The RAND Corporation

Entry and Competitive Structure in Deregulated Airline Markets: An Event Study Analysis of People Express Author(s): Michael D. Whinston and Scott C. Collins Reviewed work(s): Source: The RAND Journal of Economics, Vol. 23, No. 4 (Winter, 1992), pp. 445-462 Published by: Blackwell Publishing on behalf of The RAND Corporation Stable URL: http://www.jstor.org/stable/2555899 . Accessed: 04/01/2012 18:47
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Blackwell Publishing and The RAND Corporation are collaborating with JSTOR to digitize, preserve and extend access to The RAND Journal of Economics.

http://www.jstor.org

RAND Journal of Economics Vol. 23, No. 4, Winter 1992

Entry and competitivestructurein deregulated airline markets:an event study analysis of People Express
Michael D. Whinston* and
Scott C. Collins * *

While recent studies of deregulated airline prices considerably advance our understanding of competitive structure in this industry, they sufferfrom several weaknesses that could potentially undermine their inferences. In this article, we consider an alternative approach to this issue that uses the reactions of incumbent airlines' stock prices to announcements of entry by People Express to shed light on competitive structure. These stock reactions reveal significant route-specificprofits or sunk costs (rejecting,for example, the contestable market model) and also provide evidence on the degree of competitive localization present in the industry. We also examine the price, sales quantity, and schedule changes that followed entry. These changes corroboratethe conclusions emergingfrom our analysis of stock price reactions. 1. Introduction * The domestic airline industry has undergone dramatic changes in the past decade. The passage of the Airline Deregulation Act of 1978 led to the entry of a large number of new carriers, a dramatic change in existing carriers' route and fare structures, and a notable increase in the use of air transportation services. More recently, the industry has seen a highly publicized and controversial wave of consolidation through merger. To evaluate these changes, and to devise proper public policy toward the industry in the future, it is essential to have an understanding of the nature of competition in the industry. At the same time, however, the industry possesses a number of features that make achieving such an understanding difficult. For example, given that a typical carrier serves
* Harvard University and NBER. * * Office of the U.S. Attorney for the Northern District of Illinois.

We would like to thank Doug Bernheim, Richard Caves, Franklin Fisher, Adam Jaffe, Greg Mankiw, Nancy Rose, and Lawrence Summers, as well as seminar audiences at Berkeley, Harvard, MIT, Princeton, the NBER, and the Department of Justice for their comments on an earlier draft of this article. Severin Borenstein generously provided much of the data used in Section 5. We would also like to thank HarvardUniversity, the NBER, and the Sloan Foundation for financial support of this project, Sanseung Yi for excellent research assistance, and Ellen Dipippo, Nancy Evans, and Claudia Napolilli for their help in the preparation of the manuscript. The views expressed in this article do not necessarily represent the views of the Office of the U.S. Attorney of the Northern District of Illinois or the U.S. Department of Justice. 445

446

THE RAND JOURNAL OF ECONOMICS

hundreds of routes that form a single interrelated network, to what extent is competition sensitive to "local" market structure? Is the level of concentration on a route meaningful, or should we focus only on regional, or even national, concentration measures? Likewise, to what extent can potential competition constrain pricing on highly concentrated routes? Is the complete constraint present in the contestable market model of Baumol, Panzar, and Willig ( 1982) descriptive of deregulated airline markets, or do sunk costs and other barriers to entry greatly limit the strength of this effect?The importance of these (and other) difficult questions on the nature of airline competition was evident in the heated debates that took place during the recent merger wave, when carrierswith relatively modest national market shares proposed mergers that led to substantial regional and route-specific increases in concentration. A large number of studies have examined pricing behavior in the deregulated airline industry.' Nearly all these studies have a similar structure. Each seeks to explain the crosssectional variation in fares over various city-pair routes. The explanatory variables include proxies for demand and cost conditions, as well as measures of market structure (e.g., the route's concentration level, the number of "potential entrants," and concentration at the endpoint airports). Typically, these measures of market structure are found to exert a significant influence on fares, a finding that is taken to indicate (among other things) that local market structure matters and, by implication, that contestability theory's strong implications are contradicted. Yet while these investigations have considerablyadvanced our knowledge of competition in deregulated airline markets, they suffer as a group from several weaknesses that could, at least potentially, undermine their inferences. For example, each treats its sample as a set of unrelated routes when, as we have noted above, they actually are all part of an interrelated network. Likewise, each faces the difficult task of finding effective controls for demand and cost conditions across markets.2In the absence of convincing instruments,this latterdifficulty can be particularlyproblematic because market-structure variablesare likely to be correlated with these unobserved cost and demand conditions. Although the degree to which these problems actually affect the conclusions of these studies is unclear, it seems desirable to look for alternative sources of information about the competitive structure of deregulated airline markets that can serve both to corroboratethe findings of these studies and to provide additional insights of their own. In this article, we consider one alternative approach to examining the nature of competitive interaction in the deregulated airline industry. This approach follows recent studies of regulationin using stock price reactionsto shed light on the natureof competitive structure. Here, we examine the value changes that occur in reaction to announcements of entry into airport-pairmarkets. We use these value changes to provide information on the profitability effects of these entry events and, indirectly, about aspects of industry structure. While the use of this methodology itself entails some strong assumptions, it can provide a useful alternative source of information about these markets. The particularentry events that we focus on are those involving People ExpressAirlines during the years 1984 and 1985. During this period People Express expanded its operations into 24 new domestic (non-slot-constrained) nonstop markets. To most industry observers, People Express at this time embodied postderegulation demand and/or cost innovations that were working their way into the marketplace.3This view, if correct, makes these events
' See, for example, Graham, Kaplan, and Sibley ( 1983), Call and Keeler ( 1985 ), Bailey, Graham, and Kaplan ( 1985 ), Morrison and Winston ( 1987 ), and Borenstein ( 1989 ). In addition, two recent articles directly investigate the determinants of entry into city-pair markets, Berry (1989) and Reiss and Spiller (1989). For an interesting, more informal, examination of competition in the industry, see Levine (1987). 2 This problem is in part related to the networking issue, since the effective cost of offering service on a particularcity-pair route will depend on where the plane can fly from the destination city. 3 In this view, People Express's demand-side innovation was the introduction of low-frill air travel, while its

WHINSTON

AND COLLINS

447

ideal for our purposes for two reasons. First, if these events all possess a common driving force, they should have similar implications for airline values. Second, unlike entry events arising for other reasons, this type of event has relatively clear-cut implications for airline values under various theories of competition.4 The primary focus of our analysis is an examination of the value changes attributable to incumbency in the markets entered by People Express. Such an examination is able to shed light on several aspects of industry structure. First, the losses that carriers suffer due to incumbency provide evidence on the degree of profits or sunk costs existing in these markets. In particular, if incumbents on a route suffer value reductions in response to People Express's entry, then one of two things must be true: the incumbents could have been earning positive profits prior to entry; or they must have had sunk costs that kept them in the market despite taking these losses. Furthermore, to the extent that the losses are felt uniquely by incumbent carriers due to their positions in the entered markets, the profits or sunk costs must be tied to their operations on these routes. Hence, the losses suffered due to incumbency on routes entered by People Express provide a lower bound on the level of profits or sunk costs existing in these markets prior to entry. One implication of this point is that the value responses attributable to incumbency offer a test of the contestable market theory that is free from many of the difficulties that arise in testing the theory through an analysis of pricing (such as controlling for network effects, demand, and cost conditions). In particular, since a contestable market is characterized by the absence of both economic profits and sunk costs, a finding of significant value reductions for incumbents indicates that deregulatedairline marketsare not likely to conform to the contestable market model. Of course, if People Express indeed had lower costs than other carriers,this would itself violate the assumptions of the contestable market model. A finding of value reductions for incumbents says more, however; by revealing the presence of profits and/or sunk costs prior to People Express's entry, it indicates that these markets would not be contestable even in the absence of People Express's innovation. We do find that incumbents on routes entered by People Express suffered significant value reductions. In particular,the averageincumbent on these routes lost roughly $3 million to $6 million in value when entry was announced. At the 46% tax rate prevailing in these years, the average incumbent's loss would correspond to a pretax loss of between $6 million and $12 million. This pretax loss is equivalent to the loss that would arise were the incumbent forced to fly its planes on the entered route completely empty for roughly three to five months. Thus, reasonably significant levels of profits or sunk costs existed in these markets prior to People Express'sentry. By implication we conclude-as did the pricing-basedstudies discussed above-that the contestable market theory does not accurately describe the competitive environment of deregulated airline markets. A second aspect of competitive structure that our examination of value responses to entry sheds light on is the degree of competitive "localization" present in deregulated airline markets. This is true in two senses. First, at the simplest level, our finding that there is a unique loss attributable to incumbency on an entered route indicates that competitive outcomes on a route are directly sensitive to its market structure.While this finding corroborates

cost-side innovations involved changes in operating procedures and labor costs that were, to some extent, lower than the formerly regulated carriers. Of course, it could be argued that existing carriers were potentially able to duplicate these service and operating procedures. Likewise, it is unclear how much of People Express's labor-cost differentialcan be attributedto quality or experience differencesbetween its labor force and those of existing carriers. ' For example, in a simple model of entry with sunk costs (e.g., Mankiw and Whinston ( 1986 )), these events would lead to value declines for incumbents, while entry events caused by shocks to demand or common costs could lead to either increases or decreases in value for incumbents.

448

THE RAND JOURNAL

OF ECONOMICS

that of the pricing-based studies, it actually reveals a bit more about competitive structure than their findings did. In particular, while sensitivity of a route's pricing/output to its market structure is a necessary condition for this profitability effect, it is not sufficient; our finding also reveals that the existence of sunk costs at the route level, or other barriersto mobility, prevents a rapid equalization of profits across routes. Second, a central aspect of identifying the loss due to incumbency involves carefully distinguishing these losses from several other potential sources of losses for carriers.These include indirect losses in regulatedareas of the industry and competitive effects felt elsewhere in the system. The process of distinguishing these other effects, however, leads incidentally to some additional findings of interest concerning competitive localization. For example, we provide evidence of product differentiation across airports serving the same city-pair by establishing that the losses felt by incumbents serving the entered airport-pairare not felt by incumbents providing service between other airports in the same two cities. At the same time, however, we do detect evidence of significant value changes felt generally by carriers in the industry. While only 30%of the loss sufferedby the average incumbent on the entered airport-pair route can be attributed to these general effects, these effects amount to over 80%of the aggregateeffect on the industry. These "general"effects can arise from a number of sources, ranging from competitive effects felt throughout the industry to informational effects arising from revised beliefs about People Express's likely scale of future expansion. As such, the precise nature of these industrywide losses remains somewhat unclear. Overall, then, our findings on the value changes caused by People Express's entry corroborate some of the conclusions of the studies of airline pricing behavior cited above (the importance of local market structure, the rejection of contestability) and also provide some independent findings of interest (evidence on the extent of profits or sunk costs present in these markets, on the nonequalization of profits across routes, on the degree of product differentiation between airports in the same city, and on the general industrywide effects of these events). Finally, to provide a more complete picture of the effects of these entry events, we also examine the price, (sales) quantity, and schedule changes that incumbents in these markets undertook in response to entry. The responses to these events paint a picture similar to that emerging from our analysis of stock market data. As expected, incumbents on the entered route dramatically lowered their prices in response to entry: on average, the mean of incumbents' prices fell roughly 35%.A smaller price reduction of 15%occurred on the routes between other airports in the same city-pair. Interestingly, though, the incumbents on the entered route seem, if anything, to have increased both their scheduled service and their sales quantity following entry, a finding whose implications we discuss below. The article is organized as follows. We begin, in Section 2, by discussing our empirical methodology for examining the value changes associated with our entry events. In Section 3 we discuss the data used in this investigation. Section 4 presents our empirical findings on the valuation responses to entry. That section begins with an examination of a relatively simple specification for examining the issues raised here and then successively considers more elaborate analyses of these value changes. In Section 5 we present evidence on the price, sales quantity, and schedule responses to entry. Finally, Section 6 offers concluding comments.

2. Methodology
* The basic approach that we use to identify value changes caused by People Express entry events is the event study.5Unlike the typical event study, however, here it makes more
5 See, for example, Schwert ( 1981 ) and Rose ( 1985) for other examples of the use of this methodology in industrial organization.

WHINSTON

AND COLLINS

449

sense to model the events as causing some abnormal dollar change in value, as opposed to an abnormal return. The reason is that the dollar loss attributable to being an incumbent in a particular entered market is likely to be largely independent of the overall size of the carrier.6Letting Ai (Zt) be the expected dollar change in the value of firm i when entry events with characteristicsZt occur (Z, could include the number of events, who the incumbents are, etc.), we can derive the following return process for firm i from the CAPM model of security prices (see Whinston and Collins ( 1990) for a formal derivation):
(Rit
-

Rft) = ai

+ fB (Rmt - Rft) + at(Ai(Zt))

+?et

(1)

where Rit the rate of return on a share of firm i on day t (including any dividend payments) Rft the risk-free rate of interest on day t Rmt the rate of return on the market portfolio on day t indicator variable equal to one if some event occurs on day t at-an the value of firm i's equity at the end of day t - 1 Vt=(it

a serially uncorrelated error term.

A central featureof the analysis below is that we use differencesin carriers'characteristics to identify the sources of value changes. This leads us to model the dollar change in value for carrieri from a particularevent k as some function, f( Z, A), where Z4 are measurable characteristics of carrier i relevant to event k and A is some parameter vector that we estimate. For example, Z4 could be the number of seats that carrier i offered in the market entered in event k and A might then be the dollar change in value per seat (a parameter to be estimated). Below, we investigate a number of different specifications for f( Z, A.). For any given choice for function f(Z4, A), the total value change for carrier i due to entry on day t is then
Ai (Zt) : f(Zk,
kEEt

A),

(2)

where Et is the set of events occurring on day t. = Substituting (2) in for Ai (Zt) in (1), and noting that &t 1 if and only if Et =A0, yields our basic estimating equation for firm i:

R ORit-Rft)= az(VZ ) (Rit '\ Vt-1 )

+z(Rmt - Rft) + 2
~~keEt
t-1

) + fit. )
+Et

(3)

For completeness, we also examine the change in People Express'svalue from its entry announcements. For People Express we can simply replace f(Z4, A) in (3) with some function g(Xk, 6) whereXk are characteristics event k and 6 are parametersto be estimated. of Below we simultaneously estimate a system of M equations of form (3) and one for People Express, where M is the number of carriers other than People Express in our panel of airlines, allowing the { fit } to be contemporaneously correlated across firms.8

6 Note that we can also check for size dependence by allowing the dollar change to be a function of firm size, or even more generally, to be firm specific. Footnote 32, which discusses alternative specifications, reports the results from a specification that allows abnormal dollar changes to be firm specific. These estimates, which do not alter any of our conclusions, support our use of the abnormal dollar change specification. Note that A can include individualized parameters(e.g., firm-specific fixed effects). 8 We also estimated all of the specifications considered below, including a firm-specific constant in equation (3) to allow for misspecification in the asset pricing model. The estimated value changes for this specification were nearly identical to those reported below.

450

THE RAND JOURNAL

OF ECONOMICS

Finally, up to this point, we have not mentioned anything about the choice of a "window" over which to measure the response to the event. For example, if we want a threeday window with the actual announcement date of the entry at the end of the period (say, in case news of the event leaks out prior to the announcement), then event k will be an element of Et if and only if day t is the announcement date of event k or one of the two trading days immediately prior to that date. As is well known, expanding the event window involves a tradeoff between ensuring that all of the event's effects are captured and reducing precision of the estimated effects (from averagingmore unrelated noise). Below we examine several such windows.

3. Data
* Four sorts of information are required for our study of value responses to entry: event identification, market information for these events, stock data for a panel of airlines, and general information about these airlines' operations.

o Events. The first data issue concerns the definition and identification of relevant events. For this study, we focus on nonstop domestic entry events into non-slot-constrained airportpairs.' We focus on non-slot-constrained airport-pairsto avoid cases where regulatoryaccess restrictions are present. Similarly, international flights are subject to regulation. The time period covered here is 1984 and 1985. The choice of these years has several advantages: first, by 1984, six years had passed since the Airline Delegation Act; second, by 1984, nearly all of the flight restrictions due to the 1981 PATCO strike had been lifted;10and third, these years come prior to the recent merger wave. The set of People Express entry events were identified using Aviation Daily, The Wall Street Journal, and The New York Times.11 The date associated with a particularevent was the date of publication of the earliest report which made it clear that People Express was entering that route. Most often this date corresponded to the date on which People Express's announcement of entry was first reported. In three cases, the date used was based on an Aviation Daily "Intelligence Column" reportingthat People Expresswould enter the market in question. In total, we identified 24 events. They are summarized in the Appendix. As can be seen there, 22 of these 24 events involved Newark Airport. o Market information. For each entry event, we needed to collect information about the set of incumbents operating in the entered route. We did this by examining the edition of the Official Airline Guide issued just prior to the event date.12 The Official Airline Guide provides information on all flights offered between two airportsby each carrierand the type of plane used. For this study, two types of flight information were collected: each carrier's number of nonstop flights per week between the entered airport pair by plane type, 13 and
9 During the period of this study, four airportsfaced government-regulatedtakeoff and landing slot restrictions (New York's LaGuardiaand John F. Kennedy airports,Washington D.C.'s National Airport,and Chicago's O'Hare Airport). 10 Only two non-slot-constrained airports were still subject to these IFR constraints in 1984: Denver and Los Angeles. These constraints were removed in Februaryin Denver and in August in Los Angeles. None of our events involve these airports while they were subject to these constraints (or even shortly after their removal). " The New York Times/ Wall Street Journal Index was used to identify entry announcements in these publications. For Aviation Daily, all 1984 and 1985 issues were examined for relevant events. Aviation Daily is received in New York in the morning of the date of publication. 12 The OfficialAirline Guide is published biweekly. A complete set of back issues is available at Northwestern University's Transportation Library. 13 Actually, only flights using jet aircraftwere counted. The only effect of this decision was that in the 850717 EWR-Albany and EWR-Providence events, Delta Airlines would have been an incumbent had we counted nonjet flights.

WHINSTON

AND COLLINS

451

each carrier'snumber of nonstop flights per week by plane type for all other airport pairs in the city-pair entered (thus, this number will be zero when the cities in question each have only one airport). Using plane-capacity figures, we then converted this information into the number of seats offered by each carrier."4 Overall, 5 out of People Express's 24 entry events had no incumbents. For events that did have incumbents, the average number of incumbents on the entered route was 1.68 (with a standard deviation of .78). The average incumbent was offering 7,100 seats per week in the entered market (standard deviation = 3,283). These routes' average distance was 892 miles (standard deviation = 549).

o Stock data. A panel of airlines was constructed from two sources. First, the CRSP daily price tape was used to gather the price, number of shares outstanding, and dividends for all airline stocks traded on the New York and American Stock Exchanges. To the sixteen carriers from this source that were traded every day in our sample period, we added four carriersthat were traded on the Over-the-Counter Exchange.15In all but one case, nearly all of these companies' operations were in the air transportation business.16
4. Empirical findings on value responses

* In this section we present and discuss the results of our analysis of the value responses to entry. The presentation starts in the first subsection below with discussion of a relatively simple specification of the stock price reaction to entry into an airport-pair market. This simple specification reveals a strong drop in value for incumbents on the entered route. To allay a number of plausible concerns about the interpretation on this finding, and to investigate more thoroughly the sources of value changes, we consider in the next subsection some more elaborate specifications of these effects.

ol The simplest specification. In attempting to explain the value changes caused by People Express's entry into a nonstop airport-pairmarket, the simplest place to start is to model the changes for any particular firm as being related to the extent to which that firm was an incumbent on the entered route. For any given entry event, it is natural to model each incumbent's effect as being proportional to its share of the seats in service on that route. A more difficult question is how to achieve comparability across events. To do so, in what follows we model a given event's aggregate effect on incumbents as being proportional to the total number of seatmiles offered in the entered market prior to entry; the effect on any particular incumbent is then determined by its share of these seatmiles. This choice for capturing the aggregate incumbency effect reflects two considerations. First, it seems reasonable to think that the aggregate effect on incumbents in a given event might be related to market size as measured by available seats. At the same time, since both customer values and costs increase with distance, it seems unlikely that the total dollar value change per seat

14 This capacity information was provided by AVMARK, Inc. (a Washington, D.C.-area consulting firm) and Official Airline Guide, Inc. 15 The 20 firms in our panel were: (i) NYSE stocks: American, Alaska Air, Delta, Eastern, Northwest Orient, Pacific Southwest, Pan Am, Piedmont, Republic, Trans World, United, USAir, and Western; (ii) AMEX stocks: Continental, Ozark, and Texas Air; (iii) OTC stocks: Air California (also on AMEX in 1985), America West, Midway, and People Express. Texas Air actually held the stock of Continental and New York Air. In our analysis, all event variables for Texas Air are actually linear combinations of those for these two airlines using the percentage ownership levels that applied during the sample period. The CRSP share data was corrected in several instances to conform to known dates of stock splits. 16 The only exception was United. Even for United, however, over five-eighths of its assets during the sample period were identified with its airline operations (source: United Airlines' 1986 annual report).

452

THE RAND JOURNAL

OF ECONOMICS

would be equal for a 200-mile market and a 2,000-mile market."7Given this choice, the function f( 4, A) that captures the effect felt by carrier i from event k (discussed in Section 2) is
f(Z4,

A) = y .SM',

(4)

where SMk is carrier i's seatmiles per year in the airport-pairentered in event k and where -y,a parameterto be estimated, measures the dollar change in value for firm i per annualized seatmile (we switch to an annualized basis here to simplify comparisons that we make below)."8 To capture the change in People Express's value, we simply set g(X k, 0) = 0, a parameter to be estimated. One can, of course, think of reasons why alternative ways of capturing this aggregate incumbency effect might be superior. For example, longer nonstop routes might be more competitive because of greater competition from connecting flights, leading value losses to be decreasing rather than increasing in distance. Alternatively, if People Express's scale of entry is independent of market size, then the aggregateloss in value from an event would be roughly independent of the number of seats in service on a route. One might also think that the effects felt by incumbent carrierscould be firm specific. For these reasons, we also investigated some alternative formulations of this aggregateeffect. These alternative specifications led to little change in the resultsdescribedbelow for the simple seatmile formulation; a brief discussion of them may be found in footnote 32. The basic specification in (4) was estimated for three different event windows: a oneday window (on the announcement date), a two-day window with the announcement day at the end of the period, and a three-day window with the announcement day at the end.19 The results are reported in Table 1. The results in Table 1 paint a clear picture of the effect of entry on incumbents. The incumbents experienced a total decline in value whose point estimate ranges, depending on the window employed, between -.0109 and -.0191 per seatmile (the estimates in Table 1 are the total value change over the event window; for example, the number given for the two-day effect is twice the estimated daily average change during the event window). The estimates all allow one to reject the hypothesis of no effect with very high levels of confidence (t-statistics between -2.98 and -3.15). In addition to being statistically significant, the estimated value changes in Table 1 are also significant in economic terms. On average, these point estimates imply a total dollarvalue loss of between $6.2 million and $10.9 million per event for all incumbents combined and between $3.7 million and $6.5 million per incumbent. At the 46% tax rate prevailing in these years, this would correspond to a pretax loss of between $6.8 million and $12.0 million per incumbent. One comparison that helps to give some feel for the size of this loss arisesfrom comparing our estimated declines in value to the (weighted) average of annual revenue per seatmile for our incumbents, which is .081.20 On a pretax basis, then, the measured loss is between 25% and 43% of this quantity. Put differently, an average incumbent's pretax loss is com17 This choice is also consistent with standard industry reporting practices, which typically report values of such items as "operating profit per available seatmile." 18 That is, the variable SMk measures the total number of seatmiles that an incumbent would have flown in a year based on the number it was flying during the period just prior to the entry announcement (i.e., the number of weekly seatmilesin this priorperiod times 52). Unfortunately,it was not possibleto make any seasonaladjustments using our data. 19 In early work we also examined a three-day window with the announcement day in the middle. This was done to check whether significant information was incorporatedinto stock prices after the announcement day. Our results indicated no such effect. 20 The weights used for each airline correspond to that airline's share of the total seatmiles People Express entered against over the sample period.

WHINSTON TABLE 1

AND COLLINS

453

Total Value Response to People Express Entry: Simplest Specification Window Variable One-day -.0109 (.0035) -354 (1,103) Two-day -.0147 (.0049) 440 (1,548) Three-day -.0191 (.0061) -1,293 (1,914)

Incumbent effect per annualized seatmile (SM) People Express effect (1,000s)

Note: standard errors in parentheses.

parable to what it would have suffered had it been forced to fly its planes on the entered route completely empty for roughly three to five months. There are a number of reasons why the estimates in Table 1 may either overestimate or underestimate the true effect of entry on incumbents. In the next subsection we shall consider several possibilities that might lead to overestimates. On the other hand, at least three effects might lead to underestimates. First, we could be misdating events. Second, the market may partially anticipate these events and therefore capitalize some of the losses attributableto entry prior to our event window. Third, some value changes may also be felt by debtholders or workers (if there is rent-sharing). In interpreting the losses measured in Table 1, it is also useful to note that a seat offered in service on a route will typically be serving not only individuals traveling between the city-pair in question but also connecting passengers. Thus, the losses measured per seat are an average of the losses felt by incumbents for these two types of passengers. To the extent that People Express's entry (which was typically only into the city-pair market itself) would affect city-pairtrafficmore than connecting traffic,the estimated effect per seat will understate the loss in the city-pair market.21 Turning to the estimated value changes for People Express, we see that these are small (below $1 million in absolute value), imprecisely estimated, and of inconsistent sign. One possible interpretation of these estimates is that People Express had a number of largely equivalent entry options and that while it was not known where the airline would expand, the general scale of its expansion was fully anticipated. This would lead to little effect on People Express's stock value but a (potential) drop in value for the firms it entered against. One problem with this view, however, lies in its prediction that the expected change in value for the entire set of possible incumbents should be zero, a prediction we find falsified in the next subsection.22Thus, if the general scale and profitability of People Express's expansion was fully anticipated, the nature of the markets entered must have been bad news for incumbents overall. Interestingly, our sample period seems to be one in which People Express shifted its strategy toward entering major markets. Alternatively, People Express's scale of

21 Entry could lower the profitability of an incumbent's connecting traffic in several ways. If, for example, the entrant steals some of the incumbent's city-pair traffic, the incumbent could either lose economies of traffic density or cut back its flight frequency, either of which will hurt it in the connecting market. In addition, a fare reductionin the city-pairmarketmay necessitatefarereductionsin connecting marketsbecauseof arbitrage constraints (passengers can form a connecting trip with two separate segment tickets). Indeed, the losses in the connecting market could, in principle, be as great on a per-passengerbasis as in the entered city-pair. With data on passengers, rather than seats, it would be interesting to examine these differences. Unfortunately, we do not see any way to adequately control for the ratio of connecting to city-pair passengersacross markets with our data. 22 Admittedly, we do not have all possible incumbents in our panel, but the only notable exception is Frontier Airlines.

454

THE RAND JOURNAL

OF ECONOMICS

entry may not have been fully anticipated, but its additional entry decisions may have simply yielded the airline little return (while lowering incumbents' profits). This view is also plausible given the belief of some observers that the decision to expand into major markets was a mistake for the airline.23 0 More elaborate specifications of value responses to entry. While the results in Table 1 demonstrate that airport-pairincumbents experienced a decline in value in response to these entry events, they do not show that these losses were inherently tied to these carriers'roles as incumbents on the entered route. In this subsection we consider several more elaborate specifications of the value responses to entry that help clarify the interpretation of this finding and that also, incidentally, lead to some other findings of interest concerning competitive structure in the industry. We begin with a discussion of the motivation for these specifications before turning to a discussion of the results. Controlling for general effects. One possible interpretation of the losses documented in Table 1 is that rather than being tied to incumbency on the entered route, they are instead part of a more general reaction to a People Expressentry announcement that is felt regardless of whether a carrier is an incumbent. Such an effect could arise from a number of sources. First, a People Express entry announcement could cause general competitive effects that spread throughout the air network. These effects could, for example, be part of a general realignment of routes that serves to equalize profits across routes for existing carriers. In this case, while these losses would still be evidence of profits or sunk costs in the system, they would not be tied to incumbency on the entered route. A particular concern in this regardis the possibilitythat some of these generaleffectscould be felt in areaswhere regulation might be expected to lead to carrierrents: departuresat slot-constrained airports and international flights. A second possibility arises because carriers are not only providers of air transportation services, but also owners of airplanes. If People Express entry events are correlated with changes in airplane values, then carriers may be observed to experience value declines coincident with entry announcements. Finally, these general effects might also be the result of new information that the occurrence of an event conveys about the likely scale of People Express's future expansion. To control for such a general effect, we introduced firm-specific fixed effects into the specification discussed above. With these fixed effects added to the specification in (4), the function f(Z4, A) takes the form f(Z4, A) = ai + y-ySMk, where ai is a firm-specific parameter to be estimated. For example, in the case of value losses due to a decrease in the value of departure slots at slot-constrained airports, the ai would be related to the level of slots each carrier possessed. With informational effects about the scale of People Express's future expansion, ai would be related to the probability that carrier i is an incumbent in a People Express entry event. Controllingfor departures at other airports in the city-pair. One noticeable aspect of our sample is that 22 of our 24 events involve Newark airport, which, to some extent, may compete with two other airports (LaGuardia and Kennedy) for New York-area traffic. If there is a positive correlation between those carriersserving various airport-pairswithin a city-pair (e.g., if the Newark-to-City X incumbents tend to also serve the markets between LaGuardia or Kennedy airports and City X), then some of the incumbency effect we are This is a particularconcern measuring might be due to departuresat these other airport-pairs.

We should emphasize, however, that the use of these events to reveal aspects of competitive structuredoes not depend on the optimality of People Express's entry choices. The entry of a low-priced airline can reveal the existence of incumbent sunk costs regardlessof whether these prices are justified on the basis of costs.

23

WHINSTON

AND COLLINS

455

for our sample because these two other New York-area airports are both slot-constrained, and therefore carriersmight be expected to have regulation-generatedrents at these airports that could be dissipated by entry into a substitute airport. To control for this possibility, we included variables OSMk that measure the number of seatmiles that carrieri flew between the cities involved in event k at airport combinations other than the entered airport-pair(e.g., if the airport-pairwas Newark-Cleveland, we now measure all other seatmiles offered between airports in Cleveland and the New York metropolitan area). By adding these variables to those previously introduced, the function f(Zk, A) now becomes f(Zk, A) = ai + y *SMk + q *OSM . If the losses observed in Table 1 were actually due to departuresat these slot-constrainedairports,this value reduction should now be attributed to OSM. In our sample of events, nearly all of these other (OSM) seatmiles involve slot-constrained departures from LaGuardia or Kennedy airports to the same airport that People Express entered from Newark (only 2 of our 24 events have OSM seatmiles that do not). Compared to the average total number of seatmiles offered on the entered airport-pair (8,718,000 per week), the number at other airports in the city-pair (21,860,000 per week) was roughly two and a half times as large. Most of this difference is due to the largernumber of incumbents serving these other airports, an average of 2.71 compared to 1.33 for the entered airport-pair.Often the carriersserving the entered airport-pairalso served the other airports in the city-pair: the probability that OSMk > 0 given that SMk > 0 was .719 over all events and .793 for New York-area events. Thus, our ability to distinguish between the entered airport-pair incumbency effect (SM) and the effect at these other airports (OSM) stems largely from the difference in the value changes for these dual-operating carriers(the entered airport-pair incumbents) compared to carriers operating only at other airports in the city-pair. An additional aspect of our inclusion of the OSM variable is that it can shed some light on the degree of differentiation between airports. In particular,in the absence of either demand or cost-based differentiationbetween airportsin the same city-pair,we would expect to see similar value effects on departures at the various airports serving the same city-pair. New York-area fixed effects. Even if all of our measured incumbency effect was due to departuresat New York slot-constrained airports, one might still see little of this loss being picked up in the OSM variable if it was felt generally by all carriers at the slot-constrained airport. For example, if the basic hypotheses of contestability about the ease of entry into a route held true for carriers with operations at slot-constrained airports (that is, if any carriercould easily change the use of a slot from the service of one market to another), we would see the profit level of all departures at that airport equalized (though not to zero). Thus, entry would lead to general adjustments in all carriers' flights to that airport, and each carrierwould bear an equal loss in value per slot.24While the use of firm-specificeffects already largely controls for this possibility (since 22 of the 24 events involved Newark), we introduced separate fixed effects for New York and non-New York-area events by changing the function f(Z , A) to f(Z , A) = ai + aNY * INNY + y *SMk + 0b*OSMi, where kNY is an indicator variable equal to one if it is a non-New York event. Inclusion of these separate fixed effects also has a second benefit. To the extent that there are losses arisingfrom informational effectsrelatedto the likely scale of People Express's future expansion, the losses felt by a carriermight depend on whether People Express was thought to be planning to expand with further entries from its Newark hub or was thought to be planning entry elsewhere in its system. If an incumbent in a New York-area event might be more likely to be affected by future New York-area events than a nonincumbent, our incumbency variable (SM) could still capture some of this informational effect. The
24

This statement assumes that the distribution of each carrier'sslots by time of day is the same.

456

THE RAND JOURNAL

OF ECONOMICS

use of separate fixed effects for New York-area events helps to control for this potential
problem as well.25

Controllingfor operations at a newly entered airport. In all but one entry event (the 841114 Minneapolis-Tampa event), People Express entered a new airport for the first time. When it did so, People Express was necessarily also entering a number of other city-pair markets with connecting service (involving travel between the newly entered city and other cities in People Express's network). Thus, airlines offering service on these other routes might be expected to suffer value declines as a result of the entry announcement. In fact, carriersin the entered airport-pair(those with SM > 0) typically had disproportionately large shares at the newly entered airport. Carrierswith SM > 0, accounted for, on average, 44.3% of all entered airport departures, while those carrierswith either SM > 0 or OSM > 0 accounted for an average share of 59.1% of these departures.26 a result, incumbents on the entered As route were likely to bear a disproportionate share of the losses from People Express's connecting service entries. To isolate the loss due to incumbency on the entered nonstop route from these other effects, we constructed a set of variables, EPORT , which measured the number of departures that carrier i had at the airport entered in event k (in the year of the event).27 Thus, our most elaborate specification has
f(Zk, A)
=

ai + aiYNY INNY

y *SMk + p * OSMk +

X *EPORTk.

O Results. The results from successively introducing these more elaborate controls are presented in columns 2 through 5 of Table 2 for the case of a one-day window (for ease of comparison, column 1 reprints the one-day results from Table 1). Although not presented here, the results for two- and three-day windows paint a very similar picture; they are discussed briefly in footnotes 28 and 30 and are available in Whinston and Collins ( 1990). The results in Table 2 indicate that the introduction of these additional controls has little effect on our previous conclusion concerning the value reductions due to incumbency in an entered airport-pair. Although inclusion of these additional variables decreases the precision of our estimate of the SM effect (lowering its t-statisticto -2.18), its value actually rises slightly from our original estimates (from -.0109 in column 1 to -.0126 in column 5 ).28 Thus, the value declines we have observed for airport-pairincumbents do seem to be specifically tied to their incumbency position in these markets. Column 5 also reports the sum of the estimated fixed effects and the standard error of this sum (this statistic is also calculated for column 2, where these fixed effects are first introduced). This estimate reveals two interesting points. First, the loss suffered by a firm in response to an entry announcement is over three times larger when it is an incumbent ($6.2 million) than when it is not ($1.9 million).29 Thus, the loss attributable to being an

25 Another possible informational effect could arise if entry by People Express into a particular carrier's market signalled an intention by People Express to "go after" that carrier. Unfortunately, we see no way to control for this type of informational effect without imposing strong assumptions about its form. However, there seems to be little evidence of this type of pattern in our sample. 26 These figures are conditional on the set of events with some incumbents (either SM or OSM) and exclude the 841114 Minneapolis-Tampa event. The average number of annual departures per carrier for our 19-firm sample at the newly entered airport was 3,444. For those carriers who were incumbents in the entered airport-pair, however, the average number of annual departures at the newly entered airport was 29,616. 27 For the 841114 Minneapolis-Tampa event, EPORT was set equal to zero for all carriers. 28 The two-day window SM estimate for the specification in column 5 is -.0154 (.0082), while that for a three-day window is -.0142 (.0101); the average for the three windows, -.0141, is almost identical to that in Table 1, -.0149. 29 The figure for when a firm is an incumbent is the sum of the average incumbency effect (which, as above, is based on the average number of seatmiles offered by an incumbent) plus the estimated general effect per firm (the estimate in column 5 of Table 2 divided by 19). If, instead, we calculate a weighted average of the estimated

WHINSTON TABLE 2

AND COLLINS

457

Total Value Response to Entry: Various Specifications for a One-day Window (1) -.0109 (.0035) (2) -.0127 (.0040) (3) -.0158 (.0053) .0025 (.0028) (4) -.0144 (.0054) .0020 (.0028) (5) -.0126 (.0058) .0026 (.0029) -33.8 (38.2) -262 (1,144) Yes Yes -36,238 (19,136)

Variable Airport-pairincumbent effect (SM) Other airports incumbent effect (OSM) Entered airport effect (EPORT) People Express effect (1,000s) Fixed effects? Separate New York fixed effects? Sum of estimated fixed effects (1,000s)

-354 (1,103) No No

-256 (1,144) Yes No -30,841 (17,621)

-262 (1,144) Yes No

-268 (1,144) Yes Yes

Note: standard errors in parentheses.

incumbent in the specific market entered is a large share of the loss for an incumbent firm. At the same time, since most carriersare not incumbents in any given event, these general losses make up over 80%of the total industry loss that occurs in response to an event. Given the several possible sources for this "generaleffect" (discussed above), however, their precise nature remains somewhat unclear. Not reported in Table 2 are the estimated values of nonNew York fixed effects. These estimates are quite noisy (there were only two non-New York events), and a quasi-likelihood ratio test of the difference between the New York and nonNew York fixed effects cannot reject equality at conventional significance levels. The estimates of the OSM effect in Table 2 are small (an order of magnitude smaller than the OSM effect) and insignificantly different from zero.30They allow us to reject the hypothesis of equal effects on entered and other airport-pairincumbents with a high degree of confidence (a significance level over .025 ). Thus, on either the demand or the cost side, there appear to be significant differences between operations at the various airports within a city. This finding is, of course, consistent with the observationthat carriersbelieve departure slots at the LaGuardia and Kennedy airports to be highly valuable despite the existence of the Newark airport. One must be more cautious, however, about concluding from our estimates that there is no substitution between the airports within a city-pair, as the usual intuitions regarding the value effects of entry on substitutes need not hold here. Indeed, it is not hard to write down models where People Express's entry into the Newark-to-City X airport-pair actually benefits carriers serving City X from LaGuardia or Kennedy, even though the airports are substitutes.31

fixed effects based on carriers'extent of incumbency (as done earlier), we find a general loss of $3.75 million for the "average incumbent." Thus, as might be expected, these generalized effects were felt more heavily by those carrierswho were more likely to be incumbents in entered markets. Based on this latter calculation, the "average incumbent's" loss is 2.1 times largerwhen it is an incumbent than when it is not. 30 The results for the two-day and three-day windows, which are also small and insignificant, vary in sign; the estimate from the two-day window is .0028 (.0039) and the three-day window estimate is -.0017 (.0048). 31 One reason arises from People Express's role as a niche carrier serving primarily low-willingness-to-pay customers. If People's entry into the Newark-to-City X market leads the Newark incumbent to take actions that make it less attractive to business travelers (e.g., by decreasing flight frequency or increasing the number of coach class passengers and thereby worsening service) or if its entry does so directly (e.g., by making Newark airport generally less attractive), then the gain for LaGuardiacarriersfrom getting these business travelersmay exceed any loss they take on discount-fare customers.

458

THE RAND JOURNAL

OF ECONOMICS

Finally, the estimate of the EPORT effect in column 5 of Table 2 provides some very weak evidence of a negative effect on carriers operating at the entered airport pair (the estimate of the SM effect also declines slightly with EPORT's inclusion between columns 4 and 5). Not only is the EPORT estimate statistically insignificant, but the value losses implied by its point estimate are also relatively small. For the average carrier, this point estimate implies a value loss arising from the EPORT effect that totals only $.12 million. For carriers that are incumbents in the entered airport-pair, which (as we noted above) have much larger than average numbers of departures at the newly entered airport, the value loss due to the EPORT effect is of course much larger: $ 1 million. Nevertheless, the average value loss for these carriers attributable to incumbency on the entered airportpair ($4.3 million) far outweighs this amount.32

5. Price and quantity responses

to entry

* To get a more complete sense of the effects of our entry events, we also examined the price, sales quantity, and schedule responses that accompanied them. The evidence from this investigation broadly corroborates the conclusions from our examinations of value responses and also provides some interesting additional insights. Table 3 provides information on the year-to-year percentage changes in the mean coach class price, standard deviation of coach prices, and total quantity of coach tickets sold from 1984 to 1985 and from 1985 to 1986 for entered airport-pair incumbents in New York-area events. Since our interest is in examining the effect of entry, Table 3 breaks these events into two groups, those markets entered in 1984 and those entered in 1985, with the aim of using the set of markets not entered in a given year as a control group for those that were. Our data come from the U.S. Department of Transportation's Origin and Destination Survey for the first quarters of 1984, 1985, and 1986.33 Thus, to compute the yearly percentage change from, say, 1984 to 1985, we compare the first quarters of these years. This forces us to exclude two events that occurred in the first quarter of 1985 from consideration, as we are unable to identify a single year as the year of entry. Also excluded from Table 3 are two other events, one which did not have a Newark incumbent in the first quarter of 1984 (Nashville), and the other in which People Express switched to another airport-pair within the same city-pair before a year elapsed (San Francisco). Finally, this data is for all direct flights between Newark airport and the new city; that is, in contrast to our analysis above, both nonstop and multistop tickets are included as long as no change of plane occurred. One must, of course, be cautious about attributing changes in these variables to People Express's entry, since the underlying demand or cost conditions may also have changed in

32 For the one-day window, we also investigated an alternative specification of the aggregate incumbency effect in column 5, in which we replaced the entered seatmile measure SMi with

[w + 4*SEATSk

+ g*DISTk + -y*DISTk* SEA TSk]*SHk,

where SEA TSk and DISTk are the total available seats and distance in event k and SHMis firm i's share of the seats in the entered airport-pairfor event k, and replaced OSMi with a similar construction. The previous seatmile = measures correspond to the case where co = =f 1A 0. A test for the significance of the additional six parameters comes nowhere close to rejecting our previous specification. Of the three additional terms, the most important seemed to be DIST. The implied value responses at the sample means from an estimation incorporating this term (but not the other additional ones) had losses per seatmile of -.0118 (.0059) and -.0009 (.0047) in the entered and other airport-pairsrespectively. We also examined the specification in Table 1 (equation (4)), allowing for firm-specificseatmile coefficients. The estimates from this specification(for a one-day window) failed to rejectequal effects at conventional significance levels and led to an average seatmile effect of -.0105. 3 These data were very kindly provided to us by Severin Borenstein. They arise from a 10%sample of tickets.

WHINSTON TABLE 3

AND COLLINS

459

Year-to-Year Percentage Price/Quantity Changes of Newark Incumbents %Ain Standard Deviation of Price 1984-1985 -28 (21) +27 (48) 1985-1986 +21 (29) +1 (44)

%Ain Mean Price Markets Entered in: 1984 (#= 7) 1985 (# = 8) t-statistic for difference in means between markets entered in 1984 versus 1985 (degrees of freedom = 13) 1984-1985 -33 (26) +2 (24) 1985-1986 0 (15) -35 (21)

%Ain Sales Quantity 1984-1985 +108 (101) +36 (40) 1985-1986 +12 (22) +86 (107)

2.5

3.5

0.6

0.6

2.6

1.0

Note: standard deviations in parentheses.

these markets. Indeed, in principle such changes could be the very factors causing entry. Nevertheless, given the magnitude of the effects observed here (which we shall discuss shortly), and the general view that People Express'sentry events had more to do with People Express than with changes in the markets it entered, it may not be unreasonable to assume that much of what we are observing is in fact a direct response to People Express's entry. The striking fact about Table 3 is the remarkably similar pattern observed for the two periods, 1984-1985 and 1985-1986. As is the popular perception, there was a dramatic fall of roughly 35%in the average price of incumbents in the markets entered by People Express in a given year. In contrast, almost no change in average price occurred in markets not entered in a given year. These price reductions were presumably necessitated by People Express's dramatically lower fares. For this sample of markets, People ixpress's average fare in the firstquarterof the year following entry was an averageof 19%below the incumbent airlines' mean fare even after the price reductions noted in Table 3. In addition to these declines in mean price, People Express entry seems to be associated with a decrease or lower increase in price dispersion, though to a greater extent in the 1984-1985 period than in 1985-1986. One speculation about this difference between the two years is that it could in part reflect airlines' increasing sophistication in the use of yield management systems over this period, so that by the 1985-1986 period an incumbent airline could more easily target price reductions where they were most effective. Finally, the change in tickets sold by incumbents following entry is notable. Though there is a high degree of variation evident in the data, it appears that entry into a market by People Express is associated with, if anything, an increase in incumbents' number of tickets sold.34 To investigate further the causes of this sales increase, we also examined incumbents' schedule changes following entry (increases in scheduled capacity can increase sales holding price fixed both by increasing flight frequency and by lowering the likelihood

34 Indeed, these quantity increases made revenue fall by much less than might have been expected given the price reductions.The averagerevenue change in the 1984-1985 period for those marketsenteredwas +21% (standard deviation = 22%), while markets not entered during this period experienced a revenue change of +33% (standard deviation = 28%); in the 1985-1986 period the change for entered markets was +3% (standard deviation = 29%), while it was +10% (standard deviation = 16%) for markets not entered during that period. Note, however, that this difference understatesthe incumbent's losses both because an increase in passengerscarriedraises costs holding capacity fixed and also because, as we document in the text, incumbents seem to have increased their capacity following entry.

460

THE RAND JOURNAL

OF ECONOMICS

TABLE 4

Year-to-Year Percentage Price/Quantity Changes of Other Airport-Pair Incumbents %Ain Standard Deviation of Price 1984-1985 +2 (28) +52 (42) 1985-1986 +6 (16) -12 (11)

%Ain Mean Price Markets Entered in: 1984 (# = 7) 1985 (#= 8) t-statistic for difference in means between markets entered in 1984 versus 1985 (degrees of freedom = 13) 1984-1985 -14 (22) +9 (18) 1985-1986 +1 (13) -15 (12)

%Ain Sales Quantity 1984-1985 +12 (34) -5 (16) 1985-1986 +1 (27) +2 (21)

2.1

2.3

1.2

0.1

2.5

2.6

Note: standard deviations in parentheses.

of lost sales when flights are full). We did this by comparing the number of (nonstop) seats offered by incumbents on the entered route exactly one year following the entry announcement to the number calculated for the time of announcement (we again used the Official Airline Guide). The average change for the markets examined in Table 3 was a 25%increase in seats offered (with a standard deviation of 42%). This is to be contrasted with the average annual increase over 1984-1986 in domestic seats offered at Newark airport by carriers
other than People Express of 1 1%.35 Nearly all of this change in capacity was due to changes

in flight frequency: the average number of seats per flight in the entered markets increased by only 6% (standard deviation of 20%) compared to a 1984-1986 average increase at Newark airport for carriersother than People Express of 3%. These changes in sales and capacity are particularlyinteresting. First, they would seem to rule out at least some models of market behavior. It is difficult, for example, to imagine any simple model of price-taking behavior giving rise to both price reductions and quantity increases by incumbents following entry. One implication of this point is that the profits or sunk costs revealed in our examination of value responses cannot simply be classified as efficiency rents of the sort that may arise in a competitive market. Second, these responses indicate that incumbents' losses did not arise during a temporary period of adjustment toward a lower scale of operations. In most commonly used static models of oligopoly the typical response to entry involves decreases in capacity and sales along with decreases in price (e.g., Kreps and Scheinkman (1983)). While one can write down models in which this is not true (e.g., where quantities are strategic complements), these observed quantity responses at least raise the possibility that incumbent airlines may have been following somesort of dynamic strategy. In particular, as some industry observers claim, incumbents may have elected to respond aggressively in the hope of spurring exit or at least discouraging further expansion.36 We also performed a similar exercise for the other airport-pairs (typically, trips between the newly entered airport and either Kennedy or LaGuardia airports in New York). Table 4 provides the results for these airport-pairs.Once again, the 1984 and 1985 events

3 In fact, this comparison understates the difference, since the data yielding the 11%figure (from the Federal Aviation Administration's AirportActivity Statistics) include the entered markets. The same point applies below when we make similar comparisons. 36 In this regard, it is interesting to note that the capacity increases were larger in 1984 than in 1985 (an average of 48% versus 5%with standard deviations of 52% and 14%respectively).

WHINSTON

AND COLLINS

461

show strikingly similar patterns: markets entered in a given year see significant declines in the mean of prices and, if anything, quantity increases relative to those markets not entered in that year. Consistent with our value response finding, however, these changes are much smaller than those for the entered airport-pair.Analysis of schedule data again reveals increases in capacity on these routes: the average change in incumbent capacity in the year following an entry announcement is 1 %(standard deviation of 30%) in comparison to an averageannual change at Kennedy and LaGuardia airportsduring this period of-i %.More than all of this increase in capacity can be attributed to changes in flight frequency: seats per flight fell by 4%(standard deviation of 9%) versus a 1984-1986 decline at these airports of only 2%.As in Table 3, entry is associated with a decline or lower increase in the standard deviation of prices, which is again larger in 1984 than in 1985.

6. Conclusion
* The use of stock market data can provide a useful alternative source of information about the competitive structure of deregulated airline markets. Here we have examined in detail the value responses that accompanied announcements of entry by People Express airlines into airport-pairmarkets. The value changes that we have identified both corroborate some of the conclusions of existing studies of airline pricing behavior (such as the rejection of contestability and the importance of local market structure) and yield some independent findings of interest (evidence on the extent of profits or sunk costs present in these markets, on the nonequalization of profitsacross routes, on the differentiationbetween airportswithin a city, and on the general industrywide effects of these events).
In addition, to provide a more complete picture of these entry events, we have investigated the price, sales quantity, and schedule changes that occurred in response to People

Express'sentry. The incumbent responsesto People Express'sentry that we have documented have not only provided support for conclusions from our analysis of value responses (a dramatic decrease in incumbent fares in the entered airport-pair,a lesser decrease in other airport-pairs within the city-pair), but also offer some interesting additional insights into competitive interaction in deregulatedairline markets (the apparent increase in incumbents' service and sales following entry).

Appendix
Event Datea 840510 840817 840817 840910 840910 841003 841026 841114 841128 850111 850208 850410 850411 850417 850417 850417

Summary of People Express Entry Events.

Airport Pair' EWR-Minneapolis EWR-Miami EWR-Detroit EWR-San Francisco EWR-Denver EWR-Cleveland EWR-Orlando Minneapolis-Tampa EWR-Greensboro EWR-Rochester EWR-Cincinnati EWR-Ft. Lauderdale EWR-Birmingham EWR-Raleigh/Durham EWR-Charlotte EWR-Dayton Sourcec AD AD, NYT, WSJ AD, NYT, WSJ WSJ WSJ AD, NYT, WSJ AD WSJ AD, NYT, WSJ AD AD AD AD AD AD AD Remarks

"Probably also Cleveland and New Orleans." "Intelligence Column" report that PE had secured a gate.

"Intelligence Column" report that PE would begin service. Article saying PE was finalizing plans to serve these three markets.

462

THE RAND JOURNAL

OF ECONOMICS

Appendix
Event Datea 850618 850627 850710 850717 850717 850828 850828 850828
a
b

Continued
Airport Pairb EWR-Nashville EWR-Dallas/Ft. Worth EWR-Atlanta EWR-Albany EWR-Providence Denver-San Diego EWR-New Orleans EWR-St. Louis Sourcec AD, NYT AD AD AD, NYT AD, NYT AD, NYT, WSJ AD, NYT, WSJ AD, NYT, WSJ Remarks

"Intelligence Column" report that PE would begin service. "Intelligence Column" report that PE would begin service.

Dates are given by year/month/day. EWR stands for Newark. cAD = Aviation Daily; NYT = The New York Times; WSJ = The Wall Street Journal.

References
BAILEY, E.E., GRAHAM, D.R., AND KAPLAN, D.P. Deregulating the Airlines. Cambridge, Mass.: MIT Press, 1985. BERRY, S. "Estimation of a Model of Entry in the Airline Industry." Working paper, Yale University, 1989. BAUMOL, W.J., PANZAR, J.C., AND WILLIG, R.D. Contestable Markets and the Theory of Industry Structure. San

Diego: Harcourt, Brace, Jovanovich, 1982.

BORENSTEIN, S. "Hubs and High Fares: Dominance and Market Power in the U.S. Airline Industry." RAND

Journal of Economics, Vol. 20 (1989), pp. 344-365.

CALL, G.D. AND KEELER, T.E. "Airline Deregulation, Fares, and Market Behavior: Some Empirical Evidence."

In A.F. Daughety, ed., Analytic Studies in TransportEconomics. Cambridge: Cambridge University Press, 1985. GRAHAM, D.R., KAPLAN, D.P., AND SIBLEY, D.S. "Efficiency and Competition in the Airline Industry." Bell Journal of Economics, Vol. 14 (1983), pp. 118-138. KREPS, D.M. AND SCHEINKMAN,J.A. "Quantity Precommitment and BertrandCompetition Yield Cournot Outcomes." Bell Journal of Economics, Vol. 14 (1983), pp. 326-337. LEVINE, M.E. "Airline Competition in Deregulated Markets: Theory, Firm Strategy, and Public Policy." Yale Journal of Regulation, Vol. 4 (1987), pp. 393-494. MANKIW, N.G. AND WHINSTON, M.D. "Free Entry and Social Inefficiency." RAND Journal of Economics, Vol. 17 (1986), pp. 48-58. MORRISON, S.A. AND WINSTON, C. "Empirical Implications and Tests of the Contestability Hypothesis." Journal of Law and Economics, Vol. 30 (1987), pp. 53-66. REISS, P.C. AND SPILLER,P.T. "Competition and Entry in Small Airline Markets." Journal ofLaw and Economics, Vol. 32 (1989), pp. S179-S203.
ROSE, N.L. "The Incidence of Regulatory Rents in the Motor Carrier Industry." RAND Journal of Economics,

Vol. 16 (1985), pp. 299-318.

SCHWERT, G.W. "Using Financial Data to Measure the Effects of Regulation." Journal of Law and Economics, Vol. 24 (1981), pp. 121-159. WHINSTON, M.D. AND COLLINS, S.C. "Entry, Contestability, and Deregulated Airline Markets: An Event Study

Analysis of People Express." NBER Working Paper No. 3318, 1990.