SEARCH

SEARCH BY CITATION

Keywords:

  • L13;
  • L93

Abstract

  1. Top of page
  2. Abstract
  3. I Introduction
  4. II Related Literature
  5. III Data and Industry Background
  6. IV Empirical Model
  7. V Results
  8. VI Concluding Remarks
  9. References

We examine price dispersion in a large dataset of Australian domestic airfares. The airlines vary the lowest available fares on successive booking days by restricting the menu of available ticket types, and by changing the prices for some of those types. Our fixed-effects estimator allows us to characterise both of those mechanisms. The greatest price variation occurs on routes involving competition between the two main airlines, Qantas and Virgin; there is greater variation on monopoly routes than on routes pitting Virgin against the Qantas subsidiary, Jetstar. The lowest fares rise rapidly in the week before travel.


I Introduction

  1. Top of page
  2. Abstract
  3. I Introduction
  4. II Related Literature
  5. III Data and Industry Background
  6. IV Empirical Model
  7. V Results
  8. VI Concluding Remarks
  9. References

Airline tickets are sold to individuals and generally are not transferable; and the short-run marginal cost of carrying an extra passenger is well below the average total cost. In these circumstances, it is not surprising that airlines employ an array of price-setting techniques designed to extract as much surplus from passengers as possible. The mechanisms employed include differentiation by cabin class and route, and ticket conditions relating to transferability and cancellation. These practices give rise to significant dispersion in the fares charged to passengers.

We use an extensive self-collected set of airfares for Australian domestic routes to examine the nature of price dispersion. Given the detailed nature of our data, we are able to identify how fares vary according to both the class of ticket and the number of days between booking time and travel. The panel nature of our dataset permits us to control for individual route and flight characteristics.

Our paper contributes to the literature in three ways. First, we characterise the nature of price dispersion in the Australian domestic market for a large sample of flights on a range of routes. Our focus here is partly descriptive. We exploit variation in our data over time of booking to examine the pattern of price variation. At least since Borenstein and Rose (1994), the relationship between price dispersion and the competitive environment has been recognised as an empirical question. We use detailed information on market structure to examine the nature of price dispersion, both for different airlines and for different competitive settings. We find substantial variation in fares on a given flight, both across different classes of travel and over time as the travel date approaches. The relationship between fare dispersion and the competitive environment is not clear-cut. Monopoly routes exhibit less fare dispersion than some competitive environments, but more than others where competition appears to be intense. The greatest variation in fares is generally observed on routes in which the two larger airlines compete directly.

Second, we examine the impact of competition on airfares. Ideally, our analysis would take into account the endogenous nature of the relationship between price and the competitive environment. In the absence of valid instrumental variables, our approach provides some indication of the effect of competition on pricing behaviour, but must be read with some caution. We find that, relative to monopoly routes, duopoly routes offer fares roughly $A20–25 lower, whereas routes in which all airlines compete attract an average discount of around $A35. We present some suggestive evidence that this is an underestimate of the effect of competition.

Third, we shed some light on the mechanisms through which fare changes are implemented. Fares are altered as the travel date approaches, via two methods. The first method involves adjusting offered fares for a given ticket type. This mechanism is adopted by the low-cost carriers Virgin and Jetstar. The second method involves changing the set of available ticket classes over time, thereby altering the lowest available fare. This method is employed by all three airlines that we consider in this paper. Importantly, the use of time varying fares for a given ticket class (our first mechanism) limits the applicability of a class of theories of price dispersion discussed in the literature that presumes rigid prices.

The Australian domestic aviation market is of particular interest for several reasons. First, only three airlines account for almost all domestic air travel, making it feasible to collect fares over booking dates for all airlines on a large set of routes. Second, many of the complications of other aviation markets are absent. A very large proportion of air journeys in Australia involve direct non-stop travel, removing the necessity of considering a complicated hub and spoke network. Third, there is extensive multi-market contact, with Virgin Blue in direct competition with Qantas and/or its subsidiary Jetstar on almost all the routes that it flies. Thus, we can compare fares of different airlines on the same routes, rather than on similar routes. Finally, in recent times the Australian market has seen the exit of an incumbent and the entry of two other players, yielding a competitive environment that has not matured.

The remainder of the paper is structured as follows. In Section II, we briefly discuss some of the related empirical literature and some theories of price dispersion. We discuss the data and industry setting in Section III, and outline our empirical model in Section IV. Results are reported in Section V. We offer concluding remarks in Section VI.

II Related Literature

  1. Top of page
  2. Abstract
  3. I Introduction
  4. II Related Literature
  5. III Data and Industry Background
  6. IV Empirical Model
  7. V Results
  8. VI Concluding Remarks
  9. References

Air travel markets have long been of interest because of substantial observable variation in competitive conditions across routes. While the Borenstein and Rose (1994) study of cross-sectional data found price dispersion to be more pronounced in markets with a greater number of carriers, an examination of panel data from 1993 to 2006 for US domestic markets led Gerardi and Shapiro (2009) to favour the textbook view that more competition generally reduces price dispersion.

In a study of 1995 data for round-trip fares on several US routes, Stavins (2001) found ticket restrictions to be an important tool for price discrimination by the legacy carriers. Substantial discounts were offered to travellers purchasing tickets well in advance, and to those staying away from home on a Saturday night. While the requirements result in product differentiation, the motivation is clearly to support price discrimination. Since 1995, however, the new low-cost carriers and many legacy airlines have adopted a policy of selling one-way fares only, which rules out the use of any stay-away discriminator. And in selling one-way fares, Australian (and other) airlines now change the availability of ticket types in a way that is more flexible than the old advance-purchase requirements.

Recent studies of fare setting practices divide into two categories. The US Department of Transportation data (made available by the Bureau of Transportation Statistics) allow analysis of fares paid for travel on dates extending over a long period. Goolsbee and Syverson (2008) focus on how incumbents respond to the threat of entry, and find that in the classic case of the sustained expansion of Southwest Airlines, the incumbent legacy carriers sometimes reduced their fares on a route before actual or (even) announced entry by Southwest. Exploring similar US data, Gerardi and Shapiro (2009) find that the mechanism yielding reduced dispersion after entry by a new airline depends largely on the presence on the route of both business and leisure travel, with high fares (presumably paid by business travellers) reducing towards the level paid by leisure travellers.

The advantage of the US Department of Transportation dataset is that it includes fares paid for a large sample of tickets. However, information about booking time and ticket characteristics that one might use to characterise fare dispersion are not available. Puller et al. (2009) construct a large dataset, which includes the number of tickets sold at the various fare levels as well as ticket characteristics, and which allows estimation of load factors for individual flights. These important advantages do come at the cost of confining the study to fare-setting by the six large US legacy carriers, for a stratified sample of 90 dense routes on which there is little presence of the new ‘low-cost’ airlines. Further, a significant proportion of the sampled journeys embody travel via an intermediate port. Puller et al. (2009) suggest that for their US context, second-degree price discrimination, achieved through various mechanisms such as non-refundability or stay restrictions, is the main driver of the dispersion of fares.

A second category of studies examines the fares offered on airline booking engines, over successive booking days leading up to the day of travel. Such studies have been undertaken for routes in various countries, and for international routes. For example, Pels and Rietveld (2004) study how prices vary by booking date for a single route between London and Paris. Pitfield (2005a,b) examines the behaviour of competing low-cost carriers on five European routes originating in the United Kingdom. Button et al. (2007) look at farepaths following entry by low-cost carriers on several routes from Lisbon and Porto. Bilotkach (2006) and Bilotkach et al. (2010) study fares for travel between New York and London, and note especially that fare-setting strategies differ between airlines and that fares for business travel vary more than fares targeted on leisure travellers.

Our study too uses information on the fares offered by airlines and addresses price competition as we approach the day of travel. We harness a dataset of a substantially larger scale, having 358,597 flight × booking day observations and 1,140,367 observations for all fare types, across a variety of different routes, travel times, and competitive settings, thereby allowing us to make more general inferences.

A study that is closely related to ours is Bachis and Piga (2006); it exploits a large sample of European fares with variation by booking day. Bachis and Piga examine daily airfare data for six low-cost carriers and for full service carriers operating on similar or identical routes. Airfares are combined with monthly market structure information to form a large monthly dataset. Aside from the different geographic settings, there are several minor differences between the dataset they employ and our own dataset. They use route-specific information about seat numbers and load factors to build measures of route and airport dominance. However, this information is aggregated to the monthly level. By contrast, we rely on information about the set of competing airlines operating on each flight. While we do not have access to airline-specific information on flight load factors, our measures of market structure exhibit considerable within-month time variation. In addition, our fare data include taxes and charges, while the European data excludes those components.

Bachis and Piga (2006) report evidence consistent with intertemporal price discrimination in fare paths generally sloping upward over time, and note that the volatility of fares increases in the 4 weeks prior to departure. In our data, most, but certainly not all, flights involve a monotonically increasing fare path. Gillen and Hazledine (2006) and Hazledine (2010) undertake similar analyses of datasets for North American and Australasian airline markets respectively. Both studies find evidence of extensive price discrimination based on date of ticket purchase. Further, average prices paid are heavily influenced by the number and size distribution of airlines serving a given route.

(i) The Nature of Fare Dispersion

The problem of managing demand for a perishable commodity of fixed capacity, such as seats on an airline route, has been considered in both the economics and yield management literatures. In the yield management literature, the interest has centred on the question of how to dynamically allocate a fixed capacity through quantity-based restrictions, or through dynamic price variation. The economics literature has sought to uncover the inherent motivations for price dispersion.

In the yield management literature, in the search for an optimal pricing policy or quantity-allocation rule, the underlying motivations for yield management are often not explicitly distinguished (see Talluri and van Ryzin (2004) for a recent text treatment). For the purposes of discussion, suppose that demand arrives at a rate given by the function λ(pt,inline image,t;γ), where pt is the market price at time t, inline image is an unobservable shock reflecting demand uncertainty, and the demand function is parameterised through γ. The term t makes the time dependence of demand explicit. If demand is time invariant and there is no demand uncertainty and the parameters, γ, are known to the firm, then the firm's optimal policy is to set a constant price over time (see, for example, Gallego and van Ryzin (1994)). We can then consider the variation in the optimal pricing policy induced by uncertainty (the term inline image) as arising from inventory management concerns: intuitively, the firm optimally varies price over time in response to fluctuation in the option value of remaining inventory as sales are made and time passes. We attribute the variation in price due to the time variation in demand as price discrimination. Finally, if the demand parameters, λ, are unknown to the firm, the firm's optimal price will adjust over time due to learning.

The economics literature has directed some effort towards disentangling these motivations. It is well understood that price discrimination provides one motive for fare variation. Passengers differ in their willingness to pay and this variation is correlated with observable factors for the airline. For example, a common explanation is that business travellers are both less flexible in their plans and willing to pay more to fly. Because they do not have the luxury of booking well in advance, this induces a correlation between profit-maximising fares and the time of booking.

Demand uncertainty has been suggested as an alternative explanation for fare dispersion. Dana (1999) shuts down the price discrimination motive in order to focus on the implications of demand uncertainty for fare dispersion. Dana considers a setting in which capacity is constrained, prices are rigid, and consumers arrive in random order. In competitive, monopolistic, and oligopolistic environments, price dispersion arises in equilibrium.

Dana's (1999) theory makes predictions about the relationship between fare dispersion and realised demand. We are not in a strong position to test these predictions. Information about observed load factors by airline and route are not available in Australia, so we have no data on realised demand. Perhaps more fundamentally, Dana's assumptions break down in the Australian context. In particular, Dana assumes that firms fix a basket of fares in advance, where a pre-set number of tickets with particular characteristics or conditions are made available to potential buyers. In Section IV, we present evidence that two of the three domestic Australian airlines adjust prices over time for a given class of fare on a given flight, violating the basic premise of the model. Further, as Talluri and van Ryzin (2004, p. 176) argue, ‘many of the RM [revenue management] practices of the new low-cost airlines more closely resemble dynamic pricing than the quantity-based RM of the traditional carriers’.

Gale and Holmes (1993) consider the motives for adopting a particular price-dispersion mechanism. In a model with limited capacity (at the time of price setting), customer heterogeneity and demand uncertainty, they show that a monopolist can implement the optimal pricing scheme through the use of advance purchase discounts. Customers with a high willingness to pay for their preferred travel time will purchase a higher priced ticket for their preferred time once uncertainty is resolved. Customers with less intense travel time preferences will commit to a travel time before uncertainty is resolved to take advantage of the advance purchase discount.

The results from the analysis offered in this paper are in the spirit of Gale and Holmes in the following sense. All airlines considered in our empirical analysis offer a fully flexible economy class fare, permitting itinerary changes. Airlines also offer an economy class airfare that is more restrictive with respect to itinerary changes and refundability. While strictly speaking there are no advance purchase discounts in Australia, as we demonstrate in Section IV, there is a tendency for fares to rise on a given flight as the departure date approaches. If customers have expectations that fares tend to rise as the travel date approaches, then there is a perception of an advance purchase discount. Customers willing to pay for flexibility can purchase a fully flexible ticket. Alternatively, customers can commit early to their travel plans for a lower (in expectation) fare.

We do not isolate the source of price dispersion in our data. However, we have some reason to believe that price discrimination does play a role for the following reasons. First, in simulation exercises, the revenue gains from pure inventory management cannot explain the extent of profitability of computer reservation systems. In a model of optimal pricing of limited seat capacity, Gallego and van Ryzin (1994) argue that the potential revenue gains from dynamic (as opposed to fixed) pricing are modest in the absence of time-varying demand. Second, the recent empirical work of Puller et al. (2009) suggests that in the US market, inventory management plays only a limited role relative to price discrimination. Third, for our Australian data, too, a great deal of the variation in fares can be explained by variation in booking days. This suggests an important role for price discrimination, but does not rule out the possibility of alternative motivations for price dispersion.

III Data and Industry Background

  1. Top of page
  2. Abstract
  3. I Introduction
  4. II Related Literature
  5. III Data and Industry Background
  6. IV Empirical Model
  7. V Results
  8. VI Concluding Remarks
  9. References

(i) The Australian Airline Context

Though having a land area much the same as that of the United States, Australia has a population of only 21 million, less than one-tenth of the US population. On Australian domestic air routes in the year to 30 June 2007, there were about 46 million passenger movements, again less than one-tenth of the US total. Among Australian routes with competing airline service, about 50 (city-pair) routes have more than 100,000 passengers per annum.

For many years, the Australian government's two-airline policy (applying to all but local routes) resulted in a duopoly with parallel pricing and very little by way of fare discounting. After deregulation in 1989, a further 11 years elapsed before there was effective competition from a well-managed and financially robust entrant, Virgin Blue. Soon after Virgin's entry, one of the long-established incumbents went out of business, leaving Virgin to compete with Qantas. After Virgin had won about 30 per cent of the market, Qantas set up its own low-cost airline, Jetstar, and gradually handed to that subsidiary entire routes (mainly to tourist destinations), and certain departure times on routes still served by Qantas itself.1

In the year ending 30 June 2007, passengers on all Australian domestic routes flew about 52 billion revenue passenger kilometres. The share of the Qantas group was 66 per cent. Of this share, 46 per cent was for Qantas itself, with its wholly owned subsidiaries QantasLink and Jetstar having 5 and 15 per cent respectively. The only significant rival for Qantas was Virgin Blue, which had about 33 per cent of the market. Almost all of the very small balance was carried by Skywest (with a few jets as well as some turboprops) and Rex (34-seat turboprops only).

Relative to the aviation markets in Europe and the United States, there are several distinguishing features of the Australian market. First, the entire market is very concentrated. Over the period of our dataset, competition for domestic aviation was waged by three main airlines: Qantas, its wholly owned subsidiary Jetstar, and Virgin Blue. Second, multi-market contact was endemic. Indeed, Virgin faced competition from Qantas and/or Jetstar on every route that it served. Qantas had a monopoly on one major route, and on a number of minor routes. Third, almost all passengers in Australia travel on direct, non-stop flights, and all the fares in our database are for non-stop travel.

(ii) Data

We employ a self-collected dataset of airfares offered online by the major domestic Australian airlines for a sample of one-way travel scenarios within the period October 2003–July 2006.2 An important feature of the dataset is that it contains the price charged for each available ticket type for each flight on the given travel date, conditional on the day the flight was booked. Each observation consists of information about the fare offered, the fare class, the identity of the airline, the date of booking, and the details of the flight (origin, destination, departure time and date of travel). The fare information is inclusive of all taxes and charges. During the period, fully inclusive fares were directly available to potential passengers on the first page of their internet fare search.3

During the period, Qantas offered five different fare classes (including one for the business class cabin), and Virgin offered three fares at first and then later in the period it offered four (all for economy). From May 2004, when Jetstar entered the market, it offered only two economy fares. For each airline, its economy fare classes were generally close substitutes. The single exception is the fully flexible fare, offered by all three operators; unlike the cheaper ‘restricted’ fares, this fare permitted the ticket-buyer to change or cancel the flight at any time without penalty.

Because our dataset is collected through direct queries to airline websites, it is a necessarily selective sample. It consists of fares for non-stop travel between 118 different origin-destination pairs. For each flight in our dataset, we collected daily information on the offered fare for all available classes of travel; usually, we did so for at least the 5 weeks preceding travel. In Australia, the concentrated population distribution and increased network connectivity allow a very large proportion of journeys to be made without visit to an intermediate port. Therefore, our non-stop restriction should not greatly affect our results.

In choosing routes, we considered the apparent mix of travel purposes. Routes were classified as either ‘leisure’ or ‘business’ according to our judgement about which type of travel predominated. Broadly, routes having state capitals or major regional cities as both origin and destination were classified as business; however, all weekend travel was classified as leisure, as were routes serving holiday resorts. We use as the numeraire a third category, ‘mixed’ routes, for which both kinds of travel appear to be important.

We chose specific routes and travel dates with the goal of capturing much of the variation in the nature of Australian air travel. We included high-volume routes such as Melbourne–Sydney, predominantly leisure routes such as Sydney–Hamilton Island, and predominantly business routes such as Sydney–Canberra. To allow us to control for observable variation in several other dimensions, we collected fares for both directions of travel, for travel that falls both in and out of school holiday periods, for travel on different days of the week, and for travel starting in peak and non-peak periods. Peak periods were designated as: 6:30–8:30 am and 4:30–6:30 pm on weekdays; 7–10 am on Saturdays; and 3–7:30 pm on Sundays. School holidays vary between states; a flight was deemed to have occurred in a school holiday if the more populous of the origin and destination cities was in a school holiday period on the day of travel.

Table 1 contains summary statistics for our dataset, restricting attention to the lowest available fare on a flight. That approach allows us to capture two primary mechanisms used for intertemporal price variation: directly changing fares within a fare class over time, and restricting the set of available fares over time. By focusing on the lowest available fare, we are able to incorporate both of these elements in a single measure.

Table 1. Descriptive Statistics, Lowest Available Fares
VariableMeanStandard deviationMinMax
  1. Notes: Restricting attention to lowest available fares, an observation is a travel day × flight × booking day combination. Potential competition is characterised by the set of airlines who operate on the same route on the same day of travel. The analysis is based on 358,597 observations. Temporal competition is characterised by the set of airlines who operate on the same route within 90 min of flight departure. The analysis is based on 323,105 observations.

Potential competition
 Fare ($A)162.4684.72281533
 Days until travel25.6217.481145
 Jet fuel prices (AU c/gal)194.1733.44106.34267.02
Flight proportions operated by
 Jetstar0.0570.23201
 Qantas0.6340.48201
 Virgin0.3080.46201
Potential competition proportions
 Monopoly0.2420.42801
 Jetstar and Virgin0.0190.13701
 Qantas and Virgin0.6530.47601
 All Airlines0.0860.28001
Flight characteristic proportions
 Non-peak period0.6750.46801
 School holiday0.0990.29801
 Leisure route0.3670.48201
 Business route0.5700.49501
 Route distance (km)989.13788.472363615
Temporal competition
 Fare ($A)164.3185.69281533
 Days until travel25.7217.591151
 Jet fuel prices (AU c/gal)192.2535.65106.34267.02
Flight proportions operated by
 Jetstar0.0420.20201
 Qantas0.6600.47401
 Virgin0.2980.45801
Temporal competition proportions
 Monopoly0.3730.43401
 Jetstar and Virgin0.0020.04801
 Qantas and Virgin0.6100.48801
 All Airlines0.0140.11801
Flight characteristic proportions
 Non-peak period0.6670.47101
 School holiday0.1200.32401
 Leisure route0.3610.48001
 Business route0.5900.49201
 Route distance (km)970.72791.932363615

In Australia, market share information is not available at the level of an individual domestic route. We consider two measures of the extent of competition on a route that do not rely on market-share data. For each flight, our first measure, which we label temporal competition, is an indicator variable containing the identities of the airlines operating within 90 min (before or after) of the specified flight. This is a proxy for the extent of direct competition faced by the airline on a particular flight. Our second measure, potential competition, is an indicator variable, which identifies all airlines operating on the route on the same day as the flight, but not necessarily flying within the 90-minute limit.4 On both measures, most routes had competition between Qantas and Virgin. Of course, monopoly routes were more common for the temporal measure of competition.

IV Empirical Model

  1. Top of page
  2. Abstract
  3. I Introduction
  4. II Related Literature
  5. III Data and Industry Background
  6. IV Empirical Model
  7. V Results
  8. VI Concluding Remarks
  9. References

Our empirical work has two broad goals: to examine the determinants of airfares and to characterise the nature of fare dispersion. The primary challenge we face is the joint determination of airfares and the level of competition. We might reasonably expect the nature of competition on a route and the level of fares set on the route to be jointly determined by intrinsic characteristics of the route. To deal with this challenge and to approach our twin goals, we adopt two related empirical strategies.

Our first exercise is aimed at investigating the nature of price dispersion. We adopt a model with route-specific fixed effects. By removing unobservable heterogeneity across routes, this allows us to focus attention on price dispersion. We examine the nature of price dispersion by different carriers, and in different competitive environments. Our empirical strategy will tease out the effect of competition on price dispersion under the assumption that, conditional on operating on a given route, airlines do not choose which time period to service based on the same factors that determine price dispersion.

Our second exercise is to examine the determinants of fare levels. Our fixed effects model is ill-suited to this task because it removes all the variation across routes, allowing no inference about what characteristics of a route determine fares. The impact of competition is necessarily identified through variation in competition within a route, rather than variation across routes. Accordingly, we replace our route-specific fixed effects with route characteristics, permitting us to examine the association between airfares and characteristics of individual routes. This strategy does not address the above endogeneity problem, meaning we must treat estimates of the impact of competition on fares with caution.

(i) A Model with Route-specific Fixed Effects

Our first exercise is to characterise the nature of price dispersion. Controlling for route-specific fixed effects allows us to isolate the effect on the fare of the airline operating the route, the time of booking, and the travel time and date. We consider the following empirical specification:

  • image(1)

where pijknt is the minimum available price in Australian dollars ($A) at time t offered by airline i operating on the route linking origin j and destination k, with n days until travel.5A is a vector of indicator variables identifying the airline; R is a vector of indicators identifying the route (with separate dummies for each direction of travel); the vector N identifies the number of days until travel at the time of booking; and the vector C identifies the competitive environment, using the specifications of temporal and potential competition introduced in Section III. The vector X contains cost and demand information that varies across flights within a route. The price of jetfuel at the time of booking is included as a cost shifter. Indicators for the time of day and day of the week of travel are included to distinguish peak and non-peak periods. We also include a constant, a time trend, and an indicator for the presence of school holidays. In some specifications, we also consider interaction terms between our explanatory variables.

Note that, because we incorporate the route-specific fixed effects, it is time-series variation in the competitive environment that identifies the impact of competition on price dispersion. Variation in the set of airlines offering a service in a specific short time window identifies this impact for our first measure of competition. This variation arises because flights are not uniformly spread during the week, and airlines have different schedules. Variation within a week in the set of airlines offering a service on a given day identifies this impact for our second measure of competition.6

(ii) The Effect of Route Characteristics

In Equation (1), the route specific dummy variables, R, allow us to control for fixed characteristics of individual routes. However, their inclusion permits no insight into the determination of prices on individual routes. To this end we replace R by characteristics of individual flights.

  • image(2)

All variables from Equation (1) are retained except R is replaced by W. In the vector W, we consider route- or flight-specific characteristics relating to costs and demand: the nature of demand for travel on the route (business, leisure or mixed); the populations of the origin and destination cities; the length of the route; and an interaction term between route length and jet fuel prices.

As we alluded to above, our primary concern in interpreting Equation (2) is the joint determination of airfares and competitive conditions. Two common solutions to this problem are fixed effects estimation and instrumental variables. The limitation of fixed effects estimation is that it does not permit identification based on variation across routes. The challenge for instrumental variables estimation is to find a set of variables that are correlated with the competitive conditions in a market, but unrelated to the unobservable determinants of airfares. In the absence of convincing instruments, our estimates place a lower bound on the true impact of competition on fares.7

V Results

  1. Top of page
  2. Abstract
  3. I Introduction
  4. II Related Literature
  5. III Data and Industry Background
  6. IV Empirical Model
  7. V Results
  8. VI Concluding Remarks
  9. References

(i) Fixed Effects Model

Lowest available fares

Our fixed effects model examines price dispersion, controlling for route-specific characteristics. To abstract from price dispersion due to product differentiation, we first restrict our attention to the lowest fare made available for each flight. Because of the large number of explanatory variables, we employ a combination of tables and figures to present the results of several specifications of Equation (1).

Models 1–4 of Table 2 all contain estimates of Equation (1), but differ in the treatment of the number of days until travel, which is potentially an important instrument for price discrimination. Model 1 incorporates indicator variables for the number of days until travel, and gives an explicit relationship between booking time and fares. In Model 2, we interact these indicator variables with indicators of the airline operating the flight. This allows air fare variation over booking time to differ between airlines. In Models 3 and 4, we interact the indicator variables with indicators of the nature of competition. We return to these last two specifications shortly. Figures 1–3 contain the parameter estimates for these indicator variables; Table 2 contains the remaining parameter estimates.

Table 2. Price Dispersion, Lowest Available Fares
 Model 1Model 2Model 3§Model 4
 CoefficientStandard errorCoefficientStandard errorCoefficientStandard errorCoefficientStandard error
  1. Notes: The dependent variable is the level of fares in $A. We restrict attention to lowest available fares. Route-specific indicator variables are included in the specification, but omitted above. All standard errors are heteroskedasticity adjusted. Alternative clustering methods were considered, but did not qualitatively affect the results. Model 1 contains indicator variables for the number of days between booking and travel. Model 2 contains indicator variables for the number of days between booking and travel, interacted with the airline operating the flight. §Model 3 contains indicator variables for the number of days between booking and travel, interacted with the nature of competition on the route on the day of travel. Model 4 contains indicator variables for the number of days between booking and travel, interacted with the nature of competition on the route within 90 min of travel.

Constant96.5181.301103.2171.30699.6401.392102.4061.401
Jetstar−27.4090.479  −27.1470.482−26.4170.829
Qantas49.7460.199  49.5290.19946.9970.209
Tues am Peak39.0480.71739.2260.70839.0970.71443.1480.736
Tues pm Peak24.1250.63524.3330.62124.2410.63126.0890.644
Wednesday6.8980.3816.9500.3786.8530.3797.0170.403
Wed am Peak22.4640.66822.6630.65622.5070.66726.2450.762
Wed pm Peak39.3020.79539.3350.78439.2030.79145.3500.921
Thursday20.0440.41019.9050.40820.1770.40923.3020.460
Thur am Peak5.1010.6875.4220.6815.1280.6856.4870.803
Thur pm Peak46.0631.01346.2071.00946.0021.01147.1141.062
Friday38.9640.40539.0210.40338.8430.40440.5370.422
Fri am Peak−26.0840.516−25.8360.512−26.0340.515−27.1350.534
Fri pm Peak71.6580.96171.7760.95671.6610.95873.6441.002
Saturday6.0150.3955.9590.3916.0930.3945.6010.401
Sat am Peak8.8990.5478.9330.5428.9260.5469.0500.560
Sunday17.5470.38117.6820.37917.5150.38017.4890.393
Sun pm Peak20.4260.49620.4870.49520.3860.49621.0220.516
Monday17.5220.42317.6350.42117.3970.42317.6780.442
Mon am Peak28.0910.89228.4900.88228.0890.89028.9000.927
Mon pm Peak2.1210.7392.1750.7342.0450.7393.3050.785
Jetfuel price−0.0280.006−0.0250.006−0.0030.006−0.0350.007
School hols10.3630.35210.2040.3479.8660.35210.0270.338
Time0.0190.0010.0190.0010.0170.0010.0200.001
Observations358,597358,597358,597323,105
R20.5680.5740.5710.570
image

Figure 1. Lowest Available Fares and Booking Day

Download figure to PowerPoint

image

Figure 2. Lowest Available Fares and Booking Day, Potential Competition

Download figure to PowerPoint

image

Figure 3. Lowest Available Fares and Booking Day, Temporal Competition

Download figure to PowerPoint

First, turning to Table 2, it is clear that, for all specifications, airfares are sensitive to travel time, with an average premium for Friday night travel of around $A110 relative to a Tuesday non-peak flight (the numeraire). There is also a premium for school holiday travel. Relative to a flight on Virgin, fares are on average higher on Qantas, and lower on Jetstar. The apparently spurious negative relationship between fares and jet fuel prices presumably reflects a non-linear rise in airfares over time.

Figure 1 contains results for Models 1 and 2, illustrating a dramatic relationship between airfares and booking time. Solid lines represent point estimates, and dashed lines indicate 95 per cent confidence intervals. The line for ‘All Airlines’ is the coefficient on the number of days until travel at the time of booking for Model 1. The numeraire is a flight booked more than 5 weeks before departure. The average premium for booking a flight the day before travel is more than $A70, with the premium rising most precipitously in the last week before travel. The airline-specific lines are from Model 2. The numeraire is a Virgin flight booked more than 5 weeks before departure. Again, Qantas fares are higher and Jetstar fares are lower. The rise in fares as we approach the departure date is largest for Qantas flights, and smallest for Jetstar.

To explore the relationship between the dynamic fare path and the competitive environment, we incorporate into Equation (1) the vector C, containing dummy variables for the nature of competition. Model 3 gives results for the potential competition case. Again, we emphasise that, because Equation (1) controls for route-specific effects, we identify the effect of competition only through variation in the set of airlines operating on different days of travel for a given route. We cannot use variation in the nature of competition between different routes to identify the effect of competition. Model 4 gives corresponding results for the temporal competition case. In this case, we identify the effect of competition through variation in the identities of the temporal competitors, for a given route.

Figure 2 contains the results of Model 3. Each panel corresponds to a different market structure, and contains the coefficients on an indicator of that market structure interacted with indicators of the number of days until travel. The numeraire is a monopoly flight booked more than 5 weeks before travel. Solid lines indicate point estimates and dashed lines 95 per cent confidence intervals. The effect of the nature of competition on the day of travel on the extent of price variation by booking date is ambiguous. Considerable price variation is evident in the $A50 discount available for early booking on monopoly flights. A similar discount is apparent for routes in which all airlines offer flights on the same day. The discount rises to more than $A80 for routes in which Qantas and Virgin offer flights on the same day, but falls to around $A30 for routes involving competition between Jetstar and Virgin.

Figure 3 contains results for Model 4. There is again an ambiguous relationship between the extent of price variation and the nature of competition. Price variation is most dramatic for flights involving temporal competition between Qantas and Virgin. The extent of price dispersion is also substantial for monopoly flights. Price dispersion is less marked in routes involving competition between all airlines. Finally, there is much less dispersion evident in flights involving competition between Virgin and Jetstar.

All available fares

Table 3 contains estimates of (1) using our full dataset, which incorporates separate observations for all available fare classes. In Model 5, we include indicators for the number of days until travel (coefficients not reported) and the class of fare. In Model 6, we include indicators for the number of days until travel interacted with fare class. These are separately reported in Figure 4. Relative to our earlier results, the effect of the time and day of travel has an apparently muted relationship with fares, reflecting the reduced sensitivity of fares for flexible tickets to those factors.

Table 3. Price Dispersion, All Fare ClassesThumbnail image of
image

Figure 4. All Fare Classes and Booking Day

Download figure to PowerPoint

As we would expect, premium fare classes are clearly associated with higher average fares. After controlling for route and flight characteristics, a Qantas business class seat costs on average $A540 more than the lowest class of Qantas fare. The difference is less dramatic on Virgin and Jetstar flights, in part because those airlines do not offer business class. For those airlines, the average fare for a flexible economy ticket is around $A200 higher than the fare for the most restricted ticket.

This difference reflects both cross-sectional and time-series variation. First, at a given time, a customer chooses among a set of fare classes for a flight. The difference in prices across fare classes reflects a combination of product differentiation and price discrimination. Second, one method of intertemporal price discrimination is to reduce the set of available fare classes as we approach the date of travel. The cheapest fare classes are commonly no longer offered as the travel date becomes close. This selection effect also contributes to higher average fares on higher fare classes.

The behaviour of fares of different classes as we approach the travel date is captured in Figure 4, where each airline panel depicts the coefficients on the interaction terms between the number of days until travel and the fare class. Confidence intervals are extremely tight and are omitted. In the last 5 weeks before travel, only the cheapest Virgin and Jetstar fare classes rise in price as we approach the travel date.8

There is little intertemporal price discrimination for premium fare classes. Upon cancellation of a journey (even by ‘no show’), the Qantas business and fully flexible economy tickets retained their full value; that is there was no cancellation charge. If Qantas had made a practice of increasing these fares as the travel date approached, businesses would have had an incentive to buy such tickets well in advance, and then cancel unwanted tickets.

For the other fare classes, Virgin (and, to a lesser extent, Jetstar) secured rising fare paths by a combination of fare changes within the fare class, and restrictions placed on the availability of fare classes. Qantas relied to a greater degree on progressive withdrawal of fare classes, rather than on adjustment of fares for a given ticket type.

(ii) The Effect of Route Characteristics

Our route-specific fixed effects model allowed us to examine the relationship between airfares and the time between booking and travel, after controlling for route characteristics. To examine the effect of specific route and flight characteristics on airfares, we now turn to Table 4 which contains estimates of Equation (2) based on our lowest available fares dataset. We are particularly interested in the relationship between competitive conditions and average fares. Again, we stress that, owing to the joint determination of market structure and fares, our estimates of the effect of competition on fares should be treated with caution.

Table 4. Competition and FaresThumbnail image of

Models 7–10 differ only in their treatment of our competition variables. In all models, on average, relative to a Virgin flight, there is again a premium for travel on Qantas flights, and a discount for Jetstar travel. Naturally, longer routes are associated with higher fares. However, rising fuel prices have had no obvious positive impact on fares, with a negative relationship evident between fares and fuel prices for short routes. Fares are lower linking origins and destinations with higher populations. This may reflect economies of scale in airport operation for centres with greater population density. There is also a small fare premium for travel on routes we have designated as either predominantly business or predominantly leisure.

Model 7 has indicator variables for the identities of the airlines operating on the route on the day of the flight. The case with a single operator on the day of travel is made the numeraire. Not surprisingly, greater competition on the day of travel is associated with lower average fares, with the lowest fares observed when all three airlines operated on the day. Model 8 instead includes indicator variables for the airlines operating within 90 min of the flight. This is designed to capture the effect of more immediate temporal competition. The results are similar, with lower fares associated with a greater number of operators.

Model 9 includes both temporal and potential competition variables. Of our two proposed competition variables, one might suspect that the temporal competition variables would be more salient. After all, they target the competitive environment in the time immediately surrounding the flight. Instead, once we have controlled for the airlines operating on a given day, the effect of direct temporal competition is much reduced. Further, there is a positive and significant association between fares and temporal competition between Jetstar and Qantas relative to a monopoly numeraire. One interpretation of this last result is that times in which both Jetstar and Qantas fly are relatively attractive. In those cases where Virgin does not contest that submarket, it is not surprising that fares are relatively high. More generally, our estimates of the effect of temporal competition are confounded by a selection effect: airlines are more likely to schedule competing flights in high demand periods.

Model 10 includes interaction terms between the identities of the airlines operating on the day, and the airline operating the current flight. For Qantas flights, substantially lower fares are evident in the presence of competition. For Jetstar flights, there is still a marked negative association between competition and fares. However, for Virgin flights, the effect is substantially muted. In fact, competition with Qantas has no significant association with fares. This result further highlights the joint determination of fares and market structure. We might reasonably expect fares to be higher on routes with a single carrier. In the case of Virgin, this is not true, suggesting a selection effect; routes for which Virgin acts as a monopolist are likely to be comparatively unattractive routes to fly. This endogeneity of market structure will bias our estimates of the impact of competition. The dampened association between competition and fares on routes serviced by Virgin suggests that these routes are both less profitable and encourage lower fares relative to other routes. This is consistent with these routes exhibiting lower demand as opposed to higher operating costs. In turn, this provides suggestive evidence that our estimates understate the impact of competition.

(iii) Discussion

Most of the results we observe in Tables 2–4 are consistent with a priori expectations. For example, premiums are observed on high-demand periods including school holidays and peak times, such as Friday evenings. Intuitively, demand is likely to be more inelastic during these periods.

Our results show a substantial degree of price dispersion, much of which is explained by variation in the elapsed time between booking and travel. The average lowest available fare rises by approximately $A70 over a 35-day period prior to the day of travel. This represents around 40 per cent of the average cost of a ticket in our dataset. Although we have not explicitly tested for competing explanations for this dispersion, we believe there is evidence indicating some role for price discrimination. In particular, coefficients on our booking day indicator variables are precisely estimated, and a great deal of the variation in fares can be explained by variation in booking days. This is clearly consistent with price discrimination; it is generally accepted in the literature that some passengers, especially those travelling for business, tend to have lower price elasticity of demand and are more likely to purchase tickets close to the day of departure. However, if the explanation were principally demand uncertainty, we would expect fare paths to reflect the realised values of uncertain demand, yielding an imprecisely estimated relationship between fares and booking time.

The results for the ‘all fare classes’ database reveal the mechanism by which fare variation is achieved. All airlines make a fully flexible economy class airfare available at all times. Qantas adjusts the lowest available economy class fare mainly by varying the set of available fare classes over time as the travel date approaches. Virgin and Jetstar employ this mechanism, but often supplement it by also varying the price of the most restricted economy class fare over time. Because airlines vary prices over time within a fare class, theories of price dispersion based on rigid prices (such as Dana (1999)) have limited relevance in the Australian context.

One could argue that our observations bear some relation to the theory of Gale and Holmes (1999), who suggest that advanced purchase discounts are a profitable tool in the presence of demand uncertainty and customer heterogeneity. An expectation on the part of customers that fares will rise could play a similar role to explicit advanced purchase discounts. Customers placing a high value on flexibility can purchase a fully flexible economy class airfare, which is available on all airlines on all booking days. Customers who place a lower value on flexibility can purchase a restricted economy airfare well in advance of travel, in the expectation that they have secured a discount.

The ambiguous relationship between the extent of price dispersion and competition suggests that further work is needed in this area. We found the least price dispersion in markets in which the two low cost airlines, Virgin and Jetstar, compete head to head.9 A greater degree of price dispersion was evident in monopoly routes. However, the greatest dispersion was found in routes pitting Qantas and Virgin. An explanation might be sought in the different cost structures of the airlines, or in the different industry and firm-specific demand elasticities evident on the different routes.

A further issue that we have not addressed is the nature of the strategic interaction between the airlines. Interesting dynamics are indeed evident on individual routes in our dataset, but these dynamics are masked by the aggregative nature of our analysis. This issue is particularly interesting given the formative nature of competition in the market, and the ownership structure of the market, with two of the main airlines owned by a single firm.

VI Concluding Remarks

  1. Top of page
  2. Abstract
  3. I Introduction
  4. II Related Literature
  5. III Data and Industry Background
  6. IV Empirical Model
  7. V Results
  8. VI Concluding Remarks
  9. References

In this paper, we exploit a unique self-collected database of airfares to characterise the nature of price variation in the Australian domestic aviation market. The detailed nature of our dataset allows us to adopt a fixed effects estimator to uncover the primary mechanisms used for price variation. Because fare variation is implemented by both adjusting the set of available fares and by varying fares within a fare class over time, theories of price dispersion according an important role to rigid prices are not suited to the Australian context.

We find dramatic variation in airfares offered, both across fare classes for a given booking day, and across booking days. Lowest available fares on a given flight sometimes rise steeply over time, particularly in the week before travel. Price variation by booking day is evident for economy fares, but not premium fares, and is greatest for Qantas. We find the relationship between competition and price variation to be ambiguous. The greatest variation in prices is observed on flights involving competition between the two main airlines, Qantas and Virgin. However, compared with monopoly routes, less variation is evident on flights with competition between Virgin and the Qantas subsidiary, Jetstar. We also present evidence suggesting that the impact of competition on the level of fares is substantial, with discounts likely to exceed $A20 on duopoly routes and $A35 on triopoly routes relative to a monopoly benchmark.

Footnotes
  • 1

     For accounts of the recent Australian aviation context, see Hooper and Findlay (1998) on policy changes regarding deregulation and privatisation; Ergas and Findlay (2003) and Forsyth (2003) on the Virgin Blue entry and some issues facing regulators; and Graham and Vowles (2006) on the early development of Jetstar.

  • 2

     The majority of our data were collected through an automated routine based on original computer code, while, early in our sample, data were collected manually.

  • 3

     This practice stems from an agreement with the major airlines brokered by the Australian Competition and Consumer Commision (ACCC Press Release, 13 May 2002). This agreement broke down for approximately 3 months in early 2005. For flights in this period, we have added in the appropriate taxes and charges.

  • 4

     The modest difference between the numbers of observations for the two measures results from our omitting some cases where we failed to collect data for all the airlines flying the route on the day.

  • 5

     All empirical models that we consider have the level of fares as the dependent variable. We have also estimated each model using the logarithm of fares as the dependent variable, with qualitatively similar results. The results for logarithms of fares are available on request.

  • 6

     Unfortunately, there are few instances of entry in our sample, and none that we are aware of during the process of data collection for a specific flight. Consequently, it is not entry that is identifying our estimates.

  • 7

     More precisely, if airlines are attracted to operate on routes in which demand and consequently prices are higher, then our estimates will understate the impact of competition on fares. Conversely, if airlines are attracted to operate on a route because costs tend to be lower, our estimates will overstate the impact of competition on fares. Our results are suggestive that we understate the impact of competition. See Section V(ii), and the discussion surrounding Model 10, in particular.

  • 8

     The apparent drop in Business class and Fully Flexible Qantas fares (the top two lines in the Qantas panel of Figure 4) 35 days before travel reflects a selection effect. Only a subset of our data contains these fare classes more than 5 weeks before travel. There is no apparent drop in fares over time in these fare classes.

  • 9

     While Jetstar is a subsidiary of Qantas, Jetstar is marketed as a no-frills, low-cost alternative, and efforts are made by Qantas to separate important aspects of its operations.

References

  1. Top of page
  2. Abstract
  3. I Introduction
  4. II Related Literature
  5. III Data and Industry Background
  6. IV Empirical Model
  7. V Results
  8. VI Concluding Remarks
  9. References
  • Bachis, E. and Piga, C. (2006), ‘Hub Premium, Airport Dominance and Market Power in the European Airline Industry’, Rivisti di Politica Economica, 96, 1154.
  • Bilotkach, V. (2006) ‘Understanding Price Dispersion in the Airline Industry: Capacity Constraints and Consumer Heterogeneity’, in DarinLee (ed.), Advances in Airline Economics, Vol. 1. Elsevier, Amsterdam; 32945.
  • Bilotkach, V., Gorodnichenko, Y. and Talavera, O. (2010), ‘Are Airlines’ Price-setting Strategies Different?’, Journal of Air Transport Management, 16, 16.
  • Borenstein, S. and Rose, N.L. (1994), ‘Competition and Price Dispersion in the US Airline Industry’, Journal of Political Economy, 102, 65383.
  • Button, K., Costa, A. and Cruz, C. (2007), ‘Ability to Recover Full Costs Through Price Discrimination in Deregulated Scheduled Air Transport Markets’, Transport Reviews, 27, 21330.
  • Dana, J. (1999), ‘Equilibrium Price Dispersion under Demand Uncertainty: The Roles of Costly Capacity and Market Structure’, The Rand Journal of Economics, 30, 63260.
  • Ergas, H. and Findlay, C. (2003), ‘New Directions in Australian Air Transport’, Agenda, 10, 2741.
  • Forsyth, P. (2003), ‘Low-cost Carriers in Australia: Experience and Impacts’, Journal of Air Transport Management, 9, 27784.
  • Gale, I. and Holmes, T. (1993), ‘Advance-purchase Discounts and Monopoly Allocation of Capacity’, American Economic Review, 83, 13546.
  • Gallego, G. and Van Ryzin, G. (1994), ‘Optimal Dynamic Pricing of Inventories with Stochastic Demand over Finite Horizons’, Management Science, 40, 9991020.
  • Gerardi, K. and Shapiro, A. (2009), ‘Does Competition Reduce Price Dispersion? New Evidence from the Airline Industry’, Journal of Political Economy, 117, 137.
  • Gillen, D. and Hazledine, T. (2006), ‘The New Price Discrimination and Pricing in Airline Markets: Implications for Competition and Antitrust’, Paper for XIV Pan-American Conference of Traffic & Transportation Engineering, September, Canary Islands.
  • Goolsbee, A. and Syverson, C. (2008), ‘How do Incumbents Respond to the Threat of Entry? Evidence from the Major Airlines’, Quarterly Journal of Economics, 123, 161133.
  • Graham, B. and Vowles, T. (2006), ‘Carriers within Carriers: A Strategic Response to Low-cost Airline Competition’, Transport Reviews, 26, 10526.
  • Hazledine, T. (2010), ‘Pricing, Competition and Policy in Australasian Air Travel Markets’, Journal of Transport Economics and Policy, 44, 3758.
  • Hooper, P. and Findlay, C. (1998), ‘Developments in Australia's Aviation Policy and Current Concerns’, Journal of Air Transport Management, 4, 16976.
  • Pels, E. and Rietveld, P. (2004), ‘Dynamic Pricing in the Airline Industry’, Journal of Air Transport Management, 10, 27983.
  • Pitfield, D.E. (2005a), ‘Some Speculations and Empirical Evidence on the Oligopolistic Behaviour of Competing Low-cost Airlines’, Journal of Transport Economics and Policy, 39, 37990.
  • Pitfield, D.E. (2005b), ‘A Time Series Analysis of the Pricing Behaviour of Directly Competitive “Low-cost” Airlines’, International Journal of Transport Economics, 32, 1539.
  • Puller, S., Sengupta, A. and Wiggins, S. (2009), ‘Testing Theories of Scarcity Pricing in the Airline Industry’, NBER Working Paper No. 15555, National Bureau of Economic Research, Cambridge, MA.
  • Stavins, J. (2001), ‘Price Discrimination in the Airline Market: The Effect of Market Concentration’, Review of Economics and Statistics, 83, 2002.
  • Talluri, K. and Van Ryzin, G. (2004), The Theory and Practice of Revenue Management, Springer, New York, NY.