Common Subcontracting and Airline Prices
∗
Gaurab Aryal
†
Dennis J. Campbell
‡
Federico Ciliberto
§
Ekaterina A. Khmelnitskaya
December 27, 2023
Abstract
In the US airline industry, independent regional airlines fly passengers on behalf of
several national airlines across different markets, giving rise to common subcontracting.
On the one hand, we find that subcontracting is associated with lower prices, consistent
with the notion that regional airlines tend to fly passengers at lower costs than major
airlines. On the other hand, we find that common subcontracting is associated with
higher prices. These two countervailing effects suggest that the growth of regional
airlines can have anticompetitive implications for the industry.
JEL: D22, L13, L93.
Keywords: Airlines, Common Subcontracting, Anticompetitive Pricing.
∗
We acknowledge the Bankard Fund for Political Economy at the University of Virginia for support. We
thank Ha Pham for outstanding research assistance. We also thank participants and discussants at the 2022
SoAR Symposium, SEA 2022, IIOC 2023 and ABA Antitrust Law Section, for their suggestions.
†
Department of Economics, Washington University in St. Louis, ary[email protected]
‡
Department of Economics, University of Virginia, [email protected]
§
Department of Economics, University of Virginia, DIW and CEPR, cilib[email protected]
Sauder School of Business, University of British Columbia, ekaterina.khmelnitskay[email protected]
1
arXiv:2301.05999v4 [econ.GN] 23 Dec 2023
1 Introduction
Over the last 20 years, national air carriers, e.g., American or Delta Airlines, have increas-
ingly subcontracted their flight operations to regional airlines such as SkyWest and Trans
State Airlines.
1
Figure 1 shows the dramatic growth in the use of regional carriers and
underscores the changing nature of vertical structure in the airline industry.
Figure 1: Use of Regional Airlines
1998 2000 2002 2004 2006 2008 2010 2012 2014 2016
Years
10
20
30
40
50
60
70
80
% of Markets Served by a Regional
Note: The percentage of markets in our sample where operations of at least one flight segment were subcontracted to an
independent regional airline. The observations are at the airline-market-(second quarter of a) year level. At its lowest, in 2002,
regional airlines served at least one segment in 14.2% of the (nonstop or connecting) markets. By 2016, use had risen to 72.7%
While regional carriers operate flights on behalf of only one national airline in a nonstop
flight segment, they can serve several national airlines in different segments. For instance,
in 1998, Trans States Airlines operated flights on behalf of Alaska, Delta, Northwest, Trans
World, and United Airlines. We refer to this industry feature as common subcontracting.
We aim to determine whether this common subcontracting is associated with higher
prices. To test our hypothesis, we use panel regressions of prices on a novel measure of com-
mon subcontracting that depends on the share of passengers transported through regionals
in a given market and the extent of overlapping use of a regional airline across markets by
different national airlines.
1
We use the terms “national,” “legacy,” and “major” airlines interchangeably to mean carriers that sell
tickets. They include Alaska, American, America West, Continental, Delta, JetBlue, Northwest, Southwest,
TWA, United, and US Air. Regional airlines do not sell tickets but only operate flights for major airlines.
2
We use the US Department of Transportation Airline Origin and Destination Survey
(DB1B) dataset from 1998 to 2016 and data from the Official Aviation Guide of the Airways
(OAG) company to identify ticketing carriers and operating carriers. We supplement these
data with ownership information from SEC 10-K filings and The Regional Airline Association
to identify independent regional airlines, like SkyWest, from subsidiaries, like Envoy Air, a
subsidiary of American. Henceforth, regional airline means an independent regional airline.
Our measure of common subcontracting is a function of the shares of passengers trans-
ported by competitors through regional carriers, which is likely to be endogenous in our
panel regressions, even after including carrier, market, and year fixed effects. To address
this concern, we construct instrumental variables based on Forbes and Lederman (2009).
For each major airline, we use a measure of extreme weather conditions that the airline’s
competitors face along the routes they use to serve a market. The identification assump-
tion is that extreme weather conditions facing an airline’s competitors are correlated with
common subcontracting, but they do not directly affect the airline’s prices. Additionally,
we use the size of competitors’ networks of regional airlines as additional instruments as
they are positively correlated with overlapping use of regional carriers and hence common
subcontracting, but they should not directly affect prices.
To isolate the effect of common subcontracting on prices, we control for factor associated
with possible efficiency gains from using smaller planes flown by regional carriers and compe-
tition among national airlines across multiple markets. Specifically, we control for the share
of passengers transported by regional carriers in a market (Forbes and Lederman, 2007, 2009,
2010; He and Kosmopoulou, 2021; Gil, Kim, and Zanarone, 2022) and multimarket contact
among national airlines (Edwards, 1955; Bernheim and Whinston, 1990; Evans and Kessides,
1994; Ciliberto and Williams, 2014), respectively. These two control variables are also likely
endogenous. We, use functions of the weather conditions as an IV for airline’s decision to
use regional airlines and its competitors’ networks as an IV for multimarket contact. Section
4.3 discusses our identification strategy in detail.
3
We find that common subcontracting is associated with higher prices. In particular,
an interquartile range (IQR) increase in common subcontracting increases prices by 1.8%.
Furthermore, this price effect of common subcontracting has increased in recent years. In
particular, an IQR increase in common subcontracting pre-2004 did not affect prices be-
cause common subcontracting was negligible. However, in 2004-2012 and 2012-2016, an IQR
increase in common subcontracting increased prices by 1.7% and 5.1%, respectively.
We also find that regional share is associated with lower prices, which confirms that
relying on regional airlines in a market should lower costs and, therefore, prices. Lastly, we
confirm findings by Evans and Kessides (1994); Ciliberto and Williams (2014) that multi-
market contact increases prices.
The price effects of common subcontracting suggest that the use of regional airlines
per se is beneficial, but their common use by national carriers–common subcontracting–is
anticompetitive. Thus, we provide a novel insight into the vertical structure of the airline
industry. There is a vast literature on the role of the vertical structure of industry on market
outcomes through foreclosure (Coase, 1937; Alchian and Demsetz, 1972; Williamson, 1979;
Nocke and White, 2007). Here, we have identified a new channel, common subcontracting,
which is the exact opposite of foreclosure. In this regard, our paper is similar to Horta¸csu
and Syverson (2007), who identify a non-foreclosure channel to exert market power.
Much remains to be learned of the underlying mechanism that rationalizes our results, but
they are consistent with the conjecture that using common vertical relationships facilitates
collusion, similar to the common ad agency in Bernheim and Whinston (1985) and the trade
association in Awaya and Krishna (2020) that facilitate collusion.
2,3
The paper is organized as follows. In Section 2, we introduce regional airlines and sub-
contracting, and in Section 3, define common subcontracting. In Sections 4- 5, we present
our results. In Appendices A.1-A.2, we detail data constructions and the “Weather IVs.”
2
Regional carriers employ pilots and stewards at the same wages and fly the same types of aircraft
across different markets, independently of the major they serve, they likely reduce cost differences, increase
co-movement of their costs, and lessen cost uncertainties across national airlines that use the same regional.
3
Also see Vives (1990); Piccolo and Mikl´os-Thal (2012) and Clark, Horstmann, and Houde (forthcoming).
4
2 Regional Subcontracting
Major airlines, or legacy airlines, operate large networks of flights across the US. Examples
of major airlines are American Airlines (“AA”), Continental (“CO”), Delta (“DL”), United
(“UA”), and US Airways (“US”). These airlines earn revenue by selling flight tickets within
their network and, occasionally and primarily in the earlier years in our dataset, to flights
in other major carriers’ networks under code-share agreements. They set prices, schedule
flights, manage seat inventories, and market tickets to passengers.
Major airlines compete in many different markets, where a market is defined as a unidirec-
tional trip between any two airports. Airlines can serve a market via several combinations of
flight segments, i.e., segment paths, where a segment is a unidirectional nonstop leg between
two airports. For example, in 2016, AA served the CHO-MCO (Charlottesville-Orlando)
market via two segment paths: CHO-CLT-MCO (through Charlotte) and CHO-PHL-MCO
(through Philadelphia). The first segment path consisted of two nonstop flight segments, one
from CHO to CLT and the other from CLT to MCO.
To transport passengers, major airlines either operate their flights, subcontract operations
to regional airlines, or both. In the remainder of this section, we discuss some characteristics
of this subcontracting relationship in the airline industry over time.
2.1 Characteristics of the Subcontracting Relationship
We have collected relevant information from major airlines and publicly owned regional
airlines’ 10-K filings with the SEC. In these 10-K filings, carriers report information regarding
their contracts with each other, including the length of agreements, the scope of agreements,
and, in certain instances, the payments.
We observe subcontracting agreements (e.g., fixed-fee or revenue-sharing) between major
and (independent) regional airlines. We also observe the start and end dates (if listed) for
such agreements, the presence of termination clauses, and their terms. We collect some
5
of these variables when available. Most regional airlines are publicly traded and work for
multiple major airlines. In practice, subcontracting relationships are exclusive at the segment
level, and a regional carrier serves only one legacy airline on a nonstop flight between two
airports. However, a regional carrier can serve multiple legacy carriers in different routes.
For example, SkyWest can serve AA in the CHO-PHL route and DL in the CHO-CLT route.
We also observe that the relationships between regional and major carriers span many
years. On average, contracts between majors and independent regionals last 10.3 years, with
some lasting as long as 24 years. Some contracts contain clauses for early termination.
4
Lastly, most subcontracting agreements are “fixed-fee” or “capacity-purchase” agree-
ments wherein a major airline sells tickets, often covers certain costs incurred during op-
erations, and pays a fixed fee for service. Airlines may structure their capacity purchase
contracts differently. For example, a major airline may supply fuel or lease aircraft to a re-
gional carrier as part of a capacity purchase contract. Some contracts are “revenue-sharing”
or “pro-rate,” where the regional airline receives a percentage of ticket revenue. However,
such arrangements have become less common in recent years.
Although our data is incomplete, we find that the payments involved in these relationships
can be large. On average, across all the years in our sample, major airlines paid $1.7 billion
per year to each regional carrier they contracted with, with some payments being as high
as $5.66 billion. The total payments have risen markedly over time, consistent with the
increasing use of regional airlines (Figure 2) and the move towards a fixed-fee system.
Forbes and Lederman (2007, 2009) show that the availability of aircraft technologies de-
termines whether a major airline contracts its service on a segment to a regional carrier.
4
For example, in its 2010 10-K filing, Republic Airways Holdings notes:
Delta may terminate the code-share agreements at any time, with or without cause, if it provides
us 180 days’ written notice for the E145 regional jet code-share agreement or after July 2015
for the E175 regional jet code-share agreement. For the E145 agreement, if Delta chooses to
terminate any aircraft early, it may not reduce the number of aircraft in service to less than 12
during the 12 months following the 180-day initial notice period unless it completely terminates
the code-share agreement.
6
From the mid-to-late 1990s until the mid-2000s, airline manufacturers introduced new re-
gional jets that allowed longer flights to be operated by small regional aircraft. Before these
new regional jets, regional airlines typically operated turboprop planes with limited flight
ranges. New regional jets have allowed major airlines to subcontract more flights to regional
airlines as available regional jet technologies can be configured to comply with the major
airline’s needs. For example, Embraer’s E175SC jet is a special configuration of the E175
limited to 70 seats to take advantage of the performance improvements of the E175 while
restricting the number of seats to 70.
2.2 Ownership of Regional Airlines
While regional airlines can be either subsidiaries of major airlines or independently owned,
most are independent. For example, in 2016, seven of the ten largest regional airlines were
independently owned, accounting for 66% of all passengers served by regional airlines.
A major airline can subcontract with an independent regional or a subsidiary, but a
subsidiary regional airline cannot operate on behalf of other major airlines. For instance,
AA owns Envoy, Piedmont, and PSA Airlines, and these regionals fly exclusively for AA.
However, AA also contracts with independent regionals. Some carriers, for example, UA,
subcontract solely to independent regional airlines.
Table 1 presents a chronology of regional airline ownership information from 1998 to 2016
and is based on information we collected from various airlines’ 10-K filings and by searching
across airlines’ websites. Each row corresponds to a regional carrier and its owner, and each
event records the year the ownership changed. For example, consider Shuttle America (S5).
This regional airline was independently owned at the start of our sample and was purchased
by Wexford Capital in 2001. In 2005, it became a part of Republic Airways Holdings, with
regional airlines Chautauqua (RP) and Republic Airlines (YX). Republic Airways Holdings
is not affiliated with any major airline, so Shuttle America is an independent regional airline.
Table 1: Ownership Timeline
Carrier Name Code Parent Company (1) Time (1) Parent Company (2) Time (2)
Cape Air 9K Hyannis Air Service, Inc. 1998 - now
Nantucket Airlines ACK Hyannis Air Service, Inc. 1998 - now
Midwest Airlines
1
YX Midwest Air Group, Inc
10
1998 - 2008 TPG Capital 2008 - 2009
Skyway Airlines AL Midwest Air Group, Inc
10
1998 - 2008
Compass Airlines CP Northwest Airlines 2006 - 2008 Delta Air Lines, Inc. 2008 - 2010
Atlantic S.E. Airlines, Inc.
2
EV 1998 - 1999 Delta Air Lines, Inc. 1999 - 2005
Freedom Airlines, Inc. F8 Mesa Air Group, Inc. 2002 - 2010
GoJet Airlines LLC G7 Trans States Holdings, Inc. 2005 - now
Big Sky Airlines GQ Big Sky Transportation Co. 1998 - 2002 MAIR Holdings Inc.
11
2002 - 2008
Envoy Air
3
MQ AMR Corp. 1998 - 2013 American Airlines Group, Inc. 2013 - now
Comair OH Comair Holdings 1998 - 1999 Delta Air Lines, Inc. 1999 - 2012
Executive Airlines Inc. OW AMR Corp. 1998 - 2013
Horizon Air QX Alaska Air Group Inc. 1998 - now
Republic Airlines YX
5
Republic Airways Holdings 2004 - now
Mesaba Airlines XJ MAIR Holdings Inc.
11
1998 - 2007 Northwest Airlines Corp. 2007 - 2008
Air Midwest ZV Mesa Air Group, Inc. 1998 - 2008
Air Wisconsin Airlines ZW CJT Holdings 1998 - now
Endeavor Air
4
9E Northwest Airlines 1998 - 2003 Pinnacle Airlines Corp. 2003 - 2013
Trans States Airlines AX
6
Trans States Holdings, Inc. 1998 - now
Colgan Air 9L 1998 - 2007 Pinnacle Airlines Corp. 2007 - 2012
SkyWest Airlines OO SkyWest, Inc. 1998 - now
Chautauqua Airlines RP Wexford Capital
12
1998 - 2004 Republic Airways Holdings 2004 - 2014
Shuttle America Corp. S5 1998 - 2001 Wexford Capital
12
2001 - 2005
ExpressJet Airlines, Inc. XE
7
Continental Airlines 1998 - 2002 ExpressJet Holdings Inc. 2002 - 2010
Mesa Airlines YV Mesa Air Group Inc. 1998 - 2011 Mesa Air Group Inc. 2011 - now
America West Airlines HP America West Holdings Corp. 1998 - 2005 US Airways Group Inc. 2005 - 2007
Business Express Airlines HQ 1998 - 1999 AMR Corp. 1999 - 2000
Trans World Airlines TW 1998 - 2001 AMR Corp. 2001 - 2001
Frontier Airlines F9 1998 - 2006 Frontier Airlines Holdings, Inc 2006 - 2009
Scenic Airlines YR SkyWest Inc. 1998 - 2007 Grand Canyon Airlines
14
2007 - 2009
PSA Airlines OH
8
US Airways 1998 - 2013 American Airlines Group, Inc. 2013 - now
USAir Shuttle TB US Airways Group, Inc. 1998 - 2000
UFS Inc. U2 Trans States Holdings, Inc. 1998 - 2000
Lynx Aviation L3
9
Frontier Airlines Holdings, Inc. 2006 - 2009 Republic Airways Holdings 2009 - 2011
Piedmont Airlines PT US Airways 1998 - 2013 American Airlines Group, Inc. 2013 - now
Allegheny Airlines AL US Airways 1998 - 2004
13
Note: Ownership of regional airlines and the corresponding event years. This information is collected by the authors from
various sources including the SEC 10K filings and airlines’ webpages. For several airlines their names have changed and they
are recorded with a numbered notes as follows: (1) Midwest Express Airlines before 2003; (2) ExpressJet Airlines, Inc. since
2011; (3) American Eagle Airlines, Inc. until 2014; (4) Express Airlines I before 2002 and Pinnacle Airlines, Inc before 2013;
(5) RW before 2009; (6) also used the code 9N; (7) also used the code RU; (8) used code 16 before 2013; (9) L4 before 2009,
and 0IQ before 2007; (10) Midwest Express Holdings, Inc. before 2004; (11) Mesaba Holdings Inc. until August 2003; (12)
Wexford Management, LLC before 2000; (13) merged into Piedmont Airlines; (14) absorbed by Grand Canyon Airlines; contd.
2.3 Regional Airline Usage Patterns
We use the US DOT’s DB1B dataset to investigate regional airline usage and how it has
changed over time. This dataset consists of a 10% sample of domestic airline tickets. We use
8
Table 1: Ownership Timeline (continued)
Carrier Name Code Parent Company (3) Time (3) Parent Company (4) Time (4)
Cape Air 9K
Nantucket Airlines ACK
Midwest Airlines
1
YX Republic Airways Holdings 2009 - 2009
Skyway Airlines AL
Compass Airlines CP Trans States Holdings 2010 - now
Atlantic S.E. Airlines, Inc.
2
EV SkyWest, Inc. 2005 - now
Freedom Airlines, Inc. F8
GoJet Airlines LLC G7
Big Sky Airlines GQ
Envoy Air
3
MQ
Comair OH
Executive Airlines Inc. OW
Horizon Air QX
Republic Airlines YX
5
Mesaba Airlines XJ Delta Air Lines, Inc. 2008 - 2010 Pinnacle Airlines Corp. 2010 - 2012
17
Air Midwest ZV
Air Wisconsin Airlines ZW
Endeavor Air
4
9E Delta Air Lines 2013 -now
Trans States Airlines AX
6
Colgan Air 9L
SkyWest Airlines OO
Chautauqua Airlines RP
Shuttle America Corp. S5 Republic Airways Holdings 2005 - 2017
15
ExpressJet Airlines, Inc. XE
7
SkyWest Inc 2010 - 2011
16
Mesa Airlines YV
America West Airlines HP
Business Express Airlines HQ
Trans World Airlines TW
Frontier Airlines F9 Republic Airways Holdings 2009 - 2013 Indigo Partners LLC 2013 - now
Scenic Airlines YR
PSA Airlines OH
8
USAir Shuttle TB
UFS Inc. U2
Lynx Aviation L3
9
Piedmont Airlines PT
Allegheny Airlines AL
Note: (continued) (15) merged into Republic Airlines); (16) merged into Atlantic Southeast Airlines, but code XE was still
used till January 2012; and (17) merged into Pinnacle Airlines.
data from every second quarter from 1998 to 2016 and define a market as a unidirectional
trip between two airports, irrespective of intermediate transfer points.
5
The DB1B database reports the ticketing carrier for service in a market for each ticket
sold in a quarter, as well as the operating carrier(s) that transported the passenger along the
5
See Appendix A.1 for more detail on the sample construction.
9
segment path used for each ticket. Essentially, we observe the segment path for each ticket
sold. We observe 16 different major airlines, including legacy airlines mentioned above, low-
cost carriers such as Southwest (WN) and JetBlue (B6), and 29 regional carriers. Some
regional carriers are subsidiaries of major airlines, e.g., Envoy and Horizon, and others are
independently owned, e.g., Skywest, Mesa, and ExpressJet.
6
Our final sample has 3, 499, 463
unique market-year-ticketing carrier-segment path observations from 25, 767 unique markets.
Table 2 presents a snapshot of the regional airline landscape at the start, middle and end
of our sample that includes the name of the regional airline, its IATA code, all major airlines
it worked for, number of markets served, and its percentage of all unique markets.
7
Regional usage has been increasing over time. In 1998, American Eagle, a regional airline
owned by AA, operated in 6,272 markets, or 30.8% of the total 20,373 markets. The most-
used independent regional carrier in 1998 was Mesaba Airlines, which operated in 4,296
markets or 21.1% of the total. In 2016, the most-used regional airline was ExpressJet, an
independent regional airline operating in 15,332 markets, 70.5% of the 21,783 markets.
Figure 2 displays the share of passengers that (subsidiaries and independent) regional
airlines transport on behalf of major airlines. It shows a striking rise in the usage of regionals
over time. For instance, from 1998 to 2016, AA increased its percentage from 15.6% to 33.1%.
2.4 Regional Share
We consider the share of passenger-seat miles transported through (independent) regional air-
lines. We refer to this variable as “regional share,” and denote it by Regional Share
j,m,t
, in
year t = 1, . . . , T , market m = 1, . . . , M, and major airline j ∈ n := {AA, AS, B6, CO, . . .}.
The variable Regional Share
j,m,t
measures the reliance of a (national) airline on regional
airlines in a particular market at a given time. We consider both the number of seats flown
by the regional carrier and the miles to facilitate the interpretation of this variable as a proxy
6
We confirm operating information for regional carriers using OAG Market Intelligence-Schedules dataset.
7
The numbers in Table 2 may add up to more than 100% because airlines may use different segment paths
to serve a market, and a route may be served by different regional airlines.
10
Table 2: Markets Served by Regional Airlines
Regional Code Airlines Served # of Mkts % of Total Mkts
1998: Trans States Airlines 9N AS, DL, NW, TW, UA, US 2,486 12.2
Atlantic Southeast EV DL 4,059 19.9
American Eagle MQ AA (Owned) 6,272 30.8
Horizon Air QX AS (Owned) 712 3.5
ExpressJet RU CO (Owned) 5,623 27.6
US Air Shuttle TB US (Owned) 166 0.8
UFS U2 UA 886 4.3
Mesaba Airlines XJ NW (Owned, until 2007) 4,296 21.1
Air Wisconsin ZW UA 1,267 6.2
2007: PSA Airlines 16 UA, US 4,883 24.6
ExpressJet EV CO, DL, NW 10,897 55.0
GoJet Airlines G7 UA, US 1,542 7.8
American Eagle MQ AA (owned) 9,085 45.8
Comair OH DL (owned) 6,433 32.5
Executive Airlines OW AA (owned) 167 0.8
Horizon Air QX AS (owned) 1,495 7.5
Republic Airlines RW F9, UA, US 3,175 16.0
Shuttle America S5 CO, DL, NW, UA, US 3,214 16.2
Mesaba Airlines XJ CO, DL, NW 1,843 9.3
Mesa Airlines YV UA, US 7,385 37.3
Air Wisconsin ZW UA, US 3,705 18.7
2016: Endeavor Air 9E DL (owned) 7,361 33.9
Compass Airlines CP AA, AS, DL 3,625 16.7
ExpressJet EV AA, AS, DL, UA 15,332 70.5
GoJet Airlines G7 DL, UA 4,940 22.7
Envoy Air MQ AA (owned) 9,234 42.5
PSA Airlines OH AA (owned) 8,595 39.5
Skywest OO AA, AS, DL, UA 12,333 56.7
Horizon Air QX AS (owned) 1,150 5.3
Shuttle America S5 DL, UA 4,583 21.1
Mesa Airlines YV AA, AS, UA 7,588 34.9
Air Wisconsin ZW AA 4,315 19.9
Note: Rows correspond to regional airlines. The first column is its name; the second is its two-digit code. The third column is
the set of all major airlines the regional airline has worked for that year. The fourth column is the total number of markets the
regional airline serves. The fifth column is the percentage of regional airlines’ total markets.
11
Figure 2: Usage of Regional Airlines
1998 2000 2002 2004 2006 2008 2010 2012 2014 2016
15
20
25
30
AA
1998 2000 2002 2004 2006 2008 2010 2012 2014 2016
15
20
25
AS
1998 2000 2002 2004 2006 2008 2010
0
20
40
CO
1998 2000 2002 2004 2006 2008 2010 2012 2014 2016
0
20
40
DL
1998 2000 2002 2004 2006 2008
0
10
20
(%) Passengers Transported by Regional Ailrines
NW
1998 2000 2002 2004 2006 2008 2010 2012 2014 2016
Years
0
20
40
UA
1998 2000 2002 2004 2006 2008 2010 2012 2014
Years
0
50
US
Note: The time trends of the percentage of total passengers transported by regional airlines for each legacy airline.
for the utilization cost associated with subcontracting. We consider both the number of seats
flown by the regional carrier and the miles to facilitate the interpretation of this variable as
a proxy for the utilization cost associated with subcontracting. This measure is important
for our empirical analysis because it is a proxy for costs, and we expect it to lower prices.
For an illustration of how we calculate the share, consider an example in Table 3. Suppose
we consider the Charlottesville to Dallas Fort Worth (CHO-DFW) market being served by
AA and DL. Suppose AA serves the market through either Charlotte, i.e., CHO-CLT-DFW,
or Philadelphia, i.e., CHO-PHL-DFW, and that 100 passengers fly via CLT and 50 via PHL.
Now, suppose that OO serves the CHO-CLT segment on behalf of AA, and no other regionals
are used in this market.
Here, AA uses OO to transport 100 passengers in the CHO-CLT segment, which is 150
(nautical) miles. The total passenger seat-miles transported by AA is 100 × (150 + 300) +
50 × (50 + 450) = 70, 000, out of which 100 × 150 = 15, 000 miles are through OO. Therefore,
12
Table 3: Example of Market-Segment Paths Information
Major Mkt Seg1 Carrier1 Seg1 Dist Seg2 Carrier2 Seg2 Dist #Pass
AA CHO-DFW CHO-CLT OO 150 CLT-DFW AA 300 100
AA CHO-DFW CHO-PHL AA 50 PHL-DFW AA 450 50
DL CHO-DFW CHO-ATL DL 200 ATL-DFW DL 250 200
Note: Example of usage of regional airlines. The first column of the table, “Major,” lists the airline that sold tickets for the
given segment path. The second column, “Market,” lists the origin and destination airport of the market. The column “Seg1”
lists the first flight segment along the given segment path, “Carrier1” lists the carrier’s code which operated that flight, and
“Seg1 Dist” lists the nonstop distance of flight segment 1. The last column, “#Pass,” is the number of passengers transported
along the given segment path and who purchased a ticket from the major carrier.
the regional share for AA in the CHO-DFW market is 0.21. Likewise, the regional share for
DL in the CHO-DFW market is 0.
We calculate such shares for each national carrier, year, and market. The average regional
share across our sample is 0.154, with a median of 0. In Figure 3, we present the yearly
average of regional-share, where the average is taken across all airlines and all markets. We
observe that regional share has increased over time.
Figure 3: Regional Share
Note: Yearly average of regional-share, where the average is taken across all national airlines and markets.
13
3 Common Subcontracting and Multimarket Contact
3.1 Common Subcontracting
We now discuss the overlapping use of independent regional airlines by major airlines and
present our novel measure, or index, of the extent of this phenomenon, which we refer to as
common subcontracting.
We use a stylized example to illustrate how we construct our measure. Suppose two major
airlines (AA and DL) operate in two markets (1 and 2). First, suppose only AA subcontracts
its operations to a regional airline (OO) in Market 1. DL does not use any regional in either
of the two markets. Here, there is no overlapping use of regional airlines. Now suppose that
DL subcontracts its Market 2 operations to OO, resulting in an overlapping use of OO by
AA and DL. These two cases are shown in Figure 4.
Figure 4: Example of Common Subcontracting
Case 1
Market 1
AA DL
OO
Market 2
AA DL
Case 2
Market 1
AA DL
OO
Market 2
AA DL
OO
Note: Example of overlapping use of regional airlines. Case 1: only AA uses a regional airline (OO) in Market 1 (denoted by
the dotted line connecting OO and AA), so there is no overlap. Case 2: AA uses OO in Market 1, and DL uses OO in Market
2, creating an overlapping use of regional between AA and DL (denoted by the bold dashed lines between these two airlines).
We take the following steps to construct the index of common subcontracting. First, we
take all the segment paths each major airline uses to serve a market and record whether the
major airline subcontracts operations of one or more flight segments to regional airlines.
Second, we calculate the percentage of passengers operated by each independent regional
carrier on behalf of each major airline at the market level across all possible segment paths.
Consider the example in Table 3 again. Here the percentage of passengers served by carrier
14
OO in this market on behalf of AA is
100
150
= 0.667.
Third, we use data across all markets to determine for each period t and each major airline
j the names of all regional airlines it subcontracts, thereby determining which major and
regional airlines have vertical relationships somewhere within the larger network of flights.
Fourth, we use the passenger percentages served by regional carriers and information
on other major airlines that subcontract with these regionals anywhere in the network to
determine the overlap between each major airline and each of its competitors in the market.
Returning to Table 3, recall that only AA subcontracts out to regional airlines, and it
subcontracts 66.7% of its passengers to regional carrier OO. If DL has a relationship with
OO from other flight segments, we say that DL overlaps with 66.7% of AA’s operations in
this market. By contrast, AA has 0% overlap with DL’s operations since no other regionals
are involved in that market.
8
Fifth, and finally, we average the information shares calculated in step four across all
unidirectional major carrier pairs in a given market to measure the average common sub-
contracting in each market. In the example market with AA and DL, we have an average
measure of common subcontracting equal to 33.3%.
9
Note that in Case 1 in Figure 4, com-
mon subcontracting is zero because there is no overlapping usage of a regional airline.
Next, we define the measure formally. Let i and j denote major airlines where i, j ∈ n :=
{AA, AS, B6, CO, . . .}. Following the steps described above, we define the level of common
subcontracting in market m in period t as
CSC
mt
=
1
n (n − 1) |K
m
|
n
X
i=1
X
j̸=i
X
k∈K
m
s
kjmt
× B
ikt
, (1)
where s
kjmt
is the share of passengers in m transported by the regional carrier k ∈ K
m
on
behalf of major carrier j in market m at time t; B
ikt
= 1 if carrier i subcontracts with
8
Here, we have two national airlines. So there are two unidirectional carrier-pair-specific measures. In
markets with three major airlines, we obtain six different measures.
9
Suppose DL subcontracts with OO to transport all its 200 passengers in Segment 1. Then, AA would
have an overlap of 100%, and the common subcontracting would be (100% + 66.7%)/2 = 83.5%.
15
Table 4: Summary Statistics
Mean Median St. Dev
Market-Specific
Common Subcontracting 0.161 0.045 0.242
Multimarket Contact 5.559 5.048 2.689
Carrier-Market-Specific
Regional Share 0.154 0 0.274
Price ($) 258.14 242.56 95.33
Log Price ($) 5.495 5.491 0.339
Network- Origin (in 1,000) 0.843 0.870 0.299
Network- Destination (in 1,000) 0.841 0.870 0.298
Note: Summary statistics of key variables based on a sample of 1,001,835 carrier-market-time observations across 279,958
market-time observations. Common subcontracting is defined in Equation (1), and multimarket contact is defined in Equation
(2) (in 1,000). Price refers to one-way fare expressed in 2012 US dollars; regional share is the share of passenger seat miles
transported through regionals and is defined at the major carrier and market level; the network is the number of markets served
out of the origin-destination airports (in 100).
regional carrier k in at least one (potentially different) market, and B
ikt
= 0 otherwise. So,
for example, if DL subcontracts OO in market m
′
̸= m, then B
DL,OO,mt
= 1. Thus, CSC is
an average across major carriers of the share of passengers transported by shared regional
airlines.
The common subcontracting index takes values between 0 and 1 and is defined at the
market level. A zero value means that none of the shared regional airlines transport pas-
sengers. In contrast, common subcontracting of one means that all of the passengers in the
market are transported by shared regional airlines.
Table 4 shows that the average level of CSC
mt
in our sample across market-time observa-
tions is 0.161, while the median is 0.045. The low relative median value is because there was
relatively little regional usage in the early years of our sample (see Figure 2).
Figure 5 (left panel) displays box plots of common subcontracting from 1998 until 2016.
We observe that over our sample period, the level of the common subcontracting present
in markets increased substantially. For instance, the 50
th
and 75
th
percentiles have gone
from 0 in 2003 to 0.18 and 0.5, respectively, in 2016. In that time, the unweighted mean of
common subcontracting across markets increased by 1,350% from 0.02 to 0.29. The figure
shows a modest increase in the 2005 followed by a rapid increase in 2012 that coincided with
16
bankruptcies (Ciliberto and Schenone, 2012), consolidation among regional airlines, and with
the introduction of new regional jet technologies. These mergers were between DL and NW,
UA and CO, and US and AA.
Figure 5: Common Subcontracting and Multimarket Contact
Note: Box plots (with whiskers) of common subcontracting across markets (left) and multimarket contact (right) by year.
3.2 Multimarket Contact
Next, we define multimarket contact following the existing literature. See Evans and Kessides
(1994) and Ciliberto and Williams (2014) and the references therein. We denote this measure
of contact by MMC, and it is the average of the total pairwise contacts of the firms in a market.
To define this measure more formally, we introduce the following notation. Let i and
j be defined as before and denote major airlines where i, j ∈ n := {AA, AS, B6, CO, . . .}.
For an airline pair {i, j} ∈ n × n and a market m ∈ M, let D
ijmt
∈ {0, 1} be a dummy
variable equal to one if both i and j are present in market m in time t and zero otherwise.
Let r
ijt
:=
P
m∈M
D
ijmt
be the total number of pairwise contacts over all the markets served
17
by both i and j in period t. Then, as Evans and Kessides (1994), multimarket contact is
MMC
mt
=
Contact-mmc
mt
N
Contact-mmc
mt
:=
P
n
i=1
P
n
j=1
j̸=i
(r
ijt
× D
ijmt
)
P
n
i=1
P
n
j=1
j̸=i
D
ijmt
, (2)
where Contact-mmc
mt
is the total pairwise contact in market m and N
Contact-mmc
mt
is the total
number of possible pairwise contacts, divided by 1,000.
Table 4 shows that the average of multimarket contact is 5.559. This value is smaller
than the average values in Evans and Kessides (1994) and Ciliberto and Williams (2014)
because our sample includes all MSAs to ensure that we also capture smaller markets.
Figure 5 (right panel) displays box plots for multimarket contact across markets from
1998 until 2016. Median multimarket contact more than doubled from 3.93 in 1998 to 7.65 in
2016, and the 75th percentile increased by 170%, from 3.97 to 10.73. Of particular note are
the jumps from 2010 to 2011 and 2013 to 2014. The unweighted mean of multimarket contact
across all markets almost doubled from 4.26 pre-2010 to 7.96 post-2010. As with common
subcontracting, these changes in multimarket contact coincide with mergers of major airlines.
4 Empirical Analysis
We describe the ticket prices and carriers’ networks at the origin and destination airports.
Then, we introduce the empirical specification and discuss identification and results.
4.1 Prices and Network Service
Average fares are calculated at the carrier-market-time level, deflated using the 2012 con-
sumer price index. Whenever a ticketing carrier operates more than one segment path to
serve a market, we use weighted average prices from all segment paths where the weights are
the share of passengers transported along each segment path. In other words, the weighted
average price is the ratio of total revenue in a market to the total number of passengers. The
18
average fare in the sample is $258.14 with a standard deviation of $95.33. See Table 4.
Next, we introduce Network Origin
jmt
and Network Destination
jmt
as additional con-
trols that measure the airline’s ability to offer differentiated products at a given market and
time (Berry, 1990, 1992; Ciliberto and Tamer, 2009). These variables are the number of
markets (in 100s) airline j serves from the origin and destination airports of market m in
time t, irrespective of any stops. For example, suppose AA serves the CHO-DFW market
via CLT. Then the origin and destination networks are the numbers of airports AA serves
from CHO and DFW, respectively.
We contend that these measures of an airline’s network are exogenous relative to prices
in any particular year and market because setting up an airline network is a sticky process
that takes many years and requires sunk start-up investments. As shown in Table 4, airlines
serve an average of 84 markets from an airport.
4.2 Empirical Specification
We hypothesize that, on average, prices are higher in markets with either high common
subcontracting after accounting for the lower cost associated with using regional carriers by
airlines and higher prices from multimarket contact. To evaluate this hypothesis, we use
panel data and estimate a two-way fixed-effects model:
ln(price
j,m,t
) = β
0
+ γ
j
+ γ
m
+ γ
t
+ CSC
m,t
β
csc
+ Regional Share
j,m,t
β
share
+ MMC
m,t
β
mmc
+Network Origin
j,m,t
γ
origi n
+ Network Destination
j,m,t
γ
dest
+ ε
j,m,t
, (3)
where the dependent variable is the log of the ticket (one-way) price sold by major airline j
in market m in year t, and γ
j
, γ
m
, γ
t
are, respectively, carrier, market, and year fixed-effects.
To isolate the effect of common subcontracting on prices, we also control for two variables:
the share of passengers transported by regional carriers and multimarket contact. These
variables respectively account for factors associated with the costs of using regional carriers,
19
and forbearance among national airlines in regard to multimarket contact.
10
4.3 Identification
The main threat to identifying the coefficients in (3) is that the first three variables are likely
endogenous, as we explain next. To address that concern, we propose to use instrumental
variables. Here we introduce three sets of instrumental variables. The first two sets rely
on variations in the airlines’ network structures, and the third relies on extreme weather
patterns across airports.
First, we consider common subcontracting and regional share. Omitted variables may
affect prices, the level of common subcontracting, and the regional share. In particular,
we are concerned about unobserved costs of transporting passengers and mergers among
national airlines that may differ across regional and major carriers. As mentioned earlier,
regional carriers use smaller aircraft, the so-called “regional jets,” and have lower wage bills.
So the unobserved cost of transporting passengers may be positively correlated with prices
and negatively correlated with common subcontracting. Common subcontracting may also
(mechanically) change with mergers among national airlines, and in so far as these mergers
affect prices, it may result in omitted variable bias. For a similar reason, unobserved costs
can affect major airlines’ subcontracting decisions and, therefore, the regional share.
To address this concern, we use information about the network sizes of competitors’
regional airlines as our first set of instrumental variables. The key idea here is that a major
airline’s competitors’ regional networks in a market affect common subcontracting in that
market but do not directly affect the prices charged by that major airline. We measure the
network size of a regional airline to be the total number of airports that the regional airline
operates in a given year. To construct the instrumental variable that measures competitors’
regional networks, we first take the average network size of all the regionals a major uses
10
Besides the papers mentioned earlier, there is extensive research that shows that multimarket contact
leads to higher prices and lower quality, see Busse (2004); Chicu and Ziebarth (2013); Molnar, Violi, and
Zhou (2013); Schmitt (2018); Lin and McCarthy (2019) and Eizenberg, Shilian, and Blanga (2023).
20
in a given market. Then, for each major in the market, we aggregate the average regional
network size of the competitors. We refer to this set of IVs as the “Regional Network IV.”
The next set of variables is based on the work by Forbes and Lederman (2009), who shows
that weather conditions at an airport affect the decisions of major airlines to subcontract
operations to regional airlines in a given flight segment. At the market-major level, we
take the most extreme value of four weather condition variables (maximum precipitation,
snowfall, snow depth, and lowest minimum temperature) of all airports used along the routes
used by the major airline to serve the market. We maintain that conditional on our control
variables in the pricing regression, including the fixed effects, the weather does not directly
affect prices since there should be no “weather surcharge.”
We use two sets of excluded weather variables: one for common subcontracting and one
for regional share. For common subcontracting, we use four measures of the most extreme
weather conditions that a major airline’s competitors encounter in the routes they use to serve
a market. Airlines use different routes to serve the same market, which induces variation in
the weather across competitors. Common subcontracting is a function of the regional usage
by an airline’s competitors in a market. So any excluded variables that affect competitors’ use
of regionals (such as extreme weather) will correlate with common subcontracting. On the
other hand, for regional share, we use four measures of the most extreme weather conditions
that a given major airline encounters while serving a market as excluded variables. We refer
to this set of IVs as “Weather IVs” and provide additional details in Appendix A.2.
Finally, the key challenge in identifying the price effect of multimarket contact is that
it may be correlated with unobserved market profitability, as discussed in Ciliberto and
Williams (2014). Specifically, an airline is more likely to enter a market with a higher
expected profit and charge a higher price than in a market with a lower expected profit.
Such markets will, therefore, also have higher multimarket contact.
To address this concern, we rely on the fact that multimarket contact depends on market
structure, which in turn depends on the networks of the airlines (e.g., Berry, 1992; Ciliberto,
21
Murry, and Tamer, 2021). Therefore, we use instrumental variables that relate to the larger
network decisions of major airlines’ competitors. For each market, we construct a vector of
the number of markets that major airlines operating in the market serve both out of the
origin and destination airport associated with that market. So, the size of this vector is
twice the number of major airlines operating in that market. Then, from this vector, for
each airline, we consider the vector corresponding to the airline’s competitors. This vector
captures information about major airlines’ competitors’ larger networks, which are fixed in
the short run and thus affect multimarket contact but not directly the price. We label these
two sets of excluded variables “Network IVs.” These Network IVs differ from network controls
in that the latter are defined for a major carrier, whereas the IVs capture the network size
measures of that carrier’s market competitors.
Finally, there may be a concern about the correlation between mergers and common sub-
contracting. Mergers affect common subcontracting via the market structures. So, mergers
may indirectly affect prices through common subcontracting. They may also have direct
price effects because they change the market structure, hence the ability to exercise market
power, and, possibly, because they may reduce costs through efficiency gains. However, in
our sample, common subcontracting keeps varying over time and across markets, whereas
mergers are one-time events. So, we rely on that continued variation in common subcontract-
ing and its correlated variation in the IVs for estimation. Besides, the increase in common
subcontracting is a more recent phenomenon (Figure 5) than the mergers, suggesting that
we can identify the effect of common subcontracting on prices.
4.4 Estimation Results
Column (1) of Table 5 presents the fixed effect estimates of the first three coefficients in
Equation (3). The coefficient of common subcontracting is imprecisely estimated, and the
magnitude is economically insignificant. We also find that regional share is positively asso-
ciated with the price, while we expect it to be negative. We find that multimarket contact is
22
associated with higher prices, consistent with previous work on the price effect of multimar-
ket contact. The coefficient of multimarket contact is equal to 0.0135, and it is statistically
and economically important. In particular, if we move from a market at the 25th percentile
of multimarket contact to one at the 75th percentile, this interquartile range (IQR) move
increases multimarket contact by 3.476, which in turn is associated with a 4.69% increase in
average price. For comparison, in the largest 1,000 markets from 1984 to 1988, Evans and
Kessides (1994) find that an IQR increase in multimarket contact increases prices by 5.1%.
The results in column (1) are likely biased due to the endogeneity of all three variables.
To address this concern, column (2) of Table 5 presents estimates using a control function
approach using three instrumental variables: “Regional Network IVs,” “Weather IVs,” and
“Network IVs.” We also present coefficients for the first-stage residuals of all three variables.
We find that the coefficient for common subcontracting increases to 0.0834, and the
coefficient is precisely estimated. This result is consistent with the negative coefficients
for the first-stage residual for common subcontracting. Thus, common subcontracting is
associated with higher prices. In particular, an IQR increase in common subcontracting
from 0 to 0.214 increases average prices by 1.8%.
11
We also find that the regional share lowers the price, which confirms the reasoning that
relying on regional airlines in a market should lower costs. The coefficient of regional share
is precisely estimated at −0.208, so an IQR increase in regional share lowers prices by 3.9%.
The estimate for multimarket contact at 0.0231 implies that an IQR increase in multi-
market contact by 3.476 increases the average price by 8%. For comparison, Ciliberto and
Williams (2014), find that an IQR increase in multimarket contact increases prices by 6.5%.
We also explore whether, given the structural changes we observe in the industry (Figures
1, 2, 3 and 5), the price effects have changed over time.
To capture the time trend, we define categorical variables associated with the periods
where the variables showed different patterns. For common subcontracting and regional
11
The negative bias of the estimate without the instrumental variables is consistent with the findings of
Carlton, Israel, McSwain, and Orlov (2019), who find that mergers are pro-competitive.
23
Table 5: Common Subcontracting, Prices and Quantities
Variables (1) (2) (3) (4) (5) (6) (7) (8)
ln P ln P ln P ln Q ln Q ln P ln P ln P
CSC -0.00253 0.0834*** 0.0919*** -0.3442*** -0.8102*** 0.107*** 0.108*** 0.0883***
(0.00290) [0.00534] [0.0110] [0.0271] [0.0039] [0.00461] [0.00555] [0.00668]
Regional Share 0.0291*** -0.208*** -0.126*** 0.1081 0.2592*** -0.214*** -0.241*** -0.187*
(0.00175) [0.0180] [0.0195] [0.103] [0.1090] [0.0171] [.0182] [0.0197]
MMC 0.0135*** 0.0231*** 0.0335*** 0.0054** -0.0075*** 0.0208*** 0.0230*** 0.0181***
(0.000397) [0.000493] [0.000566] [0.0027] [0.0029] [0.000509] [0.000493] [0.000592]
Regional HHI 0.131***
[0.00694]
CSC 2004-2012 0.0112 0.5304***
[0.0103] [0.0308]
CSC Post-2012 0.0278*** 0.4102***
[0.0101] [0.0308]
Regional Share 2004-2012 -0.0278*** -0.2208***
[0.00586] [0.0211]
Regional Share Post-2012 -0.0490*** -0.2961***
[0.00583] [0.0219]
MMC Post-2010 -0.0150*** 0.0205***
[0.000371] [0.0017]
First-Stage Results
Residuals: CSC -0.106*** -0.118*** 0.332*** 0.3668*** -0.164*** -0.133*** -0.101***
[0.00596] [0.00598] [0.0299] [0.0302] [0.00510] [0.00602] [0.00727]
Regional Share 0.240*** 0.188*** -0.334*** -0.2622*** 0.251*** 0.274*** 0.212***
[0.0181] [0.0186] [0.1043] [0.1056] [0.0172] [0.0182] [0.00198]
MMC -0.0155*** -0.0144*** -0.066*** -0.0691*** -0.0111*** -0.0153*** -0.0111***
[0.000493] [0.000595] [0.0032] [0.0032] [0.000611] [0.000591] [0.000691]
F-stats (IVs): CSC 1,240.34 1,240.34 1,240.34 1,240.34 2,369.70 754.94 1,240.34
Regional Share 649.96 649.96 649.96 649.96 649.96 649.96 649.96
MMC 5,350.11 5,350.11 5,350.11 5,350.11 5,350.11 5,350.11 5,350.11
Regional HHI 1,079.66
Control Functions ✗ ✓ ✓ ✓ ✓ ✓ ✓ ✓
CSC (benchmark) ✓ ✓ ✓ ✓ ✓ ✗ ✗ ✓
CSC (count) ✗ ✗ ✗ ✗ ✗ ✓ ✗ ✗
CSC (weighted) ✗ ✗ ✗ ✗ ✗ ✗ ✓ ✗
Observations 1,000,180 1,000,180 1,000,180 1,000,180 1,000,180 1,000,180 1,000,180 802,265
Number of markets 21,390 21,390 21,390 21,390 21,390 21,390 21,390 20,815
R
2
0.092 0.094 0.096 0.29 0.29 0.094 0.093 0.102
Note: The table presents the estimates of price regression (3) in columns (1)-(3) and (6)-(8). Columns (4)-(5) present the
estimates of quantity regression where the dependent variable in (3) is replaced with the log of passenger traffic (number of
passengers transported). Here, CSC stands for common subcontracting MMC stands for multimarket contact. Estimates in
columns (2)-(8) use control functions. When we estimate the first-stage residual for multimarket contact, we omit the regional-
specific Weather IVs and Regional Network IVs. Market fixed effects, network IVs, weather IVs, and reg network IVs were used
in all regressions. Other control variables included in all specifications but whose coefficients are not reported include dummy
variables for year-quarter, markets, and carriers, and controls for network size measured by origin and destination networks.
Columns (6)–(8) use the same specification as column (2), except column (6) uses common subcontracting (count), column (7)
uses common subcontracting (weighted), and column (8) uses the baseline definition of common subcontracting but includes
market-specific Herfindahl-Hirschman Index of regional carriers (Regional HHI) based on the share of passengers transported.
For column (2), robust standard errors are in parenthesis; for the rest, bootstrapped standard errors based on 1,000 samples
are in square brackets.
24
share, we consider the period before (including) 2003, between 2004 and 2012, and after
2012. In the case of multimarket contact, we consider the period before and after 2010. We
then interact these categorial variables with the three variables of interest and include them
as additional regressors to (3).
The estimates using all the instrumental variables are presented in column (3) of Table
5. The coefficient estimate of Common Subcontracting 2004-2012 is the estimated differ-
ence between the baseline pre-2004 estimate of Common Subcontracting and the effect from
2004-2012. Similarly, the coefficient estimate of Multimarket Contact Post-2010 can be inter-
preted as the change in the effect of multimarket contact relative to the coefficient estimate
of Multimarket Contact. The coefficient estimates are precisely estimated, except for the
difference between the common subcontracting estimate pre-2004 and the estimate in the
period 2004-2012, which is consistent with the generally low level of regional share in those
periods. Thus, there is no difference in the effect of common subcontracting in 2004-2012
relative to pre-2004.
We find that the price effect of common subcontracting is larger in the post-2012 period
than before, and the negative price effect of regional share is stronger after 2004, suggesting
that the cost savings have increased. However, we find that the price effect of a multimarket
increase is smaller in later years when the absolute level of multimarket contact is larger.
For ease of comparison, Table 6 presents the price effect of increasing common sub-
contracting, regional-share, and multimarket contact by their specific IQR, separately for
periods implied by the estimates in column (3) of Table 5.
These estimates suggest that if we increased the common subcontracting by its IQR, it
would not have affected prices pre-2004 because common subcontracting was negligible, with
IQR effectively zero. In 2004-2012, if we increase common subcontracting by its IQR, the
price would increase by 1.72%, but by post-2012, the price would increase by 5.07%. Thus,
the potential anticompetitive effect of common subcontracting has recently become stronger.
The decrease in prices due to regional share shows a similar trend as common subcon-
25
Table 6: Price Effects over Time
Years Common Subcontracting Regional-Share Multimarket Contact
Pre-2004 0.06% 0%
2004-2012 1.72% -3.92%
Post-2012 5.07% -7.42%
Pre-2010 7.35%
Post-2010 9.13%
Note: Price effects of an IQR increase in common subcontracting, regional share, and multimarket contact, as implied by the
estimates in column (3) Table 5. Pre-2004 means 1996 to 2003, post-2004 means 2004 to 2012, and post-2012 means 2013 to
2016. Similarly, pre-2010 means from 1996 to 2009, and post-2010 means 2010 to 2016.
tracting. In particular, the estimates suggest that an IQR increase in regional share would
have had no effect on price pre-2004, but in the years 2004-2012 and post-2012, the price
would have decreased by 3.92% and 7.42%, respectively.
The growth rate of the price effect for multimarket contact is lower than common sub-
contracting. Our estimates suggest that an IQR increase in multimarket contact leads to
a 7.35% pre-2010 and 9.13% post-2010. Thus the increase in price effects for multimarket
contacts is relatively smaller than that for common subcontracting, which is consistent with
the difference is the increase in the level of common subcontracting and multimarket contact
(see Figure 5). As mentioned before, the growth in common subcontracting is due to the use
of regional airlines is a relatively recent phenomenon, fueled by introduction of new regional
jets, rising consolidation among regional airlines and mergers among major airlines.
Finally, we note that a difficulty in interpreting the estimated increase in price as an
anticompetitive effect of common subcontracting is that prices could be higher because of
higher quality air transport services. If so, we expect common subcontracting to either
increase or not affect passenger traffic (number of passengers transported). However, we find
that common subcontracting decreases traffic, measured by log of passengers transported.
In particular, we estimate the specifications in columns (2) and (3) of Table 5 with traffic
as the dependent variable, and the estimates are in columns (4) and (5) of Table 5. We
find that an IQR increase in common subcontracting leads to a 7.4% decrease in traffic.
Furthermore, before 2004, an IQR increase in common subcontracting decreases traffic by
26
0.6% before 2004, by 4.7% between 2004 and 2012, by 16.9% post 2012.
5 Robustness
In this section, we investigate whether the results in Section 4 are sensitive to how we define
common subcontracting and the concentration in the regional markets.
5.1 Alternative Measures of Common Subcontracting
We consider two alternative measures to assess whether our results are sensitive to our
particular method of constructing common subcontracting. In the first one, we use only the
number of overlapping regional airlines instead of weighting by the fraction of passengers
flown by regional airlines. In the second one, we use the market shares of the major airlines
to construct a weighted measure of overlap in the use of regional airlines.
The first alternative measure, which we refer to as “common subcontracting (count),”
does not depend on the share of passengers transported by a regional airline but on the
number of overlaps across major airlines created by regional usage. This measure is defined
as follows. We determine how many unidirectional major airline pairs are present in each
market and each year. This number is two for markets with only two major airlines; for
markets with three, this number is six, and likewise for other markets. Next, we determine
whether each unidirectional pair has overlapping regional usage and sum the number of
pairs with overlapping regional usage. Then we define the common subcontracting (count)
measure as the ratio of the number of unidirectional major pairs with overlapping regional
usage to the number of total unidirectional major pairs.
For example, suppose AA, DL, and UA are in a market. AA uses a regional airline in
this market, and DL uses that airline in another market. No other regionals are used. Then
there is one unidirectional major pair for which there is overlap out of six total unidirectional
pairs in the market. So the common subcontracting (count) is 1/6 = .167. Like the baseline
27
measure, this count-based measure also takes on values between 0 and 1. In our sample, the
mean of common subcontracting (count) is 0.238 with a high coefficient of variation of 1.26.
We re-estimate the model specification for column (2) in Table 5 using the control function
method with common subcontracting (count) in place of our preferred common subcontract-
ing (baseline) measure. All other variables remain the same. The results are presented in
column (6) of Table 5. The estimates are qualitatively the same as in column (2).
Next, we explore the effects of weighting the overlap across regional airlines by the market
share of a major airline. We refer to this measure as “common subcontracting (weighted).” For
instance, consider the example presented in Table 3. AA uses a regional airline, OO, to trans-
port a 0.667 fraction of its passengers. DL, the other major in the example market, used OO
in another market in the same period. To construct common subcontracting (weighted), we
multiply the fraction of passengers for which AA subcontracts operations to an overlapping
regional airline and AA’s market share. In this example market, AA’s market share is .429,
and DL’s is .571, so the weighted common subcontracting is (.667×.429+0×.571)/2 = .143.
The overall mean value of common subcontracting (weighted) at 0.163 is slightly smaller than
the mean of common subcontracting (count) but has a higher coefficient of variation of 1.62.
We re-estimate the model using the control function method with weighted common
subcontracting. The estimates in column (7) of Table 5 are similar to those in column (2).
5.2 Concentration among Regional Airlines
Our estimates suggest that the overlapping usage of regional airlines raises the price, es-
pecially in the later years in the sample. An alternative explanation for our results could
be that concentration among regional airlines is driving this association: regional airlines
with more bargaining power can extract higher fees from legacy airlines and drive up costs
and, in turn, prices. For example, we observe that the largest regional airline in our sample,
ExpressJet, served at least one segment of 55% of total markets in 2007, and it served 70.5%
in 2016. This increase could either be due to the increasing usage of regionals (see Figure 2)
28
or be the result of M&A activity and consolidation as suggested by Table 1.
To address this question, we calculated what we refer to as Regional HHI. To estimate
the underlying “market” shares for this measure, we divided the number of passengers trans-
ported by a given regional airline by the total number of passengers in a market served by
all regionals. Then we summed the square of those share terms across all regional airlines
active in each market. We find that the interquartile range of concentration stays relatively
constant over the sample and that the median concentration decreases.
To assess the explanatory power of regional concentration, we re-estimate the model
with “Regional HHI” as an additional covariate. The HHI and prices are determined simul-
taneously in equilibrium, so we treat regional HHI as an endogenous variable and use the
Regional Network IVs and Weather IVs as excluded variables. The estimation results are
in column (8) of Table 5. While prices are positively correlated with market concentration
among regional carriers, comparing the estimates in column (8) with (2), we find that the
estimated coefficients on common subcontracting are similar to our benchmark results.
6 Conclusion
In this paper, we investigate a new channel through which firms can charge higher prices:
sharing subcontractors across markets, i.e., common subcontracting. We find that common
subcontracting leads to higher prices and lower air traffic. We also find substantial benefits
to using regional airlines because they lead to lower prices. However, the increasing extent
of common subcontracting in recent years has significantly reduced those benefits.
Thus, there is a need for a nuanced approach to understanding the vertical structure in
the airline industry and how it affects competition. One hypothesis is that sharing regional
airlines across markets allows major airlines to improve information about each other’s costs,
which helps firms collude. However, testing this hypothesis is beyond the scope of our paper.
As such, there are still unanswered questions about what mechanism drives these results
29
and how they generalize to other industries and subcontracting structures. Based on a rich
literature that suggests that information and communication help collusion (Roberts, 1985;
Shapiro, 1986; Kandori, 1992; K¨uhn and Vives, 1994; Athey and Bagwell, 2001; Rahman,
2014; Awaya and Krishna, 2016; Piccolo and Mikl´os-Thal, 2012; Aryal, Ciliberto, and Leyden,
2022), we conjecture that common subcontracting softens competition by providing major
carriers with more information about each others’ costs.
Studying collusive behavior among firms is central to economics (e.g., Harrington, 2017;
Porter, 2005; Kaplow, 2013; Asker and Nocke, 2021). Nevertheless, determining how firms
reach a collusive setting is challenging. Thus, the literature has approached collusion on
a case-by-case basis, and we are cognizant that we raise more questions about the role of
common subcontracting in collusion than provide answers. We hope our results will spark
more interest in understanding of the role of subcontractors on market outcomes.
References
Alchian, A. A., and H. Demsetz (1972): “Production, Information Costs, and Economic
Organization,” American Economic Review, 62(3), 777–795.
Aryal, G., F. Ciliberto, and B. T. Leyden (2022): “Coordinated Capacity Reductions
and Public Communication in the Airline Industry,” Review of Economic Studies, 89(6),
3055–3084.
Asker, J., and V. Nocke (2021): “Collusion, Mergers, and Related Antitrust Issues,” in
Handbook of Industrial Organization, ed. by K. Ho, A. Horta¸csu, and A. Lizzeri, vol. 5,
chap. 12. North Holland: Elsevier.
Athey, S., and K. Bagwell (2001): “Optimal Collusion with Private Information,”
RAND Journal of Economics, 32(3), 428–465.
30
Awaya, Y., and V. Krishna (2016): “On Communication and Collusion,” American
Economic Review, 106(2), 285–315.
(2020): “Information Exchange in Cartels,” RAND Journal of Economics, 51(2),
421–446.
Bernheim, B. D., and M. D. Whinston (1985): “Common Marketing Agency as a
Device for Facilitating Collusion,” RAND Journal of Economics, 16(2), 269–281.
Bernheim, B. D., and M. D. Whinston (1990): “Multimarket Contact and Collusive
Behavior,” RAND Journal of Economics, 21(1), 1–26.
Berry, S. T. (1990): “Airport Presence as Product Differentiation,” American Economic
Review P&P, 79, 394–400.
(1992): “Estimation of a Model of Entry in the Airline Industry,” Econometrica,
60(4), 889–917.
Borenstein, S. (1989): “Hubs and High Fares: Dominance and Market Power in the U.S.
Airline Industry,” RAND Journal of Economics, 20(3), 344–365.
Busse, M. R. (2004): “Multimarket Contact and Price Coordination in the Cellular Tel-
phone Industry,” Journal of Economics & Management Strategy, 9(3), 287–320.
Carlton, D., M. Israel, I. McSwain, and E. Orlov (2019): “Are Legacy Airline
Mergers Pro- or Anti-Competitive? Evidence from Recent U.S. Airline Mergers,” Inter-
national Journal of Industrial Organization, 62, 58–95.
Chicu, M., and N. L. Ziebarth (2013): “Multi-market Contact and Competition: Ev-
idence from the Depression-era Portland Cement Industry,” International Journal of In-
dustrial Organization, 31(5), 603–611.
Ciliberto, F., C. Murry, and E. Tamer (2021): “Market Structure and Competition
in Airline Markets,” Journal of Political Economy, 129(11), 2295–3038.
31
Ciliberto, F., and C. Schenone (2012): “Bankruptcy and Product-Market Competiton:
Evidence from the Airline Industry,” International Journal of Industrial Organization,
30(6), 564–577.
Ciliberto, F., and E. Tamer (2009): “Market Structure and Multiple Equilibria in Arline
Markets,” Econometrica, 77(6), 1791–1828.
Ciliberto, F., and J. W. Williams (2014): “Does Multimarket Contact Facilitate Tacit
Collusion? Inference on Conduct Parameters in the Airline Industry,” RAND Journal of
Economics, 45(4), 764–791.
Clark, R., I. Horstmann, and J.-F. Houde (forthcoming): “Hub-and-Spoke Cartels:
Theory and Evidence from the Grocery Industry,” American Economic Review.
Coase, R. H. (1937): “The Nature of the Firm,” Economica, 4(16), 386–405.
Edwards, C. D. (1955): “Conglomerete Bigness as a Source of Power,” in Business Con-
centration and Price Policy, ed. by Universities-National Bureau, pp. 331–359. NBER.
Eizenberg, A., D. Shilian, and D. D. Blanga (2023): “Estimating the Strategic Effect
of Multi-Market Contact,” Working Paper.
Evans, W. N., and I. N. Kessides (1994): “Living by the Golden Rule: Multimarket
Contact in the U.S. Airline Industry,” Quarterly Journal of Economics, 109(1), 341–366.
Forbes, S. J., and M. Lederman (2007): “The Role of Regional Airlines in the U.S.
Airline Industry,” in Advances in Airline Economics, ed. by D. Lee, vol. 2. Elsevier.
(2009): “Adaptation and Vertical Integration in the Airline Industry,” American
Economic Review, 99(5), 1831–1849.
(2010): “Does Vertical Integration Affect Firm Performance? Evidence from the
Airline Industry,” RAND Journal of Economics, 41(4), 765–790.
32
Gil, R., M. Kim, and G. Zanarone (2022): “Relationships Under Stress: Relational
Outsourcing in the U.S. Airline Industry After the 2008 Financial Crisis,” Management
Science, 68(2), 1256–1277.
Harrington, J. E. (2017): The Theory of Collusion and Competition Policy. MIT Press.
He, L., and G. Kosmopoulou (2021): “Subcontracting Network Formation Among U.S.
Airline Carriers,” Journal of Industrial Economics, LXIX(4), 817–853.
Hortac¸su, A., and C. Syverson (2007): “Cementing Relationships: Vertical Integration,
Foreclosure, Productivity, and Prices,” Journal of Political Economy, 115(2), 250–301.
Kandori, M. (1992): “The Use of Information in Repeated Games with Imperfect Moni-
toring,” Review of Economic Studies, 59(3), 581–593.
Kaplow, L. (2013): Competition Policy and Price Fixing. Princeton University Press.
K
¨
uhn, K.-U., and X. Vives (1994): “Information Exchanges Among Firms and Their
Impact on Competition,” Mimeo, Barcelona: Institut d’Analisi Economica.
Lin, H., and I. M. McCarthy (2019): “Multimarket Contact in Health Insurance: Evi-
dence from Medicare Advantage,” NBER Working Paper No. 24486.
Menne, M. J., I. Durre, B. Korzeniewski, S. Mcneil, K. Thomas, X. Yin, S. An-
thony, R. Ray, R. S. Vose, B. E. Gleason, and T. G. Houston (2012): “Global
Historical Climatology Network–Daily (GHCN-Daily), Version 3.26,” NOAA National Cli-
matic Data Center.
Menne, M. J., I. Durre, R. S. Vose, B. E. Gleason, and T. G. Houston (2012):
“An Overview of the Global Historical Climatology Network-Daily Database,” Journal of
Atmospheric and Oceanic Technology, 29(7), 897–910.
33
Molnar, J., R. Violi, and X. Zhou (2013): “Multimarket Contact in Italian Retail
Banking: Competition and Welfare,” International Journal of Industrial Organization,
31(5), 368–381.
Nocke, V., and L. White (2007): “Do Vertical Mergers Facilitate Upstream Collusion?,”
American Economic Review, 97(4), 1321–1339.
Piccolo, S., and J. Mikl
´
os-Thal (2012): “Colluding through Suppliers,” RAND Journal
of Economics, 43(3), 492–513.
Porter, R. H. (2005): “Detecting Collusion,” Review of Industrial Organization, 26(2),
147–167.
Rahman, D. (2014): “The Power of Communication,” American Economic Review, 104(11),
3737–3751.
Roberts, K. (1985): “Cartel Behavior and Adverse Selection,” Journal of Industrial Eco-
nomics, 33(4), 401–413.
Schmitt, M. (2018): “Multimarket Contact in the Hospital Industry,” American Economic
Journal: Economic Policy, 10(3), 361–387.
Shapiro, C. (1986): “Exchange of Cost Information in Oligopoly,” Review of Economic
Studies, 53, 433–446.
Vives, X. (1990): “Trade Associations Disclosure Rules, Incentives to Share Information,
and Welfare,” RAND Journal of Economics, 21(3), 409–430.
Williamson, O. E. (1979): “Transtion-Cost Economics: The Governance of Contractual
Relations,” The Journal of Law & Economics, 22(2), 233–262.
34
A.1 Appendix: Sample Construction
The main data are from the domestic Origin and Destination Survey (DB1B), which is a 10%
sample of airline tickets from all reporting carriers. This dataset has three versions. First, we start
with the DB1B Coupon version, which is the basis of our analysis. This dataset provides coupon-
specific information for each domestic itinerary of the Origin and Destination Survey, such as the
name of the operating carrier, origin and destination airports, and number of passengers. We use
this dataset to construct the chain of routes each ticketing carrier uses to serve a market, which,
as defined in the paper, is a unidirectional trip between two airports, irrespective of intermediate
transfer points. This dataset also identifies the regional carriers for each nonstop segment (route).
From this dataset, we keep only domestic flights, tickets sold by domestic carriers, flights between
the 48 contiguous United States, and flights that require at most six coupons.
Second, we merge the Coupon version with the DB1B Market version, which contains informa-
tion on the directional market characteristics for each Origin-Destination domestic itinerary. We
exclude bulk fare tickets and tickets with more than three coupons in either direction. Third, we
merge in the DB1B Ticket version containing information on the reporting carrier, itinerary fare,
number of passengers, originating airport, roundtrip indicator, and miles flown. We exclude tickets
whose fare credibility is questioned by the DOT.
We further exclude tickets that are neither one-way nor roundtrip travel or include travel on
more than one airline on a directional trip (known as interline tickets), here identified by whether
there was a change in the ticket carrier for the ticket.
Next, we follow the approach in Borenstein (1989) and Ciliberto and Williams (2014) and
consider a roundtrip ticket as two directional trips on the market, and the fare for each directional
trip is half of the roundtrip fare. A one-way ticket is a one-directional trip. From the final sample,
we exclude tickets with fares less than $20 in the top and bottom one percentile of the year-quarter
fare distribution and for which the fare per mile (the yield) in the top and bottom one percentile
of the year-quarter yield distribution.
35
A.2 Appendix: Weather
In this section, we discuss the construction of our Weather IVs using airport-level weather conditions
data from the Daily Global Historical Climatology Network (Menne, Durre, Vose, Gleason, and
Houston, 2012; Menne, Durre, Korzeniewski, Mcneil, Thomas, Yin, Anthony, Ray, Vose, Gleason,
and Houston, 2012). We use four of the “core” variables in the data: precipitation (in tenths of
mm), snowfall (mm), snow depth (mm), and minimum temperature (tenths of degrees celsius).
These variables are recorded for each airport at the daily level. For our purposes, we average across
all days in a quarter to arrive at average quarterly levels. Summary statistics across airports and
years for these variables are presented in the Table (A.2.1). To construct the weather instruments
Table A.2.1: Summary Statistics of Weather Variables
Variable Mean Median Std. Dev.
Precipitation (tenths of mm) 26.46 24.84 16.70
Snowfall (mm) 1.30 0.00 3.52
Snow Depth (mm) 5.77 0.00 23.96
Minimum Temp (tenths of degrees celsius) 95.94 101.04 83.86
Note: Summary table of the weather variables.
we implement the following steps. First, we consider the quarterly average weather variables for
each airport that is used by a major to serve a market in the given year. For example, if AA
serves the market CHO-DFW by using the route CHO-CLT-DFW, then we consider the weather
variables at all three airports: CHO, CLT, and DFW. At the market-major-year quarter level, for
each variable listed above, we take the most extreme value (maximum in the case of precipitation,
snowfall, and snow depth; minimum for minimum temperature) of all airports used along the routes
the major uses to serve the market. For instance, if DFW has the worst precipitation of CHO,
CLT, and DFW, we take DFW’s average rain value. This gives us a set of four variables for each
major in a market-year that record the average weather conditions of the “worst” airport used to
serve the market for each of the weather variables. Then for each major in a market, we sum its
competitors’ extreme weather values in that market to determine the weather IVs used for common
subcontracting, as common subcontracting is a function of competitors’ regional usage in a market.
For regional share of airline j, we use the four most extreme values of weather encountered by j
itself in the market, as weather conditions are a factor that affect airlines’ subcontracting choices.
36