mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
RAND Journal of Economics
Vol. 33, No. 2, Summer 2002
pp. 298–318
Firm financial condition and airline price
wars
Meghan Busse∗
A firm that knows that cutting price may trigger a price war must weigh present versus future gains
and losses when considering such a move. The firm’s financial situation can affect how it values
such tradeoffs. Using data on 14 major airlines between 1985 and 1992, I test the hypothesis that
firms in worse financial condition are more likely to start price wars. Empirical results suggest
that this is true, particularly for highly leveraged firms. The article also explores which firms join
existing price wars and finds that a firm is more likely to enter a price war the greater the share
of its traffic on routes served by the price-war leader.
1. Introduction
Economists’ explanations for price wars differ from those of other observers of the airline
industry. Most economic models of price wars, which apply more generally than to the airline
industry alone, have emphasized the role of fluctuations in demand. Changes in demand alter
the expected profitability of undercutting a tacitly collusive equilibrium; depending on the assumptions made, the models predict that price wars occur either when demand booms or when it
slumps.
Industry insiders, meanwhile, identify the financial troubles of an individual carrier as an
important motivation in initiating the fare cuts that trigger price wars. For example:
[Mark Daugherty, airline industry analyst for Dean Witter] said weaker airlines are willing to risk losses with low ticket
prices in order to raise badly needed cash. “That is what is driving a lot of these companies,” said Daugherty. (Los Angeles
Times, February 13, 1991, p. 4.)
This comment and others like it suggest that firms’ financial situations might play an important but neglected role in determining price wars. In light of these observations, this article asks
whether price wars in the airline industry stem in part from the financial conditions of individual
firms, and, in particular, whether financially troubled firms are more likely to start price wars.
Several results from the corporate finance literature lend theoretical support to the industry observations. In particular, some models indicate that firms in poor financial condition discount future returns more heavily than do financially sound firms. This suggests that, as a result,
∗
Yale University; [email protected].
I am grateful for helpful contributions from Andrew Bernard, Severin Borenstein, Bhanu Narasimhan, Kevin Neels,
Whitney Newey, Sharon Oster, Nancy Rose, Florian Zettelmeyer, and workshop participants at MIT, Yale, Berkeley,
Northwestern, Chicago, Stanford, and the U.S. Federal Trade Commission. I also thank the Editor and two anonymous
referees for valuable suggestions. I am especially grateful to Nancy Rose and Severin Borenstein, who provided me with
some of the data used in this article, and to Asra Nomani and Bridget O’Brian of the Wall Street Journal. Michael Shonborn
provided valuable research assistance.
298
Copyright © 2002, RAND.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
BUSSE
/
299
financially troubled firms will more willingly pursue strategies that raise current returns at the
expense of lower returns in the future; for example, a firm might cut prices in order to boost
short-term demand.
Analyzing the role of firms’ financial conditions in airline price wars not only furthers
understanding of the airline industry itself, but also sheds light on two more general issues. First,
it addresses whether individual firm characteristics, in addition to market demand fluctuations,
contribute to price-war behavior. Second, it offers further evidence on the link between a firm’s
financial condition and its conduct in product markets.
The article proceeds by reviewing, in Section 2, the theoretical literature on price wars,
financial motivations for product market behavior, and the empirical literature on airline price
wars. Section 3 describes the data on airline price wars and financial characteristics used. Section
4 introduces the empirical model and discusses some estimation issues. Section 5 presents and
discusses the results. Section 6 concludes.
2. Background
Price wars. A price war is a period in which the firms in an industry or market set prices
that are significantly below the usually prevailing prices, generally implying a change in strategy
within a set of oligopolists. In game-theoretic terms, this can be modelled as a temporary period of
noncooperative behavior among players whose normal course is a (tacitly) collusive equilibrium.1
Models of price wars fall broadly into three classes, distinguished by when and why price wars
occur.
In the first class, price wars occur in periods of low demand because of firms’ cost structures.
For example, Staiger and Wolak (1992) contend that a firm facing capacity constraints will have
an incentive to cut its price only when it has idle capacity, which is more likely when demand is
low, while Scherer (1980) argues that a firm will have an incentive to cut prices when demand is
low if it has high fixed costs to cover. In the second class of models, price wars occur because a
firm cannot observe its competitors’ prices, and therefore it interprets a fall in demand for its own
output as a sign that one of its competitors has offered customers a secret price cut. A stochastic
period of low overall demand will thus prompt a retaliatory price cut, instigating a price war
(Green and Porter, 1984). In the third class of models, price wars occur when demand is high
and the immediate rewards from undercutting competitors are therefore largest (Rotemberg and
Saloner, 1986; Haltiwanger and Harrington, 1991). In this last class of models, a price war is
not a shift from collusive to noncooperative behavior. Accordingly, Ellison (1994) suggests that
“countercyclical pricing” more accurately describes what is captured by these models than “price
wars.” Ellison argues further, however, that if these models were extended to an environment of
uncertain demand, they would indeed predict price wars to occur when demand booms.2,3
In all these models, demand fluctuations, which are common in the airline industry for both
seasonal and cyclical reasons, provoke price wars. But not all the models describe the airline
industry well. The secret price deals of Green and Porter, for example, are unlikely to occur in
this industry because of computerized reservation systems and a large and dispersed customer
base. The first class of models most closely matches the characteristics of the airline industry,
including capacity constraints and high fixed costs.
Financial condition. The observations, cited in the Introduction, that a carrier’s financial
troubles may prompt it to initiate a fare war have support in theoretical results showing that a
firm’s financial condition affects its valuation of future payoffs and thus its actions. This result
1 This does not apply to a predatory price war, in which prices fall because of a strategy on the part of one (or
more) of the firms in the market to drive another firm out of the market.
2 Bagwell and Staiger (1997) offer a more general model of pricing over the business cycle in which the cyclicality
of prices depends on whether demand is positively or negatively correlated over time.
3 Athey, Bagwell, and Sanchirico (2000) present a model in which collusion is affected by privately observed cost
shocks to each firm. While their emphasis is on establishing conditions for price rigidity, their article is similar to the
present work in emphasizing change in firm-level characteristics as the determinant of the nature of collusion.
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
300
/
THE RAND JOURNAL OF ECONOMICS
has been framed in several different contexts. For example, Jensen and Meckling (1976) argue that
because equityholders are residual claimants who realize gains only in good states, they prefer
activities that raise payoffs in good states, even at the cost of lowering payoffs in bad states.4
Stiglitz and Weiss (1981) identify a similar principle in the context of credit rationing. In a related
vein, Brander and Lewis (1986) consider firms engaged in Cournot competition that must commit
to output levels before observing demand. A highly leveraged firm will compete aggressively by
producing the high level of output that is optimal in high-demand states, since only then does it
realize a nonzero residual claim. Chevalier (1995) and Chevalier and Scharfstein (1995, 1996)
provide empirical evidence that a firm’s financial condition can affect the strategy it pursues.
A firm that initiates a price war makes a tradeoff similar to those described above, raising
current profits in exchange for lower profits in the future, and possibly lower overall profits. There
are two reasons for a financially distressed firm to make this tradeoff: “bagging the bondholder”
and “gambling.” If the firm is likely to go bankrupt in the future, the current owners and managers
have reason to disregard what happens to the firm after they have lost control. They will prefer
to shift returns from the future into the present, even if doing so lowers the firm’s overall returns,
leaving the bondholder “holding the bag.”
Alternatively, if the current owners or managers can forestall bankruptcy by boosting shortterm profits at the expense of the long term, they have an incentive to do so if there is any chance
that the firm’s situation will improve in the future. They are no worse off by taking this “gamble,”
since bankruptcy costs are borne by the debtholders.
The remaining question is why a fare cut that might instigate a price war would help a
financially troubled airline to shift returns from the future to the present. One way a carrier could
benefit from a price war at the expense of its competitors would be if cutting price allowed it to
increase market share because competitors reacted with a lag or not at all.5 In practice, airlines
appear to respond to fare changes very quickly, usually within a day, although there may remain
a first-mover advantage if consumers are more likely to be aware of low prices at the airline that
initiated the fare cut, placed ads first, attracted travel agents’ attention, and so on.
It seems more likely that airlines benefit because price wars allow them to “borrow” demand
from future periods. This will be the case under the plausible assumption that some consumers are
willing to purchase tickets sooner than they had planned if they are offered sufficiently attractive
prices. There is certainly industry discussion of this as one of the major effects of price wars.
Bhattacharya (1997) formalizes this in a model of “survival price wars” in which a financially
distressed airline will cut prices in order to boost current period demand, even if it knows that its
competitor will match the price.
Price wars in the airline industry. The existing empirical literature on airline price wars
that is most similar to the present work is Morrison and Winston (1996), who consider a number
of different explanations simultaneously, concluding that price wars are more likely if demand
expectations—as measured by GDP—are incorrect, and less likely if one of the carriers serving
a route is bankrupt. Although the aim is similar, my article’s approach differs from Morrison and
Winston’s in considering financial motivations and in looking separately at price war initiators.
Brander and Zhang (1993) study airline price wars using a conjectural variations approach to
characterize particular duopolists’ interaction as quantity setting with periodic switches between
collusive and noncollusive behavior. However, they do not investigate the cause of the regime
switches. Finally, Borenstein and Rose (1995) focus on the effect of bankruptcy on airlines’
pricing strategies, although not on price wars directly. Examining major bankruptcies between
1989 and 1993, they find that bankrupt carriers’ average prices are 5% lower than industry trends
for several months before bankruptcy, although at the time of bankruptcy there is little change in
prices. This they interpret to suggest that “financial distress rather than bankruptcy per se may be
responsible for observed changes in an airline’s price preferences” (p. 399).
4 Adding agency and career concerns to this model can temper this effect by making the risky activity less attractive
to managers even if it is in the shareholders’ interests.
5 This is the incentive for cutting price in the model of Rotemberg and Saloner (1986).
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
BUSSE
/
301
3. Data
The price data that studies of the airline industry usually use come from Databank 1A of the
U.S. Department of Transportation’s Origin and Destination (O&D) Survey. This survey is a 10%
ticket sample that reports route and fare information by quarter. Morrison and Winston (1996)
and Brander and Zhang (1993) identify price wars from this data by defining classification rules
based on either the raw price data or econometric results.
The Department of Transportation (DOT) collects these data when a ticket is used, not when
it is purchased, which causes several problems in using the data to identify price wars. The first
problem is that the fares available in most fare wars are applicable for travel dates long after
the fare war itself ends. As a result, the evidence of a fare war will not necessarily show up
in the price database as a sudden concentration of very-low-priced tickets. Instead, the tickets
purchased in a fare war become observations in Databank 1A piecemeal as travellers use them.
Second, observations at any point in time contain both tickets of passengers who purchased during
a price war and tickets of passengers who did not. Third, the data are available quarterly, which
is a long interval relative to the length of most price wars.
The effect of the data collection method is to both dilute the evidence of a fare ware and
to smear it over time. This effect is evident in Morrison and Winston (1996). They define price
wars to begin when average fares fall by a certain percentage and to end when fares rise by any
amount. By this definition, they find that price wars last 1.8 quarters on average, with a modal
length of two quarters. As evidenced in the newspaper ads that usually accompany fare wars,
price reductions rarely last that long. More typically, airlines offer low, price-war fares for only a
few weeks.
In addition to complicating the identification of price wars, the collection of fare data by date
of travel makes it impossible to determine who initiated the price war. Since competitors usually
respond to a price cut within days, Databank 1A almost certainly will not show an initiator’s fares
dropping ahead its competitors’ fares. While this is of little consequence in testing hypotheses
about the timing of price wars relative to demand cycles, it is critical to identify the price-war
initiator in testing hypotheses in which the incentives of individual firms to initiate a price war
differ across firms and over time.
This article considers just such a hypothesis, namely, that financially distressed airlines
are willing to risk a price war in order to cut prices when other carriers would not. Existing
empirical work on price wars (Porter, 1983; Ellison, 1994; and Morrison and Winston, 1996) has
not distinguished between firms that led or followed in price wars. This may have been motivated
both by the theoretical emphasis on demand fluctuations in explaining price wars and by the
unavailability of appropriate data.
Identification of price wars. To avoid the shortcomings of using Databank 1A, I identify
price wars by using reports in the press. As a result of the scrutiny that the airline industry receives,
not only the airlines but also travel agents, analysts, reporters, and even the ticket-buying public
seem to agree on when a price war is occurring. This article takes advantage of that attention and
consensus by using the Wall Street Journal to identify price wars.6
I used the Wall Street Journal Index, an annually published topical guide, to direct me to
articles in the Wall Street Journal itself that contained information on price wars. For each topic,
the Index lists a summary of each corresponding article in the Wall Street Journal.7 The relevant
Index topics for airline price wars were “Airlines” and “Airlines–Rates” until 1990 and “Airlines–
United States” thereafter.
I read each summary under these topics looking for articles that described price wars. Key
6 Scott Morton (1997) uses a similar methodology, identifying price wars in British shipping cartels from shipping
company histories.
7 The listings in the Index appear to be exhaustive. For example, small blurbs called “Tapewatch” are cited in the
Index (usually in their entirety because they are so short). Similarly, articles in which airlines are a secondary topic are
nevertheless cited under the airline section.
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
302
/
THE RAND JOURNAL OF ECONOMICS
phrases indicating such articles included “fare cut,” “fare war,” “discount fares,” “price war,” “fare
reduction,” “slashed fares,” “price promotion,” “price cut,” and so on. I then read each original
article whose summary contained these or related phrases. I erred on the side of caution, looking
up an article if there was any evidence in the Index summary that it described a price war.8 The text
of the articles describing price wars typically listed who had initiated the action and the response
of competing carriers, as well as the extent of the routes affected and an indication of the size of
the price cut.9
To understand the Wall Street Journal’s coverage, I also spoke to two of the reporters who
wrote many of the articles about price wars during the sample period. According to them, reporters
during this period employed multiple sources of information to monitor price-war activity. For
example, when an airline cut fares, it usually sent out a press release over newswire services and
often placed advertisements in the Wall Street Journal and other major newspapers. Additionally,
one of the reporters checked daily for unusual numbers of filings with the Airline Tariff Publishing
Company, which does the back-end work of collecting and compiling fare data for airline
reservations systems. The reporters also had supplemental information from personal contacts
among airline pricing officials. Both reporters said it was standard procedure to contact all the
major carriers about their intended responses, including additional checks and supplementary
articles in the days following an initial price-cut announcement. One of the reporters, whose
specific assignment during four years of the sample period was airline pricing, said the coverage
during the 1985–1992 period was very aggressive; both reporters were confident that virtually
all price wars were covered. Ultimately, the features of the industry that make secret price cuts
infeasible—rapid, reliable, and easily available price information—also make it possible for the
reporters to be confident that they have identified all major price wars.
I had to use some discretion in deciding, based on the textual description, what should be
categorized as a price war for the purposes of this article. My aim was to identify substantial,
temporary drops in price affecting a broad set of routes. For example, the Index entry for the first
fare war covered in this article reads as follows:
American Airlines plans new fares on more than 2,400 routes that are about 70% below coach fares; the move was
immediately matched by a number of competitors, sending airline stocks sharply lower.
I found no instances of price cuts that went unmatched by another carrier. This could be either
because price cuts are never unmatched, or because such an event is not deemed newsworthy.
Evidence against the latter is given, for example, in a pair of articles from June 1986 in which
the first article mentions only Continental Airlines’s unilateral action, followed by TWA in a
subsequent article a few days later. If unilateral action is nevertheless not always covered, my
dependent variable in the regressions that follow will be measured with error, biasing the results
against statistical significance.
In understanding what I categorized as a price war, it may be helpful to understand what
kinds of events I did not consider to meet that definition. Instances of price decreases that were
not categorized as price wars included the following:
(i) Price cuts on limited routes (for example, routes only to Florida, or routes only out of
Chicago or Newark).10
(ii) Cuts in fares that are not applicable to most travellers (for example, senior citizen or
business class fares).
8 I chose this method over an electronic search because there are many terms used to describe price wars. It would
be difficult to guarantee that no price wars were missed in an electronic search.
9 An Appendix containing the Index entries and selected text from all the articles that contributed to the positive
identification of a fare war is available at www.rje.org.
10 These limited price wars are excluded because they appear to be prompted by very specific events, such as the
entry of a new airline. The aim of this article is to consider the general, widespread price wars that periodically grip the
industry, which seem likely to have very different causes than specific, localized price wars.
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
BUSSE
/
303
(iii) Price cuts on international routes (primarily routes to Europe).11
(iv) Price cuts for travel on very selected days (for example, travel restricted to Thanksgiving
Day).
(v) An easing of restrictions (for example, a seven-day instead of fourteen-day advance
purchase requirement for a particular class of fares).
In addition, I would have excluded price cuts had there appeared to be industry consensus that
they were necessary adaptations to changed conditions. Price increases of this sort are common
when, for example, the industry is faced with rising fuel costs. However, the airline industry does
not appear to coordinate on price decreases in the same way; there were no cases of price decreases
in which carriers’ actions reflected or seemed designed to seek consensus.
Based on the information contained in the articles, I created a series for each carrier that
indicates for each month whether a firm initiated a price war, whether it joined an existing price
war, or whether it had no participation in a price war.12 I collected data for the 14 largest firms
during the period 1985–1992.13 During those eight years there were 31 price wars. On average,
just over half the carriers participated in any given price war. All the carriers except Piedmont,
Republic, and Western were responsible for starting a price war at some point during the sample
period.
Other data. Data on airline traffic and on airline financial characteristics, which come from
publications of the U.S. DOT, supplemented the data on the occurrence of price wars. Traffic data
are reported monthly and are for domestic operations. Financial data, which are available only
quarterly, are matched to the corresponding month.
Limitations of the data. Although this approach is better able to time price wars precisely
and to distinguish leaders and followers, it also has limitations. One, already described in detail, is
that the classification of price wars is ultimately based on the comparison of a verbal description
of an event to a verbal definition of a price war. This makes the definition of a price war somewhat
arbitrary, in the sense that changes in conduct may not match perfectly with changes in the pricewar measure. A second limitation is that the binary measure necessarily misses both the depth
of the price cut and its availability in terms of the number of seats offered at that price. A final
concern is that despite the efforts of the reporters, some price wars or the participation of some
carriers may not be reported. I have tried to mitigate this concern by restricting my sample to
major carriers and by using dummy variables in the regression to capture any differences between
the reported activities of large and small carriers attributable to size alone. Although the data used
in this study have valuable advantages, it is important to keep these limitations in mind.
4. Model
If price wars are started by airlines that reduce prices in response to financial pressures, then
those that initiate the price reductions and those that later match them may have very different
reasons for their choices. In light of this, this article splits the analysis of price wars into the
decision to lead a price war and the decision to follow, or to match the price reductions of other
carriers. The primary emphasis is on the decision to lead a price war, and on the role that a firm’s
financial condition plays in that decision.
The probability that a given firm will start a price war in a given quarter is modelled as a
11 Fare wars on international routes were excluded for lack of data on the competing, non-U.S. carriers that also
served those routes, and also because fares set by U.S. carriers on those routes can be (and are) rejected by foreign
governments. Furthermore, international fare wars do not spill over into the domestic market.
12 There are months in which two carriers are identified as price-war leaders. There are two reasons for this.
Occasionally, two price-war events occur within a month. On other occasions, carriers are reported as acting simultaneously,
especially Continental and Eastern, which were both units of TexasAir during part of the sample.
13 A chronology of price wars, including who were leaders and followers, is also available at www.rje.org.
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
304
/
THE RAND JOURNAL OF ECONOMICS
function of three types of variables: demand, the financial characteristics of the firm, and controls:
Pr(Leadit = 1) = F(Demandit , Financial Measuresit , Controlsit ).
(1)
Equation (1) is estimated using a probit model, although, as noted later, estimating it with a linear
probability model produces very similar results.
Demand. The relevant demand for the firm’s decision could be either demand for the firm’s
own output or overall industry demand. The theoretical literature often assumes that firms offer
homogeneous products and face a common industry demand curve, making demand for each
firm some fraction of the total industry demand (Rotemberg and Saloner, 1986; Haltiwanger
and Harrington, 1991). In practice, it is not likely that consumers perceive service on different
airlines to be truly homogeneous. Airlines differ in scheduling and perceived quality, and because
of loyalty-inducing practices such as frequent-flyer programs. Furthermore, it seems likely that
firms can observe demand for their own product more easily than overall industry demand. Firms
may thus use their own demand as a signal for total industry demand (Green and Porter, 1984).
In the regressions, I measure demand with both industry- and firm-level enplanements,
which is the number of passengers carried in a given month. The firm-level measure of
demand is normalized enplanements, which is a firm’s enplanements in a month divided by
average enplanements for that firm over the period. Industry demand is measured by aggregate
enplanements, which is the sum of enplanements for all carriers in the period.
Since a firm’s intention in cutting price is to increase demand, enplanements is an endogenous
variable. I thus estimate (1) using instrumental variables. When (1) is estimated as a linear
probability model, this is straightforward using two-stage least squares. When (1) is estimated with
a probit model, a more involved, two-step procedure must be used. Summarizing briefly results
shown by Newey (1987), one can consistently estimate the parameters of a probit model that
has endogenous explanatory variables by first regressing the endogenous explanatory variables
on a valid set of instruments and the exogenous explanatory variables, and then including the
residuals from this regression as variables in the probit model. The probit is then estimated as
usual. The parameters estimated for the residuals are nuisance parameters, and the remaining
parameters are estimated consistently. As shown by Newey and McFadden (1994), one must alter
the usual formula for calculating estimates of standard errors to account for the fact that some
of the variables in the second-step probit are derived from the first-step regression. In all probit
results reported in this article, this correction has been made, following formulas given by Newey
and McFadden (1994).
The instruments used for enplanements are departures and a trio of regional macroeconomic
variables. Departures (the number of planes that leave the ground) are correlated with
enplanements (the number of passengers who board planes) but are unlikely to vary with price-war
behavior for several reasons. First, airlines set their schedules far in advance of the actual flight;
most airlines begin selling tickets six months before a flight. Furthermore, airport slot restrictions,
timing constraints imposed by hub-and-spoke networks, and the necessity to physically coordinate
with other airlines who use the same airports and flight routes mean that departures cannot be
changed as quickly as fares. There are prevalent anecdotal accounts of airlines cancelling flights
in response to market conditions, although the accusation is more commonly that they cancel
empty flights than flights for which fares are low. In the data used in this article, the occurrence
of a price war has virtually no predictive power in explaining the number of departures, which
gives some empirical justification for the use of departures as an instrument.
A trio of macroeconomic measures based on unemployment, output, and income are also
used as instruments. Demand for both major types of travel, business and leisure, varies with
macroeconomic conditions. For example, a 1993 Gallup Survey reported that 49% of travel in
the United States was for business, 35% to visit a friend or relative. Of the business travellers,
the survey noted that respondents most often gave an increase in new business or better company
financial condition as the reason for increased business travel. Those reporting less business travel
tended to attribute it to decreasing business or deteriorating company financial conditions (Air
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
BUSSE
/
305
Transport Association, 1993). Since enplanements, the endogenous measure, varies across carriers
as well as over time, the instruments ideally should reflect the different economic climates each of
the carriers faces because of differences in their regional presences. I construct suitable instruments
by weighting state-level measures of real per-capita personal income, real per-capita gross state
product, and the rate of unemployment to reflect the regional makeup of each carrier’s route
network. The weights are based on passengers whose flights originate at the 26 largest airports in
the country. About three-quarters of flights originate at one of these so-called large hub airports.
The state weight for a given carrier is the share of its total large hub traffic in the sample period
that originates from large hub airports in that state. Weights do not vary over time so as not to
reintroduce endogeneity between price wars and the number of passengers. For example, the
weighted unemployment measure for carrier i at time t would be
j
Pik
· U jt ,
Pi
k∈K
(2)
j
where j indicates the state, k indicates the airport, K j ⊂ K = {1, 2, . . . , 26} indicates
the
airport(s) in state j, U jt indicates the unemployment rate in state j at time t,
P
=
t Pikt
ik
indicates the total number of i’s passengers originating at airport k, and Pi = k Pik is carrier
i’s total number of large hub passengers. The result is three variables—route-weighted personal
income, output, and unemployment rate—reflecting the macroeconomic climate that each carrier
faces.14
It is particularly valuable to have both departures and the macroeconomic variables as
instruments. Departures are correlated with enplanements not only because they determine
capacity, but because airlines can adjust departures to reflect their forecasts of demand. The
latter feature means that departures will be correlated with changes in demand that are
predictable. In contrast, macroeconomic fluctuations are correlated with changes in demand that
are unpredictable.
Financial measures. The question of how best to measure a firm’s financial distress or
risk of bankruptcy is an open issue in the accounting and finance literature. A method proposed
by Altman (1968) relies on a combination of financial ratios, including measures of leverage,
liquidity, and profitability. Since Altman, there have been a number of alternative proposals,
including measures based on cash flow and on stock market returns or variability of returns.
However, Mossman et al. (1998) find that Altman’s method of financial ratios performs at least
as well as more recent proposals. This article uses financial measures that reflect several different
aspects of a firm’s financial situation. The six measures include leverage ratios, interest coverage,
liquidity measures, and operating profitability. Accounting definitions of the financial measures
are given in Table 1. The long-term debt ratio measures the share of a firm’s financing that comes
from long-term debt arrangements. The leverage ratio measures the share of financing that comes
from all debt sources. Interest coverage measures the number of times that earnings would cover
interest payments. The current ratio measures the ratio of current assets (assets that are expected
to be converted into cash within a year) to current liabilities (liabilities due within the year).
The quick ratio measures the ratio of extremely liquid, cash-like assets to current liabilities. The
operating ratio measures the ratio of operating profits to expenses.
While the first three measures reflect the underlying solvency of the firm, the last three
measure the more immediate liquidity of the firm. In an overview of empirical analyses of financial
distress, Foster (1986) identifies leverage ratios and payment coverage ratios as consistently being
among the significant predictors of financial distress and bankruptcy. He noted that liquidity and
activity-related ratios are generally of limited power in distinguishing distressed or bankrupt firms.
14 Not all the elements of the route-weighted macroeconomic measures are available with the same frequency.
State unemployment rates are available monthly; personal income is available quarterly; gross state product, airport
traffic, population, and deflators are available annually.
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
306
/
THE RAND JOURNAL OF ECONOMICS
TABLE 1
Definitions of Financial Ratios
Financial Measure
Definition
Long-term debt ratio
Long-term debt + Noncurrent obligations under capital lease
Long-term debt + Noncurrent obligations under capital lease + Net stockholders’ equity
Leverage ratio
Total equity + Net stockholders’ equity
Total equity
Interest coverage
Operating profit + Nonoperating expense + Depreciation
Interest expense
Current ratio
Total current assets
Total current liabilities
Quick ratio
Cash and short-term investments + Accounts receivable + Prepaid expenses
Total current liabilities
Operating ratio
Operating profit
Operating expense
Therefore, though both the solvency and liquidity aspects of the firm are pertinent in the concept
of financial distress, previous empirical work indicates that leverage and interest coverage may
be the most likely to be empirically significant.
The regressions incorporate financial measures both linearly and as dummy variables
indicating where in the distribution the measure falls. To benchmark where those threshold values
fall, Table 2 reports the distribution of the ratios within the sample. In addition, the regressions
instrument for financial measures with lagged financial measures. The need to instrument arises
because a firm’s involvement in a price war will affect the firm’s cash flows in that period, and
therefore it may affect current financial measures.
Another indicator of financial condition is bankruptcy. In the airline industry, bankrupt carriers
have been blamed as a destabilizing influence.15 Although bankruptcy is clearly associated with
financial distress, declaring bankruptcy actually has the effect of reducing the pressure on a firm
to meet its financial obligations. The results of an empirical study by Borenstein and Rose (1995)
examining the effect of bankruptcy on the overall level of airline prices reflect this ambiguity.
They find that on average, a carrier’s prices dip just before it declares bankruptcy, then recover
some after the firm has declared bankruptcy; competitors’ prices appear to change very little or
rise slightly over the same period.16 The regressions in this article include a dummy for whether
the firm is in bankruptcy during the month.
Controls. As controls I use dummies for each of the eight years, for each of four seasons
(quarters), and for being one of the three largest or three smallest firms in the industry. The latter
variables are included in an effort to control for possible over- or underreporting of price-war
leadership by the Wall Street Journal. They will also pick up any difference in price-war behavior
as a result of asymmetries in firm size. The largest firms during the sample period are American
Airlines® , Delta Air Lines, and United Air Lines; the smallest are America West Airlines® , Pan
Am, and Western Airlines.
A firm that is considering cutting its price may take into account the likelihood that other
firms will follow its initiative. As explored in greater detail in the next section, how likely a firm
is to respond to an initiator’s price cuts depends on the share of the respondent’s routes it shares
with the initiator. In the extreme, a firm that was a monopolist on all its routes would have little
15 “Desperate for cash, the bankrupt airlines have often started fare wars. A study this spring by Aviation Forecasting
and Economics, an airline consulting group, concluded that ‘in addition to their own losses, the bankrupt carriers have
directly cost the rest of the U.S. scheduled airline industry at least $3 billion over the last three years.’ ” (New York Times
Magazine, September 5, 1993, p. 37.)
16 Another reason for fares on bankrupt carriers to be low is to persuade consumers to buy tickets on an airline that
may cease to exist.
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
BUSSE
TABLE 2
Sample
Statistics
/
307
Distribution of Financial Ratios
Long-Term
Debt Ratio
Leverage
Ratio
Interest
Coverage
Current
Ratio
Quick
Ratio
Operating
Ratio
90%
1.033
1.245
17.258
1.221
1.063
.123
75%
.834
.959
7.725
.968
.792
.082
50%
.470
.770
2.993
.767
.627
.022
25%
.317
.629
.680
.635
.504
−.050
10%
.145
.560
.000
.526
.398
−.105
Mean
1.120
.841
6.987
.865
.712
.009
11.030
.294
13.554
.578
.483
.102
Standard deviation
reason to expect a response to its price changes. To capture the risk of response a firm faces, I
calculate a competition measure based on the sum of the shares of other carrier’s traffic on routes
served by each firm. Specifically, the “potential followers” measure for carrier i in period t is
P jr t
· I (i, r, t) ,
P jt
r
j=i
(3)
where P jr t is the number of j’s passengers on route r in period t, P jt = r P jr t is j’s total traffic
in period t, and I (i, r, t) is an indicator that carrier i serves route r in period t. Since the currentperiod number of passengers may be correlated with price-war behavior, which is the dependent
variable, a one-period lag of this measure is the variable actually included in the estimations that
follow.
5. Results and discussion
As described in Section 2, previous empirical literature on price wars has been concerned
primarily with the effect of demand on price wars. For the purpose of comparison with earlier
results, this section begins by estimating a specification reminiscent of earlier work. The result
obtained by estimating a model that does not separate decisions to lead and follow in a price war
and does not include financial characteristics is that the level of demand plays a role in predicting
price wars, as found in previous research.
As the section goes on to show, however, when the decision to initiate a price war is considered
separately, and when financial characteristics are included, demand no longer plays a significant
role. Instead, financial condition, especially as measured by leverage and interest coverage, has a
significant effect on price-war behavior.
Section 5 concludes with a brief examination of the decision to follow in a price war, finding
that an important consideration in this case is the extent to which the routes of the following firm
overlap with those of the leader.
Comparison to previous price war results. In keeping with previous analyses of price
wars, this subsection examines the timing of price wars relative to demand fluctuations. It differs
from the rest of the results reported in the article in two ways: financial measures are not included
as explanatory variables, and the dependent variable does not distinguish between whether the
firm is a price-war leader or follower, but only indicates whether it is a participant.
Pr(Participateit = 1) = F(Demandit , Controlsit ).
(4)
Equation (4) is estimated by the two-step probit model described in Section 4. The instruments
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
308
/
THE RAND JOURNAL OF ECONOMICS
TABLE 3
Instrument Regression
Firm-Level
Enplanements
Industry-Level
Enplanements
Firm-level departures
.895
(.018)
.023
(.035)
Industry-level departures
.082
(.021)
.842
(.040)
Route-weighted unemployment
−.024
(.007)
.008
(.013)
Route-weighted personal income
−1.279
(.865)
.257
(1.671)
Route-weighted gross state product
2.521
(1.195)
−.298
(2.310)
Aggregate unemployment
.083
(.021)
.170
(.041)
Aggregate personal income
−.000
(.000)
−.0004
(.00009)
Aggregate gross domestic product
.000
(.000)
.0002
(.00007)
Potential followers
.010
(.003)
−.002
(.006)
1 if bankrupt
−.033
(.013)
.005
(.026)
1 if among three largest carriers
−.036
(.014)
.006
(.027)
1 if among three smallest carriers
.024
(.013)
−.009
(.024)
Second quarter
.079
(.014)
.289
(.026)
Third quarter
.102
(.212)
.380
(.041)
Fourth quarter
.043
(.032)
.250
(.062)
.876
.843
Adjusted R 2
Notes: Dependent variable is either firm-level or industry-level enplanements, as
specified. OLS regression. Year dummies included in all specifications, 1,023
observations. These results reported without lagged financial measures instruments,
which vary by specification.
for the endogenous demand measures are departures and route-weighted macroeconomic measures, also described in Section 4. Table 3 reports the first stage, instrument regression. The results
of this estimation are reported in Table 4. Each cell in the table contains three entries. The
estimated probit coefficient is reported in the middle entry and its standard error in the bottom
entry, in parentheses. The top entry, in italics, is the marginal probability effect implied by the
estimated coefficient. For continuous variables, this should be interpreted as the rate of change in
the probability of the firm participating in a price war in response to an incremental change in the
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
BUSSE
TABLE 4
/
309
Participate Decision
Firm-level enplanements
Industry-level enplanements
1 if among three largest carriers
.114
.481
(.168)
−.118
−.499
(.201)
.086
.336
(.126)
1 if among three smallest carriers
−.093
−.470
(.110)
Second quarter
−.087
−.410
(.084)
Third quarter
−.051
−.228
(.104)
Fourth quarter
−.036
−.160
(.083)
Notes: Dependent variable is one for firm-months in which
the firm participates in a price war, zero otherwise. Two-step
probit estimation, instruments are firm-level and industry-level
departures as appropriate, and three weighted macroeconomic
measures. Year dummies included in all specifications, 1,023
observations. First entry is marginal probability, second entry is
probit coefficient, third entry is standard error of probit coefficient
adjusted for two-step procedure and for within-firm correlation.
variable. For dummy variables, this is the increase in probability if the variable were to change
from zero to one, all else equal.
The results in Table 4 give mixed evidence about the role of demand in price wars. The
negative coefficient on industry-level enplanements indicates that a price war is more likely to
occur during a month in which overall industry demand is low. This result is consistent with
the Staiger and Wolak (1992) and Scherer (1980) models of price wars, which rely on capacity
constraints and fixed costs, important features of the airline industry. The positive coefficient on
firm-level enplanements suggests the opposite conclusion. In unreported regressions, the demand
coefficients are also statistically significant when each is used alone. The seasonal effects provide
some additional evidence that fluctuations in demand may have an effect on price-war behavior.
Price wars appear to be less common in the second quarter than in the third and fourth quarters,
and most common in the first quarter. In general, estimation of (4) shows evidence that in the
absence of financial measures, demand fluctuations play some role in explaining price wars.
Decision to lead a price war. This subsection presents the results of estimating (1) in order
to address the question of whether a firm’s financial condition affects its decision to lead a price
war. The results presented are for two measures of the ability of the firm to meet its long-term
obligations, namely the leverage ratio and interest coverage ratio. Table 5 shows the results of
estimating (1) using the two-step probit model. As in Table 4, each cell in the table contains the
estimated probit coefficient and its standard error and, in italics, the marginal probability effect
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
310
/
THE RAND JOURNAL OF ECONOMICS
TABLE 5
Lead Decision and Financial Measures
Leverage ratio
.067
.877
(.395)
1 if leverage ratio in top 20%
.054
.531
(.217)
−.005
−.080
(.035)
Interest coverage
1 if interest coverage in bottom 40%
.084
.915
(.343)
1 if bankrupt
−.016
−.261
(.482)
.002
.022
(.344)
.019
.273
(.237)
.026
.282
(.219)
Firm-level enplanements
−.004
−.052
(.329)
.019
.249
(.334)
−.015
−.273
(.333)
.004
.055
(.305)
Industry-level enplanements
.011
.143
(.469)
.004
.053
(.469)
.012
.216
(.497)
.014
.191
(.476)
Potential followers
.006
.081
(.059)
.006
.079
(.053)
.005
.082
(.064)
.000
−.002
(.072)
1 if among three largest carriers
−.010
−.143
(.204)
−.005
−.065
(.216)
.000
.008
(.280)
.023
.281
(.302)
1 if among three smallest carriers
−.021
−.332
(.387)
−.021
−.325
(.304)
−.016
−.345
(.301)
−.031
−.596
(.359)
Second quarter
−.013
−.176
(.227)
−.013
−.176
(.224)
.008
.140
(.278)
.000
.004
(.256)
Third quarter
−.020
−.288
(.260)
−.019
−.281
(.252)
−.001
−.011
(.293)
−.007
−.106
(.301)
Fourth quarter
−.013
−.187
(.198)
−.014
−.201
(.202)
−.005
−.087
(.224)
−.013
−.184
(.208)
Notes: Dependent variable is one for firm-months in which the firm starts a price war, zero otherwise.
Two-step probit estimation, instruments are firm-level and industry-level departures, three weighted
macroeconomic measures, and lagged financial measures. Year dummies included in all specifications,
1,023 observations. First entry is marginal probability, second entry is probit coefficient, third entry is
standard error of probit coefficient adjusted for two-step procedure and for within-firm correlation.
implied by the estimated coefficient; the marginal probability in this case is the change in the
probability that a firm will lead a price war.
As the first two columns of Table 5 show, firms are more likely to lead price wars when they
have high leverage ratios. The estimates in the second column indicate that a firm in the top 20%
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
BUSSE
/
311
of the distribution of leverage is more likely by 5.4 percentage points to start a price war. Since the
unconditional sample probability of leading a price war is 4.3%, this is a large effect. In addition,
the first column of the table indicates that there is a significant linear relationship. The estimated
marginal probability of .067 indicates that if, for example, a firm with equal amounts of debt and
equity were to increase its debt by 10%, its probability of leading a price war would increase by
.67 percentage points.17
The second two columns of Table 5 report the results for the decision to lead a price war
as a function of interest coverage. Since a firm is better off with higher interest coverage, the
coefficient on the interest coverage variable (in the third column of the table) should be negative,
and the coefficient on the indicator that the firm is in the bottom of the distribution of interest
coverage (in the fourth column) should be positive. The coefficient estimate in the third column
is indeed significant and negative, indicating that firms are less likely to start price wars the more
able they are to make their interest payments. The estimated effect, however, is very small. The
effect is larger when using dummy variables. Namely, when a firm is in the bottom 40% of the
distribution, it is more likely by 8.4 percentage points to initiate a price war.
The bankruptcy coefficient is not statistically significant in any of the specifications in
Table 5.
One potential concern is that since the data used to estimate (1) are from a monthly panel,
the error terms might be correlated over time within a carrier. The standard errors reported in
Table 5 are corrected to allow for such correlation. Another approach would be to use random
effects. While this is not straightforward to implement in the instrumental variable probit, it is
not difficult in a linear probability model. Doing so gives numerically similar results to those
reported, although some of the financial variable coefficients become significant only at the 10%
level or are statistically insignificant.
Taken as a whole, these results are consistent with the findings of theoretical models that
imply that highly leveraged firms are more likely to take on risk or to engage in activities that
less-leveraged firms would not undertake.
The results in Table 5 indicate that firms in the worst 20% or 40% of the distribution, not
just in the extreme tail, display an increased likelihood of starting price wars. Although 40%
generally would be a large fraction of an industry to classify as distressed, as an industry airlines
were in unusually poor financial condition compared to other industries during the 1985–1992
period. This can be seen by comparing the financial ratios for the airline industry to those of
other industries.18 For example, the median leverage ratio, which was .75 in airlines, was .59
in transportation, .57 in services, and .55 in manufacturing, while the median interest coverage,
which was 2.94 in airlines, was 3.81 in transportation, 5.19 in services, and 6.51 in manufacturing.
Thus it is reasonable to expect that a higher-than-normal fraction of airlines would be operating
in a distressed condition.
Estimation results of the effect of measures of liquidity and profitability on price-war behavior
are not reported. The standard errors of the financial coefficients are too large to draw reliable
conclusions. Additionally, these measures are the most likely to be endogenous with price-war
behavior. The estimated results for the nonfinancial variables are similar to those reported in
Table 5.
The stronger performance of the solvency measures compared to the liquidity and profitability
measures is consistent with the finding in the bankruptcy prediction literature, discussed in Section
4, that leverage and fixed-payment coverage are generally more indicative of distress than liquidity.
It may also indicate that airlines that face liquidity crises, but not underlying insolvency, have
access to lines of credit or other resources that enable them to meet immediate obligations, and
17 In unreported regressions, results of similar size and statistical signficance were obtained when using the longterm debt ratio in place of the leverage ratio.
18 Ratios are calculated from firms listed in Compustat for the period 1985–1992 in the following industries:
transportation other than by air (2-digit SIC codes 40–42, 44, 46, 47), services, hotel, and recreation (SIC codes 70, 72,
73, 75, 76, 78, 79), and light manufacturing (SIC codes 20–26). Ratios for airlines taken from Table 2.
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
312
/
THE RAND JOURNAL OF ECONOMICS
need not resort to price wars. Whichever is the case, it does appear that financial condition plays
a role in initiating price wars, and that distressed carriers are more likely to initiate price wars.
In addition to the results for the financial measures, the estimates contain several other
results. First, there is some evidence in Table 5 that the smallest carriers are less likely to initiate
price wars. This estimate may indicate that the activities of small carriers are underreported, as
suggested in Section 3. It seems likely, however, that this bias would affect the reporting of small
carriers among a long list of followers more heavily than it would affect the identification of a
small carrier if it were the initiator of a price war. Alternatively, it may be that since a small
carrier will generally overlap relatively few of another airline’s routes, price cuts by small carriers
are less likely to prompt a response. This explanation is borne out in the next subsection, which
shows that firms are less likely to respond to carriers whose routes overlap with few of their own,
although it is not borne out directly by the potential-followers coefficient. The potential-followers
variable is an admittedly ad hoc measurement, which may explain its statistical insignificance.
More strikingly, in these results the decision to start a price war is not strongly related to the
level of demand. The coefficient estimates on the industry- and firm-level demand measures are
generally positive in Table 5, but the standard errors are in every case too large to reject statistical
difference from zero. Additionally, none of the seasonal effects are statistically significant. While
the financial measures may not be independent of demand, the precision with which the financial
coefficients are estimated suggests that the insignificance of the demand measures is not due
merely to multicollinearity.
This is particularly noteworthy because of the importance that the level of demand plays in
both the theoretical and previous empirical literature on price wars. There are several possible
explanations for the difference between this and previous empirical work. One explanation is
that this article differs from previous price-war literature in considering that firms may have
asymmetric incentives to initiate price wars; in particular, some firms are financially constrained
while others are not. It would be nearly impossible for an investigation of price wars that looked
at the occurrence of price wars, without distinguishing price-war leaders from followers and
without using firms’ individual circumstances as explanatory variables, to find such an effect,
while a demand effect would have been discoverable.
A second possibility is that demand is simply incorrectly specified or inadequately measured
in the regressions presented here. One alternative demand specification is that of Morrison and
Winston (1996), who use macroeconomic measures directly instead of enplanements to predict
airline price wars. Employing a similar strategy here by using the weighted macroeconomic
measures (or anticipated and unanticipated fluctuations in those measures) in these regressions
directly instead of as instruments produces results that are very similar to those reported in Table
5. Specifically, the point estimates and statistical significance of the financial measures are similar,
and the macroeconomic demand measures are insignificant. Another possible empirical model of
demand is that of Borenstein and Shepard (1996), who study the dynamics of gasoline pricing,
although not explicitly price wars. They use both a current and an expected demand measure based
on volume sales of gasoline. They predict expected volume on the basis of current volume and
dummy variables that exploit regional differences in the seasonality of gasoline demand. Such a
strategy is less successful in my data, because airlines do not face sufficiently large differences
in the seasonality of demand. As a result, expected demand and current demand measures are
colinear. A final possibility is to measure demand in terms of capacity utilization, or “load factor”
in industry parlance. This would specifically test the Staiger and Wolak (1992) model of price
wars. If load factor is used in place of or in addition to the demand measures reported here, the
results are substantially the same.
Distinguishing financial from economic distress. One question raised by the results reported so
far is whether the estimated impact of the finanical measures on price-war behavior is indeed
the result of financial distress or whether economic distress causes both poor finanical condition
and price-war behavior. The conceptual distinction is the following. A firm could be financially
solvent but face slack demand, aggressive entry, or be poorly managed and thus be unprofitable
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
BUSSE
/
313
(economically distressed). On the other hand, a firm could be profitable on an operating basis but be
highly leveraged and therefore financially distressed. While the conceptual distinction is simple,
making the empirical distinction between the two is more difficult because a period of economic
distress can lead a firm into financial distress. A related issue is raised by Gertner’s (1993) model
in which firm asymmetries, for example in the attractiveness of route structure, increase some
firms’ incentives to start a price war. A poor airline will not be able to fill its seats when prices
are high, but it can when prices are low. Because it is difficult to measure airline quality, this
is difficult to test directly. To the extent that this effect operates through economic distress or
operating losses, the Andrade and Kaplan (1998)-style test described in the next paragraph also
tests this hypothesis.
Andrade and Kaplan (1998) attempt to distinguish between financial and economic distress in
a study of the success of highly leveraged transactions by examining only those firms with positive
operating profits (i.e., economically nondistressed firms). I can perform a similar experiment by
using only observations for quarters in which a firm has positive operating profits.19 Table 6
reports the results of estimating some of the earlier regressions using data from the economically
nondistressed subsample. Results are reported only for the leverage ratio. Interest coverage is
highly correlated with operating profitability, leaving few observations with low interest coverage
and operating profits.20 While this is not a cleanly exogenous partition of the data, it is still the case
that if the financial distress measures merely proxy for economic distress, then the coefficients
on the financial measures should be smaller in magnitude in the economically nondistressed
subsample than in the full sample.
Instead, the results show the opposite; namely, financially distressed firms are more likely,
even in periods of operating profitability, to initiate price wars. Although the estimated marginal
probability effects of 12.5 and 13.7 percentage points are larger than those estimated on the full
sample, the two specifications are not directly comparable. The reason is that probit coefficients
are identified only up to the standard deviation of the error term; if the underlying variance of the
subsample in Table 6 is different from that of the full sample in Table 5, then the coefficients and
marginal probability effects are implicitly renormalized. A more appropriate comparison is to
see whether the magnitude of the financial measure coefficients relative to the other coefficients
in the specification changes between the full sample and subsample. On this basis, the relative
magnitude of the financial measure coefficients is still larger in Table 6 than in Table 5, although
the difference is more pronounced for the first column of Table 6 than the second.
Decision to follow in a price war. Although financial considerations and the decisions to
lead a price war are the primary concern of this article, for completeness I include a brief analysis
of price-war followers’ decisions. After one firm has cut its prices, other firms have to decide how
to respond. This decision is likely to depend on how the potential respondent is affected by the
price cut, which in turn depends on the number of routes the potential respondent shares with the
price cutter. Even if a particular carrier would not have chosen to cut prices on its own initiative,
it may do so when undercut by a competitor. The primary hypothesis in this subsection is that the
likelihood that a firm will respond to a competitor’s price cut depends on the extent to which the
competitor serves the same routes as the initiating firm. Note that major carriers do not always
join all price wars. In fact, in some cases carriers are reported as explicitly announcing a decision
not to join a price war.
I model the probability that a firm will join an existing price war as a function of demand,
controls, and also the extent to which the firm’s routes overlap with the routes of the price-war
leader. Although the financial condition motivation described in Section 2 applies primarily to the
decision to initiate a price war, I include some of the financial variables in the follow specification
19 All firms in the sample have at least one quarter of operating losses, so it is not possible to choose firms that are
economically nondistressed for the whole sample period.
20 Leverage is less correlated with operating profitability; for example, in the economically distressed sample 32%
of firms are highly leveraged, compared to 26% in the nondistressed sample.
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
314
/
THE RAND JOURNAL OF ECONOMICS
TABLE 6
Lead Decision for Observations with Positive
Operating Profits
Leverage ratio
.125
2.859
(1.052)
1 if leverage ratio in top 20%
.137
1.033
(.422)
−.045
−1.024
(.907)
.010
−.177
(.580)
Industry-level enplanements
.027
.624
(.807)
−.018
.315
(.752)
Potential followers
.007
.153
(.089)
−.010
.168
(.066)
1 if among three largest carriers
−.012
−.305
(.310)
.018
−.332
(.307)
Second quarter
−.012
−.294
(.273)
−.014
−.248
(.248)
Third quarter
−.026
−.733
(.218)
−.031
−.626
(.194)
Fourth quarter
.004
.080
(.375)
−.002
−.028
(.304)
Firm-level enplanements
Notes: Sample includes observations only in quarters in which firm had positive
operating profits. Dependent variable is one for firm-months in which the firm
starts a price war, zero otherwise. Two-step probit estimation, instruments are
firm-leveland industry-level departures, three weighted macroeconomic measures,
and lagged financial measures. Year dummies included in all specifications, 529
observations. First entry is marginal probability, second entry is probit coefficient,
third entry is standard error of probit coefficient adjusted for two-step procedure
and for within-firm correlation.
as well.
Pr(Followit = 1) = F(Demandit , Route Overlapit , Financial Measuresit , Controlsit ).
(5)
The demand measures and controls are the same as those used in the estimation of (1). Equation
(5) is estimated using the two-step probit, instrumenting for enplanements with departures and
macroeconomic variables and for the financial measures with lagged financial measures. Since
the decision to follow can be made only if there is an existing price war, (5) is estimated using
firm-month observations for months in which there is a price war, but in which the firm in question
is not the leader.
Route overlap with the price-war leader(s) is measured for each carrier in each month as
the share of that carrier’s traffic that is on routes also served by the price-war leader. The overlap
measure for carrier i in period t is
Pir
I (r, t) ·
,
(6)
Pi
r
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
BUSSE
TABLE 7
Sample Statistic
/
315
Distribution of
Route Overlap with
Price-War Leader
Route Overlap
90%
.943
75%
.859
50%
.737
25%
.587
10%
.385
Mean
.702
Standard deviation
.21
where r indicates theroute, I (r, t) equals one if route r is served by a price-war leader in the
sample period,
Pir = t Pir t indicates the number of i’s passengers on route r in the same period,
and Pi = r Pir is carrier i’s total passengers in the sample period.21 The information on route
traffic is calculated from the U.S. Department of Transportation’s Origin and Destination Survey.
The route-overlap variable used here is more aggregate than the ideal measure, which would be
the amount of overlap a carrier has with a leader on routes that are affected by the price war.
Since the information on price wars indicates whether a carrier is involved in a price war but not
which routes are affected, it is not possible to construct a less aggregate measure. Also, traffic
information is available quarterly, so the route-overlap measure is based on the price-war leader
of a given month, but using quarterly share information. Information on the distribution of route
overlap with the leader is given in Table 7.
The results for the decision to follow in a price war are given in Table 8. These results confirm
the intuition that a carrier is significantly more likely to respond to a price war if it shares much
of its route network with the leader(s). The estimate in the first column indicates that an increase
of 10 percentage points in the share of a carrier’s traffic on routes served by a price-war leader
increases the probability that that carrier will join the price war by 7.05 percentage points. The
sample probability that a carrier will follow in a price war (conditioning on there being a price war
and the carrier not being a leader) is 47.0%. In the specification that includes interest coverage,
the estimated coefficient is slightly smaller.
In none of the follow-decision specifications, including others not reported here, are the
financial measures statistically significant; in the unreported specifications, the route-overlap
coefficient is similar to that reported in Table 8. There is some evidence that carriers take
into account the state of demand when deciding whether to follow in a price war. In the two
specifications reported in Table 8, firm-level enplanements has a marginal probability estimate of
approximately .3, which is significant at the 5% level. This would indicate, for example, that a
firm experiencing demand that is 50% higher than usual would be more likely by 15 percentage
points to follow in a price war.
A final result is that large carriers are significantly more likely and smaller carriers less likely
to join price wars; this result holds even though the regression is controlling for the level of route
overlap. This may indicate that the overall size of a carrier’s network, not just the portion shared
with the leader, has an effect on the probability of joining a price war; it may also be evidence of
a reporting bias toward larger carriers and away from smaller carriers.
The results of this subsection must be interpreted with care. It may be that a particular carrier’s
decision to follow in a price war is dependent on whether other carriers decide to do so. If this is
21 The overlap measure could also be calculated using the number of i’s passengers in the month, but that could
well be endogenous with whether i participated in the price war.
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
316
/
THE RAND JOURNAL OF ECONOMICS
TABLE 8
Follow Decision
Fraction of routes shared
with price-war leader
.705
1.773
(.428)
1 if interest coverage in bottom 30%
.477
1.198
(.560)
.373
.972
(.681)
Firm-level enplanements
.360
.906
(.440)
.305
.768
(.388)
Aggregate enplanements
−.515
−1.294
(.986)
−.284
−.714
(1.349)
.324
.841
(.240)
.397
1.053
(.302)
−.246
−.653
(.249)
−.332
−.924
(.304)
Second quarter
.175
.444
(.331)
.269
.694
(.399)
Third quarter
.128
.321
(.653)
.133
.336
(.808)
Fourth quarter
.000
.001
(.235)
−.022
−.054
(.325)
1 if among three largest carriers
1 if among three smallest carriers
Notes: Sample includes observations only in months in which a price war occurred,
and for which the firm in question was not the leader. Dependent variable is one
for firm-months in which the firm follows in a price war, zero otherwise. Two-step
probit estimation, instruments are firm-level and industry-level departures, three
weighted macroeconomic measures, and lagged financial measures. Year dummies
included in all specifications, 281 observations. First entry is marginal probability,
second entry is probit coefficient, third entry is standard error of probit coefficient
adjusted for two-step procedure and for correlation across firms within month.
the case, then the set of firms that will enter the price war is not uniquely identified in equilibrium,
although the total number of entering firms may be. Bresnahan and Reiss (1991) describe this
problem in the context of simultaneous-market-entry games. As a result, the probit estimations
of this subsection must be interpreted as descriptive, rather than as directly representative of an
equilibrium outcome.
One way to handle this would be to base an estimation on the total number of firms that
participate in a price war, rather than on the identities of the followers (Berry, 1992). Employing
such a strategy in this particular case is hampered by the fact that there are only 31 price wars. In
unreported regressions, the number of followers in a price war was found to be positively related
to both the level of demand, at the 10% confidence level, and to the amount of overlap among all
potential followers or among one or several of the most overlapped potential followers, but not at
conventional levels of significance.
6. Conclusion
This article has examined how an airline’s financial situation affects its conduct in price
wars. The primary hypothesis is that an airline that has difficulty meeting financial obligations
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
BUSSE
/
317
and is in danger of bankruptcy stands to gain the most from an increase in demand engendered
by a significant price drop. Such an airline should thus be most likely to initiate a price war. The
article empirically tests this hypothesis using data on the participation in price wars of 14 major
U.S. carriers, and information on their financial conditions. Information on price-war participation
comes from reports in the Wall Street Journal. This procedure identifies the timing of price wars
more accurately than methods used in previous studies and makes it possible to distinguish pricewar leaders from followers.
The empirical results indicate that firms whose financial measures are in the lowest third
of the distribution are more likely by about 5 to 8 percentage points to start price wars. This is
a large effect compared to the unconditional sample probability of starting a price war of 4.3%.
The results are significant for measures of leverage and interest coverage, which are among the
financial variables that the finance and accounting literature has found to be most predictive of
bankruptcy.
In contrast to the results of previous empirical studies of price wars, fluctuations in demand
do not play a significant role. Specifications using demand measured at the level of the firm,
industry, and aggregate economy, as well as a measure of demand relative to capacity, all fail to
find a significant effect. The estimated effect of the financial measures, however, is robust to the
specification of demand.
The article makes three primary contributions. First, it suggests that price wars are motivated
not only by market conditions that are external to the firm, but by internal characteristics of the
firms operating in the market. Specifically, it has suggested that since firms in worse financial
condition discount future returns more heavily relative to current returns, such firms will find
price wars more attractive. Second, the article has suggested that the decisions to start a price
war and to join an existing price war have different motivations. For the question posed in this
article, namely the effect of financial conditions, it was necessary to examine the two decisions
separately. However, given the results here, there may be gains to separating the analysis in other
contexts as well. Finally, the article has provided additional evidence on the link between a firm’s
financial characteristics and its conduct in product markets.
References
AIR TRANSPORT ASSOCIATION. 1993 Air Travel Survey–Executive Summary, 1993.
AIR TRANSPORT WORLD. Various issues, 1985–1993.
ALTMAN, E.I. “Financial Ratios, Discriminant Analysis and the Prediction of Corporate Bankruptcy.” Journal of Finance,
Vol. 23 (1968), pp. 589–609.
ANDRADE, G. AND KAPLAN, S.N. “How Costly Is Financial (Not Economic) Distress? Evidence from Highly Leveraged
Transactions That Became Distressed.” Journal of Finance, Vol. 53 (1998), pp. 1443–1493.
ATHEY, S., BAGWELL, K., AND SANCHIRICO, C. “Collusion and Price Rigidity.” Working Paper no. 98-23, Department of
Economics, Massachusetts Institute of Technology, 2000.
BAGWELL, K. AND STAIGER, R.W. “Collusion over the Business Cycle.” RAND Journal of Economics, Vol. 28 (1997), pp.
82–106.
BERRY, S.T. “Estimation of a Model of Entry in the Airline Industry.” Econometrica, Vol. 60 (1992), pp. 889–917.
BHATTACHARYA, R. “Bankruptcy Constraints and Price Wars.” Mimeo, Australian Graduate School of Management, 1997.
BORENSTEIN, S. AND ROSE, N.L. “Bankruptcy and Pricing Behavior in U.S. Airline Markets.” American Economic Review,
Vol. 85 (1995), pp. 397–402.
——— AND SHEPARD, A. “Dynamic Pricing in Retail Gasoline Markets.” RAND Journal of Economics, Vol. 27 (1996),
pp. 429–451.
BRANDER, J.A. AND LEWIS, T.R. “Oligopoly and Financial Structure: The Limited Liability Effect.” American Economic
Review, Vol. 76 (1986), pp. 956–970.
——— AND ZHANG, A. “Dynamic Oligopoly Behaviour in the Airline Industry.” International Journal of Industrial
Organization, Vol. 11 (1993), pp. 407–435.
BRESNAHAN, T.F. AND REISS, P.C. “Empirical Models of Discrete Games.” Journal of Econometrics, Vol. 48 (1991), pp.
57–81.
CHEVALIER, J.A. “Do LBO Supermarkets Charge More? An Empirical Analysis of the Effects of LBOs on Supermarket
Pricing.” Journal of Finance, Vol. 50 (1995), pp. 1095–1112.
——— AND SCHARFSTEIN, D.S. “Liquidity Constraints and the Cyclical Behavior of Markups.” American Economic
Review, Vol. 85 (1995), pp. 390–396.
© RAND
2002.
mss # Busse; AP art. # 03; RAND Journal of Economics vol. 33(2)
318
/
THE RAND JOURNAL OF ECONOMICS
——— AND ———. “Capital-Market Imperfections and Countercyclical Markups: Theory and Evidence.” American
Economic Review, Vol. 86 (1996), pp. 703–725.
ELLISON, G. “Theories of Cartel Stability and the Joint Executive Committee.” RAND Journal of Economics, Vol. 25
(1994), pp. 37–57.
FOSTER, G. Financial Statement Analysis. Englewood Cliffs, N.J.: Prentice Hall, 1986.
GERTNER, R. “The Role of Firm Asymmetries for Tacit Collusion in Markets with Immediate Competitive Responses.”
Mimeo, University of Chicago, Graduate School of Business, 1993.
GREEN, E.J. AND PORTER, R.H. “Noncooperative Collusion Under Imperfect Price Information.” Econometrica, Vol. 52
(1984), pp. 87–100.
HALTIWANGER, J. AND HARRINGTON, J.E. “The Impact of Cyclical Demand Movements on Collusive Behavior.” RAND
Journal of Economics, Vol. 22 (1991), pp. 89–106.
JENSEN, M.C. AND MECKLING, W.H. “Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure.”
Journal of Financial Economics, Vol. 3 (1976), pp. 305–360.
MORRISON, S.A. AND WINSTON, C. “Causes and Consequences of Airline Fare Wars.” Brookings Papers: Microeconomics
(1996), pp. 85–123.
MOSSMAN, C.E., BELL, G.G., SWARTZ, L.M., AND TURTLE, H. “An Empirical Comparison of Bankruptcy Models.”
Financial Review, Vol. 33 (1998), pp. 35–54.
NEWEY, W.K. “Efficient Estimation of Limited Dependent Variable Models with Endogenous Explanatory Variables.”
Journal of Econometrics, Vol. 36 (1987), pp. 231–250.
——— AND MCFADDEN, D.L. “Large Sample Estimation and Hypothesis Testing.” In R.F. Engle et al., eds., Handbook
of Econometrics, Vol. 4. New York: Elsevier, 1994.
PORTER, R.H. “A Study of Cartel Stability: The Joint Executive Committee, 1880–1886.” Bell Journal of Economics, Vol.
14 (1983), pp. 301–314.
ROTEMBERG, J.J. AND SALONER, G. “A Supergame-Theoretic Model of Price Wars During Booms.” American Economic
Review, Vol. 76 (1986), pp. 390–407.
SCHERER, F.M. Industrial Market Structure and Economic Performance, 2d. ed. Chicago: Rand McNally College
Publishing Company, 1980.
SCOTT MORTON, F. “Entry and Predation: British Shipping Cartels 1879–1929.” Journal of Economics and Management
Strategy, Vol. 6 (1997), pp. 679–724.
STAIGER, R.W. AND WOLAK, F.A. “Collusive Pricing with Capacity Constraints in the Presence of Demand Uncertainty.”
RAND Journal of Economics, Vol. 23 (1992), pp. 203–220.
STIGLITZ, J.E. AND WEISS, A. “Credit Rationing in Markets with Imperfect Information.” American Economic Review,
Vol. 71 (1981), pp. 393–410.
U.S. DEPARTMENT OF TRANSPORTATION. Air Carrier Traffic Statistics. Various issues, 1985–1992.
———. Airport Activity Statistics. Various issues, 1985–1992.
WALL STREET JOURNAL. Various issues, 1985–1992.
WALL STREET JOURNAL INDEX, 1985–1992.
© RAND
2002.