does bankruptcy protection harm the airline...

Does Bankruptcy Protection Harm the Airline Industry?Empirical Study on Bankruptcy and Low Cost Carrier Growth in the Airline Industry

Hwa Ryung Lee

Department of Economics, UC Berkeley

April 2009

Preliminary and Incomplete

ABSTRACT

Previous discussions on bankruptcy e¤ect in the airline industry have focuses on whether bankrupt airlines

harm non-bankrupt rivals�pro�tability. However, a more relevant question would be whether bankrupt airlines

harm e¢ cient non-bankrupt rivals�pro�tability and the industry e¢ ciency deteriorates as a result. This paper

attempts to answer this question by examing fare and capacity change by bankrupt airlines and non-bankrupt

rivals and average e¢ ciency change in route-level. Focusing on the 1000 most travelled domestric routes in

each quarter from 1998Q1 to 2008Q2, we �nd that bankrupt airlines reduce fare as well as capacity signi�cantly

in the periods surrounding bankruptcy and non-bankrupt rivals reduce fare on averages but increase capacity

in response. Although average non-bankrupt rivals�pro�tability seems to decrease due to the reduced fare,

low cost airlines among those non-bankrupt rivals do not cut their fare and increase capacity even more.

As a result, the route-level capacity decreases a little but the mix of capacity seems to change in favor of

e¢ cient �rms in the sense that low cost airlines replace relatively high-cost bankrupt airlines�capacity. Low

cost airlines seem to take the capacity cut by bankrupt airlines as an opportunity to expand and the average

e¢ ciency seems to improve during the bankruptcy process. This result raises another interesting question of

what it takes for e¢ cient �rms with lower cost to take markets from ine¢ cient incumbents. Bankruptcy seems

one factor that spurs low cost airlines growth. Other factors stimulating low cost carrier expansion will be

investigated and compared to bankruptcy e¤ects.

1 Introduction

Do loose bankruptcy laws allow ine¢ cient airlines to survive and underprice their e¢ cient rivals, harming

the industry? This paper attempts to see how bankrupt airlines behave, how their non-bankrupt competitors

respond, and how the industry e¢ ciency change as a result.

Bankrupt airlines do not necessarily disappear from market right away. The United States has a unique

bankruptcy code called Chapter 11 which, unlike liquidation bankruptcy of Chapter 7, permits bankrupt

�rms to reorganize themselves under protection from creditors when the �rms have higher value as a going-

concern than immediate liquidation value. Some people criticize bankrupt �rms under protection harming

1

non-bankrupt rivals in the airline industry. Chapter 11 allows ine¢ cient airlines to survive and underprice

others, harming even their healthier counterparts, they contend. Many business press dealt with non-bankrupt

airlines complaining about bankrupt �rms�underpricing and overcapacity. The following is the quote from an

article on entrepreneur.com:

According to Robert Crandall, a former CEO of American Airlines, bankrupt airlines enjoy

competitive advantages over rivals not in bankruptcy. A bankrupt airline can defer debt payments,

modify labor agreements, and postpone pension contributions. Crandall theorizes that a bankrupt

airline can lower its �nancing and operating costs, thereby luring customers away from competitors

by o¤ering lower prices. Similarly, Nigel Milton, Virgin Atlantic�s government a¤airs manager, said,

"Chapter 11 is a type of state aid. The playing �eld gets tilted more and more against US." These

lower prices have the e¤ect of forcing nonbankrupt airlines to reduce costs and shrink their pro�t

margins, perhaps bringing these carriers closer to bankruptcy themselves.1

Indeed, under bankruptcy protection, bankrupt airlines may renegotiate with workers and suppliers so enjoy

greater cost advantage over other airlines, which can lead to underpricing that hurts other airlines�pro�tability.

Also, they may price aggressively in order to generate cash and reduce the likelihood of bankruptcy �ling or

immediate liquidation. The tendency to trigger a fare war under �nancial distress is reported by Busse (2002).

However, lower fare during bankruptcy does not necessarily mean that bankrupt airlines harm non-bankrupt

airlines by forcing them to match the low fare and lowering their pro�tability. For one, travelers would discount

bankrupt airlines to give non-bankrupt airline more room for setting higher price than bankrupt airlines, which

makes fare cut by bankrupt airlines less e¤ective. In this case, fare cut by non-bankrupt airlines will not be

signi�cant. Borenstein and Rose (1995) �nd that the fare cut by bankruptcy �ling airlines seems to start prior

to the actual �ling but dissipates quickly during bankruptcy and their rivals do not change fare signi�cantly

during the same period. Recently Ciliberto and Schenone (2008) looked at the changes in price and capacity

during and after Chapter 11 bankruptcy. They �nd that non-bankrupt rivals do not cut fares to match bankrupt

airlines� fare. They also report that bankruptc airlines reduce capacity but non-bankrupt rivals marginally

reduce or even increase capacity. Government Accountability O¢ ce (2005) also �nd that the reduced capacity

is quickly �lled by other airlines.

If bankrupt airlines shrink their operation so as to reduce expenses, this may present new openings for other

airlines to increase their presence. Then, who takes advantage of this opportunity and how does the resulting

change a¤ect industry e¢ ciency? If expanding non-bankrupt airlines are more e¢ cient than bankrupt airlines,

then the e¢ ciency may even improve, not deteriorate, on bankrupt routes. This paper con�rms that bankrupt

airlines reduce capacity and non-bankrupt airlines increase it. Collectively, the capacity in route-level does

not change much. However, composition of capacity may change. That is, if e¢ cient �rms replace ine¢ cient

bankrupt �rms, it will improve average e¢ ciency of the industry.

Figure 1 to Figure 3 show the fare change in the periods surrounding bankruptcy �lings of airlines serving

the route from ATL (William B. Harts�eld Atlanta International Airport, Atlanta, Georgia) to CLT (Char-

lotte/Douglas International Airport, Charlotte, North Carolina). Q1_fare is 25% percentile fare, Q3_fare is1http://www.entrepreneur.com/tradejournals/article/168283785_2.html

2

Figure 1: Bankruptcy and Fare Change: US Airways

Figure 2: Bankruptcy and Fare Change: Delta

75% percentile fare, and Med_fare is median fare of an airline serving the route (Figure 1: Delta, Figure 2:

US Airways, and Figure 3: AirTran Airways).2 The dashed line is when US Airways �led for bankruptcy (for

two times) and the solid line is when Delta �led for bankruptcy. The shaded areas are the periods during

bankruptcy. 3

From Figure 1 and 2, we can see the fare cuts by bankrupt �rms precede the actual bankruptcy �lings, which

may be due to negative demand shock or desperate move of �nancially distressed �rms to raise liquidity. The

fare cut seems to continue during bankruptcy. Non-bankrupt rival seems to cut fare during the same period.

These graphs suggest that bankrupt airlines may adopt aggressive pricing, potentially lowering pro�tability

for non-bankrupt rivlas serving the route.

2All fares are measured in 2000$.3For the exact date of bankrupt �lings and emergence, see Table 2 (Airline Bankruptcy Filings).

3

Figure 3: Bankruptcy and Fare Change: AirTran Airways

However, a low-cost carrier (LCC) AirTran Airways enters the route4 when Delta is about to �le for

bankruptcy. It lowers fare upon Delta�s bankruptcy �ling but do not seem to continue the fare cut, especially

for the higher range of fares while Delta is under bankruptcy protection. This suggests that the presence of

bankrupt �rm does not necessarily mean low pro�tability of non-bankrupt rivlas and may even present a new

opportunity for e¢ cient �rms to expand. As we will analyze in empirical sections, bankrupt airlines reduce

capacity and non-bankrupt rivals �ll the reduced capacity. Especially low cost airlines seem to expand their

services signi�cantly in response to bankrupt airlines�reduced capacity.

When it comes to the question of whether bankrupt airlines harm non-bankrupt airlines, an important part

we need to think about is how e¢ cient non-bankrupt airlines are relative to bankrupt airlines. If bankrupt

airlines expedite ine¢ cient airlines�exit, that is good. If bankrupt airlines harm e¢ cient airlines�pro�tability

and �nancial health, that could be bad. If bankrupt airlines improves e¢ ciency by reducing excess capacity

and e¢ cient non-bankrupt airlines �ll the gap, that would be great. All stories are possible in theory. This

paper attempts to answer which is the case empirically.

In particular, we will use price and quantity data before, during, and after bankruptcy period to see

when the bankruptcy begins to a¤ect the industry and whether the bankruptcy e¤ect persists. We separate

bankruptcy �lings into two cases: those ending up with re-emergence and those ending up with liquidation.

The comparison can be informative about whether the actual liquidation bene�ts non-bankrupt rivlas and

whether the bankrupt �rms that have chosen di¤erent strategies end up with di¤erent outcomes. Also we will

separate the low cost airlines� response to rival�s bankruptcy from that of the average non-bankrupt �rms.

The di¤erence in response is estimated to be large and signi�cant. Throughout the paper, we mean lower cost

by "e¢ ciency".

The short conclusion is that (1) bankruptcy �ling airlines start the fare and capacity cut prior to the actual

�ling, (2) average non-bankrupt rivals cut fare but not as much as bankrupt airlines do, (3) bankrupt airlines�

4We consider only carriers with at least 1% passenger share on a route. So, the entry or presence on a route means a carrierhas no less than 1% passenger share.

4

reduced capacity is �lled mostly by non-bankrupt low cost airlines and those airlines even increase fare during

rivals�bankruptcy, (4) the total capacity in route-level decreases during bankruptcy in re-emergence case and

does not change in liquidation case, (5) bankrupt airlines tend to shrink their network size and concentrate the

services on hub/focus airports where they are believed to have comparative advantages over low cost airlines,

and (6) the route average (labor) unit cost decreases during bankruptcy, possibly as result of bankrupt airlines�

cost cutting and non-bankrupt low cost airlines replacing ine¢ cient bankrupt airlines�capacity.

Low cost airlines�expansion during rival�s bankruptcy (especially when large network carriers are bankrupt)

raises another interesting question. Given the long history of the airline industry since deregulation, the growth

of low cost carrier has occurred mostly in recent years (in 2000�s). That is, lower cost is not all it takes for

e¢ cient �rm to take markets from less e¢ cient incumbents. A discrete event forcing a �rm to cut capacity

such as September 11 or bankruptcy �ling is necessary to urge an incumbent �rm to give up some capacity. So,

next question would be what are the factors that allow e¢ cient �rms, that is, low cost airlines to increase their

presence and how much fraction of the low cost carrier growth can be explained by other carriers�bankruptcy

relative to other factors. As a start, we will compare the September 11 e¤ect and bankruptcy e¤ect on capacity

composition in the industry. More factors such as merger and fuel cost shock will be examined in the future.

The remainder of this paper proceeds in the following steps. Section two describes sample construction,

variables used in empirical analysis, and summary statistics. Section three discusses econometric speci�ca-

tions and estimation results. Section four examines how the estimation results change over various other

speci�cations for robustness check. Finally, Section �ve concludes.

2 Data

2.1 Sample Construction

There are two main data sets used in the analysis; the Airline Origin and Destination Survey Data Bank 1B

(DB1B) and the Air Carrier Statistics database (T-100 data bank). Both are available from the Bureau of

Transportation Statistics of the U.S. Department of Transportation.5 First, the Airline Origin and Destination

Survey DB1B is a 10% sample of airline tickets from reporting carriers collected by the O¢ ce of Airline

Information of the Bureau of Transportation Statistics. The data set includes origin, destination and other

itinerary details such as ticket price, number of passengers transported, ticketing (i.e. marketing) carrier,

operating carrier, distance of the itinerary, number of stops (number of coupons used ina itinerary), whether

the ticket is a round trip, etc., on a quarterly basis.6

Second, we restrict our attention to U.S. domestic airline market so use T-100 Domestic Market (U.S.

Carriers) and T-100 Domestic Segment (U.S. Carriers) data from the Air Carrier Statistics database. The

5http://www.transtats.bts.gov/6The data is recorded when a ticket is used, not when it is purchased, so the timing of the change in an airline�s competitive

behavior and the market outcome may not be exact. However, if most people buy tickets within one or two monthes ahead of anactual �ight date, this may not be a big problem.

5

"market" data includes a monthly air carrier passenger tra¢ c information by enplanement for operating carrier,

origin, destination combination each time period. The "market" data records the passengers that enplane and

deplane between two speci�c points, regardless of the number of stops between the two points. This market

de�nition is comparable to the origin and destination pair in DB1B. On the other hand, the "segment" data

contains the number of seats available,the number of scheduled departures and departures performed, by

operating carrier, origin, and destination. Unlike in the "market" data, the "segment" is composed of a pair

of points served or scheduled by a single stage.7

A route is de�ned as a pair of origin and destination (on an airport basis) and each route is regarded as

one market. A route is treated in a direction-manner in the sense that, if origin and destination airports are

switched, it is considered to be a di¤erent route. Direction matters because demand conditions are di¤erent

even between the same two end points, depending on which way passengers are heading.8 Using the T-1000

Domestic Market database, we pick the 1000 largest routes in each quarter from 1998Q1 to 2008Q2, based

on passenger enplanements. The 1000 routes represent a signi�cant portion of airline market demands. For

instance, in 2007, the number of passengers who travelled the 1000 largest routes is about 60% of the total

demand. The impact of bankruptcy may be heterogenous over maket size. So, this analysis focuses on the

1000 most travelled routes in each quarter for the fourty two quarters.

The observation unit in DB1B is itinerary level. We aggreate the data to carrier level using the number

of passengers as a weight. So, we have one observation for a (ticket) carrier9 on a route in a give time (year,

quarter) in the �nal data set. In the route-level analysis, itinerary level observations are aggregated to route-

level so that we have one observation for a route in a give time. Again, observations are weighted with number

of passengers. Besides, we include the observations if a carrier has at least 1% passengers on a route in a given

time. All market fares are in�ation adjusted in 2000 dollars.10

We consider seventeen carriers with signi�cant market presence that represent each carrier group, seven

from "legacy" carrier group, seven from "low cost" carrier group, and three from "others" carrier group. Table

1 is the list of all carriers used in analysis.

Bankruptcy data is constructed mainly from Lynn M. LoPucki�s Bankruptcy Research Database (BRD)11 and "U.S. Airline Bankruptcies & Service Cessations" listed on Air Transportation Association (ATA)

website.12 BRD contains Chapter 11 �lings of public companies with assets over $100 million that are required

to �le a form 10-K with SEC. The list of bankruptcy �lings on ATA web page includes both Chapters 7 and

11, regardless of the size of a bankrupt airline. However, the web page says the list is "loose, uno¢ cial". So,

we googled the Chapter 7 bankruptcy �lings from ATA case by case. Also, when the dates of bankruptcy

�ling, emergence, or service cessation do not match between the two sources, we searched for news articles on

7For example, if Southwest operates only connecting �ights from San Francisco airport (SFO) to Chicago Midway airport(MDW), the �ights will be recorded in DB1B and the "market" data but not in the "segment" data.

8For example, when Superbowl is held in Tampa, Florida, demands for tickets going to and coming from Tampa would bedi¤erent.

9A ticket carrier and an operating carrier can be di¤erent for the same itinerary. We choose a ticket carrier over an operatingcarrier because a ticket carrier sets a price even though other carrier may actually operate the service.10Consumer Price Index - All Urban Consumers is available from http://data.bls.gov/cgi-bin/surveymos.11http://www.webbrd.com/bankruptcy_research.asp12http://www.airlines.org/economics/specialtopics/USAirlineBankruptcies.htm

6

a speci�c bankruptcy event and picked the more accurate one. From these sources, we construct the history

of airline bankruptcies that we are interested in. Table 2 shows all bankruptcy events that we will account

for in the analysis. There are twenty one bankruptcy �lings between 1998Q1 to 2008Q2. Among them,

bankruptcy �ling airlines re-emerged in ten cases,13 went out of business after bankruptcy protection in nine

cases, and ceased services right away in two cases. It is noteworthy that �ve out of ten re-emergency cases,

bankrupt airlines are legacy carriers with large network. So, the legacy carriers comprise most of observations

of bankruptcy-related variables in re-emergence case.

Table 1. Carrier List

Carrier group Carrier Name Code Status *

American Airlines AA

Continental Airlines CO

Delta Airlines DL Re-emerged from bankruptcy

Legacy Northwest Airlines NW Re-emerged from bankruptcy

United Airlines UA Re-emerged from bankruptcy

US Airways US Re-emerged from bankruptcy twice

Alaska Airlines AS

Southwest Airlines WN

ATA Airlines TZ Re-emerged but liquidated later

JetBlue Airways B6

Low Cost AirTran Airways FL

Frontier Airlines F9 Under Ch 11

Spirit Airlines NK

American West Airlines HP Merged by US

Midway Airlines JI Liquidated

Others Hawaiian Airlines HA Re-emerged from bankruptcy

Trans World Airlines TW Bankrupt then merged by American

* Status change from 1998 to 2008

Note that even though some bankrupt airlines are not used for analysis directly, the bankruptcy events

are accounted for in order to see other airlines�reactions to bankruptcy on a¤ected routes. We assume that

bankruptcy e¤ects are the same for the same group of airlines, that is, a group of bankruptcy �ling carriers that

emerged from bankruptcy, a group of their non-bankrupt competitors, agroup of bankruptcy �ling carriers that

went out of business in the end, and a group of their non-bankrupt competitors. Later, we divide non-bankrupt

carriers into two sub groups: LCC vs. non-LCC.

13Frontier Airlines �led for bankruptcy in the second quarter of 2008 and are still under bankruptcy protection. The case isregarded as an emergence case in the analysis. However, treating this case as liquidation does not change the results.

7

Table 2. Airline Bankruptcy Filings

Date of Date of Date of

Carrier Name Filing Ch. Re-emergence Service Cessation

Kiwi International (KP) Mar 23, 1999 11 Dec 8, 1999

Eastwind Airlines (W9) Sep 30, 1999 7

Tower Air (FF) Feb 29, 2000 11 Dec 7, 2000

Pro Air (P9) Sep 19, 2000 11 Sep 19, 2000

National Airlines (N7) Dec 6, 2000 11 Nov 6, 2002

Midway Airlines (JI) Aug 14, 2001 11 Oct 30, 2003

Trans World Airlines (TW)* Jan 10, 2001 11 Dec 1, 2001

Sun Country Airlines (SY)** Jan 8, 2002 7 April 15, 2002

Vanguard Airlines (NJ) July 30, 2002 11 Dec 19, 2004

United Airlines (UA) Dec 9, 2002 11 Feb 2, 2006

US Airways (US) 1st Aug 11, 2002 11 Mar 31, 2003

Hawaiian Airlines (HA) Mar 21, 2003 11 June 2, 2005

ATA Airlines (TZ) 1st Oct 26, 2004 11 Feb 28, 2006

US Airways (US) 2nd Sep 12, 2004 11 Sep 27, 2005

Aloha Airlines (AQ) 1st Dec 30, 2004 11 Feb 17, 2006

Delta Airlines (DL) Sep 14, 2005 11 April 25, 2007

Northwest Airlines (NW) Sep 14, 2005 11 May 18, 2007

Independence Air (DH) Nov 7, 2005 11 Jan 5, 2006

Aloha Airlines (AQ) 2nd Mar 31, 2008 7

ATA Airlines (TZ) 2nd April 3, 2008 11 April 3, 2008

Frontier Airlines (F9) April 10, 2008 11

* Trans World is merged by American,

** Sun Country�s bankruptcy procedure was converted from Ch.7 to Ch.11

The basic empirical approach is in the same spirit with the di¤erence-in-di¤erence. We will compare a

bankruptcy-a¤ected airlines�pricing and quantity setting behavior to the normally expected behavior of those

airlines�without bankruptcy. Su¢ cient number of observations on tickets una¤ected by bankrupty will allow

us to estimate unbiased counterfactual patterns of fare/capacity set by airlines. Those data una¤ected by

bankruptcy (so can be used to estimate the counterfactuals absent bankruptcy events) come from two sources:

data from periods prior to bankruptcy and data from routes where no airline is bankrupt. We have at least

�ve quarters ahead of bankruptcy �ling and, for most of bankruptcy cases, we have more than several quarters

ahead of bankruptcy �lings. Among the 1000 largest routes each quarter, at least part of the routes are not

bankrupt.

8

2.2 Variables

In the empirical analysis, we will see how bankrupt airlines set price and quantity right before, during, after

bankruptcy and non-bankrupt rivals behave in response to them. Thus, the bankruptcy-related variables are

constructed in the manner that we can capture how a bankrupt �rm�s and its competitors�behaviors change

over time in the periods surrounding bankruptcy. We construct bankruptcy-related variables as an interaction

between carrier identity (based on whether bankrupt or not and how a bankrupt carrier of interest end up

with; re-emergence or liquidation) and time periods (pre, during, and post bankruptcy periods).

Table 3 is the list of bankruptcy-related variables in re-emergence case and liquidation case, respectively.

By re-emergence case, we mean that a bankrutcy �ling airline of interest has been successful in emerging

from bankruptcy. By liquidation case, we mean that a bankrutcy �ling airline of interest ended up being

liquidated. In the re-emergence case (Panel 1), Em_Bankrupt[TB]irt, for example, is a dummy variable that

is triggered on if a carrier �les for bankruptcy in current quarter and Em_NonB[TB � 1]irt is a dummyvariable indicating that a carrier is serving the route where some other carriers �le for bankruptcy in next

quarter. The bankruptcy-related variables in liquidation case (Panel 2) are constructed similarly, except that

bankruptcy �ling airlines do not have post-bankruptcy periods as they are liquidated, and only one quarter

after liquidation is considered for non-bankrupt rivals as their strategic change after liquidation are expected

to be mostly done in the quarter. Li_NonB[T ]irt, for instance, is the indicator of non-bankrupt airlines that

serve the routes where some bankrupt airlines served and were liquidated in last quarter.14

Table 3. Variable List: Bankruptcy-related variables

Panel 1: Re-emergence case

Period Bankrupt NonB

Pre B [TB-2] Em_Bankrupt[TB � 2]irt Em_NonB[TB � 2]irt[TB-1] Em_Bankrupt[TB � 1]irt Em_NonB[TB � 1]irt

During B [TB ] Em_Bankrupt[TB ]irt Em_NonB[TB ]irt[TB+1] Em_Bankrupt[TB + 1]irt Em_NonB[TB + 1]irt[TB+2~T] Em_Bankrupt[TB + 2~T ]irt Em_NonB[TB + 2~T ]irt

Post B [T] Em_Bankrupt[T ]irt Em_NonB[T ]irt[T+1] Em_Bankrupt[T + 1]irt Em_NonB[T + 1]irt[T+2~] Em_Bankrupt[T + 2~]irt Em_NonB[T + 2~]irt

Panel 2: Liquidation case


Pre B [TB-2] Li_Bankrupt[TB � 2]irt Li_NonB[TB � 2]irt[TB-1] Li_Bankrupt[TB � 1]irt Li_NonB[TB � 1]irt

During B [TB ] Li_Bankrupt[TB ]irt Li_NonB[TB ]irt[TB+1] Li_Bankrupt[TB + 1]irt Li_NonB[TB + 1]irt[TB+2~T] Li_Bankrupt[TB + 2~T ]irt Li_NonB[TB + 2~T ]irt

Post B [T] (No Observation: Liquidated) Li_NonB[T ]irtTB : Quarter of bankruptcy �ling, T: First Quarter after bankruptcy procedure ends.

14Detailed descriptions about the bankruptcy related variables are in Table A1-A2 in Appendix A.

9

Table 4 is route-level bankruptcy-related variables. Route-level analysis is intended to see the capacity

change in total on bankruptcy-a¤ected routes, as a result of �nancial distress, bankruptcy, emergence, or

liquidation. The comparison group is the routes where any carrier is not bankrupt. Table 5 is the list of other

variables used in the analysis. Descriptive statistics on these variables will be reported in the next section.

Table 4. Variable List: Bankruptcy-A¤ected Routes

Period Re-emergence Liquidation

Pre B [TB-2] Em_Bankrupt_route[TB � 2]rt Li_Bankrupt_route[TB � 2]rt[TB-1] Em_Bankrupt_route[TB � 1]irt Li_Bankrupt_route[TB � 1]rt

During B [TB ] Em_Bankrupt_route[TB ]irt Li_Bankrupt_route[TB ]rt[TB+1] Em_Bankrupt_route[TB + 1]irt Li_Bankrupt_route[TB + 1]rt[TB+2~T] Em_Bankrupt_route[TB + 2~T ]irt Li_Bankrupt_route[TB + 2~T ]rt

Post B [T] Em_Bankrupt_route[T ]irt Li_Bankrupt_route[T ]rt[T+1] Em_Bankrupt_route[T + 1]irt[T+2~] Em_Bankrupt_route[T + 2~]irt

TB : Quarter of bankruptcy �ling, T: First Quarter after bankruptcy procedure ends.

Table 5. Variable List: Other Variables

Variable Unit Description

Price Med_fareirt 2000$ Median fare of a carrier i on route r at time t

Q1_fareirt 2000$ 25% percentile fare of a carrier i on route r at time t

Q3_fareirt 2000$ 75% percentile fare of a carrier i on route r at time t

Capacity N_seatsirt 1 # of available seats by a carrier i on route r at time t

N_seats_allrt 1 Total # of available seats on route r at time t

N_dprtsirt 1 # of scheduled departures by a carrier i on route r at time t

N_dprts_allrt 1 Total # of scheduled departures on route r at time t

Route LCCinrt Dummy: 1 if LCC serves route r at time t, 0 otherwise

Characteristics SWinrt 1 if Southwest serves route r at time t, 0 otherwise

HHIrt 1/1000 Her�ndahl-Hirschman Index

N_carriersrt 1 Number of carriers serving route r at time t

Distancer mile distance between origin and destination of route r

Carrier Networkit 1/1000 Number of routes a carrier i is serving at time t

Characteristics Net_originirt 1/1000 # of routes from the origin of route r a carrier i is serving at time t

Net_destirt 1/1000 # of routes to the destination of route r a carrier i is serving at time t

Net_focusirt 1 Net_originirt/Networkit%directirt 1 Ratio of direct �ights tickets of a carrier i on route r at time t

%roundirt 1 Ratio of round trip tickets of a carrier i on route r at time t

%codeshrirt 1 Ratio of tickets of a carrier i on route r at time t

operated by a di¤erent carrier

Mkt_shareirt 1 Market share of a carrier i on route r at time t

in terms of passenger enplanement

LCCi Dummy: 1 if a carrier i is LCC, 0 otherwise

10

2.3 Summary Statistics

Summary statistics give us basic information about the data, helping us to get a rough idea of what to expect

before we analyze with detailed regression models. Table 6 and 7 show basic summary statistics by various

samples. In particular, Table 6 is for carrier-level observations and Table 7 is for route-level observations.

In Table 6, we compare summary statistics between bankrupt airlines and their non-bankrupt competitors.

We divide the sample based on how the bankruptcy procedure ends. Panel 1 (Panel 2) reports summary

statistics for bankruptcy �lings ending up with re-emergence (liquidation). First of all, "All" (column (a)) is

the entire carrier-level sample used in the analysis. "Bankrupt" is the sample of bankrupt airlines only, whereas

"Non-Bankrupt" is the sample of non-bankrupt competitors that serve the same route. "PreB", "DuringB",

and "PostB" indicates pre-bankruptcy (for two quarters), during bankruptcy protection, and post-bankruptcy

periods (for two quarters if a bankrupt airline has emerged and for one quarter if a bankrupt airline went out

of business), respectively. For example, "Bankrupt/PreB" (column (b) or column (h)) is for the sample of

carriers �ling for bankruptcy in two quarters and "Non-Bankrupt/DuringB" (column (f) and column (k)) is

for the sample of non-bankrupt carriers that serve the route where some other carriers are bankrupt. Note

that there is no "Bankrupt/PostB" column in Panel 2 because all the bankrupt carriers are liquidated. Also,

"Non-Bankrupt/PostB" (column (l)) in Panel 2 is for the sample of carriers that serve the route where a

bankrupt carrier has gone out of business in last quarter, which is intended to see how these airlines perform

when a bankrupt airline disappears.

Table 6. Summary Statistics: Carrier-Level Observations

Panel 1: All & Re-emergence Sample

Re-emergence

Bankrupt Non-Bankrupt

Variable All (a) PreB (b) DuringB (c) PostB (d) PreB (e) DuringB (f) PostB (g)

Med_fare 132.29 124.22 129.40 132.57 131.43 132.70 137.99

(Unit:2000$) (50.79) (42.05) (42.50) (43.58) (40.37) (39.85) (42.88)

Network 0.453 0.496 0.538 0.501 0.443 0.427 0.424

(1=1000) (0.187) (0.166) (0.146) (0.136) (0.199) (0.194) (0.193)

N_seats 64,871 68,785 64,470 61,472 56,534 56,969 54,934

(1) (49482) (52349) (47945) (48081) (41504) (42562) (41,607)

N_dprts 483 478 469 449 408 427 425

(1) (382) (341) (329) (316) (363) (306) (310)

Mktshare 0.23 0.20 0.19 0.19 0.19 0.20 0.20

(1) (0.27) (0.26) (0.25) (0.25) (0.23) (0.24) (0.24)

LCC 0.22 0.10 0.03 0.01 0.28 0.31 0.34

(0.42) (0.29) (0.17) (0.10) (0.45) (0.46) (0.47)

LCCin 0.72 0.68 0.70 0.68 0.79 0.79 0.78

(0.45) (0.47) (0.46) (0.47) (0.43) (0.41) (0.41)

SWin 0.47 0.18 0.22 0.26 0.21 0.25 0.30

(0:44) (0:39) (0:42) (0:44) (0:41) (0:43) (0:46)

N 178,443 5,718 20,865 4,890 17,356 51,116 13,639

N_sgmt 81,593 2,112 8,446 1,913 6,738 21,617 5,839

11

Panel 2: All & Liquidation Sample

Liquidation


Variable All (a) PreB (h) DuringB (i) PreB (j) DuringB (k) PostB (l)

Med_fare 132.29 129.80 116.60 146.58 136.16 125.23

(Unit:2000$) (50.79) (46.01) (39.43) (52.95) (49.60) (43.60)

Network 0.453 0.308 0.306 0.472 0.477 0.484

(1=1000) (0.187) (0.133) (0.073) (0.186) (0.176) (0.178)

N_seats 64,871 62,623 66,340 61,194 59,146 58,280

(1) (49482) (30104) (29578) (48078) (45518) (41558)

N_dprts 483 437 489 414 396 395

(1) (382) (217) (224) (320) (291) (265)

Mktshare 0.23 0.02 0.00 0.16 0.15 0.18

(1) (0.27) (0.16) (0.00) (0.20) (0.19) (0.22)

LCC 0.22 0.66 0.68 0.18 0.20 0.19

(0.42) (0.47) (0.47) (0.39) (0.40) (0.39)

LCCin 0.72 0.17 0.19 0.74 0.77 0.71

(0.45) (0.38) (0.39) (0.44) (0.42) (0.45)

SWin 0.47 0.398 0.378 0.18 0.18 0.21

(0:44) (0:124) (0:123) (0:38) (0:38) (0:41)

N 178,443 972 1,349 6,178 9,237 1,866

N_sgmt 81,593 193 275 2,213 3,084 623

Standard Deviations are reported in parentheses. N: sample size.

N_sgmt: Nonstop �ight only sample size (N_seats, N_dprts only avilable for this segment sample)

In Panel 1, fares seem to rather increase than decrease while airlines are going through bankruptcy; median

fare goes up from pre-bankruptcy periods to bankruptcy protection periods then to post-bankruptcy periods.

However, we need to note that the airlines emerged from bankruptcy are usualy large network carriers who

used to charge high price, and also those large carriers tend to stay longer under bankrupt protection (so more

observations during bankruptcy). Thus, this result does not necessarily mean that bankruptcy leads to higher

price In fact, the regression analyses in the next section shows that bankrupt airlines and their competitors

cut price in the periods surrounding bankruptcy.

On the other hand, bankrupt airlines seem to reduce capacity (N_seats and N_dprts) during bankruptcy

whereas non-bankrupt competitors increase capacity. This pattern is con�med in the regression analyses.

Related to this pattern, a bankrupt �rm�s market share is lower under bankruptcy protection whereas its

competitor�s market share is higher in the same periods. In addition, it is noteworthy that the portion of

routes that low cost airlines such as Southwest serve increases during bankruptcy procedures (see LCCin and

SWin). This may indicate that more e¢ cient airlines �ll the reduced capacity by bankrupt �rms. In the data

periods (from 1998 to 2008), Low cost airlines growth has been signi�cant. Ine¢ cient �rms�bankruptcy and

capacity reduction may give low cost airlines more opportunities to increase their presence. This possibility

will be explored more in the next empirical section.

12

In Panel 2, we deal with the bankruptcy �lings that actually end up going out of business. These cases

involve smaller airlines with small market share, relatively to the former cases where bankrupt ailrines have

survived the process. Panel 2 shows that fares are lower during bankruptcy regardless of whether an airline is

bankrupt itself or its competitor is bankrupt. Contrary to Panel 1, bankrupt airlines�capacity level is rather

higher during bamkruptcy whereas its competitor�s capacity level is lower during the same period.15

Table 7 reports summary statistics for all route-level observations and sample of bankruptcy-a¤ected routes

only. By bankruptcy-a¤ected routes, we mean that some airlines that serve the route will �le for bankruptcy

in two quarters ("PreB": column (n) and column (q)), are bankrupt ("DuringB": column (o) and column

(r)), have re-emerged less than two quarters ago ("Re-emergence/PostB": column (p)), or have gone out of

business in last quarter ("Liquidation/PostB": column(s)).

Table 7. Summary Statistics: Route-Level Observations

Re-emergence Liquidation

Variable All (m) PreB (n) DuringB (o) PostB (p) PreB (q) DuringB (r) PostB (s)

N_seats_all 133,881 125,929 125,370 123,994 128,055 136,118 114,151

(Unite:1) (77059) (71122) (69222) (71023) (74522) (85779) (54400)

N_dprts_all 1,120 951 960 968 976 918 783

(1) (640) (610) (576 (593) (595) (585) (405)

LCCin 0.66 0.65 0.69 0.65 0.63 0.73 0.65

(0.47) (0.48) (0.46) (0.48) (0.48) (0.44) (0.48)

SWin 0.29 0.19 0.22 0.22 0.19 0.19 0.20

(0.45) (0.39) (0.41) (0.41) (0.39) (0.39) (0.40)

HHI 0.544 0.503 0.488 0.508 0.501 0.370 0.456

(1=1000) (0.219) (0.196) (0.180) (0.193) (0.198) (0.123) (0.158)

N_carriers 4.35 4.92 5.02 4.83 4.92 6.91 5.54

(1) (2.20) (2.05) (1.92) (1.99) (2.13) (2.05) (1.85)

Distance 853.28 968.73 998.91 951.46 953.40 1,332.15 1,098.15

(mile) (608.28) (624.90) (612.21) (623.77) (632.12) (658.61) (510.15)

N 42,000 9,282 14,587 8,677 6,174 1,626 338

N_sgmt 41,993 9,281 14,580 8,674 6,173 1,626 338

Standard Deviations are reported in parentheses. N: sample size


We can see route-level capacity stays pretty stable regardless of bankruptcy in "re-emergence" case. Al-

though we need more rigorous analysis controlling for other factors, the little change in route-level capacity,

combined with bankruptcy �ling �rms�tendency to reduce capacity, may indicate that bankrupt �rms�ca-

pacity is replaced by other competitors, which is inconsistent with the non-bankrupt airlines�complaint that

bankrupt airlines are exacerbating excess capacity problem in the industry. In contrast, capacity level is higher

during bankruptcy and lower after bankruptcy in "Liquidation" case. This suggests a possibility that bankrupt

15Summary statistics of other variables used in carrier-level analysis are available in Table A3 of Appendix A.

13

airlines may sell seats at very low price as a desperate attempt to raise cash just before liquidation or as a

process of liquidation, hurting other competitors.

3 Empirical Model and Results

This section discusses econometric models and estimation results as an attempt to answer the questions given

in each subsection. In the process, we will demonstrate a possible machanism by which a �nancially distressed

airline alters the industry. The �rst question is whether a bankrupt �rm harms its competitors and exacerbate

excess capacity problem.

3.1 Does a Bankrupt Firm Harm Non-Bankrupt Rivals?

Price and quantity are the main strategic tools that �rms use in market. Thus, looking at how price and

quantity level change (or do not change) in the periods surrounding bankruptcy can be informative. We

will study how bankrupt airlines behave and how their competitors respond using the following econometric

speci�cation:

Yirt = Em_Bankrupt0irt � �+ Em_NonB0irt � � + Li_Bankrupt0irt � + Li_NonB0irt � �+Xirt � �+ Timet � � + uirt

where an observation unit is a carrier i (= 1; 2; � � � ; 17) on a route r (= 1; 2; � � � ; 1447) at time t (= 1998Q1,1998Q2,� � � , 2008Q2), Yirt is a dependent variable, ln(Med_fareirt) or ln(N_seatsirt), Em_Bankruptirt isa 8 � 1 vector of bankrupt-carrier dummies of a carrier i on a route r at time t in re-emergence case; foreach time period from two quarters before bankruptcy �ling to post-bankruptcy periods, Em_NonBirt is a

8� 1 vector of non-bankrupt competitor dummies in the same period in re-emergence case, Li_Bankruptirtis a 5 � 1 vector of bankrupt-carrier dummies in liquidation case; for each time period from two quarters

before bankruptcy �ling to liquidation, Li_NonBirt is a 6� 1 vector of non-bankrupt competitor dummies inliquidation case; for each time period from two quarters before bankruptcy �ling to post-bankruptcy period

after the rival is liquidated, Xirt is a set of constant and control variables such as LCC in, SW in, HHI,

Net_origin, Net_dest, Network, %direct, %round, and %codeshr if a dependent variable is ln(Med_fare)

and LCC in, SW in, and HHI if a dependent variable is ln(N_seats),16 Timet is a set of time-speci�c

dummies for each year, quarter pair and quarter dummies for Florida route,17 and uirt is the combination of

time-invariant route-carrier �xed e¤ect (�ir) and random shock to a carrier�s fare on a route at speci�c time

(�irt), i.e. uirt = �ir + �irt.18

16See Table 5 for the description of variables. Some control variables are closely related to fare level but not to quantity level.So, those variables are dropped in quantity equation.17As for the quarter dummies for Florida route, see the paragraph on panel ID below.18Detailed descriptions on the speci�cation is in Appendix B.

14

We estimate the speci�cation with the �xed e¤ect model with carrier-route pair as a panel ID. The random

e¤ect model is not proper in this study because an airline serving a speci�c route has a non-random relationship

with fare or capacity level. That is, carrier-route pair individual e¤ect is systematically correlated with fare

or capacity level.19 For instance, it is well-documented that an airline tends to charge higher fare on routes

originated from its hub airport than other airlines or on other non-hub routes.20 We assume that the e¤ect

of a speci�c carrier-route pair on fare/capacity level has a time-invariant component (�ir) and random shock

component (�irt). While the time-invariant component is captured by carrier-route dummies, the random

component varies over time and thus are treated as usual normal error terms (i.e. �irt~N(0; �2).21 We will

adopt the �xed e¤ect model in every estimation in this paper, so the carrier-route pair dummy variables are

included in all carrier-level analyses and route dummy variables are included in all route-level analyses.

Again, the panel ID in the basic econometric speci�cation is a carrier-route pair. However, since airline

market is often characterized by seasonality (.e.g. demand conditions in the �rst quarter di¤er form those in

the third quarter), carrier-route-quarter combination may be another appropriate candidate for the panel ID.

There is a trade-o¤ between these two choices of the panel ID. If we choose carrier-route-quarter combination,

we can control for seasonal adjustment. However, we will have much shorter data periods22 that we can use

to estimate "but for" fare/capacity level, which may lead to a biased estimation of counterfactual patterns.

On the other hand, though choosing carrier-route pair has disadvantage that we do not control for quarterly

adjustment by a carrier on a route, it allows us to have much longer data periods23 that we can depend on to

estimate counterfactual fare/capacity level but for bankruptcy events.

This study chooses carrier-route pair as a panel ID as the objective of this study is to see how market

competition changes in the periods a¤ected by bankruptcy, compared to normal periods. We instead add

quarter dummies if origin or destination airports are located in Florida in addition to time speci�c dummy

variables (from 1998Q2 to 2008Q2: base.= 1998Q1). Time speci�c dummy variables are intended to control

for aggregate demand/supply shocks common to all routes and carriers or common quarterly movement in fare

or capacity. Quarter dummy variables for the route originated from or destined to Florida region are included

because quarterly pattern is similar for most of routes (demand highest in the third quarter and lowest in the

�rst quarter) but the pattern is reversed in Florida region (demand lowest in the third quarter and highest in

the �rst quarter). As we will see later in this section, the estimated coe¢ cients for time speci�c dummies and

Florida quarter dummies show the expected pattern.24

We estimate bankruptcy e¤ect in pre-bankruptcy periods (one and two quarters before bankruptcy �ling)

separately, as a bankruptcy �ling airline will begin to experience �nancial distress at some point prior to the

actual bankruptcy �ling and this may alter the airline�s pricing and quantity setting strategies. Since bankrupt

19We ran Hausman test on the empirical speci�cations above and the null hypothesis that di¤erence in coe¢ cients is notsystematic was rejected .20For one, see Borenstein (1989).21We report Robust Standard Errors in the regression analysis to account for potential heterogeneity.22Then the panel data becomes yearly data set for each carrier-route-quarter combination. So, we have eleven years of observation

at most.23The panel data is a quarterly data set for carrier-route pair. So, we have fourty two quarters of observation at most.24The estimation results do not change qualitatively even if we do not include quarterly dummies for Florida region. Choosing

carrier-route-quarter combination changes the estimation results a bit in the sense that the fare change is larger before �ling forbankruptcy than during bankruptcy procedures. Other than that, the estimation results are similar.

15

airlines usually stay under bankrupty protection for a number of quarters, we will see how their fare/capacity

levels change over time during bankruptcy. After re-emerging from bankruptcy, the bankrupt airline may

change their strategies or go back to their old strategies before bankruptcy �ling. So, post-bankruptcy periods

are also estimated separately. If the distressed ailrine change its strategy, this will lead its competitors to

change their strategies too. Thus, we will see non-bankrupt ailrines�responses as well as bankrupt ailrines

in each period. In addition, we will also look at how non-bankrupt airlines set price and quantity after a

bankrupt airline went out of business. If these airlines are better o¤ after the bankrupt �rm is liquidated, this

may imply that a quick liquidation, instead of operation under bankruptcy protection, boosts remaining �rms�

pro�tability.

Table 8 reports the estimation results. Since the dependent variable is logarithm of fare (or capacity),

the estimated coe¢ cients are interpreted as a semi-elasticity, i.e. % change in median fare (or capacity) in

response to a unit change of RHS variable. In this model, after accounting for carrier-route individual (�xed)

e¤ects, the estimates for bankruptcy-related variables are interpreted as the change in fare (or capacity) of the

same airline on the same route when a¤ected by bankruptcy.

The �rst regression estimation result shows fare change by bankruptcy �ling airlines and their rival airlines,

in re-emergence and liquication case. Fares decrease about 4.1 to 5.4% even before bankruptcy �ling in re-

emergence case. Once an airline �les for bankruptcy, the median fare is down about 9.8%. The median fare

of non-bankrupt airlines on the same route is also down about 2.8%. The fare cut persists during bankruptcy

procedures but the magnitude decreases over time. In liquidation case, on the other hand, the fare cut is

signi�cant in the later stage of bankruptcy procedure (7.3%), that is, when it is close to liquidation, indicating

a "going-out-of-business" sale. The insigni�cance of fare cut in the earlier stage may be because the sample

of liquidated airlines involve only small number of routes so only small number of observations, resulting that

estimates are signi�cant only when size of fare cut is large. After bankrupt airline re-emerge from bankruptcy,

they maintains lower fare by about 3% than before they experience �nancial distress. Though the fare goes

up almost to the original level after two post-bankruptcy quarters, this may be due to huge capacity cut by

bankrupt airlines that continues in post-bankruptcy periods as we can see in the second estimation result on

quantity (number of seats available).

As prices are strategic complements, non-bankrupt rivals seem to follow the fare cut by a bankrupt airline

but only by 1.4-2.8%. This result looks like a distressed �rm harming healthier counterparts by engaging in

aggressively low price. However, there are two things we need to note. First, the fare cut by non-bankrupt

airlines is economically insigni�cant compared to quarterly fare change. The average quarterly fare change

of an airline on a speci�c route is about 3.5% (see Figure 4 for the estimated coe¢ cients on time speci�c

dummies).25 Also, the estimated coe¢ cients of quarter dummies for Florida region, although not reported in

the table, are 0.0265 for the �rst quarter, -0.0587 for the third quarter, and almost zero for the second quarter,

meaning that the same carrier sets fare up about 3% in the �rst quarter and down about 6% in the third

quarter compared to the second and fourth quarters in the Florida region, holding other factors including the

nationwide fare change constant. Second, more importantly, we are not sure yet whether e¢ cient rivals are

25For all time speci�c dummies, base is 1998Q1. So, the estimated change is % change compared to the level in 1998Q1. Recallthat all fares are in�ation adjusted so the estimated change is also in�ation adjusted.

16

Figure 4: % Median Fare Change by Carrier, 1998 through 2008 (base: 1998Q1)

harmed by bankrupt airlines. As we will see in Section 3.3, low cost airlines do not seem to match bankrupt

airlines�low fare. Rather, they raise fare than before. This result means that e¢ cient rivals with lower cost

level are not negatively a¤ected by bankrupt airlines.

Besides, after a bankrupt �rm has gone liquidated, the non-bankrupt carriers that used to compete with

the liquidated �rm seem to charge even lower price than before (by 2.4%). This implies that either deteriorated

demand condition have forced the bankrupt �rm to liquidation and remaining carriers to keep low price. or

remaining carriers and new enterants have expanded more than the reduced capacity by liquidation. Thus, we

need to see the capacity side as well to have a complete picture.

Capacity is measured by the number of seats available.26 It is clear that bankrupt airlines are shrinking

their operations. Capacity is reduced even before the actual �ling (e.g. 15% right before the �ling) and the

magnitude of reduction gets larger during bankruptcy (from 11% at the early stage to 31% in the later stage).

This pattern goes beyond bankruptcy protection period, continueing even in post-bankruptcy periods so the

capacity level is almost cut in half after re-emergence, compared to the periods una¤ected by bankruptcy.

Similarly, in liquidation case, capacity cut becomes larger and signi�cant as the �rm is getting closer to

liquidation.

Non-bankrupt rivals, on the other hand, increase their capacity by about 7% than their normal level, in

response to the reduced capacity by bankrupt airlines in re-emergence case, which does not seem to support

the argument that bankrupt airlines exacerbate excess capaicty problem in the industry. On the other hand, if

we look at liquidation case, non-bankrupt rivals do reduce capacity while a bankrupt airline serves the route,

but it goes back to the original capacity level after the liquidation has actually occurred. So, the reduction by

non-bankrupt competitors can be either because of diminishing demand or because of aggressive new entries

intended to �ll the reduced capacity by bankrupt airlines. The possibility of new entries induced by bankruptcy

�ling �rms�capacity reduction and the total capacity in the route level will be studied in next section (Section

3.2).26Usually, airline industry capacity is measure by available seat miles (number of seats times the distance between the two end

points of a route: ASM). Here, we compare capacity change by the same carrier on the same route so the number of seats issu¢ cient to measure capacity.

17

Table 8. Estimation Results: Price and Quantity

Dependent Var. ln(Med_fare) ln(N_seats)

Re-emergence Liquidation Re-emergence Liquidation

Variable Bankrupt NonB Bankrupt NonB Bankrupt NonB Bankrupt NonB

[TB-2] -.0413*** -.0018 .0213*** .0008 -.0900*** .0141 -.0147 -.0328*

(.0043) (.0029) (.0080) (.0044) (.0257) (.0130) (.0264) (.0184)

[TB-1] -.0539*** -.0102*** -.0051 .0158*** -.1498*** .0458*** -.0869 -.0485**

(.0041) (.0025) (.0083) (.0040) (.0262) (.0126) (.0544) (.0189)

[TB ] -.0983*** -.0275*** -.0021 .0037 -.1097*** .0669*** -.0183 -.0971***

(.0045) (.0030) (.0096) (.0052) (.0267) (.0153) (.0277) (.0250)

[TB+1] -.0845*** -.0230*** -.0122 -.0150** -.1856*** .0707*** -.0642** -.1579***

(.0050) (.0035) (.0104) (.0061) (.0287) (.0157) (.0262) (.0252)

[TB+2~T] -.0634*** -.0142*** -.0730*** -.0325*** -.3136*** .0726*** -.1811*** -.0876***

(.0033) (.0024) (.0091) (.0038) (.0152) (.0096) (.0494) (.0182)

[T] -.0326*** -.0049 -.0235*** -.4134*** .0551*** .0249

(.0053) (.0036) (.0054) (.0325) (.0163) (.0268)

[T+1] -.0342*** -.0107*** -.4693*** .0254

(.0054) (.0038) (.0341) (.0175)

[T+2~] -.0028 -.0354*** -.5475*** .0110

(.0042) (.0030) (.0224) (.0128)

LCCin -.0831*** -.0486***

(.0027) (.0121)

SWin -.0992*** .0935***

(.0035) (.0196)

HHI .0853*** -.4680***

(.0079) (.0421)

Net_origin .5953***

(.2237)

Net_dest .4551**

(.2260)

Network .0425***

(.0150)

%direct -.0326***

(.0050)

%round -.5548***

(.0100)

%codeshr .1126***

(.0050)

Constant 5.3658***

(.0127)

N 178,443 81,564

Robust SE reported in parentheses. Time speci�c dummies included. N: Sample size

* Signi�cant at 10 %, ** Signi�cant at 5 %, *** Signi�cant at 1 %

18

The estimated coe¢ cients on other variables seem to make sense. First of all, in the fare equation, when

low cost airlines are present in a route, airlines cut fare by 8.3%. If the low cost airline is Southwest, the

fare cut is even larger by 9.9% so the total fare cut under the presence of Southwest is huge, about 18.2%.

Concentration in a route is positively associated with fare. So airlines set price about 0.85% higher when

Her�ndahl Index (HHI) is higher by 100 (unit of HHI variable = 1/1000). When the number of routes from

the origin (or to the destination) that an airline serves is higher by 100, fare level is also higher by 6% (4.6%).

The total number of routes is also postively related to fare level, though the e¤ect is minimal after the the

network size from the origin and to the destinationa are controlled. The portion of direct �ights and that of

round trips are negatively related to fare level (0.3% lower if 10% more direct �ights, 5.5% lower if 10% more

round trip tickets). When a ticket carrier and an operating carrier are di¤erent, that is, an actual �ight is

operated by a di¤erent carrier from a carrier that sets fare and sells the ticket, the fare level is higher.

In the capacity equation, the presence of low cost airlines is negatively related to total capacity on average

(about 5% lower when low cost airlines are present) but the presence of Southwest in particular is positively

related to total capacity (about 4% higher when Southwest is present). Concentration has negative relationship

with total capacity (4.7% lower when HHI is higher by 100).

3.2 Are Bankrupt Routes Less or More Attractive?

In the previous section, we have seen carrier-level changes in the periods surrounding bankruptcy; bankrupt

airlines cut fare as well as capacity whereas non-bankrupt rivals cut fare but expand capacity in re-emergence

case. Although both bankrup and non-bankrupt airlines cut fare and capacity during bankruptcy in liquidation

case, non-bankrupt airlines increase capacity after bankrupt rival�s liquidation. The fare cut may indicate that

bankrupt routes become less attractive but the capacity cut by bankrupt airlines may present an opportunity

for other airlines to enter the route or increase their presence. Thus we will look whether the number of airlines

serving the bankrupt routes (with at least 1% market share) increases or decreases and how the total capacity

level on the routes changes as a result. Table 9 reports the estimation results.

The �rst estimation result shows that a route tends to be more crowded when some airlines are bankrupt.

The number of carriers serving a route increases by 8 to 10% in re-emergence case when some carriers are

bankrupt. In liquidation case, the number is larger: 14 to 15%. The increase becomes much less after

bankruptcy protection is �nished either by re-emergence or by liquidation. This result indicates that airlines

enter the bankrupt routes, induced by weakened bankrupt rivals or reduced capacity by them. We will not be

able to �nd this result if bankrupt airlines hurt pro�tability of serving the route with them. It seems that at

least some airlines are �nding bankrupt routes attrative.

The second and third estimation results on capacity show little change in the total capacity level in

the routes a¤ected by bankruptcy. In re-emergence case, the total capacity is reduced in the later stage

of bankruptcy protection. In liquidation case, the total capacity is rather reduced upon bankruptcy �ling

but does not change after (if any, it rather increases after liquidation). Borenstein and Rose (2003) found

that quarterly capacity adjustment is larger than the capacity change under bankruptcy. We also �nd that

19

quarterly capacity adjustment or aggregate demand shock such as September 11 (2002Q3) has more impact

on route capacity level. Figure 5 shows the estimated coe¢ cients on time-speci�c dummy variables. We can

see a clear quarterly pattern and huge capacity cut in the aftermath of September 11. Besides, the estimates

for Florida-route quarter dummies show about 5 to 6% quarterly change in number of seats and about 3%

quarterly change in scheduled departures. So, route capacity change due to bankruptcy does not seem to be

large compared to other factors.

Table 9. Estimation Results: Competition and Capacity Change in Route-Level

Dependent Var. ln(N_carriers) ln(N_seats_all) ln(N_dprts_all)

Bankrupt_route Bankrupt_route Bankrupt_route

Variable Re-emergence Liquidation Re-emergence Liquidation Re-emergence Liquidation

[TB-2] -.0023 -.0357*** .0055 -.0016 -.0108** -.0081

(.0058) (.0090) (.0046) (.0073) (.0052) (.0083)

[TB-1] -.0025 -.0255*** .0105** -.0144** -.0015 -.0115

(.0051) (.0090) (.0043) (.0069) (.0049) (.0079)

[TB ] .0761*** .1383*** -.0024 -.0143* .0098* -.0161*

(.0063) (.0096) (.0051) (.0081) (.0056) (.0090)

[TB+1] .0911*** .1265*** -.0188*** .0016 1.42e-06 -.0050

(.0068) (.0109) (.0056) (.0094) (.0058) (.0102)

[TB+2~T] .1004*** .1540*** -.0277*** .0053 -.0038 -.0047

(.0047) (.0073) (.0038) (.0068) (.0039) (.0073)

[T] -.0025 .0232** -.0131** .0128 -.0037 .0056

(.0073) (.0104) (.0057) (.0096) (.0061) (.0109)

[T+1] .0033 -.0115** .0041

(.0072) (.0057) (.0058)

[T+2~] .0972*** .0184*** .0324***

(.0058) (.0046) (.0050)

LCCin .0591*** .0595***

(.0049) (.0050)

SWin .1273*** .0832***

(.0075) (.0070)

HHI -.3504*** -.5569***

(.0146) (.0151)

Constant 1.29*** 11.80*** 6.98***

(.0074) (.0127) (.0130)

N 42,000 41,993 41,993



In sum, bankrupt airlines do not seem to add capacity that should have been liquidated to boost other

airlines� pro�tability. Bankrupt airlines reduce capacity, non-bankrupt rivals expand, some airlines enter

the bankrupt routes, and the total capacity does not change much. Even without bankrupt �rms�capacity,

remaining airlines would have �lled the capacity gap from liquidation and the total capacity on bankrupt routes

20

Figure 5: % Route Capacity Change, 1998 through 2008 (base: 1998Q1)

would not have been reduced signi�cantly as a result of bankrupt airlines�liquidation. That is, overcapacity

problem does not seem to get worse or better as a result of bankruptcy protection.

However, the composition of capacity has been changed because bankrupt airlines reduce capacity and

other airlines �ll the gap. This leads to our next question: what kind of non-bankrupt �rms are replacing

the bankrupt airlines� capacity? This question is important because, even with the same capacity level, if

bankrupt airlines�capacity is replaced by more e¢ cient airlines with lower cost, then the industry is improving

in the sense that fhs average cost of the industry is lowered and e¢ ciency increases.

3.3 Who Replaces Bankrupt Airline�s Capacity?

We have seen the replacement of bankrupt airlines� capacity by non-bankrupt rivals in re-emergence case.

Also, new entries seem more PROMINENT? in liquidation case. It is noteworthy that bankrupt airlines that

re-emerged from bankruptcy are legacy carriers with larger networks than those ending up with liquidation.

The bankruptcy e¤ects in re-emergence case are mostly driven by legacy carriers that have gone bankrupt

during the data period, such as United, US Airways (two times), Northwest, and Delta. These legacy airlines

are characterized by higher cost structures inherited from their old history in the industry.27 The legacy

airlines also have large networks so their capacity cut would allow other carriers much room to expand.

Given that the reduced capacity by the bankrupt legacy airlines is �lled by non-bankrupt rivals, interesting

question would be who replaces the bankrupt airlines�capacity. Figure 6 shows the signi�cant di¤erence in

unit cost (per available seat mile: ASM) between legacy and low cost carriers. If low cost airlines replace

bankrupt airlines�capacity, increasing their presence, this will boost average e¢ ciency on bankrupt routes and

in the industry as a whole.

27Legacy carriers have higher cost structure due to higher labor costs for more senior workforce and retirees, higher asset-relatedcosts for various types of �eets including less fuel-e¢ cient aircrafts, and higher maintenance costs for larger networks.

21

Figure 6: Unit Cost Di¤erential, 1998 through 2003 (Source: GAO-04-836, Figure 13)

The answer to the question is unclear in theory. When some large network carriers are bankrupt and

reduce supply, other large network carriers may become more appealing to the travelers who used to choose

the bankrupt carrier due to similar product characteristics. In that case, the replacement of capacity will be

mostly done by similar large network carriers rather than low cost carriers with smaller network and no frills.

However, similar network carriers may experience similar negative shocks that forced the other airlines to �le

for bankruptcy, which makes their exansion less likely. Or travelers simply may not care about network size or

other qualities than price. Then low cost airlines will be able to take more share of the residual demand that

all non-bankrupt rivals are facing through lower price. Table 10 gives us a hint of what the empirical answer

would be.

Table 10 compares the two estimation results similar to the previous regression model of ln(Med_fare):

one is for the regression on bankruptcy-related variables only and the other is for the reggression on bankruptcy-

related variables plus route characteristics related to low cost airline presence on a route. We can see that

including LCCin (indicator of low cost carrier presence on a route) and SWin (indicator of Southwest presence

on a route) reduces the size of estimates for coe¢ cients on bankruptcy-related variables. For example, non-

bankrupt competitors�fare change when an airline �led for bankruptcy at the current quarter (i.e. coe¢ cient

on NonB[TB]) changes from -0.0302 to -0.0244 (3.02% to 2.44% fare cut) if we control for low cost carrier�s

presences, meaning the magnitude of fare cut is estimated to be smaller. This result indicates that more

e¢ cient airlines with lower cost structure �lls the gap from bankrupt �rms�capacity reduction and also push

others to cut price down.

22

Table 10. Estimation Results: Low Cost Carrier E¤ect?

Dependent Var. ln(Med_fare) ln(Med_fare)

Re-emergence Liquidation Re-emergence Liquidation


[TB-2] -.0485*** -.0038 -.0197 -.0045 -.0460*** -.0003 -.0123 .0015

(.0046) (.0030) (.0083) (.0046) (.0045) (.0030) (.0083) (.0046)

[TB-1] -.0678*** -.0122*** -.0059 .0121 -.0601*** -.0073*** .0009 .0188***

(.0043) (.0026) (.0084) (.0043) (.0042) (.0026) (.0084) (.0043)

[TB ] -.1108*** -.0302*** -.0347 .0048 -.0995*** -.0244*** -.0292*** .0093*

(.0047) (.0032) (.0104) (.0054) (.0047) (.0031) (.0104) (.0054)

[TB+1] -.0894*** -.0267*** -.0408 -.0113 -.0791*** -.0206*** -.0363*** -.0072

(.0051) (.0036) (.0105) (.0064) (.0050) (.0036) (.0107) (.0065)

[TB+2~T] -.0744*** -.0147*** -.1226 -.0323 -.0604*** -.0103*** -.1180*** -.0279***

(.0034) (.0024) (.0095) (.0040) (.0033) (.0024) (.0096) (.0040)

[T] -.0567*** -.0094** -.0238 -.0424*** -.0032 -.0219***

(.0055) (.0038) (.0057) (.0054) (.0038) (.0057)

[T+1] -.0523*** -.0138*** -.0367*** -.0061

(.0056) (.0039) (.0056) (.0039)

[T+2~] .0036 -.0337*** .0207*** -.0302***

(.0043) (.0031) (.0042) (.0031)

LCCin -.0794***

(.0029)

SWin -.1011***

(.0036)

HHI .1202***

(.0080)

Constant 4.9619*** 4.9764***

(.0047) (.0066)

N 178,443 178,443



In order to see how non-bankrupt low cost airlines respond to rivals�bankruptcy, we conduct more regres-

sions on fare and capacity, respectively. Table 11 reports regression results of the same econometric speci�ca-

tion from Section 3.1 except that we add one more category to non-bankrupt competitors (NonB_lcc). The

estimated coe¢ cients on the variables under this new catergory (NonB_lcc column) are interpreted as the

incremental e¤ect of LCC among non-bankrupt competitors. So, the sum of estimates from NonB column

and NonB_lcc column is the total e¤ect of being a non-bankrupt competitor. As before, Panel 1 is about

pricing and Panel 2 is about quantity setting. We compare these results to the those in Table 8.

23

Table 11. Estimation Results: Non-bankrupt LCC Responses to Rivals�Bankruptcy

Panel 1: Pricing

Dependent Var. ln(Med_fare)


Variable Bankrupt NonB NonB_lcc Bankrupt NonB NonB_lcc

[TB-2] -.0350*** -.0058* .0180*** .0206** -.0087* .0493***

(.0043) (.0033) (.0043) (.0080) (.0049) (.0090)

[TB-1] -.0487*** -.0143*** .0203*** -.0059 .0087** .0370***

(.0041) (.0029) (.0041) (.0082) (.0044) (.0080)

[TB ] -.0971*** -.0501*** .1007*** -.0044 -.0012 .0241***

(.0045) (.0034) (.0044) (.0096) (.0057) (.0101)

[TB+1] -.0874*** -.0456*** .0916*** -.0150 -.0388*** .1256***

(.0050) (.0039) (.0048) (.0104) (.0065) (.0116)

[TB+2~T] -.0681*** -.0277*** .0548*** -.0782*** -.0573*** .1173***

(.0033) (.0025) (.0030) (.0091) (.0041) (.0074)

[T] -.0280*** -.0120 .0351*** -.0500*** .1392***

(.0054) (.0042) (.0052) (.0061) (.0099)

[T+1] -.0303*** -.0177*** .0349***

(.0055) (.0044) (.0055)

[T+2~] -.0025 -.0531*** .0575***

(.0042) (.0035) (.0042)

LCCin -.0788***

(.0028)

SWin -.0987***

(.0035)

HHI .0891***

(.0079)

Net_origin .3235

(.2239)

Net_dest .1617

(.2263)

Network .0344**

(.0152)

%direct -.0355***

(.0050)

%round -.5610***

(.0099)

%codeshr .1201***

(.0051)

Constant 5.38***

(.0126)

N 178,443

NonB_lcc (=LCC�NonB) Non-bankrupt LCC dummyRobust SE reported in parentheses. Time speci�c dummies included. N: Sample size


24

Panel 2: Quantity Setting

Dependent Var. ln(N_seats)



[TB-2] -.0808*** .0261* -.0248 -.0297 -.0270 -.0636

(.0257) (.0143) (.0201) (.0284) (.0200) (.0405)

[TB-1] -.1449*** .0313** .0420** -.0975* -.0376* -.0855**

(.0263) (.0148) (.0198) (.0545) (.0212) (.0349)

[TB ] -.1063*** .0149 .1849*** -.0267 -.1142*** .0498

(.0269) (.0171) (.0238) (.0277) (.0274) (.0541)

[TB+1] -.1895*** .0326* .1243*** -.0720*** -.1649*** -.0024

(.0287) (.0184) (.0227) (.0262) (.0270) (.0529)

[TB+2~T] -.3147*** .0404*** .1156*** -.1869*** -.0909*** -.0081

(.0153) (.0116) (.0136) (.0496) (.0204) (.0383)

[T] -.3893*** -.0444** .2777*** .0192 -.0117

(.0326) (.0191) (.0233) (.0313) (.0552)

[T+1] -.4435*** -.0649*** .2532***

(.0341) (.0210) (.0256)

[T+2~] -.5327*** -.1111*** .2929***

(.0227) (.0157) (.0179)

LCCin -.0400***

(.0121)

SWin .0902***

(.0196)

HHI -.4261***

(.0421)

Constant 10.85***

(.0290)

N 81,564



In Panel 1 of Table 11, the average fare cut by non-bankrupt competitors in each period surrounding

bankruptcy almost doubles the previous estimates. While a rival �rm is under bankruptcy (see [TB], [TB+1],

and [TB+2] rows), for instance, non-bankrupt airlines�fare cut was estimated to be around 2.8% at �rst and

1.4% later during bankruptcy in Table 8, and they are now estimated to be about 5% at �rst and 2.8% later

during the same period. More importantly, although average non-bankrupt rivalss maintain lower fare than

normal, low cost airlines maintain higher fare than normal. When a rival �les for bankruptcy (see [TB] row),

non-bankrupt low cost carrier raises fare by about 5% (=-0.0501+0.1007). So, aggressive pricing by bankrupt

airlines seems to a¤ect only non-LCC among non-bankrupt airlines.

The estimation result in Panel 2 is interesting since it shows that capacity expansion by non-bankrupt

competitors are mostly done by low cost airlines in re-emergence case. Bankrupt airlines cut their capacity

25

even before actual �ling and the size of cut is about 15% right before the �ling (see [TB � 1] row). Averagenon-bankrupt ailrines increase their capacity then by 3% and low cost airlines increase capacity more by 4%

(so about 7% higher than normal fare, in total). The capacity cut by bankrupt airlines increases during

bankruptcy (see [TB], [TB+1], and [TB+2] rows). Meanwhile, average non-bankrupt airlines expand a little, 4%

at most, but low cost airlines expand their capacity signi�cantly about 12 to 18%. In liquidation case, low

cost airlines do not seem to increase their capacity, which could be due to that liquidated airlines are rather

small and often are low cost airlines themselves so their impact on other carriers are not signi�cant.

Although low cost airlines increases its fare, it would be still lower than other carriers�fares. Thus, the

capacity expansion by low cost airlines to �ll the reduced capacity by bankrupt airlines could pose a signi�cant

price competitive pressure on other non-bankrupt airlines. These results imply that non-bankrupt competitors

may look like they are hurt by bankrupt airlines�aggresstive pricing since they tend to lower fares on bankrupt

routes, but that low cost airlines take an opportunity to expand capacity and increase their presence may have

more impact on price competitive pressure rather than bankrupt �rms�low price. That is, bankrupt carrier

may have triggered fare cut in the beginning, it could be their capacity cut that increases price competition

by allowing low cost airlines to expand. Thus bankruptcy protection per se do not seem to harm average

non-bankrupt airlines. Moreover, e¢ cient airlines with low cost structure are bene�ted by bankrupt airlines�

capacity cut. That is, the industry transtiton in favor of more e¢ cient players may have been facilitated by

bankruptcy �lings and capacity cut that followed.

In Table 12, the analysis on market share shows market share change in the periods surrounding bankruptcy.

The loss in market share of bankrupt airlines is signi�cant even before the actual bankruptcy �ling. The loss

is larger during bankruptcy. Then who are the bankrupt airlines losing their market share to? Average non-

bankrupt airlines seems to lose market shares during the same period. However, if we look at low cost airline,

they actually are increasing their presence in the bankrupt market. In re-emergence case, when an airline �les

for bankruptcy, average non-bankrupt rivals experience loss in market share by about 4.7% but, among them,

low cost rivals�market share increases by 4% (=-.0469+.0879). This pattern continues during bankruptcy. We

can see similar pattern in liquidation case too. So low cost airlines take an opportunity to expand and increase

their market share at the expense of bankrupt airlines or other non-bankrupt airlines, even with higher fare

level than usual. Although, the low cost airlines�market share increase becomes insigni�cant after bankrupt

airlines re-emerged. But when bankrupt airlines are actually liquidated, the size of market share increase is

even larger after liquidation.

26

Table 12. Estimation Results: Market Share Change

Dependent Var. ln(MktShare)



[TB-2] -.0433*** -.0095*** .0173*** -.0207** -.0136*** .0463***

(.0046) (.0035) (.0046) (.0083) (.0052) (.0095)

[TB-1] -.0610*** -.0163*** .0226*** -.0063 .0076 .0246***

(.0043) (.0031) (.0042) (.0084) (.0047) (.0085)

[TB ] -.1063*** -.0469*** .0879*** -.0362*** .0061 .0007

(.0047) (.0034) (.0044) (.0104) (.0060) (.0106)

[TB+1] -.0872*** -.0397*** .0720*** -.0403*** -.0250*** .0941***

(.0050) (.0038) (.0048) (.0106) (.0069) (.0121)

[TB+2~T] -.0733*** -.0277*** .0522*** -.1255*** -.0547*** .1014***

(.0033) (.0026) (.0030) (.0095) (.0045) (.0079)

[T] -.0397*** .0056 -.0021 -.0491*** .1335***

(.0055) (.0043) (.0052) (.0064) (.0103)

[T+1] -.0333*** .0017 .0014

(.0056) (.0044) (.0053)

[T+2~] .0161*** -.0282*** .0222***

(.0041) (.0052) (.0037)

Constant 4.96

(.0047)

N 178,443



In addition to carrier-level analyses, we want to check whether low cost airlines enter bankrupt routes. We

conduct the conditional �xed e¤ect logit regression of indicator of low cost carrier presence (or the presence

of Southwest in particular) on bankruptcy-related variables. Table 13 shows the estimation results. Note that

the number of observations available are substantially reduced in the �xed e¤ect logit model since individulas,

routes in this case, that have not experienced change in low cost carrier presence (or Southwest presence) are

dropped in the estimation as they do not contribute to the likelihood. For example, if some low cost carriers

are present in the route, entry by another low cost carrier is not accounted for. Since all bankruptcy-related

variables are binary, we interpret the estimated coe¢ cient on those variables in terms of the log of the odds

ratio involved. That is, the log odds ratio increase by the estimated coe¢ cient when an indicator is triggered

on. So positive sign of estimates mean the likelihood of low cost airline entry or Southwest entry is more likely.

From Table 13, we can see that bankrupt routes that used to be served by non-low-cost carriers are more

likely to experience low cost carrier entry, compared to usual time and non-bankrupt routes. Also, Southwest

is more likely to enter a route when some carriers are bankrupt on the route in re-emergence case. Unlike

average low cost carriers, Southwest does not enter the route where bankrupt airlines are about to be liquidated.

Within two quarters after bankrupt airlines re-emerged or went out of business, the likelihood of the entry is

27

not di¤erent from or lower than normal routes or times. So the entry by low cost carriers including Southwest

seems to be active while some airlines are bankrupt but serving the route.

Table 13. Estimation Results: Low Cost Carrier Entry?

Dependent Var. LCCin SWin

Bankrupt_route Bankrupt_route

Variable Re-emergence Liquidation Re-emergence Liquidation

[TB-2] -.2463* .0138 -.4290 -.4321

(.1306) (.1835) (.3126) (.3886)

[TB-1] -.0221*** .2024 .4928** -.2204

(.1271) (.1821) (.2449) (.4108)

[TB ] .6008*** .5406** 1.0322*** -.5093

(.1573) (.2419) (.3048) (.6279)

[TB+1] .4207** .4335 1.1085*** -1.3989

(.1675) (.2933) (.3438) (.9738)

[TB+2~T] .5865*** .5182** .2812 -3.6155***

(.1121) (.2083) (.2211) (1.1788)

[T] -.3022** .1646 .2071 -1.1406

(.1518) (.2580) (.2970) (1.0354)

[T+1] -.1288 .1953

(.1533) (.3148)

[T+2~] .2163* .1029

(.1232) (.2378)

Log Likelihood -5124.2009 -1107.3817

N_iteration 3 4

N 14,924 5,723

SE reported in parentheses. Time speci�c dummies included. N: Sample size


3.4 How Does a Bankrupt Airline Adjust Route Structure?

This section discusses how bankrupt airlines transform themserlves and survive in the aspect of network

formation. Bankrupt airlines with high cost but larger network may abandon the routes where low cost

airlines can compete with price e¤ectively and move in to the routes where their larger network can have

comparative advantage, e.g. hub/focus airports or international routes. Table 14 is the regression result on

change in route structure. Since the bankrupt airlines ending up with liquidation have rather smaller network,

they do not have particular hub airports. So the regression result shows the route structure change when large

network carriers go bankrupt.

The dependent variables is the network size out of origin divided by the total network size that a carrier

is serving among the 1000 largest routes. The value of this variable of a route gets higher when network size

28

shrinks or services are concentrated on the route. We look at whether bankruptcy �ling airlines concentrate

their services in the periods surrounding bankruptcy and the concentration is more signi�cant in their hub

or focus airports. The result is consistent with the expectation that bankrupt airlines focus their services by

reducing the number of routes they serve (i.e. network size). Also, this pattern is more signi�cant in hub/focus

airports. So bankrupt network carriers seem to focus their services on hub or focus airports where they have

comparative advantages over low cost carriers.so can compete more e¤ectively.

Table 14. Estimation Result: Bankrupt Airline Network Formation

Dependent Var. ln(Net_focus)

Re-emergence Liquidataion

Variable Bankrupt Hub_B NonB Bankrupt NonB

[TB-2] .0649*** -.0452** -.0771*** -.0613 .0088

(.0234) (.0214) (.0172) (.0969) (.0212)

[TB-1] .1538*** .0332 -.0008 -.1388 -.0043

(.0207) (.0214) (.0147) (.1257) (.0240)

[TB ] .1593*** .1761*** .0037 .0045 .0018

(.0234) (.0230) (.0182) (.0630) (.0215)

[TB+1] .1402*** .2116*** -.0644*** .3573** -.0083

(.0264) (.0253) (.0207) (.1614) (.0336)

[TB+2~T] .2244*** .3089*** -.0433*** .4841*** .0059

(.0184) (.0157) (.0156) (.1036) (.0206)

[T] .4667*** .3840*** .0617*** -.0057

(.0277) (.0297) (.0210) (.0278)

[T+1] .4528*** .2839*** .0043

(.0408) (.0349) (.0293)

[T+2~] .4212*** .0455* .0544*

(.0290) (.0246) (.0271)

Constant 4.59***

(.0218)

N 178,443

Hub B (=Hub_origin�Bankrupt): Bankrupt hub carrier dummyRobust SE reported in parentheses. Time speci�c dummies included. N: Sample size


3.5 Does Average Unit Cost Decrease?

[Incomplete]

The question we want to answer in this section is whether the average unit cost goes down in the process

of bankruptcy. There are two channels by which the average unit cost is reduced in bankrupt routes. First,

bankrupt airlines may be able to achieve cost cut by, for instance, renegotiating with suppliers and workers.

29

This would be a direct e¤ect of bankruptcy on productive e¢ ciency. There is another e¤ect of bankruptcy

that alters the production composition of markets. Even if bankrupt airlines cannot achieve cost cut and

other airlines maintain the same cost level, the share of market supplied by each airline changes as a result of

capacity cut spurred by bankruptcy. In the previous sections, we saw that more cost-e¢ cient airlines catch

the opportunity to expand while bankrupt airlines cut their capacity. This will change the average unit cost

in bankrupt market (where some �rms are bankrupt). In a welfare perspective, this is good in the sense that

it boosts average e¢ ciency in a market.

The section discusses whether e¢ ciency deteriorates or improves as a result of bankruptcy and, if it

improves, how much of the e¢ ciency increase is explained by the change in production composition. First

of all, we want to see whether the average route-level unit cost decreases during bankruptcy as a proxy for

e¢ ciency change. The unit cost, i.e. cost per AMS (Available Seat Mile) is the standard cost measure in the

ailrine industry. So, we will look at how route average unit costs are a¤ected by bankruptcy. In particular,

system total expense (excluding fuel) per ASM (i.e. CASM ex fuel, unit: cents per ASM) and system labor

expense (excluding management and other) per ASM (i.e. LASM ex mngmnt/other, unit: cents per ASM)

will be analyzed. Fuel expenses are exclued form CASM as it is more subject to external shock than carrier

strategy and management related expenses (e.g. CEO pay) are excluded from LASM as it does not seem much

related to the marginal productivity of a �rm.

For the construction of route average unit cost, we currently use the data from the Airline Data Project

(ADP) established by the MIT Global Airline Industry Program.28 The data source has various �nancial

and operational data in both industry level and carrier level, constructed from the DOT data set. Carrier-

level data is available only for several major carriers; American, Continental, Delta, Northwest, United, US

Airways, America West, Southwest, JetBlue, AirTran, Frontier, ATA, Alaska, Hawaiian, Midwest. The data

set contains yearly observation of the unit costs for these carriers from 1995 to 2007. From the data set, we

construct route average unit cost by taking capacity-share-weighted average of the cost of carriers serving the

route.29

Table 15 shows the unit cost change on bankruptcy-a¤ected routes. First of all, in the re-emergence case,

CASM ex fuel seems to be higher than usual prior to and at the early stage of bankruptcy �lings. The unit cost,

however, seems to get lower as time under bankruptcy protection passes by. In liquidation case, on the contrary,

the unit cost is rather lower prior to bankrutcy �ling and gets higher during bankruptcy protection. However,

all the estimates are of little magnitude and so the route average cost do not seem change signi�cantly. LASM

ex management and other, on the other hand, seem to show signi�cant bankruptcy impact. The labor unit cost

is usually high in pre-bankruptcy period (by 1.3% or 3.3% higher than usual) and then gets lower signi�cantly

during bankruptcy protection periods (-2.9% upon bankruptcy �ling, -4.3% in next quarter, and -8% later) in

emergence case. The reverse is true in liquidation case, though the estimates are almost insigni�cant.

The results may be interpreted as a temporary high cost shock worsened the situation for the �nancially

distressed airlines and the bankruptcy �ling �rms emerge from bankruptcy when things get better with favor-

able cost shock. However, since we controlled for industry-wide common cost shock by including time-speci�c

28http://web.mit.edu/airlinedata/www/default.html29Capacity share of a carrier is calculated from the segment data used in the previous sections.

30

dummy variables, the results do not seem to be driven totally by external cost shocks. After accouting for

common cost shocks, signi�cant estimates suggest that bankrupt routes experience incremental e¤ects. For

example, the following story is possible. In emergence case, bankrupt �rms may have su¤ered more from

negative cost shock, becoming more cost-conscious, so start to reduce capacity to reduce total expenses. The

reduced capacity is �lled by non-bankrupt airlines, especially LCCs, driving the route average (labor) unit cost

down. Of course, bankrupt airlines themselves may be cutting total expenses .e.g by terminating pension plans

or so, but the unit costs would not decrease dramatically since they reduce capacity as well. The regression

result is inconclusive but show the possibility that more e¢ cient airlines replace bankrupt ailrines�ine¢ cient

capacity.

Table 15. Estimation Results: Route Average Unit Cost

Dependent Var. ln(CASM ex fuel) ln(LASM ex mngmnt=other)

Bankrupt_route Bankrupt_route

Variable Re-emergence Liquidation Re-emergence Liquidation

[TB-2] .0187*** -.0145*** .0333*** .0293***

(.0025) (.0035) (.0034) (.0044)

[TB-1] .0135*** -.0108*** .0134*** .0282***

(.0023) (.0034) (.0035) (.0044)

[TB ] .0127*** .0006 -.0293*** .0097*

(.0029) (.0046) (.0041) (.0057)

[TB+1] .0123*** .0024 -.0432*** .0093

(.0033) (.0056) (.0042) (.0065)

[TB+2~T] -.0041** .0142*** -.0799*** .0155***

(.0020) (.0034) (.0027) (.0043)

[T] -.0006 .0136*** -.0542*** .0163**

(.0029) (.0050) (.0043) (.0065)

[T+1] -.0054* -.0629***

(.0031) (.0043)

[T+2~] .0074*** -.0037

(.0026) (.0036)

N 39,033 39,033



This section is still very incomplete. Next work that should follow would be, �rst, to take advantage of all

the information available and, second, to decompose the e¢ ciency change into two factors: bankrupt airlines�

cost cut and production composition change. ADP data depends on US DOT database. In particular, CASM

ex fuel data is calculated from US DOT Form 41 via BTS, Schedule T2 & P6 & P52 and LASM ex management

and other data is calculated from US DOT Form 41 via BTS, Schedule P6 & T2. We will construct the unit

cost data on quarterly basis for every carrier used in our analysis and for all data periods.

31

3.6 How Much Fraction of LCC Growth is Explained by Rivals�Bankruptcy?

[Incomplete]

Given the long history of the airline industry nce deregulation in 1978, the low cost carrier, even with

a signi�cantly large cost e¢ ciency relative to legacy carriers, have not expanded that much (see �gure 7).

Also, most of the growth has occurred only after 1990 (low cost carriers�passenger share is less than 5% in

1990). This raises a question. What else does it take for e¢ cient airlines to take markets from less e¢ cient

incumbents? Incumbent legacy airlines can be very adverse to reducing capacity for various reasons. For one,

it may be hard to get terminals or other airport facilities back once they lose them to other airlines. Or, since

they have extra aircrafts anyway, they can add capacity at a very low marginal costs. These reasons may

be holding back the incumbent airlines from reducing capacity in normal times when they do not need any

dramatic change immediately.

Figure 7: LCCs�Share of Domestic O&D Passengers in the U.S. Since 1990 (Source: Darin Lee�s webpage:http://www.darinlee.net/data/lccshare.html)

This section moves on for a bigger picture. If bankruptcy is one factor that spurs low cost airlines�expansion

as we have seen so far, then what else would be there? How much fraction of the low cost airlines�growth

is explained by rivals�bankruptcy relative to other factors? Some candidates are mergers, external demand

shocks such as September 11, 2001, or supply shocks such as soaring fuel costs. We will investigate such factors

that a¤ect low cost expansion discretely and compare the size of e¤ects to bankruptcy e¤ects.

32

4 Robustness Checks

This section checks whether the results are robust over di¤erent samples and econometric speci�cations. We

will begin with balanced sub-panel data with route seletivity under bankruptcy issue. We then discuss the

inclusion of origin/destination linear time trend variable that Ciliberto and Schenone (2008) suggested. Finally,

upper and lower quartile fare will be analyzed.

4.1 Balanced Sub-Panel

This section uses only carrier-route pairs that exist for all quarters throughout the data periods. So, the

carriers have not entered into a new route or exited from currnet routes in this subsample. The estimates then

measure only the change in incumbents�fare or capacity. The original sample is unbalaned panel data as a

carrier that is observed on a route at some point may disappear at another time or a new carrier shows up on

a route.

The route selection in normal times (una¤ected by bankruptcy) can be captured by individual �xed e¤ects

so the estimates would be consistent. However, the route selection may occur more actively or di¤erently

in the periods surrounnding bankruptcy either because bankrupt carriers have to change their strategies

discretely or because demand changes more signi�cantly (enough to push some airlines to �le for bankruptcy)

in those periods. As airlines enter into pro�table routes where they have comparative advantage and exit from

unpro�table routes that they cannot set high price, the non-random route selection can bias the estimated

bankruptcy e¤ect on fare and capacity to zero. The selectivity bias, if present, may lead to smaller estimates

for bankruptcy e¤ects than the actual e¤ects. The balanced sub-panel have only carriers that are present on

a route throughout the data periods so the potential selectivity bias may be reduced by using this subsample,

though the balanced sub-panel have drawbacks that we do not use all the information available (the sample

size shirinks signi�cantly, from 178,443 to 93,197 (81,564 to 53,699) for the fare (capacity) regression.)

Table 16 repeats the same econometric speci�cations in Table 11 with the balanced sub-panel. Note that

liquidated airlines are dropped form the sample since it banishes at some point in data periods (except some

airlines went bankrupt right after the data periods (2008Q2) so the pre-bankruptcy observations exist for these

cases). The estimation results show that the direction of bankrupt e¤ects do not change overall but the size of

the e¤ects on fare is larger for bankrupt �rms and average non-bankrupt competitros, with balanced sub-panel.

Bankrupt ailrines, for example, are estimated to cut fare 3.5% and 4.9% prior to the �ling when the unbalanced

panel is used. The numbers are 4% then 6.3% with the balanced sub-panel. During bankruptcy, the estimated

fare cuts by bankrupt ailrines are 9.7%, 8.7%, then 6.8% with the unbalanced panel and 10%, 10.8%, then

9% with the balanced sub-panel. In the same period, the fare cuts by non-bankrupt rivals are also bigger by

about 1.5% with the balanced sub-panel. The estimates on bankruptcy e¤ect on capacity remain essentially

unaltered with slight increase for average non-bankrupt airlines�capacity expansion during bankruptcy. This

may suggest a potential selectivity bias. Without building up a structural model of entry and exit decision, it

is hard to �x the potential selectivity bias. We will work on improving the estimates in this aspect also.

33

Table 16. Robustness Check: Balanced Sub-Panel

Panel 1: Non bankrupt LCC response to bankruptcy in price

Dependent Var. ln(Med_fare)



[TB-2] -.0404*** -.0045 .0307*** -.0067 -.0022 .0173

(.0057) (.0041) (.0069) (.0808) (.0063) (.0130)

[TB-1] -.0628*** -.0208*** .0335*** .0664 -.0002 .0767***

(.0055) (.0037) (.0064) (.0516) (.0056) (.0137)

[TB ] -.1001*** -.0627*** .1015*** -.0034 -.0312*

(.0062) (.0043) (.0068) (.0074) (.0160)

[TB+1] -.1081*** -.0570*** .0909*** -.0532 .1406***

(.0063) (.0049) (.0070) (.0078) (.0166)

[TB+2~T] -.0900*** -.0444*** .0564*** -.0618*** .0964***

(.0041) (.0032) (.0042) (.0049) (.0131)

[T] -.0330*** -.0341*** .0586*** -.0429*** .0951***

(.0068) (.0053) (.0085) (.0071) (.0128)

[T+1] -.0323*** -.0415*** .0530***

(.0068) (.0056) (.0093)

[T+2~] -.0139*** -.0741*** .0811***

(.0052) (.0043) (.0059)

LCCin -.0885***

(.0034)

SWin -.1179***

(.0047)

HHI .0502***

(.0101)

Net_origin .9040***

(.2845)

Net_dest .6703**

(.2814)

Network -.0357*

(.0187)

%direct -.0324***

(.0065)

%round -.4522***

(.0145)

%codeshr .1047***

(.0075)

Constant 5.36***

(.0171)

N 93,197



34

Panel 2: Non bankrupt LCC response to bankruptcy in quantity

Dependent Var. ln(N_seats)



[TB-2] -.0642** .0196 .0467** -.4052*** -.0159 -.1367*

(.0297) (.0156) (.0234) (.1330) (.0215) (.0822)

[TB-1] -.1090*** .0184 .0753*** -.1991 -.0293 -.0891

(.0287) (.0162) (.0255) (.1291) (.0216) (.0567)

[TB ] -.1095*** -.0020 .1519*** -.0957*** .0127

(.0317) (.0189) (.0349) (.0275) (.0860)

[TB+1] -.2024*** -.0072 .1568*** -.1658*** .1128***

(.0334) (.0206) (.0310) (.0289) (.0554)

[TB+2~T] -.2963*** .0150 .1945*** -.1130*** -.0443

(.0166) (.0128) (.0163) (.0202) (.0785)

[T] -.3369*** -.0177 .3099*** -.0257 .0727

(.0371) (.0209) (.0317) (.0329) (.0664)

[T+1] -.4131*** -.0583** .3167***

(.0393) (.0236) (.0360)

[T+2~] -.5011*** -.1128*** .3050***

(.0248) (.0174) (.0211)

LCCin -.0122

(.0130)

SWin .1143***

(.0240)

HHI -.1505***

(.0468)

Constant 10.90***

(.0319)

N 53,699



4.2 Origin/Destination Linear Time Trend?

Second, Ciliberto and Schenone (2008) includes origin and destination linear time trend in their regressions.

They argue that, since airlines serving the routes with diminishing demands may be more likely to �le for

bankruptcy, the downward demand trend can complicate the estimated fare change upon bankruptcy to be

biased in negative.direction. They suggest this problem can be lessened by controlling for inear time trend of

origin and destination airport. If there is the positive relationship between bankruptcy and diminishing time

trend of demand, removing linear time trend will be appropriate. However, removing origin and destination

speci�c linear time trend could be problematic for the following reasons.

35

To begin with, the demand shocks or trends pushing airlines to bankruptcy are rather industry-wide

aggregate one than individual one speci�c to a route. Thus, if we want to control for diminishing demand

time trend, an industry-wide linear time trend would be a better choice. In addition, if demand is declining in

speci�c routes, then airlines will adjust their route structure in the manner they move out of declining routes

and enter into more �ourishing routes. That is, airlines will not stay in declining routes until they �le for

bankruptcy.

Moreover, an important question is whether there is the speci�c linear time trend in bankrupt routes in

the �rst place. If we look at some routes where a dominant carrier is bankrupt, it is hard to say that demand

is declining in those routes compared to other routes. If there is no speci�c demand time trend before any of

the ailrines serving the route �les for bankruptcy and we include a linear time trend variable to control for

the non-existing "trend", then the estimated "trend" will be picking up all the bankruptcy-related e¤ects and

we will have biased estimates for bankruptcy e¤ects. For example, if fare cut starts prior to bankruptcy �ling

and continues, then the linear time trend variable will pick up this negative e¤ect of bankruptcy on fare level

and the bankrupt e¤ect will be biased upward.

Table 17 repeats the fare and capacity regressions in Table 8. The di¤erence is, the linear time trend

variables for each origin and destination airport are added this time. Now the estimated fare cut by non-

bankrupt competitors decreases almost to zero (see [TB], [TB+1], and [TB+2] rows) and their fare is estimated

to even increase after bankrupt airlines emerged from bankruptcy (see [T ] and [T + 1] rows). On the other

hand, since the total capacity serving a route does not change that much by some carriers�bankruptcy, the

estimated origin/destination trend of capacity of each origin and destination, though not reported in the paper,

is much less signi�cant than in the fare regression results. Thus adding origin/destination linear time trends

do not change the estimation results much. Ciliberto and Schenone (2008) report that non-bankrupt rivals

do not cut their fares and even increase the fares after bankrupt �rms emerge, but only after accounting for

the linear time trends. This result could be problematic since the estimated normal pattern of non-bankrupt

airlines�pricing could have been biased, resulting in biased estimation of bankruptcy e¤ect.

4.3 Upper and Lower Quartile Price Change

In the main empirical results, we look at median fares. This section looks at upper and lower quartile fares if

there is any meaningful di¤erence over the range of fares. Table 18 reports the estimation results that repeat

the same regressions in Table 11 using lower (panel 1) and upper (panel 2) quartile fares instead of median

fare. The fare cut in the periods surrounding bankruptcy seems to be a bit larger for the upper quartile fare,

though the di¤erence does not seem big. This result may imply that airlines on bankrupt routes become to

have weaker market power that allows them to price-discriminate.

Other than these robustness check regressions, we also ran the same econometric speicifcations with a

subsample that dropped a major bankruptcy case one by one (US Airways, United, Northwest, and Delta) to

see if one of these major bankruptcy events are driving the results. The results are not reported for brevity

but the results are qualitatively the same.

36

Table 17. Robustness Check: Origin/Destionation Linear Time Trend

Dependent Var. ln(Med_fare) ln(N_seats)

Re-emergence Liquidataion Re-emergence Liquidataion


[TB-2] -.0394*** -.0042 .0212*** -.0073*** -.0847*** .0175 -.0278 -.0165

(.0043) (.0028) (.0078) (.0043) (.0256) (.0129) (.0288) (.0179)

[TB-1] -.0423*** -.0071*** -.0093 .0038 -.1601*** .0329*** -.1002** -.0318**

(.0041) (.0025) (.0081) (.0040) (.0264) (.0124) (.0573) (.0183)

[TB ] -.0740*** -.0142*** -.0050 -.0081 -.1309*** .0480*** -.0324 -.0808***

(.0045) (.0030) (.0095) (.0051) (.0275) (.0153) (.0301) (.0246)

[TB+1] -.0552*** -.0036 -.0143 -.0272*** -.2091*** .0525*** -.0833*** -.1442***

(.0049) (.0034) (.0105) (.0060) (.0294) (.0158) (.0294) (.0245)

[TB+2~T] -.0320*** .0011 -.0756*** -.0373*** -.3377*** .0505*** -.1965*** -.0728***

(.0033) (.0023) (.0092) (.0037) (.0173) (.0098) (.0560) (.0178)

[T] -.0064 .0153*** -.0275*** -.4315*** .0268 .0187

(.0051) (.0035) (.0053) (.0336) (.0165) (.0267)

[T+1] -.0044 .0116*** -.4930*** -.0060

(.0053) (.0037) (.0349) (.0178)

[T+2~] .0377*** -.0138*** -.5920*** -.0154

(.0041) (.0029) (.0252) (.0132)

LCCin -.0523*** -.0559***

(.0028) (.0126)

SWin -.0761*** -.0008

(.0038) (.0213)

HHI .0893*** -.3685***

(.0084) (.0437)

Net_origin -.2959

(.2317)

Net_dest -.3393

(.2313)

Network .0476***

(.0148)

%direct -.0333***

(.0049)

%round -.5770***

(.0093)

%codeshr .1092***

(.0050)

Constant 5.38*** 10.83***

(.0121) (.0296)

N 178,443 81,564



37

Table 18. Robustness Check: Upper/Lower Quartile Fare Change

Panel 1: E¤ect on 25% Percentile (lower quartile) Fare

Dependent Var. ln(Q1_fare)



[TB-2] -.0311*** -.0186*** .0364*** .0388*** -.0040 .0380***

(.0039) (.0030) (.0039) (.0066) (.0041) (.0072)

[TB-1] -.0421*** -.0213*** .0278*** -.0253*** -.0020 .0394***

(.0038) (.0026) (.0036) (.0078) (.0038) (.0072)

[TB ] -.1021*** -.0466*** .1011*** -.0294*** -.0115** .0398***

(.0045) (.0031) (.0040) (.0085) (.0049) (.0082)

[TB+1] -.0825*** -.0388*** .0972*** -.0221** -.0464*** .1308***

(.0046) (.0036) (.0044) (.0090) (.0057) (.0094)

[TB+2~T] -.0674*** -.0277*** .0654*** -.0545*** -.0557*** .1081***

(.0031) (.0024) (.0027) (.0069) (.0038) (.0063)

[T] -.0069 -.0078* .0228*** -.0303*** .1253***

(.0050) (.0040) (.0049) (.0051) (.0081)

[T+1] -.0108** -.0151*** .0234***

(.0051) (.0041) (.0051)

[T+2~] -.0069* -.0527*** .0423***

(.0039) (.0033) (.0039)

LCCin -.0602***

(.0024)

SWin -.1046***

(.0033)

HHI .0423***

(.0071)

Net_origin .6385***

(.2035)

Net_dest .5673***

(.2078)

Network -.0037

(.0137)

%direct -.0377***

(.0050)

%round -.4041***

(.0077)

%codeshr .1164***

(.0045)

Constant 5.01***

(.0104)

N 178,443



38

Panel 2: E¤ect on 75% Percentile (upper quartile) Fare

Dependent Var. ln(Q3_fare)



[TB-2] -.0367*** .0019 .0027 -.0166 .0033 .0618***

(.0051) (.0039) (.0051) (.0103) (.0060) (.0109)

[TB-1] -.0576*** -.0030 .0061 -.0241** .0099* .0484***

(.0048) (.0035) (.0047) (.0098) (.0054) (.0096)

[TB ] -.1022*** -.0393*** .0669*** -.0331*** .0182** .0186

(.0052) (.0041) (.0052) (.0121) (.0073) (.0125)

[TB+1] -.0841*** -.0320*** .0633*** -.0292** -.0191** .0979***

(.0056) (.0046) (.0054) (.0131) (.0082) (.0139)

[TB+2~T] -.0711*** -.0183*** .0324*** -.0489*** -.0725*** .1268***

(.0037) (.0029) (.0033) (.0114) (.0050) (.0087)

[T] -.0508*** -.0204*** .0467*** -.0543*** .1288***

(.0062) (.0049) (.0056) (.0073) (.0121)

[T+1] -.0516*** -.0188*** .0403***

(.0063) (.0052) (.0061)

[T+2~] -.0198*** -.0498*** .0664***

(.0048) (.0041) (.0047)

LCCin -.1012***

(.0034)

SWin -.0927***

(.0040)

HHI .0829***

(.0091)

Net_origin .2691

(.2617)

Net_dest -.0670

(.2661)

Network -.0719***

(.0174)

%direct -.0032

(.0058)

%round -.7586***

(.0116)

%codeshr .0873***

(.0058)

Constant 5.96***

(.0149)

N 178,443



39

5 Conclusion

The paper investigates a potential mechanism by which bankrtupcy �lings a¤ect non-bankrupt �rms and the

industry as a whole. We found little evidence supporting the argument that bankruptcy protection harms

the non-bankrupt e¢ cient �rms and the industry. Instead we found some evidence against it: bankrupt

airlines cut fares but also cut capacity, non-bankrupt rivals cut fares a little but expand capacity, and low

cost airlines mong non-bankrupt rivals do not cut fare and increase capacity. This results in the increase

in market share and entry of low cost airlines. That is, bankrupt airlines do not seem to harm their rivals

by distressed pricing and exacerbating overcapacity problem. Rather, e¢ cient airlines with lower cost are

not negatively a¤ected by bankrupt airlines�pricing and increase their presence. They replace the bankrupt

airlines�ine¢ cient capacity and possibly improve industry e¢ ciency. Meanwhile, large bankrupt airlines that

re-emerged from bankruptcy seem to transform themselves by cutting unpro�table capacity and focus on the

routes where they have comparative advantages ove low cost carriers. In sum, bankruptcy protection seems

to make distressed airlines to change their strategies in a more pro�table way and the capacity cut triggered

by bankruptcy �ling allows more e¢ cient �rms to expand. This raises an interesting question about low cost

airline growth. What does it take for �rms to expand in addition to e¢ cientcy? External demand/supply

shocks, or an immediate structural changes such as bankruptcy or merger woule be potential candidates for

factors that spur e¢ cient �rms�growth. Future works would be on how much fraction of low cost airlines�

growth is explained by bankruptcy and the capacity cut that followed, compared to other factors.

References

Borenstein, Severin. 1989. �Hubs and High Fares: Dominance and Market Power in the US Airline

Industry.�RAND Journal of Economics. 20: 344-365.

Borenstein, Severin and Joseph Farrell. 1998. "Competition and Productive E¢ ciency." Working Paper.

University of California, Berkeley.

Borenstein, Severin and Nancy L. Rose. 1995. �Bankruptcy and Pricing in U.S. Airline Markets.�American

Economic Review. 85(2): 397-402.

Borenstein, Severin and Nancy L. Rose. 2003. "Do Airline Bankruptcies Reduce Service?" NBER Working

Paper 9636.

Busse, Meghan. 2002. "Firm Financial Condition and Airline Price Wars" RAND Journal of Economics.

33(2): 298-318.

Chevalier, Judith. 1995. � Capital Structure and Product Market Competition: Empirical Evidence from

the Supermarket Industry.�American Economic Review. 85(3): 415-435.

40

Ciliberto, Federico and Carola Schenone. 2008. "Bankruptcy and Product-Market Competition: Evidence

from the Airline Industry." Working Paper. University of Virginia.

Farrell, Joseph and Carl Shapiro. 1990. �Horizontal Mergers: An Equilibrium Analysis.�American Eco-

nomic Review 80(1): 107-126.

Hofer, Christian, Martin E. Dresner b, and Robert J. Windle. "The Impact of Airline Financial Distress

on US Air Fares: A Contingency Approach." Transportation Research Part E

Kennedy, Robert E. 2000. �The E¤ect of Bankruptcy Filings on Rivals�Operating Performance: Evidence

from 51 Large Bankruptcies.�International Journal of the Economics of Business. 7(1): 5-26.

United States Government Accountability O¢ ce. 2004. "Commercial Aviation: Legacy Airlines Must

Further Reduce Costs to Restore Pro�tability." GAO-04-836.

United States Government Accountability O¢ ce. 2005. "Commercial Aviation: Bankruptcy and Pension

Problems Are Symptoms of Underlying Structural Issues." GAO-05-945.

Verbeek, Marno and Theo Nijman. 1992. "Testing for Selectivity Bias in Panel Data Models." International

Economic Review. 33(3): 681-703.

Wolfers, Justin. 2006. "Did Unilateral Divorce Laws Raise Divorce Rates? A Reconciliation and New

Results." American Economic Review. 96(5): 1802-1820.

Wooldridge Je¤rey M. 2001. "Econometric Analysis of Cross Section and Panel Data." MIT Press. Cam-

bridge, MA.

41

Appendix A: Tables

S

Table A1. Variable List: Bankruptcy-related variables

Panel 1: Re-emergence case


Pre B [TB-2] Em_Bankrupt[TB � 2]irt Em_NonB[TB � 2]irt=1 if a carrier i �les for B at t+ 2 =1 if a non-bankrupt carrier i serves route r

0 otherwise at t and other carrier on r �les for "B" at t+ 2

0 otherwise

[TB-1] Em_Bankrupt[TB � 1]irt Em_NonB[TB � 1]irt=1 if a carrier i �les for B at t+ 1 =1 if a non-bankrupt carrier i serves route r

0 otherwise at t and other carrier on r �les for B at t+ 1

0 otherwise

During B [TB ] Em_Bankrupt[TB ]irt Em_NonB[TB ]irt=1 if a carrier i �les for B at t =1 if a non-bankrupt carrier i serves route r

0 otherwise at t and other carrier on r �les for B at t

0 otherwise

[TB+1] Em_Bankrupt[TB + 1]irt Em_NonB[TB + 1]irt=1 if a carrier i �led for B at t� 1 =1 if a non-bankrupt carrier i serves route r

0 otherwise at t and other carrier on r �led for B at t� 10 otherwise

[TB+2~T]irt Em_Bankrupt[TB + 2~T ]irt Em_NonB[TB + 2~T ]irt=1 if a carrier i �led for B at t� 2 =1 if a non-bankrupt carrier i serves route r

of before and is still bankrupt at t and other carrier on r �led for B at t� 20 otherwise or before and is still bankrupt, 0 otherwise

Post B [T] Em_Bankrupt[T ]irt Em_NonB[T ]irt=1 if a carrier i emerged from B =1 if a non-bankrupt carrier i serves route r

at t� 1; 0 otherwise at t and other carrier on r emerged from B

at t� 1; 0 otherwise[T+1] Em_Bankrupt[T + 1]irt Em_NonB[T + 1]irt

=1 if a carrier i emerged from B =1 if a non-bankrupt carrier i serves route r

at t� 2; 0 otherwise at t and other carrier on r emerged from B

at t� 2; 0 otherwise[T+2~] Em_Bankrupt[T + 2~]irt Em_NonB[T + 2~]irt

=1 if a carrier i emerged from B =1 if a non-bankrupt carrier i serves route r

at t� 2 or before, 0 otherwise at t and other carrier on r emerged from B

at t� 2;or before, 0 otherwiseBankruptcy is abbreviated as B

42

Panel 2: Liquidation case


Pre B [TB-2] Li_Bankrupt[TB � 2]irt Li_NonB[TB � 2]irt=1 if a carrier i �les for B at t+ 2 =1 if a non-bankrupt carrier i serves route r

0 otherwise at t and other carrier on r �les for "B" at t+ 2

0 otherwise

[TB-1] Li_Bankrupt[TB � 1]irt Li_NonB[TB � 1]irt=1 if a carrier i �les for B at t+ 1 =1 if a non-bankrupt carrier i serves route r

0 otherwise at t and other carrier on r �les for B at t+ 1

0 otherwise

During B [TB ] Li_Bankrupt[TB ]irt Li_NonB[TB ]irt=1 if a carrier i �les for B at t =1 if a non-bankrupt carrier i serves route r

0 otherwise at t and other carrier on r �les for B at t

0 otherwise

[TB+1] Li_Bankrupt[TB + 1]irt Li_NonB[TB + 1]irt=1 if a carrier i �led for B at t� 1 =1 if a non-bankrupt carrier i serves route r

0 otherwise at t and other carrier on r �led for B at t� 10 otherwise

[TB+2~T]irt Li_Bankrupt[TB + 2~T ]irt Li_NonB[TB + 2~T ]irt=1 if a carrier i �led for B at t� 2 =1 if a non-bankrupt carrier i serves route r

of before and is still bankrupt at t and other carrier on r �led for B at t� 20 otherwise or before and is still bankrupt, 0 otherwise

Post B [T] Li_NonB[T ]irt(No Observation: Liquidated) =1 if a non-bankrupt carrier i serves route r

at t and a bankrupt carrier that served r was

liquidated at t� 1, 0 otherwiseBankruptcy is abbreviated as B

43

Table A2. Variable List: Bankruptcy-A¤ected Routes

Period Re-emergence Liquidation

Pre B [TB-2] Em_Bankrupt_route[TB � 2]rt Li_Bankrupt_route[TB � 2]rt=1 if some carriers on r �les for B at t+ 2 =1 if some carriers on r �les for B at t+ 2

0 otherwise 0 otherwise

[TB-1] Em_Bankrupt_route[TB � 1]irt Li_Bankrupt_route[TB � 1]rt=1 if some carriers on r �les for B at t+ 1 =1 if some carriers on r �les for B at t+ 1


During B [TB ] Em_Bankrupt_route[TB ]irt Li_Bankrupt_route[TB ]rt=1 if some carriers on r �les for B at t =1 if some carriers on r �les for B at t


[TB+1] Em_Bankrupt_route[TB + 1]irt Li_Bankrupt_route[TB + 1]rt=1 if some carriers on r �les for B at t� 1 =1 if some carriers on r �les for B at t� 1


[TB+2~T]irt Em_Bankrupt_route[TB + 2~T ]irt Li_Bankrupt_route[TB + 2~T ]rt=1 if some bankrupt carriers on r �led =1 if some bankrupt carriers on r �led

for B two or more quarters ago for B two or more quarters ago


Post B [T] Em_Bankrupt_route[T ]irt Li_Bankrupt_route[T ]rt=1 if some carriers on r emerged from B =1 if some carriers that were serving r

at t� 1; 0 otherwise at t� 1 were liquidated at that quarter

[T+1] Em_Bankrupt_route[T + 1]irt=1 if some carriers on r emerged from B

at t� 2; 0 otherwise

[T+2~] Em_Bankrupt_route[T + 2~]irt=1 if some carriers on r emerged from B

at t� 2 or before, 0 otherwise

Bankruptcy is abbreviated as B

44

Table A3. Summary Statistics: Carrier-Level Observations

Panel 1: All & Re-emergence Sample

Re-emergence


Variable All (a) PreB (b) DuringB (c) PostB (d) PreB (e) DuringB (f) PostB (g)

Med_fare 132.29 124.22 129.40 132.57 131.43 132.70 137.99

(Unit:2000$) (50.79) (42.05) (42.50) (43.58) (40.37) (39.85) (42.88)

Q1fare 102.54 97.17 99.70 104.58 103.78 105.58 110.32

(2000$) (35.03) (28.16) (29.36) (30.52) (28.24) (29.40) (32.62)

Q3fare 193.69 176.97 184.38 183.99 190.56 185.07 189.32

(2000$) (98.43) (75.96) (74.71) (73.06) (78.17) (69.12) (69.47)

Network 0.453 0.496 0.538 0.501 0.443 0.427 0.424

(1=1000) (0.187) (0.166) (0.146) (0.136) (0.199) (0.194) (0.193)

Net_origin 0.016 0.017 0.018 0.017 0.016 0.016 0.016

(1=1000) (0.013) (0.014) (0.014) (0.013) (0.012) (0.013) (0.012)

Net_dest 0.017 0.018 0.018 0.017 0.016 0.016 0.016

(1=1000) (0.013) (0.014) (0.014) (0.013) (0.013) (0.013) (0.012)

N_seats 64,871 68,785 64,470 61,472 56,534 56,969 54,934

(1) (49482) (52349) (47945) (48081) (41504) (42562) (41,607)

N_dprts 483 478 469 449 408 427 425

(1) (382) (341) (329) (316) (363) (306) (310)

Mktshare 0.23 0.20 0.19 0.19 0.19 0.20 0.20

(1) (0.27) (0.26) (0.25) (0.25) (0.23) (0.24) (0.24)

LCC 0.22 0.10 0.03 0.01 0.28 0.31 0.34

(0.42) (0.29) (0.17) (0.10) (0.45) (0.46) (0.47)

LCCin 0.72 0.68 0.70 0.68 0.79 0.79 0.78

(0.45) (0.47) (0.46) (0.47) (0.43) (0.41) (0.41)

SWin 0.47 0.18 0.22 0.26 0.21 0.25 0.30

(0:44) (0:39) (0:42) (0:44) (0:41) (0:43) (0:46)

HHI 0.468 0.471 0.476 0.480 0.42 0.438 0.438

(1=1000) (0.181) (0.180) (0.175) (0.176) (0.15) (0.154) (0.155)

%direct 0.51 0.42 0.46 0.46 0.44 0.47 0.48

(1) (0.42) (0.40) (0.41) (0.40) (0.40) (0.41) (0.41)

%round 0.77 0.80 0.78 0.77 0.78 0.77 0.74

(1) (0.13) (0.12) (0.13) (0.12) (0.12) (0.12) (0.13)

%codeshr 0.08 0.07 0.12 0.13 0.06 0.09 0.11

(1) (0.21) (0.19) (0.24) (0.25) (0.19) (0.23) (0.25)

N 178,443 5,718 20,865 4,890 17,356 51,116 13,639

N_sgmt 81,593 2,112 8,446 1,913 6,738 21,617 5,839

Standard Deviations are reported in parentheses. N: sample size


45

Panel 2: All & Liquidation Sample

Liquidation


Variable All (a) PreB (h) DuringB (i) PreB (j) DuringB (k) PostB (l)

Med_fare 132.29 129.80 116.60 146.58 136.16 125.23

(Unit:2000$) (50.79) (46.01) (39.43) (52.95) (49.60) (43.60)

Q1fare 102.54 104.85 93.77 113.61 105.67 97.03

(2000$) (35.03) (30.76) (27.33) (35.48) (34.42) (27.45)

Q3fare 193.69 176.87 165.57 234.06 215.67 192.79

(2000$) (98.43) (95.63) (84.17) (130.19) (120.04) (90.34)

Network 0.453 0.308 0.306 0.472 0.477 0.484

(1=1000) (0.187) (0.133) (0.073) (0.186) (0.176) (0.178)

Net_origin 0.016 0.013 0.013 0.017 0.018 0.017

(1=1000) (0.013) (0.010) (0.009) (0.012) (0.012) (0.012)

Net_dest 0.017 0.013 0.012 0.017 0.018 0.017

(1=1000) (0.013) (0.010) (0.008) (0.013) (0.012) (0.013)

N_seats 64,871 62,623 66,340 61,194 59,146 58,280

(1) (49482) (30104) (29578) (48078) (45518) (41558)

N_dprts 483 437 489 414 396 395

(1) (382) (217) (224) (320) (291) (265)

Mktshare 0.23 0.02 0.00 0.16 0.15 0.18

(1) (0.27) (0.16) (0.00) (0.20) (0.19) (0.22)

LCC 0.22 0.66 0.68 0.18 0.20 0.19

(0.42) (0.47) (0.47) (0.39) (0.40) (0.39)

LCCin 0.72 0.17 0.19 0.74 0.77 0.71

(0.45) (0.38) (0.39) (0.44) (0.42) (0.45)

SWin 0.47 0.398 0.378 0.18 0.18 0.21

(0:44) (0:124) (0:123) (0:38) (0:38) (0:41)

HHI 0.468 0.31 0.29 0.374 0.348 0.423

(1=1000) (0.181) (0.38) (0.37) (0.123) (0.113) (0.140)

%direct 0.51 0.82 0.84 0.40 0.38 0.38

(1) (0.42) (0.10) (0.11) (0.38) (0.38) (0.39)

%round 0.77 0.00 0.00 0.79 0.80 0.82

(1) (0.13) (0.03) (0.00) (0.11) (0.11) (0.11)

%codeshr 0.08 0.12 0.11 0.02 0.02 0.03

(1) (0.21) (0.18) (0.17) (0.09) (0.11) (0.12)

N 178,443 972 1,349 6,178 9,237 1,866

N_sgmt 81,593 193 275 2,213 3,084 623

Standard Deviations are reported in parentheses. N: sample size.


46

Appendix B: Econometric Speci�cation

Yirt = Em_Bankrupt0irt � �+ Em_NonB0irt � � + Li_Bankrupt0irt � + Li_NonB0irt � �+Xirt � �+ Timet � � + uirt

where

an observation unit is a carrier i (= 1; 2; � � � ; 17) on a route r (= 1; 2; � � � ; 1447) at time t (= 1998Q1,

1998Q2,� � � , 2008Q2),

Yirt is a dependent variable, ln(Med_fareirt) or ln(N_seatsirt),

ln(Med_fareirt): logarithm of median fare of a carrier i on a route r at time t;

ln(N_seatsirt): logarithm of number of seats available by a carrier i on a route r at time t;

Em_Bankruptirt is a 8 � 1 vector of bankrupt-carrier dummies of a carrier i on a route r at time t inre-emergence case; for each time period from two quarters before bankruptcy �ling to post-bankruptcy periods,

i.e. Em_Bankruptirt = (Em_Bankrupt[TB � 2]; Em_Bankrupt[TB � 1];Em_Bankrupt[TB]; Em_Bankrupt[TB + 1]; Em_Bankrupt[TB + 2~T ];

Em_Bankrupt[T ]; Em_Bankrupt[T + 1]; Em_Bankrupt[T + 2~])0; 30

� is a 8� 1 vector of coe¢ cients conformable to Em_Bankruptirt,i.e. � = (�preB2; �preB1; �B0; �B1; �B2+; �postB0; �postB1; �preB2+)0;

Em_NonBirt is a 8�1 vector of non-bankrupt competitor dummies of a carrier i on a route r at time t inemergence case; for each time period from two quarters before bankruptcy �ling to post-bankruptcy periods,

i.e. Em_NonBirt = (Em_NonB[TB � 2]; Em_NonB[TB � 1]; Em_NonB[TB]; Em_NonB[TB + 1];Em_NonB[TB + 2~T ]; Em_NonB[T ]; Em_NonB[T + 1]; Em_NonB[T + 2~])0;

� is a 8� 1 vector of coe¢ cients conformable to Em_NonBirt,i.e. � = (�preB2; �preB1; �B0; �B1; �B2+; �postB0; �postB1; �preB2+)

0;

Li_Bankruptirt is a 5 � 1 vector of bankrupt-carrier dummies of a carrier i on a route r at time t inliquidation case; for each time period from two quarters before bankruptcy �ling to liquidation,

i.e. Li_Bankruptirt = (Li_Bankrupt[TB � 2]; Li_Bankrupt[TB � 1];Li_Bankrupt[TB]; Li_Bankrupt[TB + 1]; Li_Bankrupt[TB + 2~T ])0;

is a 5� 1 vector of coe¢ cients conformable to Li_Bankruptirt,i.e. = ( preB2; preB1; B0; B1; B2+)

0;

Li_NonBirt is a 6� 1 vector of non-bankrupt competitor dummies of a carrier i on a route r at time t inliquidation case; for each time period from two quarters before bankruptcy �ling to the �rst quarter after the

rival is liquidated,

30See Table 3 or Table A1 for details.

47

i.e. Li_NonBirt = (Li_NonB[TB � 2]; Li_NonB[TB � 1]; Li_NonB[TB];Li_NonB[TB + 1]; Li_NonB[TB + 2~T ]; Li_NonB[T ])0;

� is a 6� 1 vector of coe¢ cients conformable to Li_NonBirt,i.e. � = (�preB2; �preB1; �B0; �B1; �B2+; �postB0);

Xirt is a set of constant and control variables such as LCC in, SW in, HHI, Net_origin, Net_dest,

Network, %direct, %round, and %codeshr if a dependent variable is ln(Med_fare) and LCC in, SW in,

and HHI if a dependent variable is ln(N_seats),31

Timet is a set of time-speci�c dummies for each year, quarter pair and quarter dummies for Florida route,

and

uirt is the combination of time-invariant route-carrier �xed e¤ect (�ir) and random shock to a carrier�s fare

on a route at speci�c time (�irt), i.e. uirt = �ir + �irt.

31See Table 5 for the description of variables. Some control variables are closely related to fare level but not to quantity level.So, those variables are dropped in quantity equation.

48

does bankruptcy protection harm the airline...

Documents