preliminary version: do not cite without authors’ … · 2005. 3. 18. · enrico bachis...
TRANSCRIPT
Do prices grow more in Euro-land? Evidence from the
airline industry
Enrico Bachis
Nottingham University Business School
Claudio Piga*
Economics Department,
Loughborough University
March 2005
Abstract
We study annual changes in fares posted by airlines on their web sites over the period June 2002- June
2004. Using panel data techniques, annual fare changes are regressed against a set of country specific,
fixed effects and control variables accounting for costs on a specific route, inflation and exchange rate,
route competitive characteristics and seasonal fixed effects. The estimates provide various insights on
the process underlying price setting by airlines. In particulars, they reveal an important relationship
between airfare changes and exchange rate fluctuations. Finally, the study suggests the presence of
country effects, that do not conform however to the notion that airfares to Euroland destinations
followed a similar trend.
JEL classification:L11, L13, L93
Keywords: Price discrimination; Panel Data.
PRELIMINARY VERSION: DO NOT CITE
WITHOUT AUTHORS’ PERMISSION
*Contact Author. Address for correspondence:
Economics Department
Loughborough University,
Leicestershire, UK, LE11 3TU
Tel: +44 (0) 1509 222701
Fax: (+44)-(0) 1509 223910
2
1. Introduction
“Although […] nominal exchange rates [of the euro-zone members] have been set in stone,
their real exchange rate – i.e., adjusted for differences in their rates of inflation – have shifted
significantly. […] Real exchange rates measure international competitiveness, and are therefore more
important than nominal rate in terms of their economic impact.”
The Economist, February 19th-25th 2005
“[…] the European Commission has announced it will investigate allegations that Apple is
unfairly charging up to 20 per cent more for music downloads in the UK than it does in the rest of
Europe. Apple charges UK users about 83p a track compared with about 52p in France and Germany.
Apple says its pricing was based on "the underlying economic model in each country”.
Financial Times, 4th March 2005
As Motta (2003) posits: “price discrimination is a pervasive phenomenon, of which examples
from our daily life abound” (p.491). A recent example is given by the differential pricing made by
Apple in the electronic music market. Quite relatedly, a visit to the web site of Tele2, the international
telecommunication company, will reveal that it charges 12.5 eurocent per minute (plus a fixed charge of
15.5 eurocent at the conversation start) for calls from Italy to UK, while residents in UK pay their calls
to Italy only 2.5 pence per minute (and no fixed charge at the start).1 At the current exchange rate of 0.7
£/euro, this corresponds to a situation where the Italian residents are charged four times as much as UK
ones.
A company’s ability to charge different prices for the same product in different countries is an
instance of “third degree price discrimination” (Motta, 2003). To succeed, such a strategy entails the
absence of arbitrage, thus enabling firms to charge in accordance to each country’s willingness to pay
for a product/service. Two recent developments have brought back the issue of price discrimination to
the forefront of economic analysis. First, the Internet, whose features includes low search costs, low
barriers to entry and easier price comparability, was considered a technological advancement capable of
bringing about a business environment replicating the characteristics of a perfectly competitive market.
However, doubts were immediately cast on the likelihood that the Internet would represent “frictionless
markets” (Brynolfsson and Smith, 2000).
Second, on the 1st of January 2002, twelve European countries started to adopt a common
currency, the Euro. In his speech in Maastricht on 6th February 2002, the then President of the European
Central Bank stated that “the introduction of the euro has increased the transparency of prices between
1 Information retrieved from the sites www.tele2.it and www.tele2.co.uk on March, 14th 2005.
3
countries and regions…This increased price transparency will trigger more cross-border trade and
commerce and, hence, competition”. Contrary to these predictions, anecdotal evidence reported on the
press seems to suggest a huge rise in the price level after the Euro changeover, although this was
particularly severe in sectors protected by international competition. Formal studies of the impact of the
Euro introduction have also been conducted, where price comparisons are made before and after the
changeover. Baye et al. (2002) find a significant increase for the case of electronic consumer goods,
while Goldberg and Verboven (2004) find a reduction in car prices, with some evidence of common
dynamics in price differentials within the euro-zone with respect to no euro-zone members.
It is against this backdrop that we aim to study the determinants of annual price changes in the
airline industry. To do so, we rely on primary data obtained by retrieving the fares posted on their web
sites by the airlines for flights within UK, and from UK to other European countries. Given the period
covered (June 2002 until June 2004), we cannot evaluate changes before and after the changeover, but
we can ascertain whether the airlines systematically increased their fares more to take advantage of the
turmoil following the euro adoption, that seemed particularly severe in the second half of 2003. In the
jargon previously used, we aim to check whether airlines discriminate across countries.
Because of the seasonality characterizing airfares, the annual price changes are worked out, for
the same company operating on a given route, by using a twelve months lag. To check for robustness,
changes were obtained for the mean, median and minimum price values from the distribution of daily
prices collected within a company-route-month combination. These were in turn regressed against a set
of country specific, fixed effects and control variables accounting for costs on a specific route, inflation
and exchange rate, route competitive characteristics and seasonal fixed effects.
As far as country differences are concerned, the evidence suggests that they are important (even
after costs, inflation and exchange rate differentials are taken into account). Thus, it would seem that
airlines can identify factors that are unknown and unobserved by the authors, but that affect, in each
point in time and in each country, a person’s willingness to pay for a ticket. However, the evidence
neither supports the results in Baye et al (2002) of price increases in the Euro area, nor those in
Goldberg and Verboven (2004) of systematic fares reductions. Indeed, the evidence is more mixed, with
increases and decreases in both Euro and non-Euro adopters, although relative to UK domestic route,
the minimum fares seem to have decreased significantly in almost all European countries, thereby
suggesting a strategy used by airlines to boost demand via promotional offers.
As the citation from the Economist indicates, it is still possible that important country effects
operate through the joint impact of inflation and the fluctuations of the exchange rate. The estimates
reveal the importance of the latter: the devaluation of the British Sterling relative to the Euro appears to
be associated with significant fare increases of about 3%. Prima facie, such a result may be counter-
intuitive: why would airlines increase fares when the Sterling devaluation is already making a trip to a
European destination more costly for UK residents? There are various simple explanations. First, flights
4
are used not only by UK residents, but also by visitors from other European countries, who benefit from
their currency appreciation. Hence, a fare increase following, say, a Euro appreciation is a simple way
to extract some surplus from visitors to UK. Second, some studies reveal how the demand of British
tourists for trips to other European countries is generally inelastic (Li et al. 2003).
The paper is organized as follows. The next section describes how the fares were collected, and
the nature of the secondary data used to describe the traffic on a route. Section 3 provides some
descriptive statistics derived from the original price dataset, plus information on the estimation sample
characteristics. Section 4 outlines the econometric models, whose results are commented in the Section
5. Some conclusive remarks are included in the final Section 6.
2. Data Collection
Most of the empirical contributions on pricing behaviour in the Civil Aviation Sector have
focused, so far, mainly on the U.S. market, whereas few contributions have been devoted to the
European market. Moreover, the analysis of the U.S. aviation sector has been mostly conducted relying
on the same dataset, namely the Databank of the U.S.A. Department of Transportation’s Origin and
Destination Survey, which is a 10 percent yearly random sample of all tickets that originate in the
United States on U.S. carriers (Evans and Kessides, 1993; Borenstein and Rose, 1994; Hayes and Ross,
1998; Alam et al., 2001; Stavins, 2001; Baylis and Perloff, 2002; Liu, 2003). In this dataset prices are
measured as one-way fares and are computed as one-half the reported fare round-trip tickets. All tickets
other than one-way and round trips are excluded.
In contrast, our analysis is based on primary data on fares and secondary data on routes traffic.
Initially, when the project began in May 2002, fares were collected using an “ electronic spider”, which
connected directly to the websites of only the main Low Cost Airlines (henceforth LCA) (i.e., Ryanair,
Buzz, Easyjet, GoFly) operating in Britain. Collection of fares for flights operated by traditional carriers
(i.e., British Airways, Air Lingus, Air France, Lufthansa, KLM, Alitalia, Volare, Iberia, SAS, Tap
Portugal, Air Europa and Maersk) started in March 2003: in this case, fares were collected only for
flights that traditional carriers operated on routes similar or identical to those where a LCA also flew.2
This decision was necessary to reduce the number of markets under study, where each market is
identified in this study as an airport pairs combination.
The datasets includes daily flights information operated from June 2002 up to, and including,
June 2004, for a total of 25 months. Over the period of analysis, a number of important events took
place which are somehow reflected in the dataset. First, a series of takeovers occurred: GoFly was
2 The airfares of the traditional companies were collected from the website www.opodo.co.uk, which is owned and managed by British Airways, Air France, Alitalia, Iberia, KLM, Lufthansa, Aer Lingus, Austrian Airlines, Finnair and the global distribution system Amadeus. This means that prices listed on opodo are the
5
acquired by Easyjet (December 2002) and Buzz by Ryanair (March 2003). Second, new LCA began
their operations: the “spider” was updated to retrieve fares from the Bmibaby and MyTravelLite sites.
However, due to technical difficulties, fares from Flybe, which was already an established LCA, and
Thomson Fly, a new entrant could not be obtained. This is not important, as far as sample selection
issues are concerned, as traffic operated by every airline is used to construct market indicators.
Prices were gathered for both the flights originating in a British airport bound to a continental
European destination (or another UK domestic airport) and the flights originating from a continental
European airport (or another UK domestic airport) arriving to a British destination. We collected the
fares for departures due, respectively, 1, 4, 7, 10, 14, 21, 28, 35, 42, 49, 56, 63 and 70 days from the
date of the query. So, for instance, if we consider London Stansted-Rome Ciampino as the airports pair
of interest, and assume the query for the flights operated by given airline was carried out on April 1st
2003, the spider would retrieve the prices for both the London Stansted-Rome Ciampino and the Rome
Ciampino-London Stansted routes for departures on 2/4/03, 5/4/03, 8/4/03, 11/4/03 and so on. The
return flight for both types of journeys was scheduled one week after the departure. In this way, we
were able to retrieve fares for any flights due to depart any number of days from the date of the query.
For the case of routes where an airline operates more than one flight per day, all fares for every flight
were collected. Thus, for every daily flight we obtain up to 13 prices that differ by the distance from the
day of departure. The main reason to do so was to satisfy the need to identify the evolution of fares -
from more than two months prior to departure to the day before departure – which has been noted to be
very variable for the case of LCA (Pels and Rietveld, 2004; Giaume and Guillou, 2004). While the
spider could have retrieved any number of prices, in practice the need to reduce both the number of
queries made to an airline server and the time of programme execution to a manageable level, led to the
choice of 13 prices. Furthermore, given the site characteristics, it was impossible to collect traditional
carriers’ fares 1 and 4 days prior to departure: it was also decided to omit collecting fares from these
companies for flights due to depart more than 49 days after the query. Thus, for traditional carriers, up
to 8 fares per daily flight are available.
Over the 25 months period, prices from UK for flights to and from Austria, Belgium, Czech
Republic, France, Germany, Greece, Ireland, Italy, Netherlands, Norway, Portugal, Spain, Sweden,
Switzerland were considered. Furthermore, flights for the main UK domestic routes were also included.
The collection of the airfares has been carried out everyday at the same time in order to prevent
inconsistencies due to spanning price changes during the inquiries. In fact, prices are updated
dynamically based on yield management algorithms and could change during a day3. Our approach was
official prices of each airline, although Opodo may not report promotional offers that each airline may offer on its web sites.
3 Actually, we have observed that airlines change the prices posted on Internet overnight, keeping airfares unchanged during the day.
6
then such as to reduce the probability to get spurious variation due to fare changes. In addition to
airfares we collected the name of the company, the date of the query, the departure date, the scheduled
departure and arrival time, origin and destination airports and the flight identification code.
Fares are collected before tax and handling fees4 and recorded in the currency of the country
where the journey originates. Therefore, fares for flights from a European destination are converted in
British Sterling using the current and appropriate exchange rate available from Thomson Datastream.
To complement the price data, secondary data on the traffic on all routes and all airlines flying
to the countries indicated above was obtained from the Civil Aviation Authority (henceforth, CAA).
These data report, for each flight, the name of the company, the origin and destination airports, the
departure date, the actual arrival time, the number of seats for airplane, the number of actual carried
passengers and the load factors. Moreover, the CAA dataset gives account of the flight identification
code, which proved to be precious for our research, as it made possible to merge the information on
traffic with the data retrieved from the airlines websites on prices. However, this feature is not
particularly useful in this study, as price changes are calculated as differences in the same months but in
different consecutive years, for all the flights on a route.
3. Data Analysis
Tables 1 to 3 provide a breakdown of mean, median and minimum fares, denoted in British
Sterlings, by airline, day of the week and year.
Given the data collection discussed in the previous section, these fares’ values were first
obtained by averaging out the 13 (or less, if not all available) fares for each daily flight, and then
derived for each sub-category in the tables. Thus, the mean, median and minimum fares for Ryan Air in
the week-days of 2002 were obtained from the distribution of average fares for the 15633 daily Ryan
Air flights in our price dataset. The last column reports the total number of daily flights per company in
the price dataset. A similar approach was followed for the construction of our dependent variables (see
below).
Average prices appear to have changed over the years, with a clear downward trend for the
LCA, and a less clear-cut variation for the traditional carriers, although an overall upward trend seems
to emerge. Changes in the median prices are in line with these findings, whereas minimum prices
appear to have decreased across all companies over the years. As expected, weekdays fares tend to be
cheaper than in week-ends, although the difference is not as conspicuous as one might expect.
Furthermore, note that median prices are lower than the average. Because this may be due to the
presence of high fares that airlines may post for a number of strategic reasons (i.e., trial and error
7
evaluation of demand, signal confusion to competitors etc), but that do not translate into sales, it is
likely that the median price constitutes a better indicator of the representative price at which sales are
realised.
The main strength of the data set is its size (more than 739000 daily observations) and its
duration (25 months). As far as its composition is concerned, Table 1 to 3 clearly highlight how the
majority of observations is for LCA operated flights. This is, however, a misleading viewpoint, as the
observations from traditional carriers, taken as a whole, make up for almost a third of all the
observations, thus matching the importance of the two main LCA, Ryan Air and Easyjet. Recall that a
market may be identified as either an airport pair (route) or a city pair (say, London to Paris). Based on
these definitions, the datasets provide a good description of the rivalry between LCA and traditional
carriers in markets where they both operate. Furthermore, it also includes information on pricing in
markets where LCA operate alone.
Table 4 reports the monthly structure of the civil aviation market in the United Kingdom during
the period under investigation. The entire population-headed columns report the market structure as a
whole on the base of the CAA statistics, where a market is identified as a route. Around 78% of the
total routes appear to have been operated by one monopolistic company, whereas approximately 17%
are duopolistic. This leaves a merely 5% to potential competitive routes, in which 3 or more airlines
operate. It is a clear indication of a market in which companies try to avoid direct competition by
operating flights that serve small and not serviced markets.
When it comes to our price sample, it obviously includes less markets, as the majority of routes
reported by the CAA are operated by traditional carriers (in particular, in UK, by British Airways and
BMI – British Midlands). However, in terms of percentage composition, it turns out to be a good
representation of the UK market, with little differences in the distribution of routes among different
types of markets. These differences are most likely due to the way routes for traditional carriers were
chosen. Indeed, the price sample does not include any route where a traditional carrier is the only firm.
This may also account for the fact that routes with two or more airlines are slightly over-represented.
In terms of the composition of the estimation sample, we have included in the analysis only
those carrier-route combinations for which we have collected at least 15 daily flights (and
corresponding average daily prices) in a month. Although this leads to a more varied distribution of
routes among different markets as shown in Table 4, it certainly makes the analysis more precise, as it
avoids the calculations of annual changes based on a limited number of observations. Furthermore, the
estimation sample is reduced by the need to have, for the same company-route combination, reliable
information on traffic and prices for the same months in two consecutive years.
4 This allows to avoid spurious information due to changes in the agreements with the airport local authorities, or to variations in the local tax rates. The second reason is that this made more feasible the collection
8
More precisely, as far as our estimation sample is concerned, the analysis is carried out on 189
route-company pairs, where a route is a pair of origin and destination airports located in the European
countries, and company refers either to one of the 3 low-cost airlines (excluding Buzz and GoFly, for
which no comparison for fares lagged twelve months can be done) or to one of the 10 traditional
carriers reported in Tables 1 to 3. A maximum number of 90 routes is considered. As indicated above,
the data are available between June 2002 and June 2004, with a total number of 25 monthly. However,
because we focus the attention on the annual percentage change of various measures of airfares, the
econometric analysis will be actually carried out on 13 time units, from June 2003 to June 2004. That is,
observation tagged as June 2003 will refer to changes in various measures of price and price indexes
that occurred to June 2003 prices relative to prices for the same company-route pair in June 2002. This
implies that the base period is always twelve months before the period the observation refers to.
4. The empirical model
Two types of dependent variables are used in the study. The first represents the percentage
change in the price ratio of two consecutive years, obtained as:
12,
,ln__ln_−
=ti
ti
P
PpricesXratio
where i denotes a company-route combination, and t a time index expressed in months. As
discussed above, the X index corresponds to the average, median and minimum values over the relevant
months’ distributions of fares.
The second set of dependent variables is obtained by deriving a Fisher price index which is
calculated as:
Pti
Lti
Fti PPP ,,, = where L
tiP , is the Laspeyres’index calculated as: ∑∑
−−
−=12,12,
12,,,
titi
titiLti qp
qpP where
pi,t (pi,t-12) is the price per company-route in month t (t-12) and qi,t-12 is the number of passengers per
company-route in month t-12, whereas PtiP , is the Paasche’index calculated as:
∑∑
−
=titi
titiPti qp
qpP
,12,
,,, where pi,t (pi,t-12) is the price per company-route in month t (t-12) and qi,t-12 is the
number of passengers per company-route in month t. As usual, the relevant prices used are the average,
median and minimum values over a given company-route-month distribution of prices.
We consider two specifications. In the first, the dependent variables are assumed to vary with
the number of companies at route level. In the second, annual changes are tested on the number of
companies at city pair level. In doing this, we check whether airlines compete within city markets rather
of data.
9
than at route level. The remaining factors likely to exert an influence on the year-to-year airfare changes
are then grouped by costs, market characteristics, seasonal dummies and geographical characteristics.
Costs
A potential source of the annual changes in the airfares is certainly represented by the cost of
oil, a proxy for the cost of the fuel used by the airlines. What we expect is a positive relationship
between the year-to-year change in the price of airfares and the cost of oil. However, the consumption
of fuel varies with the different phases of the flight, such that when the aircraft flies at cruising speed
fuel employment is much lower than when the airplane takes off. Thus, in order to capture these
economies of scale we interact the cost of oil with the distance - as expressed in miles - between the two
endpoints of each route. The variable ln_cost represents this interaction.
Other potential determinants of the annual change in the airline prices are represented by the
inflation and the exchange rate of the euro and the other European currencies with respect to the British
pound. The variable inflation is expected to exert a positive influence on the annual fare changes, as it
represents a broad measure of the price dynamics within each country. In contrast, D_exchange_rate -
the annual percentage change in the exchange rate - has a less straightforward interpretation. A decrease
in the value of this variable indicates an appreciation of the European currencies with respect to the
sterling pound. This, in turn, means airlines face higher costs for goods and services denominated in
European currencies. Thus, a depreciation of the pound increases costs for the majority of UK based
airlines we survey. However, there exists a more compelling, demand-side hypothesis. An appreciation
of the continental Europe currencies should make travelling to UK cheaper for continental European
residents, whereas British travellers are expected to face higher costs. Thus, a fare increase is simply
likely not to influence the former’s decision to travel, as the holiday costs are reduced thanks to their
currency’s appreciation. Given this situation, it would be sensible for airlines to increase fares following
a continental European currency appreciation. On the other hand, such an attempt to extract more
surplus from continental Europeans may reduce the traffic of British travellers. The expected effect of
D_exchange_rate should therefore be neutralized by these opponent forces, assuming that price
elasticities are identical among these two groups of travellers. However, since the majority of route
served by LCA can be considered, to a great extent, tourist destinations and given the documented short
run inelasticity of the British tourist demand towards European countries (Li et al. 2003), we expect a
negative coefficient and, therefore, a positive contributions in terms of the annual change in the posted
airfares.
Market characteristics
An important set of control variables is the one related to market characteristics. Market
structure is indeed commonly seen as an important determinant of changes in prices, and so are market
density and external shocks to the market composition. In the former subset of variables we control for
the effect of the number of competing firms both at route level and at city pairs level. At route level we
10
are able to control for monopolistic, duopolistic and three-firm markets, whereas at city pairs level we
can check for up to six companies. In line with the literature we expect that prices are declining with the
number of firms operating in the same market.
Another variable capable to represent market structure is Dummy_charter, which is a
categorical variable for the presence of charter flights on a route. The effect of this variable should be
negative, as a consequence of an augmented competition and a higher cross-elasticity of demand across
flights. However, the presence of charter flights also indicates seasonal variations on the demand of a
specific service, and as such it could capture the increase in the shadow costs of aircraft capacity. Peak-
load pricing, as a consequence of variations in these shadow costs, could result in higher prices, and so
in a positive effect on the dependent variable.
In addition, we check for the role of market density -measured as the number of flights operated
in the preceding year (Lag_n_of_flights). We treat this as a proxy for the elasticity of demand in a
specific market, and its effect could be positive or negative. In fact, more frequent service could simply
reflect a higher demand for air travels and therefore a lower elasticity of demand for flights. This, in
turn, should enable companies to charge more. However, a high density of flights on a route also
decreases the cost of switching flights, increasing the substitutability across flights, which leads to more
competitive markets. Under these circumstances, firms would find difficult to increase prices. The
Hausman test ruled out the endogeneity of Lag_n_of_flights.
A second subset of variables aims at describing the effect on the annual change of airfares
exerted by structural breaks in the market composition. New_route_entry is a dummy variable that
identify those routes launched in the previous year by a LCA. Given its particular composition,
New_route_entry should capture the increased prices in the following year with respect to the
promotional airfares that usually accompany the launch of a new route. What we expect is, therefore, a
positive effect on the dependent variable. On the other hand, the dummies Easyjet_takeover and
Ryan_takeover capture changes that occurred when GoFly and Buzz were taken over by Easyjet and
Ryanair in December 2002 and April 2003, respectively. The effect on the regressands is expected to
be negative, as more efficient business models should enable the buying companies to operate the same
flights at lower costs, therefore reducing the monthly average airfares. Another dummy, Low_cost is
finally meant to reflect the negative impact of the low-cost airlines on the changes in the average prices.
Seasonal dummies
In order to control for seasonal variations and peak-load pricing, we introduced the three
dummy variables Jun-Sep03-03, Oct-Dec03-03, Jan-Mar04, which, ceteris paribus, should tell us about
the changes in the dependent variables with respect to the period April 2004-June 2004. Over this
period European airlines have faced one of the most difficult period in their history, due to a sluggish
growth of the main continental economies and an increase in the competition to attract reluctant
customers. Both facts have undoubtedly pushed for a reduction of the airfares, whereas less clear has
11
proved the long run impact of the currency union intervened in 2001 in most of the countries under
investigation. Although the gains from the euro changeover were presumed to be huge especially in
terms of cross border price transparency, in many euro-area members there has been a lot of complaints
about rising inflation rates. The analysis of these seasonal dummies will enable us to identify the
presence of any trend in the price changes.
Geographical characteristics.
The simplest test for the presence of third degree price discrimination, i.e., differential price
changes in different countries, is to see how much of the price difference variation is explained by each
country. To this purpose we include a set of dummies for all nations included in our sample, using UK
domestic flights as the baseline for comparison. This will give an insight into whether the airlines have
followed the same price strategy across countries, or whether they have implemented any third-degree
price discrimination. In the former case we should expect, ceteris paribus, a uniform variation of the
average airfare, whereas in the price discriminatory scenario fares should move systematically
differently on national lines.
The presence of countries who joined the euro-zone along with no euro members gives us the
opportunity to see how much of the price difference variation is explained by a nation’s participation in
the currency union. Again, were the euro members really part of a more and more integrated market -
and so subject to the same macroeconomics long run effects – we should expect a common trend in the
variation of their airfares. Whether this trend will be positive or negative will depend on which
macroeconomics force will prevail.
5. The results
We present several models in which the dependent variables are defined as both the percentage
change in the price ratio of two consecutive years (Ln_ratio_of_average_prices ) and the percentage
change in the Fisher price index of two consecutive years (Ln_Fisher_of_average_prices). We control
for average, median and minimum prices, in order to give more robustness to our findings. We use
Prais-Winsten estimations to control for heteroschedasticity, first order panel-specific correlation and
correlated panels.
Table 5 shows models testing for the presence of discriminatory pricing with the route-level
specification. In terms of variation in the average prices, it appears that, ceteris paribus, each European
country has witnessed a specific pattern with respect to the UK market. Netherlands, Germany and
Switzerland show a significant decrease in the average airfares, whereas in Sweden average prices have
fallen with respect to the UK. Among the other nations, Spain and Portugal appear to have seen a
significant decrease in the ratio of their average prices, but this result it is not robust to the use of the
Fisher index. The remaining nations show either positive or negative coefficients, but they are not
statistically significant. This fact leads to the conclusion that European countries do not compose an
12
integrated market in which airlines apply the same pricing strategy, and this proves to be true even
when we limit the comparison between no-euro, i.e. Switzerland, Sweden, Norway and Czech Republic,
and euro members, i.e. Austria, Belgium, France, Germany, Ireland, Italy, The Netherlands, Spain,
Portugal. Yet, that country-dummy coefficients are generally not statistically significant with respect to
UK does not allow to conclude either that airlines systematically price discriminate among countries.
What emerges, then, is that there are other decisive factors affecting pricing decisions at route level, and
on these grounds airlines tend to approach each country as a separate market.
Among cost variables, ln_cost exert a positive and significant effect on the year-to-year changes
of both the price ratio and Fisher price, whereas the inflation rate turns out to be not significant. This is
an expected result as it shows airfares do not follow the price dynamics of normal goods, but are very
sensitive to the oil price.
D_exchange_rate is probably one of the most interesting control variables. Any decrease of this
variable corresponds to an appreciation of the European currencies with respect to the British sterling.
As said before, this means UK based airlines face higher costs for goods and services traded in the
European currencies, while British travellers pay more for their holidays in Europe. On the other hand,
Europeans find convenient to spend their holidays in UK. On average, the estimates reveal that a 1%
appreciation of a continental European currency is accompanied by a 3% increase in airfares. This is
consistent with the airlines’ attempt to extract the surplus enjoyed by continental European travellers,
who benefit from their currencies’ appreciation, and for whom the airfares costs, considering the figures
reported in Tables 1 to 3, constitute a tiny portion of the entire trip costs. By the same token, British
travellers would not be deterred by these small airfares increases. Moreover, since Britons show a short
run inelastic demand for holidays in European destinations, airlines were probably able to charge higher
fares without endangering growth. While such a result hold for the mean and median prices, notice how
statistical significance is lost when the dependent variables based on the minimum prices are studied.
Very low prices are used by airlines, in particular the low cost ones, to increase their load factors: to
pursue such a goal, it would be pointless to take into account fluctuations of the exchange rate. The
need to cover costs remains a focus point: indeed, we find that oil prices remain highly correlated with
changes in minimum prices.
In terms of market variables, it appears that at route level the presence of charters on the same
route has not affected the change in the listed airfares, probably because charters meet the demand of a
specific section of customers, who generally buy a flight as part of a broader tourist package. Despite
expectations, charter flights appeared not be in competition with scheduled flights.
Table 5 also shows through the variable New_route_entry that airlines tend to charge very low
fares when launching a new route, as a way to attract customers. However, these promotional prices are
doomed to fade away in the long run, as companies rely on more sophisticated yield management
techniques. Among the big changes to the market structure intervened during the period under analysis,
13
the take-over of Buzz by Rynair is the one that seems to have exerted the biggest effect on the change
of average airfares. As expected, prices on the routes once operated by Buzz have decreased,
confirming that the Ryanair has a more efficient business model. Strangely enough, the takeover of
GoFly by Easyjet did not make the same impact on prices, as the coefficient turns out to be negative but
not significant. At first this appears to show the two companies had the same pricing policy, but further
and deeper analysis is necessary to identify the reasons behind this fact.
The most striking result is, however, represented by the absence of any impact of the number of
companies per route on the dependent variables. The coefficient of the variables Duopoly_route and
Three_firms_route are not significant, indicating that with respect to monopoly the presence of more
companies does not add anything to the competitive level of markets, and therefore does not exert any
pressure on the annual changes of prices. A plausible explanation may simply be that since we are
looking at price changes, and that the structure of a market tend to be rather fixed, the market structure
effects are wiped out by differencing. Table 6 shows that such a finding is robust to the analysis at the
city pair level too, confirming that we are dealing with a structural feature of the UK civil aviation
market. It would be interesting to check whether the analysis of price levels, rather than changes, yields
similar outcomes.
The positive and significant coefficient of the number of flights in the previous year
(Lag_n_of_flights) validates the hypothesis that this variable is a proxy for those routes with a higher
demand for air travels and therefore a lower elasticity of demand. In these particular routes airlines are
probably more capable of putting in place all those yield management techniques that usually allow
companies to extract more profits from customers. On the contrary, the categorical variable Low_cost
does not prove to have a robust impact on the annual changes of average prices. In fact, although Table
5 shows the expected negative and significant effect on the change of the
Ln_Fisher_of_average_prices, when it comes to the Ln_ratio_of_average_prices the significance of
Low_cost coefficient vanishes. We will see, however, that Low_cost has a robust impact on the changes
of the annual median airfare changes.
Finally, the seasonal dummies show that in the sub-periods June 2003-September 2003 and
October 2003-December 2003 airfares have decreased with respect to what happened between April
2004 and June 2004. This is in line with the difficult period airlines have faced in 2002 and 2003, when
they were forced to reorganize their activities and to compete fiercely to attract reluctant customers on
their airplanes. The positive trend in 2004 is also consistent with the statements of many UK airlines’
CEOs, who talked of an “avoided bloodshed” in the reduction of airfares for the year 2004.
The third and forth columns of Table 5 report similar results for the annual changes in the
median prices. The only difference relies on the robust significance of the country-dummy Portugal,
which points out a negative change in the year-to-year variation of the median airfares for flights from
and to Portuguese destinations. However, this little difference does not change the broad picture.
14
Estimates for the median prices confirm that neither Europe nor the euro-zone are treated as integrated
markets by the airlines. The evidence also rules out the presence of any third-degree price
discrimination, given that only 6 country-dummy coefficients out of 14 are significant and robust across
models. As for average prices, the annual changes in the median prices are rather explained by other
control variables, among which ln_cost, D_exchange_rate, New_route_entry and Lag_n_of_flights
stand out. It is worth noting that the categorical variable capturing the presence of the LCA, Low_cost,
is now significantly robust across different dependent variables, in line with the hypothesis that, ceteris
paribus, LCA have decreased airfares with respect to what traditional carriers have done. Finally, the
last two columns of Table 5 summarize the results of the estimations on the variables
Ln_ratio_of_minimum_prices and Ln_Fisher_of_minimum_prices. Again, the reported results confirm
what said so far, with the only difference that now the variable Three_firms_route shows a negative and
significant coefficient, robust across different dependent variables. This interested finding suggests that
when airlines compete with each other, their way to attract customers is by launching sensational
promotional offers like the “free flight5”one, firstly launched by Ryanair. This does not mean that
consumers enjoy lower prices on the more competitive routes, but simply that offers on competitive
markets are more eye-catching.
Table 6 reports exactly the same models as Table 5, the only difference being that we now
control for the number of firms per city pairs. This was meant to be a double-check on the robustness of
the previous estimates, but it was also meant to represent a deeper analysis on the kind of price
strategies implemented by the LCA. The hypothesis was that the absence of competition at route level
could be balanced by a strong competition at the city pair level, where many more companies are
involved. However, the estimates of Table 6 rule out this hypothesis, as none of the coefficients relative
to the number of firms proves significantly different from zero. This amounts to confirm that in the UK
civil aviation market airlines tend to avoid competition by flying on small airports, in which their
struggles do not come from stiff price competition, but from successfully attracting passengers on their
aircrafts. Besides that, Table 6 estimates represent a further proof that neither discriminatory pricing nor
euro-effect are among the determinants of the annual changes in the listed airfares.
6. Conclusions
Our empirical analysis shows that third-degree price discrimination has not exerted any strong
impact on the annual changes of the on-line listed airfares within the European civil aviation market.
We also rule out the possibility that euro-zone members compose an integrated market, in which
airlines implement homogenous pricing strategies. Consequently, we can conclude that the euro
changeover has failed so far to integrate the euro-zone members’ markets as Baye et al. (2002) and
5 Taxes excluded.
15
Goldberg and Verboven (2004) have shown for the electronic products and the car markets,
respectively. Additionally, we found that several control variables have a strong impact on the year-to-
year airfare variations, such as oil price, exchange rate, promotional offers and flight density. Finally, it
is worth noting how the UK-based LCA tend to fly to small airports in order to avoid direct
competition. However, this strategy exposes them to uncertainties on the demand side, which airlines
overcome through periodical promotional offers. The result of this strategy is to have a reduced market
power even in monopolized markets. All these conclusions are robust to a variety of controls.
These results should lead to a more precise investigation of discriminatory pricing, based on an
empirical analysis that clearly separate different European markets. In addition, the analysis of the
determinants of the price levels should try to test whether market structure is actually irrelevant for
price changes, as found for the annual percentage change.
References
Alam, Ila M. S., Leola B. Ross and Robin C. Sickles. (2001). ‘Time Series Analysis of Strategic
Behaviour in the US Airline Industry’, Journal of Productivity Analysis 16, 49-2.
Baye, Michael R., Rupert Gatti, Paul Kattuman, and John Morhan. (2002b). ‘Online Pricing and
the Euro Changeover: Cross-Country Comparisons’, mimeo.
Baylis, Kathy and Jeffrey M.Perloff. (2002). ‘Price Dispersion on the Internet: Good Firms and
Bad Firms’, Review of Industrial Organization 21, 305-324.
Borenstein, Severin, and Nancy L. Rose. (1994). ‘Competition and Price Dispersion in the U.S
Airline Industry’, Journal of Political Economy 102, 653-683.
Brynjolfsson, Erik, and Michael D. Smith. (2000). ‘Frictionless Commerce? A Comparison of
Internet and Conventional Retailers’, Management Science 46: 563-585.
Evans, William N., and Ioannis N. Kessides, ‘Localized Market Power in the U.S. Airline
Industry’, Review of Economics and Statistics 75 (1), 66-75.
Giaume, Stephanie, and Sarah Guillou. (2004). Price Discrimination and Concentration in
European Airline Market, Journal of Air Transport and Management 10(5), 293-370.
Goldberg, Pinelopi K., and Frank Verboven. (2004). EU car prices, Economic Policy, 19(40),
484-521.
Hayes, Kathy J., and Leola B. Ross. (1998). ‘Is Airline Price Dispersion the Result of Careful
Planning or Competitive Forces?’ Review of Industrial Organization 13, 523-541.
Li, Gang, Haiyan Song, and Stephen F. Witt. (2004). Modelling Tourism Demand: A Dynamic
Linear AIDS Approach, Journal of Travel Research 43(2), 141-150.
16
Liu, Qihong. (2003). ‘The Effect of Market Structure on Price Dispersion: An Analysis of the
U:S. Airline Industry’, April 2003, SUNY Stony Brook.
Motta, Massimo. Competition Policy. Theory and Practice.Cambridge University Press, 2003.
Pels, Eric, and Piet Rietveld. (2004). ‘Airline pricing behaviour in the London-Paris market’,
Journal of Air Transport Management 10, 279-283.
Stavins, Joanna. (2001). ‘Price Discrimination in the Airline Market: The Effect of Market
Concentration’, The Review of Economics and Statistics 83 (1), 200-202.
17
Table 1. Average airfares by groups of days and years. 2002 2003 2004
weekdays weekend weekdays weekend weekdays weekend
Total observations
Bmibaby - - 48.41 (12364) 51.83 (15509) 36.62 (6500) 38.21 (7890) 42263
Ryanair 49.26 (15633) 49.28 (20450) 35.94 (50082) 36.53 (65475) 34.57 (29106) 35.06 (37908) 218654
Easyjet 52.74 (13618) 55.14 (19076) 45.83 (53503) 47.40 (74196) 39.87 (32612) 41.65 (43009) 236014
Buzz 58.40 (5574) 59.85 (7445) 33.37 (3083) 36.23 (4065) - - 20167
GoFly 72.04 (8375) 74.80 (10782) - - - - 19157
Aer Lingus - - 55.90 (3737) 56.13 (4177) 49.60 (1996) 50.49 (1998) 11908
Air Europa - - 55.59 (216) 55.92 (207) 52.05 (55) 56.99 (83) 561
Air France - - 76.50 (2708) 74.08 (2989) 82.29 (1464) 79.04 (1502) 8663
Alitalia - - 73.67 (1341) 73.78 (1585) 85.93 (1018) 85.89 (1186) 5130
BMI-British Midland - - 58.90 (9839) 59.24 (12037) 60.67 (12540) 60.69 (15562) 49978
British Airways - - 71.95 (18304) 72.76 (21485) 75.02 (11526) 76.96 (13343) 64658
Czech Airlines - - 82.77 (210) 82.92 (230) 74.46 (234) 74.65 (261) 935
Iberia - - 98.39 (2356) 98.97 (2955) 102.06 (1732) 102.13 (2304) 9347
KLM - - 71.54 (1537) 68.99 (1657) 76.90 (1343) 66.29 (1598) 6135
Lufthansa - - 68.24 (7534) 68.35 (9462) 72.67 (4683) 73.35 (5837) 27516
Maersk Air - - 69.44 (276) 69.51 (338) 58.34 (235) 58.58 (237) 1086
Scandinavian Airlines - - 69.22 (2065) 69.43 (2502) 79.83 (2036) 79.92 (2532) 9135
Swiss - - 72.23 (2096) 70.63 (2296) 93.92 (1814) 93.58 (1974) 8180
Total observations 43200 57753 171251 221165 108894 137224 739487 Source: own calculations based on data collected. (number of observations)
18
Table 2. Median airfares by groups of days and years. 2002 2003 2004
weekdays weekend weekdays weekend weekdays weekend
Total observations
Bmibaby - - 42.79 (12364) 46.06 (15509) 33.88 (6500) 35.32 (7890) 42263
Ryanair 44.67 (15633) 44.79 (20450) 31.73 (50082) 32.37 (65475) 29.22 (29106) 29.50 (37908) 218654
Easyjet 48.61 (13618) 50.69 (19076) 41.30 (53503) 42.88 (74196) 34.64 (32612) 36.17 (43009) 236014
Buzz 54.35 (5574) 55.50 (7445) 31.28 (3083) 33.78 (4065) - - 20167
GoFly 69.91 (8375) 72.69 (10782) - - - - 19157
Aer Lingus - - 53.50 (3737) 53.86 (4177) 47.77 (1996) 48.86 (1998) 11908
Air Europa - - 52.00 (216) 52.21 (207) 46.71 (55) 49.19 (83) 561
Air France - - 73.81 (2708) 71.40 (2989) 73.28 (1464) 70.62 (1502) 8663
Alitalia - - 66.76 (1341) 66.95 (1585) 80.08 (1018) 79.90 (1186) 5130
BMI-British Midland - - 56.45 (9839) 56.77 (12037) 56.94 (12540) 56.89 (15562) 49978
British Airways - - 68.12 (18304) 68.88 (21485) 70.30 (11526) 72.00 (13343) 64658
Czech Airlines - - 81.09 (210) 81.55 (230) 74.97 (234) 75.03 (261) 935
Iberia - - 87.73 (2356) 88.42 (2955) 89.59 (1732) 89.52 (2304) 9347
KLM - - 64.98 (1537) 62.98 (1657) 79.45 (1343) 66.98 (1598) 6135
Lufthansa - - 62.01 (7534) 61.98 (9462) 67.05 (4683) 67.58 (5837) 27516
Maersk Air - - 65.27 (276) 65.28 (338) 58.14 (235) 58.33 (237) 1086
Scandinavian Airlines - - 64.65 (2065) 64.85 (2502) 80.04 (2036) 80.11 (2532) 9135
Swiss - - 68.08 (2096) 66.44 (2296) 88.91 (1814) 88.46 (1974) 8180
Total observations 43200 57753 171251 221165 108894 137224 739487 Source: own calculations based on data collected. (number of observations)
19
Table 3. Minimum airfares by groups of days and years. 2002 2003 2004
weekdays weekend weekdays weekend weekdays weekend
Total observations
Bmibaby - - 23.60 (12364) 25.81 (15509) 17.80 (6500) 19.11 (7890) 42263
Ryanair 13.57 (15633) 13.74 (20450) 11.27 (50082) 11.71 (65475) 8.40(29106) 8.44 (37908) 218654
Easyjet 20.80 (13618) 22.02 (19076) 20.27 (53503) 20.94 (74196) 17.62 (32612) 18.49 (43009) 236014
Buzz 32.44 (5574) 33.26 (7445) 12.32 (3083) 14.21 (4065) - - 20167
GoFly 36.76 (8375) 38.10 (10782) - - - - 19157
Aer Lingus - - 39.68 (3737) 39.66 (4177) 37.38 (1996) 37.90 (1998) 11908
Air Europa - - 42.18 (216) 42.22 (207) 39.97 (55) 40.19 (83) 561
Air France - - 66.05 (2708) 63.51 (2989) 61.94 (1464) 58.15 (1502) 8663
Alitalia - - 49.47 (1341) 49.79 (1585) 45.63 (1018) 45.56 (1186) 5130
BMI-British Midland - - 39.62 (9839) 39.91 (12037) 37.42 (12540) 37.24 (15562) 49978
British Airways - - 48.60 (18304) 49.28 (21485) 42.11 (11526) 42.77 (13343) 64658
Czech Airlines - - 57.76 (210) 58.00 (230) 48.60 (234) 48.60 (261) 935
Iberia - - 58.03 (2356) 59.48 (2955) 49.30 (1732) 49.60 (2304) 9347
KLM - - 46.29 (1537) 45.59 (1657) 39.73 (1343) 37.43 (1598) 6135
Lufthansa - - 41.84 (7534) 41.90 (9462) 40.50 (4683) 40.19 (5837) 27516
Maersk Air - - 52.03 (276) 52.18 (338) 33.63 (235) 34.19 (237) 1086
Scandinavian Airlines - - 59.40 (2065) 59.45 (2502) 56.07 (2036) 56.03 (2532) 9135
Swiss - - 56.29 (2096) 54.92 (2296) 52.50 (1814) 51.54 (1974) 8180
Total observations 43200 57753 171251 221165 108894 137224 739487 Source: own calculations based on data collected. (number of observations)
20
Table 4. Market composition by number of companies per route. Routes with one airlines (%) Routes with two airlines (%) Routes with three or more airlines (%) Total number of routes
Entire population
Sample Estimation Entire
population Sample Estimation
Entire population
Sample Estimation Entire
Population Sample Estimation
Jun 2002 424 (78.66) 78 (88.64) - 91 (16.88) 10(11.36) - 24 (4.45) 0 (0.00) - 539 88 -
Jul 2002 425 (78.85) 77 (84.62) - 89 (16.51) 12(13.19) - 25 (4.64) 2 (2.20) - 539 91 -
Aug 2002 425 (78.41) 92 (82.14) - 92 (16.97) 18(16.07) - 25 (4.61) 2 (1.79) - 542 112 -
Sep 2002 420 (78.65) 92 (81.42) - 90 (16.85) 18(15.93) - 24 (4.49) 3 (2.65) - 534 113 -
Oct 2002 427 (77.50) 94 (80.34) - 96 (17.42) 21(17.95) - 28 (5.08) 2 (1.71) - 551 117 -
Nov 2002 419 (77.31) 91 (79.13) - 98 (18.08) 21(18.26) - 25 (4.61) 3 (2.61) - 542 115 -
Dec 2002 410 (73.87) 74 (63.25) - 104 (18.74) 30(25.64) - 41 (7.39) 13 (11.11) - 555 117 -
Jan 2003 426 (76.48) 97 (75.19) - 102 (18.31) 28(21.71) - 29 (5.21) 4 (3.10) - 557 129 -
Feb 2003 427 (77.50) 96 (73.28) - 91 (16.52) 27(20.61) - 33 (5.99) 8 (6.11) - 551 131 -
Mar 2003 441 (77.78) 103 (76.30) - 97 (17.11) 27(20.00) - 29 (5.11) 5 (3.70) - 567 135 -
Apr 2003 438 (78.78) 100 (65.36) - 98 (17.63) 41(26.80) - 20 (3.60) 12 (7.84) - 556 153 -
May 2003 466 (78.19) 115 (68.05) - 107 (17.95) 41(24.26) - 23 (3.86) 13 (7.69) - 596 169 -
Jun 2003 476 (78.03) 115 (67.65) 23 (85.19) 111 (18.20) 42(24.71) 4(14.81) 23 (3.77) 13 (7.65) 0 (0.00) 610 170 27
Jul 2003 475 (78.38) 114 (66.67) 24 (82.76) 109 (17.99) 42(24.56) 5(17.24) 22 (3.63) 15 (8.77) 0 (0.00) 606 171 29
Aug 2003 473 (76.91) 141 (67.46) 28 (84.85) 123 (20.00) 54(25.84) 5(15.15) 19 (3.09) 14 (6.70) 0 (0.00) 615 209 33
Sept 2003 479 (78.65) 144 (69.23) 32 (91.43) 107 (17.57) 48(23.08) 3(8.57) 23 (3.78) 16 (7.69) 0 (0.00) 609 208 35
Oct 2003 485 (77.11) 143 (68.42) 32 (84.21) 119 (18.92) 50(23.92) 6(15.79) 25 (3.97) 16 (7.66) 0 (0.00) 629 209 38
Nov 2003 478 (77.35) 142 (68.27) 23 (88.46) 116 (18.77) 48(23.08) 3(11.54) 24 (3.88) 18 (8.65) 0 (0.00) 618 208 26
Dec 2003 474 (75.00) 146 (66.06) 31 (86.11) 127 (20.09) 54(24.43) 5(13.89) 31 (4.91) 21 (9.50) 0 (0.00) 632 221 36
Jan 2004 461 (73.17) 142 (63.96) 35 (77.78) 133 (21.11) 59(26.58) 9(20.00) 36 (5.71) 21 (9.46) 1 (2.22) 630 222 45
Feb 2004 483 (77.78) 150 (69.12) 37 (80.43) 116 (18.68) 52(23.96) 8(17.39) 22 (3.54) 15 (6.91) 1 (2.17) 621 217 46
Mar 2004 510 (77.27) 150 (69.12) 42 (79.25) 120 (18.18) 47(21.66) 9(16.98) 30 (4.55) 20 (9.22) 2 (3.77) 660 217 53
Apr 2004 522 (76.88) 149 (68.04) 56 (64.37) 126 (18.56) 54(24.66) 27(31.03) 31 (4.57) 16 (7.31) 4 (4.60) 679 219 87
May 2004 547 (78.03) 149 (69.63) 56 (63.64) 123 (17.55) 47(21.96) 26(29.55) 31 (4.42) 18 (8.41) 6 (6.82) 701 214 88
Jun 2004 539 (77.33) 149 (69.95) 58 (64.44) 125 (17.93) 47(22.07) 26(28.89) 33 (4.73) 17 (7.98) 6 (6.67) 697 213 90
Source: own calculations based on United Kingdom Civil Aviation Authority’ data set.
21
Table 5. Results of estimation of models based on routes. Dependent variable
Ln_ratio_of_
average_prices Ln_Fisher_of_ average_prices
Ln_ratio_of_ median_prices
Ln_Fisher_of_ median_prices
Ln_ratio_of_ min_prices
Ln_Fisher_of_ min_prices
Ln-cost 0.130 (2.85*** 0.060 (2.03)** 0.189 (3.70)*** 0.113 (2.82)*** 0.543 (4.36)*** 0.407 (4.89)*** Inflation 0.036 (1.56) 0.061 (2.70)*** 0.030 (1.03) 0.031 (0.99) -0.051 (0.72) -0.032 (0.41) D_exchange_rate -0.032 (5.62)*** -0.027 (4.85)*** -0.038 (5.68)*** -0.032 (4.36)*** 0.001 (0.07) -0.016 (1.18) Dummy_charter 0.005 (0.42) -0.002 (0.20) 0.009 (0.52) -0.004 (0.37) -0.004 (0.12) 0.023 (0.70) new_route_entry 0.097 (3.91)*** 0.064 (2.47)** 0.237 (6.25)* 0.168 (6.30)*** 0.364 (4.51)*** 0.182 (2.53)*** Ryan_takeover -0.099 (2.55)** -0.084 (2.19)** -0.109 (1.21) -0.139 (2.38)** 0.182 (1.71)* -0.017 (0.17) Easyjet_takeover -0.021 (0.56) 0.029 (0.91) -0.037 (0.68) 0.016 (0.46) 0.096 (1.16) 0.217 (2.91)*** Duopoly_route 0.012 (0.59) -0.004 (0.35) 0.025 (0.86) -0.012 (0.82) -0.168 (1.99)** -0.060 (1.44) Three_firms_route -0.063 (1.23) -0.039 (1.28) 0.076 (1.33) -0.026 (0.55) -0.568 (1.70)* -0.142 (2.18)** Lag_n_of_flights 0.341(4.21)*** 0.231 (2.89)*** 0.301 (3.32)*** 0.275 (3.47)*** 0.946 (2.95)*** 0.734 (2.39)** Low_cost -0.043 (0.92) -0.145 (3.79)*** -0.116 (1.75)* -0.140 (2.16)** -0.220 (1.29) -0.141 (1.20) Jun-Sep03-03 -0.646 (6.74)*** -0.554 (6.22)*** -0.647 (6.01)*** -0.586 (4.96)*** 0.187 (0.87) -0.066 (0.31) Oct-Dec03-03 -0.588 (7.38)*** -0.491 (6.26)*** -0.659 (7.17)*** -0.615 (5.82)** -0.121 (0.68) -0.397 (2.26)** Jan-Mar04-04 -0.086 (1.64) -0.028 (0.55) -0.114 (1.81)* -0.083 (1.19) -0.207 (1.71)* -0.249 (2.17)** Italy -0.096 (1.23) -0.096 (1.48) -0.186 (1.69)* -0.131 (1.29) -0.706 (3.36)*** -0.616 (3.75)*** France 0.022 (0.30) 0.090 (1.23) -0.020 (0.22) -0.010 (0.11) -0.250 (1.07) -0.082 (0.37) Spain -0.164 (2.89)*** -0.052 (0.86) -0.200 (3.08)*** -0.095 (1.32) -0.429 (2.14)** -0.300 (2.00)*** Netherlands 0.148 (3.10)*** 0.149 (2.88)*** 0.196 (3.34)*** 0.189 (3.43)*** 0.281 (1.72)* 0.280 (1.92)* Germany 0.164 (2.43)** 0.151 (2.33)** 0.157 (1.96)** 0.196 (1.91)* -0.536 (2.20)** -0.501 (2.15)** Belgium -0.020 (0.28) -0.042 (0.51) -0.008 (0.06) -0.026 (0.19) 0.203 (0.28) 0.223 (0.29) Greece -0.104 (1.39) -0.028 (0.34) -0.211 (2.42)** -0.085 (0.75) -0.490 (1.49) -0.267 (1.03) Ireland -0.058 (1.05) -0.070 (0.71) 0.006 (0.11) -0.005 (0.07) -0.347 (2.24)** -0.355 (2.67)*** Portugal -0.173 (2.03)** -0.120 (1.40) -0.261 (3.18)*** -0.179 (1.93)* -0.518 (1.65)* -0.354 (1.34) Switzerland 0.192 (2.16)** 0.244 (2.30** 0.178 (1.61) 0.188 (1.86)* -0.148 (0.66) -0.016 (0.07) Sweden -0.342 (4.61)*** -0.190 (2.41)** -0.418 (4.82)** -0.290 (2.97)*** -0.343 (0.98) -0.398 (1.24) Norway 0.242 (1.49) 0.231 (1.61) 0.322 (1.96)* 0.302 (1.88)* -0.788 (2.51)** -0.381 (1.20) Austria -0.017 (0.22) -0.009 (0.13) -0.053 (0.70) 0.008 (0.12) -0.232 (1.15) -0.138 (0.81) Czech Republic -0.059 (0.71) -0.064 (0.81) -0.052 (0.47) -0.037 (0.34) -0.288 (1.92)* -0.187 (1.39) Observations 1303 1303 1303 1303 1303 1303 R2 0.4383 0.4864 0.3880 0.4527 0.1522 0.2609 Wald χ2 2489.38 856.84 1031.05 2315.09 858.96 1514.94 Prob> χ2 0 0 0 0 0 0 Note: Prais-Winsten regression: corrected standard errors (PCSEs), controlling for first order panel-specific correlation and correlated panels. (standard errors) *** p-value of hypothesis test <0.01; ** p-value of hypothesis test <0.05; * p-value of hypothesis test <0.10;
22
Table 6. Results of estimation of models based on city pairs. Dependent variable
Ln_ratio_of_
average_prices Ln_Fisher_of_ average_prices
Ln_ratio_of_ median_prices
Ln_Fisher_of_ median_prices
Ln_ratio_of_ min_prices
Ln_Fisher_of_ min_prices
Ln-cost 0.126 (2.99)*** 0.062 (2.16)** 0.180 (3.71)*** 0.121 (3.23)*** 0.536 (4.20)*** 0.456 (5.21)*** Inflation 0.035 (1.53) 0.062 (2.74)*** 0.030 (1.01) 0.033 (1.07) -0.052 (0.75) -0.034 (0.43) D_exchange_rate -0.033 (5.71)*** -0.027 (4.87)*** -0.038 (5.72)*** -0.032 (4.42)*** -0.001 (0.14) -0.017 (1.28) Dummy_charter 0.008 (0.52) -0.0002 (0.02) -0.007 (0.39) -0.001 (0.13) 0.013 (0.35) 0.024 (0.73) new_route_entry 0.094 (3.90)*** 0.063 (2.49)** 0.233(6.21)*** 0.164 (6.40)* 0.360 (4.54)*** 0.178 (2.50)** Ryan_takeover -0.102 (2.46)** -0.087 (2.20)** -0.121 (1.31) -0.144 (2.47)** 0.130 (1.24) -0.041 (0.39) Easyjet_takeover -0.010 (0.24) 0.038 (1.12) -0.019 (0.35) 0.020 (0.61) 0.135 (1.38) 0.216 (2.53)** Duopoly_citypairs -0.014 (0.53) -0.010 (0.63) -0.020 (0.60) -0.010 (0.57) -0.169 (1.69)* -0.056 (1.10) Three_firms_citypairs -0.034 (1.17) -0.033 (1.85)* -0.036 (1.09) -0.030 (1.50) -0.160 (1.20) -0.064 (0.86) Four_firms_citypairs -0.027 (0.81) -0.0005 (0.02) -0.055 (1.55) -0.009 (0.36) -0.288 (2.06)** 0.029 (0.50) Five_firms_citypairs -0.024 (0.63) -0.019 (0.73) -0.026 (0.74) 0.021 (0.61) 0.059 (0.44) 0.070 (0.81) Six_firms_citypairs -0.008 (0.10) -0.012 (0.14) 0.001 (0.02) -0.052 (0.61) 0.216 (0.97) 0.160 (0.91) Lag_n_of_flights 0.379 (4.48)*** 0.257(2.83)*** 0.380 (3.99)*** 0.244 (2.98)*** 0.830 (1.89)* 0.697 (2.35)** Low_cost -0.059 (1.30) -0.144 (3.73)*** -0.134 (2.07)** -0.139 (2.17)** -0.121 (0.84) -0.108 (0.95) Jun-Sep03-03 -0.649 (6.82)*** -0.552 (6.28)*** -0.649 (6.04)*** -0.580 (4.94)*** 0.180 (0.86) -0.067 (0.32) Oct-Dec03-03 -0.591 (7.47)*** -0.491 (6.32)*** -0.661 (7.20)*** -0.614 (5.84)*** -0.150 (0.85) -0.405 (2.35)** Jan-Mar04-04 -0.089 (1.71)* -0.027 (0.54) -0.118 (1.87)* -0.082 (1.18) -0.221 (1.91)* -0.253 (2.25)** Italy -0.090 (1.20) -0.101 (1.50) -0.169 (1.50) -0.134 (1.35) -0.654 (3.55)*** -0.617 (3.87)*** France 0.021 (0.30) 0.088 (1.16) -0.016 (0.17) -0.008 (0.09) -0.277 (1.44) -0.069 (0.32) Spain -0.157 (2.96)*** -0.056 (0.89) -0.184 (2.89)*** -0.097 (1.36) -0.386 (2.61)*** -0.311 (2.20)** Netherlands 0.146 (2.88)*** 0.155 (3.03)*** 0.191 (3.17)*** 0.204 (3.69)*** 0.178 (1.01) 0.299 (1.91)* Germany 0.172 (2.65)*** 0.163 (2.30)** 0.172 (2.25)** 0.218 (2.03)** -0.463 (2.31)** -0.451 (2.10)** Belgium -0.010 (0.12) -0.036 (0.43) 0.008 (0.05) -0.027 (0.19) 0.224 (0.28) 0.218 (0.28) Greece -0.073 (1.15) -0.002 (0.03) -0.169 (2.24)** -0.070 (0.61) -0.365 (1.35) -0.245 (1.05) Ireland -0.049 (0.89) -0.068 (0.68) 0.006 (0.12) 0.011 (0.17) -0.388 (2.45)** -0.289 (2.04)** Portugal -0.171 (2.07)** -0.127 (1.44) -0.250 (3.21)*** -0.175 (2.01)** -0.492 (2.06)** -0.364 (1.46) Switzerland 0.213 (2.50)** 0.263 (2.42)** 0.205 (1.86)* 0.214 (2.12)** -0.071 (0.36) 0.031 (0.13) Sweden -0.341 (4.68)*** -0.182 (2.25)** -0.413 (4.87)*** -0.280 (2.80)*** -0.332 (1.03) -0.391 (1.29) Norway 0.267 (1.70) 0.251 (1.73)* 0.347 (2.18)** 0.324 (2.00)** -0.639 (2.22)** -0.317 (1.03) Austria -0.021 (0.28) -0.016 (0.23) -0.050 (0.71) 0.004 (0.06) -0.275 (1.47) -0.151 (0.92) Czech Republic -0.048 (0.59) -0.061 (0.77) -0.039 (0.34) -0.026 (0.24) -0.193 (2.00)** -0.178 (1.68)* Observations 1303 1303 1303 1303 1303 1303 R2 0.4384 0.4882 0.3878 0.4541 0.1647 0.2647 Wald χ2 2929.56 521.39 6782.10 13572.88 944.99 313.74 Prob> χ2 0 0 0 0 0 0 Note: Prais-Winsten regression: corrected standard errors (PCSEs), controlling for first order panel-specific correlation and correlated panels. (standard errors) *** p-value of hypothesis test <0.01; ** p-value of hypothesis test <0.05; * p-value of hypothesis test <0.10;