009). I report the summary statistics (mean, 25th percentile, and 75th percentile) for the full sample and the truncated sample in Table 2.4.

The summary statistics in Table 2.4 report that low-cost carriers and regional airlines are both particularly attracted to routes that are aﬀected by three variables: distance, secondary airports, and slot-controlled airports. First, low-cost carriers and regional airlines can both be found competing against each other on short-haul routes. The mean and interquartile range for the distance of the route, distance, in the truncated sample are both much lower than that in the full sample. In fact, the 75th percentile distance in the truncated sample is nearly equivalent to the 25th percentile distance in the full sample. Some metropolitan areas are serviced by multiple airports.

Brueckner, Lee, and Singer [5] identify the primary airport and secondary airport

48 Table 2.4: Summary Statistics (Truncated Sample)

All Routes Routes with Both Low-Cost Carriers and Regional Airlines 25th 75th 25th 75th Variable Mean Percentile Percentile Mean Percentile Percentile Route distance (distance) 723 440 988 425 308 493 Hub airport (hub) 0.737 0 1 0.779 1 1 Multi-airport market (multiapt) 0.589 0 1 0.357 0 1 Primary airport (primaryapt) 0.461 0 1 0.357 0 1 Secondary airport (secondaryapt) 0.181 0 0 0 0 0 Slot-controlled airport (slots) 0.136 0 0 0 0 0 Population (pop) 3.402 1.867 4.255 2.380 1.130 3.595 Per capita income (income) 3.582 3.181 3.946 3.444 3.085 3.822 Tourist market (tourism) 1.745 0.546 1.211 2.483 0.502 1.036 Airport congestion (pdelay) 0.202 0.142 0.252 0.172 0.119 0.217

Note: This table reports the summary statistics for all routes and for routes where at least one low-cost carrier and one regional airline were present throughout the entirety of the sample period (1998-2009).

in these multi-airport markets.36 Using their airport deﬁnitions, I construct two indicator variables that identify if either endpoint on the route is a primary airport

(primaryapt) or a secondary airport (secondaryapt). If either airport is classified as either a primary airport or a secondary airport, then it is also coded to service a multi-airport city (multiapt). To be sure, it can be the case that neither endpoint is in a multi-airport market.37 According to the summary statistics, no routes in the truncated sample appear to be identified as a secondary airport. In other words, low- cost carriers and regional airlines would never be found to simultaneously operate routes to or from a secondary airport. Finally, the summary statistics for slots, which is an indicator variable that assumes the value of 1 if either endpoint is a slot-controlled airport, is different between the truncated sample and the full sample.

36For example, the city of Dallas contains two airports: Dallas/Fort Worth International Airport (primary airport) and Dallas Love Field (secondary airport). 37In this case, primaryapt, secondaryapt, and multiapt for the observation would all equal 0.

49 Slot-controlled airports regulate the number of takeoﬀs and landings that airlines are allowed each hour.38 Table 2.4 shows that no route with an endpoint that uses a slot allocation would be serviced by both a low-cost carrier and a regional airline.

I use the interquartile range for distance, secondaryapt, and slots to set the parameters for the selected sample incorporating the most attractive routes for both low-cost carriers and regional airlines. I start with the full dataset, but only keep the routes that fulfill three criteria: 1) the route must have a distance between 300 and 500 miles, 2) neither endpoint of the route can be a secondary airport, and 3) neither endpoint of the route can be a slot-controlled airport. Using this new selected sample, I then run the two-way fixed effects model outlined in Equation 2.2 using entry by low-cost carriers (LCCentry) as the dependent variable. As with before, I am interested in the estimated sign and significance of the nREG variable.

Table 2.5 presents the regression estimates for the selected sample that includes routes in which the potential effect of regional airlines as a barrier to entry would be at its highest. The results suggest that low-cost carriers seem to shy away from dense markets (lndensity) with high per capita income (income), which reflects the trend that smaller, less affluent markets tend to offer attractive operational conditions, including lower operating costs. Most importantly, the number of regional airlines operating on a route (nREG) has a negative, yet statistically insignificant effect on entry by low-cost carriers. Moreover, the number of low-cost carriers operating on the route (nLCC) is one of the variables that seems to affect low-cost carrier entry.

38The four major airports using slot allocation in the United States include Chicago O’Hare International Airport, New York LaGuardia Airport, New York John F. Kennedy International Airport, and Ronald Reagan Washington National Airport.

50 Therefore, it is the presence of rival low-cost carriers, not regional airlines, that could deter entry by low-cost carriers.

Table 2.5: Entry by Low-Cost Carriers (Selected Sample)

Dependent variable LCCentry Logit Standard Variable coeﬃcient error Market density (lndensity) -2.641* (1.307) Airport congestion (pdelay) 0.311 (6.741) Population (pop) 3.553 (5.060) Per capita income (income) -14.925* (7.518) Tourist market (tourism) 0.500 (0.782) Route concentration (HHIroute) 6.530 (9.589) Maximum market share (maxshare) -5.686 (10.301) Number of legacy carriers (nLEG) 2.019* (0.902) Number of low-cost carriers (nLCC) -5.729** (1.498) Number of regional airlines (nREG) 0.550 (0.453) Number of other airlines (nOT HER) 0.146 (0.549) N 418

Note: This table presents the results for the two-way fixed effects logit regression model on entry by four low-cost carriers (AirTran Airways, JetBlue Airways, Southwest Airlines, and Spirit Airlines) using the selected sample. The data truncates the full sample using three criteria: 1) the route distance is between 300 and 500 miles, 2) neither endpoint airport is slot-controlled, and 3) neither endpoint airport is a secondary airport. Observations are at the route-year level. Route and year fixed effects suppressed. * indicates significance at 5% level. ** indicates significance at 1% level.

This specification corrects for a possible selection bias that arises when legacy carriers focus their use of regional airlines on routes where low-cost carriers are most likely to enter. Moreover, it is in this subsample of routes where regional airlines would exhibit the most potential to serve as an effective barrier to entry. If regional airlines are found to have no effect even in these markets, then the presence of regional airlines would not seem to inhibit low-cost carriers from entering a route. The results

51 suggest that the use of regional airlines does not signiﬁcantly reduce the likelihood of entry by low-cost carriers. Thus, legacy carriers do not seem to be able to eﬀectively use regional airlines to preclude future entry by low-cost carriers.

Although it seems plausible that a low-cost carrier might ﬁnd a route with regional airline operation to be unattractive, the regression results in Tables 2.2, 2.3, and 2.5 suggest that this is not the case. Legacy carriers who outsource the operation of a route to a regional airline do not inhibit low-cost carriers from entering that particular route in the future. In fact, the regression results consistently suggest that the biggest factor that could deter low-cost carrier entry is the existence of a rival low-cost carrier.

The upshot is that legacy carriers do not seem to eﬀectively use regional airlines as a barrier to entry to low-cost carriers.

The results from Table 2.1 imply, however, that legacy carriers use regional airlines as a response to competition against existing low-cost carriers. Legacy carriers are more likely to start using regional airlines on routes where low-cost carriers are present. Thus, legacy carriers may respond to current competition with low-cost carriers by using regional airlines, which could subsequently alter their pricing strategies.

2.4 The Eﬀect of Regional Airline Entry on Pricing

This section investigates how outsourcing to regional airlines alters the pricing strategy of the outsourcing legacy carrier and competing low-cost carriers. Legacy carriers could potentially charge lower prices by exploiting the regional airlines’ more eﬃcient cost structure, thus allowing them to better compete with low-cost carriers.

Low-cost carriers might respond to this change in competition by changing their average airfares once a legacy carrier switches to a regional airline. I run separate

52 two-way ﬁxed eﬀects regressions on logged average airfares in order to study price

changes associated with regional airline entry. I attempt to rationalize the regression

results by examining the frequency that legacy carriers price match competing low-

cost carriers before and after outsourcing occurs.

Observations from the DB1B for every quarter between 1998:Q1 to 2009:Q4 is

merged with the annual population estimates of metropolitan statistical areas (MSA)

in the United States for every year between 1998 to 2009 from the Bureau of Economic

Analysis. The population for the airport’s MSA is assumed to be constant for each

quarter in a particular year. An observation in the resulting dataset is at the route-

carrier-year-quarter level. Thus, the main distinction between this dataset and the

ones used in the previous section is that observations are at the route-carrier-year-

quarter combination here, whereas the dataset used for the entry models is aggregated

to the route-year level.

I construct one dependent variable and seven control variables to be used in the

price regressions. The dependent variable (lnprice) is the logged average airfare set

by the ticketing carrier - operating carrier combination for a particular route in a

given year-quarter time period. I create two market share variables called msroute

and msapt, which calculated the market share of the carrier on a particular route and

the arithmetic mean of the carrier’s market share at the endpoint airports on a par-

ticular route in a given year-quarter, respectively. I proxy for market concentration

by constructing the route’s Herﬁndahl-Hirschman Index in a particular year-quarter

(HHIroute) and the arithmetic mean of the Herﬁndahl-Hirschman Indexes at endpoints on a route in a given time period (HHIapt). Both market share variables

and market concentration variables are based on the number of passengers ﬂown by a

53 given ticketing carrier. In order to control for market size, I use the population data to create pop, which is the geometric mean of the endpoint’s MSA population in millions.

Finally, I construct the two variables of interest: REGoperating and REGcompeting.

REGoperating assumes the value of 1 the year-quarter that a legacy carrier starts to outsource to a regional airline and every subsequent year-quarter that the regional airline operates on behalf of that legacy carrier, and 0 otherwise. REGcompeting assumes the value of 1 the year-quarter that a low-cost carrier competes against a legacy carrier that switches to a regional airline and every subsequent year-quarter that the low-cost carrier competes against the regional airline / legacy carrier partnership, and

0 otherwise.

I create two subsamples that isolate observations relevant to legacy carriers and low-cost carriers. In the legacy carrier subsample, I keep observations that pertain to a legacy carrier being the ticketing carrier and a regional airline as the operating carrier.

I then keep observations if the legacy carrier decided to start using the regional airline partner as the operating carrier on the route during the sample time period (1998:Q1 -

2009:Q4). This subsample contains 91,777 observations on 1,313 routes from 1998:Q1 to 2009:Q4. In the low-cost carrier subsample, I keep observations where a low-cost carrier serves as the ticketing carrier and was an incumbent when the legacy carrier switched to a regional airline during the sample time period. This subsample contains

25,220 observations on 590 routes from 1998:Q1 to 2009:Q4. Summary statistics and details on variable creation can be found in Appendix B.

By exploiting the panel structure of the legacy carrier subsample, I use a ﬁxed eﬀects approach to perform an event study that measures the change in the legacy carrier’s price once it outsources to a regional airline while controlling for time-invariant

54 and route-speciﬁc factors. As such, the dependent variable used is the legacy carrier’s logged average airfare (lnprice). The speciﬁcation is as follows:

lnpriceij,t = γij + νt + αXij,t + βREGoperatingij,t + ij,t, (2.3)

where lnpriceij,t is the average airfare for legacy carrier i on route j in time t, γij is the route-carrier fixed effect, νt is the year-quarter fixed effect, REGoperatingij,t is the indicator variable used to identify when a legacy carrier i switches the operation of a route to a regional airline on route j in time period t or prior to time period t, and Xij,t are the other control variables discussed above. Standard errors are clustered by route-carrier to account for correlation between a route-carrier combination over time. REGoperating, the key variable of interest, captures the change in price charged when a regional airline operates on a route relative to what the legacy carrier charged when it previously operated the route itself. Given that regional airlines have a more efficient cost structure than legacy carriers, the sign for REGoperating is hypothesized to be negative and significant, implying that legacy carriers decrease their average prices once they outsource the operation of a route to a regional airline, on average.

Table 2.6 summarizes the price regression results that estimates how legacy carriers change their average price once they switch to regional airlines. The coeﬃcient for

REGoperating (-0.111) is both negative and signiﬁcant at the 1% level, which implies that legacy carriers decrease their average price by 10.5% (exp(-0.111) - 1 = -0.105), on average, when they outsource a route to a regional airline. This is consistent with the idea that outsourcing allows the legacy carrier to charge lower prices, which may allow it to better compete with rival carriers on that route.

55 Table 2.6: Price Response to Outsourcing

Type of Carrier Legacy Carrier Low-Cost Carrier Standard Standard Variable Coeﬃcient error Coeﬃcient error Route-level market share (msroute) 0.007 (0.020) -0.091** (0.033) Route concentration (HHIroute) 0.203** (0.021) -0.145** (0.029) Airport-level market share (msapt) -0.138* (0.067) 0.406** (0.103) Airport concentration (HHIapt) 0.210** (0.075) -0.082 (0.097) Population (pop) 0.364** (0.096) -0.093 (0.103) Operates using regional airline (REGoperating) -0.111** (0.006) – – Competes with regional airline (REGcompeting) – – 0.004 (0.005) N 91,777 25,220

Note: This table reports the results of the two-way fixed effects price regressions outlined in Equations 2.3 and 2.4. Observations are at the route-carrier-year-quarter level. Route and year-quarter fixed effects suppressed. Standard errors are clustered by route-carrier to account for correlation between a route-carrier combination over time. * indicates significance at 5% level. ** indicates significance at 1% level.

Competing low-cost carriers could respond to the lower price that legacy carri-

ers charge once they switch to a regional airline by altering its pricing strategy as

well. In order to study how low-cost carriers react to outsourcing by legacy carri-

ers, I run a separate two-way ﬁxed eﬀects regression using REGcompeting in lieu of

REGoperating. This second price regression uses the low-cost carrier subsample to

estimate how a low-cost carrier responds to a legacy carrier switching from operating

a route itself to outsourcing to a regional airline while controlling for time-invariant

and route-speciﬁc variables. As such, the dependent variable used here is the low-cost

carrier’s logged average airfare, lnprice. The speciﬁcation is as follows:

lnpriceij,t = γij + νt + αXij,t + βREGcompetingij,t + ij,t, (2.4)

where lnpriceij,t is the average airfare for low-cost carrier i on route j in time t, γij is the route-carrier fixed effect, νt is the year-quarter fixed effect, REGcompetingij,t is the indicator variable used to identify when a low-cost carrier i competes with a

56 legacy carrier that switches the operation of a route to a regional airline on route j in time period t or prior to time period t, and Xij,t are the other control variables discussed above. Standard errors are clustered by route-carrier to account for correlation between a route-carrier combination over time. REGcompeting assumes the value of 1 if a low-cost carrier competes against a legacy carrier that switches to a regional airline and every subsequent year-quarter that the low-cost carrier competes against the regional airline / legacy carrier partnership, and 0 otherwise. As such, a positive (negative) and statistically signiﬁcant coeﬃcient for REGcompeting would imply that low-cost carriers increase (decrease) their average price in response to a competing legacy carrier outsourcing to a regional airline.

Table 2.6 reports the results for the price regression pertaining to low-cost carriers.

The coefficient for REGcompeting is statistically insignificant, suggesting that the pricing behavior of low-cost carriers does not systematically change when competing legacy carriers switches from operating the route itself to outsourcing to a regional airline. This result is consistent with the key findings in Tan [32], which finds that incumbent low-cost carriers do not significantly alter prices once a rival low-cost carrier enters a route. One interpretation is that low-cost carriers treat competing regional airlines as if they were a rival low-cost carrier.

In order to further examine these price responses, I look at the likelihood that legacy carriers price match competing low-cost carriers. I deﬁne price matching as a legacy carrier charging a price within 20% of a low-cost carrier’s price.39 Tables

2.7 and 2.8 report the frequency that legacy carriers match the price of a competing

39I check for the sensitivity of the results by using diﬀerent price windows. The results for these robustness checks are reported in Appendix C.

57 low-cost carrier in the quarter before they outsource to a regional airline40 and in the quarter that the switch occurs, respectively.

Table 2.7: Frequency of Price Matching Before Outsourcing

Legacy Carrier’s Price > Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 28 (52.8%) 2 (100.0%) 24 (30.0%) 2 (100.0%) Continental 12 (75.0%) – 9 (30.0%) – Delta 146 (53.5%) 13 (52.0%) 54 (29.3%) 0 (0.0%) Northwest 72 (63.2%) – 41 (50.0%) 8 (100.0%) United 33 (70.2%) 9 (90.0%) 29 (46.8%) 2 (100.0%)

Legacy Carrier US Airways 67 (59.3%) 4 (33.3%) 47 (70.1%) 14 (100.0%) Legacy Carrier’s Price = Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 20 (37.7%) 0 (0.0%) 49 (61.3%) 0 (0.0%) Continental 2 (12.5%) – 21 (70.0%) – Delta 108 (39.6%) 12 (48.0%) 123 (66.8%) 0 (0.0%) Northwest 31 (27.2%) – 37 (45.1%) 0 (0.0%) United 14 (29.8%) 1 (10.0%) 24 (38.7%) 0 (0.0%)

Legacy Carrier US Airways 44 (38.9%) 6 (50.0%) 15 (22.4%) 0 (0.0%) Legacy Carrier’s Price < Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 5 (9.4%) 0 (0.0%) 7 (8.8%) 0 (0.0%) Continental 2 (12.5%) – 0 (0.0%) – Delta 19 (7.0%) 0 (0.0%) 7 (3.8%) 1 (100.0%) Northwest 11 (9.6%) – 4 (4.9%) 0 (0.0%) United 0 (0.0%) 0 (0.0%) 9 (14.5%) 0 (0.0%)

Legacy Carrier US Airways 2 (1.8%) 2 (16.7%) 5 (7.5%) 0 (0.0%)

Note: This table reports the frequency and percentages of the occurrence that a legacy carrier sets an average price that is higher than, similar to, or lower than that of incumbent low-cost carriers in the quarter prior to when a legacy carrier switches the operation of a route to a regional airline. A dash signiﬁes that the legacy carrier never switched to a regional airline when the low-cost carrier exists on the route. A window of 20% deﬁnes price matching when the legacy carrier’s average price is within 20% of the price charged by the low-cost carrier.

40The results still hold even if looking at one year before outsourcing instead of just one quarter prior to the switch.

58 Table 2.8: Frequency of Price Matching After Outsourcing

Legacy Carrier’s Price > Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 16 (30.2%) 0 (0.0%) 17 (21.3%) 2 (100.0%) Continental 9 (56.3%) – 8 (26.7%) – Delta 145 (53.1%) 1 (4.0%) 23 (12.5%) 0 (0.0%) Northwest 68 (59.6%) – 32 (39.0%) 7 (87.5%) United 24 (51.1%) 8 (80.0%) 28 (45.2%) 2 (100.0%)

Legacy Carrier US Airways 73 (64.6%) 6 (50.0%) 37 (55.2%) 13 (92.9%) Legacy Carrier’s Price = Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 35 (66.0%) 1 (50.0%) 54 (67.5%) 0 (0.0%) Continental 4 (25.0%) – 17 (56.7%) – Delta 117 (42.9%) 21 (84.0%) 142 (77.2%) 0 (0.0%) Northwest 39 (34.2%) – 41 (50.0%) 1 (12.5%) United 21 (44.7%) 2 (20.0%) 25 (40.3%) 0 (0.0%)

Legacy Carrier US Airways 36 (31.9%) 6 (50.0%) 30 (44.8%) 1 (7.1%) Legacy Carrier’s Price < Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 2 (3.8%) 1 (50.0%) 9 (11.3%) 0 (0.0%) Continental 3 (18.8%) – 5 (16.7%) – Delta 11 (4.0%) 3 (12.0%) 19 (10.3%) 1 (100.0%) Northwest 7 (6.1%) – 9 (11.0%) 0 (0.0%) United 2 (4.3%) 0 (0.0%) 9 (14.5%) 0 (0.0%)

Legacy Carrier US Airways 4 (3.5%) 0 (0.0%) 0 (0.0%) 0 (0.0%)

Note: This table reports the frequency and percentages of the occurrence that a legacy carrier sets an average price that is higher than, similar to, or lower than that of incumbent low-cost carriers in the quarter when a legacy carrier switches the operation of a route to a regional airline. A dash signiﬁes that the legacy carrier never switched to a regional airline when the low-cost carrier exists on the route. A window of 20% deﬁnes price matching when the legacy carrier’s average price is within 20% of the price charged by the low-cost carrier.

The before and after evidence shows that legacy carriers tend to decrease their average price toward the lower price levels of competing low-cost carriers once they outsource to a regional airline. Price matching is more frequent in the quarter that the

59 legacy carrier starts outsourcing to the regional airline (Table 2.8) relative to the quarter prior to the switch (Table 2.7). Moreover, the legacy carrier is less likely to charge an average price that is more than 20% higher than that of the competing low-cost carrier once it makes the switch. For example, Table 7 shows that American Air- lines matched the price of AirTran Airways on 20 of 53 routes (37.7%) in the quarter prior to American Airlines switching to one of their three regional airline partnerships

(American Eagle, American Eagle/Executive, or Chautuaqua Airlines).41 American

Airlines set an average price higher than AirTran’s average price by at least 20% on 28 of 53 routes (52.8%), while undercutting AirTran Airways on 5 of 53 routes

(9.4%). However, Table 8 shows that American Airlines increased the frequency of price matching AirTran Airways to 35 of 53 routes (66.0%) when American Airlines outsourced to a regional airline. In fact, American Airlines decreased the amount of times it charged a higher price than AirTran to only 16 of 53 routes (30.2%), while undercutting AirTran on just 2 of 53 routes (3.8%) once they switched to a regional airline. Thus, the results suggest that not only are legacy carriers more inclined to decrease their price once they switch to a regional airline, but this price also tends to match that of a competing low-cost carrier. Legacy carriers seem to exploit the cost advantages of regional airlines in order to charge a lower price and better compete with existing low-cost carriers, who do not tend to subsequently alter their price strategy.

In order to test for the significance of the increase in the frequency of price matching after outsourcing, I run two specifications of an OLS regression using the price matching outcome as the dependent variable. The first specification uses observations

41A full list of the regional airline partnerships can be found in Appendix B.

60 that pertain to the quarter prior to outsourcing and the quarter when outsourcing occurs for each route where a legacy carrier switches the operation of the route to a regional airline. The second speciﬁcation serves as a robustness check by comparing price matching outcomes one year before outsourcing, instead of the quarter immediately preceding outsourcing, with observations in the quarter that outsourcing occurs.

The general regression speciﬁcation is as follows:

pricematchingijr,t = β1AAijr,t + β2COijr,t + β3DLijr,t + β4NWijr,t

+ β5UAijr,t + β6USijr,t + β7outsourceijr,t + ijr,t, (2.5) where pricematchingijr,t is equal to 1 when legacy carrier i price matches low-cost carrier j on route r in time period t and 0 otherwise, AAijr,t, COijr,t, DLijr,t, NWijr,t,

UAijr,t, and USijr,t are dummy variables that indicate whether the outsourcing legacy carrier is American Airlines, Continental Airlines, Delta Air Lines, Northwest Airlines,

United Airlines, or US Airways, respectively, and outsourceijr,t is equal to 1 if legacy carrier i outsources to a regional airline on route r in time period t and 0 otherwise.

Standard errors are robust and clustered by route - legacy carrier - low-cost carrier combinations in order to account for intertemporal correlation between an outsourcing legacy carrier and a competing low-cost carrier on a particular route. The coeﬃcient for outsource, the key variable of interest, can be interpreted as the change in the share of price matching after outsourcing.

Table 2.9 presents the regression results for Equation 2.5. Column 1 reports the regression results using the first specification, in which observations pertain to the quarter prior to outsourcing and the quarter when outsourcing occurs. The estimated coefficient for outsource (0.065) is positive and statistically significant, implying that price matching occurs on 6.5% more routes, on average, once a legacy

61 Table 2.9: Price Matching by Legacy Carriers

(1) (2) Standard Standard Variable Coeﬃcient error Coeﬃcient error American Airlines (AA) 0.560** (0.035) 0.503** (0.030) Continental Airlines (CO) 0.446** (0.068) 0.534** (0.065) Delta Air Lines (DL) 0.539** (0.024) 0.498** (0.024) Northwest Airlines (NW ) 0.330** (0.034) 0.279** (0.034) United Airlines (UA) 0.329** (0.041) 0.334** (0.044) US Airways (US) 0.269** (0.030) 0.295** (0.032) Occurrence of outsourcing (outsource) 0.065** (0.017) 0.086** (0.020) N 2,137 2,003

Note: This table presents the results for the OLS regression model on price matching by legacy carriers outlined in Equation 2.5. Price matching occurs when a legacy carriers sets an average price within 20% of a competing low-cost carrier on a given route. Column (1) reports the results for the specification using observations that pertain to the quarter prior to outsourcing and the quarter when outsourcing occurs, whereas column (2) reports the results for the specification using observations that pertain to the year before outsourcing and the quarter when outsourcing occurs. Observations pair an outsourcing legacy carrier with an existing low-cost carrier on a route in a particular year-quarter. Standard errors are robust and clustered by route - legacy carrier - low-cost carrier combinations in order to account for intertemporal correlation between an outsourcing legacy carrier and a competing low-cost carrier on a particular route. ** indicates significance at 1% level.

carrier outsources the operation of a route to a regional airline relative to the quarter

immediately preceding outsourcing. Thus, legacy carriers are more inclined to price

match competing low-cost carriers once they start outsourcing to a regional airline.

This result still holds qualitatively for the second specification (Column 2),42 which uses observations related to the year prior to outsourcing and the quarter that outsourcing occurs, since the estimated coefficient for outsource (0.086) is also positive and significant. Moreover, the carrier dummies report the share of routes that are price matched prior to outsourcing. In the quarter immediately preceding outsourcing (Column 1), American Airlines price match on 56% of routes where a low-cost

42The second specification includes less observations than the first specification because some routes experienced low-cost carrier entry within one year of outsourcing by legacy carriers to a regional airline.

62 carrier exists, whereas United Airlines price match on 32.9% of routes. Based on the

results in Column 2, Delta Air Lines and US Airways price match competing low-cost

carriers on 49.8% and 29.5% of routes in the year prior to outsourcing, respectively.

These results conﬁrm the statistical signiﬁcance of the increase in the frequency of

price matching after outsourcing.

Overall my ﬁndings provide complementary insights to those of previous stud-

ies that have investigated how legacy carriers respond to competition from low-cost

carriers. Goolsbee and Syverson [16] ﬁnd that airlines decrease prices when facing

potential competition from Southwest Airlines.43 In other words, airlines decrease

price even before the low-cost carrier services that route. The mere potential, and

likelihood, that Southwest Airlines will enter that route is suﬃcient for incumbent

carriers to respond by undercutting their price. Tan [32] uses similar methodology to

ﬁnd that legacy carriers decrease their price when a low-cost carrier44 actually enters the route. This paper shows that legacy carriers use regional airlines as a means to lower prices in order to better compete with low-cost carriers.

2.5 Conclusion

Some firms face a “make-or-buy” decision to either produce a product internally or outsource to an outside supplier. One reason that a firm would choose to outsource is to reduce operating costs, especially if the benefit of the lower production cost from outsourcing outweigh the transactions costs associated with search frictions and

43Potential competition with Southwest Airlines occurs on a route from airport A to airport B when Southwest Airlines ﬂies to both airports, but does not service the direct route linking airport A to airport B. 44Tan [32] examines entry not only by Southwest Airlines, but also by three other low-cost carriers (AirTran Airways, JetBlue Airways, and Spirit Airlines).

63 incomplete contracting. This issue is prevalent in the U.S. airline industry, where legacy carriers decide to operate a route using their own ﬂeet and aircrew or outsource to a more cost-eﬃcient regional airline.

This paper investigates how low-cost carriers influence where regional airlines are used by legacy carriers. The results suggest that legacy carriers are more likely to start outsourcing to regional airlines on routes where they currently compete with a low-cost carrier. I also find that legacy carriers tend to decrease average prices once they switch to a regional airline and that this lower price tends to match the average airfare charged by existing low-cost carriers. However, the presence of a regional airline does not serve as an effective barrier to entry. Therefore, I conclude that legacy carriers use regional airlines as a response to current competition with low-cost carriers and not as an attempt to preclude future entry by low-cost carriers.

Although regional airlines provide a more cost-eﬃcient alternative to operating a route themselves, legacy carriers are unable to use regional airlines on all routes. First, regional airlines use smaller aircraft that can only carry between 50-100 passengers at a time. As such, legacy carriers would not want to use regional airlines if the distance is too far or if the demand for a particular route is too high. In these cases, it would be more proﬁtable for a legacy carrier to operate the route with their own

ﬂeet and aircrew. Moreover, “scope clauses” in labor agreements with legacy carriers limits the number of routes that can be outsourced to regional airlines. Despite these limitations, regional airlines serve as a means for legacy carriers to better compete with current competitors on certain routes.

Industrial organization economists have long been interested in pricing phenomenons, particularly in the U.S. airline industry. Previous papers have found evidence that

64 airlines charge higher prices at their hub airport, that competition aﬀects the ability for airlines to price discriminate, and that some incumbent airlines decrease their price in response to entry by a low-cost carrier, particularly Southwest Airlines. This paper analyzes yet another facet of price competition between airlines by investigating how outsourcing to regional airlines enables legacy carriers to set a lower price in order to better compete with existing low-cost carriers. It would be interesting to study how the recent mergers between legacy carriers (e.g. Delta Air Lines - North- west Airlines and United Airlines - Continental Airlines) aﬀect their relationship with their respective regional airline partners and how price competition adapts to the im- minent change in the competitive environment. Alas, I leave this topic up for future research.

65 Chapter 3: The Eﬀect of De-Hubbing on Airfares

Hub-and-spoke networks have become the predominant route network structure for legacy carriers since the U.S. airline industry deregulated in 1978. Under this system, a legacy carrier moves passenger traﬃc between spoke airports through one of its hub airports in order to exploit economies of scope and economies of traﬃc density.

Each of these airlines has several hub airports strategically located in diﬀerent regions of the United States.45 However, some of the legacy carriers have recently de-hubbed an airport by ceasing hub operations at that airport. Delta Air Lines de-hubbed

Cincinnati/Northern Kentucky International Airport (CVG) in 2006 and Orlando

International Airport (MCO) in 2001, while US Airways and American Airlines de- hubbed Reagan National Airport (DCA) in 2006 and Lambert-St. Louis International

Airport (STL) in 2004, respectively. This paper analyzes the impact of de-hubbing on airfares at these four de-hubbed airports.

There is a dearth of academic research on de-hubbing, and what little has been done has studied its eﬀect on capacity. Redondi, Malighetti, and Paleari [29] formally deﬁne the criteria for identifying cases of de-hubbing and identify 37 airports

45For example, American Airlines currently utilizes Dallas/Fort Worth International Airport, John F. Kennedy International Airport, Los Angeles International Airport, Miami International Airport, and O’Hare International Airport as hub airports within the United States.

66 that have been de-hubbed between 1997 and 2009 world-wide. They find that de- hubbing, which can occur due to weak demand or a strategic decision to focus on other nearby hub airports, results in a significant and permanent decrease in the number of scheduled flights and seats offered. However, they do not take into consideration the ramifications of de-hubbing on airfares. As such, this paper is the first to study the effect of de-hubbing on airfares.

In contrast to the lack of attention spent on de-hubbing, the existing literature has been focused on the hub premium, in which prices are higher, on average, when at least one of the route’s endpoints is a hub airport for the servicing airline. Legacy carriers experience more market power at their hub airports because passengers are attracted to the higher frequency of flights and the increased variety of destinations that they offer from the hub airport. Moreover, Lederman [25] finds that certain passengers are willing to pay higher prices in order to receive future awards from the airline’s frequent-flyer program. Early works empirically estimated the value of the hub premium by regressing logged airfares on airport market shares, while controlling for other factors. Borenstein [2] and Evans and Kessides [9] both find that airport market shares has a positive and statistically significant effect on airfares. More recently, Lee and Luengo-Prado [26] use hub dummy variables as a more explicit proxy for the hub premium and find that prices are between 12.2% and 13.0% higher, on average, when the flight travels to or from an airline’s hub airport. However, in contrast to these studies which rely mainly on cross-sectional variation to identify a hub premium, this paper utilizes variation over time in the hub status of particular airports.46

46I focus on de-hubbing because hub creation is usually too gradual.

67 Since de-hubbing leads to a reduction in the frequency of flights and number of destinations serviced to and from the de-hubbed airport, two possible price effects could occur. On the one hand, passengers could become less brand loyal to that airline since they no longer dominate the flights to and from that airport. As such, the airline’s local market power decreases, which could eliminate the hub premium.

Thus, de-hubbing could potentially lead to lower airfares as demand becomes more elastic due to the unraveling of the airline’s market power at the de-hubbed airport.

On the other hand, the reduction in capacity by the de-hubbing airline could soften price competition, assuming that rival airlines do not respond by altering their flight schedules. If the de-hubbed airline’s competitors maintain a steady number of flights and seats offered, then airfares could potentially increase even though the market is less concentrated. This paper empirically tests these two stories using data from the

U.S. airline industry.

In order to analyze this pricing puzzle, I study four instances of de-hubbing at domestic airports between 2001 and 2006. Using a difference-in-differences estimation approach, I find a positive and statistically significant price increase after de-hubbing.

Moreover, I allow for the possibility that the de-hubbing airline might alter its pricing strategy differently than its competitors at the de-hubbed airport. The results of this difference-in-difference-in-differences estimation strategy implies that the de-hubbing airline increases its average prices by a similar percentage as its competitors. Thus, the key result of this paper is that de-hubbing softens price competition, which leads to an increase in average airfares.

The results of this paper are somewhat surprising given the tremendous amount of research on the rationale behind the existence of the hub premium. Despite the

68 removal of upward pressure on prices that are found at hub airports, prices further elevate once that airport has been de-hubbed as the decrease in capacity leads to a weaker competitive environment. The results shed light on the ramiﬁcations of an interesting, yet relatively unexplored strategic decision by legacy carriers. Namely, the pricing phenomenon that follows once an airline decides to stop using a particular airport as a hub airport.

3.1 Data

I construct a dataset from three sources in order to test the eﬀect of de-hubbing on airfares. The main dataset used in this paper is the Airline Origin and Destination

Survey (DB1B), which is a 10% survey of domestic travel that is published quarterly by the Bureau of Transportation Statistics. An observation in the raw data provides information on the number of passengers who paid a certain price to ﬂy with a particular airline on a given route, as well as the distance between the two endpoints of the route. Using this data, I am able to calculate both the mean average airfare and the number of passengers for a particular airline on a given route in each year-quarter.

I also obtain data on the number of scheduled ﬂights and available seats from the T-

100 database, which is also published by the Bureau of Transportation Statistics. The

T-100 dataset, which provides monthly traﬃc and capacity data of travel within the

United States by domestic airlines, is aggregated for a year-quarter combination to the route-carrier level so that it can be merged with the DB1B. Finally, I augment this dataset with yearly estimates on population and per capita income for metropolitan statistical areas (MSA), which can be found in the Local Area Personal Income tables

69 that are created and distributed by the the Bureau of Economic Analysis. Data from

1993 to 2009 are collected from each of the three data sources.

I focus on coach class fares on nonstop or one-stop ﬂights between the top 100 airports within the continental United States.47 Following Ito and Lee [21], I eliminate any observations with a reported price that is less than $25 or more than $1500 as these observations are believed to be either frequent-ﬂyers or incorrectly coded. I also drop any observations where the airline has a market share of less than 1% as is done in Borenstein [2]. The resulting dataset consists of 1,517,163 observations on 8,158 routes. Summary statistics are summarized in Table 3.1.

Table 3.1: Summary Statistics

Variable Deﬁnition Mean (Std. Dev.) priceijt Average one-way market fare for carrier i on route j in time period t 200.36 (64.52) distancej One-way distance (in miles) between the endpoints of route j 1,262.85 (657.70) passengersijt Number of passengers for carrier i on route j in time period t 2,059.75 (5,922.27) nLEGit Number of legacy carriers operating route j in time period t 3.89 (1.55) nLCCit Number of low-cost carriers carriers operating route j in time period t 0.17 (0.41) popjt Geometric mean of population (in hundreds of thousands) of origin 25.66 and destination airports’ MSA on route j in time period t (20.76) incomeit Geometric mean of per capita income (in tens of thousands) of origin 3.33 and destination airports’ MSA on route j in time period t (0.72) Routes Number of routes in the sample 8,158 N Number of observations 1,517,163

47Rankings are based on the number of boardings in 2009 and are obtained from the Federal Aviation Administration. The airport with the most amount of boardings is Hartsﬁeld-Jackson Atlanta International Airport, while the 100th-ranked airport is Myrtle Beach International Airport.

70 (a) CVG (Delta Air Lines) (b) DCA (US Airways)

Figure 3.1: Number of Flights by De-Hubbing Airline

The number of flights is the sum of the scheduled departure and arrival flights at the particular airport. The gray dashed line indicates the year-quarter that the airport became de-hubbed by the airline identified in parentheses.

Redondi, Malighetti, and Paleari [29] identiﬁed 37 airports world-wide that were de-hubbed between 1997 and 2009. Four of these airports are in the United States, including Cincinnati/Northern Kentucky International Airport (CVG), Reagan Na- tional Airport (DCA), Orlando International Airport (MCO), and Lambert-St. Louis

International Airport (STL). I define the de-hubbing time period at these airports as the year-quarter after the largest drop in the number of scheduled flights, which is then verified using public sources. Figure 3.1 graphs the total number of flights scheduled to and from each of these four airports for the de-hubbing airline between

71 1993 and 2009. The gray dashed line indicates the time period when the airport is considered to become de-hubbed.

Figure 3.1 shows a sharp decline in the number of ﬂights following de-hubbing at each of the four airports. Delta Air Lines de-hubbed CVG in 2006:Q1 prior to their merger with Northwest Airlines, which was announced in April 2008. In 2001:Q4,

Delta de-hubbed MCO, which served as the hub airport for Song, its currently defunct subsidiary. Moreover, other legacy carriers also de-hubbed during this time period.

Upon acquiring Trans World Airlines (TWA) in 2001, American Airlines set up hub operations at STL, where TWA had been headquartered. STL was supposed to alle- viate the traﬃc congestion at Chicago O’Hare International Airport and Dallas/Fort

Worth International Airport, two of American’s other hub airports. However, the merger resulted in a ﬁnancial drain and American Airlines subsequently de-hubbed

STL in 2004:Q1. Finally, US Airways de-hubbed DCA in 2006:Q4 following their merger with America West in 2005.48 All four examples of de-hubbing experienced a stark decrease in capacity, which is reported in Table 3.2.

Table 3.2 presents two ways to study the effect of de-hubbing on quantity: the total number of flights, which is the sum of the scheduled departure and arrival flights at the airport, and the total number of seats, which is the sum of the seats offered on all of those flights. In order to analyze the de-hubbing effect, I compare the time period before de-hubbing, defined to be the three to six quarters prior to the airport becoming de-hubbed, with the time period after de-hubbing, defined to be the three to six quarters following de-hubbing. The percent change from these two time periods

48There is a sharp decrease in the number of ﬂights by US Airways to and from DCA in 2001:Q4. It is believed that this is a response to a negative shock to demand following the September 11 attacks and not a concerted de-hubbing eﬀort as evidenced by a spike in capacity beginning in 2004:Q4.

72 Table 3.2: Capacity Before and After De-Hubbing

Total Number of Flights Total Number of Seats Before After Percent Before After Percent De-Hubbing De-Hubbing Change De-Hubbing De-Hubbing Change All Airlines 108,695 44,296 -59.2% 15,398,463 6,802,906 -55.8% Delta 106,544 42,803 -59.8% 15,168,678 6,633,355 -56.3% CVG Competitors 2,151 1,493 -30.6% 229,785 169,551 -26.2% All Airlines 159,000 135,120 -15.0% 19,339,702 17,539,441 -9.3% US Airways 76,344 53,627 -29.8% 8,584,479 6,829,585 -20.4% DCA Competitors 82,656 81,493 -1.4% 10,755,223 10,709,856 -0.4% All Airlines 209,031 182,093 -12.9% 30,571,678 28,154,311 -7.9% Delta 64,787 37,699 -41.8% 9,807,581 6,562,253 -33.1% MCO Competitors 144,244 144,394 0.1% 20,764,097 21,592,058 4.0% All Airlines 231,089 102,125 -55.8% 31,145,731 13,449,236 -56.8% American 162,454 38,915 -76.0% 22,196,619 5,369,376 -75.8% STL Competitors 68,635 63,210 -7.9% 8,949,112 8,079,860 -9.7%

Note: This table reports changes in capacity by all airlines, the de-hubbing airline, and other airlines present at the de-hubbed airport. The total number of flights is the sum of the scheduled departure and arrival flights at the particular airport, whereas the total number of seats is the sum of the offered seats on all of those flights. The before and after periods includes the the three to six quarters prior to and following de-hubbing, respectively. This allows for a one-year excluded period around the de-hub date that accounts for the transitional time period in which de-hubbing takes effect.

is also reported in the table. These two measures of quantity are summarized for all airlines servicing the airport, which is composed of the de-hubbing airline and its competitors. The results show that de-hubbing leads to an overall decrease in the total number of flights and the total number of seats at each of these four airports; yet, this is largely due to the fact that the de-hubbing airline substantially decreases its operation at the airport, by definition. For example, the total number of flights and seats offered by all airlines servicing MCO decreased by 12.9% and 7.9%, respectively, while Delta Air Lines, the airline that de-hubbed MCO, reduced the number of flights by 41.8% and the total number of seats offered by 33.1%. With the exception of CVG, there is a relatively small change in capacity by competitors of the de-hubbing airline after the airport has become de-hubbed. In fact, competitors at DCA decreased the

73 total number of scheduled ﬂights and seats oﬀered by 1.4% and 0.4% following de- hubbing by US Airways. MCO is the only airport in which competitors increased their capacity in response to de-hubbing, albeit by a relatively small amount. Regardless, both proxies for quantity show that legacy carriers reduce capacity once they de-hub an airport, while competing airlines generally maintain a steady level of capacity.

These results are not surprising in light of Redondi, Malighetti, and Paleari [29], who

ﬁnd that operations are permanently scaled-back at de-hubbed airports. With this in mind, I turn to analyzing the price eﬀect of de-hubbing.

3.2 Empirical Analysis

In order to test the effect of de-hubbing on airfares, I conduct an event study on four airports that were de-hubbed between 2001 and 2006. By using a difference-in- differences approach, I am able to infer an overall before and after effect of de-hubbing on airfares. In order to determine whether the price response of the de-hubbing airline differs from its competitors, I use a difference-in-difference-in-differences estimation technique. The results from both estimation strategies suggest that average airfares increase following de-hubbing. The following section formalizes the estimation strategy and discusses the regression results.

I use a two-way fixed effects model in order to yield a differences-in-differences

(DID) estimate on the eﬀect of de-hubbing on airfares. The dependent variable is logged average airfares (lnprice). I control for the geometric mean of the population

(pop) and per capita income (income) of the two endpoint airports’ MSAs, as well the number of legacy carriers (nLEG) and low-cost carriers (nLCC) that service the route. I also include a dummy variable (airport) that indicates whether the de-hubbed

74 airport is one of the route’s endpoint airports, another dummy variable (dehub) that indicates whether the time period is pre- or post-de-hubbing, and the interaction term of the two dummy variables (airport × dehub). The airport variable is constructed to limit the data sample to a before and after period, where the before period starts three quarters prior to the identified de-hubbing date and ends six quarters before de- hubbing whereas the after period starts three quarters following the de-hubbing time period and ends six quarters after de-hubbing. This allows for a one-year excluded period around the de-hub date that accounts for the transitional time period in which de-hubbing takes effect. The specification for the difference-in-differences regression model is as follows:

lnpriceij,t = γij + νt + αXij,t + β1airportij + β2dehubij,t

+ β3(airportij × dehubij,t) + ij,t, (3.1)

where lnpriceij,t is the average one-way airfare for airline i on route j in time t, γij is the carrier-route fixed effects, νt is the year-quarter fixed effects, airportij is the

49 airport dummy variable, dehubij,t is the de-hub time dummy variable, and Xij,t are the other control variables explained above. I cluster the standard errors by route-carrier in order to account for intragroup correlation over time. The variable of interest is the interaction term (airport × dehub), which signiﬁes the DID estimate.

A positive and statistically signiﬁcant coeﬃcient for the interaction term implies that airfares increase, on average, after the airport has been de-hubbed.

Table 3.3 reports the results of the DID regression for each of the four de-hubbed airports. The coeﬃcient for airport×dehub, the control variable of interest, is positive and statistically signiﬁcant for all four de-hubbed airports, meaning that the DID

49By construction, the airport dummy variable becomes absorbed by the route-carrier ﬁxed eﬀects.

75 Table 3.3: Diﬀerence-in-Diﬀerences Results

Variable CVG DCA MCO STL De-hub time dummy (dehub) 0.163** 0.232** -0.155** -0.071** (0.004) (0.007) (0.004) (0.003) Airport/de-hub interaction (airport × dehub) 0.289** 0.039** 0.047** 0.069** (0.008) (0.009) (0.006) (0.007) Population (pop) -0.260** -0.586** 0.661** 0.727** (0.044) (0.050) (0.086) (0.066) Per capita income (income) -0.351** -0.718** 0.960** 0.083 (0.053) (0.058) (0.051) (0.060) Number of legacy carriers (nLEG) -0.006** -0.008** -0.009** -0.011** (0.001) (0.001) (0.001) (0.001) Number of low-cost carriers (nLCC) -0.023** -0.025** -0.110** -0.062** (0.003) (0.003) (0.005) (0.003) N 194,570 190,970 183,915 190,799

Note: This table reports the results of the two-way fixed effects price regressions outlined in Equation 3.1. Observations are at the route-carrier-year-quarter level. Route and year-quarter fixed effects suppressed. Standard errors, which are presented in parentheses, are clustered by route-carrier to account for correlation between a route-carrier combination over time. * indicates significance at 5% level. ** indicates significance at 1% level.

method implies that average airfares at de-hubbed airports increases after it has been de-hubbed by a legacy carrier. The results suggest that de-hubbing contributed to a 3.9%, 4.7%, and 6.9% increase in average airfares at DCA, MCO, and STL, respectively, yet the eﬀect is more pronounced at CVG. Recall that Table 3.2 showed that DCA, MCO, and STL experienced a substantial reduction in capacity by the de-hubbing airline with little response by its competitors, whereas Delta Air Lines and its competitors all substantially reduced their capacity once Delta de-hubbed

CVG. Thus, the strong eﬀect at CVG can be attributed to the substantial capacity changes by the de-hubbing airlines and its competitors. Nonetheless, the reduction

76 in capacity by the de-hubbing airline seems to soften competition in every case of

de-hubbing.

I specify a separate two-way ﬁxed eﬀects regression model in order to ascertain

whether the de-hubbing airline changed its prices at a diﬀerent rate than its com-

petitors once it de-hubs an airport. In order to yield a diﬀerence-in-diﬀerence-in-

diﬀerences (DDD) estimate, I include three additional variables to Equation 3.1: a

dummy variable (carrier) that indicates the de-hubbing airline, the interaction term

(dehub × carrier) of the time dummy variable and the de-hubbing airline dummy variable, as well as the triple interaction term (airport × dehub × carrier). The specification for the differences-in-differences-in-differences regression model is as follows:

lnpriceij,t = γij + νt + αXij,t + β1airportij + β2dehubij,t + β3carrierij

+ β3(airportij × dehubij,t) + β4(dehubij,t × carrierij)

+ β5(airportij × dehubij,t × carrierij) + ij,t, (3.2)

where lnpriceij,t is the average one-way airfare for airline i on route j in time t

and carrierij is equal to 1 if airline i de-hubs one of the endpoint airports on route

j.50 The dehub × carrier interaction term controls for any nationwide change in

price for airline i that occurs following de-hubbing. The interpretation of the other

variables follows directly from Equation 3.1. Given this speciﬁcation, the estimate

for the airport × dehub interaction term represents the percent increase in average

airfares by competitors of the de-hubbing airline after the airport has been de-hubbed.

However, I am particularly interested in the airport×dehub×carrier interaction term,

which signiﬁes the DDD estimate. A positive and statistically signiﬁcant estimate for

50By construction, carrier becomes absorbed by the route-carrier ﬁxed eﬀects.

77 airport × dehub × carrier implies that the de-hubbing airline increases its airfares at a higher percentage than its competitors once it has de-hubbed an airport.

Table 3.4: Difference-in-Difference-in-Differences Results

Variable CVG DCA MCO STL De-hub time dummy (dehub) 0.165** 0.213** -0.164** -0.069** (0.004) (0.007) (0.004) (0.003) Airport/de-hub interaction (airport × dehub) 0.282** 0.026** 0.046** 0.051** (0.012) (0.009) (0.007) (0.008) De-hub/carrier interaction (dehub × carrier) -0.009** 0.116** 0.052** -0.015** (0.003) (0.004) (0.003) (0.003) Airport/de-hub/carrier interaction (airport × dehub × carrier) 0.020 0.021 0.010 0.069** (0.016) (0.027) (0.012) (0.017) Population (pop) -0.252** -0.542** 0.594** 0.727** (0.044) (0.050) (0.086) (0.066) Per capita income (income) -0.344** -0.669** 0.934** 0.098 (0.053) (0.057) (0.050) (0.059) Number of legacy carriers (nLEG) -0.006** -0.006** -0.009** -0.011** (0.001) (0.001) (0.001) (0.001) Number of low-cost carriers (nLCC) -0.023** -0.028** -0.110** -0.062** (0.003) (0.003) (0.005) (0.003) N 194,570 190,970 183,915 190,799

Note: This table reports the results of the two-way fixed effects price regressions outlined in Equation 3.2. Observations are at the route-carrier-year-quarter level. Route and year-quarter fixed effects suppressed. Standard errors, which are presented in parentheses, are clustered by route-carrier to account for correlation between a route-carrier combination over time. * indicates significance at 5% level. ** indicates significance at 1% level.

Table 3.4 reports the results for the DDD estimation strategy for the four de- hubbed airports. The airport × dehub interaction term is positive and significant for all four de-hubbed airports, which implies that the de-hubbing airline’s competitors increase average airfares after de-hubbing occurs. Moreover, the airport × dehub × carrier term accounts for the difference in airfares between the de-hubbing airline and its competitors once of of the route’s endpoint airports becomes de-hubbed. The estimate for airport × dehub × carrier is positive, yet statistically insignificant for

78 CVG, DCA and MCO, implying that the de-hubbing airline increases its airfares at the same rate as its competitors after it has de-hubbed an airport. However, the coeﬃcient for airport × dehub × carrier for STL (0.069) is positive and statistically signiﬁcant, meaning that American Airlines further increased its average airfares by

6.9% relative to its competitors after it de-hubbed STL. The upshot of these regression results is that the price increase is attributed not only to the de-hubbing airline but also the other airlines servicing the de-hubbed airport.

The regression results discussed in this section shed light on the price puzzle regarding de-hubbing on airfares. Although prices are found to increase after an airline ceases hub operations at an airport, it would have been theoretically plausible if prices decreased since de-hubbing results in a less concentrated market. Thus, the results overwhelmingly support the story that the decrease in capacity by the de-hubbing airline reduces the availability of substitutes at the de-hubbed airport, which weakens competition and leads to higher average airfares.

3.3 Conclusion

Legacy carriers have de-hubbed certain airports due to changes in demand and competitive environment. The potential effect of de-hubbing on prices is non- trivial. Airfares could decrease due to the market becoming less concentrated once the dominant airline scales back its operation at the de-hubbed airport. However, the reduction in the frequency and variety of flights diminishes the availability of substitutes, which leads less competition and subsequent increase in price. By using an event study estimation approach, this paper serves as the first attempt to test

79 which of these two stories characterizes the eﬀect of de-hubbing on airfares in the

U.S. airline industry.

The results suggest that de-hubbing puts upward pressure on prices at de-hubbed airports. Not only do average airfares increase after the airport becomes de-hubbed, but the percent change in airfares is also shown to rise at a similar rate for both the de-hubbing airline and its competitors. A possible explanation for this pricing phenomenon is that competition is softened once an airport is de-hubbed as the de- hubbing airline drastically reduces the number of scheduled ﬂights and seats oﬀered whereas its competitors maintain a steady level of capacity.

Several extensions can be explored in future research. First, it is possible that legacy carriers have de-hubbed other domestic airports. If this is true, then it would be helpful to add these other examples to the empirical analysis. Second, studying the effects of de-hubbing on the 10th and 90th percentile airfares could provide more insight on how de-hubbing affects other sections of the price distribution. Finally, it would be interesting to investigate whether low-cost carriers respond to de-hubbing differently than legacy carriers. Nonetheless, the results in this paper serve as the foundation for future empirical work on de-hubbing in the U.S. airline industry.

80 Appendix A: Tables for “Incumbent Response to Entry by Low-Cost Carriers in the U.S. Airline Industry”

Table A.1: Summary Statistics

Variable Definition Mean (Std. Dev.) priceijt Average one-way market fare for carrier i 170.35 on route j in time period t (72.68) giniijt Gini coefficient of carrier i’s prices 0.249 on route j in time period t (0.089) distancej One-way distance (in miles) between the endpoints of route j 1196.28 (663.10) passengersijt Number of passengers for carrier i 1014.65 on route j in time period t (1641.07) mktshrrouteijt Market share for carrier i on route j in time period t 0.223 (0.275) HERF routejt Herfindahl Index for route j in time period t 0.484 (0.189) mktshraptijt Arithmetic mean of carrier i’s market share 0.157 at endpoints on route j in time period t (0.128) HERF aptjt Arithmetic mean of Herfindahl Indexes 0.247 at endpoints on route j in time period t (0.079) popjt Geometric mean of population (in millions) of origin and 4.04 destination airports’ MSA on route j in time period t (2.43) Routes Number of routes in the sample 1000 N Number of observations 263,272

81 Table A.2: Incumbent Price Response to Actual Entry

(1) (2) (3) (4) Entrant: AirTran Airways JetBlue Airways Southwest Airlines Spirit Airlines Incumbent: All Carriers All Carriers All Carriers All Carriers t0 − 12 0.0139 0.0040 0.0236* 0.0159 (0.0072) (0.0137) (0.0100) (0.0148) t0 − 11 0.0194* 0.0080 0.0183 0.0097 (0.0076) (0.0135) (0.0107) (0.0162) t0 − 10 0.0105 -0.0242 0.0367* 0.0191 (0.0073) (0.0132) (0.0103) (0.0186) t0 − 9 0.0107 -0.0408* 0.0044 -0.0059 (0.0080) (0.0181) (0.0108) (0.0167) t0 − 8 -0.0155* -0.0209 -0.0019 -0.0246 (0.0076) (0.0152) (0.0110) (0.0191) t0 − 7 -0.0132 -0.0278* 0.0059 -0.0395* (0.0083) (0.0134) (0.0100) (0.0167) t0 − 6 -0.0302* -0.0481* 0.0326* -0.0275 (0.0089) (0.0171) (0.0103) (0.0186) t0 − 5 -0.0320* -0.0492* 0.0274* -0.0021 (0.0089) (0.0149) (0.0104) (0.0168) t0 − 4 -0.0067 -0.0603* 0.0337* -0.0056 (0.0088) (0.0181) (0.0105) (0.0156) t0 − 3 -0.0237* -0.0126 0.0246* 0.0101 (0.0090) (0.0152) (0.0110) (0.0156) t0 − 2 -0.0224* -0.0048 0.0262* -0.0210 (0.0087) (0.0157) (0.0122) (0.0157) t0 − 1 -0.0362* -0.0406* -0.0017 -0.0433* (0.0092) (0.0200) (0.0109) (0.0161) t0 -0.0691* -0.0447* -0.0820* -0.0365* (0.0088) (0.0153) (0.0112) (0.0200) t0 + 1 -0.1143* -0.0573* -0.1306* -0.0551* (0.0102) (0.0155) (0.0120) (0.0224) t0 + 2 -0.1274* -0.1044* -0.1137* -0.0344 (0.0103) (0.0168) (0.0119) (0.0224) t0 + 3 -0.1330* -0.0980* -0.1198* -0.0408* (0.0095) (0.0157) (0.0130) (0.0204) t0 + 4 -0.1131* -0.0676* -0.1230* -0.0466* (0.0095) (0.0165) (0.0121) (0.0182) t0 + 5 -0.1093* -0.0619* -0.1369* -0.0600* (0.0095) (0.0177) (0.0120) (0.0203) t0 + 6 -0.1184* -0.0871* -0.1057* -0.0591* (0.0094) (0.0170) (0.0115) (0.0191) t0 + 7 -0.1145* -0.0914* -0.1131* -0.0364 (0.0094) (0.0152) (0.0118) (0.0247) t0 + 8 -0.0872* -0.0858* -0.1184* -0.0577* (0.0098) (0.0146) (0.0122) (0.0196) t0 + 9 -0.0897* -0.0136 -0.1086* -0.0653* (0.0093) (0.0205) (0.0125) (0.0212) t0 + 10 -0.0887* -0.0121 -0.0883* -0.0782* (0.0101) (0.0205) (0.0118) (0.0200) t0 + 11 -0.0776* -0.0564* -0.1026* -0.0717* (0.0098) (0.0223) (0.0123) (0.0223) t0 + 12 -0.0663* -0.0570* -0.0800* -0.0467* (0.0105) (0.0239) (0.0120) (0.0214)

Note: The dependent variable is the logged average airfare (lnprice). The variables shown are the lagged/forward time dummies, where t0 is quarter of entry. Fixed eﬀects, competition variables, and market variables suppressed. Standard errors are in parentheses. Number of observations: 263,270. * denotes signiﬁcance at 5% level. 82 Table A.3: Legacy Carrier Incumbent Price Response to Actual Entry

(1) (2) (3) (4) Entrant: AirTran Airways JetBlue Airways Southwest Airlines Spirit Airlines Incumbent: Legacy Carriers Legacy Carriers Legacy Carriers Legacy Carriers t0 − 12 0.0069 0.0003 0.0225 0.0123 (0.0085) (0.0152) (0.0123) (0.0173) t0 − 11 0.0080 -0.0013 0.0154 0.0117 (0.0088) (0.0149) (0.0130) (0.0195) t0 − 10 -0.0021 -0.0290 0.0346* 0.0183 (0.0086) (0.0160) (0.0123) (0.0196) t0 − 9 0.0006 -0.0281 -0.0027 -0.0025 (0.0096) (0.0166) (0.0129) (0.0192) t0 − 8 -0.0310* -0.0144 -0.0062 -0.0449* (0.0087) (0.0147) (0.0130) (0.0154) t0 − 7 -0.0360* -0.0276 0.0004 -0.0556* (0.0095) (0.0155) (0.0117) (0.0165) t0 − 6 -0.0479* -0.0523* 0.0294* -0.0417* (0.0103) (0.0161) (0.0122) (0.0189) t0 − 5 -0.0478* -0.0620* 0.0250* -0.0032 (0.0103) (0.0175) (0.0123) (0.0169) t0 − 4 -0.0276* -0.0552* 0.0396* -0.0100 (0.0095) (0.0166) (0.0123) (0.0170) t0 − 3 -0.0440* -0.0138 0.0241 0.0017 (0.0104) (0.0174) (0.0131) (0.0173) t0 − 2 -0.0489* -0.0065 0.0324* -0.0247 (0.0093) (0.0184) (0.0147) (0.0181) t0 − 1 -0.0644* -0.0295 -0.0033 -0.0504* (0.0102) (0.0209) (0.0125) (0.0179) t0 -0.0916* -0.0511* -0.0925* -0.0652* (0.0096) (0.0173) (0.0130) (0.0202) t0 + 1 -0.1428* -0.0734* -0.1403* -0.0831* (0.0103) (0.0166) (0.0139) (0.0248) t0 + 2 -0.1474* -0.1183* -0.1188* -0.0674* (0.0109) (0.0188) (0.0139) (0.0226) t0 + 3 -0.1564* -0.1163* -0.1227* -0.0612* (0.0103) (0.0171) (0.0151) (0.0229) t0 + 4 -0.1368* -0.0789* -0.1185* -0.0660* (0.0102) (0.0181) (0.0142) (0.0201) t0 + 5 -0.1309* -0.0674* -0.1253* -0.0856* (0.0103) (0.0193) (0.0136) (0.0226) t0 + 6 -0.1401* -0.0848* -0.0974* -0.0862* (0.0101) (0.0182) (0.0132) (0.0206) t0 + 7 -0.1405* -0.0852* -0.1086* -0.0718* (0.0099) (0.0170) (0.0139) (0.0218) t0 + 8 -0.1170* -0.0859* -0.1170* -0.0865* (0.0102) (0.0167) (0.0138) (0.0209) t0 + 9 -0.1124* -0.0236 -0.0999* -0.0859* (0.0100) (0.0240) (0.0146) (0.0231) t0 + 10 -0.1090* -0.0149 -0.0928* -0.0963* (0.0110) (0.0242) (0.0136) (0.0219) t0 + 11 -0.1014* -0.0613* -0.1104* -0.0942* (0.0104) (0.0259) (0.0144) (0.0247) t0 + 12 -0.0913* -0.0571* -0.0823* -0.0694* (0.0106) (0.0283) (0.0140) (0.0238)

Note: The dependent variable is the logged average airfare (lnprice). The variables shown are the lagged/forward time dummies, where t0 is quarter of entry. Fixed eﬀects, competition variables, and market variables suppressed. Standard errors are in parentheses. Number of observations: 263,270. * denotes signiﬁcance at 5% level. 83 Table A.4: Low-Cost Carrier Incumbent Price Response to Actual Entry

(1) (2) (3) (4) Entrant: AirTran Airways JetBlue Airways Southwest Airlines Spirit Airlines Incumbent: Low-Cost Carriers Low-Cost Carriers Low-Cost Carriers Low-Cost Carriers t0 − 12 0.1307* -0.0110 -0.0148 0.0511* (0.0132) (0.0400) (0.0429) (0.0203) t0 − 11 0.1263* 0.0457 -0.0950 0.0236* (0.0167) (0.0452) (0.0572) (0.0116) t0 − 10 0.1370* -0.0067 -0.0648 -0.0058 (0.0219) (0.0395) (0.0566) (0.0267) t0 − 9 0.0895* -0.0422 -0.1020* 0.0047 (0.0249) (0.0449) (0.0410) (0.0324) t0 − 8 0.0659* -0.0296 -0.1083* 0.0631* (0.0305) (0.0438) (0.0427) (0.0306) t0 − 7 0.0704* -0.0592 -0.1203* 0.0434 (0.0252) (0.0405) (0.0436) (0.0398) t0 − 6 0.0543* -0.1188* -0.0890 0.0341 (0.0215) (0.0475) (0.0456) (0.0456) t0 − 5 0.0686* -0.0922* -0.1023* 0.0691 (0.0286) (0.0395) (0.0467) (0.0454) t0 − 4 0.0777* -0.0827 -0.1242* 0.0767 (0.0292) (0.0516) (0.0490) (0.0419) t0 − 3 0.1177* -0.0981 -0.0968 0.1004* (0.0257) (0.0573) (0.0499) (0.0376) t0 − 2 0.0943* -0.1449* -0.1387* 0.0684 (0.0234) (0.0460) (0.0490) (0.0359) t0 − 1 0.0609* -0.1445* -0.0776 0.0441 (0.0233) (0.0471) (0.0631) (0.0438) t0 0.1069* -0.0516 -0.0571 0.0580 (0.0233) (0.0425) (0.0599) (0.0455) t0 + 1 0.1162* -0.1139* -0.1040 0.0763* (0.0232) (0.0333) (0.0545) (0.0297) t0 + 2 0.1134* -0.1070* -0.0982* 0.0365 (0.0254) (0.0374) (0.0497) (0.0311) t0 + 3 0.0832* -0.1225* -0.1270* 0.0293 (0.0360) (0.0485) (0.0551) (0.0431) t0 + 4 0.1153* -0.0741* -0.2183* 0.0452 (0.0310) (0.0373) (0.0446) (0.0306) t0 + 5 0.0825* -0.0709* -0.1957* 0.0829* (0.0309) (0.0352) (0.0454) (0.0319) t0 + 6 0.0654* -0.0870* -0.2146* 0.0640 (0.0307) (0.0348) (0.0575) (0.0369) t0 + 7 0.0564 -0.1541* -0.1840* 0.0505 (0.0351) (0.0385) (0.0500) (0.0433) t0 + 8 0.0881* -0.1190* -0.1881* 0.0857 (0.0359) (0.0342) (0.0464) (0.0453) t0 + 9 0.1224* 0.0010 -0.2821* 0.0058 (0.0343) (0.0436) (0.0467) (0.0340) t0 + 10 0.0847* 0.0073 -0.1791* 0.0056 (0.0322) (0.0307) (0.0474) (0.0418) t0 + 11 0.0912* -0.0221 -0.1488* 0.0382 (0.0304) (0.0350) (0.0468) (0.0416) t0 + 12 0.1506* -0.1140* -0.1580* 0.0129 (0.0370) (0.0440) (0.0469) (0.0452)

Note: The dependent variable is the logged average airfare (lnprice). The variables shown are the lagged/forward time dummies, where t0 is quarter of entry. Fixed effects, competition variables, and market variables suppressed. Standard errors are in parentheses. Number of observations: 263,270. * denotes significance at 5% level. 84 Appendix B: Data Construction for “The Influence of Low-Cost Carriers on the Use of Regional Airlines”

I collect data from three sources. The ﬁrst is the Airline Origin and Destination

Survey (DB1B), which is published quarterly by the Bureau of Transportation Statis- tics. Each observation in this dataset, which is a ten percent survey of domestic air travel, provides information on the number of passengers who paid a particular price to an identified ticketing carrier in order to fly on a certain route that is operated by an identified operating carrier. The data also provides the distance of the route and the number of coupons associated with the itinerary. I obtain data on the number of delayed flights at the route level from the Airline On-Time Performance Data, which is published monthly by the Bureau of Transportation Statistics. Finally, the Local

Area Personal Income tables, which is published annually by the Bureau of Economic

Analysis, provides annual estimates on population, per capita personal income, and personal income by major source and earnings at the metropolitan statistical area

(MSA)-level. This section details the step-by-step process used in order to create the main dataset used in this paper.

The sample time period for this paper is 1998 - 2009. I begin by downloading all the DB1B ﬁles for 1998:Q1 to 2009:Q4. I start with 1998 because the Bureau of

Transportation Statistics did not ask carriers to report whether the operating carrier

85 diﬀered from the ticketing carrier until that year, while 2009 is chosen because the

Delta Air Lines / Northwest Airlines merger was oﬃcially completed in 2010. By using 2009 as a cutoﬀ year, I am able to isolate a pristine time period before the competitive nature of the airline industry was drastically altered. I then download and merge the Airline On-Time Performance Data from January 1998 to December

2009, as well as the annual estimates in the Local Area Personal Income tables from

1998 to 2009.

The following steps enumerate the process taken of cleaning the data and constructing the variables (in italics) used in this paper:

1. I merge all of the DB1B ﬁles from 1998:Q1 to 2009:Q4.

2. I drop observations where either the ticketing carrier or operating carrier is

unidentiﬁed.

3. Following Gerardi and Shapiro [13], I drop observations where the market fare

was less than $10 in order to eliminate frequent-ﬂyer tickets.

4. I drop observations if either endpoint is an airport in Alaska, Hawaii, or Puerto

Rico. Forbes and Lederman [12] justify this criteria since the nature of the

routes is diﬀerent between ﬂights within the continental United States and these

states and territories. Karlsson, Odoni, and Yamanaka [22] comment that routes

involving Alaska and Hawaii are subject to special, complex tax rules, which

alters the route structure to and from this area.

5. I keep observations pertaining to nonstop ﬂights.

86 6. I generate msroute to be the ticketing carrier’s market share at the route-level

based on the number of passengers ﬂown in a particular year-quarter.

• I generate maxshare to be the maximum market share of all ticket carriers

servicing a given route in a particular year-quarter.

7. I generate HHIroute to be the sum of squared market shares at the route-level

in a particular year-quarter.

8. Following Evans and Kessides [9], I drop observations where the ticketing carrier

had less than a 1% market share on the route.

9. Following Forbes and Lederman [11], I limit the data to routes with a maximum

distance of 1,500 miles since none of the planes typically used by the regional

airlines can travel farther than this distance.

10. I generate lndensity to be the logged total number of passengers ﬂying on a

route in a particular year-quarter.

11. Following Reiss and Spiller [30], I limit my data to the top 2,500 routes in order

to account for small airline markets.

12. I generate nLCC, nLEG, nREG, and nOT HER to be the number of operating

carriers on a route in a particular year-quarter that is identiﬁed as either a low-

cost carriers, legacy carrier, regional airline, or other airline, respectively.

• Low-cost carriers: AirTran Airways, JetBlue Airways, Southwest Airlines,

Spirit Airlines

87 • Legacy carriers: American Airlines, Continental Airlines, Delta Air Lines,

Northwest Airlines, United Airlines, US Airways

• Legacy carrier - regional airline partnerships are outlined in Table B.1.

13. REGentry and LCCentry are generated to equal 1 in the year-quarter that a

regional airline operating carrier - legacy carrier ticketing carrier combination

or low-cost carrier ticketing carrier, respectively, starts servicing a route that

it was not on for the previous four quarters and stays on the route for at least

two consecutive quarters afterwards. The entrant must also ﬂy at least 100

passengers in the year-quarter of entry.

• I create W Nentry and otherLCCentry to equal LCCentry when the tick-

eting carrier is Southwest Airlines or one of the other low-cost carriers

(AirTran Airways, JetBlue Airways, and Spirit Airlines), respectively.

14. Using the Airline On-Time Performance Data aggregated to the quarterly level,

I construct pdelay, which is constructed to be the percentage of ﬂights that

are delayed over 15 minutes on a given route in a particular year-quarter, and

merge that into the dataset.

15. Using the Local Area Personal Income tables, I construct pop, income, and

tourism and merge them into the dataset.

• pop: geometric mean of population (in millions) of endpoint airports’ MSA

in a particular year

• income: geometric mean of per capita income (in tens of thousands) of

endpoint airports’ MSA in a particular year

88 • tourism: the maximum of endpoint airports’ percentage of accommodation

income in nonfarm income in a particular year

16. I drop any routes with missing tourism or income data.

This is where the process diverts between the datasets used for the entry regressions analyzed in Section 2.3 of Chapter 2 and the price regressions discussed in Section 2.4.

In order to get the data used for the entry regressions I continue with the following steps:

1. I collapse the variables at the route-year-quarter level.

2. I keep observations pertaining only to quarter 4 of each year in order to aggre-

gate the data from a quarterly basis to a yearly basis.

3. I only keep routes if a legacy carrier was a ticketing carrier at any point in the

sample time period.

Summary statistics for the dataset used for entry regressions are reported in Table B.2.

In order to get the data used for the price regressions I commence from step 16 with the following steps:

1. I generate price to be the passenger-weighted average price charged by the

ticketing carrier.

• I create lnprice to be the natural log of price.

89 2. I create msapt to be simple average of the ticketing carrier’s market share at

the airport-level based on the number of passengers ﬂown in a particular year-

quarter.

3. I generate HHIapt to be the sum of squared market shares at the airport-level

in a particular year-quarter.

4. I create REGcompeting to equal 1 if a low-cost carrier competes against a

legacy carrier that switches to a regional airline and every subsequent year-

quarter that the low-cost carrier competes against the regional airline / legacy

carrier partnership, and 0 otherwise.

5. I generate REGoperating to equal 1 the year-quarter that a legacy carrier starts

to outsource to a regional airline and every subsequent year-quarter that the

regional airline operates on behalf of that legacy carrier, and 0 otherwise.

I create two subsamples that isolate observations relevant to legacy carriers and low- cost carriers. In the legacy carrier subsample, I keep observations that pertain to a legacy carrier being the ticketing carrier and a regional airline as the operating carrier.

I then keep observations if the legacy carrier decided to start using the regional airline partner as the operating carrier on the route during the sample time period (1998:Q1

- 2009:Q4). Summary statistics for the legacy carrier sample set are reported in Table

B.3. In the low-cost carrier subsample, I keep observations where a low-cost carrier serves as the ticketing carrier and was an incumbent when the legacy carrier switched to a regional airline during the sample time period. Summary statistics for the legacy carrier sample set are reported in Table B.4.

90 Table B.1: Regional Airline Partnerships

Legacy Carrier Regional Airline Partners American Eagle American Airlines American Eagle/Executive Chautauqua Airlines Chautauqua Airlines Cape Air Continental Airlines Colgan Air CommutAir Express Jet Gulfstream International Airlines Atlantic Southeast Airlines Chautauqua Airlines Comair Delta Air Lines Freedom Airlines Mesaba Airlines Pinnacle Airlines Shuttle America SkyWest Airlines Comair Northwest Airlines Compass Airlines Mesaba Airlines Pinnacle Airlines Chautauqua Airlines Colgan Air ExpressJet Airlines GoJet Airlines United Airlines Great Lakes Mesa Airlines Shuttle America SkyWest Airlines Trans States Airlines Air Wisconsin Chautauqua Airlines Colgan Air US Airways Mesa Airlines Piedmont PSA Republic Airlines Trans States Airlines

Source: Regional Airline Association 2009 Annual Report (www.raa.org)

91 Table B.2: Summary Statistics (Entry Sample)

Variable Definition Mean (Std. Dev.) REGentryit Indicator equal to 1 if a regional airline enters on route i 0.122 in time period t, and 0 otherwise (0.345) LCCentryit Indicator equal to 1 if a low-cost carrier enters on route i 0.028 in time period t, and 0 otherwise (0.166) W Nentryit Indicator equal to 1 if Southwest Airlines enters on route i 0.006 in time period t, and 0 otherwise (0.078) otherLCCentryit Indicator equal to 1 if a low-cost carrier other than 0.023 Southwest Airlines enters on route i in time period t, and 0 otherwise (0.149) distancei Distance (in miles) between the endpoints of route i 723.65 (339.84) passengersit Number of passengers on route i in time period t 29859.29 Note: lndensityit = ln(passengersit) (29346.83) pdelayit Percentage of flights delayed over 15 minutes on route i in time period t 0.202 (0.084) popit Geometric mean of population (in millions) of origin and destination 3.402 airports’ MSA on route i in time period t (2.204) incomeit Geometric mean of per capita income (in tens of thousands) of origin 3.582 and destination airports’ MSA on route i in time period t (0.547) tourismit Maximum of the percentage of accommodation income in nonfarm income 1.745 of origin and destination airports’ MSA on route i in time period t (20.415) HHIrouteit Herfindahl-Hirschman Index for route i in time period t 0.578 (0.206) maxshareit Maximum of the market share of carriers on route i in time period t 0.691 (0.184) nLEGit Number of legacy carriers operating route i in time period t 2.647 (1.456) nLCCit Number of low-cost carriers carriers operating route i in time period t 0.572 (0.593) nREGit Number of regional airlines servicing route i in time period t 1.057 (1.269) nOT HERit Number of other carriers (not legacy carrier, low-cost carrier, or regional 0.596 airline) servicing route i in time period t (0.890) Routes Number of routes in the sample 1,161 N Number of observations 12,790

92 Table B.3: Summary Statistics (Legacy Carrier Subsample)

Variable Definition Mean (Std. Dev.) priceijt Average one-way market fare for carrier i 200.50 on route j in time period t (89.40) msrouteijt Market share for carrier i on route j in time period t 0.400 (0.340) HHIroutejt Herfindahl Index for route j in time period t 0.598 (0.209) msaptijt Arithmetic mean of carrier i’s market share 0.235 at endpoints on route j in time period t (0.137) HHIaptjt Arithmetic mean of Herfindahl Indexes 0.333 at endpoints on route j in time period t (0.099) popjt Geometric mean of population (in millions) of origin and 3.24 destination airports’ MSA on route j in time period t (2.02) REGoperatingijt Indicator variable equal to 1 if carrier i is a legacy carrier 0.313 and switches to a regional airline on route j in time period t (0.464) or prior to time period t, and 0 otherwise Routes Number of routes in the sample 1,313 N Number of observations 91,777

Table B.4: Summary Statistics (Low-Cost Carrier Subsample)

Variable Definition Mean (Std. Dev.) priceijt Average one-way market fare for carrier i 134.10 on route j in time period t (36.87) msrouteijt Market share for carrier i on route j in time period t 0.344 (0.300) HHIroutejt Herfindahl Index for route j in time period t 0.539 (0.192) msaptijt Arithmetic mean of carrier i’s market share 0.267 at endpoints on route j in time period t (0.179) HHIaptjt Arithmetic mean of Herfindahl Indexes 0.308 at endpoints on route j in time period t (0.106) popjt Geometric mean of population (in millions) of origin and 2.82 destination airports’ MSA on route j in time period t (1.89) REGcompetingijt Indicator variable equal to 1 if carrier i is a low-cost carrier 0.361 and a legacy carrier switches to a regional airline on route j (0.480) in time period t or prior to time period t, and 0 otherwise Routes Number of routes in the sample 590 N Number of observations 25,220

93 Appendix C: Robustness Checks for “The Inﬂuence of Low-Cost Carriers on the Use of Regional Airlines”

This section includes the description and results of the robustness checks discussed in the paper.

C.1 Pooled Logit Regression Results

The regressions reported in the paper are based on a two-way fixed effects logit model (Equations 2.1 and 2.2 in Chapter 2). I run several specifications without the fixed effects as a robustness check. The first set of specifications provides a robustness check for the results on the use of regional airlines as a response to current competition, whereas the second set of specifications provides a robustness check for the results on the use of regional airlines as a barrier to entry.

Tables C.1 and C.2 report the robustness checks for the results in Table 2.1 in

Chapter 2. Both specifications in the appendix use a pooled logit model with the key difference being that the regression reported in Table C.2 includes time dummies to account for year effects, whereas the regression reported in Table C.1 does not. In both cases, the estimated coefficient for nLCC, the variable of interest, is positive and significant as is the case in Table 2.1 in Chapter 2.

94 Tables C.3 and C.4 report the robustness checks for the results in Table 2.2 in

Chapter 2. Both specifications in the appendix use a pooled logit model with the key difference being that the regression reported in Table C.4 includes time dummies to account for year effects, whereas the regression reported in Table C.3 does not. In both cases, the estimated coefficient for nREG, the variable of interest, is statistically insignificant as is the case in Table 2.2 in Chapter 2.

C.2 Price Matching Windows

Tables 2.7 and 2.8 in Chapter 2 of the paper discuss how legacy carriers are more likely to price match competing low-cost carriers, if they exist, once they outsource the operation of a route to a regional airline. Price matching in that table was deﬁned as the legacy carrier’s price being within 20% of an existing low-cost carrier after the switch to a regional airline. Tables C.5 and C.6 report the frequency and percentage of legacy carrier’s setting an average price that is higher, the same, or lower than that of low-cost carriers using a 15% window in the quarter prior to the legacy carrier outsourcing to a regional airline and in the quarter of the switch, respectively, while Tables C.7 and C.8 similarly report the respective before and after results utilizing a 25% window. The results suggest that legacy carriers are more inclined to price match existing low-cost carriers once they outsource to regional airlines regardless of the window size. The caveat is that a larger price window tends to deﬁne more instances of price matching.

95 Table C.1: Entry by Regional Airlines (Pooled Logit Model)

Dependent variable REGentry Logit Standard Variable coeﬃcient error Route distance (lndistance) -0.688** (0.062) Hub airport (hub) 0.313** (0.064) Multi-airport market (multiapt) -0.404** (0.063) Slot-controlled airport (slots) 0.551** (0.075) Market density (lndensity) -0.077* (0.031) Airport congestion (pdelay) 1.116** (0.253) Population (pop) -0.013 (0.015) Per capita income (income) 0.056 (0.048) Tourist market (tourism) -0.045** (0.009) Route concentration (HHIroute) -1.190* (0.547) Maximum market share (maxshare) -0.388 (0.581) Number of legacy carriers (nLEG) 0.026 (0.024) Number of low-cost carriers (nLCC) 0.113** (0.019) Number of regional airlines (nREG) -0.367** (0.057) Number of other airlines (nOT HER) -0.068* (0.029) N 18,889

Note: This table presents the results for the logit cross-section regression model on entry by regional airlines. Entry is defined when a legacy carrier switches operation of a route to a regional airline. Standard errors are clustered by route to account for correlation within a route over time. Observations are at the route-year level. * indicates significance at 5% level. ** indicates significance at 1% level.

96 Table C.2: Entry by Regional Airlines (Pooled Logit Model with Time Dummies)

Dependent variable REGentry Logit Standard Variable coeﬃcient error Route distance (lndistance) -0.773** (0.065) Hub airport (hub) 0.315** (0.068) Multi-airport market (multiapt) -0.415** (0.066) Slot-controlled airport (slots) 0.599** (0.080) Market density (lndensity) lndensity -0.021 (0.033) Airport congestion (pdelay) 1.637** (0.284) Population (pop) -0.021 (0.016) Per capita income (income) -0.121 (0.089) Tourist market (tourism) -0.048** (0.009) Route concentration (HHIroute) -1.273* (0.572) Maximum market share (maxshare) -0.319 (0.611) Number of legacy carriers (nLEG) 0.047 (0.025) Number of low-cost carriers (nLCC) 0.050* (0.022) Number of regional airlines (nREG) -0.449** (0.062) Number of other airlines (nOT HER) -0.058 (0.030) N 18,889

Note: This table presents the results for the logit cross-section regression model on entry by regional airlines. Entry is defined when a legacy carrier switches operation of a route to a regional airline. Observations are at the route-year level. Time dummies suppressed. Standard errors are clustered by route to account for correlation within a route over time. * indicates significance at 5% level. ** indicates significance at 1% level.

97 Table C.3: Entry by Low-Cost Carriers Airlines (Pooled Logit Model)

Dependent variable LCCentry Logit Standard Variable coeﬃcient error Route distance (lndistance) 0.367* (0.147) Hub airport (hub) -0.774** (0.168) Multi-airport market (multiapt) -0.912** (0.140) Slot-controlled airport (slots) -0.756** (0.222) Market density (lndensity) 0.292** (0.095) Airport congestion (pdelay) -0.310 (0.661) Population (pop) 0.113** (0.033) Per capita income (income) -0.272* (0.112) Tourist market (tourism) -0.024 (0.015) Route concentration (HHIroute) -0.789 (1.291) Maximum market share (maxshare) 0.587 (1.405) Number of legacy carriers (nLEG) 0.111* (0.048) Number of low-cost carriers (nLCC) -1.172** (0.177) Number of regional airlines (nREG) -0.025 (0.059) Number of other airlines (nOT HER) -0.339 (0.079) N 12,789

Note: This table presents the results for the logit cross-section regression model on entry by four low-cost carriers (AirTran Airways, JetBlue Airways, Southwest Airlines, and Spirit Airlines). Standard errors are clustered by route to account for correlation within a route over time. Observations are at the route-year level. * indicates signiﬁcance at 5% level. ** indicates signiﬁcance at 1% level.

98 Table C.4: Entry by Low-Cost Carriers (Pooled Logit Model with Time Dummies)

Dependent variable LCCentry Logit Standard Variable coeﬃcient error Route distance (lndistance) 0.398** (0.149) Hub airport (hub) -0.704** (0.172) Multi-airport market (multiapt) -0.978** (0.143) Slot-controlled airport (slots) -0.657** (0.227) Market density (lndensity) 0.213* (0.103) Airport congestion (pdelay) -1.679* (0.796) Population (pop) 0.112** (0.032) Per capita income (income) 0.167 (0.203) Tourist market (tourism) -0.017 (0.015) Route concentration (HHIroute) -0.127 (1.309) Maximum market share (maxshare) -0.145 (1.411) Number of legacy carriers (nLEG) 0.100* (0.049) Number of low-cost carriers (nLCC) -1.082** (0.187) Number of regional airlines (nREG) -0.038 (0.067) Number of other airlines (nOT HER) -0.322** (0.082) N 12,789

Note: This table presents the results for the logit cross-section regression model on entry by four low-cost carriers (AirTran Airways, JetBlue Airways, Southwest Airlines, and Spirit Airlines). Observations are at the route-year level. Time dummies suppressed. Standard errors are clustered by route to account for correlation within a route over time. * indicates signiﬁcance at 5% level. ** indicates signiﬁcance at 1% level.

99 Table C.5: Frequency of Price Matching Before Outsourcing: 15% Window

Legacy Carrier’s Price > Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 29 (54.7%) 2 (100.0%) 30 (37.5%) 2 (100.0%) Continental 13 (81.3%) – 9 (30.0%) – Delta 171 (62.6%) 13 (52.0%) 62 (33.7%) 0 (0.0%) Northwest 76 (66.7%) – 46 (56.1%) 8 (100.0%) United 35 (74.5%) 10 (100.0%) 32 (51.6%) 2 (100.0%)

Legacy Carrier US Airways 74 (65.5%) 6 (50.0%) 49 (73.1%) 14 (100.0%) Legacy Carrier’s Price = Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 16 (30.2%) 0 (0.0%) 41 (51.3%) 0 (0.0%) Continental 0 (0.0%) – 21 (70.0%) – Delta 82 (30.0%) 12 (48.0%) 107 (58.2%) 0 (0.0%) Northwest 25 (21.9%) – 28 (34.1%) 0 (0.0%) United 11 (23.4%) 0 (0.0%) 21 (33.9%) 0 (0.0%)

Legacy Carrier US Airways 37 (32.7%) 4 (33.3%) 13 (19.4%) 0 (0.0%) Legacy Carrier’s Price < Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 8 (15.1%) 0 (0.0%) 9 (11.3%) 0 (0.0%) Continental 3 (18.8%) – 0 (0.0%) – Delta 20 (7.3%) 0 (0.0%) 15 (8.2%) 1 (100.0%) Northwest 13 (11.4%) – 8 (9.8%) 0 (0.0%) United 1 (2.1%) 0 (0.0%) 9 (14.5%) 0 (0.0%)

Legacy Carrier US Airways 2 (1.8%) 2 (16.7%) 5 (7.5%) 0 (0.0%)

Note: This table reports the frequency and percentages of the occurrence that a legacy carrier sets an average price that is higher than, similar to, or lower than that of incumbent low-cost carriers in the quarter prior to when a legacy carrier switches the operation of a route to a regional airline. A dash signiﬁes that the legacy carrier never switched to a regional airline when the low-cost carrier exists on the route. A window of 15% deﬁnes price matching when the legacy carrier’s average price is within 15% of the price charged by the low-cost carrier.

100 Table C.6: Frequency of Price Matching After Outsourcing: 15% Window

Legacy Carrier’s Price > Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 16 (30.2%) 0 (0.0%) 26 (32.5%) 2 (100.0%) Continental 11 (68.8%) – 9 (30.0%) – Delta 160 (58.6%) 2 (8.0%) 29 (15.8%) 0 (0.0%) Northwest 72 (63.2%) – 38 (46.3%) 8 (100.0%) United 28 (59.6%) 9 (90.0%) 31 (50.0%) 2 (100.0%)

Legacy Carrier US Airways 76 (67.3%) 9 (75.0%) 43 (64.2%) 13 (92.9%) Legacy Carrier’s Price = Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 31 (58.5%) 1 (50.0%) 44 (55.0%) 0 (0.0%) Continental 1 (6.3%) – 15 (50.0%) – Delta 91 (33.3%) 18 (72.0%) 125 (67.9%) 0 (0.0%) Northwest 32 (28.1%) – 33 (40.2%) 0 (0.0%) United 15 (31.9%) 0 (0.0%) 21 (33.9%) 0 (0.0%)

Legacy Carrier US Airways 30 (26.5%) 3 (25.0%) 24 (35.8%) 1 (7.1%) Legacy Carrier’s Price < Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 6 (11.3%) 1 (50.0%) 10 (12.5%) 0 (0.0%) Continental 4 (25.0%) – 6 (20.0%) – Delta 22 (8.1%) 5 (20.0%) 30 (16.3%) 1 (100.0%) Northwest 10 (8.8%) – 11 (13.4%) 0 (0.0%) United 4 (8.5%) 1 (10.0%) 10 (16.1%) 0 (0.0%)

Legacy Carrier US Airways 7 (6.2%) 0 (0.0%) 0 (0.0%) 0 (0.0%)

Note: This table reports the frequency and percentages of the occurrence that a legacy carrier sets an average price that is higher than, similar to, or lower than that of incumbent low-cost carriers in the quarter when a legacy carrier switches the operation of a route to a regional airline. A dash signiﬁes that the legacy carrier never switched to a regional airline when the low-cost carrier exists on the route. A window of 15% deﬁnes price matching when the legacy carrier’s average price is within 15% of the price charged by the low-cost carrier.

101 Table C.7: Frequency of Price Matching Before Outsourcing: 25% Window

Legacy Carrier’s Price > Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 25 (47.2%) 2 (100.0%) 22 (27.5%) 2 (100.0%) Continental 11 (68.8%) – 7 (23.3%) – Delta 129 (47.3%) 13 (52.0%) 48 (26.1%) 0 (0.0%) Northwest 65 (57.0%) – 34 (41.5%) 7 (87.5%) United 32 (68.1%) 6 (60.0%) 25 (40.3%) 2 (100.0%)

Legacy Carrier US Airways 62 (54.9%) 4 (33.3%) 44 (65.7%) 13 (92.9%) Legacy Carrier’s Price = Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 24 (45.3%) 0 (0.0%) 51 (63.8%) 0 (0.0%) Continental 4 (25.0%) – 23 (76.7%) – Delta 128 (46.9%) 12 (48.0%) 133 (72.3%) 0 (0.0%) Northwest 39 (34.2%) – 46 (56.1%) 1 (12.5%) United 15 (31.9%) 4 (40.0%) 28 (45.2%) 0 (0.0%)

Legacy Carrier US Airways 49 (43.4%) 6 (50.0%) 19 (28.4%) 1 (7.1%) Legacy Carrier’s Price < Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 4 (7.5%) 0 (0.0%) 7 (8.8%) 0 (0.0%) Continental 1 (6.3%) – 0 (0.0%) – Delta 16 (5.9%) 0 (0.0%) 3 (1.6%) 1 (100.0%) Northwest 10 (8.8%) – 2 (2.4%) 0 (0.0%) United 0 (0.0%) 0 (0.0%) 9 (14.5%) 0 (0.0%)

Legacy Carrier US Airways 2 (1.8%) 2 (16.7%) 4 (6.0%) 0 (0.0%)

Note: This table reports the frequency and percentages of the occurrence that a legacy carrier sets an average price that is higher than, similar to, or lower than that of incumbent low-cost carriers in the quarter prior to when a legacy carrier switches the operation of a route to a regional airline. A dash signiﬁes that the legacy carrier never switched to a regional airline when the low-cost carrier exists on the route. A window of 25% deﬁnes price matching when the legacy carrier’s average price is within 25% of the price charged by the low-cost carrier.

102 Table C.8: Frequency of Price Matching After Outsourcing: 25% Window

Legacy Carrier’s Price > Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 11 (20.8%) 0 (0.0%) 13 (16.3%) 2 (100.0%) Continental 7 (43.8%) – 7 (23.3%) – Delta 120 (44.0%) 1 (4.0%) 15 (8.2%) 0 (0.0%) Northwest 60 (52.6%) – 26 (31.7%) 5 (62.5%) United 23 (48.9%) 6 (60.0%) 22 (35.5%) 2 (100.0%)

Legacy Carrier US Airways 65 (57.5%) 5 (41.7%) 31 (46.3%) 13 (92.9%) Legacy Carrier’s Price = Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 41 (77.4%) 2 (0.0%) 60 (75.0%) 0 (0.0%) Continental 6 (37.5%) – 22 (73.3%) – Delta 149 (54.6%) 23 (92.0%) 158 (85.9%) 0 (0.0%) Northwest 50 (43.9%) – 50 (61.0%) 3 (37.5%) United 22 (26.8%) 4 (40.0%) 35 (56.5%) 0 (0.0%)

Legacy Carrier US Airways 47 (41.6%) 7 (58.3%) 36 (53.7%) 1 (7.1%) Legacy Carrier’s Price < Low-Cost Carrier’s Price Low-Cost Carrier AirTran JetBlue Southwest Spirit American 1 (1.9%) 0 (0.0%) 7 (8.8%) 0 (0.0%) Continental 3 (18.8%) – 1 (3.3%) – Delta 4 (1.5%) 1 (4.0%) 11 (6.0%) 1 (100.0%) Northwest 4 (3.5%) – 6 (7.3%) 0 (0.0%) United 2 (4.3%) 0 (0.0%) 5 (8.1%) 0 (0.0%)

Legacy Carrier US Airways 1 (0.9%) 0 (0.0%) 0 (0.0%) 0 (0.0%)

Note: This table reports the frequency and percentages of the occurrence that a legacy carrier sets an average price that is higher than, similar to, or lower than that of incumbent low-cost carriers in the quarter when a legacy carrier switches the operation of a route to a regional airline. A dash signiﬁes that the legacy carrier never switched to a regional airline when the low-cost carrier exists on the route. A window of 25% deﬁnes price matching when the legacy carrier’s average price is within 25% of the price charged by the low-cost carrier.

103 Bibliography

[1] Gustavo Bamberger and Dennis Carlton. Predation and the Entry and Exit of Low-Fare Carriers. Elsevier, 2006.

[2] Severin Borenstein. “Hubs and High Fares: Dominance and Market Power in the U.S. Airline Industry”. The RAND Journal of Economics, 20:344–365, 1989.

[3] Severin Borenstein. “The Evolution of U.S. Airline Competition”. The Journal of Economic Perspectives, 6(2):45–73, 1992.

[4] Severin Borenstein and Nancy Rose. “Competititon and Price Dispersion in the U.S. Airline Industry”. Journal of Political Economy, 102(4):653–683, 1994.

[5] Jan Brueckner, Darin Lee, and Ethan Singer. “City-Pairs vs. Airport Pairs: A Market-Deﬁnition Methodology for the Airline Industry”. Working paper, 2011.

[6] Nicola Cetorelli and Philip Strahan. “Finance as a Barrier to Entry: Bank Competition and Industry Structure in the Local U.S. Markets”. The Journal of Finance, 61(1):437–461, 2006.

[7] Ronald Cotterill and Lawrence Haller. “Barrier and Queue Eﬀects: A Study of Leading US Supermarket Chain Entry Patterns”. The Journal of Industrial Economics, 40(4):427–440, 1992.

[8] Kenneth Elzinga and David Mills. Predatory Pricing in the Airline Industry: Spirit Airlines v. Northwest Airlines. 5 edition.

[9] William Evans and Ioannis Kessides. “Localized Market Power in the U.S. Airline Industry”. The Review of Economics and Statistics, 75(1):66–75, 1993.

[10] Silke Forbes and Mara Lederman. The Role of Regional Airlines in the U.S. Airline Industry. Elsevier, 2007.

[11] Silke Forbes and Mara Lederman. “Adaptation and Vertical Integration in the Airline Industry”. American Economic Review, 99(5):1831–1849, 2009.

104 [12] Silke Forbes and Mara Lederman. “Does Vertical Integration Aﬀect Firm Perfor- mance? Evidence from the Airline Industry”. The RAND Journal of Economics, 41(4):765–790, 2010.

[13] Kristopher Gerardi and Adam Shapiro. “Does Competition Reduce Price Disper- sion? New Evidence from the Airline Industry”. Journal of Political Economy, 117(1):1–37, 2009.

[14] David Gilbert. “Relationship Marketing and Airline Loyalty Schemes”. Tourism Management, 17(8):575–582, 1996.

[15] Christopher Goetz and Adam Shapiro. “Strategic Alliances and Entry Deter- rence: Evidence from Airline Codesharing”. Working paper, 2011.

[16] Austan Goolsbee and Chad Syverson. “How Do Incumbents Respond to the Threat of Entry? Evidence from the Major Airlines”. The Quarterly Journal of Economics, 123(4):1611–1633, 2008.

[17] Henry Grabowski and John Vernon. “Brand Loyalty, Entry, and Price Com- petition in Pharmaceuticals after the 1984 Drug Act”. Journal of Law and Economics, 35(2):331–350, 1992.

[18] Kathy Hayes and Leola Ross. “Is Airline Price Dispersion the Result of Careful Planning or Competitive Forces?”. Review of Industrial Organization, 13:523– 541, 1998.

[19] Barry Hirsch. Wage Determination in the U.S. Airline Industry: The Role of Unions and Product Market Competition. Elsevier, 2007.

[20] Abraham Hollander. “On Price-Increasing Entry”. Economica, 54(215):317–324, 1987.

[21] Harumi Ito and Darin Lee. “Domestic Codesharing Practices in the U.S. Airline Industry”. Journal of Air Transport Management, 11:89–97, 2005.

[22] Joakim Karlsson, Amedeo Odoni, and Shiro Yamanaka. “The Impact of Infrastructure-Related Taxes and Fees on Domestic Airline Fares in the U.S.”. Journal of Air Transport Management, 10:285–293, 2004.

[23] Byung-Do Kim, Mengze Shi, and Kannan Srinivasan. “Reward Programs and Tacit Collusion”. Marketing Science, 20(2):99–120, 2001.

[24] Paul Klemperer. “Markets with Consumer Switching Costs”. The Quarterly Journal of Economics, 102(2):375–394, 1987.

105 [25] Mara Lederman. “Are Frequent-Flyer Programs a Cause of the Hub Premium?”. Journal of Economics & Management Strategy, 17(1):35–66, 2008.

[26] Darin Lee and Mar´ıaJos´eLuengo-Prado. “The Impact of Passenger Mix on Reported Hub Premiums in the U.S. Airline Industry”. Southern Economic Journal, 72(2):372–394, 2005.

[27] Steven Morrison. “Actual, Adjacent, and Potential Competition: Estimating the Full Eﬀect of Southwest Airlines”. Journal of Transport Economics and Policy, 35(2):239–256, 2001.

[28] Jeffrey Perloff and Steven Salop. “Equilibrium with Product Differentiation”. The Review of Economic Studies, 52(1):107–120, 1985.

[29] Renato Redondi, Paolo Malighetti, and Stefano Paleari. “De-Hubbing of Airports and Their Recovery Patterns”. Journal of Air Transport Management, 18:1–4, 2012.

[30] Peter Reiss and Pablo Spiller. “Competition and Entry in Small Airline Mar- kets”. Journal of Law and Economics, 32(2):179–202, 1989.

[31] Robert Rosenthal. “A Model in which an Increase in the Number of Sellers Leads to a Higher Price”. Econometrica, 48(6):1575–1579, 1980.

[32] Kerry Tan. “Incumbent Response to Entry by Low-Cost Carriers in the U.S. Airline Industry”. Working paper, 2011.

[33] Timothy Vowles. “The Southwest Eﬀect in Multi-Airport Regions”. Journal of Air Transport Management, 7:251–258, 2001.

106